Wireless Sensor Networks For Marginal Farming in India: Jacques Panchard

WIRELESS SENSOR NETWORKS FOR MARGINAL FARMING IN INDIA
` PRESENTEE A LA FACULTE INFORMATIQUE ET COMMUNICATIONS
ECOLE POLYTECHNIQUE FEDERALE DE LAUSANNE

` POUR LOBTENTION DU GRADE DE DOCTEUR ES SCIENCES
PAR
JACQUES PANCHARD
` Ingenieur Diplome en Systemes de Communication (M.Sc.), Ecole Polytechnique Federale de Lausanne, Suisse de nationalite suisse
jury: ` Prof. Jean-Pierre Hubaux, directeur de these Dr. Pearl Pu Faltings, presidente de jury Prof. H.S. Jamadagni, rapporteur Prof. Andre Mermoud, rapporteur Dr. Kentaro Toyama, rapporteur
Lausanne, EPFL 2008
ii
Abstract
In this dissertation, we explore the potential of wireless sensor networks (WSNs) in an original context, the small agriculture of Developing Countries (DCs). Our goal is to confront an emerging technology with a concrete problem of world-wide dimensions, the sustainability of farming for small land-holders living in conditions of water scarcity. Based on a survey about information needs, we design a series of precise use cases, provide system design, implementation and deployment guidelines for the technology, present a toolkit including an original interface to wireless sensors for non-specialists, and bring to the attention of the research community the lessons we learned in the process. In the rst part, we present the environmental challenges faced by the developing world and identify relevant applications of environmental monitoring in this context. Then, we proceed with a review of the technology of environmental monitoring in the broad context of agriculture and formally present the opportunity represented by WSNs. Finally we show how this can be applied to addressing a crucial problem of DCs, namely rural poverty. The second part of the dissertation is devoted to the collaborative design of a decision-support tool for marginal agriculture using wireless sensor networks. We rst describe a survey that was made in 2004 in three villages of Karnataka, India. The results highlighted the potential that environment-related information has for the improvement of farming strategies in the face of highly variable conditions, in particular for risk management strategies (choice of crop varieties, sowing and harvesting periods, prevention of pests and diseases, efcient use of irrigation water etc.). The results were used to identify potential use cases for an environmental monitoring system for agriculture, and to make crucial design decisions for this system. At this point, we present our toolkit in detail and proceed with its assessment. Deployment issues are covered in detail, as they are critical for the success of such a system. The results of our eld deployments, both in Switzerland and in India, highlighted the potential of the technology and demonstrated its applicability in the eld. However, the direct use of this technology by the farmers themselves did not foster the expected participation of the population. This made it difcult to develop the intended decision-support system. The third part of this dissertation addresses the lessons learned and their consequences for upcoming experiments and deployments. We take the following position: Currently, the deployment of WSN technology in developing regions is more likely to be effective if it targets scientists and technical personnel as users, rather than the farmers themselves. We base this nding on the lessons learned from the COMMON-Sense system deployment and the results of an extensive user experiment with agriculture scientists, which is extensively described. We also took steps to make the deployment and maintenance of wireless sensors easier. Their limited resources, indeed, make them a challenging tool to handle in the eld. In particular, they lack a proper display, which makes them difcult to deploy and to manage, once they are deployed. Accordingly, we present Sensor-Tune, a light-weight deployment and maintenance support tool for wireless sensor networks. This tool is based on an auditory user interface using sonication. Sonication refers to the use iii
of audio signals (mostly non-speech) to convey information. We explore the potential of this approach, in particular how it allows users to overcome the inherent limitations of visual interfaces. We then justify our design choices, and present typical WSN applications where sonication can be particularly useful. Finally, we present the prototype that we built, and we describe a user experiment that we conducted in early 2008, which is the rst reported attempt to put a multi-hop wireless sensor network deployment in the hands of non-specialists. In a conclusive part, we go beyond the mere technology and technology use, by advocating an original use of Information and Communication Technologies (ICT) in the context of developing countries. We believe our demand-driven approach for the design of appropriate ICT tools that are targeted at the resource poor to be relatively new. In order to go beyond a pure technocratic approach, we adopted an iterative, participatory methodology. Keywords: tem, toolkit wireless sensor networks, sustainable development, developing countries, agriculture, sys-
R sum e e
Dans cette dissertation, nous explorons le potentiel des r seaux de capteurs sans l (RCSFs) dans e un contexte particulier et original. Notre but est de confronter cette technologie en devenir avec un ` probl` me concret aux enjeux mondiaux, a savoir la petite agriculture dans les r gions arides des pays en e e d veloppement (PDs). e A partir dune etude de terrain sur les besoins en information de populations rurales du Karnataka (Inde du Sud), nous avons concu une s rie dapplications pr cises, pour lesquelles nous fournissons les e e d tails de design, impl mentation et d ploiement. Nous pr sentons aussi un syst` me incluant une intere e e e e face originale qui permet de mettre des capteurs sans l dans les mains de non-sp cialistes. Finalement, e nous attirons lattention de la communaut scientique sur les lecons que nous avons apprises dans le e cadre de ce projet de d veloppement et coop ration. e e Dans la premi` re partie de la dissertation, nous pr sentons les d s environnementaux que doivent e e e relever les pays en d veloppement, et nous identions des domaines dapplications pour le monitoring e de lenvironnement dans ce cadre. Puis, nous passons en revue l tat de lart de cette technologie dans e le contexte plus large de lagriculture et pr sentons formellement la fen tre dopportunit ouverte par e e e ` les r seaux de capteurs. Finalement, nous montrons comment appliquer ces technologies a un probl` me e e ` crucial des PDs, a savoir la pauvret rurale. e ` La deuxi` me partie est consacr e au design participatif dun outil daide a la d cision pour la petite e e e ` agriculture (ou agriculture marginale) bas sur les RCSFs. Tout dabord, nous d crivons une etude a e e laquelle nous avons particip en 2004 dans trois villages du Karnataka. Les r sultats obtenus ont mis en e e evidence le potentiel de linformation environnementale pour lam lioration des strat gies agricoles dans e e ` un climat semi-aride soumis a des fortes variations saisonni` res. Cela concerne en particulier la gestion e du risque, avec le choix des vari t s a cultiver, les p riodes de semailles et de moisson, la pr vention ee ` e e des maladies et des parasites ou lutilisation efcace de leau dirrigation. Les r sultats nous ont servi e ` a identier des cas dutilisation pour un syst` me de monitoring de lenvironnement pour lagriculture e marginale. Certaines d cisions cruciales quant au design en d pendent aussi directement. A ce stade, e e nous pr sentons egalement notre bote a outils et l valuation que nous en avons faite. Nous couvrons e ` e en d tail les questions de d ploiement, parce quelles sont critiques pour le succ` s dun tel syst` me. e e e e Les r sultats de nos d ploiements, en Suisse comme en Inde, ont mis en evidence le potentiel des e e RCSFs et d montr leur application sur le terrain. Pourtant, lutilisation directe de cette technologie par e e des agriculteurs indiens na pas b n ci de la collaboration esp r e de la population. En cons quence, e e e ee e ` le d ploiement du syst` me daide a la d cision pr vu sest av r difcile. e e e e ee La troisi` me partie de cette dissertation aborde les leons apprises et leurs cons quences pour exp riences e e e ` et d ploiements a venir. Nous d fendons le choix suivant: pour linstant, le d ploiement de RCSFs dans e e e des r gions en d veloppement a plus de chance d tre efcace sil est dirig vers des utilisateurs sciene e e e tiques ou techniciens, plut t que des agriculteurs. Nous basons cette recommendation sur nos propres o d ploiements, ainsi que sur une exp rience-utilisateurs men e aupr` s de scientiques de lagriculture, e e e e v
exp rience que nous d crivons en d tail. e e e Nous avons egalement pris des mesures pour rendre le d ploiement et lentretien des capteurs sans e ` e l plus facile. Leurs ressources limit es, en effet, en font un outil difcile a g rer sur le terrain. En e ` e ` e particulier, ils nont pas un afchage graphique, ce qui les rend d licats a d ployer et a g rer une fois e quils sont d ploy s. En cons quence, nous pr sentons Sensor-Tune, un outil de support au d ploiement e e e e e ` et a la maintenance de capteurs sans l. Cet outil est bas sur une interface-utilisateur auditive utilisant le e concept de sonication. La sonication consiste en lutilisation des signaux audio (essentiellement nonverbaux) pour transmettre de linformation. Nous explorons le potentiel de cette approche, en particulier comment elle permet de surmonter les limitations inh rentes aux interfaces visuelles. Nous justions e nos choix de design et pr sentons des applications typiques pour lesquelles la sonication peut etre e particuli` rement utile. Enn, nous pr sentons le prototype que nous avons d velopp en laboratoire, et e e e e ` d crivons une exp rience-utilisateurs que nous avons men e au d but de 2008, et qui constitue a notre e e e e connaissance la premi` re tentative de mettre un r seau sans l auto-organis entre les mains de none e e sp cialistes. e En conclusion, nous allons au-del` des consid rations purement technologiques en pr conisant une a e e utilisation originale de linformation et des technologies de la communication (TIC) dans le contexte des pays en d veloppement. Pour d passer une approche purement technocratique, nous avons adopt une e e e d marche it rative et une m thodologie participative. e e e Mots-cl s: e syst` mes e r seaux de capteurs sans l, d veloppement durable, pays en d veloppement, agriculture, e e e
Acknowledgements
I want to thank rst my advisor, Professor Jean-Pierre Hubaux, for allowing me to pursue this exciting and original research topic, and for guiding me during all my time at EPFL. His trust and help in all aspects of the PhD never faltered. I want to express my gratitude to the members of my thesis committee, Prof. H.S. Jamadagni, Prof. Andr Mermoud, Dr. Kentaro Toyama, as well as the president of the committee, Dr. Pearl Pu, for the e time and effort that they invested in criticizing my dissertation, and for the interest they expressed for my research. The work presented in this thesis was supported by the National Competence Center in Research on Mobile Information and Communication Systems (NCCR-MICS) and by the EPFL-SDC Fund. I am grateful for this support. During my PhD, I had the opportunity to collaborate with wonderful people, who made these years an exceptional learning experience, not only professionally, but at a personal level as well. In particular, many thanks to Seshagiri Rao, T.V. Prabhakar and M.S. Sheshshayee, who unveiled for me a tiny portion of the Great Indian Novel. Special thanks go to my friend Andr Pittet and his wife Catherine, for their e unconditional hospitality, their incomparable insight and their constant support during my stays in India. I want also to pay tribute to the Sensorscope group at LCAV (Thomas Schmid and Henri DuboisFerri` res, and later Guillermo Barrenetxea), the giant on whose shoulders I stood. e Many thanks to my colleagues at LCA for making this PhD such an enjoyable experience. I am particulary indebted to my successive ofce mates, Michal Piorkowski, Jun Luo and Julien Freudiger, for bearing with me in such a friendly way. I am thankful to the staff of LCA: Danielle Alvarez, Holly Cogliati, Angela Devenoge, and Patricia Hjelt for helping me with all administrative issues, as well as Philippe Chammartin, Jean-Pierre Dupertuis and Marc-Andre L thi for keeping the computing inu frastructure up and running. Finally, my gratitude goes to my family for their love, support and encouragement during all my studies. Most of all, thank you my love, Sandra, for sharing my deepest moments of joy, and for supporting and encouraging me whenever fear or discouragement where looming over me.
vii
Main Abbreviations
CEDT: CSN: DC: ICT: ICT4D: IISc: MICS: NCCR: NGO: SDC: WSN: Centre for Electronics Design and Technology COMMON-Sense Net Developing Country Information and Communication Technology Information and Communication Technologies for Development Indian Institute of Science, Bangalore, India Mobile Information and Communication Systems National Center of Competence in Research Non-Governmental Organization Swiss Agency for Development and Cooperation Wireless Sensor Network
ix
Contents
Introduction 1 Environmental Challenges in Developing Countries 1.1 The Millennium Development Goals and the Environment . 1.2 The Importance of Environmental Monitoring . . . . . . . . 1.3 Agriculture . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3.1 Rationale . . . . . . . . . . . . . . . . . . . . . . . 1.3.2 Potential Applications . . . . . . . . . . . . . . . . 1.4 Air Pollution and Trafc . . . . . . . . . . . . . . . . . . . 1.4.1 Potential Applications . . . . . . . . . . . . . . . . 1.5 Water Quality Monitoring . . . . . . . . . . . . . . . . . . . 1.6 Agriculture and Water in India: a Brief Historical Perspective 1.7 Indias Agriculture Today . . . . . . . . . . . . . . . . . . . 1.7.1 Facts and Figures . . . . . . . . . . . . . . . . . . . 1.7.2 A Recent and Growing Concern: Water Scarcity . . 1.7.3 The Specic Case of Karnataka . . . . . . . . . . . 1.8 India Today: the Current Institutional Framework . . . . . . 1.8.1 States and Central Government . . . . . . . . . . . . 1.8.2 Local Institutions . . . . . . . . . . . . . . . . . . . 1.9 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . 1 5 5 6 7 7 7 8 9 10 10 12 12 12 13 13 13 14 14 17 17 17 18 18 18 18 19 20 20 21 21 22 23
. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . .
New Opportunities for Environmental Monitoring and Agriculture 2.1 Usual Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1.1 Stand-Alone Sensors . . . . . . . . . . . . . . . . . . . . 2.1.2 Laboratory Analysis . . . . . . . . . . . . . . . . . . . . 2.1.3 Remote Sensing . . . . . . . . . . . . . . . . . . . . . . 2.1.4 Telemetry . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Wireless Sensor Networks . . . . . . . . . . . . . . . . . . . . . 2.3 Design Dimensions in Environmental Monitoring . . . . . . . . . 2.4 Where Do WSNs Stand? . . . . . . . . . . . . . . . . . . . . . . 2.4.1 Wireless Networking . . . . . . . . . . . . . . . . . . . . 2.4.2 Self-Organization . . . . . . . . . . . . . . . . . . . . . . 2.4.3 Efcient Power Management . . . . . . . . . . . . . . . . 2.4.4 Modularity . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.5 Web-based Data Management . . . . . . . . . . . . . . . xi
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
2.5
2.6
Sensors and Agriculture . . . . . . . . . . 2.5.1 Soil Moisture . . . . . . . . . . . 2.5.2 Soil Salinity and PH . . . . . . . 2.5.3 Climatic Variables . . . . . . . . WSNs in Agriculture . . . . . . . . . . . 2.6.1 Vineyard Temperature Monitoring 2.6.2 Potato Disease Prevention . . . . 2.6.3 Tomato Disease Prevention . . . . 2.6.4 Cattle Monitoring . . . . . . . . . 2.6.5 Paddy Field Monitoring . . . . . 2.6.6 Discussion . . . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
23 23 25 25 26 26 27 27 28 28 29 31 31 31 32 33 34 35 36 36 36 39 43 43 44 45 46 47 47 50 51 51 52 53 54 54 56 56 57 58 58 59 59 59
WSNs and Developing Countries 3.1 Existing WSN Projects . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1.1 Groundwater Arsenic Contamination Assessment in Bangladesh 3.1.2 SenSlide, A Sensor Network Based Landslide Prediction System 3.1.3 Wireless Sensor Network for Water Quality Management . . . . 3.1.4 Flood Detection System for Honduras . . . . . . . . . . . . . . 3.1.5 Road Surface Condition Monitoring . . . . . . . . . . . . . . . 3.1.6 Other Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 A New Tool for Developing Regions? . . . . . . . . . . . . . . . . . . 3.2.1 Assets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.2 Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . Wireless Sensor Networks for Marginal Agriculture in India 4.1 Project, Consortium and Funding . . . . . . . . . . . . . . . . . 4.2 COMMON-Sense Net: a Decision-Support Tool for Agriculture 4.3 Setting the Context . . . . . . . . . . . . . . . . . . . . . . . . 4.3.1 The Pavagada Region . . . . . . . . . . . . . . . . . . . 4.3.2 The Chennakeshavapura Village . . . . . . . . . . . . . 4.3.3 Type of Agriculture . . . . . . . . . . . . . . . . . . . . 4.3.4 Marginal Farmers . . . . . . . . . . . . . . . . . . . . . 4.4 A Survey and Analysis on Farmers Needs . . . . . . . . . . . . 4.4.1 Survey Methodology . . . . . . . . . . . . . . . . . . . 4.4.2 Summary of Results . . . . . . . . . . . . . . . . . . . 4.4.3 Interpretation and Motivation . . . . . . . . . . . . . . 4.5 Use Cases and Related Environmental Data . . . . . . . . . . . 4.5.1 Crop Modeling . . . . . . . . . . . . . . . . . . . . . . 4.5.2 Water Conservation Measures . . . . . . . . . . . . . . 4.5.3 Pest and Disease Prediction/Prevention . . . . . . . . . 4.5.4 Water Management for Decit Irrigation . . . . . . . . . 4.6 Design Guidelines . . . . . . . . . . . . . . . . . . . . . . . . . 4.6.1 Technical Point-of-View . . . . . . . . . . . . . . . . . 4.6.2 Scientic Point-of-View . . . . . . . . . . . . . . . . . 4.6.3 Economical and Sociocultural Point-of-View . . . . . . 4.7 Methodology: Science and Farmers . . . . . . . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . .
System Design and Implementation 5.1 Design Options . . . . . . . . . . . . . . . . . . . . 5.1.1 Data Generation Strategy . . . . . . . . . . . 5.1.2 Data Transport Strategy . . . . . . . . . . . 5.2 Design Choice: Overview . . . . . . . . . . . . . . . 5.3 Embedded Probes . . . . . . . . . . . . . . . . . . . 5.4 Wireless Sensor Network: Data Collection Subsystem 5.4.1 Radio Range . . . . . . . . . . . . . . . . . 5.4.2 Power Consumption . . . . . . . . . . . . . 5.5 Hybrid Network: Data Transit Subsystem . . . . . . 5.5.1 WiFi Bridge . . . . . . . . . . . . . . . . . . 5.5.2 GPRS Bridge . . . . . . . . . . . . . . . . . 5.5.3 Performance Evaluation . . . . . . . . . . . 5.6 Data Management and Processing . . . . . . . . . . 5.6.1 Tables . . . . . . . . . . . . . . . . . . . . . 5.7 A Web-based Tool . . . . . . . . . . . . . . . . . . . 5.7.1 Data Display . . . . . . . . . . . . . . . . . 5.7.2 Network Statistics . . . . . . . . . . . . . . 5.7.3 Commands . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . .
61 61 61 62 63 63 64 65 66 67 68 69 70 71 71 72 73 74 74 77 77 79 80 80 81 85 86 89 89 93 94 96 97 97 99 99 99 100 100 100 101 102 103 104
A Wireless Sensor Network Toolkit for Rural India 6.1 Changins . . . . . . . . . . . . . . . . . . . . . . . . 6.2 Chennakeshavapura . . . . . . . . . . . . . . . . . . . 6.3 Issues of a Rural Deployment . . . . . . . . . . . . . . 6.3.1 Hardware Issues . . . . . . . . . . . . . . . . 6.3.2 Probe Assessment . . . . . . . . . . . . . . . 6.3.3 Power Management . . . . . . . . . . . . . . 6.3.4 Environment . . . . . . . . . . . . . . . . . . 6.3.5 Power and Telecommunications Infrastructure . 6.3.6 Connectivity Issues . . . . . . . . . . . . . . . 6.4 Human Issues . . . . . . . . . . . . . . . . . . . . . . 6.5 System Overview . . . . . . . . . . . . . . . . . . . . 6.6 Lessons Learned . . . . . . . . . . . . . . . . . . . . Making the Invisible Audible 7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . 7.2 State of the Art . . . . . . . . . . . . . . . . . . . . 7.2.1 WSN deployment and Maintenance Support . 7.2.2 Sonication . . . . . . . . . . . . . . . . . . 7.2.3 WSNs and their End-Users . . . . . . . . . . 7.3 Sonication for Sensor Networks . . . . . . . . . . . 7.3.1 Applications . . . . . . . . . . . . . . . . . 7.3.2 Advantages . . . . . . . . . . . . . . . . . . 7.3.3 Challenges . . . . . . . . . . . . . . . . . . 7.3.4 Signal and Noise Metaphor . . . . . . . . . . 7.4 System Design . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
7.5
7.6 7.7
7.8 8
7.4.1 System Model . . . . . . . . . 7.4.2 Tool and Scenarios . . . . . . . 7.4.3 Protocols . . . . . . . . . . . . 7.4.4 Sonication Mapping Strategy . Initial Exploration: User Survey . . . . 7.5.1 Description . . . . . . . . . . . 7.5.2 Results . . . . . . . . . . . . . Prototype Implementation . . . . . . . 7.6.1 Prototype Description . . . . . Experimental Validation . . . . . . . . 7.7.1 Comparable Graphical Interface 7.7.2 Experimental Design . . . . . . 7.7.3 Participants . . . . . . . . . . . 7.7.4 Results . . . . . . . . . . . . . Discussion . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
104 105 108 111 112 112 113 114 114 117 117 118 119 119 120 123 123 124 124 126 126 128 129 133 134 135 135 136 137 137 138 138 140 141
Usability and Usefulness of the System 8.1 Charting the Paradigm Shift . . . . . . 8.1.1 Choosing the Target Population 8.1.2 Goal and Methodology . . . . . 8.2 Experiment Results . . . . . . . . . . . 8.2.1 Questionnaires . . . . . . . . . 8.2.2 User Activity Logging . . . . . 8.2.3 Debrieng Meetings . . . . . . 8.3 Discussion . . . . . . . . . . . . . . . . 8.3.1 Usefulness . . . . . . . . . . . 8.3.2 Usability . . . . . . . . . . . . 8.3.3 Use . . . . . . . . . . . . . . . 8.3.4 Sectoral Analysis . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
Building a Knowldedge Society with the Use of WSNs? 9.1 Experimental Technology for Social Change? . . . . . . . . . . . . 9.2 Design/Implementation Gaps . . . . . . . . . . . . . . . . . . . . . 9.3 Knowledge Creation, Context and Knowledge Assets . . . . . . . . 9.4 Apprenticeship & Participatory Methods to Develop ICT Capacities 9.5 From Theory to Practice . . . . . . . . . . . . . . . . . . . . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
Conclusion 143 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143 References Index 145 156
Introduction
Wireless Sensor Networks (WSNs) are increasingly considered by the scientic community as the future of Environmental Monitoring: Providing at a low cost the possibility to gather and process all sorts of data with a space and time resolution unimaginable before, these networks are viewed as a critical element of the revolution of ubiquitous computing . The idea of automating the collection of physical data in order to monitor environments is not new. But recent technological advances have allowed for the networking of a wide variety of sensors, independently from any preexisting infrastructure. Whenever physical conditions in a milieu change rapidly over space and time, WSNs allow for real-time data processing at a minimal cost. Their capacity to organize spontaneously in a network makes them easy to deploy, expand and maintain, and provides resilience to the failure of individual measurement points. After a variety of test beds have been reported in the scientic literature [SPMC04, ea02, BTB04, SDFV, BIS+ 08], the rst commercial applications of WSNs started appearing on the market, in the area of home or building automation (Control4 [Con], HomeHeartBeat [Hea]), safety (LifeTag [Ray]), automatic meter reading (Wellspring [Irrb]), storage monitoring (ip01 [IP0]), oil industry and agriculture ( ko [Cro]). If such systems are to be adopted by a wide customer basis, early adopters will have to e demonstrate the gains of using this technology. The search of a killer application is by many means still ongoing. In this context, it is legitimate to ask whether the wireless sensing technology can nd new niches of applications. The work presented in this thesis aims at broadening the scope in this quest. We investigate whether the WSNs application paradigm can be adapted from ubiquitous computing for mass consumption markets to a decision-support tool for sustainable development in Developing Countries (DCs). To some extent, researchers have also tried to apply WSNs to this new context. However, such projects remain rare, and their socioeconomic effect remain to be proven. Throughout this thesis, we explore the potential of such networks in developing countries with a particular use case: environmental monitoring for marginal agriculture. Despite their rapid urban growth, DCs have to sustain an important, if not predominant, rural population. The livelihood of these populations has changed dramatically with the cumulative effects of globalization and the Green Revolution. The new situation has raised formidable challenges for the small land holders, who still represent the bulk of the rural population. In 2008, they still live in a fragile equilibrium, dramatically illustrated by the renewed threat of a worldwide food crisis [Edi08]. In this dissertation, we investigate the use of the WSN technology in the context of rural development, and we seek to determine if -and how- WSNs can help tackle some of the environmental challenges awaiting the developing countries in the face of globalization and climate change. This work is resolutely multidisciplinary, as we aim at solving relevant social problems, with the appropriate use of environmental data for agriculture, through wireless networking. This triple (social, scientic and engineering) challenge is reected in the choice of projects partners, who come from both 1
INTRODUCTION
the academic world and the civil society, and from two cultural backgrounds: India and Switzerland. The choice of India as a test-case for developing countries might be questioned by some, because this country is emerging as a major economic power of the 21st century. India, however, is a country with two faces. In opposition to the rising power of mega-cities, the life conditions are deteriorating in the countryside, still home to almost two-thirds of the total population. As a consequence, the country is by no means alien to the issues faced by developing countries, but it possesses the brain-power and the technological skills to address these problems efciently. Hence, it is an ideal setting for the sort of investigation we wanted to pursue, providing both the appropriate problem and highly skilled local partners to solve it. In Chapter 1, we set the context by emphasizing the role of the environment in the sustainable development of emerging economies and less-developed countries alike. Setting our focus on India as a test-case, we highlight the role of agriculture, and draw a brief historical and institutional outline of this sector. In Chapter 2, we present the state of the art in environmental monitoring, with a special focus on agriculture. First we draw a panorama of wireless sensor networks, then we give an outline of the new sensing technologies that are relevant in the cropping eld. We also give a description of agricultural practices that benet from a precise knowledge of the environmental conditions in and around the cultivated plot. Finally, we propose a taxonomy of environmental monitoring, whose goal is to help decide when WSNs are a suitable option for a given task. In Chapter 3, we highlight the potential of WSNs in developing countries. Firstly, we review exhaustively the existing initiatives, then identify and explain through them the main assets and challenges associated with this technology. Chapter 4 is devoted to the user requirements of such an application. First, we introduce the test case that we chose in order to explore the risks and opportunities offered by WSNs: a wireless sensor network as a support tool for small land-holding agriculture in India. Then we present the survey about farmers information needs in which we participated in 2004, and which resulted in the broad denition of WSN applications for marginal agriculture. In Chapter 5, we present the design and implementation choices that we made in the course of the COMMON-Sense Net project. Several alternatives, such as hybrid networks, delay-tolerant networking, periodic data collection and data on-demand are discussed. Chapter 6 highlights the challenges that are still at hand after all the technical hurdles have been solved. It makes it clear that to this day, deploying a self-organized wireless sensor network is still a complex process, especially in a remote areas of a developing country. We proceed to a formal evaluation of the performance of our system, evaluation that was compiled in the light of two years of operating networks in a rural environment. The lessons learned led us to design original solutions for deployment support, which we detail in Chapter 7. In this chapter, we describe the elements of the toolkit that we propose for the deployment of wireless sensor networks in rain-fed agriculture research. This toolkit is based on an auditory user interface using sonication1 . We explore the potential of this approach, particularly how it allows users to overcome the inherent limitations of visual interfaces. We then justify our design choices and present typical WSN applications where sonication can be particularly useful. Finally, we present the prototype that we built, and we describe a user experiment that we conducted in early 2008, which is the rst reported attempt to put a multi-hop wireless sensor network deployment in the hands of non-specialists. Chapter 8 focuses on the users again, in this case agriculture scientists who gave us their feedback
1
Sonication refers to the use of audio signals (mostly non-speech) to convey information.
on the technology developed in the framework of this project. Based on our experience, we take the following position in this chapter: Currently, the deployment of WSN technology in developing regions is more likely to be effective if it targets scientists and technical personnel as users, rather than the farmers themselves. We base this claim on the lessons learned from the COMMON-Sense Net system deployment and the results of an extensive user experiment with agriculture scientists. The societal and methodological issues are covered in Chapter 9, before we draw a conclusion on the work accomplished so far, and we set landmarks for future developments.
INTRODUCTION
Chapter 1
Environmental Challenges in Developing Countries

Developing countries face many challenges on their way to industrialization and economical well-being. Among them, ensuring environmental sustainability while promoting economic growth is more and more regarded as a critical objective. As was illustrated by the negotiations that took place during the UN Climate Talks held in Bali in December 2007, the role of emerging economies in the ght against global warming can no longer be ignored [FR07], even if their levels of greenhouse gas emissions per capita is still signicantly lower than those of industrialized nations. Already today, DCs are confronted with severe problems linked with the change of climatic patterns and the increased strain on their water resources caused by a booming population and improving living standards. In this chapter, we investigate the potential of environmental monitoring in DCs from the perspective of sustainable development. We do not address yet the specic issue of technology, although the utility of a network of sensors will become apparent in most of the use cases.
1.1
The Millennium Development Goals and the Environment
In general, the environmental threats affecting DCs are due to a variety of factors, from the decrease and degradation of water resources caused by unsustainable agricultural practices, to the problem of urban air pollution, to the general health concerns caused by unsatisfactory access to clean fresh water. There is a growing tension between short-term economic development goals and a long-term sustainable environmental policy, as exemplied by the eight Millennium Development Goals [UN 02]. Endorsed by 189 United Nations member states in 2000, the Millennium Declaration lists the eight following goals as the primary targets for an equitable and sustainable human development: 1. Eradicate extreme poverty and hunger 2. Achieve universal primary education 3. Promote gender equality and empower women 4. Reduce child mortality 5. Improve maternal health 6. Combat HIV/AIDS, malaria, and other diseases 7. Ensure environmental sustainability 5
CHAPTER 1. ENVIRONMENTAL CHALLENGES IN DEVELOPING COUNTRIES
8. Develop a global partnership for development In order to highlight the importance of environmental monitoring in developing countries, we show in the following section what role this technique can play in helping to meet at least three of the eight Millennium Goals: Eradicate extreme poverty and hunger: A large percentage of the population living in emerging and developing countries is still rural. The subsistence difculties of the poor farmers represent a major cause of both rural and urban poverty, because mass migration to the cities results in an increase in unemployment and slum population. A better understanding and monitoring of the environmental conditions in which farming is done can make it more protable and sustainable (irrigation, crop management, etc.). Reduce child mortality: About 15% of the child mortality in developing countries is due to diarrhoeal diseases. In this case an environmental factor, water quality, is a direct culprit. In a more general sense, child mortality rates are correlated with poverty levels [Wag02], so it can be assumed that reducing poverty by providing better means of subsistence in rural areas will decrease mortality. Ensure environmental sustainability: Sustainable development involves pollution monitoring, as well as the monitoring and protection of natural resources and the prevention of natural disasters.
1.2
The Importance of Environmental Monitoring
There are several reasons environmental issues are not addressed properly today. They tend to be low on the agenda of the countries whose main objective is to build a prosperous economy and to reach higher standards of living [MC]. Often the longer term effects of environmental degradation are not well known or understood. Information is a key issue in this regard, because only a precise knowledge of the environment can lead to a proper assessment of the situation and a clear understanding of the consequences. Only with this knowledge can timely alerts be issued and appropriate environmental policies be implemented. As a consequence, we claim that environmental monitoring is not anymore a luxury for wealthy societies, but a necessity for all. Table 1.1 presents three applications of environmental monitoring that are particularly adapted to the developing countries, along with their relevance with regard to the MDGs listed above. We develop them in detail in the following section. Of course this list is not exhaustive. With proper collaboration with the scientic communities concerned with the Millennium Goals, such as agronomists, hydrologists, health professions etc., other relevant use cases can probably be found.
Poverty and Hunger YES NO ? Child Mortality YES YES YES Environmental Sustainability YES YES YES
Agriculture Air Pollution and Trafc Water Quality Monitoring
Table 1.1: Environmental Monitoring and MDGs: correspondence matrix
1.3. AGRICULTURE
1.3
1.3.1
Agriculture
Rationale
Agriculture is by far the human activity that uses the most of freshwater resources (65-70%). Worldwide, only 17% of the croplands are irrigated, but they account for 40 % of the global harvest. More than ever, water management has become vital for agriculture. Poor farmers around the globe, who survive on rainfed farming, would more than double their average crop yield if they could use irrigation [HDM+ 06], [FotUN06], [TGCB04]. But for this to happen, irrigation techniques need to be made sustainable. Generally speaking, it has been shown that irrigation efciency in developing countries can be low [MTY05]. It is estimated that the overall efciency of agricultural water use is currently lower than 30 % [Wal00]. Large losses occur in conveyance and distribution systems. Considerable volumes of water are wasted due to inadequate irrigation management, which can also lead to water-logging or to soil salinization (10 to 15 % of irrigated lands worldwide) [TM93], [SOF99]. Furthermore, due to overuse, irrigation is frequently responsible for the depletion of the groundwater table and loss of aquifer storage capacity [DC07]. Better operation and management of irrigation water leads to signicant savings and to a more sustainable use of water resources, as well as enhanced soil productivity [HDM+ 06, MTY05]. To achieve this, systematic monitoring is required for the soil water content in the root zone, as well as for other relevant variables (soil temperature, depth and salt content of the groundwater, climatic parameters, etc.). Indeed, real-time knowledge of these parameters would allow the farmer to dene precisely the time and the amount of water needed at each irrigation, based on the estimated needs of the plant at the corresponding stage of its development, and on the water available in the soil. As a result, the amount of water used could be minimized, which would result not only in cost savings, but also in better preservation of the soil against salinization or water logging. A proper environmental monitoring system for water management in agriculture has demanding requirements. Because of the spatial and time variability of the relevant parameters, continuous monitoring at several locations is necessary. In the cases where the amount of water is limited, an overview of the elds is necessary to determine the optimal distribution of water. This is especially true if the parcels are fragmented, as is the case in most developing countries, where the average land holding does not exceed a few hectares. In this situation, a community-based management of scarce water resources could lead to a more efcient use of these resources throughout the cropping season. As the optimal irrigation timewindow is small, there is a strong real-time component. Finally, it is necessary to minimize intrusiveness, because free access to the elds is necessary throughout the cropping season.
1.3.2
1.3.2.1
Potential Applications
Irrigation Management
As the limits of groundwater exploitation are reached, farmers are beginning to invest in micro-irrigation technologies to conserve water. The cost of these technologies has been declining. For example, the cost of trickle (or drip) irrigation equipment is US $0.03 per square meter or US$300 per hectare, easily affordable for those growing high-valued crops, and the equipment can be manufactured locally [WCS04], [Bra01]. This means that this technique is becoming attractive even for small farmers, as trickle irrigation can improve water use efciency by up to 60 to 80 percent, compared to traditional surface irrigation techniques [RTB98].
Because micro-irrigation efciency depends on the adjustment of water provided to the water demand of the plant, improved environmental information is instrumental to its successful implementation, especially in a demand-based system. A likely method is to determine signicant thresholds in soil moisture as a function of soil, crop type and crop growth stage, and to monitor the soils water content in order to know when to add water and how much. Similar approaches have been proposed in the literature [GSR00], [MHH+ 96]. As shown in the survey on sensors that we present in Chapter 2 (Section 2.5), inexpensive soil moisture sensors exist today. The ne-grained instrumentation of agricultural parcels is hence possible at an affordable cost. 1.3.2.2 Pest and Disease Control
There is an interest in monitoring the probability of occurrence of pests or diseases based on the evolution of climatic parameters (temperature, humidity, soil moisture). In this way, farmers can schedule more efciently the application of pesticides and fungicides and limit the associated monetary and environmental costs. There are several environmental parameters susceptible to inuence the emergence of pests and diseases in plants. Air humidity and temperature are known factors for a variety of crops, e.g. potato fungi [Bag05] or grapevines [CW03]. Soil moisture and leaf wetness often also play a role [AW00]. We present appropriate sensors for such measurements in the next chapter. The question of the time and space variability remains open. Air temperature and humidity can evolve rapidly over time but usually slowly over space. However, cropping elds may present features of mesoor microclimates [BTB04], justifying dense environmental monitoring systems, if only for research and validation purposes. As for soil moisture, it may present high space variability but slow time response. Here again, clear scientic evidence is lacking to dene the optimal granularity of measurements. 1.3.2.3 Other Applications
Many agricultural practices designed to improve productivity in the short-term can have devastating long-term effects. Intensive use of nitrogenous fertilizers can result into soil acidication [Moo]. Soil salinization can be caused by excessive irrigation [TM93]. The same can provoke waterlogging or, on the contrary, water-table depletion. Control of all these phenomena would benet from enhanced environmental monitoring techniques. Whereas it is still difcult today to assess the pH of the soil in-situ, simple piezometers can be used to monitor the level of the water table. The appropriateness of installing measurement systems in the eld for continuous monitoring needs to be carefully examined, however. Contrary to the previous use cases, these applications do not necessarily need a fast response, nor high granularity data.
1.4
Air Pollution and Trafc
In 2006, according to the last UN-habitat report, over 50% of the worlds population lives in cities. In many urban areas of developing countries, air pollution is a major concern, especially in the mega cities of Asia [HKV02]. According to Romieu et al. [RKB04], acute respiratory infections (ARIs) are the most common cause of illness and death in children in the developing world. One major challenge is to develop monitoring tools that can assess precisely pollution levels as a function of the location and the time, in order to identify the precise causes of the problem (e.g. industrial
1.4. AIR POLLUTION AND TRAFFIC
plants, light or heavy vehicles trafc), to raise the awareness of the authorities and the civil society, and to take appropriate measures. Using modeling tools, trafc emission factors and local measurements to ponder them, Zarate and Clappier [Zar07] ran such a study in Bogota. They concluded that trafc is the major contributor to the plume of pollutants in Bogota and that strategies directed to mitigate air pollution might have contradictory effects depending on the pollutant to be tackled. They also recognized the importance of instrumenting a eet of cars with sensors in order to obtain a more precise assessment of trafc emissions over time and space for different types of vehicles, which was not possible at an affordable cost at the time of their study. Urban air pollution monitoring pursues a double goal. On one side,individual cars can have their gas emissions monitored. On the other side, instrumenting the main crossroads and trafc routes with pollutant sensors would allow for warnings to be issued when pollution levels reach critical thresholds, and possibly for the trafc to be monitored accordingly, either based on instantaneous measurements or longer trends (create one-way streets, modify trafc-lights periods, enforce limitation of trafc volume during peak periods, etc...). The requirements for an air pollution monitoring system include several aspects. The real-time data is crucial for decision-making processes, such as: issuing air quality alerts, interrupting or rerouting the trafc in certain areas, etc. The collection of extensive data over time and space is crucial, because it will enable a better understanding of urban air pollution and circulation behavior, and of the identication of critical periods and zones.
1.4.1
Potential Applications
Two different strategies can be adopted. One can instrument cars at the exhaust pipe, in order to assess their impact on the environment in a concrete context: a given city, a characteristic quality of fuel, typical trafc patterns and driving habits, etc. One would need to design a portable system making it possible to store on the vehicle the data regarding speed, RPMs and gas emissions. These data would have to be transmitted directly to a data management system for processing. A wireless technology seems particularly appropriate to facilitate this transfer. Known locations (such as domiciles or work places) could be equipped with data collection points (also called sinks, or base stations). The whole system would then apply a store-and-forward mechanism that allows vehicles to keep the data in memory until they opportunistically pass near a data collection point (e.g. at certain crossroads or gas stations). Public vehicles, taxis and private cars selected on a voluntary basis could be used. Findings are targeted at policy makers primarily. Another possibility is to instrument with sensors a grid of positions in the city - typically crossroads in order to see how the emissions evolve over time at different time scales. In this model, sensors record the ow data about particle emissions and transmit them over wireless to a data management server. An unprecedented wealth of data could then be used to rene and possibly adapt existing pollution models. The implications are again for policy making. The ability to use low-cost automated sensors in order to conduct the needed experiments is crucial. Such sensors exist today, even if at the moment not all pollutants can be detected. Kanjo et al. [KL07] reported in 2007 on an experiment on air pollution monitoring in Cambridge (U.K.) using lightweight sensors mounted on bikes, which could detect carbon monoxide, nitrogen oxide, and nitrogen dioxide.
10
1.5
Water Quality Monitoring
In developing countries, diarrhoeal diseases are one of the main causes of child mortality [Wor03]. This high death rate is directly correlated with the lack of sanitation and the consumption of polluted water. The United Nations Human Settlements Programme estimates that systematic and sustained tracking and review of progress of the sanitation system is important to develop policies and programmes at national and city-level targeted to improve services for the urban poor [UNH]. This primarily concerns biological contamination, mostly by fecal matters, because such contaminations lead rapidly to diarrheal diseases that present high mortality rates, especially among small children. The chemical pollution of water, however, is also a serious environmental hazard, which is harder to track but leads to death by cancer or other diseases several years after exposure, claiming probably hundreds of thousands lives every year [RBE+ 06]. Sensors may be used to sense water quality in the tanks and the distribution networks (as was done in 2007 in Boston for leakage detection [SNM07]). In the large cities of developing countries, these networks are often under-documented. Hence, a dense network of sensors placed in the pipes has the potential to help localize problems such as punctual and diffuse pollution sources or spills. It can also be used as an early warning system, in case of sudden degradation of water quality. In rural areas, water is also directly extracted from wells or streams: these water sources can also be instrumented. In this application, a major design issue is the development of inexpensive and reliable sensors for water quality. More specically, sensors monitoring biological contaminations -such as detection of faecal bacteria like E. Coli - are necessary. Currently, water analysis is usually performed in the laboratory where samples are cultivated to detect the growth of bacteria. This method is costly, takes time, and does not scale well. What is needed is the development of affordable automated sensors, that can monitor fresh water sources in situ, without manual sampling and laboratory analysis. Methods for this have been developed. E. Coli can be numbered in situ using gene-based remote detection technologies [LCF+ 05]. However, this method is not yet commercially available. Whereas, chemical water sensors exist for pH and a variety of contaminants [RBE+ 06]. The system requirements of water quality monitoring are similar to the previous applications. Time and space variability have to be dealt with, because contamination can occur at any time or anywhere in the distribution network. The real-time aspect is also critical, due to the importance of reacting rapidly in a case of water contamination. Finally, an automated collection is desirable, as pipes and wells are not easily accessible by human staff. As for potential applications, the general strategy would consist in instrumenting water distribution systems at critical points, for both bacterial and chemical risks.
1.6
Agriculture and Water in India: a Brief Historical Perspective
In the previous sections, we have identied environmental monitoring applications that are of particular interest for developing countries. Among them, agriculture seems an ideal candidate, because, on one hand it is related to well dened problems, and on the other hand the technology to solve them seems mature enough to be deployed in the eld. Now we focus on the special case of India by drawing a historical perspective that highlights the critical needs facing Indian agriculture today. Large scale irrigation has been used for thousands of years throughout Asia, leveraging essentially on two lines of development: at the community level by the design and implementation of local water management infrastructures such as dams and distribution networks, and through the action of emerging powerful states, referred to as the rst hydraulic societies [Wit57].
1.6. AGRICULTURE AND WATER IN INDIA: A BRIEF HISTORICAL PERSPECTIVE
11
At the local level, community irrigation systems were developed, essentially located in mountainous or hilly areas. Those systems were based on the intake from water streams (Himalayas) or on the construction of small tanks (India, Sri Lanka). They required community labor and management to gain access to and share water, and to minimize conicts. Later on, increased socioeconomic heterogeneity as well as the intervention of the state in the construction or maintenance of weirs has often weakened social cohesion and collective action. Nevertheless, the structures developed in the pre-colonial era serve even today a signicant portion of the total irrigated area [BM04]. In the Indus valley, a powerful agrarian society emerged as early as 3000 BC, based on large-scale and government-led irrigation works [Wit57]. This early model of an hydraulic society led to a massive economic development, because it provided a signicant increase in food supply, which permitted population growth, urbanization and development of alternative economic activities, such as trading and handicrafts. However, it was always constrained by the availability of the one critical resource it relied on: water. Eventually, its sustainability was threatened by an increase of environmental problems, such as salinity or water shortage, which surfaced due to the intensication of irrigated agriculture. In the words of Barker [BM04], it is worth noting that many of the ancient systems collapsed because societies could not manage environmental problems. Under colonial rule, the occupant had to meet two conicting goals: satisfying local market needs, so as to avoid unrest, and extracting as much surplus as possible. In semi-arid regions, dominant irrigation strategies were to develop protective irrigation for famine prevention in years of drought, which often resulted in suboptimal land productivity [JMW96]. To this day, most large-scale systems in the IndoGangetic Plain are protective irrigation systems , spreading the water thinly over a large area, regardless of the degree of scarcity experienced [BM04]. During the Cold War, the main concrete priority was to increase cereal grain production in order to attain food security. This triggered a Green Revolution (use of fertilizer, high-yielding varieties) but also a Blue Revolution (development and expansion of irrigation systems). The highest food-grain prices were reached in the 1960s and 1970s, the same period where about 80% of the dams existing today in Asia were built [McC96]. Since then, however, Asia suffered a sharp decline in investments and large dam construction: cereal grain prices were divided by two. At the same time, construction prices rose, as did the opposition of the environmentalists. During this period, following a top-down approach, public irrigation systems grew faster than the corresponding regulation bodies, resulting in a failure to build community ownership and to foster cooperative behavior at the local level [Hor98], [Jon95]. Foreign consultants promoted designs often inappropriate for the developing-country situation; typically, optimistic design assumptions produced insufcient ows in pipes and led users to destroy facilities . More recently, there has been a surge in private initiatives, as farmers, unsatised with government policies and projects started installing tube-wells or pumping from canals and drains. Such initiatives are hard to identify and quantify, as they are not ofcially acknowledged. Nevertheless, they enhanced the productivity of the public sectors investment in irrigation. To summarize the effects of the Cold War years, the positive impact of irrigation on poverty reduction and in enhancing rural livelihoods is felt through increased employment, lower food prices, and more stable outputs. There are also multiplier effects and indirect effects. However, an unclear denition of water rights and the unequal distribution of water yielding assets created inequality. Concurrently, large scale irrigation works (typically dams) and unregulated use of groundwater and canal pumping at the micro level represent new threats to the environment [BM04].
12
1.7
1.7.1
Indias Agriculture Today

Facts and Figures
Since the early 1970s India has achieved food self-sufciency. As reected by the growing per capita income (USD 450, with purchasing power parity - ppp - USD 2150), the rural poverty itself is decreasing: today, it reaches 33%, down from 56% in 1973-1974. Even if the agriculture part of GDP is relatively small at around 24%, its share of employment is about 67%. This agriculture remains largely rain-fed (60%). Nevertheless, the irrigation part is growing, as it is needed to sustain the general trend of a shift to a market economy. Although India is using only about 57 percent of its total water resource potential at present, the country is already using about 66 percent of its irrigation potential [oWR00]. Being a vast and monsoon-dependent country, India displays a wide variation across time and space for water resources availability. However, an average can be drawn for the effective water resource potential: 1122 bn m3 per year. The projections for water requirements are sharply increasing. From 644 bn m3 per year today, likely forecasts assess 784-850 bn m3 in 2025 [Sal04]. As a consequence, one witnesses an increasing supply-demand gap and a continuous decline in per capita water availability (in 1955: 5277 m3 , today: 1970 m3 [oWR00]). The canal irrigation sector is developed and managed by public agencies. Its importance for distributing water and recharging wells must not be underestimated. However, the inadequacy of the water institutions projects and policies, and inequality of water distribution has led to a ourishing of private initiatives, which are mostly centered on exploitation of groundwater. Groundwater irrigation is developed and managed by independent farmers, often illegally. It is estimated that 9.8 million electric and 4.4 diesel pump-sets are scattered across the country, as well as about 10 million dug-wells [Sal04].
1.7.2
A Recent and Growing Concern: Water Scarcity
In rural India, the new era of globalization is marked by a shift from a subsistence society to a market economy. This new economy is characterized by a new pressure on productivity, and by the shift to new, often water-demanding crops, such as cotton for instance. This situation is not exceptional in developing countries, where irrigation demands recently grew to consume well over 70 percent of the total developed water supplies [BM04]. Irrigation expansion has come to an end because developing more of the potentially utilizable water resources is costly. Raising ecological concerns has also led to the abandon of large scale irrigation projects. As a consequence, the attention has turned to the improvement in the management and performance of existing irrigation systems. Concomitantly, India (like the rest of Asia) is experiencing rapid growth in demand for water targeted at non-agricultural uses. The consequence of this global trend is easy to imagine. Within the rst quarter of this century, a projected 400 million Indians will live in regions that experience severe water scarcity [SAM+ 98]. This situation is made worse by the sharp increase in the use of groundwater as primary source of irrigation. Today it exceeds surface irrigation and threatens to alter the hydrology of the river basins, and to provoke irremediable environmental damage. While groundwater has contributed much to the growth in agricultural productivity, the over-exploitation of groundwater in the semi-arid regions is affecting both the quantity and quality of water available for agriculture, domestic use, and other purposes [BM04].
1.8. INDIA TODAY: THE CURRENT INSTITUTIONAL FRAMEWORK
13
The culprit is often the poor level of public irrigation services [Jon95], [Hor98], which led individual farmers to invest in the acquisition of pumps and to drill wells. These farmers may be less willing to participate in participatory irrigation schemes. But failure to maintain the surface irrigation systems can, in turn, affect groundwater recharge and increase the cost of pumping as groundwater tables fall. Groundwater exploitation increasingly leads to the drying up of wells and rivers, as well as the salinization of soils [SOF99]. Paradoxically, it can also lead to waterlogging in other areas, a state in which the subsoil water table is located at or near the surface. Excess water is accumulated in the root zone of the soil. If the land is cultivated this results in a reduced yield of crops commonly grown. Uncultivated land is limited in its use because of the high subsoil water table. Waterlogging can also lead to irreversible soil salinization. Groundwater depletion, on the other hand, is not a fatality. A 10 year study in Uttar Pradesh shows that surplus monsoon water can be used to recharge underground aquifers and simultaneously provide farmers with better crop security [SC02]. This highlights the potential that innovative practices can have in the area of water management.
1.7.3
The Specic Case of Karnataka
It is to be noted that the region of main interest for the COMMON-Sense Net project, i.e. the semi-arid area encompassing the main part of Karnataka and part of Andhra Pradesh has probably exhausted its potential for rain-fed agriculture. A study by Droogers et al. from the International water Management Institute [DSM01] comes to the conclusion that this area belongs to those absolutely needing more irrigation to meet the needs of their growing populations. As a consequence, there is a large potential for increasing food production through small-scale water harvesting systems that provide partial irrigation when water is most needed by the crops.
1.8
India Today: the Current Institutional Framework
During colonization, the British provided a highly centralized institutional framework. As a result most community-centered institutions and practices have lost their relevance and gradually disappeared. Unfortunately, most water institutions, which were developed in an era of water surplus, especially during the colonial period, are becoming increasingly ineffective in addressing water challenges as the country enters an era of water scarcity [Sal04]. Although the State and nationwide governance is often ineffective at tackling the problem globally, the situation in the villages shows local improvisations that try to cope with the new constraints faced by the population.
1.8.1
States and Central Government
The water legal framework can be broken down into three components: the law itself, the policy established by the government to achieve the intended goals, and the structure of the administration put in place to reach the target. There is no separate water law, but a profusion of water-related legal provisions that fail to reect the current conditions of water scarcity and water conicts. While States have jurisdiction over waterresources within their borders, the central government has the prerogative to resolve interstate disputes and to foster inter-state legal harmonization. However, it does not have the means to enforce these decisions because of the present constitutional division of power. The water rights themselves are illdened. Individual rights to both surface water and groundwater are recognized indirectly through land
14
rights, which accentuates discrimination and inefcient water use . There is a groundwater permit system, but it fails to set withdrawal limits [Sal04]. The water policy relates to the declared statements, as well as the intended approaches, of the central and state governments for water-resource planning, development, allocation, and management. Today, it recognizes the role of private sector participation. This marks a shift from water development to performance improvement. However, it failed to identify the necessary institutional mechanisms and to enforce its recommendations. As for water pricing policy, despite the traditional view of water as a public good, it is more and more acknowledged that if the method and level of water rates are such as to capture and convey the scarcity value of the resource, they can both induce efciency and ensure full cost recovery at the same time [Sal04]. The water administration framework is more effective. Although the state government has a dominant role, local governments such as municipalities and panchayats (village councils) unions also play an important role in the drinking water supply. Similarly, while irrigation departments have a larger role in the provision and management of irrigation, users and stakeholder groups are also encouraged to get involved in cost recovery and management at the outlet and system levels. As for regulatory arrangements, however, a very ineffective top-down approach is still in place. There exists little harmonization between states, and no credible enforcement mechanism at the top-level [Sal04].
1.8.2
Local Institutions
The situation at the local level is characterized by a sense of improvisation and self-organization. Admittedly, there are a variety of cooperatively-operated and community managed irrigation activities ranging from lift irrigation schemes in canal and groundwater areas, to water harvesting and sharing arrangements in arid and mountain areas. These can help developing a form of water rights in a canal region. This inheritance from the past, however, has been damaged by an inefcient centralization, which often resulted in a degradation of the proposed service [Sal04]. Private initiative, as we saw, widely took over. The expanding phenomenon of pump-set rentals is an indication of the existence of surplus pumping capacity, particularly in the case of diesel pump-sets. In view of the monopolistic or oligopolistic tendencies in these markets, not only are the water rates several times higher than the pumping cost but also the price and non-price discriminations remain pervasive. The root cause for the sub-optimality of these groundwater markets lies not so much in their economic and organizational aspects but in the legal and institutional vacuum within which they operate at present [Sal04].
1.9
Conclusion
In this chapter, we have sought to emphasize the role of the environment in the social and economic development of DCs. Agriculture itself represents a huge challenge in this area, as the outlook that we have given of Indian farming very well illustrates. Many DCs, India among others, suffer from increased water scarcity. This situation will only get worse because of demographic pressure, improved living standards and climatic changes. It has become urgent to investigate new ways of increasing productivity, especially as this sector is in crisis, not only because of the water issue, but because of institutional aws and infrastructure disrepair. Often, the institutional trend has been a progressive deterioration in the authority of operating agencies. One clear symptom of that trend is the destruction of system infrastructure by the farmers, because they have no means of verifying fairness of water delivery and there is no clarity or transparency in the operation.
1.9. CONCLUSION
15
A crucial point is on-the-spot environmental knowledge, which requires enhanced environmental monitoring. In this context, it seems appropriate to try and apply new information technologies in order to build a modern infrastructure that farmers can benet at the local level. A bottom-up approach has our preference. The crisis of institutions would make a top-down approach ineffectual. Although rural communities suffer from the obsolescence of once thriving commonly-maintained infrastructure, their capacity for improvisation is an asset in the search of applications. In the next chapter, we investigate the state of the art in environmental monitoring. In the subsequent ones, we will see how this can be transposed to the problems specic to developing countries. We will use rural India as a test-case throughout.
16
Chapter 2
New Opportunities for Environmental Monitoring and Agriculture

Before moving to the specic case of the rural areas of developing regions, let us examine the current status of environmental monitoring for agriculture, regardless of the context where it is used. Attempts at monitoring the cropping elds environment are as old as agriculture. The close observation of climate-related phenomenons and their effect on soil and crop alike are instrumental in the denition of any efcient cropping strategy, be it for crop selection, choice of sowing and harvesting windows, or irrigation schedules. More recently, the Green Revolution put to the use advanced fertilizing and irrigation techniques that benet from improved sensing technologies, especially for soil monitoring. Traditionally, however, a precise monitoring of the farming environment has always be seen as costly, for two reasons: The price of the sensing technology, and the human cost of manually collecting information from the sensors. More recently, automated sensors have started to appear in the cropping eld. Until the recent advent of wireless sensor networks, however, such systems were cumbersome to deploy and to maintain. In this chapter, we describe the new opportunities for agriculture represented by state-of-the-art sensors (Section 2.5) and wireless sensor networks (Section 2.3), before presenting concrete test-cases in the domain.
2.1
2.1.1
Usual Techniques
Stand-Alone Sensors
Inexpensive, portable and reliable sensors for air temperature, atmospheric pressure or humidity have existed for decades. More recently, stand-alone sensors with data-logging capability have been developed for a wealth of physical phenomena, which has fostered the development of the technique of manual insite reading, the most widely used today. Sometimes the sensors are connected to a data logger with the ability to store data that can be retrieved at a later stage. This technique, which is essentially cumbersome and work intensive, remains attractive in the developing countries, due to the low price of labor. However, it still limits signicantly the time resolution and the responsiveness of the system, and leaves a large part to human error. 17
18
CHAPTER 2. NEW OPPORTUNITIES FOR ENVIRONMENTAL MONITORING AND AGRICULTURE
2.1.2
Laboratory Analysis
Traditionally, precise assessment of soil status was made by manual sampling and laboratory analysis. Such a technique is still widely used today for the analysis of chemical and biological contaminants, and as a reference technique to assess the performance of in-situ sensors.
2.1.3
Remote Sensing
We refer to remote sensing as the use of imaging sensor technologies to assess physical parameters, in general aboard aircrafts and satellites. For instance, remote sensing can be used to derive drought conditions from electromagnetic radiation. It has been shown that even crop yields can be predicted 5 to 13 weeks prior to harvests using remote-sensing techniques [UK98]. The satellite used to assess soil moisture until recently was the AVHRR (Advanced Very High Resolution Radiometer), which has a time resolution of 10 days and a space resolution 10 km. Its recent successor is the Moderate-Resolution Imaging Spectrometer (MODIS), an advanced narrow bandwidth sensor, from which reectance data are made available at no cost every 8 days by NASA and USGS, through the Earth Resources Observation Systems (EROS) data center. Raw images are available on a daily basis, but their use involves considerable extra processing. MODIS spatial resolution is around 500m [TGS04]. Such a solution is minimally intrusive and allows for the monitoring of wide areas, including countries or continents. However, it only works for electromagnetic radiation, which limits it to phenomena affecting the atmosphere and the shallow layers of the soil (down to 10 cm at most). The deeper layers (the root zone) are beyond the reach of such a system. For this reason, and because in remote sensing the physical parameters are assessed indirectly through interpretation of the electro-magnetic spectrum obtaining complementary data from ground sensors is usually desirable. Another limitation of satellite sensing is that the frequency and delay of data depend on the satellites orbit. It is not suitable for a real-time application if the continuous monitoring of a parameter is needed, especially if the required space resolution is high.
2.1.4
Telemetry
Telemetry using cellular networks such as GSM is widely used today. It presents the advantage of wide and rapidly expanding coverage. The main issue to take into consideration for the use of such systems is the issue of cost. In particular, communication costs are prohibitive for messages sent several times per hour by a large number of sensors over a long period of time. Here a store-and-forward method can be used to send data in bulk, at the expense of responsiveness.
2.2
Wireless Sensor Networks
A wireless sensor is a self-powered computing unit usually containing a processing unit, a transceiver and both analog and digital interfaces, to which a variety of sensing units typically sampling physical data, such as temperature, humidity etc. can be adapted (see Figure 2.1 as an example). These sensors automatically organize themselves into an ad-hoc network, which means they do not need any preexisting infrastructure, as do cellular networks such as GSM. For this reason, we refer to such a network as an ad-hoc Wireless Sensor Network, which we denote as WSN throughout this document. The sensor nodes communicate with each other in order to exchange and process the information collected by their sensing units. In some cases, nodes can use other wireless sensors as relays, in which
2.3. DESIGN DIMENSIONS IN ENVIRONMENTAL MONITORING
19
Figure 2.1: Wireless sensor with 2 alkaline batteries, a connector to 2 soil moisture probes, and its weatherproof casing
case the network is said to be multi-hop. If nodes communicate only directly with each other or with a base station, the network is single-hop. In a data-collection model, sensors communicate with one or several base stations connected to a database and an application server that stores the data and performs extra data-processing. The result is available typically via a web-based interface 2.2. Recently, WSNs have raised considerable interest in the computing and communication systems research community. They have decisive advantages, compared with the technologies previously used to monitor environments via the collection of physical data. Whenever physical conditions change rapidly over space and time, WSNs allow for real-time processing at a minimal cost. Their capacity to organize spontaneously in a network makes them easy to deploy, expand and maintain, as well as resilient to the failure of individual measurement points. Wireless sensors are order of magnitudes cheaper than traditional weather stations connected to cellular networks. Although they remain expensive at the moment (a few hundred US dollars for a typical weather station [BIS+ 08]) because they have yet to evolve from laboratory prototypes to off-the-shelf products, most analysts rely on Moores Law to predict a price per unit of a few US dollars within 5 to 10 years for light-weight applications using cheap off-the-shelf sensors, such as temperature monitoring in buildings.
2.3
Design Dimensions in Environmental Monitoring
In any environmental monitoring application, many design dimensions need to be taken into account before choosing the appropriate technology to deploy. Based on the requirements of typical applications and our own experience, we propose the following dimensions to be the building blocks of any multicriteria decision for a system designer. Spatial scale: What is the size of the area to be instrumented? This can vary from single point if the phenomenon is to be observed at a single location, to local if the area spans a few hectares or square
20
Figure 2.2: Typical (Hybrid) Wireless Sensor Network Architecture
kilometers, to regional if an entire city or district must be instrumented, and even to global for larger areas (provinces, countries etc.). Time scale: How long must the phenomenon be observed? We can distinguish one-time phenomena, where a single measure is sufcient, short-term phenomena, lasting a few days or weeks, seasonal phenomena, lasting several months, and permanent phenomena, supposed to last an indenite time. Spatial variability: How many measurement points are necessary to model a given phenomenon over the monitored area? We can distinguish dense phenomena and sparse phenomena. Time variability: How fast does the phenomenon evolve over time? We can distinguish fast varying phenomena, which vary at a time scale in the order of the second or the minute, and slow varying phenomena, which remain constant several days, weeks or more. Responsiveness: What is the time period, within which the environmental information must be made available to the user? We can distinguish off-line systems, where data can be retrieved after an arbitrary long time and real-time systems, with stringent response-time requirements. Non-accessibility: Is the area to monitor remote or difcult to access? Non-Intrusiveness: Must the monitoring system be invisible and non-conicting with any activity happening in the monitored area? Deployment and Maintenance Costs What is the cost to deploy and maintain the system? This includes the hardware and software costs incurred by the system, and also the price of labor necessary to accomplish these tasks. Wireless sensor networks present attractive characteristics when the requirements of one or several of these dimensions are high. We detail these technical characteristics in the following sections.
2.4
Where Do WSNs Stand?
In this section, we describe the main technical characteristics and capabilities of WSNs with regard to the design dimensions that we introduced in the previous section.
2.4.1
Wireless Networking
Organizing the sensors into a network allows for real-time collection of a large number of measurement points at a minimum human cost and effort. As such it enables the monitoring of phenomena of high variability, both in time and space. It also makes it possible to retrieve data in real-time from locations
2.4. WHERE DO WSNS STAND?
21
that are difcult to access, either temporarily or permanently, thus addressing stringent responsiveness and accessibility requirements. The wireless component makes the monitoring system minimally intrusive in places where wires would disturb the normal operation of the environment to monitor. It reduces the installation costs signicantly, as it is estimated that typical wiring cost is US$ 130650 per meter and adopting wireless technology would eliminate 2080% of this cost [WZW04]. Because WSNs are constrained by size and power consumption, they use low-power radios. In order to achieve spatial scalability, most networks use multi-hop transmissions, meaning that each node can play the role of a relay between two or more communicating nodes.
2.4.2
Self-Organization
By self-organization, we mean that the nodes do not need any conguration in order to be operational once installed. In order to achieve this, each node is programmed to run a discovery of its neighborhood, specically to recognize which are the nodes that it can hear and talk to over its radio. In typical wireless networks, each sensing node is supposed to send back its information to a basestation, or sink, which is connected to a data management system. This is known as the data collection paradigm. In this model, nodes send data either periodically or as responses to events, typically thresholdbased alerts or user explicit queries (e.g. what is the current temperature at a particular location, what is the mean soil moisture content in a given area, etc.). In order to send data in an optimal manner to the base station, nodes organize themselves spontaneously into a data-collection tree, meaning that each node will choose a preferred parent, to which it will send its data. On the other hand, each node can have several children. Several metrics can be used to build the tree. Usually, the technique is based on a shortest path algorithm using the number of hops to the base station. The protocol typically runs as follows: The base station sends periodical beacons, which are selectively retransmitted by the nodes that receive them. Each beacon contains a hop-count, set to 0 by the base station and incremented by 1 at each retransmission. Upon reception of a beacon, a node stores its hop-count as its own if it is smaller than the one it currently has (initially the hop count is innite). A node retransmits a beacon only if its hop-count is smaller than its own. A typical example is the Collection Tree Protocol (CTP) dened within the TinyOS operating system for WSNs [FGJ+ ]. Using self-organization reduces notably the maintenance costs and allows for spatial scalability.
2.4.3
Efcient Power Management
Wireless Sensor Networks can be deployed in remote or difcultly accessible areas. As a result, they often cannot rely on the power grid for their energy source, and need to use either batteries or energy harvesting techniques such as solar panels [KYH+ 05]. In order to reach lifetimes of several months or years, WSNs not only use energy-efcient micro-controllers and radios, but implement power-efcient schemes at the Medium-Access layer. Thanks to that, they allow for long-lasting deployments in locations that are difcult to access or where minimum intrusion is required. 2.4.3.1 Medium Access Control
At the Medium Access Control layer, two large classes of protocols can be identied. Asynchronous schemes rely on the nodes coordinating their activity simply by listening to the medium, whereas in synchronized schemes, extra trafc is exchanged to keep the communicating devices in sync.
22
Asynchronous Schemes In this class of protocols, no synchronization is assumed between the nodes. In order to save energy, nodes go to sleep whenever they are inactive (which in a typical environmental application is most of the time). They only wake-up periodically to perform their task and to listen for possible incoming packets. As there is no synchronization, nodes that send packets need to send preambles that have a length corresponding to a duration at least equal to the period of sleep. This incurs extra energy for emitting the packet (100% of the energy to emit the preamble) and for receiving it (in average 50% of the preamble) but allows for duty-cycles in the order of 1%. In order to reduce the cost of overhearing, the preamble can repeat the address of destination. B-MAC is an example of such a scheme [PHC04], widely used in TinyOS [tos]. Synchronized Schemes In synchronous schemes, nodes use a TDMA1 -based transmission schedule. In other terms, communicating nodes send regularly synchronization beacons to prevent their internal clocks from drifting, to advertise which time-slots are available at a certain point in time, and to reserve free time-slots for their own use. This way, the periods of sleep are perfectly synchronized, which removes the necessity to send long preambles before the packets. In a multi-hop data collection environment, each node has to maintain two schedules: the schedule set by its parent, and the schedule it denes itself for its children. Dozer [BvRW07], designed and implemented on the Tinynode [tin] platform, uses such a scheme, achieves a duty cycle of 0,2% and claims a lifetime of up to 5 years for a typical data collection application. Asynchronous schemes are notoriously simpler to implement than synchronized schemes, but are less-efcient for applications requiring very low duty-cycles. In any case, the choice of an appropriate MAC scheme is highly application-dependent. 2.4.3.2 Mobility for Load Balancing
A main concern for multi-hop data collection networks is that the nodes closest to the base station, relaying all the trafc en-route to the sink, will deplete their batteries faster. Mobility is a possible answer to that question, either by rotating or moving the base station according to a carefully established schedule, as is done for instance in the Mobi-Route protocol [LH05], [LPP+ 06], or by using mobile agents to manually bring the sink to the nodes, as in the work on data mules by Shah et al. [SRJB03]. The data mule approach is related to the body of work on Delay-Tolerant Networks (DTN) [HF04], because nodes need to wait for the sink to come to them. In other words, they cannot send directly the data, they have to use a store-and-forward strategy. Such an approach is hence not adapted to applications with stringent response-time requirements.
2.4.4
Modularity
For widely distributed applications for instance agricultural plots , purely ad-hoc networks are an unsatisfying answer because of the limited range of their radio. This is an issue for: 1. minimizing the numbers of nodes to deploy by limiting the number of necessary relays, 2. ensuring Internet connectivity: a WSN cannot always be deployed close to an Internet access point, 3. addressing energy issues: servers usually need to be connected to the power grid because of their power consumption requirements.
1
Time Division Multiple Access
2.5. SENSORS AND AGRICULTURE
23
Hybrid solutions have been widely investigated [LLT03], [MLV03], [SM05], etc. In general, the answer to that problem revolves around the use of bridges (GSM, 802.11) between the WSN and external modules of the application (typically, a web-based management and data-processing system). This allows for the partitioning of the network into possibly far-away clusters, each connected to the central server through a bridge. Modularity allows for an increase in responsiveness, accessibility and space scale.
2.4.5
Web-based Data Management
With all the features described above, WSNs allow for the collection of an unprecedented wealth of data. The storage, mining and processing of these data then becomes an application-specic challenge in itself. One might argue that this part is not specic to WSNs, as other environmental systems have the same data display and processing requirements. However, the requirements for a data storage and processing system are particularly high for wireless sensing, because of the amount of data that are collected. As such web-based management can be considered an integral part of a WSN-based environmental monitoring system. Generic platforms exist for the collection and display of data in a geographical context. For example, Global Sensor Networks (GSN) [AHS06] and SenseWeb [NLZ07], both are web-based tools integrating Geographic Information Systems (GIS) features for the display of sensing data.
2.5
Sensors and Agriculture
Sensors have been used in precision agriculture for years. They are used in convergence with other technologies like the Global Positioning System (GPS), Geographic Information Systems (GIS), miniaturized computer components, automatic control, remote sensing, mobile computing and advanced information processing and telecommunications. In the following per-parameter state-of-the-art review, we focus on sensors that work in an automated fashion, come at a reasonable price, and can be easily adapted to an existing system.
2.5.1
Soil Moisture
The soil moisture (or soil water) content is dened as the quantity of water contained in the soil. This quantity is calculated as a volumetric or gravimetric ratio between water and the soil. The most commonly used metric is water content per volume , which links water volume Vw to the total volume Vt : = Vw /Vt (2.1) There are several methods to assess this ratio. Direct Methods consist in weighing explicitly the water contained in a portion of soil, and in deriving the gravimetric or volumetric relative water content. Indirect methods consist in assessing this content by computing soil characteristics that change as a function of soil moisture. 2.5.1.1 Direct Method: Gravimetric Method
This method is used as a reference to assess the efciency of other techniques. The experimenter takes soil samples, weighs them, dries them in an oven (105 o C), reweighs them, and infers mass humidity: w = (Mh Md )/Md (2.2)
24
where Mh is the weight humid and Md is the weight dry. The volume humidity can be derived: = w(d /w ) (2.3)
where d is the density of dry soil, and w is the density of water. If d is not known, it is possible to assess the volume with the Archimedes method. This method remains the most precise (provided that the drying is well executed). However, it has several drawbacks: it is destructive, very localized, labor-heavy and slow. And it is not adapted to in-situ measurements. 2.5.1.2 Indirect Methods
In-situ sensors are used to assess soil water content indirectly. A physical parameter correlated with soil moisture is assessed, and empirical formulas are used to convert the result into actual water content or hydric pressure. Nuclear methods In particular, the neutron probe, which uses the emission of neutrons that are slowed down by the hydrogen particles surrounding the probe. The slow neutrons are measured by a bore triuoride detector. This method remains the most precise in-situ measurement. However, it uses radioactive material. As a consequence, it is impossible to leave unattended in the eld. The equipment itself remains costly, at around USD 10000 apiece [NMDvI07]. It needs the establishment of a calibration curve. Dielectric methods Among all the constitutive elements of soil, water is the by far the one with the highest dielectric constant. As a consequence, there is a relationship between the dielectric constant of the soil and the water content . Topp and al. [TF02] proposed the following equation: = (0.043
3
5.5
292 530) 104
(2.4)
This relationship is supposed to be fairly independent of the composition of the soil, of the temperature and of the salt content. Among the dielectric methods, the Time Domain Reectometry Method uses the propagation time of an electric signal, which is a function of the dielectric constant of the medium where it travels. = ct L
2
(2.5)
Where c is the speed of light in the void, t is the measured time of propagation, and L is the length of the poles between which the signal travels. As of 2008, the cost of a commercial TDR probe was typically between USD 500 - 1000. The Capacitive Method, determines the dielectric constant by measuring the capacity of a capacitor made of electrodes and of the soil as dielectric medium. When a capacitor is introduced into wet soil, its capacity increases and so does the time necessary to charge it. Capacitive probes typically measure the time needed to charge the capacitor to a predened value, and infer the capacity from the measure.
2.5. SENSORS AND AGRICULTURE
25
Q(t)d (2.6) A Where A is the surface of the capacitor, V the predened voltage differential, t the time to reach this differential and Q an exponentially decreasing function. In Frequency Domain Reectometry, the capacitor is connected together with an oscillator to form an electrical circuit. Changes in soil moisture can be detected by changes in the circuit operating frequency. There is a wide variety of probes using the fequency domain reectometry method, and they usually come at an affordable price [sms]. As an example, the ECH2O sensors are available at a price of around USD 80. Their simple mode of operation: the input is a short pulse of DC, the output a voltage, make them an attractive option. The experimental results via dataloggers [NMDvI07] or wireless sensors [Rou08] are encouraging. The observed margin of error was around 4% [Rou08]. The Resistivity Method, uses a resistor embedded in a semi-porous material. The resulting probe is buried in the soil, where the semi-porous material comes little by little in a water-potential equilibrium with the surrounding soil. Gypsum blocks and watermarkT M sensors are examples of such probes, which cost between 5 and 30 USD. Watermarks typically have a 10% error margin, although the rate of faulty probes is reported on web-forums to be quite high (up to 30%) V =
2.5.2
Soil Salinity and PH
In the current state of the technology, it is not possible to assess separately in-situ the different chemical components of the soil, such as nitrates, phosphates or potassium [Phi08], which at the proper concentration are important nutrients for the plant. Instead, salinity and pH are measured, using similar electrical methods as described in the previous section about soil moisture. Salinity indicates the total concentration of soluble elements. PH is a measure of the acidity or alkalinity of the medium and as such indicates the availability of nutrients to plants. The shortcoming of these methods is their inability to discriminate between the components, and to know their respective concentrations. Moreover, it is impossible to ignore elements of little interest, such as carbonates, sulfates or chlorides. Soil salinity and pH in-situ sensors exist on the market today [Mat04], [SCH07] [OI67]. To the best of our knowledge, there are no commercially available in-situ soil nutrient sensors. However, some promising technologies are under development, and commercial products are likely to appear in the years to come. In particular, the use of electrochemical sensors yielded promising results in the recent past [LNTN+ 07].
2.5.3
Climatic Variables
We restrict ourselves to the usual components of a weather station, and we focus our attention on inexpensive sensors that can be easily adapted to analogous or digital channels of a custom data logger (or a wireless sensor). The sensors we took into consideration take a simple AC or DC current input, and provide either a digital or analogous (voltage) response. In fact, all the sensors described below have been tested in the context of a wireless sensor network deployment [SDF]. Air Temperature and Humidity Many commercially available sensors exist for this. We have had experience with the SHT75 manufactured by Sensirion. The SHT75 is a digital temperature and humidity sensor2 that has a relative error of 0.3 C in temperature, and of 2% in humidity.
2
Its cost in 2008 was around USD 30 apiece
26
Wind velocity and direction Many commercially available sensors exist. Among the most popular products in the scientic community, Davis proposes an anemometer3 assessing wind direction and speed at an accuracy of 5% in speed and 7 in direction. Precipitations Here again, there are many rain gauges to choose from on the market. The Davis Rain Collector4 is advertised as a high-accuracy tipping bucket. Experimentally, its performance was assessed at 10% [SDF].
2.6
WSNs in Agriculture
Wireless sensor networks, a light-weight sensing and communication system necessitating little, if any, network conguration and maintenance, are entering their maturity phase as this thesis is being written. There are no examples of commercial applications to date. However, in recent years a number of investigations have been conducted by scientists in realistic agricultural settings.
2.6.1
Vineyard Temperature Monitoring
In 2004, Beckwith, Burrell et al. [BBB04], [BTB04] reported on the use of sensor networks for integrated management of a vineyard. Using an ethnographic approach, they rst assessed the needs of vineyard managers, before designing and deploying a system in the eld. Their work was motivated by the primary importance of temperature in the development of grapes to ensure wine quality. Because of the costs of environment monitoring, most vineyard owners have so far used single sensors for this purpose. However, at the vineyard scale, the climate is a microclimate. Field work was conducted by Beckwith et al.[BTB04]. 48 nodes were deployed over a period of more than 6 months in an Oregon vineyard, reporting temperature every ve minutes. The results were logged in a centralized way and could be displayed on a map and retrieved on a per-sensor basis. Moreover, alarms were sent when the temperature decreased below 0, indicating a risk of frost. The history of the temperature variations throughout a cropping season are especially critical. Variation in fruit maturity within a management block (i.e., intrablock heterogeneity) is a well-known phenomenon and vineyard owners adapt their harvesting strategy accordingly. Organization Intel Research Application Vineyard ne-grained temperature monitoring Deployment duration 6 months Distance between nodes 15 to 25 meters Size of the network 48 nodes. 16 backbone nodes, and 3 leaf nodes for each of them Average lifetime 6 weeks for the backbone nodes, which were running at a 20% life-cycle. Networking platform Mica2 [xbo] Sensing platform Sensirion SHT Performance 77 % packet delivery rate
3 4
Cost: around USD 120 (2008) Cost: around USD 75 (2008)
2.6. WSNS IN AGRICULTURE
27
2.6.2
Potato Disease Prevention
In 2005, Baggio [Bag05] presented the initial results of the Lofar project in the monitoring of microclimates in a crop eld. The deployed WSN monitored humidity and temperature in order to better ght phytophtora in a potato eld. In Baggios words, phytophtora is a fungal disease which can enter a eld through a variety of sources. The development and associated attack of the crop depends strongly on the climatological conditions within the eld. Humidity is an important factor in the development of the disease. Both temperature and whether or not the leaves are wet are also important indicators. The goal of the WSN deployment is to predict the emergence of the disease and to schedule fongicide treatment only when needed. The authors only reported on the pilot study, however. The full-size network has not been deployed yet. Application Lofar Application Fungus management in a potato eld Deployment duration 12 months Distance between nodes 1530 meters Size of the network 150 nodes (pilot study 12) Average lifetime Not disclosed Networking platform Mica2 [xbo] Sensing platform Sensirion SHT Performance Not disclosed
2.6.3
Tomato Disease Prevention
Mancuso and Bustaffa deployed in 2006 a small wireless sensor network in order to monitor the emergence of certain diseases in a greenhouse: gray mould, leaf mould and powdery mildew. The goal of the deployment was to explore the potential of WSNs for the control and maintenance of temperature, relative humidity and CO2 concentrations within optimal limits. Application Rinnovando S.R.L, Nemo S.R.L Application Tomato disease monitoring Deployment duration Unknown Distance between nodes 6.512.5 meters Size of the network 7 nodes Average lifetime Not calculated (1 year claimed) Networking platform Sensicast Sensing platform Sensirion SHT71 (air temperature and relative humidity), Picotech PT100 platinum resistance thermometer (soil temperature) Performance Not disclosed
28
2.6.4
Cattle Monitoring
Recently, two projects have addressed the possibility of using WSNs to monitor cattle in a farm [BHSA+ 07], [RW06]. Radenkovic and Wietrzyk [RW06] explore the potential of wireless sensor networks for nationwide cattle monitoring systems. Each wireless sensor acts as an extended RFID collar storing the identity and health status of the animal, which can be tracked at different locations, such as pasture or farm buildings. Each location is equipped with a base station opportunistically recording the information from the collars as the animals come into its range. The system was so far evaluated through extensive rounds of simulations. Bishop-Hurley et al. [BHSA+ 07], tested in-situ the responsiveness of cattle to electrical and audio stimuli designed to modify their behavior and prevent them from crossing a line in an experimental alley. Cattle were equipped with collars containing a GPS receiver for positioning and a wireless transceiver similar to a wireless sensor. The sensed data here is the positioning of the animal. Each collar communicated to a base station connected to a server responsible to analyze the received signals and to generate the appropriate cues. The goal was to design a virtual fencing application replacing expensive wired fences in extensive grazing systems. Organization CSIRO Application Virtual fencing Deployment duration punctual experiments Distance between nodes 0-40 meters Size of the network 25 nodes Average lifetime Not calculated Networking platform Proprietary Sensing platform GPS receiver Performance Not disclosed
2.6.5
Paddy Field Monitoring
Hirafuji et al. developed the concept of Field-Monitoring Server, a Wi-Fi based wireless sensing platform that was applied in settings as various as Earth observation, urban image monitoring and agriculture [HNK+ 07]. They report on the deployment of a network of 5 nodes in paddy elds. Taking an agnostic approach to the current view on WSNs, they contend that agricultural monitoring systems need enhanced capabilities, such as wireless broadband communication and high-resolution image-monitoring technology. In their words, specic data such as images of emerging rice blast are indispensable to revise the prediction system. However, concrete results on how to process this information and for what benet have not been published yet. Organization Japanese National Agricultural Research Center Application Paddy elds Monitoring Deployment duration Unknown Distance between nodes < 100m
2.6. WSNS IN AGRICULTURE
29
Status Architecture
Bridge Size (number of nodes) Radio range Data sensed Energy source Node life-time
Vineyard temperature Full deployment (6 months) Multihop WSN (backbone and leaves) None 48 15 30 m Air temperature Batteries ~ 6 weeks (battery failure)
Potato disease Tomato disease Cattle monitoring Paddy monitoring Pre-deployment (2 months) 2-Tier Network
Clustered WSN Server bridge
Pre-deployment Multihop WSN
Prototyping Single-hop WSN
Prototyping Mesh Network
None 12 (150 expected) 15 30 m Air temperature Air humidity Unknown Unknown
None 7 6.5 12.5 m Air and soil temperature Battery Unknown (1 year claimed)
None 25 0 40 m Position (GPS) Battery Unknown
Wi-Fi 5 10 100 m Unknown Solar energy Unknown
Figure 2.3: Summary of WSN agriculture-related projects
Size of the network 5 nodes Average lifetime Unknown (solar energy used) Networking platform Wi-Fi Sensing platform Unknown Performance Unknown
2.6.6
Discussion
Precision agriculture has been using state-of-the-art sensors for decades. However, the possibilities offered by environmental monitoring were limited due to the infrastructure and the labor costs it incurred. In this section, we have presented four typical projects in the area of wireless sensor networks for agriculture. The projects are summarized in Fig. 2.3. Although it is generally admitted that ne-grained environmental monitoring holds great promise for agricultural sciences, related projects are still few in the scientic literature. A possible explanation is that the wireless sensor networking is just reaching its maturity phase. Seminal works on WSNs in environmental monitoring in general [SPMC04], [BCI+ 08], [ARE05] have nally demonstrated the feasibility of deploying and maintaining such networks for periods of time in the order of a few months. Such endeavors were a prerequisite to the collection and analysis of the amount of data necessary to develop useful applications for agriculture. Another observation is that most of the projects were focused on few and simple data measurements, usually air temperature and humidity. A possible explanation is that such sensors come as standards on most platforms, which makes them easy to use. Whereas, the type of soil moisture probes simple and inexpensive enough to be adapted on wireless sensors necessitated until recently the design of special data acquisition boards, and in most cases needed to be properly validated in this new (wireless) context of use. A closer look at the individual projects leads to the following observations. Firstly, event-detection emerges as a strong theme in the envisioned applications. Two of the projects deal with early detection of diseases, one with prediction of frost, and one with virtual fencing (i.e., redirecting cattle when they risk to leave their grazing area). In all these applications, the capacity of networked sensors to report events in real-time is leveraged. In two occurrences, vineyard and paddy eld monitoring, continuous
30
monitoring is also used to adapt farming strategies, in the short or long term. In all cases, spatial and time ne-grained resolutions are perceived as a critical improvement. Secondly, most of the networks focus on sensing rather than actuating. The goal is to provide the user with enhanced information that lets him take his or her own decision. Only for cattle monitoring is the sensor coupled with an actuator, namely an audio or electrical stimulus. As for a power source, batteries are used in most cases, for different reasons. In vineyard monitoring, the constraint was to deploy light-weight sensing nodes on the vines themselves. Solar panels would be difcult to adapt in this situation, because of their size and the effect of vegetation on solar energy collection over time. Similar concerns were probably considered by Bishop-Hurley et al., as solar panels would be problematic to install on cows collars. The tomato disease prediction application is aimed at greenhouses, where solar energy is not directly available. Finally, the size of the networks remains small, in the order of a few tens of nodes in the largest case. This indicates both the investigative nature of the experiments, which are primarily aimed at research rather than production, and the scalability challenges raised by WSNs to this day. In particular, Beckwith et al. [BTB04] acknowledge resorting to a planned network conguration rather than a self-organizing one, for deployment facilitation. Such an approach would not scale to large networks. To summarize briey, we can consider these projects as proofs of concepts, whose transposition to commercial products will be the measure of success in years to come. This situation is likely to change, as proper commercial tools are soon going to be in the hands of agricultural scientists. Tim Wark et al. deployed in 2007 a network of 16 of such nodes equipped with ECH2O soil moisture probes in order to observe the effects of irrigation on agricultural plots [WCS+ 07]. The results of the experiment, however, have yet to be disclosed. Chapters 5, 6 and 7 of the present document relate our own efforts in the development of such a toolkit. But for the time being, we set our sights on Developing Countries again, this time scrutinizing the use of WSNs in this context, and drawing lessons from existing deployments (see Chapter 3) before proposing our own problem statement and design (Chapter 4).
Chapter 3
WSNs and Developing Countries

As we mention in the introduction, typical applications of WSNs include home automation, forest re prevention or monitoring of industrial processes. We have also seen that some pilot projects exist in the area of agriculture as well. Such applications are often targeted at and tailored for industrialized countries. But researchers have also tried to apply WSNs to issues that concern the developing countries. As of today, such projects remain few, and their effect on development issues remains to be proven. This chapter explores existing initiatives and, from them, draws some design guidelines, identifying both opportunities and challenges brought by these systems.
3.1
Existing WSN Projects
In this section, we outline the characteristics of several WSN projects in developing countries. We briey describe their aim and focus, and we summarize their technical characteristics. These projects are still few, and sometimes do not go beyond the design and simulation stage. However, a current trend can be observed, consisting in attempts to apply state-of-the-art wireless sensing research to developmentrelated problems.
3.1.1
Groundwater Arsenic Contamination Assessment in Bangladesh
The CENS unit of UCLA is involved in the deployment of a WSN in Bangladesh, with the goal of better understanding the presence of arsenic in groundwater [RBE+ 06]. Every day, the Bangladeshi population living in the Ganges Delta consumes water that is contaminated with arsenic. This situation, which concerns millions of people, represents a major humanitarian disaster in the making. If nothing is done, it is estimated that hundreds of thousands of people every year will be affected by diseases such as arsenicosis and skin cancer, and that the incidence of death by cancer will be approximately 3000 cases per year. Say Ramanathan et al: A full understanding of the factors controlling arsenic mobilization to ground water is lacking. A current working hypothesis in some regions is that the inux of dissolved arsenic to ground water is greatly enhanced where irrigation for rice cultivation provides the primary source of aquifer recharge. In order to verify this hypothesis, CENS, in collaboration with the Bangladesh University of Engineering and Technology and MIT, deployed a sensor network in a rice eld in the area of Dhaka. The authors justify the usage of a network by the heterogeneity of soil, which requires dense spatial sampling. In the case of arsenic contamination, daily or longer period variations in the concentrations of 31
32
CHAPTER 3. WSNS AND DEVELOPING COUNTRIES
pollutants were of primary interest in order to understand the underlying mechanisms of contamination. In particular, they highlight the paradox between the criticality of water quality concerns and the fact that analysis is still primarily conducted in a laborious manner by physical collection of a sample that is analyzed back in a laboratory. Ramanathan et al. also introduced the concept of a wireless sensor as a shared resource, meaning that several users should be able to benet from one single sensor, by reusing it over time and space. In this model, the sensor itself becomes a possibly mobile resource, so that the necessary resources can be minimized. Sensor sharing encompasses two contexts of use: moving a smaller number of sensors around in a deployment to emulate density, and gradually removing redundant sensors from a deployment to go from dense to sparse deployments. In early 2006, the authors deployed three pylons containing 3 complete suites of sensors (soil moisture, temperature, carbonate, calcium, nitrate, chloride, oxidation- reduction potential, ammonium, and pH), each deployed at a different depth (1, 1.5, and 2 meters below ground). Water depth was monitored through a pressure transducer at the base. In total, they deployed 48 sensors in the eld for a period of 10 days. The authors claim that even such a limited experiment allowed for the observation of daily trends in several redox active geochemical parameters [RLL+ 07]. Organization: University of California, Los Angeles Application: groundwater arsenic contamination assessment Data monitored: soil moisture, temperature, carbonate, calcium, nitrate, chloride, oxidation- reduction potential, ammonium, pH and arsenic Deployment duration: 10 days Distance between nodes: unknown Size of the network: 3 pylons, 48 sensors Average lifetime: 10 days Networking platform: Crossbow Mica2, 802.11b access point, GPRS bridge Sensing platform: unknown Results obtained so far: observation of daily cycles in geochemical parameters WSN added value: space and time variability of soil contaminants
3.1.2
SenSlide, A Sensor Network Based Landslide Prediction System
Sheth et al. [STM+ 05, STM+ 07] designed, simulated, and built a laboratory test-bed of a landslide prediction system via wireless sensor networks using 2-axis strain gauges to predict landslides. The usual approaches to detect rocky landslides involve drilling 20 30 meter holes into the surface. Both sensors and installation are costly, making it difcult to proceed to wide deployments. In contrast, SenSlides approach to measure slope stability is to combine observations from a large number of distributed inexpensive wireless sensors connected to one-axis strain gauges. The strategy is to observe the cause of the landslide, specically the increasing strain in the rocks. A Bayesian statistical approach is used to link the data of a sensor patch to the imminence of a landslide. For this approach to be effective, geologists assess the optimal sensor-separation to be 30 - 40 meters. In this model, data needs to be sampled periodically to help earth scientists collect trend information over time. But the lifetime
3.1. EXISTING WSN PROJECTS
33
of the network must not be adversely affected by frequent sampling. As a direct consequence, SenSlide shares features of a rare-event detection network, as well as a very low sampling-rate data-collection network. The primary objective of this work is to provide a distributed sensor system that is robust against failures. Redundancy is used throughout the network architecture. The point measurements made by individual sensors are propagated to a set of base stations that are connected to GPRS and/or 802.11 bridges. A laboratory testbed of 65 sensor nodes was used, as well as simulation results for larger systems up to 400 sensor nodes. A prototype of the system should be deployed in-situ during an upcoming monsoon season. Organization: Microsoft Research, University of Colorado, Boulder, Indian Institute of Technology, Mumbai Application: landslide prediction Data monitored: strains in rocks Deployment duration: not deployed yet Distance between nodes: 30-40 meters (projected) Size of the network: patches of 600 nodes (65 nodes deployed in the laboratory so far) Average lifetime: 4 months (projected) Networking platform: Telosb from moteiv [tel] Sensing platform: single-axis strain gauge (manufacturer not specied) Results obtained so far: simulation of the operation of a full-scale network, laboratory tests run on 65 nodes WSN added value: diversity provided by the dense deployment of inexpensive sensors
3.1.3
Wireless Sensor Network for Water Quality Management
In 2007, Zennaro et al. undertook a project on water quality management in Malawi [ZYFP07]. The overall goal of this project is to develop an infrastructure that will be used to measure water quality using Wireless Sensor Networks. The project focuses both on the design, implementation and deployment of an innovative wireless sensing application, and on the dissemination of results. The network is envisioned to be single-hop, with a periodic data collection paradigm. Samples are going to be collected once per day. Because the network is to be deployed in a remote region without easy access to technology and expertise, automated fault recovery is a major theme in the system design. Throughout 2007, the project completed the initial system design and implementation phase. All the probes used in the network were tested, validated and calibrated. The next phase consists in the deployment of the system in the catchment area of the Mudi Dam, which spans about 1 km2 . Organization: Royal Institute of Technology, Stockholm Application: water quality monitoring Data monitored: pH, water reduction/oxidation (redox) and turbidity
34
Deployment duration: not deployed yet Distance between nodes: 200300 meters (with relays in between) Size of the network: 4 nodes (rst phase, planned in 2009) Average lifetime: upcoming (solar powered) Networking platform: Sun SPOT [Sun] Sensing platform: Ionode for the pH and redox sensors [ion], OBS-3+ for turbidity sensor [DAI] Results obtained so far: software and hardware design and implementation, sensors calibration and testing WSN added value: automated data collection for continuous monitoring. POssibility to raise alarms automatically.
3.1.4
Flood Detection System for Honduras
Basha and Rus [BR07] began from the realization that natural disasters have aggravated effects in developing countries, because of the lack of infrastructure and the absence of well established procedures and responsibilities. They derive the necessity of developing integrated warning and evacuation systems that take into account the specic technological, social and political constraints of this context. Say Basha and Rus: The complexity of these systems and the need for autonomy within the context of a developing country - while remaining maintainable and accessible by nontechnical personnel - provides a challenge not often solved within developed countries, much less the developing. The authors focus their attention on the design and implementation of a ood detection for Honduras, a country repeatedly hit by heavy rainfalls and devastating hurricanes in recent years. The Aguan River basin was chosen as deployment site, as it constitutes a particularly exposed area. Taking a holistic approach, Basha and Rus address the problem by subdividing the necessary actions in four tasks: event prediction, authority notication, community alert, and community evacuation. From the technology point-of-view the main issues they address are: protection of the system from environmental and human damages, appropriate coverage of the area at risk, effective prediction and electricity supply. For the authors, prediction entails a model of the physical system, an understanding of the relevant variables that this model requires as input and output, physical measurements of these variables, communication of this data to the computation locations, and a computational system to run the variables through the model. As for the physical variables, they decided to focus on water pressure sensors to monitor the evolution of river level. As for the model, the authors contend that statistical computerized methods extend the prediction time to up to 48 hours. For prediction to be effective, monitoring has to run continuously during critical periods at critical points, hence justifying the use of a network of sensors reporting autonomously to a centralized processing unit. As for coverage issues, the authors designed a two-tier architecture. 8 km radius clusters are organized as local single-hop networks in the 900MHz band. Inter-cluster communication necessitates a radio-range of up to 25 km, for which analog 144MHz radios with a custom modem were used. Among the lessons learned from the project, the authors emphasize the importance of creating local partnerships and of relying on local knowledge. In their case, they created a partnership with a Honduran non-governmental agency. Another important point is the questioning of the usual approach of developing and testing the system in the lab before deploying it in the eld. The authors claim that such a strategy is likely to lead to a failure of the end-product. They opted instead for a hybrid approach of trial
3.1. EXISTING WSN PROJECTS
35
and error, where the system was partially tested in the lab, then deployed in the eld as it was, before going back to the lab to address the specic issues discovered in-situ. Organization: Massachusetts Institute of Technology Application: ood monitoring Data monitored: water pressure (river level) Deployment duration: not deployed yet (preliminary eld tests conducted) Distance between nodes: up to 8 km in a cluster. Up to 25 km between clusters Size of the network: Average lifetime: unknown (solar panels) Networking platform: proprietary Sensing platform: unknown Results obtained so far: design and early deployment tests WSN added value: autonomous, distributed data collection
3.1.5
Road Surface Condition Monitoring
De Zoysa et al. [ZKSS07] started from the realization that in Sri Lanka one of the main reasons for the deteriorated condition of the road system is the lack of continuous monitoring of the surface condition. They designed BusNet, a delay-tolerant sensor network for monitoring the road surface condition, which is mounted on buses belonging to the public transport network. This way, the sensor-equipped buses use the very roads that the authors want to monitor. Road surface conditions are assessed through the use of accelerometers. The data are stored while the bus is on the road, and transferred once a day to a collection point at the bus Main Station. A delay-tolerant solution is appropriate in this case, because the data gathered on road surface condition are not needed in real time. The critical point for the usability of this system is the ability to translate the acceleration data into actual information about the road condition. The authors are currently developing an analytical model for this. Precise positioning of the sensors over time is also mandatory. The authors plan to adapt GPS receivers to the wireless sensors. Lifetime is not a serious issue in this case, since the sensors can be powered by the bus battery, and maintenance can be performed regularly when the bus is at the Main Station. Organization: University of Colombo, Sri Lanka Application: road surface monitoring Data monitored: vertical and horizontal acceleration Deployment duration: not deployed yet (preliminary eld tests conducted) Distance between nodes: variable. Nodes act as data mules and bring back data to a centralized collection point. Size of the network: prototype phase Average lifetime: unknown (Buses provide power) Networking platform: proprietary
36
Sensing platform: MicaZ Results obtained so far: design and early deployment tests WSN added value: autonomous, distributed data collection
3.1.6
Other Work
InteleSense [Int] is deploying water and weather sensors and to integrate them with public health data to research links for water-borne illness, both in Vietnam and Ethiopia. Johnson and Margalho studied wireless transmissions in the Brazilian Amazon, with the aim to deploy an environmental monitoring system at a later stage [JM02]. Their work focused on simulation studies, using parameters collected from the environment. They identied a link between signal attenuation and climatic conditions. Dargie and Zimmerling investigated the scope and usefulness of specic WSN application domains in the context of developing countries [DZ07]. They concluded that wireless sensor networks have the potential of aiding developing countries to carefully utilise scarce resources, to protect and maintain infrastructures, and to prevent undesirable occurrences. Accordingly, they advocate their use for environment, infrastructure and habitat monitoring, agricultural management and disaster prevention. However, they do not propose any design or design guidelines at this stage. Similar high level work has been conducted by others (e.g. [PHL07]).
3.2
A New Tool for Developing Regions?
From the perspective of the specicities presented by Developing Countries, there are general lessons to be drawn from the projects described in the previous section (see Fig. 3.1 for a summary). The challenges and opportunities presented by WSNs in this particular context are deeply inuenced by its geographical, social, economical and cultural features. Most of them are highlighted by the projects described in the previous section.
3.2.1
Assets
The potential that distributed, wireless solutions present for environmental monitoring was highlighted by Estrin et al. as early as 2001 [EGPS01]. Among the expected benets, we can list: 1. Easiness of deployment and maintenance 2. Flexibility of the solution 3. Automation of data collection 4. Unattended operation 5. High space- and time-resolution at a low cost 3.2.1.1 Deployment and Maintenance
A recurring intended benet of WSNs is their independence from any preexisting infrastructure. The nodes themselves are the network, and can be added, moved or removed in a seamless fashion. The possibility of instrumenting phenomena continuously for extended periods of time is often mentioned. Being specically designed at the hardware and software level for low power operation, WSNs have a
3.2. A NEW TOOL FOR DEVELOPING REGIONS?
37
Arsenic detection Location Status Bangladesh Full deployment
Landslide prediction India Simulation Laboratory deployment 2-Tier Network

Multi-cluster WSN Server bridge
Water quality management Malawi Laboratory testing
Flood detection Honduras Laboratory testing Field pre-deployment 2-Tier Network
Road Monitoring Sri Lanka Laboratory testing
Architecture
3-Tier Network
Multi-cluster WSN Inter-cluster access point Server bridge
Multihop WSN
Unknown 4 nodes (projected) 200300 m (projected) pH Redox Turbidity
Multicluster WSN Server bridge
Delay-Tolerant Network
Bridge Size Radio range
Wi-Fi (802.11b) GPRS 48 sensors 3 access points < 100m
Wi-Fi (802.11b) GPRS 65 (deployed) 600 (projected) 30 40 m (projected) Rock strains
Proprietary (144MHz) Unknown 8 km (in a clsuter) 25 km (between clusters) Water pressure Unknown 30 40 m
Data sensed
Energy source Node life-time
Soil moisture Temperature Carbonate Calcium Nitrate Chloride Ammonium pH Arsenic Solar panel 10 days (deployment duration)
Acceleration
Battery 3 4 months (projected)
Solar panel Unknown
Solar panel Unknown
Car battery Unknown
Figure 3.1: Summary of WSN projects in developing countries
longer autonomy than cellular-phone networks used for telemetry, which also incur extra communication costs. WSNs are also supposed to be conguration and maintenance free, the nodes being able to organize spontaneously into a network and to work unattended until failure or energy depletion. The easiness of deployment and operation is repeatedly mentioned as a major argument by the designers of existing projects [RBE+ 06], [STM+ 07], [ZYFP07], [BR07]. The argument is that, in developing countries, the existence of a reliable infrastructure is often questionable. On the front of telecommunications, the traditional land-line infrastructure is usually old, sometimes totally obsolete. Cellular coverage is still partial at best, although this situation is quickly changing. In India, for instance, GSM operators claim to have coverage in 100,000 villages out of a total of 650,000, and gures suggest almost tripling the rural coverage by 2010 [CTT06]. However, whether elds outside of villages will be satisfactorily covered remains uncertain. The electricity supply infrastructure itself can often not be taken as granted. The example of India, once again, is telling, with daily power cuts lasting several hours being the rule in most rural areas. 3.2.1.2 Flexibility
An environmental monitoring system often does not operate autonomously. It communicates with data storage and management systems, with alert dissemination infrastructure, etc. Pure ad-hoc wireless sensor networks cannot provide global connectivity. For this, an access point to the traditional public switched telecom network (PSTN) or to the Internet is needed. Moreover, even local connectivity is a challenge for resource limited devices such as wireless sensors. Typically, their radio range will vary from 50 to a few hundreds meters in outdoor conditions, depending on the platform. This means that for
38
applications necessitating sparse networks with a large coverage, hybrid solutions need to be provided, as described in Chapter 2. Fortunately, their exibility allows wireless sensor networks to act as a module in a larger system, a capacity leveraged by several projects [RLL+ 07], [STM+ 07], [BR07]. 3.2.1.3 Unattended Operation
Human infrastructure is a possible source of concern [BR07], or, more precisely, the level of technical expertise that can be expected in rural areas of developing regions. Even with the increased exibility of the technology, deploying a wireless sensor network remains to this day a specialists task. This is generally taken for granted by the existing projects. However, once the network is installed, engineers will not be available through its deployment life. If WSNs manage to fulll their promise of maintenance free operation (which remains to be proven for long-term deployments, see [BTB04], [SPMC04]), they will emerge as a primary option for environmental monitoring in developing countries. 3.2.1.4 Real-time Response
The capacity to collect and process data automatically on the ight is especially attractive for alert-based systems. In this case, the user needs to be warned as soon as a monitored environment threatens to drift from its desirable state [STM+ 07], [ZYFP07], [BR07]. Direct human intervention is today the primary resort for environment monitoring in developing countries. However, such monitoring is usually ill-adapted to fast response-time requirements. Basha and Rus pinpoint typical obstacles that would make such an approach unreliable in the case of a ood detection system: The geographic area involved hinders any form of volunteer-based system. The communities upstream that would need to perform the measurements and/or the communication of those measurements have almost no connection to the communities affected by the ooding. This removes any level of self-interest and peer-pressure in voluntarily performing any system tasks. Second, the nature of the problem involves measuring the river and surrounding area during heavy rains, hurricanes, and at all times of day and night. Very few volunteers would stand outside in a hurricane to perform a measurement or radio information to a central ofce. Neither would they perform these tasks in the middle of the night. Although paying someone may allow for night-time measurements, few people would remain during a hurricane, especially if that hurricane affects their own community with small-scale ooding, building leakages, or potential agriculture crop damage. Yet it is at these times that the measurements are most needed. [BR07] These are very context-specic issues, but similar problems can easily be transposed to other applications. In particular, 24-hour a day manual monitoring is always costly. And the presence of persons in an area at risk should always be limited as much as possible. 3.2.1.5 High Resolution
Wireless Sensor Networks allow for the collection of data at a high spatial and temporal resolution [EGPS01], [ASSC02]. For phenomena presenting a high variability in these two domains, the benet is obvious, and is repeatedly mentioned as a major asset. Contaminants concentrations in the soil are shown to vary due to the heterogeneity of the soil, and to present daily cycles [RLL+ 07]. Geologists recognize the importance of dense instrumentation for the prevention of landslides in rocky environments [STM+ 07]. Early signs of ooding can appear at any time at several locations in a river basin [BR07]. In general, ne grained data, if it does not come at a higher cost, will always be considered welcome if it can ltered out on-demand.
39
3.2.2
Challenges
Many technical challenges related to the design and deployment of WSNs have been widely discussed and addressed by the research community in the recent years. For a list of issues, the reader can refer to the survey done by Akyildiz et al. [ASSC02]. In this section, we focus our attention on those issues that bear some relevance with the particular context of developing countries. We will show how we took them into account in our own system in Chapter 5. 3.2.2.1 Technical Challenge
Most of technical issues are resolved today... in a laboratory environment. The critical challenge remains to be able to operate a wireless sensor network for an extended period of time and in concrete, existing environments. In 2002-2003, Szewczyk et al. [SPMC04] deployed a sensor network for wildlife habitat monitoring. This system ran unattended for four months. Although researchers reported on several deployments, a large, long-term WSN deployment for environmental monitoring has yet to be demonstrated. Lifetime: Reliable electricity supply remains a recurrent issue in developing countries, especially in rural settings. When wireless sensors are deployed in environments without an easy access to a permanent power source (as would be the case in a house or a car) the question of lifetime of the system becomes critical. Most WSN platforms available today advertise lifetimes in the order of years, thanks to optimized software and hardware power management techniques. However, real deployments tend to show that networks do not last that long, because: 1. In multi-hop data collection networks, intermediate nodes lifetimes can decrease rapidly, because they need to relay other nodes packets. 2. Node failures are frequent in challenging environments. 3. One cannot manually restart a node that entered a non-functioning state. 4. The failure of intermediate nodes can make upstream nodes unreachable from the sink, especially in sparse networks, where routing alternatives are few if existing at all. Solar energy is a possible solution if the sensor is to be deployed outdoor, but if systems are to be deployed unattended in widely available environments, thefts are likely to occur frequently, as was observed by Basha et al. for instance. A solar panel is a visible, attractive and easy to resell part. The design of efcient synchronized communication protocols, and the efciency of sleep / wakeup cycles are likely design directions [BvRW07]. Energy scavenging techniques, that would in the end suppress the need of using and changing batteries, are also being investigated, in particular the use of vibrations [RWR03]. Reliable data collection: Due to the limited radio range of wireless sensors, which can not exceed in the best cases a few hundred meters in line-of-sight, WSNs need to implement multi-hop communication protocols, where intermediate nodes relay the information towards the base station. Low-power operation of multi-hop networks is particularly fragile. Experience shows that the throughput of nodes decreases rapidly with the number of hops to the base station [BTB04]. This necessitates implementing retransmission schemes and store-and-forward techniques that are costly in energy and affect the network lifetime. Currently, the standard multihop protocols, such as CTP in tinyOS, suffer from high variations in the packet delivery rate, especially in cases where nodes are outdoor, and distant from each other, resulting in a sparsely connected network.
40
In most applications intended at industrialized countries, this effect can be mitigated by adding redundancy in the network. Home automation or industrial applications usually consider dense network deployments, where a packet transmitted through the network has the luxury to choose several paths and to rely on a predictable radio channel. In most applications targeted at developing countries, however, the density of the network is likely to be low, because of the expected coverage of the system, and because cost issues become more stringent [BR07], [DZ07]. For this reason, the quest for simple, robust and energy efcient multihop protocols will be a critical issue in the years to come, with an emphasis on multiple-layers protocols that can achieve more efciency with specialization and customization of the communication stack [BvRW07]. Data management: The likely topology of WSNs in our context is a large number of small clusters that need to be interconnected, possibly across wide geographical areas. This can be done at the server side, using middleware applications that make the creation, removal of new clusters plug and play, and treat them as a single entity. This necessitates the development of back-ends that make it easy to dynamically query, store, process and aggregate data over widespread geographical locations. To display data in a meaningful way, one should consider the integration of Geographical Information Systems (GIS). 3.2.2.2 Scientic Challenge
A major challenge regards the usage of the data collected. So far, sensor networks projects result essentially from initiatives taken by computer and communication systems scientists. The connection with the environmental sciences is often loose, if existent at all. Szewczyk et al., in their 2004 seminal work [SPMC04], admit that the wealth of data they collected during their 4 months deployment was of no scientic use in the end. A recurring problem is the capacity for distinct scientic communities to work on a common problem from complementary perspectives. Another problem is that the unprecedented capability of sensor networks to collect data represents a disruptive point in environment monitoring. Often, the tools and theoretical backgrounds to process such an amount of data have yet to be created. For instance, we have been conducting interviews with Indian agricultural scientists since 2004. These scientists always had to cope with limited environmental information in their research. They perceive this as a limitation, yet miss currently the models appropriate to process large amounts of data. Accordingly, the application design can become a chicken-and-egg problem, where the existence of the system is a prerequisite to the denition of requirements, while such requirements are necessary for a successful system design. We address in detail the scientic challenges raised in our project in Chapter 4 and in Chapter 8. 3.2.2.3 Operational Challenge
Deploying a wireless sensor network raises issues well beyond the technical aspect of things. Survivability of the system in the face of adverse environmental conditions, maintenance of its components, but also interaction with the local population can be experiences full of surprises. Among our examples, the only project that went all the way to an actual deployment was only left in the eld for a few weeks, and attended by specialists throughout [RBE+ 06]. The lessons to be learned for a long-lasting deployment represent still an open question, which we address in Chapter 6. 3.2.2.4 Sociocultural Challenge
The hardest to grasp precisely, this challenge is also the most important to address. In the words of Brewer et al., as far as technology is concerned, (...) Western market forces will continue to meet the
41
needs of developing regions accidentally at best [BDD+ 05]. In the same spirit, we advocate the importance of exploring the potential of an emerging technology - sensor networks - in concrete situations, in order to take the ecological, social, cultural and economic conditions of developing countries into account in the development of hardware and software platforms alike. However, as is always the case with a disruptive technology, special care must be taken to avoid most common pitfalls. This challenge is addressed in Chapter 9. Design/Implementation Gaps: Heeks [Hee01] argues that the failures of information systems projects in developing countries are often caused by design-actuality gaps. Country context mismatches (in terms of institutions, infrastructures etc.), as well as hard-soft gaps (rational design versus cultural and political actuality) play a role all the more important if the system was designed in an industrialized context. To summarize, failures can generally be explained by the distance (geographical, cultural or socioeconomic) between the designers of the system and its intended community of users. Using participatory design can mitigate this risk. Heeks warns, however, that participatory design in itself is no guarantee for success in developing countries, because these techniques have usually been developed in and for industrialized countries organizations. A lesson to be drawn is that a participatory approach in a developing country is instrumental to success if and only if it integrates a tool to bridge the contextual gap between design and use. In order to bridge this gap, Heeks advocates the usage of hybrids, specically individuals who understand both the alien worlds of the community of users and of the community of designers/builders of the artifact. Wireless sensor networks also present an important feature, in the fact that they constitute an emerging technology in constant evolution. This leaves a signicant place for experimentation, and presents the advantage of being able to develop a technology specically for the developing countries context, instead of tweaking existing systems made to operate in a different context, which is a criticism made recurrently to projects dealing with ICT for development [Hee01]. Computer Literacy and Application Ownership: It is not enough for an information system to satisfy adequately the needs of its intended target population. When this population is living in poor and remote areas with a low level of literacy (not to mention computer literacy), a major issue is the capacity of the user base to understand, use and nally own the system (we dene ownership as the ability and willingness to maintain the system in a working state and to integrate it in daily activities). For this to happen in our case, we have to satisfy two conditions: 1. The ability of the sensor network to function autonomously, without the need of skilled maintenance. As we saw, this is a design goal of sensor networks, not yet fully realized, but on which will depend the success or failure of the whole technology. 2. The capacity of the population to learn about the use cases of the system. Here, one can use the concept of capacity building and knowledge creation through apprenticeship [Pan04]. Our hypothesis is that there are some aspects of apprenticeship that make it particularly suited in the acquisition and integration of radically new paradigms of knowledge. It is a self-organized process in which every individual takes ownership of the knowledge he or she is acquiring. Not relying on formal teaching, it can be more integrated in the social structure and possibly more equitable, as people without the time, resources or will to attend classes can be reached through it. 3.2.2.5 Economic Challenge
Business models for WSN deployment must be looked at carefully: Sensor networks are still a maturing technology, which remains costly. Today, the price of a single wireless sensor is typically between USD 100.- to USD 300.-. The price of wireless transmitters and common sensors (temperature for
42
instance) is expected to drop sharply in the upcoming years. Even so, customized sensors (such as soil moisture sensors), will remain expensive unless new designs and local manufacturing can be realized. This requires an expertise that emerging economies, such as India, Brazil or China, possess already. By building partnerships with companies and institutions from these countries, one can leverage on this capacity. In a rst phase, one can privilege research projects that use sensor networks as a tool for validating or developing the theoretical frameworks (in hydrology, agronomy, environmental sciences etc.) relevant to the pursued goals. Such small scale projects also play a role in the technology maturation process. More generally, due to the technologys current high cost, it is important to raise awareness among potential sponsors, such as the development agencies and international organizations, about the potential of wireless sensor networks in the context of development. As mentioned later on in this thesis, we followed this approach in the COMMON-Sense Net project, with the involvement of the Swiss Agency for Development and Cooperation as main funding source [PRP+ 06]. However, a precise economical analysis of the sustainability of our system is beyond the scope of this dissertation.
Chapter 4
Wireless Sensor Networks for Marginal Agriculture in India

In this chapter, we present our own contribution to the application of novel environment monitoring techniques to the context of developing countries: a decision-support tool for marginal farming in India. In Section 4.1, we start by a summary of the activities that led to the project denition and consortium. In Section 4.2, we explain the general objectives of the project. In Section 4.3, we give a description of the village where we concentrated our activities. In Section 4.4, we describe the survey on information needs that was run in this village and two others in Karnataka (India). The applications that we derived from this survey are detailed in Section 4.5. Finally, Section 4.7 explains the methodology that we decided to use throughout the project.
4.1
Project, Consortium and Funding
The initial idea for the project came in 2004. Initial personal contacts between EPFL and the Centre for Electronic Design and Technology (CEDT) at the Indian Institute of Science (IISc) were strengthened when EPFL issued a call for proposals for projects centered on scientic collaboration with academic institutions from developing countries. The funding came from the Swiss Agency for Development and Cooperation (SDC) via a research grant endowed with CHF 3 million (the NCCR MICS1 and CEDT provided a matching fund for our project). The primary goal of this call for proposal was to promote interdisciplinary research focused on critical problems faced by emerging and developing countries. The author of this thesis traveled to Bangalore for a set of introductory meetings where the theme of agricultural management quickly surfaced as one of the most pressing issues in Karnataka. During these meetings, the need for a local partner in the eld appeared as an important prerequisite before any action to be undertaken. The Chennakesha Trust, a Non-Governmental Organization (NGO) active in the district of Tumkur (see Fig. 4.1) was invited on board at this stage. We chose to focus on small land-holding farmers, because they represent an important part of the farming population in Karnataka (see Section 4.3.4). Moreover, they are more concerned than any other by the environmental issues that we identied in Chapter 3. As a local organization regrouping resourcepoor farmers, the Chennakeshava Trust (see Fig. 4.2) seemed ideally placed to identify and address the challenges faced by this category of population.
1
Swiss National Centre for Competence in Research in Mobile Information and Communication Systems
43
44
CHAPTER 4. WIRELESS SENSOR NETWORKS FOR MARGINAL AGRICULTURE IN INDIA
Figure 4.1: Tumkur district in Karnataka, India
The project was submitted in Fall 2003, accepted in early 2004, and ofcially launched in Summer 2004. Its background and rationale are described in the next section.
4.2
COMMON-Sense Net: a Decision-Support Tool for Agriculture
Since 2001, drought has hit India repeatedly. A wave of farmers suicides ensued, claiming probably tens of thousands of lives throughout the country, although ofcial gures are lacking [Mis06], [Zub06]. What is certain, however, is that the principal cause of this outbreak is a vicious circle of borrowing money to buy seeds, and getting into increasing debts because of crop failure [Sai05]. Adverse climatic events can often be blamed, but farmers bear their part of responsibility, since they tend more and more to replace subsistence crops with cash crops2 , sometimes ill-adapted to the local conditions, often inefciently grown due to lack of knowledge and experience. Farmers lack information and knowledge to face the new challenges raised by the shift of paradigm in their activity. Improved environment monitoring may be part of the answer. Although it cannot prevent drought or replace a political solution to the structural problems of Indian agriculture, environment monitoring can help to improve the lives of resource-poor farmers by mitigating the effects of extreme events, allowing the farmers to adapt their strategy to abnormal or changing climatic features when they occur. Information on the temporal and spatial variability of environmental parameters, their impact on soil, crop, pests, diseases and other components of farming, play a major role in formulating the farmers strategy [Gla03], [HNM00], [GRR02]. Today, large mechanized farms in developed countries take this factor
2
A crop which is grown for money
4.3. SETTING THE CONTEXT
45
Figure 4.2: Some members of the Chennakeshava Trust
into account and utilize the convergence of several technologies, including in-eld sensors, geographic information system (GIS), remote sensing, crop simulation models, prediction of climate and advanced information processing and telecommunications. Similar techniques can be highly useful to farmers in the semi-arid regions of developing countries, provided they can be adapted to small land holdings and labor intensive, low productivity agriculture. However, traditional approaches are too expensive f and do not scale down to the size of a marginal farmers plot. Moreover, the implications of climatic variability in developing countries are a largely unexplored area for agriculture research [SGB00]. Designing and implementing a decision-support tool based on environmental information for Indian marginal farmers is an ideal occasion to investigate the use of wireless sensor networks in developing countries (see Fig. 4.3). The rest of this chapter describes the research conducted around the design of such a decision-support tool. This project is named COMMON Sense Net (for Community Oriented Monitoring and Management of Natural resources via a Sensor Network) and, as we mentioned, stems from a multidisciplinary partnership between India and Switzerland, with both academic institutions and the civil society involved. Because of the novelty of the issues that this project addresses, it uses extensively a participatory and iterative design (see Chapter 9). The initial goal is to provide farmers with more strategic options for their crop management, thanks to enhanced environmental information. The long-term goal of the project is to help developing replicable strategies for agricultural practices.
4.3
Setting the Context
In India the political institutions are organized hierarchically according to a 5-level scale: 1. Central Government in Delhi. 2. State: in our case, Karnataka. 3. District: around 10 taluks. Chennakeshavapuras district is Tumkur.
46
Figure 4.3: Technical components of a decision-support system for agriculture
4. Taluk: around 100000 citizens, corresponding to 10 panchayats. Chennakeshavapuras taluk is in Pagavada. 5. Panchayat: around 10000 citizens. A panchayat usually comprises 5 villages. Only the 3 upper-layers have legislative and judiciary bodies. Up to the taluk level, there is only an executive council. Each assembly is elected once every 5 years, but not in a synchronized fashion.
4.3.1
The Pavagada Region
The Pavagada region is a part of the large semi-arid tract of Southern India. It is centered on 14o N and 77o E and is situated in the Eastern part of Karnataka state. The central part of the region is a plateau with an elevation of about 600 to 700m, and several chains of rocky hills found in the landscape form series of watersheds. The upper catchment areas of the watersheds are utilized for rain-fed groundnut cultivation. Hills and rocky outcrops constitute the grazing lands for the livestock. In the lower reaches of the watersheds, manmade tanks storing runoff for irrigation were constructed in the 19th century and rst half of the 20th century. In addition, large open wells, as well as tube wells, support small patches of irrigated farms. For economical reasons, however, about 85% of the total cultivated area depends exclusively on rainfall for the growing of groundnut during the rainy season (June-November). Indeed, water for irrigation is too costly for the resource-poor farmers. Their farms are usually located on the upper reaches of the local watershed, and thus cannot benet from the water stored in traditional surface storage reservoirs in the valleys below. Since the drilling of bore wells is costly and has a history of high failure rate, the risk is too high for them to take. The major climatic feature of the Pavagada region is the low amount of rainfall and its high variability. The annual average is 561mm, with a standard deviation as high as 190mm. The distribution of the rainfall within the year is bimodal [RGK+ 04]. The maximum rainfall occurs in the second half of September. The second mode is between the last week of May and the rst week of June. Another major
47
Figure 4.4: Chennakeshavapura village surrounding elds
characteristic of the climate of the region is the frequent occurrence of long dry spells. Consequently, the crop is highly prone to moisture stress, a risk enhanced by the low moisture retention capacity of the shallow sandy loam soils. As a result, the cost of cultivation is not recovered in 60% of the harvests [RG99].
4.3.2
The Chennakeshavapura Village
Chennakeshavapura (see Figs. 4.4, 4.5, 4.6) is a village of approximately 2000 inhabitants in the Pavagada area. At the local level, wealth remains by and large the most important factor of political inuence. According to a local contact-person, the elected members of the panchayat can be the target of signicant pressure coming from the richest landowners, who usually prefer to act unofcially than to be exposed as ofcials. Thus, corruption is not uncommon, since the main task of the panchayat is to distribute the funds coming from the higher levels in order to realize community-based projects: building roads or buildings, distribute grants etc. [Rao08]
4.3.3
Type of Agriculture
As in other areas in Karnataka, rain-fed farming is the most practiced form of agriculture, revolving essentially around groundnut, with other crops such as pigeon pea and cereals, and areca nut trees. Irrigation remains possible thanks to the community tanks. The three biggest ones have a command area respectively of 85ha, 17ha and 12ha. Fig. 4.8 shows the dyke built more than one century ago to circumscribe the main tank. Smaller reservoirs, of the type shown in Fig. 4.7 are scattered around the area. However, the recent trend is one of building individual electricity-powered wells, sometimes with retentions tanks (as in Fig. 4.9). There is an estimated 55 bore-wells in around the village, most of them unregistered.
48
Figure 4.5: A neighborhood in Chennakeshapura
Figure 4.6: The villages movie theater
49
Figure 4.7: CKpura, a community tank
Figure 4.8: Walk on the dyke of the main community retention basin
50
Figure 4.9: The reservoir of an individual well
4.3.4
Marginal Farmers
In India, the share of agriculture in employment is still about 67% [BM04], with a majority of small land holdings. In Karnataka, 87% of the farming families own farms of less than 4 ha, accounting for more than 50% percent of the total cultivated area. Families with very small farms (less than 1 ha) constitute 39% of the total. They usually lack access to irrigation facilities and depend on rain-fed farming for their livelihood. Their crop yields are highly unreliable due to the variability in both rain-fall amount and its distribution [GAR99]. For all these reasons, we refer to this group as resource-poor farmers, more traditionally called marginal farmers. Marginal farmers in India and South Asia in general are the category of population that benetted the less from the economic boom and overall poverty reduction of the last 15 years [BS05], [DKMU03]. They constitute a recurrent target for development agencies, but are difcult to help efciently, because of their economic helplessness. Our experience with marginal farmers in Karnataka shows that these farmers feel an estrangement with the outside world. The feeling of distrust towards the scientic community is extremely widespread. It emerged from informal meetings held in CKPura with CK Trust members that farmers have the impression they did much more for the career of scientists than the latter did for the improvement of their life conditions. In our experience, this leads to a general feeling of instrumentalization in the marginal farming community. On top of that, three consecutive years of severe drought have taken their toll on the small and marginal farmers of the area. Some of them, incapable of repaying their debts are now desperate. Marginal farmers can also be bitter about former agricultural development projects that have consistently left them behind while focusing on irrigated agriculture. Communication with outsiders is difcult, because of language issues, most farmers speaking Telugu or Kannada only. Moreover, if a foreigner comes to talk to them, they are likely to expect something immediately, typically a loan. As a consequence, the dialogue of scientists with resource-poor farmers is a challenge. The only way to reach them is to bridge the gap with the use of a hybrid [Hee01], namely a person that has a foot in both worlds, on one hand practicing science and technology, on the other hand, knowing the marginal farmers and working with them for long enough to have gained their trust. We recognized this aspect early on in our project and accordingly, built a partnership with an agronomist, farmer and founder of an
4.4. A SURVEY AND ANALYSIS ON FARMERS NEEDS
51
NGO working for marginal agriculture3 , Mr. P.R. Seshagiri Rao4 . Another characteristic of marginal farmers is their inability to make any investment that would improve their productivity, due to unavailability of nancial resources. As an obvious consequence, their rst and foremost claim is getting affordable loans [RGK+ 04]. Along these lines, we defend the development of appropriate microcredit [Sap06] schemes as a prerequisite to any rural development in India (and probably in developing countries at large). However, this is beyond the scope of an academic project such as ours, and as a consequence, our solution is orthogonal to microcredit. Our goal was to look beyond the current horizon of local farmers and to try and anticipate what advances in sciences, enabled by the proper technology, could improve their livelihood through better land and water productivity. The gap is wide between the mental models of those who have rst-hand experience of daily agricultural challenges and engineers. Hence the necessity to involve another partner in the project: agricultural scientists who are trying to understand better the processes at work in the eld, where countless physical, biological and meteorological parameters interact constantly. In order to address the subject in a credible way, we had to build a communication chain involving ICT researchers, agricultural scientists, hybrids and marginal farmers. We initiated a partnership with the Centre for Atmospheric and Oceanographic Sciences (CAOS) at the IISc, which lasted until 2005. At that point, we approached a Professor at the University of Agriculture Sciences (UAS) in Bangalore, which allowed us to conduct several interviews and surveys on UAS campus. The rst step was then to conduct an assessment of the needs perceived by a community of marginal farmers for the improvement of their livelihood.
4.4
A Survey and Analysis on Farmers Needs
The results and discussion of this section are based on a eld survey conducted over a period of ten months from August 2003 to May 2004 in three villages of the Pavagada region (Southern India): Chennakeshavapura (CKPura), Venkatapura and Ponnasamudra [RGK+ 04]. The goal of this enquiry was to identify and categorize the information needs of the population living in the semi-arid regions of India, and to assess the relevance of environment monitoring in such a context.
4.4.1
Survey Methodology
Before beginning the assessment of information needs, Rao et al. classied the different user groups, with the family as the basic unit. Each family can have more than one livelihood activity (e.g. farming, sheep keeping, trade, fuel wood gathering etc.). The various livelihood activities of the families are listed on the basis of effort allocated by the family for the activity. Livelihood activities with maximum allocation of effort are categorized as major livelihood activities. During the initial survey and mapping of the village, for each neighborhood (cluster of houses) or caste group (endogamous group signifying social status) the authors of the survey identied a set of knowledgeable individuals. Discussions with these people allowed determining the major livelihood and other livelihood options of the families belonging to the relevant user group. In the second phase, Rao et al. collected information needs of various groups. For this part, they held group meetings and complementary semi-structured interviews. For the group meetings, the resident
The Chennakeshava Trust, introduced in Section 4.1 Mr. Rao is the author and coauthor of several leading articles in the area of Indian ecology and agriculture. He lives and farms in the village of CKPura, and works as a consultant for agro-business companies and international organizations alike.
4 3
52
User group Rain-fed Farmers Irrigated Farmers Irrigated Orchards Owners
Number of families 160-200 40-60 10-12
Meetings held 11 4 2
Participants (average) 29 18 10
Table 4.1: User survey participation

Rain-fed Farm 1 2 3 4 Irrigated Orchard 4 2 1 3 5 Irrigated Farm 1 4 2 3
Crop yield prediction Rain prediction Plant disease prediction Daily jobs opportunities Water level in bore wells Groundwater survey Electricity supply
Table 4.2: Priority of information needs per user group (1 denotes the highest interest)
families were grouped along patterns of resource use (such as irrigated agriculture, rain-fed agriculture, animal grazing, daily labor etc.). Table 4.1 show the three categories of main interest, with their weight in the community in terms of number of households. It also gives the number of households that participated actively to the survey, and indicates in each case the number of meetings that were held. During group discussions, the farmers identied relevant issues and prioritized them. Several group discussions with the members of the user group were held to determine focal issues of their information needs. The identied focal issues were prioritized by consensus. Any disagreements in choice of focal issues or assignment of priorities were also documented. Separate discussions were then held with interested individuals, in order to gather the details of information on focal issues. These discussions typically lasted for 2 - 4 hours with 3 to 6 users and usually took place at the farms or houses of user group members. The interviews were not based on questionnaires, but on open-ended interactions. The interviewers focused on the general categories highlighted in the group discussions, and sought to extract information from the individuals in an interactive manner. The questions and answers were collected in a written form and interpreted by the survey main author in order to create the themes that are presented in detail in the survey document [RGK+ 04]. The following section focuses on the analysis of the different farming groups, at the expense of shepherds, shop owners, craftsmen etc. Special emphasis is given to the resource-poor farmers, since they constitute the target population of the COMMON-Sense Net project. Richer farmers are also considered, since they are likely to be directly affected by a deployment of the system. More information can be found in the survey report.
4.4.2
Summary of Results
The information requirements of the rural families were very diverse. They covered a wide range of needs including weather prediction, market conditions on a particular day, or legal advice on land-holding rights. A signicant nding, however, is that environment-related information ranks high in the perceived needs of the rural families. Drawing directly from the user survey document, we constructed a prioritization of information needs per user group, as depicted in Table 4.2, in which a 1 designates the highest priority, 4 a low priority and - an absence of interest. In this table, one can distinguish different types of issues. Concerns about electricity cuts or ground-
4.4. A SURVEY AND ANALYSIS ON FARMERS NEEDS
53
water and wells are specic to farmers rich enough to afford to pay for irrigation. As for resource-poor farmers, their wish for better weather forecast or employment opportunities can hardly be satised by better agricultural practices. On the other hand, the two themes of crop yield prediction and disease control stand out prominently in all farmers categories. For these subjects, the management options available, their costs, risks and benets are largely inuenced by the high variability of environmental parameters.
4.4.3
Interpretation and Motivation
At rst sight, the realization that crop yield is an important concern for farmers seems obvious. However, the non-trivial nding of the survey is the fact that crop yield prediction is critical mainly for poor farmers, because their lack of resources forces them to constantly adapt their strategies to the evolution of the environment. Hence, expected yield plays an important role in the choice to invest or not in a tactical option, such as buying fertilizer or pesticides, borrowing and carrying water, etc. As we showed in the previous subsection, environment monitoring and understanding the impact of variability constitute a leitmotif for farmers. This calls for an extension of the usual paradigm of rural development projects centered on ICT [PH03]. Whereas projects currently consider primarily interpersonal communications such as rural phone and Internet connectivity, the COMMON-Sense Net project wants to advocate a different category of applications that will allow the farmers to connect to- and act on the constraints of their own environment in a more precise way. In semi-arid regions, the amount of rainfall and its distribution during the season inuence most of the farming: crop yields, disease and pest incidence, farming operations, level of inputs, etc. Because they are farming under such a high-risk situation (uncertainty of expected benet), poor families try to minimize their risk by investing as little as possible, be it for soil fertilizers, soil water conservation or spraying for pest and disease management. The downside of such a strategy is that in good rainfall years their crop yields are much lower than the eld potential. Experience shows that poor farmers usually achieve about half of the yields of large farmers, who use better soil-fertility- and pest-management. In situations of uncertain output, the use of a decision-support system able to give information on the benets and risks of all the available options will help resource-poor farmers to make an informed choice for the best strategy. It is in this area that a sensor network can help them in several ways: Firstly by making it a tool in the hands of agricultural scientists who work on more sustainable practices and strategies. Simulation models of crops, pests, diseases and farming operations are important tools to answer several of the farmers information requirements. The environment monitoring data provided over time and space by sensors can be used to validate and calibrate existing models. Finally, it can help to assess the efciency of simple water conservation measures, such as planting trees or mulching. Secondly, when used directly in the eld, sensors can improve farm-level decision making by providing important benchmarks for the impact of moisture decits, and monitor in real-time the eldconditions with regard to these benchmarks, providing the farmers with a decision support system adapted to their needs, and encouraging them to invest in order to get higher prots from their farms. In particular, resource-poor farmers resort to rain-fed farming not out of choice, but out of necessity. Irrigation practices in the semi-arid areas of developing countries are usually inefcient and require large quantities of water. This necessitates drilling wells, which is either too risky or unaffordable for them. A reliable decision-support system is a component of a decit irrigation system that seeks to maximize the impact of irrigation on crop yield while minimizing the intake of water. For poor farmers, this could mean applying new strategies of partial irrigation, such as transporting water from community tanks on
54
Information Needs Crop yield prediction
Specic Questions of Marginal Farmers 1) Assess appropriateness of crop choices 2) Assess appropriateness of farming operations. Provide forecasts of occurrences during weeding. Cost/benet analysis of using bunds and trees. Increase yield with minimal water use.
Strategy to Provide Information Use existing scenariobased models. Validate these models with local data. Determine environmental parameters that have an inuence, and their respective values. Compare effectiveness of different measures Dene critical thresholds in soil water content at different stages in crop growth. Issue warnings when water is needed (if possible, indicate amount of water needed)
Role of Sensors Soil moisture measurements to validate groundnut crop model.
Pest and Disease Prevention
Gather soil moisture, air temperature and relative humidity data Gather soil moisture measurements in different conditions. Gather soil moisture information in each homogenous parcel
Water Conservation Measures Decit Irrigation
Table 4.3: Environmental data for marginal agriculture
carts, renting rich farmerss wells, etc.
4.5
Use Cases and Related Environmental Data
At this early stage of the project (we were in Spring 2005), it seemed easier to collaborate with agriculture scientists in order to design possible applications, because a direct interaction with the farmers was feared to generate either incomprehension (their immediate attention being more focused on loans) or high expectations leading to disappointment and disinterest, since our prototype will take time to be fully operational. As a consequence, we dened system functionalities and use cases jointly with a crop physiologist from the University of Agriculture Sciences, Bangalore, a farmer with high-education training in agronomy and environmentalists from the IISc. The goal of these meetings was to translate perceived needs into scientic solutions. Drawing on the surveys analysis of the needs of small-farm families in terms of environmental data (Rao, 2005), we extracted the most promising and rapidly implementable applications and analyzed them (Table 4.3). These considerations led us to the use cases that we detail in the next section.
4.5.1
4.5.1.1
Crop Modeling
Rationale
The rst and foremost concern expressed by marginal farmers was about crop yield prediction. Several crop simulation models are available for simulating the growth of various crops and crop mixes with different environmental constraints such as moisture stress, nutrient stress and water logging. These
4.5. USE CASES AND RELATED ENVIRONMENTAL DATA
55
models are an important component of a decision support system in this area. If the results of envisaged scenarios can be communicated to the farmers, that would signicantly enrich their strategic options. In our case, agricultural scientists identied DSSAT (Decision Support System for Agrotechnology Transfer) [MS02] and APSIM (Agricultural Production Systems sIMulator) [MHH+ 96] as the most promising models for the Pavagada region. They have, however, certain limitations. Both DSSAT and APSIM have a narrow and deep focus on certain components of decision making crop growth and yield - and neglect other pertinent areas ([MHH+ 96], article of Stephens and Middleton in [MS02]). In decision making for farmers, precision should not be provided at the expense of relevance. In other words, it is more important to be roughly right than precisely wrong. A specic criticism of DSSAT is that it is highly crop-plot centric, whereas the users consider farming processes at the higher scale of a whole agricultural ecosystem [Wal02]. Both models based their decision making on simulations that take as input a handful of physical and climatic parameters, such as soil type, crop type, precipitations seasonal trends and irrigation schedule. A key element to the capacity of these systems is the ability to predict, based on the inputs, the water content of the soil at any stage of the cropping season. Recently, this ability has been questioned in the scientic literature [GJJ03], [FK03]. A validation methodology seems needed to assess the performance of the models, especially if they are to be used in contexts different from the ones in which they were developed in the rst place. Comparing actual yields with predicted ones would be a possibility. However, it is impractical because of the number of options to take into consideration. Moreover, if predicted and observed yields do not match, no underlying cause will be identied and the possible recalibration of the model will be difcult. Collecting data about soil moisture at different depths throughout a cropping season and comparing them with values computed by the model ensures that the main factor of crop yield is well understood, and allows to narrow down possible discrepancies to a limited set of coefcients, with the hope of being able to tune them appropriately.
4.5.1.2
Use of a WSN
Manual collection of soil moisture data at different depths for a duration of six months is a cumbersome and unreliable process. Deploying automated sensors in well-identied plots and let them collect data over the period could simplify the process and provide for high resolution data specic to different eld conditions (soil physics, type of crop and of agriculture, geographic location with regard to slopes, exposition to the sun, proximity of water sources). The Chennakeshava Trust identied two groundnut-growing areas, one rain-fed and one irrigated, with 10 measurement points each, where a preliminary experiment could be run. The application is as follows. Once the sensor network is deployed, the data are gathered repetitively, saved into a database and uploaded regularly by crop modeling specialists, who:
tune the model coefcients to the relevant parameter space in the region of interest; validate the model with the new set of data; reect on the models performance as improved environmental data become available.
56
4.5.2
4.5.2.1
Water Conservation Measures

Rationale
Farmers who cannot resort to irrigation need to make the maximum use of precipitation water throughout the cropping season. They already do so, however a precise assessment of the efciency of such measures is still lacking. Comparative readings of soil-moisture can be used to assess the efciency of different water conservation measures, such as building bunds and planting trees to trap water in the shallow layers of the soil, or using mulch and gypsum to reduce evaporation. 4.5.2.2 Use of a WSN
This use case is similar to the previous one, except that, here, soil moisture readings are used directly. Sensors are placed in elds that are comparable from a physical point-of-view, but where different water conservation measures are used. Here again, different parameters are relevant, including the location of the cropping plots. For this, spatial variability has to be taken into account, justifying the use of a wireless sensor network. Information would be eventually exchanged with farmers through participatory meetings.
4.5.3
4.5.3.1
Pest and Disease Prediction/Prevention

Rationale
Pests and disease are a major concern for farmers. They realize that environmental parameters play a role in the emergence of such phenomena. However, the nature and the value of these parameters is still unclear. As a consequence, farmers who can afford it tend to treat their crop no matter what, whereas poor farmers leave their crop unprotected because of the cost of spraying. Observing the correlation of different parameters with the outbreak of pests and diseases could lead to the denition of statistical models of pest or disease prediction. If such models can be developed, they could be used subsequently in the eld in order to issue warnings. 4.5.3.2 Use of a WSN
In a rst phase, sensors can be deployed in a semi-controlled environment in order to observe correlations between environmental parameters and outbreaks. The necessary granularity of such measurements is still unclear, but eld- and crop-variability, as in the previous applications, should not be underestimated. The time necessary to conduct such a study is not precisely specied by the agricultural scientist whom we interviewed, but denitely spans several years. If this phase is successful, in the longer term wireless sensors can be deployed in the eld, in order to issue warnings to farmers. One could argue that in this case, the action depends on the local environmental status, and that a network is not necessary in this case. However, there are two arguments to support such a solution: 1. Not all environmental parameters have the same level of variability. It would be an overkill to instrument air temperature or humidity at each and every plot. Using a networked solution allows for the efcient aggregation of data. 2. In case of risk, warnings need to be issued to farmers. One might think of an SMS-based system, the relevant sensor sending a message directly to the concerned owner. Or each sensor could have
4.5. USE CASES AND RELATED ENVIRONMENTAL DATA
57
a display that can be read by the farmer. However, both solutions increase the cost of a single sensor. Using a network to centralize the collection of information at a local information kiosk is another solution. In this case, farmers can pass by the kiosk whenever they want it. This solution has the advantage of fostering potential exchange of information among the stake-holders.
4.5.4
4.5.4.1
Water Management for Decit Irrigation

Rationale
The situation of marginal farmers with regard to irrigation varies depending on the location of their elds. Some can access to community tanks, because their plot is located directly downstream from one, but others are totally exposed to the unpredictability of weather. Their only lever is their willingness to build their own retention tank (which they usually do not do) or to transport water from tanks to their eld. However, community tanks regularly dry out, and transporting water is an extenuating task. If water is not used optimally to the last drop, the result is often not worth the effort. As a consequence, marginal farmers can benet from the technology of decit irrigation, an agricultural water management system in which the water needs of the crop (potential evapotranspiration) during the growing period can only be met partially by a combination of soil water, rainfall and irrigation [D.R05]. Decit irrigation management requires optimizing the timing and degree of plant stress within restrictions of available water. Of particular use to the farmers is the knowledge of benchmark points for crop/trees water requirements (those points are specic to a particular crop). Using the recent trend of soil moisture values recorded by sensors and the knowledge of these points, the farmer could predict the behavior of his crop and use simple water management techniques. 4.5.4.2 Use of a WSN
For such an application, in addition to deploying soil-moisture sensors, other parameters are needed. Climatic parameters such as daily rainfall, sunlight hours, wind speed, and air humidity are homogenous enough to necessitate the deployment of only one weather station every few square kilometers. Soil characteristics, however, can vary signicantly due to composition and situation. This means that the soil moisture content has to be assessed every few hectares at least. This requirement is even reinforced when the average size of individual plots is small. Concretely, it is reasonable to deploy between 2 and 4 of sensors (for cross-checking) per homogenous parcel, compute the model coefcients for this parcel over a calibration phase, and retrieve them from a table when a prediction has to be made. In order to assess the inuence of a particular feature of the landscape (such as trees, bunds, etc.) on the soil conditions, a sensor should be added at this particular location. The use case is as follows: Calibration As a one time effort, soil moisture probes need to be calibrated with measurements from the gravimetric method, an accepted standard procedure of determining soil moisture. Climatic probes are also calibrated. Then, in normal mode of operation, the calibration continues to take place, in a feedback loop based on the difference between the predicted and measured value in order to take local variations into account. Alert Real-time alerts are given whenever the measured soil-moisture of a parcel reaches a threshold in the benchmark values. These alerts are automated, but farmers have to be notied by the system operator. Once the alert is given, the farmer should be able to look at weather forecast data
58
and know, based on historical climatic data for the region, what is the probability of rain in the near future. Soil Moisture Prediction Based on the model and the actual measurements, the system uses a real-time learning process to give predictions on soil-moisture values over time. Water Requirements Assessment Based on the same type of request as above, the system gives an estimate of the minimum irrigation water needed according to the benchmarks.
4.6
Design Guidelines
In all the use cases developed in the previous sections, we gave arguments to advocate the use of a wireless sensor network. The two rst use cases do not take direct advantage of the possible real-time features of a sensor network, because the response-time is not critical, and because they aim at improving the situation of marginal farmers indirectly, through observation and scientic analysis. The third and fourth use cases constitute direct applications in the eld. Regardless of their context of deployment, however, each of them has to be positioned according to the challenges that we highlighted in the previous chapter, so that proper design guidelines can be drawn.
4.6.1
Technical Point-of-View
Lifetime: to yield usable results, all the use cases suppose a eld deployment of several months. This requires adapted energy sources. Unfortunately, deploying sensors with solar panels is usually problematic in agriculture. Solar panels limit the choice of locations for sensors, because orientation and placement have to be thought in relation with insolation. They also increase the total bulk of the equipment to deploy, making it more difcult to handle and more attractive to steal. For all these reasons, we found it necessary to resort to batteries, whose energy prole must be carefully studied to insure the necessary lifetime. An additional difculty comes from the sparse nature of the network, which means the likely partitioning of the network into clusters, with some nodes bearing the extra burden to connect clusters to a single data collection point. These nodes will have to implement some bridge technology (typically GSM or Wi-Fi) and their energy requirements will increase signigcantly. Solar energy will become an alternative in this case, because it is impossible to rely on the power grid, and because the size of the batteries necessary to power the nodes will become prohibitive. The small number of such nodes justies a more careful placement. On the hardware side, the choice of the communication platform and of the environmental sensors to use will be conditioned in part by their energy prole. On the software side, low-power protocols (MAC and routing) need to be assessed precisely to choose the best t. Reliable data collection: Sparse networks make it more difcult to collect data in a reliable fashion. Long distances between nodes and few routing alternatives are the main culprits. In our case, growing vegetation can signicantly impair the communication as the cropping season advances. This means that the choice of the communication platform is especially crucial. Also, this has implications in the network architecture. A clustered two-tier architecture will be necessary in most cases. Assessment of routing protocols in such challenging settings is also a necessary prerequisite, because results of long-term WSNs are still lacking in the literature. Sensing accuracy:Finally, one question to solve is that of the effectiveness of probes. Whereas temperature and relative humidity embedded sensors have been used for years, soil moisture probes have
4.7. METHODOLOGY: SCIENCE AND FARMERS
59
to the best of our knowledge not been formally evaluated in the context of wireless sensing.
4.6.2
Scientic Point-of-View
What use to make of the data is a complex challenge to address. As we realized, use cases are easy to nd, but precise usage of data is not known, because scientists are not used to work at the level of granularity allowed by WSNs. There is a need rst to collect extensive data, then to provide visualization and processing tools for scientists, in order to develop full decision support system and/or recommendations based on this. A possible strategy is to start with the simplest use cases, the ones that need minimum data processing, such as assessment of water conservation practices. This would allow for the development of the technology and system, its deployment and deployment assessment and the collection of an extensive set of data. At this point, it would also be possible to reuse the collected data by presenting them to agricultural scientists as a proof of concept, in order to foster the precise denition and development of other use cases.
4.6.3
Economical and Sociocultural Point-of-View
It is to be noted at this point that a major question mark is the affordability of such a system by farmers. A wireless sensor, all equipped, costs in average USD 200 300.- in 2008, a price a priori too expensive for an Indian farmer5 . The technologys cost is widely believed to drop dramatically in the years to come, but it remains to be seen to what extent. Before we reached this state, the effectiveness of WSNs in the eld needs to be proven. All the applications that we mentioned earlier necessitate the deployment of a sparse network with an average distance between the nodes of 100m or more. Each node will have to survive at least one cropping season in order to minimize the maintenance. These are major design issues, as highlighted by Chapter 5. Before a system can be made commercially viable, external sources of funding need to be sought, in order to make a proof of concept of the whole technology. In this way, one can build a case for potential commercial applications, increase awareness of the users and hence their willingness to invest in a new technology, and be prepared for the time when the price of technology decreased enough. Sources of funding depend on the use case and of the context of deployment. For applications directly targeted at poor farmers, and meant to be deployed in the eld, development and cooperation organizations are ideal targets. In our case, we obtained a grant from the Swiss Agency for Development and Cooperation.
4.7
Methodology: Science and Farmers
As shown in the previous sections, meetings with agriculture specialists led to general use cases. However, it proved to be difcult to go beyond this initial stage, and to design experiments based on these ndings. As mentioned by P.R. Seshagiri Rao [Rao08], WSNs seem to represent a disruptive technology whose full potential is difcult to grasp for scientists who have been always working with the constraint of having to cope with sparse environmental data. Another difculty in working upfront with scientists is the lack of funding. As the meetings took place after the SDC-funded project started, we could not nance scientists to conduct experiments. On
5
Protability studies are yet to be conducted, which is beyond the scope of this work.
60
their side, these scientists expected to see a concrete system working before considering to spend time and money on related experiments. Experiments in the eld, on the other hand, have several advantages: 1. They allow for a thorough testing of the system in a real setting 2. They enable the collection of data that can attract interest from scientists and trigger the denition of precise experiments. 3. They allow for the testing of the initial response of farmers to the intrusion of technology in their elds. Accordingly , we opted for a similar strategy as Basha and Rus [BR07] where we design and test to some extent the system in the lab (both in Lausanne and Bangalore), before deploying it in the eld as it is. We then would go back to the lab to address the identied issues. The results of the preliminary deployment were expected to conrm the technical viability of the system, and to reect on the initial efforts to involve farmers in the use of wireless networking in their plots. As we explain in Chapter 6, the results obtained in Chennakeshavapura encouraged us to recontact and probe scientists in order to pursue collaboration with them.
Chapter 5
System Design and Implementation

As was made clear in the previous chapter, the agricultural issues faced by farmers in Karnataka illustrate the use that ne-grained environmental data can have in this context. In this chapter, we undertake the task of designing and implementing a system that will provide the necessary data granularity. The following sections are based on our own experience, that of the setup of a wireless sensor network in rural Karnataka: COMMON-Sense Net (Community Oriented Monitoring and Management of Natural resources via a SENSor network) [PRP+ 06], [PRS+ 07]. The functionalities described above have two features in common: the usefulness of gathering data over a period of time spanning at least one crop season, and the necessity to collect locally heterogenous data. If we refer to Chapter 3, wireless sensor networks are ideally suited for this type of applications. But it remains to be seen what type of such a network would correspond to the constraints we have. Except for decit irrigation, our use cases do not necessitate the collection of real-time data. However, there is a need to centralize the data, even in the case of decit irrigation, since: The farmer needs to be notied, the soil moisture predictions require the use of historical data and of a general mathematical model, whose operation is beyond the computing capabilities of wireless sensors. A pure data collection model seems hence the most appropriate. In this model, data are generated locally but processed globally. Nodes do not analyze the data they collect, they transmit them to a central server, where they are stored and processed.
5.1
Design Options
Even with the data collection model in mind, there remain two main design questions to be solved. Will the data be produced periodically or on demand (i.e. in response to events or explicit queries)? And will the data be transmitted on the ight, or will we use a store-and-forward strategy?
5.1.1
Data Generation Strategy
Data can be generated periodically or as a response to an event, which can be human-generated typically the explicit emission of queries to the network or environment-based for instance when a parameter reaches or exceeds a threshold. This latter model is the most reactive, and allows to capture on the 61
62
CHAPTER 5. SYSTEM DESIGN AND IMPLEMENTATION
ight important events, whose freshness is important for the proper operation of the system. Both eventbased models offer the best energy consumption prole, because radio resources are used only when it is considered necessary by the application. However, in our case, two constraints call for the use of a periodic model. Firstly, data must be collected periodically in order to feed the crop models, to assess continuously the effectiveness of water conservation measures, or to run predictive scenarios about the soil moisture level. Secondly, the data requirements are still unclear, because as it was mentioned in the previous chapter, the models that are used today take into account the limitation of the technology at the time they were developed. As a consequence, agricultural scientists are curious to observe ne-grained data in order to determine what level of granularity is signicant. Accordingly, we seek to generate as much data as possible, while not compromising the lifetime of the network, so that it remains operational throughout a full season at the minimum. For the decit irrigation use case, a hybrid strategy will be followed: a periodic data collection model, with a variable rate of emission depending on the data variability at the time. For instance, it is not necessary to collect soil moisture data at more than one sample per hour, or even per day, when no rain is falling. However, when water is brought in either by precipitation or irrigation, a ner resolution might be desirable.
5.1.2
Data Transport Strategy
We envision networks spanning several hectares. With the communication capabilities of wireless sensors, this means that a direct connection to a single base station is unfeasible. On the other hand, a pure ad-hoc multihop network, where full connectivity is required, would not be adapted to our case, because we require the exibility of instrumenting patches that can be distant from each other. As a consequence, we need to resort to a hybrid sensor network. The term was rst coined by Sharma and Mazumdar in 2005 [SM05]. The authors refer to a hybrid sensor network as to a wireless sensor network where the use of limited infrastructure, in the form of wires, is designed to improve the energy efciency of the system. Our concern is more the sparseness of the network, although energy gains should not be overlooked. In our case, however, a wired infrastructure would be impractical, due to the distances between nodes, and to the intrusiveness of such deployments in the elds. Accordingly, we expand the notion of hybrid network to an infrastructure, where mutually disconnected clusters of sensors communicate with each other via an alternative technology. 1. IEEE 802.11 bridge: this option, although attractive in a rural setting with partial connectivity, suffers from distance limitations. Emerging technologies, such as WiMax, might overcome this limitation in the near future. 2. Cellular network bridge: This solution is appealing because the use of a global telephony network allows global connectivity. Its major drawback is the cost of communication, which can be mitigated if we aggregate data packets at each cluster-sink and send them in bulk (to the expense of some responsiveness) 3. Store-and-forward strategy: Also known as delay-tolerant networking [HF04], this option means that the cluster-sink stores the data until a mobile agent, also know as a data mule [JSS05], comes into communication range, at which point the data is downloaded to the agent. The data are then transported to the central server, where they are downloaded again. This solution has the advantage of cost, both in terms of energy and money, to the expense of responsiveness. In our case, the data mule would be a farmer equipped with a hand-held device.
5.2. DESIGN CHOICE: OVERVIEW
63
Finally, we chose the solution of the wireless bridge in order not to compromise the implementation of the decit irrigation use-case, which necessitates timely measurements. We rst used a WiFi bridge, because GSM connectivity was poor in our deployment area, but moved to GPRS as soon as base stations were deployed, because the range limitations of 802.11 proved to be severe.
5.2
Design Choice: Overview
This section and the following ones present the ongoing design and implementation of COMMON-Sense Net (CSN), a decision support system for resource-poor farmers using the wireless sensor networks technology for environment monitoring. The system design is as shown in Figure 5.1. This corresponds to a logical architecture summarized in Figure 5.3 for the 802.11 architecture, and in Figure 5.4 for its GPRS equivalent. On the left is depicted the data collection subsystem. This subsystem contains the wireless sensors, to which the embedded probes are attached, and which self-organize into a multi-hop data collection tree. This network is designed to be deployed without conguration, the only restriction concerning the connectivity between the nodes, which depends on the radio used. Each network can serve an homogenous agricultural patch, or a series of contiguous patches. Because deployment locations can be distant from each others, we introduce a data transit subsystem connecting each cluster from a data collection subsystem to the centralized server. The data transit connection is typically wireless, but with a more powerful radio than the typical sensing nodes contain (Wi-Fi bridge) or relying on the cellular telephony infrastructure, if available at the deployment location (GPRS bridge). The node, on which the data transit bridge is installed (the base station - or sink - of the corresponding cluster), needs a complementary source of electrical power. On the server side, one nds rst the data logging and network management subsystem. This component is responsible for listening to the packets coming from the nodes via the data transit subsystem, to interpret them, to log them into the database, and to rely commands that a user could want to issue to one or several clusters. The data processing subsystem is envisioned as a modular component, where the processing of each use case can be implemented. For the rst phase, this subsystem contains only data visualization and upload modules. Finally, the data access subsystem, a web-based graphical user interface, can be collocated with the network server, or installed on another machine. The data actuation part, is left open for future developments. We detail all the sub-systems in the following subsections.
5.3
Embedded Probes
For meteorological parameters, CSN uses a battery of sensors for temperature and humidity (Sensirion SHT11), ambient light (TAOS TSL2550D), and barometric pressure (Intersema MS5534AM). In the absence of a microclimate, such parameters do not vary signicantly over the deployment area, so only 2 weather stations are deployed, for redundancy and detection of measurement drifts. Soil moisture is a parameter of higher variability. We chose to equip several sensors with two ECH2O probes each. CSN does not measure solar radiation at this point, although this should be included in the near future, as it is a major input for predicting the productivity of the crop. The Leaf Area Index (LAI) based on the intercepted radiation provides information on the useful biomass of the crop and thus its yield.
64
Figure 5.1: System overview
5.4
Wireless Sensor Network: Data Collection Subsystem
CSN uses a centralized data-collection model, where individual wireless sensor nodes perform minimal data processing and send back the data via a base station (a node connected to a computer) to a single server where they are processed. Since neighboring nodes of the network can be more than hundred meters apart, a majority of them are unable to reach the base station directly. They have to resort to multi-hop transmissions, where nodes can relay data from other nodes in addition to sending their own. That means that every node in the network can perform three tasks: collecting data, sending them towards the base station, and, if needed, relay data sent by other nodes. The chosen embedded operating system is TinyOS [tos], because it is widely used by the scientic community, quickly becoming a de facto standard. Moreover, this operating system makes libraries of components readily available, such as Medium Access layer, and multihop routing. As for routing, since there is no mobility in the network, and since topology changes are rare (node failure, occasional moving or addition of a node), CSN uses a simple tree construction algorithm, based on neighboring radio links quality and hop-counts to the base station. There are two main issues affecting the platform choice for the wireless sensors. The rst is radio range. Given the data variability and sparse density of the network, a range of more than 100 meters is mandatory, and up to one km is desirable. The second important issue is the power consumption, although this characteristic can be mitigated by an appropriate power management scheme such as duty cycling. Ideally, the nodes have to perform autonomously for the duration of the cropping season (roughly 6 months), either on batteries or with a small solar panel. Given all these considerations, the most adapted platform available in late 2004 (when the initial choice was made) was the Mica2 mote manufactured by Crossbow, because its power consumption is reasonably low, and its radio range is the highest among candidate technologies. The short range of Zigbee and Bluetooth radios disqualied them, and technologies such as 802.11 did not match the power consumption requirements. Still, the radio range of Mica2 is sometimes stretched. Tests conducted in typical landscapes of the deployment area indicated a higher bound of 100 meters in the best case with quarter wave antennas connected to a ground plane. In 2006, we moved to another platform, TinyNode [tin] by Shocksh, whose range we tested to be at least two times better, without compromising on the power consumption. The next two subsections explain in details the platform choices.
5.4. WIRELESS SENSOR NETWORK: DATA COLLECTION SUBSYSTEM
65
30
60
90
120
150
180
210
240
Distance (meters)
Figure 5.2: Mica2 range measurements in the eld
5.4.1
Radio Range
We carried out range measurements on mica2 motes in CKPura [PRS+ 07] in Spring 2005. Since the proposed deployment has to work in a multihop paradigm, it becomes imperative to determine the usable radio communication distance between two sensors. The experiments were conducted in an open plain land with antennae placed at a height of 1.2 meters from the ground level. Transmit power levels were set to -10, 0 and 10 dBm. The data rates at each power level were 1, 5 and 10kbps. A packet transmission consisted of 8 bytes of preamble, 2 bytes of SYNC and 34 bytes of data. Fig. 5.2 shows the results of the achieved communication range for a radio sensitivity of -100dBm. Using a square ground plane with edge length equal to one quarter of the carrier wavelength ( ) and 4 transmit power of +10dBm, one may deploy sensors up to about 200m. Once the data collection networks were deployed in the eld, it became obvious that this communication range was too limited for the kind of deployments we were looking at (see Chapter 6). In the meanwhile, a new platform of wireless sensors had surfaced, the tinynodes[tin], with an advertised range of up to 1km, making it the longest range attainable by commercially available wireless sensors. We made comparative experiments between tinynodes and mica2 in Fall 2005. The tests were carried out in a at and open eld devoid of obstacles (a soccer eld), in full Line of Sight, and at different heights above the ground. Bidirectional connectivity of nodes was assessed using a Ping-Pong application. The settings were chosen as follows: a transmission power of 0dBm , a bit rate of 19.2kbps. These settings were chosen because they represent the higher bound for Mica2 performance. It is to be noted that tinynodes can go up to 15dBm and 255kbps. We also tested in parallel the 802.14 micaZ nodes, although their performance showed that they are not adapted to outdoor sparse deployments. We observed connectivity by looking at blinking LEDs. The range was determined as the distance for which at least 80% of the packets were received1 . It is to be noted that when this approximate value is reached, the packet delivery quickly drops to 0. These tests were not as thorough as the ones performed
1
measured perceptually by counting the LEDS blinks
66
Height 0m 1m 2m
MicaZ unreliable 35m 80m
Mica2 unreliable 110m 150m
Tinynode unreliable 275m 280m
Table 5.1: Range measurements (indicative): micaZ, mica2, tinynode
State Tinynode Mica2
Sleep 4 A < 15 A
Active 2 mA 8 mA
Radio Receive 16 mA 18 mA
Radio Transmit (5dBm) 32 mA 27 mA
Table 5.2: Tinynode and Mica2 respective current consumptions in typical states
initially on mica2. However, they concluded to an average improvement by a factor close to 2 for the tinynode compared with mica2 (see Table ). Another advantage of tinynodes is their ability to operate at higher transmission power, up to 15 dBm. As a consequence, we chose at this point to migrate to this platform.
5.4.2
Power Consumption
Mica2 and Tinynode have similar power proles in active state, and for radio transmit or receive states. However, Tinynode performs better in active and sleep states, as shown in Table 5.2. With a low datarate application such as is the case with COMMON-Sense Net, one might expect the idle-state energy consumption to become predominant. However, this depends on the efciency of the power management schemes implemented at the nodes level, in particular the duty-cycling algorithm that allows the node to go to sleep and save on all its radio and CPU resources whenever they are not needed. Such a performance depends particularly on the efciency of the Medium Access Control layer, since radio operations consume most of the energy (we observed that at a 5 minute sampling rate, probes power consumption became negligible). We explain the options available in the next two paragraphs. B-MAC To achieve low power operation, B-MAC[PHC04] employs an adaptive preamble sampling scheme to reduce duty cycle and minimize idle listening. When idle, nodes go to sleep in an asynchronous way, for a xed period of time - around 100ms at the lowest duty cycle. They wake up periodically to listen to the radio channel or send their own packet. Since there is no schedule agreement between the nodes, they prex each packet with a preamble whose duration is the same as the sleeping period 271 bytes at the lowest duty cycle. This scheme allows the nodes to reach low duty cycle without the need to send extra monitoring trafc over the channel. A signicant drawback, however, is the increase in emission energy, as well as the energy wasted at the reception end in overhearing messages that are not addressed to the node itself (since a node needs to wait for the whole preamble before reading the packet header). A later optimization of the scheme includes the message destination in the preamble, so that nodes can go back to sleep immediately if a message is not addressed to them. Dozer In order to maximize power efciency, Dozer [BvRW07] merges the MAC and the routing layers of the communication stack. First, it establishes a tree structure on top of the physical network. In its bootstrap phase, a node tries to join the tree as quickly as possible. It starts listening for beacon messages that are sent periodically by all nodes belonging to the network. The node then chooses its
5.5. HYBRID NETWORK: DATA TRANSIT SUBSYSTEM
67
potential best parent based on its distance to the sink (in number of hops) as well as its load (the number of the nodes direct children). The actual connection setup is initiated after the transmission of the next beacon of the selected neighbor. After sending its beacon the potential parent stays in receive mode for a short amount of time. The prospective parent will send the new node its TDMA slot. Beacons are used to maintain slot synchronization. Comparison Both B-MAC and Dozer are designed to enable low duty cycles. Typically, they both advertise up to 99% of sleep time. For a given platform, as we have seen, the power consumption of different states is well dened (Table 5.2). The difference in power consumption could come from the overhead induced by the signalling messages, in particular synchronization messages. With this regard, both B-MAC and Dozer have minimal overhead, B-MAC because it is asynchronous by essence, Dozer because it leverages the routing beacons to keep synchronization once the schedules have been established (that is, in regular operation mode). The only difference that can be observed between Dozer and B-MAC is the duration of transmission and reception. Since more recent B-MAC extensions managed to reduce overhearing (of long preambles from packets addressed to another node), the power consumed in reception is not itself really discriminant. At the transmission side, however, the difference is marked. Dozer sends normal TinyOS packets that are 36 bytes long, while at at duty cycle of 1%, the frames transmitted by B-MAC are 2160 bytes long. This difference can be observed at the oscilloscope. While a packet transmitted by Dozer takes 5ms, a preamble plus packet sent by B-MAC takes about 350ms. We consider a typically low data emission rate of one message every 120 seconds. With the values written down in Table 5.2, that yields: Current consumption of idle listening: 0, 99 7A 7A = 160A Current consumption of reception: 0, 01 16mA = Current consumption of emission with Dozer: Current consumption of emission with
532mA 1201000 = 1.3A B-MAC: 35032mA 90A 1201000 =
As a result, one can estimate the average current consumption of Dozer to be around 160A, while the same for B-MAC is around 250A. Hence, Dozer would be 20% more efcient than B-MAC. Accordingly, we chose Dozer as our MAC protocol. What really matters, however, is the performance in the eld, which has to be experienced rst hand. We will reect on effective power consumption in Chapter 6.
5.5
Hybrid Network: Data Transit Subsystem
In order to interconnect disconnected patches to one single server for data logging and network management, CSN makes use of bridges between individual network clusters. Unlike individual sensor nodes, those bridges are connected to the power grid via electric poles that can be found regularly in the deployment area. They are not power-constrained, and expand signicantly the scalability of the network. The rst solution that we investigated makes use of classical 802.11 access points and a rugged PC for the bridge (see Fig. 5.3). This solution is both expensive and power hungry. GSM connectivity, which was not satisfactory at the time of the deployment, since improved considerably in the region of CKPura in 2006. Accordingly, we implemented a GPRS bridge that aggregates and transmits the data directly to the central server located at the Indian Institute of Science (IISc) in Bangalore (see Fig. 5.4). This work
68
Figure 5.3: System architecture: 802.11 bridge
is described with all its technical core details in [Sta07]. The next paragraphs provide an overview of the two systems.
Figure 5.4: System architecture: GPRS bridge
5.5.1
WiFi Bridge
The system consists of several network clusters, each with their own Base Station (BS) that communicates with a local server (LS) over 802.11 (or WiFi). Data packets from the sensor nodes are transmitted in a multi-hop fashion to the BS, which gathers data packets from the sensor nodes and transfers them over a WiFi link to the LS, where it is temporarily stored. From time to time, the data is retrieved by the central server (CS) over a dial-up connection, and stored into a database. Base Station The BS consists of three hardware elements: A base-station node (BSN), an 802.11 access point and a Single Board Computer (SBC). The SBC is a simple, Linux-based computer equipped with an Intel processing unit, USB, serial and ethernet connectivity. The FDC unit is mounted on an metal pole and connected to the electricity network. The power control section of the FDC includes a voltage converter, a charger and a 12V, 7 ampere-hour battery. The voltage converter converts the 12V
5.5. HYBRID NETWORK: DATA TRANSIT SUBSYSTEM
69
from the battery source to three different output voltages (5V, 6.5V, 12V) required for the BS, the SBC and the access point. The battery is charged from mains during the periods when electricity from the grid is available. The BSN is connected to the serial port of the SBC. The ethernet port of the SBC is connected to the ethernet port of the access point. Local Server The LS in CKPura village acts as an intermediate data storage server and offers local access to the sensor data. It runs a dial-in PPP server over a telephone link. An external modem and an access point with external 13dBi omnidirectional antenna are connected to the server. A UPS power supply is also deployed to bridge frequent power cuts in the village. The LS runs a periodic scheduler that fetches the log les stored on the BS. This is only possible during the periods when there is power available in the village. Central Server The central server is connected to an external modem for telephone link connectivity to the LS at CKPura village. The central server periodically dials to the LS (usually once a day), fetches the logged data and updates the central database.
5.5.2
GPRS Bridge
General Packet Radio Service (GPRS) is a mobile, packet switched data service available to users of GSM mobile phones. GPRS uses network resources and bandwidth only during data transmission. It is therefore well suited for a range of applications that typically require bulky data transfer that occur in bursts. Examples of such applications are mobile Internet or location-based services. Since GPRS is packet-switched, networks are loaded more efciently if the data stream is very irregular. With GPRS, subscribers are charged for the amount of data they transmit and not for the duration of the connection. Packet-switched GPRS operates alongside existing circuit-switched services in mobile networks. Fig. 5.4 shows the COMMON-Sense system using the GPRS bridge. The system consists in the sensor network cluster, the Shocksh MamaBoard and the central server. The sensors are represented as a small cluster on the left side of the schema. Data packets, which are generated by the sensor nodes, are sent in a multi-hop fashion to the basestation node on the MamaBoard. The TC65 module sends the packets over GPRS to the central server. On the server, the packets are stored in a database. The data is then available for analysis and consultation over the Internet. The MamaBoard forms the link between the sensor network and the GPRS system. It serves as a base station for the sensor network and as a gateway to the GPRS cellular network. On the server, the same software is running as on the WiFi server. Only a few modications had to be made to the software in order to receive the GPRS connections from the wireless module. We have selected Airtel India as our cellular operator. Airtel offers the broadest coverage of the GSM operators in India and namely has good coverage at the test site in CKPura village. Airtel offers atrate connection for unlimited data transfer over GPRS. Hardware Overview Fig. 5.5 depicts the Mamaboard and its components. The MamaBoard from Shocksh combines a TinyNode Standard Extension Board (SED) and a cellular GPRS module on a single device. The MamaBoard is intended to bridge a wireless sensor network to wired ethernet (LAN), WLAN or GPRS. Each connectivity type is enabled by plugging an appropriate external module to the MamaBoard. Hence, it offers a wide variety of connectivity choices. Furthermore, the MamaBoard provides a slot for SD memory cards. This can be used as a storage buffer, which allows to gather large
70
amounts of data on the MamaBoard, before sending them to an adjacent network. The SD slot can be accessed both by the TinyNode and the GPRS module. The Siemens TC65 is the GPRS module that can be mounted on the Mamaboard. It includes a simplied version of the J2ME (Java 2 Mobile Environment) [24]. Hence, it is possible to execute standard Java applications on the GPRS module. Furthermore, the TC65 comes with an integrated TCP/IP stack. This allows to establish standard Java socket connections to a server by using AT commands2 . The TC65 is connected to the MamaBoard via a 80 pin board-to-board connector.
Figure 5.5: Mamaboard equipped with Siemens TC65 wireless module. 1. TinyNode, 2. GPRS antenna, 3. SD card slot, 4. TC65, 5. SIM card
5.5.3
Performance Evaluation
Major advantages of the GPRS bridge are simplicity, energy efciency and deployment exibility. The system design is simpler than with the WiFi system. The BS runs a simplied stack, since it only needs to feed the GPRS module with packets to send. The LS is no longer necessary, neither is the dial-up link to the CS. The MamaBoard is the only device required to establish communication from the eld to the central server. The two factors mentioned above, simplicity and energy efciency, open up new possibilities for deployment of the network clusters. Setting up the network is faster, less laborious and easy to debug. Ideally, the GPRS enabled basestation can even run on battery power for reasonable amounts of time. In this way, the clusters become completely independent of a power source and could therefore be deployed even in regions which are out of reach of other systems. Such an architecture already approaches the classical concept of a sensor network which operates in an independent, self-organized manner. Finally, it also solves the problem of the power cuts, which caused many problems in the WiFi system. However, there is one important drawback of the GPRS system. The command link has a large delay. Commands issued are transmitted to the network only once the MamaBoard sets up a GPRS connection to the central server. Other solutions to minimize this delay have been studied, but were not implemented. GPRS also introduces an ongoing operating cost. Right now, the weekly cost is INR 99.- (or USD 2.30) per cluster. In networks with many clusters, an intermediate aggregation layer could be used to gather data from several clusters on one GPRS Base Station to keep the operating cost low. Finally,
AT (for attention) commands are the components of the Hayes command set, a specic command-language for modems. The command set consists of a series of short strings which combine together to produce complete commands for operations such as dialing, hanging up, and changing the parameters of the connection (denition from Wikipedia).
2
5.6. DATA MANAGEMENT AND PROCESSING
71
network coverage at the deployment sites can be a limitation for the GPRS system. While GPRS services are nowadays widely available in urban areas throughout India, it remains a limiting factor in rural areas.
5.6
Data Management and Processing
The collected data is logged into a PostGreSQL database, whose (simplied) structure is shown in Fig. 5.6.
Figure 5.6: Database Structure
The essential features of the database are the following: 1. Login-based access to the application. 2. Several deployments can be managed jointly. 3. Clusters are regrouped by geographical location. 4. Data can be retrieved at any time based on the geographical location of the network, its identier and time parameters. 5. A user can dene an experiment, selecting one or several nodes from different clusters without geographical constraint.
5.6.1
Tables
Log Command and Log Query This table is used to log queries and command messages sent to the network. The command is stored as a string, with its associated address, value and cluster id. Cluster This table is used to store information about clusters. A cluster usually refers to a physical instance of a sensor network. A cluster is associated with the name of a data table (which are created
72
dynamically) and a unique group id . Duplicates for data tables and group ids must be looked for. Group ids are unique but reusable. Data table names are unique. Cluster Group This table is used to regroup clusters by geographical location. Mote Info This table is used to store information about motes, such as their associated cluster id, their location, and the platform type. Data xxxx This table stores the data received from a particular node id over time. Each sensor has its own data table. This is necessary because of the size that an aggregated table would have. Given the amount of sensor data that is stored into the database (more than 700 per node and per day), this would make the data access time prohibitive, especially for interactive display. Experiment This table is to store information about application groups, namely individual experiments conducted on a set or subset of nodes from one or several clusters. Experiment Node This is a joint table to associate a mote to an application group. Routing Parent This table is used to store parent statistics over time. For the moment, the data can be viewed based on their type in different graphs, and may be downloaded as well. There is no integrated data processing at the moment. We will see in the chapter about System Assessment why we had to review our ambitions on that side.
5.7
A Web-based Tool
A proprietary java front-end, based on an original design by the sensorscope group [SDFV], is used to send commands to the wireless network and to log data and meta-data into a database, from which they are extracted for display and processing. The java front-end is also used to send commands and queries to the network (such as transmission power and radio channels change etc.) The system contains a web-based interface for the display and upload of data. This interface has two roles. Firstly to allow a user to select clusters and obtain the relevant data, either in order to display or to download them for further analysis. Secondly, to allow an administrator to monitor the state of any cluster, and to issue commands to the corresponding networks.
Figure 5.7: Web application: home page
From the home page (Fig. 5.7), one can choose 4 different functionalities: Displaying environmental data by cluster or experiment
5.7. A WEB-BASED TOOL
73
Looking at the network statistics Sending a command to the network Learning more about the project (about)
5.7.1
Data Display
Figure 5.8: Web application: network selection page
Figure 5.9: Web application: display of data
From the Environmental Data tab, one can select a cluster group, then a cluster. Once the cluster is selected, the map is displayed (Fig. 5.8). The map has several interactive features: 1. On mouse-over a semi-transparent window allows to display details about the corresponding node. 2. The nodes can be selected by a mouse click.
74
3. On the menu in the right-hand side, one can choose to show the last value for a given type of sensor. 4. Finally, one can open a window allowing to display the graphs. The window used to select the graphs to display (Fig. 5.9) also has dynamic content. Using the righthand menu, the user rst needs to select one or several sensors, then a time period, then whether he/she wants all the graphs one one or several images, then whether he/she wants to display graphs or a data table. Then the left panel of the window will display the data accordingly. The graphs can be zoomed by clicking on the corresponding part of the image. It is to be noted that currently this process can be quite slow, depending on the amount of data that you want displayed on screen. The user can also choose to download the corresponding data sets.
5.7.2
Network Statistics
Figure 5.10: Web application: display of routing tree
One can display the connectivity graph and the routing tree (Fig. 5.10) in the following way: 1. First select the cluster group, then the cluster. 2. Display the network graph. 3. Allows to display the network graph for a given time window.
5.7.3
Commands
The web interface enables to send commands to the nodes (Fig.5.11). A pop-up window appears with the replies. The authorized user rst chooses the type of command and the corresponding parameter (where applicable). Then he chooses the destination node ID. Finally he presses the Submit Query button. A pop-up window appears and a Python application is called within that window. After a few seconds replies from all the nodes (where applicable) are shown in the pop-up window.
5.7. A WEB-BASED TOOL
75
Figure 5.11: Web application: command interface
76
Chapter 6
A Wireless Sensor Network Toolkit for Rural India

This chapter describes work that was carried out between 2005 and 2008 in partnership with the Centre for Electronic Design and Technologies (CEDT) of the Indian Institute of Science (IISc). In particular, most of the eld work was accomplished by a research team led by Prabhakar T.V. of the CEDT. Engineers who worked in the village are : Sujay M.S., Aswath Kumar M., Vinay S. and Amar Sahu. A local team provided us support under the supervision of Mr. P.R. Seshagiri Rao. We proceeded to the rst deployment of our system in India in 2005. This took place on the campus of the Indian Institute of Science, in Bangalore, with Mica2 motes. A second experiment was conducted in early 2007 with the Tinynode platform. In parallel, a test eld has been used in Switzerland (at the Changins agronomic station) from 2006 (rst tests on Tinynodes) to 2007-2008 (rst deployment of a GPRS bridge). The success of these various deployments and the support from the local NGO Chennakeshava Trust encouraged us to go for live deployment in Chennakeshavapura in 2006 (Mica2), and again in the summer of 2007 (Tinynode) (Fig. 6.2 shows a view of the elds from the servers location). Concomitantly, we extended our network deployment in Changins. This chapter details the history of these deployments and the lessons learned in the process.
6.1
Changins
The Changins deployment took place at the agronomic research station of the same name. A map of this deployment is depicted in Fig. 6.1. 23 nodes were deployed from July 2007 to March 2008. The base station of the network, connected to a GPRS bridge, was installed on the third oor of the stations main building, on the south balcony and was connected to a power plug inside the building (node 1 on the map). Other nodes were deployed outside. 10 nodes were installed in the orchard and vineyard located to the south-east of the building1 . 3 nodes were located on a small vine research plot on the south of the building2 . Two nodes were deployed in a vineyard on the north of the building3 , and two more in a nearby orchard4 . Two nodes were set close to an existing meteorological station, further up the hillock
1 2
1200-1207, 1217, 1221, 1232 1226, 1229, 1230 3 1241, 1245 4 1258, 1260
77
78
CHAPTER 6. A WIRELESS SENSOR NETWORK TOOLKIT FOR RURAL INDIA
100 meters
Figure 6.1: Changins test-bed (Switzerland) : 25 nodes deployed for a period of 6 months without interruption
on the north side of the building5 . Finally, we placed two relay nodes on the south and east side of the balcony6 , and one relay node on a pole north-west from the base station7 . The whole area is located on the side of a hillock. There are trees close to the building and in the orchards, and a block of greenhouses on the east side of the main building. This setting allowed us to test the longevity of the batteries in a multi-hop setting. We observed consistently three hops connections in this network, with some temporary topologies going up to 4. A general lesson to be drawn is the unpredictability of the topology. Nodes located 400m from the base station regularly connected directly to it (node 1242 in Fig. 6.1), whereas nodes nearby sometimes used intermediate nodes that are possibly further away from the base station than themselves (e.g. 1200-12061263). An extreme example of this situation is illustrated in Fig. 6.1 by node 1207, which communicates with 1253 on the north-west side of the test-eld, which in turn communicates with 1232 at the southeast side 400m away, which communicates with the base station. 1207 is about 80m away from the base station. We did not investigate this phenomenon in detail, as in our opinion it is linked with the unpredictable nature of the radio channel. In any realistic setting, phenomena such as multi-path effect or fading are
5 6
1258, 1260 1253, 1263 7 1256
6.2. CHENNAKESHAVAPURA
79
Figure 6.2: CKPura deployment: view of deployment eld from the central server roof
extremely hard to predict because of the difculty in building an exact model of the environment. We used the results obtained in Changins to analyze the performance of the system in terms of throughput and lifetime. We present these results in Sections 6.3.3 and 6.3.6.
6.2
Chennakeshavapura
The rst eld deployment in India was carried out in 2006, for which we obtained data throughout the whole year. The platform used at that time was Mica2, and the clusters were connected to the central server through a 802.11 bridge. The total number of nodes deployed during this period was 18. 10 nodes remained active throughout the deployment period. Figure 6.3 details the settings of this deployment consisting of 2 separated clusters, (note: the bodies of water indicated on the map are dry most of the year), from which the network has already collected a wealth of data that were used in three ways: to validate the data collected by the different probes; to assess the performance of the network in terms of range, lifetime and connectivity; and to test and rene the design. A second deployment, this time with tinynode hardware, took place in the summer of 2007. It consisted of 7 nodes. These deployments are sparse. The average distance between the nodes is 150 meters for the Mica2 stations, and 300 meters for the Tinynode stations. We deployed only one weather station containing temperature, humidity and pressure sensors (Fig. 6.5). All the other nodes were equipped solely with ECH2O probes (Fig. 6.4). By default, two probes were deployed at each measurement point (i.e. by wireless sensor), at two different depths : 10 cm and 30 cm.
80
Figure 6.3: Chennakeshava initial deployment
This was done in order to assess the water needs of a plant at two depths of its root zone. It is to be noted that for subsequent deployments, the depth of the sensors for a given plot should depend on the type of crop cultivated at this location. The base stations of each cluster were connected to the electrical grid, and also contained a rechargeable battery to cope with power cuts (Fig. 6.6). The Chennakeshavapura deployment is particularly illustrative of the difculties that remain after technical design and lab testing have given satisfactory results.
6.3
6.3.1
Issues of a Rural Deployment

Hardware Issues
Memory corruption of motes contributes to the overall unreliability of the system. The experience in live deployment with large leaf canopy cover resulted in unpredictable node ID changes in at least 3 occasions. We also experienced a complete freezing of nodes in the eld deployment at CKpura. The node ID change is mostly a one or two bit ips in the node ID eld structure. Although the node ID may be brought back to its original value by a software reboot of the running code, a node freeze proved to be a corruption of the ash memory. We suspect high package temperatures to be the cause for the ash corruption seen in the eld deployment.
6.3. ISSUES OF A RURAL DEPLOYMENT
81
Figure 6.4: CKPura deployment: typical node in weather-proof package, powered by two AA batteries
6.3.2
6.3.2.1
Probe Assessment
Climatic Probes
The results obtained from the sensor network deployed at CEDT were compared to benchmark measurements from the Center of Atmospheric and Oceanic studies (CAOS) from the Indian Institute of Science in Bangalore, in order to see if the trend matched. As shown in Figure 6.7, the results for temperature are an exact match. The same result holds for humidity, which uses the same Sensirion SHT11 probe (see Fig. 6.8). 6.3.2.2 ECH2O
We validated the soil moisture readings indirectly by superposing them with rainfall data (Fig. 6.9). As one can see, the trend clearly matches. Rain falls of late May, when the soil water content is low, provoked a sharp increase in soil moisture. A similar phenomenon can be observed in mid-July and on September 30th. On the other hand, important rain falls of late August only stabilized the soil moisture at a high value, since the water content at that time was probably close to eld capacity (around 25%).
82
Figure 6.5: CKPura deployment: Temperature, pressure and humidity micro-station
However, the measurements appeared to be noisier than hoped, although they remain in the 5 % range specied in the ECH2O user manual. This problem can be solved by averaging over a larger number of samples (which is what is done a traditional data logger), but this increases the power consumption and decrease the lifetime signicantly. Instead, when we moved to the Tinynode platform, we designed in collaboration with Shocksh a new data acquisition board ltering out high frequency signal variations. This proved to reduce signicantly the effect of noise, while not compromising on the accuracy, as shown in the next paragraphs. The laboratory of Hydrologogy at EPFL (HYDRAM) assessed the use of ECH2O probes through a wireless sensor. In order to do so, 6 samples of soils were gathered from the Changins test-site at different locations. Two more samples of sand were added to the lot. These samples were fully dried and put into pots, where soil moisture probes were inserted. During two months, precise amounts of water were added to the pots, while evaporation was prevented with the use of aluminum sheets. The ECH2O probes measurements were then compared with punctual gravimetric measurements. The details of the methodology can be found in [Rou08]. The results obtained during these tests are summarized in Table 6.1 for a probes default calibration as provided by the manufacturer, and in Table 6.2 for a soil specic calibration based on the gravimetric method. Based on these data, the authors conclusion (as can be found - in French - in [Rou08]) was the following: The performed measures led to a result which is overall less satisfactory than the benchmark provided by the manufacturer. The observed drift between the expected precision and the experimental
83
Figure 6.6: CKPura deployment: base station box containing linux box for 802.11 connectivity, wireless sensor and battery
Changins 1 0.0620 0.0378 0.1313 Changins 2 0.0518 0.0280 0.0993 Changins 3 0.0666 0.0574 0.2050 Changins 4 0.0679 0.0456 0.2030 Changins 5 0.0362 0.0274 0.1057 Changins 6 0.0565 0.0259 0.1069 Sand 1 0.0604 0.0158 0.1038 Sand 2 0.0470 0.0197 0.0938
Average Sigma Maximum
Table 6.1: ECH2O and Data Acquisition Board Assessment over Wireless: Precision of measurement with default calibration
results, however, remain small and localized. As a consequence, they can be considered to be acceptable depending on the intended use of the probes. (...) The zone of inuence, i.e. the volume of the soil representative of the measured water content, plays an important role for soil moisture probes in general, and in particular for our experiment. According to the manufacturers documentation, this zone covers a width of 2cm along the probe. The remaining uncertainty linked with the homogeneity of the nonmanipulated soil emphasizes the importance of this zone. An extended experiment, as well as trials on various soil volumes and structures should be considered (...). The ECH2O probe that was tested in our framework seems to have a particularly interesting future. It needs to be put into its context of use. As a matter of fact, this tool must not be seen as a precision tool targeted at laboratory research, but as a eld-tool intended for long-term outdoor measures. 6.3.2.3 Watermark
Although our experience with the ECH2O probes is so far satisfactory, there are some concerns about the usability of the measurement method itself in the context of agriculture. As a matter of fact, the volumetric water content of the soil is independent of soil physics. However, the capacity of a plants
84
Figure 6.7: Temperature: Performance of SHT11

Changins 1 0.0310 0.0176 0.0842 Changins 2 0.0357 0.0326 0.1218 Changins 3 0.0344 0.0242 0.1453 Changins 4 0.0245 0.0159 0.0976 Changins 5 0.0377 0.0254 0.0969 Changins 6 0.0233 0.0197 0.0926 Sand 1 0.0367 0.0164 0.0970 Sand 2 0.0413 0.0159 0.0995
Average Sigma Maximum
Table 6.2: ECH2O and Data Acquisition Board Assessment over Wireless: Precision of measurement with soil-specic calibration
roots to draw water is conditioned by the soil composition. This means that with ECH2O probes, a proper soil calibration might be needed in order to get relevant information for agriculture. There exist, however, probes that assess the suction potential directly. Watermark probes are an example. These inexpensive probes use a resistor embedded into a semi-porous material that mimics the action of the root itself. With this method, the result obtained is independent of the soil type. The interfacing of such probes is a problem, however. The probe manufacturer is not keen on disclosing its technical sheet, because it interfaces the probes with data loggers itself. There exists a chip on the market, the SMX [Sys], that translates the resistance into a voltage that can be used with an appropriate table of values. CEDT is currently performing tests on the watermark probe and its usage with the COMMON-Sense Net data acquisition board. 6.3.2.4 Field Data
The data that were collected on the test-bed at Chennakeshavapura (Figs. 6.10 and ??) were used in the user experiment described in Chapter 8. Being presented to the participants via the Web application de-
85
Figure 6.8: Humidity: Performance of SHT11
scribed in the previous chapter, they were used for the denition of use cases and for a general reection on the role and capabilities of environmental monitoring via Wireless Sensor Networks. However, the processing of these data for the purpose of rain-fed farming is beyond the scope of this dissertation. At this time, we will have to perform further experiments in a controlled environment with the help of agricultural scientists to put these data to a concrete use.
6.3.3
Power Management
Power management issues are a critical aspect of wireless sensor network deployments. In our initial tests with B-MAC [PHC04] and mica2 motes, we found that the lifetime of nodes in the eld was on average no more than one month, even with the lowest obtainable duty cycle of 1%. As a consequence, we moved to a new MAC protocol known as Dozer [BvRW07] when we changed the platform. This protocol held promises for improvement, because it uses nodes synchronization to improve power efciency. At this point, an explanation of the two protocols underlying principles is useful. The results obtained with Dozer in our eld deployment in Switzerland were very encouraging. After 7 months of deployment with a data period of 2 minutes for each node all the nodes were still alive, with only a slight degradation in the average voltage. Fig. 6.12 displays this evolution from August 19th, 2007 to February 19th, 2008. The lower limit for proper operation of the system is 2.8V. To the exception of 3 nodes, which started malfunctioning in February, the nodes stock of energy remained remarkably stable. It is impossible to extrapolate the nodes lifetime with this graph, because the operation of a lithium battery is known to be non-linear. However, the theoretical lifetime of several years is credible at this time. However, the results obtained in CKPura were very different. There, an average lifetime of two
86
Figure 6.9: Soil moisture: Correlation of ECH2O with rain fall
weeks was observed in preliminary tests. A software problem is one possible explanation, should some application parameters differ in one implementation and the other. Another possibility is the role played by particularly challenging conditions, in particular distance between nodes, which might cause frequent losses of synchronization and increase the nodes power consumption signicantly. Indeed, the distance between the nodes is in average 200-300m in CKPura, while in Changins nodes are at most 100m form their nearest neighbor. Failed links might indeed cause numerous retransmissions, exchange of resynchronization packets and undue active-radio time. CEDT is currently performing tests on these eventualities, until the boxes used in Changins can be used in CKPura.
6.3.4
Environment
This aspect is especially critical for the packaging issues. In this subsection, we explain why we chose to develop self-contained light-weight stations comprising only soil moisture probes. Such a model is quite unusual in environment monitoring, where people usually try to integrate as many probes as possible into a single station. First comes the issue of temperature. In the semi-arid region the ambient temperature can go over 37C. Packaged electronic systems can experience even higher levels. Semiconductor memories, microcontrollers and radio devices should work predictably in these conditions. Then comes the issue of intrusiveness. The hardware must be minimally invasive, so as not to interfere with the normal operation in the eld. It is not practical to deploy bulky weather stations throughout the cropping elds. The sensors should be self-contained in a box that can be xed to existing infrastructure or on simple poles planted wherever needed. To summarize, this means that: 1. The number of probes used within a single sensor should be kept minimal
87
Figure 6.10: Data collected by soil moisture sensors at 15 cm below surface in Chennakeshavapura between September 2006 and February 2007
88
Figure 6.11: Data collected by soil moisture sensors at 30 cm below surface in Chennakeshavapura between September 2006 and February 2007
Measurements for Voltage [mV]
Figure 6.12: Evolution of voltage over time for a 25 nodes network deployed in Changins, Switzerland
89
2. The use of solar panels is not an option, as it requires installing a special infrastructure. Moreover, such panels are prone to theft, as was indicated by the local NGO we work with in the eld [Rao05]. 3. The boxes and hardware must withstand high temperatures and heavy rain. As we focus mostly on soil moisture data (see previous chapter), we opted for a self-powered lightweight design with an enclosure providing a protection index IP67 [IP], meaning total protection against dust and protection against strong jets of water. We adapted to the box a pressure equalizer plug (compatible with the norm IP67), in order to prevent condensation of water. Two membrane cable glands allow for the connection of the cables to the probe wires. This design has a decisive advantage in that it does not impair sensing and is easy to deploy. We powered each box with a 3.6V lithium battery, enough to ensure a 6 months lifetime according to our tests in Changins. One essential argument in favor of WSNs is their low price, expected to drop even more in the years to come. We face an essential tradeoff here: the need to look for packaging that lives up to the standards of environment monitoring, while remaining cheap. Hence a challenge in choosing and adapting off-theshelf boxes so that they withstand temperatures and precipitation strains. In our project, we chose the enclosures and accessories provided by FIBOX [FIB]. These components are listed in Sec. 6.5.
6.3.5
Power and Telecommunications Infrastructure
In rural India, power cuts are frequent, often happening every day and lasting for hours. The design for deployment should consider intermittent power as a major issue. Sometimes, the operation of auxiliary generators is not enough to cope with extended power outage. This is important when one considers the cluster heads that contain Wi-Fi or GPRS equipment (see Fig. 6.13). GPRS has a further advantage, because GSM base stations are not affected by power cuts, as they have their own generator. In this case, we can bypass a local server that would suffer from the power outages as well, and send the data directly to a server located in a urban area, where power is more reliable. This brings the question of cellular coverage in the deployment area. The price of GPRS data has dropped sharply, while the coverage in rural India is progressing steadily [CTT06]. There was no GSM base station close to the Chennakeshavapura village when we proceeded to our initial deployment in 2005. In 2007, several companies had started to operate in the area. As for us, we initially connected our base stations to the regular power grid, via a rechargeable battery, but as we experienced problems with power cuts, we decided to move to a solar powered solution. Sensorscope [SDF] designed a GPRS-compatible station powered by a solar cell. We plan to reuse their design in upcoming releases of the system.
6.3.6
Connectivity Issues
On the bright side, a deployment in a rural region does not have to take into account the same amount of interferences that can be observed in an urban environment. The radio channel can be expected to be more predictable in the absence of tall buildings, metallic infrastructure and heavy trafc. But this picture needs to be nuanced. Terrain in marginal farmers land can have undulations, making Line-of-Sight (LOS) connectivity difcult. Then, there is the issue of vegetation, which grows consistently throughout the cropping season. That means the connectivity will tend to degrade as time passes. Having to cope with a sparse network does not help either. When two nodes get disconnected, there is little other alternative than to deploy an intermediate node as a bridge.
90
Figure 6.13: CKPura deployment: base station with WiFi bridge
We made an in-depth analysis of connectivity from the deployment we made in Changins between August 2007 and February 2008. This deployment represents the most extended data set we have to date with the Dozer protocol (see Section 5.4.2). This connectivity study was based on effective throughput gures obtained throughout the period. As the nodes report data every two minutes continuously, we analyzed the evolution of the Packet Delivery Ratio (PDR, which we also call normalized throughput) over time. Throughput gures in Changins are mixed. Fig. 6.14 shows the evolution of normalized throughput over time, by 24 hours increments. Apart from the network disruption that is observable from October 7 to October 17, it is possible to make the following observations. On one hand, losses happen continuously in the network, and the average PDR stagnates around 55 to 60 %. On the other hand, disconnected nodes always recover and manage to send messages at least every few hours. At this point, the most likely cause is recurring losses of synchronization at the MAC layer (Dozer), which force the nodes to go over the synchronization and schedule establishment procedure over and over again. As the networks topology is heterogenous, we show in Fig. 6.15 the connectivity gures for the nodes located on the north side of the base station. These nodes are positioned on the upper side of the hillock, in small clusters separated by distances from 100m to 400m. We do the same for the south nodes in Fig.6.16. As one can see in Fig. 6.1, the density of the network is higher in this area, and nodes are also closer to the base station. Although connectivity gures are still mixed, one can observe that the
91
Figure 6.14: Throughput in Changins, September 2007 - February 2008 (all nodes)
Figure 6.15: Throughput in Changins, September 2007 - February 2008 (north nodes)
Figure 6.16: Throughput in Changins, September 2007 - February 2008 (south nodes)
92
Figure 6.17: Throughput in Changins, December 2007 (north nodes)
Figure 6.18: Throughput in Changins, December 2007 (south nodes)
south region displays a better behavior. This becomes apparent if we zoom to a period of one month, as is done in Figs. 6.17 and 6.18. The inuence of the vegetation and of the network density is hence perceptible in the connectivity graphs. Not surprisingly, throughput gures are also highly correlated with the distance of a node towards the base station, as illustrated by Fig. 6.19. This gure shows a clear decrease in PDR for nodes directly connected to the base station ( 75%), nodes at two hops ( 50%) and nodes at three hops ( 25%). This is consistent with a cumulative packet error rate over successive links, the only abnomaly being the occasional raise of connectivity at 3 hops over the connectivity at 2 hops when the latter drops below 50%. This can be explained by a change of topology, some of the weakest 2-hop links being replaced by 3-hop links at this point in time. We did not display the 4-hop links because they mostly happen in a transient fashion. Based on these observations, we reached the following conclusions 1. The network is robust, in the sense it is able to function over a period of at least 6 months unattended, and deliver packets consistently during this period. 2. The average PDR, however, is low, which can have several causes:
6.4. HUMAN ISSUES
93
Figure 6.19: Throughput per number of hops in Changins, December 2007
(a) Synchronization issues at the MAC layer, (b) Sparseness of the network, (c) Additive effect on the multi-hop links. The general lessons to be drawn from the connectivity issues that we faced in the eld is the pressing need of an appropriate deployment and maintenance support tool that helps with the deployment of a wireless sensor network. Moreover, such a tool, if it is to be put in the hands of a non-specialist user, has to be intuitive and must not require a priori knowledge of networking.
6.4
Human Issues
In order to achieve the successful implementation of the uses cases described in Chapter 4, the collaboration of the farmers was required to protect the hardware, to report regularly on eld conditions and to give feedback on the added value brought by the technology. Unfortunately, we observed an initial distrust of the population towards the technology and the presence of scientists in the eld. Informal discussions with local stake-holders indicated that the population of small farmers has an experience of being systematically left behind in the innovation process. A second obstacle was the difculty in translating the scientic terminology of environmental science (soil moisture, evapotranspiration, etc.) into the language of the eld. Finally, the current uncertainty about the benet to cost ratio of the technology did not encourage active collaboration. In 2007, CEDT reassessed the situation in an internal project document[Jam]. Informal interviews with some marginal farmers indicated that they are not interested in any project that does not bring them rains, a road to the village, a perennial bore well, or cash (as loans for example) and subsidies.
94
In this assessment, the results of the survey on information needs [RGK+ 04] were not deemed invalid, but considered in the light of the circumstances: 1. Farmers were asked to think about a very specic question, which might bear limited relevance in the face of more immediate problems, in particular the threat of bankruptcy. 2. There is a minority of individuals who recognize the commitment of scientists to improve farming conditions through this project. However, their importance was exaggerated in the survey because they were the ones more willing to participate actively to the survey 3. Farmers tend to have condence in Mr. Sheshagiri Rao (our hybrid), because he is one of them On three occasions, we experienced theft of hardware. Although such events could not be traced with certainty to the local population, the Chennakeshava Trust eventually concluded to the necessity of locking the hardware in protected chambers in order to prevent more stealing. This clearly contradicts the initial goal of using a participatory approach, and also conicts with the exibility required of a light-weight system.
6.5
System Overview
We provide in this section a summary of the components that we used in our toolkit, with useful links to manufacturers whenever appropriate. Operating System: tinyOS-2.x, with proprietary MAC plus Routing protocol Dozer by Shocksh (instructions on how to use it are available with the code). Wireless Sensing Platform: Tinynode [tin] by Shocksh with Linx quarter-wave antennas soldered on the board [Tec]. Data Acquisition Board: Custom design (schematics in Fig. 6.20) with interfaces for two ECH2O and two Watermark sensors, and a battery casing for one Lithium 3.6V battery. Hybrid Network Bridge: GPRS transmitter over shocksh mama-board [tin], necessitating Siemens TC65 GPRS card [Sie]. Currently powered by the electrical grid. Probes: temperature/humidity: Sensirion SHT15 soil moisture: ECH2O [ECH] soil hydric pressure: Watermark [Irra] Enclosure: FIBOX [FIB] Boxes: PC 150/50 LG 2 x PG 16 cable glands 1 x MB 10894 pressure equalizer plug Cables going in the soil need to be protected up to 1 meter above the ground Server side: Web server Apache2 PostGresql database Java servlet and python scripts The full application code is available at http://commonsense.ep.ch
K12
ECHO soil moisture - Probe 1 ADC Voltage Reference

EVREF_EN 1k R9 D5 2.5v 1.5% C4 4.7uF GND GND R1 E_EVREF GND C29 4 3 2 1 Audio Jack 100nF K7 A
VCC VReg GND nRST
6.5. SYSTEM OVERVIEW
0.33-1.32v
V_ECHO1 ECHO1_EN
EVREF_EN nACTION LED_GREEN LED_YELLOW 100 D7 C30 2.5v GND 100nF 1.5% GND GND
V_ECHO2 V_ECHO1 V_WATERM2 V_WATERM1
URXD1 UTXD1 P1.2 P1.3 E_EVREF P6.7 P6.6 P6.5 P6.4 P6.3 P6.2 P5.0 P5.1 P5.2 P5.3 P1.6 P2.3 P2.4 P4.0 P4.1 P5.0 P5.1 P5.2 P5.3 P1.6 P2.3 P2.4 P4.0 P4.1 ECHO1_EN ECHO2_EN WATERM1_EN WATERM2_EN Header
P2.2 P1.1 P1.2 P1.3 E_EVREF P6.7 P6.6 P6.5 P6.4 P6.3 P6.2
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29
2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
ECHO soil moisture - Probe 2 Action Button

VCC GND S2 R8 R32 0 nRST R31 nACTION 10k GND C18 100nF K6
0.33-1.32v
V_ECHO2 ECHO2_EN
4 3 2 1 100 D6 C19 2.5v GND 100nF 1.5% GND GND Audio Jack
Watermark - Probe 1
C9
K4 U4 WATERM1_EN BAT1 R10 100 R17 47k C17 4.7uF LiSo2 LMC555
1 2
4.7uF
R6 150k
Power
R33 C36 4700uF GND C34 33pF C20 100nF C33 4.7uF GND R40 DNF 0 VReg VCC
C10
R21
4.7uF
390
1 2 3 4 GND VCC TRIGGER DIS OUTPUT THRES RESET V_CTRL
8 7 6 5
C V_WATERM1 GND R4 1k
C31 100nF
0.2-1v
Watermark - Probe 2
C5
LEDs
LED_GREEN U3 WATERM2_EN LMC555 R28 220 LED_YELLOW R29 330
D8
Figure 6.20: Data Acquisition Board Schematics

D9 1 2 3 4 GND VCC TRIGGER DIS OUTPUT THRES RESET V_CTRL R7 100 R11 47k C8 4.7uF 8 7 6 5 GND
K2
1 2
4.7uF
R5 150k
C6
R20
4.7uF
390
D GND R3 1k
C22 100nF
0.2-1v
V_WATERM2
Title CommonSense Size A4 Date: File: 22/01/2007 \\..\CommonSense.schdoc 3 Number
(C) 2006 Revision V1.0
Shockfish SA
Sheet 1 of 1 Drawn By: Roger Meier 4
95
96
6.6
Lessons Learned
Deploying a sparse network in a remote and uncontrolled environment raised critical deployment issues. Fluctuations of the radio channel caused by the growth of vegetation during the cropping season had a severe impact on the network connectivity. To diagnose such problems and other software or hardware failures would require a constant presence of communication engineers on the deployment site, which is difcult to achieve in a rural setting. The main lesson to be learned from this deployment, however, is the difculty of implementing and testing convincingly our use cases in the eld. Firstly, stabilizing the system software and hardware took longer than expected, meaning that we started to collect reliable data only in 2007. The volume of collected data is not yet important enough to draw conclusions on the themes identied in Chapter 4. This prevented us from establishing convincing contacts with the marginal farmers, as their attitude towards scientists is one of disbelief unless proven otherwise, as we already identied in Chapter 4. Another reason for the inability to implement use cases yet was the particular situation of communitybased irrigation. Because of several drought occurrences in recent years, community tanks have displayed a low recharge level since the project started. Closer study of irrigation practices also showed that farmers practice submersion irrigation, which consists in ooding the eld at early stages in the cropping season. Such practice would receive little benet from the use of wireless sensors. Currently, marginal farmers consider drip-irrigation as too costly a practice. Experiments conducted in controlled environments will be necessary if we are to prove the protability of drip irrigation to marginal farmers. Both the difculty in tracing the technical problems and the impossibility to create a partnership with farmers represented serious obstacles. This called for a change of paradigm: in our experience, the promise of ubiquitous computing will have to wait for maturation of both the technology and the users before being fullled. Instead, managing the technology in a controlled environment with the participation of committed users can lead to rapid results, provided we can ensure a spill-over effect to the farming population. In the next chapters, we explain 1. How we developed a user friendly original deployment and maintenance support tool, designed to be used by non-network specialists (Chapter 7) 2. Why we decided to focus on scientists working on applied research for rain-fed agriculture, and how we veried the appropriateness of this new approach (Chapter 8).
Chapter 7
Making the Invisible Audible

In this chapter, we draw on lessons learned from our deployment in Chennkeshavapura (see Chapter 6). Deploying a sparse wireless sensor network is not an easy task. Currently, this requires the intervention of networking specialists. One major issue in determining nodes placement is the connectivity between them and its evolution over time. In particular, in an environment characterized by dense vegetation, partial line-of-sight and low network density, deploying sensors requires precisely analyzing the connectivity between nodes while they are being installed. Once the nodes are deployed, specialists are not always available in the eld. We realized that maintaining the network in operational state was a challenging task for the local staff. In particular, diagnosing the status of each node is not trivial in the absence of any embedded display. Some health status can be assessed directly from the server, but if nodes stop responding, it can be difcult to trace down the origin of the failure. In this chapter, we present Sensor-Tune, a light-weight deployment and maintenance support tool for wireless sensor networks. This tool is based on an auditory user interface using sonication. Sonication refers to the use of audio signals (mostly non-speech) to convey information. We explore the potential of this approach, in particular how it allows to overcome the inherent limitations of visual interfaces. We then justify our design choices, and present typical WSN applications where sonication can be particularly useful. Finally, we present the prototype that we built, and describe a user experiment that we conducted in early 2008, which is the rst reported attempt to put a multi-hop wireless sensor network deployment in the hands of non-specialists. We claim that the tool that we developed can be used in a variety of deployments. Accordingly, we extend our focus to any data collection wireless sensor network. However, the reader should keep in mind that a context such as a developing country, where a large part of the population has still limited computer literacy, would particularly benet from such an approach.
7.1
Introduction
It is an experience commonly reported in the literature that deploying a wireless sensor network can be a cumbersome and labor-intensive task [RYR06], [SWC+ 07], [RBE+ 06]. In particular the inuence of the environment on network connectivity is often difcult to diagnose due to the limited display capabilities of wireless sensor nodes. These difculties are exacerbated when the network topology is sparse (for instance WSNs for agriculture), or when the environment is particulary challenging for the radio channel (indoor environment with metallic walls or pipes, etc.) 97
98
CHAPTER 7. MAKING THE INVISIBLE AUDIBLE
In the current state of affairs, the tools at disposal are ill adapted. The wireless sensors themselves lack a proper interface to convey precious information back to the user, in particular connectivity. Usually, the only available feedback to the users is through a series of LEDs. A more sophisticated graphic display would not be practical in most cases, as it would consume too much energy to be adapted to a resource-constrained device such as a wireless sensor. With this in mind, the options left to a deployment team are few. The staff can work blindly in a long and painful trial-and-error process. It can use ad-hoc nodes blinking their LEDs in order to assess one-to-one connectivity. Such a measure involves moving around two nodes that run a Ping-Pong application, and constantly observing the LEDs indicating metrics such as Packet Error Rate (PER). Or it can use more complex network monitoring systems, to the extra cost of relying on extra infrastructure or software services. Selavo et al. developed a portable graphical display for deployment time validation [SWC+ 07]. However, our experience indicates that traditional displays are usually not convenient during a workintensive deployment task. Indeed, portable devices generally use LCD displays that are difcult to read outdoors, especially on sunny days. It is also to be noted that the necessity to actively look at a signal or a screen is an important distraction from the work to be accomplished. To this day, a light-weight tool that makes it easy to assess the quality of the radio channel while performing the necessary deployment tasks is still lacking. Designing such a tool for the average user is challenging. Currently, deploying a WSN remains a task requiring a high level of expertise, while end-user installation is crucial for cost reduction, scalability and users acceptance of the technology [BCL04]. In particular, the deployment-support tools that have been developed so far (see Section 7.2) require advanced computer skills and knowledge in networking. If WSNs are to become as ubiquitous as foreseen by many analysts, it will be necessary to develop intuitive interfaces for this technology. In this context, an important issue is the ability of untrained users to deploy a WSN effectively by assessing connectivity and placing nodes appropriately. A deployment and maintenance support tool for wireless sensor networks should satisfy a basic set of requirements. First of all, the system should interfere minimally with the task to be carried out. In particular, it should not require the installation of an extra infrastructure. Because deployments can take place in challenging environments, it should provide information that is easy to read in all circumstances. Finally, it should not require any special expertise to be interpreted by the person in charge for the deployment, while providing expert users with extended information about the network. The contribution of this chapter is three-fold. First, we introduce the application of sonication (the use of non-verbal audio signals to convey information) to wireless sensor networks, discussing advantages and challenges of this approach. Second, we present the design and implementation of Sensor-Tune (see Fig. 7.1), a light-weight deployment and monitoring tool based on sonication. To the best of our knowledge, it is the rst tangible example of a sonication-based solution to WSN problems. Third, we report on the eld evaluation of our tool, showing the impact of our design choice through the analysis of users performance on a network deployment task. This experiment represents the rst reported attempt to put a wireless sensor network deployment in the hands of non-specialists. The rest of this chapter is structured as follows: in the next Section, we begin by presenting the state of the art, both in deployment support systems and in sonication. In Section 7.3, we explain how we used sonication and justify the choices we made in order to make the sound feedback as intuitive as possible. In Section 7.4, we explain in detail the scenarios and system design. In Section 7.5, we present a survey on sonication validating our design choices. We present the implementation of our prototype in Section 7.6. Results from a eld experiment are presented and discussed in Section 7.7. In Section 7.8, we summarize the contributions of this work and draw guidelines for future extensions.
7.2. STATE OF THE ART
99
7.2
7.2.1
State of the Art

WSN deployment and Maintenance Support
In wireless sensor networking, a traditional way to assess the connectivity between two points is to use a ping-pong application that requires two wireless mobile nodes communicating with each other. Uni- and bi-directional connectivity can be assessed in this way. The idea of using a PDA for eld-inspections was mentioned before in several publications. A concrete example is the TASK project [BGH+ 05]. The TASK eld tool provides the ability to ping a single node, issue a command to turn on a LED, or to reset the node. Similarly, Ringwald et al. [RYR06] propose an in-eld inspection tool on a compact device that not only simplies the process of collecting information about the nodes state but also enables the actual users of the WSN to perform routine checks such as displaying the network topology, or uploading new rmware versions. In the same vein, Selavo et al. [SWC+ 07] recognize what they call the deployment time validation (DTV) as an indispensable part of elding a real system. They developed a deployment time validation approach, named SeeDTV, based on a simple communication protocol between a master node and a deployed network, and an in-situ user interface device, called SeeMote. The feedback is given to the user through a small screen adaptable to a mote. In all these cases, the feedback given to the user is visual, not sound-based. As mentioned earlier, we contend that sound is better suited to deployment tasks, the challenge being to provide intuitive feedback in this form. In the area of deployment-support tools, Ringwald and R mer emphasize the necessity to passively o inspect the network in order not to disturb it nor modify its state [RR07]. Consequently, they designed a deployment-support network (DSN) that superposes itself onto the network to be monitored, communicating with it on a back-channel. This approach supposes to deploy a second network in parallel with the monitored network, and it requires the extra-nodes to have dual radios. In contrast, our approach is resolutely light-weight. The interference caused to the network by the exchange of messages with the PDA is tolerable, because we only want to have a snapshot of the nodes state and of its connectivity with the rest of the network. As such, the perturbation of the normal trafc is limited both in time and in intensity.
7.2.2
Sonication
Sonication refers to the use of audio signals (mostly non-speech) to convey information. The use of sound to display information is not new, early examples include alarms, the telephone bell, the Geiger counter and medical instrumentation [KWB+ 97]. However, over the last decade this eld has drawn increasing attention, mainly because of the growing amount of scientic data to display and the improved technology capabilities to process audio. A presentation of sonication, its usefulness, approaches and issues, as well as a list of resources can be found in [KWB+ 97, BK99]. In most of sonication systems, selected features of the data display such as power onsets, spectral features, crossing of thresholds are used to control parameters of a sound synthesis process (such as pitch, amplitude, timbre...). The choice of these parameters - both features and synthesis parameters and their relationship is known as mapping strategy. Sonication research has often investigated applications targeted at expert users: either users expert in the acoustic domain (e.g. people with a music background) [PH06a] or experts in the domain of application [GC00, MMea03, QMK+ 07]. Therefore mapping strategies generally leverage on users advanced knowledge or ability to detect sound qualities in order to provide a rich output that displays
100
multiple data dimensions and at the same time associates each of them to different audio synthesis parameters such as pitch, loudness, duration and timbre. As discussed below, our approach is targeted at non-expert users, so it favors simplicity at the expense of multidimensional display. Different projects investigate the application of sonication to the monitoring of computer networks. The Peep [GC00] and NeMoS [MMea03] systems provide a framework for associating different network trafc conditions and events to the generation of sound, while Qi et al.[QMK+ 07] focus on intrusion detection and denial-of-service attacks. All of them differ from what we propose in the present chapter in that they are targeted at advanced users - network administrators - rather than non-experts. Moreover, no usability experiments are reported for any of these systems.
7.2.3
WSNs and their End-Users
User studies are still a rarity in the eld of wireless sensor networks. Beckmann et al. [BCL04] explored end-user installations of sensors for domestic use. Based on the results of their study, they proposed ve design principles to support this task. Their experiment considered the placement of sensors in the environment from the perspective of the sensor operation (proximity of the phenomenon to monitor), but did not consider communication issues. Some of the derived principles, however, still make sense in our context, in particular the benets of detecting incorrect installation of sensors, and of providing value for partial installations. The authors also emphasize how important it is to make appropriate use of user conceptual models for familiar technologies, which is what guided us when designing our audio interface (see Subsection 7.3), although in a different sense than meant by Beckmann et al. Williams et al. [WKD05] ran a user experiment about the impact that augmented objects (such as sensor-equipped appliances) will have on the perception people have of their surrounding space. They equipped toys with sensors generating sounds, in order to understand how people will encounter and understand these spaces, and how they will interact with each other through the augmented capabilities of such spaces. The authors of the study used an auditory interface; however, unlike our present work, they were not interested in the specicities of sound as a helper for WSN deployments. The feedback provided was not intended to improve user performance with the system.
7.3
Sonication for Sensor Networks
In this section we outline the advantages that sonication can bring to the eld of sensor network deployment and management, compared to the use of graphic displays. We also identify a number of challenges that need to be considered. Later in the chapter, we will discuss how we specically addressed some of these in the design and development of our deployment-support tool.
7.3.1
Applications
The precise description of the system model that we use is provided in Section 7.4. For the time being, let us only mention that we consider a self-organized multi-hop data collection network where nodes send packets to one or several base stations (sinks) either periodically or as responses to local events. We also suppose that the nodes are installed manually, and not at random. In this context, we want to design a deployment and maintenance support tool for wireless sensor networks that allows primarily for the monitoring of the connectivity of a node with the rest of the
7.3. SONIFICATION FOR SENSOR NETWORKS
101
Next hop node (Slave)
Next hop node (Slave) Serial cable
Monitored node (Master) Next hop node (Slave)
Wireless monitor
PDA
Next hop node (Slave) Headphone
Figure 7.1: Sensor-Tune system: The monitored node has a wireless connection with the eld manager, and queries its immediate neighbors for connectivity information. Both the quality of the local link, and the distance to the sink are taken into account.
network. This depends on the quality of the radio channel between the monitored node and its neighbors, and on the general topology of the network. A rst application is to inform the user about the appropriateness of nodes locations during deployment. Since the radio channel can vary considerably due to the presence of natural obstacles or interferences with other systems, it is important to get immediate feedback while nodes are being installed. In this way, one can move nodes in order to nd a better radio channel whenever possible, or to install efciently relaying nodes if needed. A second application is to report on a nodes activity after it has been deployed. In a multi-hop data collection network, it can be difcult to identify the points of failure when some nodes start reporting erratically or stop transmitting altogether. In this case, one wants to be able to interrogate the nodes individually about their recent packet-delivery-rate history in order to assess their connectivity status. There are also other possible use cases for an in-eld inspection system using sonication. For instance, to verify the proper functioning of the probes attached to a single node. A sound feedback would allow to check the responsiveness of a given probe. Also, if security becomes an issue, the detection and localization of possible attacks - such as jamming or malicious modication of the routing topology - are important features of the network. Here again, a sound feedback can inform a maintenance person present in the vicinity of a potential problem. Nevertheless, we focus in this chapter on the two rst use cases. A precise description of the rst two use cases will be provided in Section 7.4. For the moment, we are building a general case for sonication in the context of WSNs.
7.3.2
Advantages
The deployment and monitoring of WSNs normally requires users to physically navigate in the environment. In the deployment case, it is necessary to place the nodes in suitable locations, both in terms of sensing ability and radio coverage. Nodes maintenance may also require physical proximity because of the wireless sensors limited radio range. Navigation is primarily a visual task, which may be particularly demanding in areas that are not easily accessible. The use of auditory displays can be highly advantageous in these situations, because it frees completely the users visual resources, eliminating the need to
102
switch visual attention between the display and the environment. This visual attention switch is known to be a frequent cause of distraction. For instance, it has been shown that medical students, faced with the concomitant tasks of simulating an operation on patients while monitoring several of their health parameters, performed better when these parameters were represented as sounds rather than graphs [FK94]. Similar experiments performed on drivers and pilots led to comparable results [Bal94]. Moreover, considering that the most common portable graphic displays are hand-held, the use of audio outputs also frees the users hands, which can be used to support or balance the body in impervious situations. From the hardware and physical construction standpoint, an audio display, such as loudspeaker or headphones, presents considerable advantages compared to a graphic display. Audio displays are available at a fraction of the cost of the visual counterparts, and the same is true for the rendering and display driver systems. Because audio requires less processing and displaying power than image, the power consumption is also reduced, which is highly desirable in the context of WSN. From the ergonomics point of view, visual displays are often problematic for outdoor usage. Under bright daylight, the contrast provided by common LCD screens is often insufcient. Graphic screens also tend to be more fragile than their audio counterparts, which can be problematic in remote or industrial environments, where WSNs are often deployed [MPS+ 02], [HC05]. Sonication applications in other elds show the potential of the human ear to integrate simultaneously several dimensions of information into a single auditory experience. Experiments with auditory display of scientic data [MF95] tend to conrm the effectiveness of auditory display in conveying information and complex structures. Sonication has been demonstrated to be effective for the human recognition of patterns in data, both from experts [PH06a] and non-experts. For example it has been shown [PH06b] how in the context of physiotherapy it can be helpful to create real-time sonications corresponding to the patients movements and to let them compare the sounds that they produce to the target sound of a healthy person. Another appealing aspect is that sonications can allow alternative perceptions and new insights into the data [BK99]. This can give a certain level of intuitive understanding of specialized data sets to non-specialists. As it is illustrated below, with the design of evaluation of our system, sonication allows for the denition of interface metaphors that can be well understood from novice users but at the same time convey ne details to trained users.
7.3.3
Challenges
In order to apply sonication to the context of wireless sensor networks, it is important to consider domain-specic constraints and challenges. The interpretation of sound by a user can be decomposed into two parts. First a sound creates a sensation - a rst contact between the sense organ and the stimulus. Shortly after comes the perception of the sound, namely the attempt to identify and classify it [IH06]. In a system using sonication for extended periods of time, the sound must be designed not to generate fatigue in the sensation part. At the same time, it must be complex enough to convey information during the perception phase. In particular the design of the mapping strategy has to take into account: Prolonged use: deployment and monitoring sessions can span from a few minutes to several hours. It is hence important that the interface sound be pleasant or at least not annoying over an extended period of time. High-level metaphor: the overall impression of the interface should present a sound easy to interpret , if necessary conveying integrated and preprocessed information, making the tool accessible
7.3. SONIFICATION FOR SENSOR NETWORKS
103
by a non-specialist. Low-level details: specic aspects of the sound should allow advanced users to perceive detailed information about the status of the network or the node under examination. Ergonomics: The tool must work well in an outdoor environment. Sounds used in the interface need to be easily distinguishable from ambient sounds. Non-invasiveness: The tool must not disrupt networks operation. User acceptance: The tool must be acceptable regardless of the cultural background of the users. Using audio output in noisy environments (such as construction sites, highways or windy environments) can be problematic. The use of headphones with efcient sonic insulation could be a potential solution. However, the trade-off between sound insulation and the awareness of the environment required for the users safety has yet to be carefully examined.
7.3.4
Signal and Noise Metaphor
To address the challenges described in the previous subsection, we propose here an audio metaphor adapted to sonication for WSNs. The starting point for the design is the following observation: For the deployment and maintenance of WSNs it is generally possible to dene a good or desirable state in which the network is in a working state, all nodes being active and connected; and a bad or undesirable state in which the system is in a non-working state, sensing and communication not functioning properly. These states may just be ideal and conceptual, because in reality the good state may correspond to several actual network congurations. However they can be easily understood by even non-expert users. In fact, users will normally maintain some model of how the system actually works, which may or may not reect the reality depending, among other things, on their technical literacy. Based on this observation we decided to associate the desirable state to a sound that can be immediately identiable as pleasant and undistorted, and to use a gradual degradation of this sound to signify that the system state moves away from the desirable condition. The proposed mapping strategy can also be interpreted as a metaphor for the tuning of an FM radio, an action that is familiar to most people around the world. The emphasis is not on realistically mimicking the FM tuning effect, but just on providing users with a model easy to interpret. Excessively realistic metaphors are known to be problematic in HCI [SRP07]. The proposed strategy leverages the assumption that even non-expert users of WSNs will have some understanding of a system relying on radio transmission. The use of sound noise or distortion seems not to be very common in the auditory display literature, despite its strong metaphorical valence. This is perhaps due to the concern of confusing degradation generated by the interface with real degradation affecting the system. The advent of digital technology, however, allows for an easier control of the presence of noise or distortion, to the extent of completely eliminating analogue noise, as demonstrated by the adoption of comfort noise in digital communication systems [New04]. 7.3.4.1 What Sound to Play?
The proposed sonication strategy requires the choice of what we dened as a pleasant and undistorted sound. One advantage of the proposed mapping is that such a sound can be selected by the end-users according to their taste. However, some consideration needs to be taken into account in this choice. First
104
and foremost, the sound needs to be easily distinguishable from the distortion. Second, it should not generate fatigue in the listener. Finally, in order to optimize the use of storage memory, it is convenient to use an audio loop of relatively short duration . Our experience suggests that repeated speech clips can very quickly induce fatigue. Natural sounds (e.g. animals, water, wind) can be considered suitable background sounds, as they do not capture the attention of the auditor and are generally perceived as pleasant, but they are not always easy to distinguish from noise especially wind and water ow sounds can be quite similar to white noise. One alternative strategy could have been to translate into audio only the errors or problems in the network, and to have silence signify perfect connectivity. However, while this strategy would have the advantage of limiting the listeners fatigue, as well as reducing energy consumption, we deem it fundamental to give clear feedback about the monitoring tool status the use of silence makes it nontrivial to understand if the system is correctly working or the tool is just powered off. As a consequence, we opted to play a piece of music customizable by the user according to his/her preferences. 7.3.4.2 What Degradation?
There are several ways to degrade the quality of a sound, examples include: adding noise (including different types); reducing the resolution (increasing quantization noise); modifying the pitch or playback speed; reducing the signal bandwidth (bandpass ltering); convolving with another sound; adding a delayed copy of the same sound (echo); multiplying with a square wave of variable duty cycle (introducing silent gaps for duty cycle < 100%). Each method has its advantages and disadvantages, in terms of computational complexity and control of perceived degradation. Different types of degradation can be used at the same time, mapping the intensity of each of them to a different variable (e.g. signal power level, packet error rate, SNR, ...). In this way, novice users can perceive the general status of the system from the overall sound quality, even without distinguishing different degradation types, whereas advanced users can get more precise information recognizing what exactly is affecting the network. As detailed later in this chapter, for Sensor-Tune we used two types of additive colored noise to represent local and global properties of the network.
7.4
7.4.1
System Design
System Model
The context we consider is a multi-hop data collection network where nodes send packets to one or several base stations (sinks) that are connected to a server either directly or through a bridge (typically GSM or 802.11). The trafc can be either periodic, query-based or event-based. We assume that nodes
7.4. SYSTEM DESIGN
105
are capable of organizing into a data collection tree (or forest in the multiple-sink case). A critical issue for each node is to nd a suitable parent to route its data towards a sink. The placement of each node is constrained by the landscape and the data it is supposed to collect. This means that for each node to install, there is a region within which this node must be placed. We do not make any assumption about the size of the region, as this depends on the type of application considered. We assume that the radio channel is highly unpredictable. That can apply to both indoor and outdoor environments, depending on the presence of obstacles and interferences. We must deploy a total of N sensors. M < N sensors are already deployed. We add nodes one by one and want to place them as well as possible within their allowed region, which is determined by the phenomenon to observe. Extra nodes can be deployed in between the measurement points to insure connectivity, but as these nodes do not provide useful data, their number should be kept at a minimum.
7.4.2
7.4.2.1
Tool and Scenarios

Sensor-Tune: A Sonication Toolkit for WSN
We designed and implemented a deployment-support system that we call Sensor-Tune. It consists in a lightweight tool integrating a wireless sensor with a sonication module connected to earphones. This tool can interact with any node present in the network (see Fig. 7.1). It is designed to be carried easily by any person deploying the network. A small set of buttons are used to turn Sensor-Tune on and off, and to choose the mode of operation of the tool. Once Sensor-Tune is started in the proper mode of operation, no visual interaction with Sensor-Tune is necessary. In this way, the staff can focus on handling the nodes to deploy or to maintain, possibly in places that are difcult to access and require full physical availability (see Fig. 7.3). The acoustic feedback is intended to convey information that cannot be easily retrieved due to the limited interface capabilities of wireless sensors. As a proof of concept of the use of sonication in this framework, we decided to implement the following two use cases: Deployment support: Optimization of the placement of new nodes into a multi-hop network, Maintenance tool: Retrieval of recent connectivity history of a deployed node
7.4.2.2
Scenario 1: Live Information
In this scenario, we want to assess the connectivity of nodes as we are deploying them. In order to achieve this, we imagine the following ow of events: 1. The member of the deployment team carrying Sensor-Tune produces a new node from his stock. 2. As he/she turns it on, this node connects itself to Sensor-Tune and probes its neighbors in order to assess their potential as a parent. 3. This information is relayed to Sensor-Tune and displayed in real-time as audio data. 4. The deployment staff positions the node based on the obtained feedback. 5. When the node has been placed, a new node is turned on, which automatically takes over, while the previous node enters its normal mode of operation.
106
Figure 7.2: Carrying the Sensor-Tune: hand-free WSN deployment
Figure 7.3: Example of usage of Sensor-Tune, with a sensor difcult to access physically
7.4. SYSTEM DESIGN
107
In the event of total loss of connectivity, a continuous tone is played in order to spare the ears of the user. We implemented this scenario, evaluated it and used it for the experiments that we describe in Section 8.1. Information and Metrics When deploying one sensor i, we evaluate its connectivity with its neighbors Clocal , and its distance to the sink Cglobal . For Clocal , we use the Packet Delivery Ratio (PDR) (or, which is equivalent, the Packet Error Rate PER), while the expression of Cglobal depends on the routing protocol used. For CTP [FGJ+ ], the metric used is called CTP, and consists roughly in an aggregation of the link qualities from the node to the base station. Local connectivity: For Clocal , we use the information about the quality of the radio link between the node and its neighbors. For this information, we use the Packet Error Rate (PER) from this node to all its neighbors. It is dened as follows: P ER = where sf is the number of packets whose emission failed at the sender, nack is the number of packets that were not acknowledged, ack is the number of acknowledged packets. (sf + nack) (ack + sf + nack) (7.1)
General topology information: This information, Cglobal , reects how well the neighbors of the current node are positioned in the network with regard to the base station. For each potential parent, we take into account the metric of the multi-hop protocol used. This metric is customizable in our system through an API. Generally, it is based on the hop count and/or the aggregation of the connectivity levels (packet delivery ratios) of the nodes along the path to the base station. A good metaphor for this metric is the distance with the base station. 7.4.2.3 Scenario 2: Connectivity History
In this scenario, the connection is with an individual node, without local communication with its neighbors. It is assumed that this node has recorded relevant information over the last 24 hours, typically storing parameters in its ash memory for a succession of time steps that last 10 minutes each. It will replay it on-demand. The parameter that can be easily stored in the ash is the PER, information coming from the radio itself. More precisely: 1. The maintenance staff carrying Sensor-Tune walks around the deployment area 2. He/she denes a minimum PER performance Pmin . 3. Automatically, Sensor-Tune beacons the neighboring nodes 4. When a node hears Sensor-Tunes beacon, it answers if its average performance over thelast 24 hours is lower than Pmin . 5. The nodes that answered are queried sequentially in a FIFO manner 6. The history is downloaded to Sensor-Tune, where it is played a xed number of times (the total sound sequence should last a few tens of seconds)
108
7. The user can interrupt the sequence by pressing on a button, either deleting it or saving it in the memory for later retrieval and ner grained analysis. We implemented this scenario, but have not yet fully evaluated it. 7.4.2.4 Other Scenarios
Local Connectivity In the use case described above, we only monitor a nodes parent. This means that when moving a node, we do not know anything about its connectivity with its potential children. If we want to know what effect the moving of a node will have on its childrens connectivity, we need to test the Packet Delivery Ratio, namely the percentage of packets received by this node coming from its children. The sonication technique for this use case can be directly derived from the previous ones. Probes Operation When deploying a node, the proper operation of the probes can be tested as a sound as well. In this way, the deployment staff can make simple tests such as covering a solar radiation sensor, warming a thermometer, lling a rain gage, etc. In this case, the noise metaphor does not hold anymore. An appropriate sonication would be the synthesis of a sound whose pitch varies as a function of the sensed data.
7.4.3
Protocols
When monitoring a wireless network, it is important to do so in a minimally invasive way. Ideally, a fully passive system should be used. In our case, however, it is not possible. Most of the time, indeed, the node that we monitor is not part of the network yet. It needs to interrogate its neighbors about their position and to run a decision process to choose its parent. In a normal operation mode, this procedure takes time, and we cannot rely on the regularly exchanged routing messages to send instantaneous feedback to the user as he/she moves the node to nd its best placement. Accordingly, we designed a communication protocol that provides real-time connectivity feedback, and is as minimally invasive as possible. The number of messages exchanged is compatible with a typical environmental monitoring application, even if it might conict with applications requiring very high data rates and a nearly instantaneous response. We analyze its overhead in Section 7.4.3.4. 7.4.3.1 The Actors
We distinguish several actors in the unfolding of the protocol. 1. Sensor-Tune: the monitoring device. 2. Master: the node to deploy, which will query its neighbors for a suitable parent. 3. Slave: any node in the neighborhood of the master node, which is going to answer its queries. The basic idea behind Sensor-Tune operation is to run a one-hop multicast protocol between the node to deploy (the master) and its neighbors. The radio link between the PDA and the master is used to bootstrap the process, to forward periodically data to the PDA, and to switch nodes. 7.4.3.2 Live Data
Fig. 7.4 describes the exchange of messages for this use case.
7.4. SYSTEM DESIGN
109
Figure 7.4: Communication protocol for the Live Data use case. The master initiate the session, rst querying the cluster heads, then all the nodes in its neighborhood. N requests are sent, then the neighbors have a time slot of duration D to answer if their local and global connectivity metrics are good enough
110
1. Sensor-Tune receives a message from the PDA as soon as the latter is ready to accept candidates (meaning that the user pressed the on button). 2. When a new node is turned on in the vicinity of Sensor-Tune, it sends a INIT message to it, thus applying to become a master. 3. If it does not receive an answer within a given (customizable) time, it enters its normal mode of operation (meaning that Sensor-Tune was off or not present). 4. If Sensor-Tune hears the INIT message, it answers with a START, turning the new node into a master. 5. At this point, the master starts a series of rounds that last one second each. During the rst 500ms, it sends bursts of INFO QUERY messages (customizable, but typically 10), and waits for an INFO RESPONSE from its best potential parents during the next 400ms. In order to reduce collisions, the neighbors use a random back-off timer during this period. 6. The last 100ms of each round are left for regular data trafc to take place. 7. Based on the metrics we dened in the previous sections, the master will select the best potential parent and forward its local and global connectivity parameters to Sensor-Tune. This information will ultimately be translated into sound. 8. When we are satised with the placement of the node, we simply turn a new node on. Upon reception of the new INIT message, Sensor-Tune rst stops the previous master (which enters then the normal mode of operation), before starting the new one. In order to avoid too many answers from the neighbors, the target value for the PER is included in the INFO QUERY message. We denote it Cglobal . This value depends on the last value received (and increases exponentially if no messages have been received in the last rounds). Nodes only respond if, based on the INFO QUERY messages they received, their own PER from the master combined with their own distance to the base station is close enough to Cglobal . Cglobal < Cglobal C Where C is a parameter. A higher value of C improves the responsiveness to channel variations, but increases the trafc. Cglobal is updated at each round with the best value of the last round. 7.4.3.3 Clustering
Initially, the value of the threshold has to be set arbitrarily, which means that many answers can be expected. In order to avoid a congestion at this point, we use an algorithm that partitions the network into different clusters. Each cluster is composed of a cluster head and a subset of its single-hop neighbors, the cluster members. At the rst round of the Live-data protocol, only cluster heads can respond. The clustering protocol takes place at the deployment of each new node (see Fig. 7.5). 1. The new node sends a Request message to the network, and starts a one-shot random timer. 2. A node whose timer is expired sends a packet to all its single-hop neighbors, declaring its type as a cluster-head. 3. A node that receives a Request message will answer to it by a message that declares its type (clustered or non-clustered).
7.4. SYSTEM DESIGN
111
Figure 7.5: State machine of the clustering algorithm
4. If the new node receives a message from a cluster head, it stops its timer and becomes member of the corresponding cluster. Each cluster is uniquely identied by the ID of its cluster-head. 7.4.3.4 Overhead
The protocol described above is based on a fast exchange of packets between a new node and its neighbors. During each one-second round, at least 10 messages are received by the neighbors. The number of messages that are sent back depends on the number of neighbors and on the quality of their link with both the new node and the base station. We tested this scheme successfully for up to 5 potential parents. If, above this limit, messages are lost due to collisions, this will not affect signicantly the performance of the system, because there will be more than enough parents to choose from. The resort to clustering ensures a successful bootstrapping of the protocol. Another question is whether this protocol will have an effect on the operation of the WSN main application, because it is communication-intensive, if only for a short period of time. If we are installing a whole network from scratch, the possible disruption is not an issue, as the network is usually not supposed to be fully operational at the moment of deployment. If a node is to be added at a later stage, while the network is operational, we may disrupt the network operation locally during the period of time it takes to install the new wireless sensor. As this task typically takes from a few minutes to up to an hour, this is not a problem for a typical environmental monitoring application, with data rates in the order of the minute or more, without tight response-time constraints, and with some tolerance to errors or missing data. For an alert-based system, the 100 ms window will allow for the delivery of an alert message even in a dense network.
7.4.4
Sonication Mapping Strategy
As outlined in section 7.3.4, the interface is based on a simple yet powerful model: a pleasant sound indicates that the network is in good condition, whereas additive colored noise indicates a degradation in the network status. The tool allows us to monitor one node at a time. The existence of a connection between the current node and the sink is represented by a piece of music sm (t), corrupted by an amount of additive colored noise depending on the connection quality ng (q, t). This can be expressed as follows: so (t) = sm (t) + ng (q, t) (7.2)
112
where so (t) is the sound output, t is time and q is the connection quality. The base sound sm (t) can be selected according to the taste of the end-user, so that different cultural backgrounds can be accommodated. As discussed above, however, it is important that the sound chosen is easily distinguished from noise. For minimizing storage requirements, an audio loop is used for sm (t), selected to be long enough not to be annoying and to loop in a seamless way (discontinuities could be perceived as signal degradations). In the prototype evaluation described below we used a 16-second clip of classical piano music. The noise signal is the weighted sum of two distinct colored components: a lower frequency component indicates local connectivity the PER between the node and its parent and the other component indicates the global connectivity the routing metric of the current nodes parent. This can be more precisely expressed as follows: ng (q, t) = nL (qP ER , t) + nH (qroute , t) (7.3)
where and are two parameters; nL (t) and nH (t) are colored noises produced from the same white noise source ltered respectively with a low-pass lter with cut-off frequency 200 Hz and band-pass with center frequency at 2.7kHz and a bandwidth of 20 Hz. The two components are distinguishable when needed, as shown by the user survey described in the next section. Although normal usage does not rely on users distinguishing the two types of noise, this feature can provide an additional layer of information for advanced users. We tuned and manually, in order to ensure a comfortable level of noise in desirable cases (low PER, small distance to the base station). In particular, was chosen signicantly smaller than , in order to give more importance to the local connectivity, because this is the parameter of primary importance when placing a node. The function that we chose to generate noise as a function of PER acts almost linearly for low values of PER and becomes exponential as the PER increases1 . This is because we want to monitor more closely the low values of PER, as above a certain threshold of packet errors, experience shows a rapid degradation towards total disconnection. At the same time, the human ear functions on a logarithmic scale, so higher intensities of noise become harder to distinguish. Since the power of the output signal depends on the level at which the user will tune its headphones, we normalized the music waveform and added noise with an increasing envelope. For instance, a 5% PER corresponds to a normalized amplitude of 0.02.
7.5
Initial Exploration: User Survey
In the previous section, we discussed how sonication techniques, such as altering a sound le with noise, are useful for the deployment of WSNs. We suggested the use of the PER and Cglobal to alter a sound le with high and low frequency noise, respectively. In this section, we describe the results of a user survey which explores the perception of noise by users.
7.5.1
Description
Given the generic nature of this survey on noise perception, we do not require the users to have any prior knowledge of sensor networks, nor musical predispositions. Similarly, we do not constrain the user auditive environment: Sensor-Tune should be usable in any milieu. The survey is thus available online
1
This function behaves as the ETX value dened in [CABM03]
7.5. INITIAL EXPLORATION: USER SURVEY
113
Intensity Low (P ER < 10%) Middle (P ER 20%)
High (P ER > 30)
|| 0 1 0 2 2
Correct 85% 100% 92% 92% 64%
Entropy 0.60 0 0.40 0.40 0.94
Table 7.1: Survey results for the rst part. Users seem able to recognize noise intensity variations in low and middle intensities (results based on 24 answers).
and was advertised at EPFL via email2 to users with different academic backgrounds. To stop the users from taking the survey several times, we use persistent cookies3 . Sonication techniques are usually evaluated by measuring how helpful they are for users to accomplish their task. In our case, we wish to know how precisely can users perceive the variations of noise in intensity and the frequency. The survey is composed of two parts. In the rst part, the users are given eight sound les containing a sequence of classical piano music altered with noise of low, middle and high intensity. We introduce noise intensity variations and test whether users can perceive these variations by asking: Do you perceive a change in the noise intensity? Among the possible answers, users are asked to choose whether they perceive an increasing/decreasing or no change at all. With this question, we evaluate the granularity of noise intensity perception by users. On our normalized scale, a of +1 corresponds to an increase of 0.02 in the noise envelope. In the second part, we generate twelve sound les of the same piece of music, but this time, not only do we alter the les with varying noise intensities, we also use two types of noise: a low frequency and high frequency noise. We examine whether users could recognize the noise types (i.e frequencies) by asking: Ignoring changes in intensity, do you perceive different types of noise? Users are asked to say yes or no. We consider various scenarios where both the intensity and frequency vary (DD), where only the intensity varies (DS), and where none varies (SS)4 . Finally, we ask users for their age and whether they used a headset while taking the survey.
7.5.2
Results
Over a period of two weeks, 24 users took the survey online. 95% of the users were in their twenties (18-30) and 66% used headphones. Overall, we did not observe any changes in the quality of answers between users with and without headphones. As suggested in [FBB05], we use the entropy Hq to measure the uncertainty of the answer to a question q:
l
Hq =
i=1
pi,q log2 (pi,q )
(7.4)
where pi,q is the probability to answer i to the q, and l is the number of categories for answers to q. Hq is measured in bits and tells how easy it is for users to identify a sound.
2 3
http://csn.ep.ch/sonication A stronger authentication mechanism could be used but was not deemed necessary because of low risk of attacks. 4 D stands for Dynamic, S for Static
114
Variations SS DS DD
Correct 4% 46% 100%
Entropy 0.24 0.99 0
Table 7.2: Survey results for the second part. Noise type (i.e. frequency) and intensity varies (DD), only intensity varies (DS) and none varies (SS) (results based on 24 answers).
In the results of the rst question (Table 7.1), we found that 90% of the users seem to be able to distinguish noise intensity variations at low and middle intensities (Hq [0, 0.60]). This is much better than for high intensities (Hq = 0.94). This result conrms that because the human ear works in a logarithmic manner, users cannot efciently recognize noise variations at high intensities. Accordingly, we empirically dimension our system with respect to the noise variations that we introduce when the PER and Cglobal vary. For instance, a value of +1 of corresponds to a PER of 5% in our nal system, +2 to 10%, +5 to 20%, and +30 to 50%, etc. With the second question (Table 7.2), we observe that people tend in SS and DS cases to aggregate both noise types as one. In SS, they are even convinced that there is only one type of noise being played (i.e., note the low entropy). It appears that users could distinguish two noises only when one noise replaces another over time (DD). In other words, when the relative importance of the two noises changes - the dominated noise becomes dominant - users can distinguish the two noise types. With Sensor-Tune, a user must rst optimize the local connectivity (PER). During this operation, while the PER is not good, the low and high frequency noises will not be distinguishable: the user can concentrate on nding a good location for a node. Once a good location is found, the high frequency noise vanishes, and the low frequency noise appears clearly. Thus, users can alternately focus on optimizing the local connectivity and global connectivity to the base station. The results of the survey allowed us to verify our system design. We realized that we must carefully select noise intensity variations for the users to be able to notice them, and noise frequencies for the users to recognize them when necessary.
7.6
7.6.1
Prototype Implementation
Prototype Description
In this section, we describe the implementation of a prototype of Sensor-Tune, using a PDA running Linux Maemo (Nokia N800) (see Fig. 7.7). We emphasize the fact that a real commercial system can be implemented in a much less expensive way than with the off-the-shelf components that we used 5 . The PDA is connected through a serial interface with a wireless device compatible with each node that is to be deployed. The wireless device communicates with the monitored node and forwards the received information to the PDA, where this information is analyzed and passed on to a sound-generator. The user can listen to the sonied data via headphones. Once packed, the system is quite compact (see Fig. 7.8).
5
the technical details and software implementation are available on-line at http://csn.ep.ch/anonymous
7.6. PROTOTYPE IMPLEMENTATION
115
Figure 7.6: Sensor-Tune simplistic graphical user interface
We used a simple graphical interface to start the tool and set it to the desired mode of operation (see Fig. 7.6). We decided to keep this interface at a bare minimum so that user options can be magnied and be readable in outdoor conditions.
Figure 7.7: Sensor-Tune prototype: A linux-based PDA connected through a serial port to a wireless sensor
Figure 7.8: Sensor-Tune prototype once packed
We distinguish the embedded part from the PDA part. On the PDA, a Java subsystem is responsible for the message interface and the data analysis, and a dedicated software, pure data (PD), takes care of the sound generation part. 7.6.1.1 Software and Hardware
On the embedded side, we use the TinyOS [tos] operating system, as it has evolved to become the preferred choice of the research community to design and implement wireless sensor network systems.
116
Figure 7.9: Sensor-Tune functional blocks
We implemented our tool on the tinynode [tin] platform. For the PDA, we used the Nokia N800, which runs Linux Maemo 3.2, thus making it easy to add custom software to it. Java in particular is easy to install. Moreover, it runs PDa, the embedded version of the open source Pure Data sound generator. All different software components communicate through sockets. 7.6.1.2 Embedded Part
There is minimal change to be brought to any multi-hop application, whose performance we want to monitor. The application comes as a plug-in to be added to the conguration le of the deployed application. The global metric needs to be passed back to Sensor-Tune through an interface, because it depends on the routing protocol to be used. All other metrics are dealt with at a lower layer, so they are independent from the particular context. 7.6.1.3 PDA
Java subsystem The java subsystem has four tasks: 1. State machine: managing the PDA state machine,in order to keep synchronization with the node to deploy. 2. Message interface: sending and receiving messages exchanged with the master. 3. Data processing: analyzing the incoming data, logging them if appropriate, and processing them so that they can be translated into sounds. 4. Sending the result to the sound generator through a socket. Pure Data subsystem PD (Pure Data) is a real-time graphical programming environment for audio, video, and graphical processing. PD is an example of Dataow programming languages. In such
7.7. EXPERIMENTAL VALIDATION
117
Figure 7.10: PDA: PD subsystem description, with music and two additive noises
languages, functions or objects are linked or patched together in a graphical environment that models the ow of the control and audio. PD is an open source project and has a large developer base working on new extensions to the program. This tool, initially designed for desktop computers, has been ported on small handheld devices running Linux, under the name PDa (PD anywhere) [Gei03]. As mentioned earlier, we chose a method consisting in superposing to the background music two noises at different frequencies: a high frequency noise whose volume increases as the packet error rate between the node and its best potential parent increases, and a low-frequency noise (perceptually less annoying) whose volume increases as the distance from the base station in terms of hops increases. The (simplied) PD subsystem that we designed is described in Fig. 7.10.
7.7
Experimental Validation
To validate the proposed design, and in particular the audio-based interface, an experiment was designed and performed. The rst objective of the experiment was to assess whether, with an appropriate interface, it is possible for non-specialists to deploy a wireless sensor network in a challenging setting, and this with minimal training. We then wanted to evaluate the effects of the auditory presentation independently of the underlying technical system and the actual information presented. For this reason, the audio interface described in Section 7.4 was compared with a graphical user interface (GUI) that presented the same information on the screen of the Nokia PDA.
7.7.1
Comparable Graphical Interface
In order to assess the sonication based interface independently from the amount of information provided and of the underlying technical implementation of the system, we decided to compare it with a graphical user interface that would present the same information. Therefore we designed and implemented an interface that displayed two horizontal bars of variable length, as illustrated in Figure 7.11, one related to the PER and the other to the ETX of the monitored node (see Section 7.4). We decided to mimic the
118
signal bar common in all mobile phones, so the bars are full when the connection is perfect and become shorter when the connection quality decreases in other words the length of each bar was inversely proportional to the PER and ETX, respectively.
Figure 7.11: Screen capture of graphical user interface used for the experiment as shown on the Nokia PDA. The horizontal bars convey information about the connection quality.
7.7.2
Experimental Design
The experiment consisted of 2 network deployment tasks, in each of them subjects had to create a linear multi-hop network that connected specic start and destination points in a building on EPFL campus. For both tasks the destination point was the same and it was located in the parking garage in the basement of the building, and marked with an x sign on the oor. The starting points for the two task were on two different ends of the 4th oor of the building (there was a 5 oors distance between start and destination). The building has 5 different stairways and 3 elevator towers and it includes a mix of glass, metal and concrete partitions that attenuate the radio signals of our wireless nodes in different ways (often drastically). A number of movable elements, such as doors and elevators, made the radio path variable with time, which contributed to make the tasks even more challenging, given especially that the experiment took place during business hours, when many people walk around the building. For each task, subjects had a maximum time of 20 minutes and a maximum of 6 nodes (but emphasis was put on the fact that they could complete the task with less). As a benchmark, both tasks could be completed by experts in less then 5 minutes, using only 3 nodes. Subjects received instructions in written form (to ensure consistency), informing them about the system and the two tasks, asking them to try and complete them as quickly as possible, using the smallest number of nodes as possible, and making the connection quality as good as possible. The instructions were kept concise, with total length of two A4 pages. The instruction simply reported that the audio degradation through noise, or the length of the bars in the GUI indicated the quality of the connection of the current node to the base station, but did not provide any details about the PER nor the ETX. Several participants asked what was the difference between the two bars, but they were answered that they reected different aspects of the connection quality but that the details were irrelevant to the experiment. After each subject read the instructions, and before the start of each task, the experimenter showed the start and destination point, and a specic path between them, even though during the experiment subjects were free to take any path they liked between the two points. A maximum duration of 20 minutes was given for each task. If the subjects did not reach the destination point within this interval, the attempt was considered failed.
7.7. EXPERIMENTAL VALIDATION
119
All subjects tried both interfaces, each on a different deployment task in alternate order: half of the subjects used the audio interface in the rst task and the GUI in the second, while the other half used the GUI for the rst task and the audio interface in the second. The two tasks, however, were performed in the same order. The completion time, the number of nodes needed to achieve the task and the resulting network performance were recorded for all tasks. Participants were shadowed by an experimenter, who took notes about their behaviour and performance. At the end of the experiment subjects were asked to ll a short questionnaire related to their previous experience with computers, with wireless networks and with music as well as their preferences between the audio and graphical interfaces in terms of ease of use, efciency and overall favour.
7.7.3
Participants
Participants were 14 males, of age between 26 and 48 (avg. 32.6, st. dev. 7.4), all volunteers. All subjects were naive, in that they had not used our system before the experiment and all had no experience in deploying a multi-hop wireless network. Four subjects reported having set-up a home wireless network (Wi-Fi access point).
7.7.4
Results
Overall, the network deployment was successful in 17 of the 28 trials (60.7%). The rst task was completed successfully in 7 of the 14 cases (50%), while the second task was completed sucessfully in 10 of the 14 cases (71.4%). When the audio interface was used, the rst task was successful in 4 out of 7 cases (57.1%), while with the GUI the rst task was successfully completed in 3 out 7 cases (42.9%). For the second task, subjects using the audio based interface were always successful (7 out of 7, 100%) while subjects using the GUI where successful in 3 out of 7 cases (42.9%). The results are summarized in Table 7.3 Out of the 14 subjects, 5 succeeded in both tasks (2 started with the audio interface, 3 started with the GUI); 2 subjects, who started with the GUI, failed in both tasks; 5 subjects failed in the rst task but succeeded in the second (1 of them started with the audio interface and 4 started with the GUI); 2 subjects completed successfully the rst task using the audio interface, but failed in the second task using the GUI. At the qualitative level, we noticed a number of frequent behaviors that were detrimental to the task completion or even resulted in failure. First, most participants tried to let the radio waves follow their same path in particular, most participants tried to bring the radio signal down the stairways, even though these are interrupted by a number of glass and metal doors that block the radio waves of the nodes. Often, it was noticed that these participants were aware of the fact that the radio waves can go though walls, but simply did not actively use this information. Only 3 of the 14 subjects attempted to let the wireless connection go through the oor, which results in a more efcient solution. All subjects who attempted this alternative strategy were successful in completing the task and used a minimal (3) number of nodes. A second common source of problems was the fact that the very rst node was placed in a position where it was not very well connected with the base station, which compromised the connection of the following nodes to the base station. In turn, the bad positioning of the rst node was often the result of the two following behaviours: before choosing the position for a node subjects monitored its connection quality for a period that was too short to notice signal drops due to transient events such as other people
120
Task 1 Success
Task 2 Success
Total Success
Audio
4 of 7 (57.1%) 3 of 7 (42.9%) 7 of 14 (50.0%)
7 of 7 (100.0%) 3 of 7 (42.9%) 10 of 14 (71.4%)
11 of 14 (78.6%) 6 of 14 (42.9%) 17 of 28 (60.7%)
GUI
Overall
Table 7.3: User experiment results: successful completion of the deployment tasks by untrained participants.
passing by, doors opening and closing, or elevators moving; subjects monitored the connection quality only when they were very close to the nodes, while their body somehow inuenced the EM eld in favor of the connection. As soon as they walked away, the connection dropped. Regarding the expressed preferences, 8 of the 14 subjects (57%) indicated the audio interface as easier to use, while 9 (64%) indicated that they deemed the GUI let them perform better, and the same number reported it as generally preferable.
7.8
Discussion
Throughout the literature, one cannot help but feel there is a paradox in the fact that wireless sensor networks are envisioned as the ubiquitous communication technology of the near future, while they remain cumbersome to deploy and difcult to maintain. In this chapter, we have investigated a novel approach for interfacing the wireless sensing world, relying on acoustic feedback. We have presented the advantages of such an approach in terms of deployment efciency, reliability, intuitiveness and cost, and have developed an original metaphor for the analysis of connectivity based on the metaphor of noise. The implementation of a prototype allowed us to conrm that this approach is promising for wireless sensor networks. The overall success rate of 60.7% in the experiment indicates that the interface is effective in supporting non-expert users deploying a multi-hop wireless network, validating the proposed design for Sensor-Tune. The results indicate no large differences between the performance with the audio interface and with the GUI, suggesting that the two interfaces perform as well as each other. The additional advantages provided by the audio interface, namely eyes-free and hands-free operation, are therefore available without any penalty compared to a graphic counterpart. This experiment was conducted indoors in a technical institution, although not all participants had a formal technical training. Since we want to apply this strategy in the eld, by putting wireless sensors in the hands of agriculture scientists if not farmers, another experiment needs to be conducted by the intended users. Only then will be the tool formally validated. This validation should include an improve-
7.8. DISCUSSION
121
ment whose necessity has been unveiled by the user experiment: from the observation that often one specic link between two nodes is the cause of major problems in the entire network, the modication of the interface so that users can easily select which link to monitor, or even monitor several links at the same time, may dramatically increase its performance. In terms of future developments, the rst step will be to integrate the Sensor-Tune application with the Common-Sense Net system, which uses a proprietary MAC and Routing protocol, designed for extremely low duty-cycling. Other applications than deployment support could also be implemented and tested, such as history of connectivity, on-board sensors validation, etc. Security applications can also be sought. Finally, transposition of the sonication paradigm to other wireless technologies (such as 802.11 access points) could be envisaged.
122
Chapter 8
Usability and Usefulness of the System

In the COMMON-Sense Net project, our goal was to help the farmers monitoring the physical farming environment, in order to understand more precisely the physical processes at hand, and to react optimally to changing conditions. Our initial vision had been to bring the benets of technology directly to farmers, in a participatory way. We deployed our system after identifying use cases with the locals. Over a long period of data collection and usage, we reached an impasse: numerous difculties emerged, essentially hindering our efforts to bring in a participatory manner the added value of WSN technology to the farmers. This has been primarily due to the farmers alienation with the worlds of science and technology. After we presented the lessons learned through our three-year long effort and deployment in Chapter 6, we are currently forced to consider as somewhat idealistic our initial objective. The resource-poor farmers could not really put enhanced environmental data in use effectively. Based on these experiences, we moved on with a different approach: we investigated controlledenvironment strategies related to rain-fed farming, such as developing new crop varieties or pest prediction measures. Accordingly, the position this chapter takes is as follows: Under the current conditions in developing regions, such as Karnataka, the targeted users for WSN-enabled applications should be researchers, scientists, and technicians. As our study case and eld experience indicate, this orientation towards a new user group (scientists), which can then advise or guide the farmers, appears currently the only method to have an effective decision-support system for rain-fed farming. This position was corroborated by an experiment that we conducted in Bangalore, from November 2007 to February 2008. This chapter brings this seemingly pessimistic yet realistic position to the attention of the community. Essentially, our concerted effort of deploying and running the COMMON-Sense Net system points that still scientists remain the preferred customers of WSN technology. Section 8.1 describes the methodology for our user experiment to determine if our new user target group was an appropriate choice. In Section 8.2, we present our results, followed by a discussion in Section 8.3.
8.1
Charting the Paradigm Shift
The path from user needs to precise specications of a system is not an easy one to trod. In the previous section, we identied a strong necessity to nd a mediator between the technology and the target population. Agricultural scientists are ideally placed to dene use cases. However, this is no trivial matter, because for them sensor data represents a new and unfamiliar context. Most of the scientists we 123
124
CHAPTER 8. USABILITY AND USEFULNESS OF THE SYSTEM
interacted with are not familiar with sensor data at high resolution in time and space provided by a large number of data gathering points with uniform accuracy.
8.1.1
Choosing the Target Population
For these two reasons, we decided to set-up an experiment where we confront soil physicists, crop physiologists, entomologists, pathologists and agronomists with the results gathered from the eld by our deployed prototypes. The reason for selecting such a various user basis is twofold. On one hand, we wanted to nd the largest scope of use for the WSN technology in the context of rain-fed farming, and did not want to restrict ourselves to our own preconceptions. On the other hand, as different disciplines can have various data requirements, it was important to know whether an appropriate system design could meet all of them. There are several types of institutions where such professionals are likely to work, each of them with its own goals and agenda: 1. Academia: scientists doing research in agricultural departments. For them, two competing goals are at stake. Doing research that provides them with scientic impact and visibility, as well as solving practical problems. 2. Government: scientists working either as advisors for policy makers or as implementers of programs at the local level. Marginal farming is only one aspect of their concern, which is agriculture as a socioeconomic sector. 3. Non-governmental agencies: NGOs focusing on rural development often are innovative in terms of agricultural practices. As such, they are interested in applied research. We did not extend our survey to the industry of agriculture inputs or to corporate agriculture, although these two sectors are likely users of the wireless sensor networking technology. In the rst case, we did not want to get involved in the controversy surrounding the effects of large seeds providers on the livelihood of small and marginal farmers in India. In the second case, the type of agriculture practiced (mechanized farming, on-demand irrigation, precision agriculture with high added-value) was considered too different from our focus of interest, namely rain-fed farming.
8.1.2
Goal and Methodology
We interviewed 30 people from the backgrounds detailed above, following both a qualitative and a quantitative approach (Table. 8.1). The goal of the experiment was to identify the use that agriculture scientists would make of the data that are collected by the COMMON-Sense Net system. The experiment was scheduled to run for 2 weeks in November 2007. We asked the scientists a series of questions about the value of environmental data for them. The goal of the experiment was to understand precisely: 1. What are the types of environmental data they can make use of, and how? 2. What is the spatial diversity they will use for the data? 3. What is the time granularity they will make use of in their task? In order to answer to these questions, we used three complementary approaches: Structured interviews: As a preliminary, some general questions were asked to the participants in a general questionnaire before they tested the interface during a two-weeks long study. These questions served to assess their current view of the eld. Participants were asked to answer more concrete questions
8.1. CHARTING THE PARADIGM SHIFT
125
Name Prof. Ali Prof. Bhaskar Prof. Bhaskar Prof. Gowda Prof. Kumar Prof. Mohan Raju Prof. Murthy Prof. Parama Dr. Ananth Dr. Ashar Dr. Beena Dr. J.N. Madhura Dr. Masaki Dr. Reddy Dr. Shahidhar Dr. Sheshshayee Dr. Suvarna Dr. Reddy -Anonymous-
Field of Research Agronomy Crop Physiology Forestry Farming Systems Entomology Crop Physiology Soil Physics Soil Science Genetics and Plant Breeding Crop Physiology Crop Physiology Crop Physiology Crop Physiology Crop Physiology Agronomy Crop Physiology Microbiology Veterinary Medicine Information Extension
Afliation UAS UAS UAS UAS UAS UAS UAS UAS UAS UAS UAS UAS UAS UAS UAS UAS UAS BAIFF Karnataka State Agriculture Dpt
Table 8.1: Names, research specialty and afliation of researchers interviewed during the user experiment. The list is restricted to faculty members and senior researchers
126
in a detailed questionnaire during the experiment. A detailed description of the questionnaires can be found in [PRS08]. Behavior observation: It is risky in such an experiment to rely solely on users opinions. This is why the scientists were encouraged to provide data sets in order to substantiate their answers (in the form of graphs or numbers). Moreover, the queries they made to the database in order to retrieve data were recorded. In particular, the time and space granularity of the data that they consulted was logged into a database. This makes it possible to analyze the scientists behavior as well as their discourse. Semi-structured meetings: Finally, a debrieng took place after the experiment, in order to allow the participants to share their impressions in an informal discussion. As it will appear clearly in the next section, this part turned out to be surprisingly rich in information.
8.2
8.2.1
Experiment Results
Questionnaires
The questionnaire were answered thoroughly. Each and every participant wrote detailed comments. 8.2.1.1 General Questionnaire
This questionnaire was lled as a preliminary one, before participants had the occasion to use the application. All participants but one identied environmental information as an important input for the study of rain-fed farming. The participants chose primarily soil water content as the relevant constraint for rain-fed agriculture, followed by temperature. Soil type, humidity and rain fall are also mentioned a signicant number of times (see Table 8.2). As for the characteristics per parameters (desired rate, spatial granularity and precision), they are depicted in Figs 8.1, 8.2, 8.3. Regarding spatial density, one can classify the parameters into two categories: the parameters considered with a low spatial variability, i.e. one kilometer and above, such as temperature, rain fall and atmospheric pressure. And the parameters demanding high spatial variability (from 500m downwards): only soil moisture belongs to this category. Interestingly, soil type shows a bimodal result, about half the users considering that a single measurement point is enough, and the other half considering that it should be performed at least every few hundred meters. The required measurement frequency shows a wider distribution. It is interesting to note, however, that in all cases but one (soil moisture), the lower limit is the day. Only for soil moisture were hourly or more frequent measurements deemed appropriate, and only for a minority of users (less than 10 percent). As for the error tolerated for each parameter, the participants tend to require high precision (less than ve percent error). 8.2.1.2 Detailed Questionnaire
The detailed questionnaire was lled after the users had some time to play with the application. Fig. 8.4 shows the choice users made as for crop. Groundnut, which is the crop currently most grown in the area, comes rst, with almost half of the expressed opinions. Other indigenous crops are mentioned, such
8.2. EXPERIMENT RESULTS
127
Rain-Fall Pattern 16%
Soil Type 17%
Humidity 16%
Soil Water Content 29%
Temperature 22%
Table 8.2: General Questionnaire: Constraints identied as important by the participants (% of answers)
as ragi, pigeon pea and sorghum. Cotton and potato are mentioned only once, while sun ower is not mentioned at all. Farming constraints are depicted in Fig. 8.5. Not surprisingly, crop water stress is considered by a wide margin to be the most stringent constraint to take into account for marginal agriculture (27 % of expressed opinions). Pest and disease also rank high (at respectively 16 % and 18 %), while soil related concerns -physical properties and nutrient content- represent also a signicant group. Weather (temperature, humidity) and insolation are not considered important parameters. The results relative to the important parameters (Fig. 8.6) are consistent with those of the general questionnaire. The only notable difference is that temperature is now considered more important than soil moisture, which still remains the second most important parameter, however. Fig. 8.7 shows the parameter distribution per crop. This gure conrms that for each crop the parameters to take into account are the same, except pest, which is not considered to be an issue for ragi. The results about parameter variability and tolerated error are not depicted, because the results are are consistent with those of the general questionnaire. In other terms, users did not noticeably change their mind after using the web application. The next section clearly explains why.
128
Measurement Rate
12
10
Monthly Weekly Daily Hourly 5 minutes
0 Humidity Soil Water Content Soil Type Temperature Rain Fall Pattern
Figure 8.1: General Questionnaire Desired rate (y-axis: number of times a rate was chosen by the participants)
8.2.1.3
Comments
The participants wrote extensive comments. The time they spent reecting on the application seemed encouraging, as they pondered the usage of this technology by marginal farmers, commented on the technical accuracy required by such as system and elaborated on useful sensors that could be added to the data acquisition kit. A complete set of comments can be found in [Pan07]. We mention some characteristic remarks hereunder. 1. Several users questioned the capacity of farmers to act based on the environmental data collected. 2. Sensors for soil nutrient status are often mentioned as extra sensing devices. 3. Soil physical characteristics are considered important as well 4. Bio-sensors are mentioned 5. The accuracy of the system is mentioned as a constraint.
8.2.2
User Activity Logging
This is the set of meta-data that was generated by the logging of participants interactions with the web application. This part of the experiment led to inconclusive results. Out of the 30 participants, only six actually used the on-line application at some point. All of them were PhD students and post-docs. No senior researcher used spontaneously the on-line application.
129
Measurement Granularity
8 7 6 5 4 3 2 1 0 Humidity Soil Water Content Soil Type Temperature Rain Fall Pattern 100 meters 500 meters Kilometer Single Measure Meter
Figure 8.2: General Questionnaire Desired spatial granularity, per constraint (y-axis: number of times a granularity was chosen by the participants)
The participants who used the application did in average 3 queries to the system, mostly to look at the soil moisture status. This paradoxical disinterest for the on-line application is discussed in the Discussion undertaken in the next question. It made the debrieng meetings very important, in order to understand the mismatch between the interest manifested in the survey and the actual usage of the application.
8.2.3
Debrieng Meetings
Initially, the debrieng meetings were intended to gather the opinions of the participants in a more informal manner than during the experiment. However, in light of the mismatch mentioned in the previous section, they became a crucial element of the experiment. The goal was to nd out why the users had not used the application as expected, and to assess their real level of interest. The meetings were conducted with 8 Professors (out of the 10 who initially answered the survey). Additionally, we discussed the application with a top NGO executive, and a local ofcial of the agriculture department of Karnataka. Instead of asking these questions directly, which would have been likely to bias the answers, we chose to address concrete use cases. If people were able to come up with original use cases, that meant they had conducted a reection about the tool. Moreover, we could then talk concretely about upcoming partnerships, an extra- measure of their interest, and a critical conditions for the continuation of the project. The results were encouraging. We found four compelling use cases, and in each case a concrete interest in using the technology provided. Precise details about the requirements of each use case can be
130
Required Precision
12 10 8 1% 6 4 2 0 Humidity Soil Water Content Soil Type Temperature Rain Fall Pattern 5% 10% 30%
Figure 8.3: General Questionnaire Desired precision of the measure, per constraint (y-axis: number of times a precision was chosen by the participants)
found in [PRS08]. 8.2.3.1 Soil Science
Provided we can adapt nutrient sensors to the wireless nodes, there is research to be conducted in the response at the root zone to different strategies of nutrient application and irrigation. The main objective would be to observe the variation in nitrogen , phosphorus and potassium content, in the context of nutrient dynamics under a system of multiple crops and trees. Appropriate sensors are of interest in order to understand the dynamics of nutrients and soil moisture, soil PH, etc. Experiment would be meaningful if conducted in the following way: 4 ha divided in 4 plots, 10 sensors per plot. Sensors placed at 2 different depths Experiment should last 2-3 years minimum Bi-weekly measurements 1 8.2.3.2 Entomology
The observation of pests present in the crop eld shows that their activity depends on the weather, especially rain fall, soil moisture and soil temperature. There is a clear correlation between the rain patterns
For research purposes, having multiple measures per day would be useful, but that would not give meaningful results from an agriculture point-of-view, because the time-scale of farming operations is much larger.
1
131
Crop Choice
cotton 4% sorghum 12%
potato 4%
ragi 15%
groundnut 50%
pigeon pea 15%
Figure 8.4: Detailed Questionnaire Crops that are considered the most adapted to the region given the data provided by the application (% of the answers)
and the emergence of adults of the insects from the soil. This happens in a xed time of the year between the last week of April and the end of June. If there is rain before April 20th, then there is no emergence. Moreover, after October all the larvae enter pupation and emerge as adults after 20 days. If the soil moisture conditions are not favorable to them until the end of June, a large percentage of the population might die. Another pest, the Red Hairy Caterpillar, has a similar biology but 2 months later than the white grubs. Emergence happens in two cycles, one in early July and the second in late September. The hypothesis to verify is whether the insects activity depends on soil moisture evolution and accumulation of soil heat in the weeks prior to emergence. If the soil moisture conditions are not favorable to them during pupation, a large percentage of the population might die. Soil moisture, temperature sensors in specic regions of endemic populations of these pests (sampling various soil texture typologies) will help to investigate, understand their biology . This would make it possible to provide advance information on the intensity of pest damage to farmers.
132
Critical Constraints
30
25
20
15
10
0 Soil Moisture Disease Pest Nutrient Soil Weather Day length
Figure 8.5: Detailed Questionnaire Constraints to take into account in priority (y-axis: number of times a constraint was chosen by the participants)
The interviewee expressed keen interest on the usage of WSNs, in a rst stage to test the technologys reliability and effectiveness. 8.2.3.3 Crop Physiology
This use case is about the precise assessment of the ratio between the water that is transpired by the plant and the water that is evaporated, in other terms the plants water efciency. The possibility to achieve crop improvement through selection would have a positive impact on yields achieved in rain-fed farming. For this, it is necessary to test plants with different genotypes obtained by cross-breeding and to assess which one has the best ratio of biomass production per water used. The method used for this test today is gravimetric method. For this, plants in pots are used. The pots are lled daily with water up to eld capacity. The next day, they are weighted to assess the water lost in evapotranspiration. Bare plots are used as a benchmark to assess the effect of pure evaporation. The goal is to replace the gravimetric method with soil moisture sensors that would give directly the volumetric content of water of the soil. The tedious weighting procedure could then be avoided. A typical experience contains 120-200 pots. Each pot should contain 1-4 probes (ECH2O) connected to one wireless sensor. The experiment duration is typically 80 days from plant sowing (out of which 50 days of measures) The number of measures should be at least 4 per day, without upper limit. His expressed interest is high. The interviewee would like to conduct as soon as possible a rst experiment with 30 pots, as a proof of concept.
8.3. DISCUSSION
133
Critical Environmental Parameter

80 70 60 50 40 30 20 10 0 Temperature Soil Water Content Rain-Fall Pattern Soil Type Humidity Atmospherique Pressure VaporPressure Deficit
Figure 8.6: Detailed Questionnaire Physical parameters most important to monitor, regardless of the corresponding constraint (y-axis: number of times a constraint was chosen by the participants)
8.2.3.4
Water Management
For a large NGO conducting applied research in the area of rain-fed farming, wireless sensor networks are perceived as a promising validation tool. Two experiments are envisioned: 1. the possibility to increase soil water-retention capacity through different measures, such as fertilizer, mulching, etc. 2. assessing the efciency of underground drip irrigation. Here, the goal is to bring the water directly to the root zone of the plant. For both experiments, soil moisture is the ultimate measure of success or failure. A second use case distinct from research validation is information sharing in rural kiosks. The NGO implemented a network of internet kiosks in several districts of Karnataka. At the central server, extension specialists are analyzing data they receive from local kiosks in the villages. This data consists in questions and environmental information. Then, they redistribute the information they analyzed to the kiosks. Obtaining live data about soil moisture content for different types of soil would be an interesting complementary source.
8.3
Discussion
Potential users expressed keen interest in several cases. In particular, a major NGO working in the eld of dry-land farming throughout India expressed interest in the WSN technology. Such promising results
134
Critical Constraints per Crop

30
25
20
Potato Cotton Sorghum Ragi Pigeon Pea groundnut
15
10
0 soil moisture desease pest nutrient soil temperature daylength
Figure 8.7: Detailed Questionnaire: Specic characteristics per Parameter
must be tempered by the low response obtained by the application use, which we address in section 8.3.3.
8.3.1
Usefulness
From the questionnaires, there is a large consensus as to the usefulness of using ner-grained environmental data for rain-fed agriculture. The level of detail, at which scientists answered the on-line survey indicates a high level of interest and curiosity on their part. Such an interest was already perceptible at the inception of the project. However, the creation of precise use case was not possible then. This gap was lled during the individual interviews with scientists coming from academia, as well as the non-governmental sector. With four precise use cases and potential partners clearly identied, the initiative is now in the hands of the information and communication systems specialists. One central question is the potential of information-sharing with farmers. Will the results ever leave the lab and scientic reports to materialize in the eld? According to our interviews with government ofcials [PRS08], the Indian institutional framework is very clear: The agriculture scientists are expected to provide scientic evidence of phenomena, to investigate preventive or corrective actions when appropriate, and to publish recommendations that are used by the agriculture department to relay information to the public (which is referred to as extension work). The case of the non-governmental sector is different. Large NGOs are conducting applied research aimed at improving farming practices in dry-land management. In this case, the scientists work directly in contact with the farmer. This makes them privileged partners.
8.3. DISCUSSION
135
8.3.2
Usability
New types of sensors were mentioned in the course of the experiment. Most prominently, in-situ chemical sensors that could sense the concentration of nutrients (nitrogen, phosphorus, potassium) in the soil are mentioned repeatedly. The development of such sensors is still at an experimental stage, but some recent advances have been made for low-cost sensors using ion-selective electrodes [KHSM07]. The accuracy is a recurring concern. However, the exact precision to which sensors need to operate is still an open question. In general, it is more important to be roughly right than precisely wrong. Parameters linked with the soil are pinpointed as having a high space variability. As a consequence, any high-precision technology that comes at a price such that it is not possible to diversify the readings is not going to be usable. With low cost sensors, a relatively high error can be compensated by spatial diversity, which allows for the statistical elimination of outlying measurements. The sampling period of the data (time variability) is still the object of uncertainty. Measures shorter than a day are not a priori taken into consideration by scientists in the framework of an applied research for agriculture. At present, the researchers mostly want daily data, since they use data of similar granularity obtained from conventional measurements without sensors. During debrieng sessions, however, we gathered evidence that the responses are likely to change with time spent on reection and experience with high-resolution data. Indeed, when prompted by one of the authors with background in agriculture about possible uses for research, the participants acknowledge in a majority of cases that such data could be used in the framework of their research. It appears that certain elements of the current responses, particularly those related to time resolution, will change after some experience and / or contemplation on use of high time resolution data. Researchers have always viewed data gathering as a major constraint in research design and conceptualization. The current experiment presents a completely contrasting situation with the provision of very rich data in both time and space for parameters of interest. It is in this light that we suggest a co-learning process for agriculture researchers and sensor technology providers to evolve better and meaningful use cases. On the technical side, more work has to be done to integrate new, more complex sensors to the current data acquisition kit.
8.3.3
Use
We tried to provide possible explanations about the paradoxical low level of use recorded during the experiment. It cannot be ruled out that this reects the actual disinterest of the participants. However, that would be contradictory with all the other results of the survey. Moderate computer literacy is a possible explanation. A similar result was obtained during a previous experiment conducted in 2006, specically about interface design [Pan06]. At that time, we had seen a clear difference of behavior between scientists used to working with environmental sensors and computers, and those who did not have this expertise. As a consequence, we sought to improve the interface, but such a gap remains, as illustrated by the difference in handling the application from senior scientists and their younger, computer-savvy students. However, this is only a partial explanation to the observed phenomenon. As a matter of fact, despite their computer literacy, the students and post-docs did not extensively use the application. A posteriori, this can be explained by the difculty in nding a one-size-ts-all scenario for the participants of the survey. We had not realized how diverse the concerns of agricultural scientists could be.
136
8.3.4
Sectoral Analysis
The response that we received from the government ofcial illustrates the functioning of the Indian public sector when it comes to design and implementation of policies. The academic institutions are expected to provide scientic evidence of phenomena, to investigate preventive or corrective actions when appropriate, and to publish recommendations. The administrative role is one of information spreading and relaying to the public (which is referred to as extension work), and of control. A valid concern to raise at this point is to question how the information is collected and forwarded in the other direction, in our case, how do the needs and wishes of the farmers come to the scientics desk. But we did not investigate this question. The case of the non-governmental sector is different. Large NGOs do not restrict themselves to extension and implementation. They are conducting applied research aimed at improving farming practices in dry-land management. In this case, the scientist works directly in contact with the farmer. This makes him or her a privileged partner in the eld. We did not extend our survey to the industry of agriculture inputs or to corporate agriculture, although these two sectors are likely users of the wireless sensor networking technology. In the rst case, there is currently a controversy surrounding the effects of large seeds providers on the livelihood of small and marginal farmers in India. It seemed premature to contact such rms in the framework of a rural development project. In the second case, the type of agriculture practiced (mechanized farming, ondemand irrigation, precision agriculture with high added-value) was considered too different from our focus of interest, namely rain-fed farming.
Chapter 9
Building a Knowldedge Society with the Use of WSNs?

Can an ICT engineering project contribute to building local ICT capacities for the Information Society in a developing country context? How can it achieve this? We address this question by applying a set of theoretical concepts to the COMMON Sense Net project. We particularly look at human capacity building through participation as a form of ICT education. We believe that rural communities and developing regions ask for innovative methods that go beyond traditional classroom learning. Therefore we outline three categories of ICT capacities, introduce a process of knowledge and capacity creation, study apprenticeship as a form of knowledge and capacity appropriation and analyze it all in the execution of the COMMON Sense Net project.
9.1
Experimental Technology for Social Change?
The COMMON-Sense Net project deals with an experimental technology: wireless sensor networks. As such, it is likely that it will not lead immediately to concrete economically protable applications. However, as Brewer et al. [BDD+ 05] reected about technology needs, (...) Western market forces will continue to meet the needs of developing regions accidentally at best. In the same spirit, we advocate the importance of exploring the potential of an emerging technology - sensor networks - in the particular case of rural development, in order to take the ecological, social, cultural and economic conditions of developing countries into account in the design of hardware and software platforms, and to develop applications that are well adapted to this context. Several authors have discussed the formidable potential of ICTs to foster development in the South (Heeks, 2002 [Hee01], [Neg98], [Wal01], etc.). They show that ICTs can be applied to a wide spectrum of different areas to leverage development projects. Such literature exclusively deals with Person-toPerson ICTS, namely systems that interconnect people, such as phones or Internet-enabled computers. We address the value and the issues of another important area of ICT for development that in our opinion is still rather poorly researched: Environment-to-Person Information Systems (EPISs). These are systems that collect environmental information and communicate them to machines and people. With the goal to improve living conditions, this sub-area of ICTs helps individuals and communities develop a better knowledge of the physical parameters that make up their environment (e.g. pollution monitoring, agricultural management, etc.). We argue that development projects that focus on designing and building the tools for collecting and disclosing environmental information have a direct impact through 137
138
CHAPTER 9. BUILDING A KNOWLDEDGE SOCIETY WITH THE USE OF WSNS?
the artifacts they build, but can also have an indirect impact through the ICT capacities they create via dynamic knowledge generation. We will analyze this hypothesis in the COMMON-Sense Net project for ICT-based agricultural water management in rural India, which we introduce in the following lines. It is difcult at this stage of the project to talk about demonstrable gains, since we are using a technology still in its maturation phase and not yet widely available on the market. As a consequence, rather than studying economic feasibility, we aim at verifying the hypothesis that resource-poor farmers can take benet from a system similar to ours. This being said, it is important to keep in mind the ultimate benets that local farmers will get from the system. The involvement of the agronomical scientic community and the ability to disseminate the obtained results to the population in a credible way are the key points. This is no simple task, but leveraging on existing experience and success stories is possible (Sakthivadivel et al., 2001) [SLAH01].
9.2
Design/Implementation Gaps
Heeks [Hee01] argues that the failures of information systems projects in developing countries are often caused by design-actuality gaps. Country context mismatches (in terms of institutions, infrastructures etc.) as well as hard-soft gaps (rational design versus cultural and political actuality) play a role all the more important if the system was designed in an industrialized context. To summarize, failures can generally be explained by the distance (geographical, cultural or socioeconomic) between the designers of the system and its intended community of users. As stated above, the CSN project uses participatory design extensively, which mitigates this risk. Heeks warns, however, that participatory design in itself is no guarantee for success in developing countries, since these techniques have usually been developed in and for industrialized countries organizations. A lesson to be drawn is that a participatory approach in a developing country is instrumental to success if and only if it integrates a tool to bridge the contextual gap between design and use. In order to bridge this gap, Heeks advocates the usage of hybrids, namely individuals who understand both the alien worlds of the community of users and of the community of designers/builders of the artifact. In the CSN case, the hybrid is a local farmer who is also an agronomist and who is familiar with information systems for having worked with them for more than a decade. Despite all these precautions, the success at bridging the cultural gap with farmers gave mixed results. After an initial positive response from farmers, who willingly participated to the user survey described in Chapter 4, the interest of the rural community faltered in the absence of immediate impact of the CSN prototype on their livelihood. As explained in Chapters 6 and 8, we eventually refocused our efforts towards the scientists who had participated to the initial design of the system. Possible design/implementation gaps remained with this new user target, but they proved to be addressable with the action of our hybrid, the organization of regular meetings and the presentation of environmental data that are immediately interpretable by scientists. This entertained a sustained interest in our system and allowed a progressive understanding of the wireless sensor networks technology and its potential applications.
9.3
Knowledge Creation, Context and Knowledge Assets
Nonaka, Toyama et al. [NOO+ 00] outline four elements in the process of knowledge creation: The knowledge creation cycle, a shared context for knowledge creation and the circulating knowledge assets.
9.3. KNOWLEDGE CREATION, CONTEXT AND KNOWLEDGE ASSETS
139
Figure 9.1: Knowledge creation cycle (Nonaka, Toyama et al. 2000)
The rst element of knowledge creation, which Nonaka, Toyama et al. call SECI (acronym for Socialization, Externalization, Combination, Internalization), functions like a spiral describing the interactions between actors in order to transmit knowledge in it tacit or explicit form, and the actions of individuals or groups in order to translate knowledge from tacit to explicit, and vice-versa. This process follows four modes feeding each-other in a spiral (Fig. 9.1). First, the socialization process of transmitting and converting new tacit knowledge through shared experiences. Socialization typically occurs in a traditional apprenticeship. Second, the externalization process of articulating tacit knowledge into explicit knowledge. The success of such a conversion depends on the sequential use of metaphor, analogy and model. Third, the combination process of converting explicit knowledge into more complex and systematic sets of explicit knowledge. Finally, the internalization process of embodying explicit knowledge into tacit knowledge. In the COMMON Sense Net we aim at analyzing these four phases with regard to capacity building. The rst phase of socialization concerns discussions between agricultural scientists to emphasize their desires and aspirations regarding agricultural water management. The second phase of articulating their desires concerns discussions between scientists and technical specialists (the hybrid mentioned in the previous section and the system designers). The third phase of combining knowledge concerns the connection of the aspirations of the scientists with technical knowledge in rural engineering, water management and ICT in order to design a system. The fourth phase consists of applying the system in a controlled environment, then extending the acquired knowledge to the scientists. The loop starts over with phase one, with scientists feedback and/or recommendations on the proposed information. The second element of knowledge creation that Nonaka, Toyama et al. mention is the context, which they call Ba, a Japanese concept that roughly translates into the English word place. This is particularly interesting for the case of ICT-for-Development projects such as COMMON Sense Net. It is often stressed that the particularities of the developing country context (stakeholders and environment) and technology development are highly dependent [BS98]. The COMMON Sense Net project covers a wide diversity of contexts, which must be carefully considered during the execution of the project and the analysis of the ICT capacities. Among several other diverse contexts most importantly gure the involved
140
rural Indian village, the laboratories at EPFL in Switzerland and the laboratory at the University of Agriculture in Bangalore. The last element of knowledge creation that Nonaka, Toyama et al. mention are the knowledge assets. Knowledge assets are the inputs, outputs and moderating factors of the knowledge-creating process. Those assets are experiential (e.g. skills, know-how), conceptual (e.g. concepts, designs, methods), systemic (technological platforms, manuals, libraries of software components) and routine-based (e.g. organizational routines). All these assets need to be mapped in order to be usable. This mapping process is at the core of the dynamic knowledge creation. In the COMMON Sense Net project we particularly aim at observing the interaction between the tradition skills and know-how of the agricultural scientists in terms of agricultural water management and the modern concepts and ICT systems brought in through the project.
9.4
Apprenticeship & Participatory Methods to Develop ICT Capacities
In the previous sections we presented the three axes along which capacities are built for creating an Information Society and argued that analyzing the knowledge creation process was central to understanding capacity building. In this section we study apprenticeship as the main mechanism through which we believe ICT knowledge and capacity will be created in the COMMON Sense Net project. We dene apprenticeship as a situation in which a learner works intensively with an expert to learn a new task that may necessitate the understanding of new concepts. We present it as an alternative to traditional classroom learning that can be very effective to instrumentalize knowledge as capacity in rural communities of developing regions. Particularly for Environment-to-Person Information Systems a participatory approach seems an appropriate tool that can help overcome some underlying barriers to the development of innovative environmental technologies [FH01], [Sot03]. The question is how much of a spill-over effect participatory learning can have on the development of more general ICT capacities. Misconceptions, what Heeks calls design-actuality gaps, namely the gap between the technocrats who design systems using scientic knowledge and the local context characterized by irrational cultural features, seem to be at the root of most failures for Information Systems in developing countries. This recurring aw calls for the concept of participatory design and implementation. In participatory approaches, the end-user is constantly involved in the design and assessment of the product or service being developed for him. Cooper [Coo00] emphasizes the role that group working and end-user involvement can play in a successful implementation. However, Heeks [Hee01] warns that this is no guarantee to success in developing countries, since these techniques have usually been developed in and for industrialized countries organizations. A lesson to be drawn is that a participatory approach in a developing country is instrumental to success if and only if it integrates a tool to bridge the contextual gap between design and use. We claim that the resort to apprenticeship is such a tool. Freemans denition of apprenticeship is learning by doing [Fre02]. Adapting this denition to our context and trying to be more specic, we consider apprenticeship as the process by which a person acquires a new knowledge or skill by imitation and interaction with someone who possesses that skill or knowledge already, rather than in a formal way in the classroom with a teacher. Our hypothesis is that there are some aspects of apprenticeship that make it particularly suited in the acquisition and integration of radically new paradigms of knowledge. It is a self-organized process in which every individual takes ownership of the knowledge he or she is acquiring. Not relying on formal teaching, it can be more integrated in the social structure and possibly more equitable since people not
9.5. FROM THEORY TO PRACTICE
141
having the time, the resources or the will to attend classes can be reached through it. Solving concrete issue one after another insures that people are interested in the process and increases the likelihood of them persevering in the endeavor. It allows for unexpected forms of organization to develop and is adaptive. Ultimately, it is empowering. It reserves surprises for the teacher as well as for the student. The challenge lies in bootstrapping the process, in other terms in convincing the local stake holders that a new formerly unheard of form of knowledge can be of value to them. One possibility is nding a local partner who speaks both languages, who understands and uses the indigenous knowledge, but masters also the language of technology and science. At this stage, a more formal teaching approach may be needed in order to form such a partner. But here again, knowledge exchange, rather than knowledge provision proves to be a key-concept in integrating new forms of knowledge in traditional societies without losing the value of what indigenous knowledge brought to the community in the rst place. The COMMON-Sense Net project is proposing to local stake-holders an ICT system that will help them accomplishing more efciently daily tasks in accordance with specications they laid down themselves (in our case the information requirements for agricultural management).
9.5
From Theory to Practice
During the four years that the project lasted, we completed one full cycle of the Knowledge Creation spiral. The initial phase of socialization happened through the initial survey about information requirements, where several meetings among farmers were organized in the villages. The second phase of externalization provided a shared experience where communication scientists, agriculture specialists and farmers exchanged information with the facilitation of our hybrid. In particular, the expectations of farmers were clearly identied, the constraints of rain-fed farming and research were understood, and the local stake-holders were presented for the rst time with a disruptive technology through several participatory meetings. Farmers became less involved at this point, because it was clear that what they expected were concrete demonstrable results before considering novel ways to practice agriculture. The scientists, although they did not know anything about the technology and had problems initially to conceptualize it, showed keen interest. Discussing the general case of rain-fed agriculture with them, we were able to dene very general use cases for an environmental monitoring system, but not to lay down precise requirements and conduct targeted experiments. This was enough, however, to build and test a prototype, which we did in collaboration with our hybrid. This took a full two years before the system could be considered operational. We reected on the many challenges involved in this phase in Chapter 6. Then we came back to the scientists with early results for a new phase of socialization and externalization, which is described in Chapter 8. Seeing an interactive map of the deployment and being able to look at extensive environmental data, the scientists were able to rene greatly their use case and determine constraints for our data collection system (parameters, their time and space granularity and their error-tolerance). In dening the use cases, the scientists showed an improved understanding of WSN and their capabilities. We are now at the combination phase of the second cycle. If deployments of the tool can be nalized in a controlled environment, the role of apprenticeship, already apparent in the use of our early system - where scientists developed their understanding of ne-grained environmental monitoring through the observation of actual data - will become fundamental in the assessment of the system, since scientists will have to operate wireless sensor networks and process their data themselves.
142
Conclusion
Wireless Sensor Networks are emerging as a fundamental block of the Internet of Things that is likely to emerge in the years to come in highly technological societies. In this thesis, we broadened the scope and sought to nd relevant applications of this technology for issues specic to developing countries. We began by setting the context and showing the strengths of wireless sensor networks to tackle problems linked with the Millennium Goals. Then, we presented an on-going research and implementation work on an environmental monitoring system primarily aimed at resource-poor farmers of developing countries. Using participatory design and a rigorous technical approach, we developed an integrated wireless sensor-network system that we tested in the eld. This system was deployed for an extended period of time in a village of Karnataka (India). Based on our experience, we highlighted the challenges that await similar initiatives. For the benet of the research community, we presented our environmental monitoring toolkit and the lessons learned in a rural deployment over two years. This lead us to design and implement Sensor-Tune, a sonicationbased deployment-support tool that would enable non-specialists to handle wireless sensor networks in the eld. We also came to the conclusion that using the toolkit in a controlled or semi-controlled environment was the most promising approach for the time being. Accordingly, we proceeded to a user experiment with scientists from the University of Agricultural Sciences in Bangalore. Promising use cases and user interest have been clearly identied at this point. This work is novel because it is the rst example of an actual wireless sensor network in rural India and, to the best of our knowledge, in developing regions as a whole. It is also the rst documented example of an holistic approach in designing a WSN with a particular purpose through a participatory method including the main stakeholders - in our case, farmers, scientists and NGOs alike. As such, it is meant to be a thorough case study outlining a reusable methodology and a reusable platform. We also reported honestly all the pitfalls that could appear in similar projects.
Future Work
Despite the successful design and implementation of an environmental monitoring toolkit, more work needs to be done in order to assess the usefulness of this toolkit in real conditions. Initial tests in the elds led us to consider a more conservative strategy by putting our toolkit at the disposal of scientists. The initial response, based on environmental data collected in the eld, was positive. However, successfully implementing the use cases isolated with the scientists will require more partnerships and deployments spanning one year or more. Ultimately, the technoology will have to be brought back to the cropping eld, with convincing results to show to the farmers. 143
144
CONCLUSION
Future work also includes the formal testing of our deployment-support tool in the Indian context (typically with the agriculture scientists mentioned above). We successfully put the deployment of a WSN in a challenging indoor setting into the hands of non-specialists. The use of a sonication-based interface proved to be useful, but only a full deployment will be the measure of the tool. Finally, the other environmental challenges identied in this thesis should be investigated in a similar project as COMMON-Sense Net. Bangalore itself is an ideal candidate for the air pollution monitoring example.
References
[AHS06] K. Aberer, M. Hauswirth, and A. Salehi. Global Sensor Networks. EEE Communications Magazine, special issue on Advances in Service Platform Technologies for Next Generation Mobile Systems, 2006. A. Arora, R. Ramnath, and E. Ertin. Exscal: Elements of an extreme scale wireless sensor network, 2005. I.F. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci. A Survey on Sensor Networks. IEEE Communications Magazine, Vol. 40, No. 8, 2002. G. S. Abawi and T. L. Widmer. Impact of soil health management practices on soilborne pathogens, nematodes and root diseases of vegetable crops. Applied Soil Ecology, 15(1):3747, 2000. A. Baggio. Wireless sensor networks in precision agriculture. In The REALWSN05 Workshop on Real-World Wireless Sensor Networks, IVA, Stockholm, June 2005. J.A. Ballass. Delivery of Information Through Sound. In G. Kramer, editor, Auditory Display: Sonication, Audication, and Auditory Interfaces, volume XVII of SFI Studies in the Sciences of Complexity. Addison Wesley, 1994. J. Burrell, T. Brooke, and R. Beckwith. Vineyard Computing: Sensor Networks in Agricultural Production. IEEE Pervasive Computing, 2004. G. Barrenetxea, O. Couach, F. Ingelrest, M. Krichane, K. Aberer, M. Parlange, and M. Vetterli. SenSorscope : an Environmental Monitoring Network. Water Resources Research Journal, 2008. C. Beckmann, S. Consolvo, and A. Lamarca. Some assembly required: Supporting enduser sensor installation in domestic ubiquitous computing environments. UbiComp 2004: Ubiquitous Computing, pages 107124, 2004. E. Brewer, M. Demmer, B. Du, M. Ho, M. Kam, S. Nedevschi, J. Pal, R. Patra, S. Surana, and K. Fall. The case for technology in developing regions. IEEE Computer magazine, 38(6):2538, 2005. P. Buonadonna, D. Gay, J. M. Hellerstein, W. Hong, and S. Madden. Task: Sensor network in a box. In EWSN, 2005. 145
[ARE05]
[ASSC02]
[AW00]
[Bag05]
[Bal94]
[BBB04] [BCI+ 08]
[BCL04]
[BDD+ 05]
[BGH+ 05]
146
REFERENCES
[BHSA+ 07] G.J. Bishop-Hurley, D.L. Swain, D.M. Anderson, P. Sikka, C. Crossman, and P. Corke. Virtual fencing applications: Implementing and testing an automated cattle control system. Computers and Electronics in Agriculture, 56(1):1422, 2007. [BIS+ 08] G. Barrenetxea, F. Ingelrest, G. Schaefer, M. Vetterli, O. Couach, and M. Parlange. SensorScope: Out-of-the-Box Environmental Monitoring. In The 7th International Conference on Information Processing in Sensor Networks (IPSN 2008), St. Louis, Missouri, USA, 2008. S. Barrass and G. Kramer. Using sonication. Multimedia Syst., 7(1):2331, 1999. R. Barker and F. Molle. Evolution of Irrigation in South and Southeast Asia. Technical report, International Water Management Institute, 2004. E. Basha and D. Rus. Design of early warning ood detection systems for developing countries. In Proceedings of the Conference on Information and Communication Technologies and Development, Bangalore, India, December 2007. T. E. Brabben. Affordable irrigation technology: prospects for smallholders in South Africa. In SABI Congress, 2001. S. Biggs and G. Smith. Beyond methodologies: Coalition-building for participatory technology development. World Development, 26(2), 1998. P. Basu and P. Srivastava. Scaling-up Micronance for Indias Rural Poor. National Council of Applied Economic Research, India, 2005. R. Beckwith, D. Teibel, and P. Bowen. Unwired wine: Sensor Networks in Vineyards. In IEEE International Conference on Sensors, 2004. N. Burri, P. von Rickenbach, and R. Wattenhofer. Dozer: ultra-low power data gathering in sensor networks. In Tarek F. Abdelzaher, Leonidas J. Guibas, and Matt Welsh, editors, IPSN, pages 450459. ACM, 2007.
[BK99] [BM04] [BR07]
[Bra01] [BS98] [BS05] [BTB04] [BvRW07]
[CABM03] D. De Couto, D. Aguayo, J. Bicket, and R. Morris. A high-throughput path metric for multi-hop wireless routing. In MOBICOM, 2003. [Con] [Coo00] [Cro] [CTT06] [CW03] [DAI] Control4. Home Automation System. http://www.control4.com/. R. B. Cooper. Information technology development creativity:A case study of attempted radical change. MIS Quarterly, 24, 2000. Crossbow. Eko Environmental Monitoring System. http://www.xbow.com/eko/. Documentation available at
F. Chau, J. C. Tanner, , and S. Tippu. Country & Wireless. Wireless Asia, May 2006. J. E. Carroll and W. F. Wilcox. Effects of Humidity on the Development of Grapevine Powdery Mildew. Phytopathology, 93(9):11371144, September 2003. OBS-3+ turbidity sensor. Documentation available at http://www.d-a-instruments.com.
REFERENCES
147
[DC07]
H. Diwikara and M.G. Chandrakanth. Beating negative externality through groundwater recharge in India: a resource economic analysis. Environment and Development Economics, Cambridge University Press, 12:271296, 2007.
[DKMU03] A. Dorward, Jonathan Kydd, J. Morrison, and I. Urey. A Policy Agenda for Pro-Poor Agricultural Growth. World Development, 32:7389, 2003. [D.R05] D.R. Upchurch and J.R. Mahan and D.F. Wanjura and J.J. Burke. Concepts in Decit Irrigation: Dening a Basis for Effective Management. In Environmental & Water Resources Institute Conference, 2005. P. Droogers, D. Seckler, and I. Makin. Estimating the Potential of Rain-fed Agriculture. Technical report, International Water Management Institute, 2001. W. Dargie and M. Zimmerling. Wireless sensor networks in the context of developing countries. In IFIP World IT Forum (WITFOR), 2007. P. Juang et al. Energy-Efcient Computing for Wildlife Tracking: Design Tradeoffs and Early Experiences with ZebraNet. In ASPLOS X, 2002. ECH2O Soil Moisture and Microclimate Monitoring. http://www.ech2o.com/. Editorialist. The World Food Crisis. New York Times, 2008. D. Estrin, L. Girod, L. Pottie, and M. Srivastava. Instrumenting the world with wireless sensor networks. Acoustics, Speech, and Signal Processing, 2001. Proceedings. (ICASSP 01). 2001 IEEE International Conference on, 4:20332036 vol.4, 2001. M. Fernstrom, Eoin Brazil, and Liam Bannon. Hci design and interactive sonication for ngers and ears. IEEE Computer Society, 2005. R. Fonseca, O. Gnawali, K. Jamieson, S. Kim, P. Levis, and A. Woo. The Collection Tree Protocol (CTP). TinyOS Extension Proposal 123. V1.8. B. Fleming and D. Henkel. Community-Based Ecological Monitoring: A Rapid Appraisal Approach. Journal of the American Planning Association, 67, 2001. FIBOX Enclosures. Documentation available at http://www.box.com/. W.T. Fitch and G. Kramer. Sonifying the Body Electric: Superiority of an Auditory over a Visual Display in a Complex Multivariate System. In G. Kramer, editor, Auditory Display: Sonication, Audication, and Auditory Interfaces, volume XVII of SFI Studies in the Sciences of Complexity. Addison Wesley, 1994. N. Fodor and G.J. Kovac. Sensitivity of 4m maize model to the inaccuracy of weather and soil input. Applied Ecology and Environmental Research, 1, 2003. Food and Agriculture Organization of the United Nations. Water at a glance, 2006. T. Fuller and A.C. Revkin. Climate plan looks beyond Bushs tenure. New York Times, 2007. Documentation available at
[DSM01] [DZ07] [ea02] [ECH] [Edi08] [EGPS01]
[FBB05] [FGJ+ ] [FH01] [FIB] [FK94]
[FK03] [FotUN06] [FR07]
148
REFERENCES
[Fre02]
H. A. Freeman. Comparison of Farmer-Participatory Research Methodologies: Case Studies in Malawi and Zimbabwe. Technical report, International Crops Research Institute for the Semi-Arid Tropics (ICRISAT), 2002. S. Gadgil, Y. P. Abrol, and P.R.S. Rao. On growth and Fluctuation of Indian Foodgrain Production. Current Science, 1999. M. Gilx and A. L. Couch. Peep (the network auralizer): Monitoring your network with sound. In LISA 00: Proceedings of the 14th USENIX conference on System administration, pages 109118, Berkeley, CA, USA, 2000. USENIX Association. G. Geiger. Pda - real time signal processing and sound generation on handheld devices. In International Computer Music Conference (ICMC), 2003. A.J. Gijsman, S.S. Jagtap, and J.W. Jones. Wading through a swamp of complete confusion: how to choose a method for estimating soil water retention parameters for crop models. European Journal of Agronomy, 18, 2003. M. Glanz. Climate Affairs - A Primer. Island Press, Washington, USA, 2003. S. Gadgil, P.R.S. Rao, and K.N. Rao. Use of climate information for farm-level decision making: rainfed groundnut in southern India. Agricultural Systems, 74, 2002. B. A. George, S. A. Shende, and N. S. Raghuwanshi. Development and testing of an irrigation scheduling model. Agricultural Water Management, 46(2), 2000. M. Hatler and C. Chi, editors. Wireless Sensor Networks for the Oil & Gas Industry. ON World, 2005.
[GAR99] [GC00]
[Gei03] [GJJ03]
[Gla03] [GRR02] [GSR00] [HC05]
[HDM+ 06] I. Hussain, D.Wijerathna, S. Arif Murtiningrum, A. Mawarni, and Suparmi. Irrigation, productivity and poverty linkages in irrigation systems in java, indonesia. Water Resources Management, 20(3):313, 2006. [Hea] [Hee01] [HF04] [HKV02] Home Heartbeat. Home Awareness System. http://www.homeheartbeat.com/. Documentation available at
R. Heeks. Information Systems and Developing Countries: Failure, Success and Local Improvisations. The Information Society, 2001. M. Ho and K. Fall. Delay Tolerant Networking for Sensor Networks. In 1st IEEE Conference on Sensor and Ad Hoc Communications and Networks, 2004. G. Haq, C. Kim, and H. Vallack. Benchmarking Urban Air Quality Management and Practice in Major and Mega Cities of Asia. Technical report, Air Pollution in Mega Cities of Asia (APMA), 2002. M. Hirafuji, S. Ninomiya, T. Kiura, T. Fukatsu, H. Hu, H. Yuoichi, K. Tanaka, K. Sugahara, T. Watanabe, T. Kameoka, A. Hashimoto, R. Ito, R., and H. Shimamura. Field server projects. In SAINT-W 07: Proceedings of the 2007 International Symposium on Applications and the Internet Workshops, page 75, Washington, DC, USA, 2007. IEEE Computer Society.
[HNK+ 07]
REFERENCES
149
[HNM00]
G.L. Hammer, N. Nicholls, and C. Mitchell, editors. Applications of seasonal climate forecasting in agriculture and natural ecosystems: The Australian experience. Kluwer Academic Publishers, Dordrecht, The Netherlands, 2000. L. Horst. The dilemmas of water division: Considerations and criteria for irrigation system design. Technical report, International Water Management Institute, 1998. A. A. A. Ibrahim and A. Hunt. An hci model for usability of sonication applications. In K. Coninx, K. Luyten, and K. A. Schneider, editors, TAMODIA, volume 4385 of Lecture Notes in Computer Science, pages 245258. Springer, 2006. InteleSense. Public Health Monitoring http://www.intelesense.net/deployments/vietnam. in Vietnam and Ethiopia.
[Hor98] [IH06]
[Int] [ion] [IP] [IP0] [Irra] [Irrb] [Jam] [JM02]
Ionode Pty. Ltd. Documentation available at http://www.ionode.com.au/. Ingress Protection Rating. http://en.wikipedia.org/wiki/IP Code. Documentation available at
IP01. Cold Chain Management System. Documentation available at http://www.ip01.com/. Irrometer. Watermark Sensor. http://www.irrometer.com/agcat.htm. Documentation available at
Irrometer. Wireless Meter Reading. http://www.wellspringwireless1.com/pdfs/aqura.html. H.S. Jamadagni. COMMONSense Net Internal Report. T. M. Johnson and M. Margalho. Performance Evaluation of Wireless Transmissions in an Amazonian Climate. In IEEE Wireless Communications and Networking Conference (WCNC), 2002. M. Jurriens, P. O. Mollinga, and P. Wester. Scarcity by design. Protective irrigation in India and Pakistan. International Livestock Research Institute, Nairobi, 1996. W. I. Jones. The World Bank and irrigation. Technical report, Operations Evaluation Department, the World Bank, 1995. D. Jea, A. A. Somasundara, and M. B. Srivastava. Multiple controlled mobile elements (data mules) for data collection in sensor networks. In IEEE/ACM International Conference on Distributed Computing in Sensor Systems (DCOSS), Marina del Rey, California, USA, June 2005.
[JMW96] [Jon95] [JSS05]
[KHSM07] H.J. Kim, J.W. Hummel, K.A. Sudduthb, and P.P. Motavalli. Simultaneous Analysis of Soil Macronutrients Using Ion-Selective Electrodes. PhD thesis, 2007. [KL07] Eiman Kanjo and Peter Landshoff. Mobile phones to monitor pollution. IEEE Distributed Systems Online, Urban Computing and Mobile Devices, 8(7), 2007.
[KWB+ 97] G. Kramer, B. Walker, T. Bonebright, P. Cook, J. Flowers, N. Miner, and Neuhoff. Sonication report: Status of the eld and research agenda. Prepared for the National Science Foundation by members of the International Community for Auditory Display, 1997.
150
REFERENCES
[KYH+ 05]
K.Lin, J. Yu, J. Hsu, S. Zahedi, D. Lee, J. Friedman, A. Kansal, V. Raghunathan, and M. Srivastava. Heliomote: enabling long-lived sensor networks through solar energy harvesting. In SenSys 05: Proceedings of the 3rd international conference on Embedded networked sensor systems, pages 309309, New York, NY, USA, 2005. ACM. M.J. LaGier, C.A. Scholin CA, J.W. Fell, J. Wang, and K.D. Goodwin. An electrochemical RNA hybridization assay for detection of the fecal indicator bacterium Escherichia coli. MARINE POLLUTION BULLETIN 50, November 2005. J. Luo and J.-P. Hubaux. Joint Mobility and Routing for Lifetime Elongation in Wireless Sensor Networks. In Proc. of the 24th IEEE INFOCOM, 2005. B. Liu, Z. Liu, and D. Towsley. On the Capacity of Hybrid Wireless Networks. Infocom, 2003.
[LCF+ 05]
[LH05] [LLT03]
[LNTN+ 07] S. G. Lemos, A. R. Noguera, A Torre-Neto, A. Parra, and J. Alonso. Soil calcium and ph monitoring sensor system. Journal of Agricultural and Food Chemistry, 55:46584663, 2007. [LPP+ 06] J. Luo, J. Panchard, M. Piorkowski, M. Grossglauser, and J.-P. Hubaux. MobiRoute: Routing towards a Mobile Sink for Improving Lifetime in Sensor Networks. In the 2nd IEEE/ACM DCOSS, 2006. H. Matthiesen. In situ measurement of soil ph. Journal of Archaeological Science, 31(10):13731381, 2004. M. Munasinghe and W. Cruz. Macroeconomics and the Environment. P. McCully. Silence drivers: The ecology and politics of large dams. Zed Books, London, 1996. R. Minghim and A.R. Forrest. An Illustrated Analysis of Sonication for Scientic Visualisation. In 6th IEEE Visualization Conference (VISUALIZATION 95), 1995.
[Mat04] [MC] [McC96] [MF95]
[MHH+ 96] R.L. McCown, G.L. Hammer, J.N.G Hargreaves, D.P. Holzworth, and D.M Freebairn. APSIM: a Novel Software System for Model Development, Model Testing and Simulation in Agricultural Systems Research. Agricultural Systems, 50(3):255271, 1996. [Mis06] [MLV03] [MMea03] [Moo] [MPS+ 02] S. Mishra. Suicide of Farmers in Maharashtra. Indira Gandhi Institute of Development Research, 2006. M. Miller, W. List, and N. Vaidya. A Hybrid Network Implementation to Extend Infrastructure Reach. Technical Report, University of Illinois at Urbana-Champaign, 2003. D. Malandrino, D. Mea, and A. Negro et al. Nemos: Network monitoring with sound. ICAD, 2003. P. Moody. A. Mainwaring, J. Polastre, R. Szewczyk, D. Culler, and J. Anderson. Wireless Sensor Networks for Habitat Monitoring. ACM International Workshop on Wireless Sensor Networks and Applications, 2002.
REFERENCES
151
[MS02] [MTY05]
R.B. Matthews and W. Stephens, editors. Crop-Soil Simulation Models: Applications in Developing Countries. CaB Intl, 2002. A. Mermoud, T. Tamini, and H. Yacouba. Impacts of different irrigation schedules on the water balance components of an onion crop in a semi-arid zone. Agricultural Water Management, 77, 2005. N. Negroponte. The Third Shall Be First. Wired, 1998. H. Newton. Newtons Telecom Dictionary. CMP Books, 20 edition, 2004. Suman Nath, Jie Liu, and Feng Zhao. Sensormap for wide-area sensor webs. Computer, 40(7):9093, 2007.
[Neg98] [New04] [NLZ07]
[NMDvI07] K. S. Nemali, F. Montesano, S. K. Dove, and M. W. van Iersel. Calibration and performance of moisture sensors in soilless substrates: Ech2o and theta probes. Scientia Horticulturae, 112(2):227234, 2007. [NOO+ 00] R. Nelson, R. Orrego, O. Ortiz, J. Tenorio, C. Mundt, M. Fredrix, and N.V. Vien. SECI, Ba and Leadership: a Unied Model of Dynamic Knowledge Creation. Long Range Planning, 33, 2000. J. D. Oster and R. D. Ingvalson. In Situ Measurement of Soil Salinity with a Sensor. Soil Science Society of America, 1967. Ministry of Water Resources. Annual Report 1999-2000. Technical report, Government of India, 2000. J. Panchard. Water Sensors for Agriculture: A Technology Survey. Technical report, EPFL, Lausanne, 2004. J. Panchard. Computer-assisted Cognition: Using Wireless Sensor Networks to Assist the Monitoring of Agricultural Fields. Technical report, EPFL, Lausanne, 2006. Available at http://people.ep.ch/jacques.panchard. J. Panchard. Survey on Wireless Sensor Networks for Agriculture. Technical report, EPFL, Lausanne, 2007. C.K. Prahalad and Allen Hammond. What Works: Serving the Poor, Protably. Technical report, World Resources Institute, 2003. S. Pauletto and A. Hunt. The sonication of emg data. In Proceedings of the 12th Meeting of the International Conference on Auditory Display (ICAD), 20-23 June 2006, London, 2006. S. Pauletto and A. Hunt. the sonication of emg data. In ICAD, 2006. J. Polastre, J. Hill, and D. Culler. Versatile low power media access for wireless sensor networks. In SenSys 04: Proceedings of the 2nd international conference on Embedded networked sensor systems, pages 95107, New York, NY, USA, 2004. ACM. Philippe Monney, Agronomist. Private Communication, 2008.
[OI67] [oWR00] [Pan04] [Pan06]
[Pan07] [PH03] [PH06a]
[PH06b] [PHC04]
[Phi08]
152
REFERENCES
[PHL07]
A. K. Pathan, C. S. Hong, and H.-W. Lee. Smartening the environment using wireless sensor networks in a developing country. Computing Research Repository (CoRR), abs/0712.4170, 2007. J. Panchard, S. Rao, T. Prabhakar, H. Jamadagni, and J.-P. Hubaux. Common-sense net: Improved water management for resource-poor farmers via sensor networks. In International Conference on Communication and Information Technologies and Development (ICTD), 2006. T. V. Prabhakar, N. V. Chalapathi Rao, M. S. Sujay, J. Panchard, H. S. Jamadagni, and A. Pittet. Sensor network deployment for agronomical data gathering in semi-arid regions. In COMSWARE, 2007. J. Panchard, S. Rao, and S. Sheshshayee. Wireless sensor networks for applied research on rain-fed farming in India: an exploratory user experiment. Technical report, 2008. Available at: http://people.ep.ch/jacques.panchard.
[PRP+ 06]
[PRS+ 07]
[PRS08]
[QMK+ 07] L. Qi, M. V. Martin, B. Kapralos, M. Green, and M. Garca-Ruiz. Toward sound-assisted intrusion detection systems. In OTM Conferences (2), pages 16341645, 2007. [Rao05] [Rao08] [Ray] [RBE+ 06] S. Rao. Minutes of Project Meeting, 2005. S. Rao. Private Communication, 20042008. Raymarine. Lifetag: http://www.raymarine.com/. Wireless Man Overboard Monitoring System.
N. Ramanathan, L. Balzano, D. Estrin, M. Hansen, T. Harmon, J. Jay, W. Kaiser, and G. Sukhatme. Designing wireless sensor networks as a shared resource for sustainable development. In International Conference on Information and Communication Technologies and Development, 2006. S. Rao and S. Gadgil. Groundnut Cultivation in Variable Climate - Farmers Perspective. Technical report, Centre for Atmospheric and Oceanic Studies (CAOS), Indian Institute of Science, Bangalore, 1999. S. Rao, M. Gadgil, R. Krishnapura, A. Krishna, M. Gangadhar, and S. Gadgil. Information Needs for Farming and Livestock Management in Semi-arid Tracts of Southern India. Technical Report AS 2, CAOS, IISc, September 2004. I. Romieu, J.M. Sametand K.R. Smith KR, and N. Bruce. Outdoor air pollution and acute respiratory infections among children in developing countries. Technical report, Instituto Nacional de Salud Publica, Mexico, 2004. N. Ramanathan, C. Lee, T. Lin, R. Neumann, S. Rothenberg, C. Harvey, T. Harmon, E. Kohler, D. Estrin, and J. Jay. Sensor-based investigation of biogeochemical controls on arsenic mobilization in rural bangladesh. In American Chemistry Society, General Meeting, 2007. C. Roussel. Fiabilite de capteurs de mesure de la teneur en eau du sol. Hydrology laboratory technical report, HYDRAM, EPFL, February 2008.
[RG99]
[RGK+ 04]
[RKB04]
[RLL+ 07]
[Rou08]
REFERENCES
153
[RR07] [RTB98]
M. Ringwald and K. R mer. Snif: A comprehensive tool for passive inspection of sensor o networks. http://www.vs.inf.ethz.ch/publ/papers/mringwal-snif-kuvs.pdf, July 2007. J.N. Raina, B.C. Thakur, and A.R. Bhandari. Effect of Drip Irrigation and Plastic Mulch on Yield, Water Use Efciency and Benet-Cost Ratio of Pea Cultivation. Journal of the Indian Society of Soil Science, 46(4), 1998. M. Radenkovic and B. Wietrzyk. Mobile Ad Hoc Networking Approach to Detecting and Querying Events Related to Farm Animals. In IEEE International Conference on Networking and Services (IEEE ICNS), Silicon Valley, USA, July 2006. S. Roundy, P.K. Wright, and J.M. Rabaey. Energy Scavenging for Wireless Sensor Networks with Special Focus on Vibrations. Springer, 2003. M. Ringwald, M. Y cel, and K. R mer. Demo abstract: Interactive in-eld inspection of u o wsns. In Adjunct Proceedings of the 3rd European Workshop on Wireless Sensor Networks (EWSN 2006), Zurich, Switzerland, February 2006. P. Sainath. Vidharbha: whose suicide is it, anyway? The Hindu Newspaper, June 25 2005. R. Maria Saleth. Strategic Analysis of Water Institutions in India. Technical report, International Water Management Institute, 2004.
[RW06]
[RWR03] [RYR06]
[Sai05] [Sal04]
[SAM+ 98] D. Seckler, U. Amarasinghe, D. Molden, R. de Silva, and R. Barker. World Water Demand and Supply, 1990 to 2025: Scenarios and Issues. Research report 19, International Water Management Institute, 1998. [Sap06] [SC02] V.K Sapovadia. Micro Finance: The Pillars of a Tool to Socio-Economic Development. Development Gateway, 2006. R. Sakthivadivel and A.S. Chawla. Articial recharging of river water: An experiment in Madhya Ganga Canal Project, India. Technical report, International Water Management Institute, 2002. S. A. Staggenborg, M. Carignano, and L. Haag. Predicting Soil pH and Buffer pH In Situ with a Real-Time Sensor. Agronomy Journal, 99:854861, 2007. T. Schmid and H. Dubois-Ferrires. Sensorscope (LCAV laboratory, EPFL). Documentation available at http://sensorscope.ep.ch. T. Schmid, H. Dubois-Ferriere, and M. Vetterli. SensorScope: Experiences with a Wireless Building Monitoring Sensor Network. In Workshop on Real-World Wireless Sensor Networks (REALWSN05). M.V.K Sivakumar, R. Gommes, and W. Baier. Agrometeorology and Sustainable Agriculture. Agricultural and Forest Meteorology, 103:1126, 2000. Siemens. TC 65 GPRS Card. Documentation available https://pia.khe.siemens.com/eles/wireless/datasheets/TC65 Datasheet.pdf. at
[SCH07] [SDF] [SDFV]
[SGB00] [Sie]
154
REFERENCES
[SLAH01]
R. Sakthivadivel, R. Loeve, U.A. Amarasinghe, and M. Hemakumara. Water Scarcity and Managing Seasonal Water Crisis: Lessons from the Kirindi Oya Project in Sri Lanka. Technical Report Research Report 55, International Water Management Institute, September 2001. G. Sharma and R. Mazumdar. Hybrid sensor networks: a small world. In MobiHoc 05: Proceedings of the 6th ACM international symposium on Mobile ad hoc networking and computing, pages 366377, New York, NY, USA, 2005. ACM. Soil moisture sensors. Documentation http://www.microirrigationforum.com/new/sensors/. available at URL
[SM05]
[sms] [SNM07]
I. Stoianov, L. Nachman, and S. Madden. Pipenet: A wireless sensor network for pipeline monitoring. IEEE International Conference on Information Processing in Sensor Networks (IPSN), 2007. R. B. Salama, C. J. Otto, and R. W. Fitzpatrick. Contributions of groundwater conditions to soil and water salinization. Hydrogeology Journal, 7:4664, 1999. M. Sotoudeh. Participatory methods: a tool for the improvement of innovative environmental technologies. International Journal of Environmental Technology & Management, 3:336348, 2003. R. Szewczyk, J. Polastre, A. Mainwaring, and D. Culler. Lessons From a Sensor Network Expedition. In Proceedings of the First European Workshop on Sensor Networks (EWSN 04), Berlin, Germany, January 2004. R. Shah, S. Roy, S. Jain, and W. Brunette. Data mules: Modeling a three-tier architecture for sparse sensor networks. In IEEE SNPA Workshop, May 2003. H. Sharp, Y. Rogers, and J. Preece. Interaction Design: Beyond Human Computer Interaction. Wiley, March 2007. S. Staehli. Evaluation of Bridge Architectures for the COMMON-Sense Net Project. Master thesis, EPFL, 2007. A. Sheth, K. Tejaswi, P. Mehta, C. Parekh, R. Bansal, S. Merchant, T. Singh, U. B. Desai, C. A. Thekkath, and K. Toyama. Senslide: a sensor network based landslide prediction system. In SenSys 05: Proceedings of the 3rd international conference on Embedded networked sensor systems, pages 280281, New York, NY, USA, 2005. ACM Press. A. Sheth, C. A. Thekkath, P. Mehta, K. Tejaswi, C. Parekh, T. N. Singh, and U. B. Desai. Senslide: a distributed landslide prediction system. SIGOPS Oper. Syst. Rev., 41(2):7587, 2007. Sun Small Programmable Object Technology. http://www.sunspotworld.com/. Documentation available at
[SOF99] [Sot03]
[SPMC04]
[SRJB03] [SRP07] [Sta07] [STM+ 05]
[STM+ 07]
[Sun]
[SWC+ 07] L. Selavo, A. Wood, Q. Cao, T. Sookoor, H. Liu, A. Srinivasan, Y. Wu, W. Kang, J. Stankovic, D. Young, and J. Porter. Luster: wireless sensor network for environmental research. In SenSys 07: Proceedings of the 5th international conference on Embedded networked sensor systems, pages 103116, New York, NY, USA, 2007. ACM.
REFERENCES
155
[Sys] [Tec] [tel] [TF02] [TGCB04]
EME Systems. SMX: Electrical Interface for Watermark or Gypsum Block Sensors. Documentation available at the URL http://www.emesys.com/smx.htm. Linx Technologies. Telosb wireless sensor from moteiv. http://www.moteiv.com. G. C. Topp and T. P. A. Ferre. Methods of Soil Analysis: Water content. Madison, WI, 2002. K. Turner, S. Georgiou, R. Clark, and R. Brouwer. Economic valuation of water resources in agriculture : From the sectoral to a functional perspective of natural resource management. Food and Agriculture Organization of the United Nations, 2004. P. S. Thenkabail, M. S. D. N. Gamage, and V. U. Smakhtin. The Use of Remote Sensing Data for Drought Assessment and Monitoring in Southwest Asia. Research Report 85, International Water Management Institute, 2004. tinynode wireless sensor from Shocksh. http://www.tinynode.com. D.S.G. Thomas and N.J. Middleton. Salinization: new perspectives on a major desertication issue. Journal of arid environments, 24(1):113127, 1993. TinyOS operating system. Documentation available at URL http://www.tinyos.net. L.S. Ungani and F.N. Kogan. Drought monitoring and corn yield estimation in Southern Africa from AVHRR data. Remote Sesnsing of Environment, 63:219232, 1998. UN Genral Assembly. United Nations Millennium Declaration, 2002. Water and Sanitation Programme. http://www.unhabitat.org/categories.asp?catid=270. Adam Wagstaff. Poverty and health sector inequalities. Bulletin of the World Health Organization, 80:97 105, 00 2002. J. S. Wallace. Increasing agricultural water use efciency to meet future food production. Agriculture, Ecosystems & Environment, 82(1-3):105119, December 2000. G. Walsham. Making a World of Difference: It in a Global Context. John Wiley & Sons, Inc., New York, NY, USA, 2001. D.H. Walker. Decision support, learning and rural resource management. Agricultural Systems, 73:95105, 2002. S. Von Westarp, S. Chieng, and H. Schreier. A comparison between low-cost drip irrigation, conventional drip irrigation, and hand watering in nepal. Agricultural Water Management, 64(2):143160, 2004.
[TGS04]
[tin] [TM93] [tos] [UK98] [UN 02] [UNH] [Wag02] [Wal00] [Wal01] [Wal02] [WCS04]
[WCS+ 07] T. Wark, P. Corke, P. Sikka, L. Klingbeil, Y. Guo, C. Crossman, P. Valencia, D. Swain, and G. Bishop-Hurley. Transforming agriculture through pervasive wireless sensor networks. IEEE Pervasive Computing, 6(2):5057, 2007.
156
REFERENCES
[Wit57] [WKD05] [Wor03] [WZW04]
K.A. Wittfogel. Oriental despotism: A comparative study of total power. New Haven and London: Yale University Press, 1957. A. Williams, E. Kabisch, and P. Dourish. From interaction to participation: Conguring space through embodied interaction. In Ubicomp, pages 287304, 2005. World Health Organization. World Health Report. Technical report, WHO, 2003. N. Wang, N. Zhang, and M. Wang. Wireless sensors in agriculture and food industry Recent development and future perspective. Computers and Electronics in Agriculture, 50:114, January 2004. Mica2 Wireless Sensing Platform. Documentation available at URL http://www.xbow.com. E. Zarate. Understanding the origins and fate of air pollution in Bogota, Colombia. PhD thesis, EPFL, Lausanne, 2007. Kasun De Zoysa, Chamath Keppitiyagama, Gihan P. Seneviratne, and W. W. A. T. Shihan. A public transport system based sensor network for road surface condition monitoring. In NSDR 07: Proceedings of the 2007 workshop on Networked systems for developing regions, pages 16, New York, NY, USA, 2007. ACM. A. Zubair. Indian farmer suicide toll rises, 2006. BBC News, Mumbai, July 14, 2006. M. Zennaro, A. Yemane, N. Farahbakhshzad, and B. Pehrson. WIRELESS SENSOR NETWORKS FOR WATER QUALITY MANAGEMENT. Technical report, Royal Institute of Technology, Stockholm, 2007.
[xbo] [Zar07] [ZKSS07]
[Zub06] [ZYFP07]
Index
Agriculture, 7 Actors, 124, 136 Irrigation Management, 7 Pest and Disease, 8 Agriculture, Indian Facts and Figures, 12 History, 10 Institutions, 13 Karnataka, 13 Water Scarcity, 12 Air Pollution, 8 Apprenticeship, 140 Chennakeshavapura, 47 Agriculture, 47 Marginal Farming, 50 COMMON-Sense Net Data collection, 64 Database, 71 Design, 61 Data Generation, 61 Data Transport, 62 Design guidelines, 58 Embedded, 63 GPRS, 69 Information survey, 51 Methodology, 59 Partners, 43 Project, 44 System Overview, 63 Usability, 135 Use, 135 Use cases, 54 Usefulness, 134 User Survey, 123 Web, 72 Wi-Fi, 68 Deployment Support, 100 Developing Country Environmental Challenges, 5 Environmental monitoring Design dimensions, 19 ICT4D, 137 Gaps, 138 Karnataka, 45 Knowledge Creation, 138 Knowledge Society, 137 Medium Access Control B-MAC, 66 Dozer, 66 Millennium Development Goals, 5 Participatory Methods, 140 Pavagada, 46 Sensing Cellular telemetry, 18 Remote sensing, 18 Stand-alone, 17 Wireless sensor networks, 18 Sensor Agriculture, 23 Precipitation, 26 Soil moisture, 23 Soil pH, 25 Soil salinity, 25 Temperature, 25 Wind, 26 Sensor-Tune, 104 Design, 108 History, 107 Live, 105 Prototype, 114 User Experiment, 117 157
158
INDEX
User Survey, 112 Shocksh Mamaboard, 69 Siemens TC65, 70 Signal and Noise Metaphor, 103 Degradation, 104 Sound, 103 Sonication, 99 Advantages, 101 Challenges, 102 Strategy, 123 TinyNode, 64 Power consumption, 66 Radio range, 65 TinyOS, 64 Trafc, 8 Use Case Crop Physiology, 132 Entomology, 130 Soil Science, 130 Water Management, 133 Water Quality, 10 Wireless Sensor Networks Developing Countries Strategy, 123 User Experiments, 100 Wireless sensor networks, 20 Agriculture, 26 Cattle monitoring, 28 Developing countries (assets), 36 Developing countries (challenges), 39 Developing countries (examples), 31 Disease prevention, 27 Examples, 26 Medium Access Control, 21 Mobility, 22 Modularity, 22 Multihop, 21 Power management, 21 Self-organization, 21 Vineyard, 26 Web, 23 Wireless, 20 Wireless Sesnor Networks
Interfaces, 99
JACQUES PANCHARD
Research and teaching assistant Laboratory for computer Communications and Applications (LCA) School of Computer and Communication Sciences (IC) EPFL (Ecole Polytechnique F d rale de Lausanne), Switzerland e e Station 14 CH-1015 Lausanne, Switzerland PERSONAL Born in Lausanne, Switzerland on March 25, 1970. Citizen of Switzerland. Languages: French - native, English - uent, German - uent RESEARCH Senior Communication Systems Engineer, with an experience in development projects. I will complete my PhD at EPFL in summer 2008. Prior to that, I worked as IT consultant, system designer and project leader in Switzerland and California, in the area of mobile communications, Internet and Web. My PhD focuses on the use of Information and Communication Technologies for rural development in India. EDUCATION PhD. student in communication systems, July. 2004 present EPFL, S WITZERLAND thesis title: Wireless Sensor Networks for Marginal Farming in India advisor: Prof. Jean-Pierre Hubaux expected graduation: August 2008 Diploma (Dipl. Ing.) in Communication Systems, Sep. 1990 July. 1995 E COLE P OLYTECHNIQUE F E D E RALE DE L AUSANNE (EPFL), S WITZERLAND thesis title: Assessment of the VCELP Vocoder for Mobile Communications advisor: Pr. Dirk Slock (Ericsson Research, Hungary) major in telecommunications, minor in mobile communications Master (MSc.) in Communication Systems, Sep. 1990 July. 1995 I NSTITUT E UR E COM , S OPHIA A NTIPOLIS , F RANCE major in telecommunications, minor in mobile communications PROFESSIONAL EXPERIENCE Research and teaching assistant, Sep. 2004 present S CHOOL OF C OMPUTER AND C OMMUNICATION S CIENCES (IC) EPFL, S WITZERLAND Project Manager, 1999 2002 I NFO D ESIGN C OMMUNICATIONS SA G ENEVA , S WITZERLAND Design and implementation of strategic web portals in the area of banking, insurances, international organizations (programming in Perl, Java and WebObjects)
e-mail: jacques.panchard@ep.ch url: http://people.ep.ch/jacques.panchard phone: +41 21 6935613
Senior IT Consultant, 1998 1999 E LLIPSIS C OMMUNICATIONS C ORP. P ETALUMA , CA, USA Projects for telecommunications companies including Cerent Corporation (now part of Cisco) and Alcatel USA. Ellipsis Communications Corp. is now part of Turin Networks. Software Design Engineer, 1997 1998 N USANTARA C OMMUNICATIONS P ETALUMA , CA, USA Software design and development on the Advanced Rural Telephone System (ARTS), a wireless telephone system designed and developed for use in rural areas IT Junior Consultant, 1996 L OGICA Z URICH , S WITZERLAND TEACHING Teaching assistant Mobile Networks, EPFL 2006, 2007, 2008 Computer Networks, EPFL 2005 Embedded Systems, EPFL 2007 Supervised projects Ga l Ravot, Sudden Node Death in WSNs: Causes, Detection and Prevention, master, 2007 e Stefan Staehli, Bridge Architectures for the COMMON-Sense Net Project, master, 2007 Behnaz Bostanipour, Sound-based Monitoring of Wireless Sensor Networks, semester, 2007 Julien Giraudi, WSN for Environmental Monitoring: Data Exchange, semester, 2006 Jean Rossier and Ga l Charri` re, Prototype of WSN for Water Management in India, sem., 2005 e e Sathya Anand, Power Management Issues in the COMMON-Sense Net Project, intern., 2004 AWARDS Selected paper at ICTD 2006 for MIT Press Journal Publication EPFL- DDC Grant for Projects in Development and Cooperation (CHF 500000) REFERENCES Prof. Jean-Pierre Hubaux, Full Professor, EPFL, Switzerland Room BC 207, +41 21 6932627, jean-pierre.hubaux@ep.ch Prof. H.S. Jamadagni, Chairman, C ENTRE FOR E LECTRONIC D ESIGN AND T ECHNOLOGIES I NDIAN I NSTITUTE OF S CIENCE, Bangalore, India +91 (80) 2 360 08 08, hsjam@cedt.iisc.ernet.in Dr. Andr Pittet, Chief Project Advisor, e S WISS AGENCY FOR D EVELOPMENT AND C OOPERATION, Bangalore, India +91 (80)2 3600 809, apittet@cedt.iisc.ernet.in
PUBLICATIONS Journals J. Panchard, S. Rao, T. V. Prabhakar, J.-P. Hubaux, and H. S. Jamadagni, COMMON-Sense Net: A Wireless Sensor Network for Resource-Poor Agriculture in the Semiarid Areas of Developing Countries, In Information Technologies and International Development, MIT Press, 4(1):51-67, 2007. Conferences, Workshops 1. 2. J. Panchard, P. S. Rao, M. Sheshshayee, P. Papadimitratos, and J.-P. Hubaux, Wireless Sensor Networking for Rain-fed Farming Decision Support User Survey, In ACM SIGCOMM Workshop on Networked Systems for Developing Regions, Seattle, 2008. J. Panchard, S. Rao, T. Prabhakar, H. Jamadagni, and J.-P. Hubaux, COMMON-Sense Net: Improved Water Management for Resource-Poor Farmers via Sensor Networks, In International Conference on Communication and Information Technologies and Development (ICTD), Berkeley, 2006. J. Panchard and A. Osterwalder, ICTs and Capacity Building through Apprenticeship and Participatory Methods - Applied to an ICT-based agricultural water management system, In Conference of the International Federation of Information Processing - Social Implications of Computers in Developing Countries (IFIP WG9.4), Abuja, 2005. T. V. Prabhakar, N. V. Chalapathi Rao, M. S. Sujay, J. Panchard, H. S. Jamadagni, and A. Pittet, Sensor Network Deployment For Agronomical Data Gathering in Semi-Arid Regions, In IEEE International Conference on COMmunication System softWAre and middlewaRE (COMSWARE), Bangalore, 2006. J. Luo, J. Panchard, M. Piorkowski, M. Grossglauser, and J.-P. Hubaux, MobiRoute: Routing towards a Mobile Sink for Improving Lifetime in Sensor Networks, In the 2nd IEEE/ACM International Conference on Distributed Computing in Sensor Systems (DCOSS), San Francisco, 2006.
3.
4.
5.
6.
Technical Reports 7. 8. J. Panchard, S. Rao, and S. Sheshshayee. Wireless sensor networks for applied research on rainfed farming in India: an exploratory user experiment, Technical report, 2008. J. Panchard. Computer-assisted Cognition: Using Wireless Sensor Networks to Assist the Monitoring of Agricultural Fields, Technical report, 2007.

Wireless Sensor Networks For Marginal Farming in India: Jacques Panchard

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Wireless Sensor Networks For Marginal Farming in India: Jacques Panchard

Uploaded by

Copyright:

Available Formats

WIRELESS SENSOR NETWORKS FOR MARGINAL FARMING IN INDIA

` PRESENTEE A LA FACULTE INFORMATIQUE ET COMMUNICATIONS

ECOLE POLYTECHNIQUE FEDERALE DE LAUSANNE

Lausanne, EPFL 2008

61 61 61 62 63 63 64 65 66 67 68 69 70 71 71 72 73 74 74 77 77 79 80 80 81 85 86 89 89 93 94 96 97 97 99 99 99 100 100 100 101 102 103 104

Conclusion 143 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143 References Index 145 156

Environmental Challenges in Developing Countries

The Millennium Development Goals and the Environment

CHAPTER 1. ENVIRONMENTAL CHALLENGES IN DEVELOPING COUNTRIES

The Importance of Environmental Monitoring

Agriculture Air Pollution and Trafc Water Quality Monitoring

Table 1.1: Environmental Monitoring and MDGs: correspondence matrix

CHAPTER 1. ENVIRONMENTAL CHALLENGES IN DEVELOPING COUNTRIES

Air Pollution and Trafc

1.4. AIR POLLUTION AND TRAFFIC

CHAPTER 1. ENVIRONMENTAL CHALLENGES IN DEVELOPING COUNTRIES

Water Quality Monitoring

Agriculture and Water in India: a Brief Historical Perspective

1.6. AGRICULTURE AND WATER IN INDIA: A BRIEF HISTORICAL PERSPECTIVE

CHAPTER 1. ENVIRONMENTAL CHALLENGES IN DEVELOPING COUNTRIES

Indias Agriculture Today

A Recent and Growing Concern: Water Scarcity

1.8. INDIA TODAY: THE CURRENT INSTITUTIONAL FRAMEWORK

The Specic Case of Karnataka

India Today: the Current Institutional Framework

States and Central Government

CHAPTER 1. ENVIRONMENTAL CHALLENGES IN DEVELOPING COUNTRIES

CHAPTER 1. ENVIRONMENTAL CHALLENGES IN DEVELOPING COUNTRIES

New Opportunities for Environmental Monitoring and Agriculture

CHAPTER 2. NEW OPPORTUNITIES FOR ENVIRONMENTAL MONITORING AND AGRICULTURE