Professional Documents
Culture Documents
Received 4 January 1999; received in revised form 30 June 1999; accepted 8 October 1999
Abstract
0168-1699/99/$ - see front matter © 1999 Elsevier Science B.V. All rights reserved.
PII: S 0 1 6 8 - 1 6 9 9 ( 9 9 ) 0 0 0 6 4 - 2
196 N.D. Clarke et al. / Computers and Electronics in Agriculture 24 (1999) 195–204
1. Introduction
Although, many computer software programs have been developed for use in the
agricultural sector, few have being used to any extent. Edwards-Jones (1993)
reported that over 67 knowledge-based systems (KBS) have been developed specifi-
cally for crop protection, but the use of these KBS was small (B 12% actually
became fully operational). The low acceptance rate of KBS in agriculture has been
attributed to many factors (Udink ten Cate, 1991; Edwards-Jones, 1992; Lambert,
1993; Freeman et al., 1994). However, two factors have been frequently mentioned
for the low transfer rate of KBS from the laboratory to the field: lack of
consultation with the end user during development (Gould et al., 1991; Crassweller
et al., 1993; Parham and Poley, 1993) and lack of long-term support and mainte-
nance (Crassweller et al., 1993; Foster and Berry, 1993; Lambert, 1993). Software
must be developed in cooperation with end users to solve their actual needs, not
their assumed needs. Direct contact with and early focus on users are required to
make software usable, useful and likeable (Gould et al., 1991). A software develop-
ment project must be viewed as a long term project instead of a short term research
project with a definite ending point (Crassweller et al., 1993) and the product must
be continuously improved and supported through annual training of users, revi-
sions of documentation, telephone support, and on-farm visits (Lambert, 1993).
Without well organised plans for support and maintenance, KBS in agriculture are
doomed to a limbo of perpetual prototype (Foster and Berry, 1993).
The Harrow Greenhouse Manager (HGM) is a suite of decision-support tools to
assist growers in the management of greenhouse cucumber and tomato crops
(Agriculture and Agri-Food Canada, 1998) and is based upon a general model of
integrated crop-management of greenhouse crops (Clarke et al., 1994). HGM
consists of five basic modules: information, diagnostic, control, databases, and
tools. The information module consists of six hypertext documents, three each for
the cucumber and tomato crops. The diagnostic module consists of knowledge
bases for the identification of diseases, nutritional/physiological disorders, and
insects/mites of cucumber and tomato. The control module is a single module for
viewing and scheduling appropriate biological, chemical and cultural control op-
tions and for determining if there are any conflicts with previously scheduled or
applied controls. The database module is a single database for recording and
analysing data for all aspects of the greenhouse vegetable operation. Finally, the
tools module consists of two programs for assisting growers with common tasks:
the nutrient calculator determines the amount of individual fertilisers to add to the
stock tanks while the climate parameter calculator calculates the psychometric
properties of the air in the greenhouse.
This report describes the development of the HGM. Our DSS developmental
approach attempts to overcome the above mentioned barriers to adoption by
extensively involving the end-user. Some of the significant contributions made by
the end users are discussed.
N.D. Clarke et al. / Computers and Electronics in Agriculture 24 (1999) 195–204 197
The following milestones outline the major tasks undertaken to develop the
HGM. A number of the tasks overlapped, in particular, tasks 2.8 through 2.11. The
project was initiated in 1992 and the finished product was commercialised in 1998.
The first step in the project was recognition by Agriculture and Agri-Food
Canada (AAFC) scientists that a problem existed, namely that greenhouse veg-
etable crop management is a complex and difficult task, involving expertise of crop,
insect/mite pest, and disease management. Many factors must be taken into account
such as plant nutrition, greenhouse climate, pest incidence, disease incidence,
chemical control options, biological control options, disease prevention methods,
etc. Conflicting recommendations need to be identified and, where possible, be
resolved. The scientists routinely received numerous requests from local growers to
help with management decisions. It was determined that a knowledge-based ap-
proach would be the best solution to the problem. Funding was obtained to finance
the development of a prototype KBS and a knowledge engineering firm (AI
Solutions) was retained.
The knowledge engineer and experts met several times to determine the goals,
objectives, and function of the KBS within the greenhouse management framework.
Of the three most common greenhouse vegetable crops produced in Ontario
(tomato, cucumber, and sweet pepper), cucumber was selected for developing the
KBS because this crop has the greatest number of pest and disease problems which
can result in many conflicting recommendations. The overall goal of the KBS was
to assist the greenhouse grower in making management decisions to reduce energy
consumption by optimising greenhouse environmental conditions and fertiliser
applications, and to eliminate the use of pesticides through nonchemical disease and
pest-management strategies. These goals were achieved while maintaining accept-
able and profitable cucumber production levels. A conceptual diagram was devel-
oped illustrating how the KBS would be used in the management process. The
results of these meetings were presented to greenhouse extension advisors from the
Ontario Ministry of Agriculture, Food and Rural Affairs (OMAFRA). After the
suggestions of the advisors were incorporated, a presentation was made to the
greenhouse industry (see task 2.3) and further revisions were made.
Two meetings were held to present the project to the greenhouse industry.
Greenhouse growers, consultants, extension advisors, suppliers, marketing board,
and experts were present. Comments and suggestions were incorporated into the
198 N.D. Clarke et al. / Computers and Electronics in Agriculture 24 (1999) 195–204
stated goals and objectives. This milestone was very important, as the potential end
users must first understand what KBS are and how they are developed. Then, they
need to be convinced of the benefits of KBS for greenhouse production. Without
support and acknowledgement from the industry that KBS are useful to them, the
project would not adequately address the needs of the industry and likely would not
be adopted.
The various human and computer resources were identified. Three experts (an
entomologist, a plant physiologist and a plant pathologist) from the Agriculture
and Agri-Food Canada Greenhouse and Processing Crops Research Centre
(GPCRC) agreed to commit time and effort to the project and serve as the domain
experts. A greenhouse systems engineer from the GPCRC provided engineering
assistance. Two OMAFRA greenhouse advisors undertook to critique and test the
system to ensure recommendations were complete and current. These experts and
advisors, along with the knowledge engineer, formed the steering committee, which
met every month to review project progress. Greenhouse growers were selected as
the main user group, although the extension service, greenhouse service industry,
and educators also found the software useful. Eighteen local growers were asked to
participate in a Grower committee to review and suggest changes to the software
throughout the developmental process. The development and delivery platform was
the personal computer and the operating system was Microsoft® Windows™ 3.1.
The purpose of this task was to determine and describe what the KBS will do.
Meetings with the knowledge engineer and experts were held until a consensus was
reached on the objectives, constraints and specific tasks that the KBS must perform
to accomplish the design objectives. Knowledge required to perform the tasks was
organised into modules. Each module was described in terms of its basic algorithm
and the required inputs, outputs, and knowledge sources. Benchmarks were estab-
lished for completing the different modules. The end result was a design document,
which aided in the planning and programming of the KBS. The design document
didn’t contain detailed specifications, but rather general guidelines because the
prototyping approach to development was used.
N.D. Clarke et al. / Computers and Electronics in Agriculture 24 (1999) 195–204 199
A prototype system was built to test the functional specifications. Knowledge was
acquired from the experts through interviews, formalised, and coded into the
development tool to build a working prototype. The prototype identified cucumber
pests and diseases and recommended appropriate control actions. KnowledgePro®
Windows (Knowledge Garden, 1991) was the development tool. The prototype was
evaluated and the decision was made to expand on the prototype to build the
complete system.
Workshops were held with the Grower committee every 2 or 3 months. During
a typical workshop, the committee was first shown the new features of the KBS that
had been added since the last meeting. The growers then had an opportunity to
work with the KBS. All comments and suggestions made by the growers were
recorded. Every attempt was made to incorporate these comments and suggestions
into the KBS.
The steering committee met once a month to review progress and to ensure the
project was on track. Benchmarks were revised and new goals established.
Based upon the design document, the prototype was expanded into the complete
KBS. Knowledge acquired from experts and literature was formalised and coded
into the KBS. Different techniques such as case studies, decision trees and decision
charts were used to acquire the experts’ knowledge. A data management system was
developed to record and analyse all aspects of a greenhouse operation. The system
design for the cucumber KBS (see task 2.6) was also used for tomatoes. As a result,
programming of the tomato system proceeded very quickly, because many of the
basic routines and user interface screens were already developed. The combined
system became known as the HGM. The prototype development process continued
to be used; the knowledge engineer refined and modified the software after
reviewing changes with the Grower committee.
Testing and verifying the HGM was an ongoing process that complemented the
development process. Testing consisted of four components. First, the HGM was
constantly being tested by the experts and knowledge engineer during development.
Second, OMAFRA extension advisors had the HGM loaded on laptops which they
took on farm visits. On such visits, the HGM was used to diagnose crop disorders
200 N.D. Clarke et al. / Computers and Electronics in Agriculture 24 (1999) 195–204
A long-term support and maintenance plan was devised. Updates to the data files
will be required on a yearly basis as registered pesticides change and new biological
control options become available. The system was designed so additional informa-
tion or modules could be easily added. The HGM is marketed, through a private
engineering company that is interested in expanding its business interests in
agriculture and software systems, in accordance to AAFCs mandate is to promote
and partner with the agriculture business sector. The company is responsible for
marketing and distributing the software, as well as, providing technical software
support. It is critical for the government to form partnerships with private industry
and grower organisations. This way they have a stake in the system and will benefit
from its successful adoption and future improvements.
3. End-user contributions
modules and the control module share the same basic screen layouts. Another
feature that the growers desired was the ability to access as much information as
possible from any point in the HGM. For example, the control module was
originally designed so that if the grower knew what the disorder was, control
actions could be accessed without entering the diagnostic module. In addition to
accessing the control actions, the growers also wanted to view a picture and read
text describing the disorder. These features were subsequently added to the control
module.
The growers were the driving force behind the development of the nutrient
calculator and the climate parameter calculator. These tasks are tedious and
complex and growers complained about the difficulty and amount of time it took
to solve the tasks. With the nutrient calculator, many different combinations of
fertilisers can be selected. It is a difficult task to determine the amounts of fertiliser
that will result in exact nutrient concentrations in the final solution. Previously, the
growers had to arrive at the final solution by trial and error. Now, the nutrient
calculator automates this task and reduces the amount of time that growers spend
calculating fertilizer combinations.
With the database module, the growers contributed in several ways. First, they
brought the issue of data recording and analysis to the forefront. They had been
recording items such as crop yields in booklets and then they would try to make
Fig. 1. A sample prompt screen from the cucumber insect/mite diagnostic component. The growers use
an expert rule base to answer questions to determine the disorder.
202 N.D. Clarke et al. / Computers and Electronics in Agriculture 24 (1999) 195–204
Fig. 2. Screen layout from the slide viewer. When users find the disorder that matches their crop
symptoms, they can click on the info button to get more information about the disorder or choose the
select button to get a list of possible control measures.
comparisons with data from previous years. As a result, it was very difficult and
time consuming to make meaningful comparisons among years and to determine
the cause of any differences. The growers felt that a database system would enable
them to make better comparisons between data from different years. Second, the
growers wanted assistance in determining the cause of changes in their crops. They
felt that it would be a great benefit if they could generate a report or graph
containing two or more variables that may influence the crop. For example,
graphing crop yield, temperature and light intensity could help determine if changes
in the greenhouse climate had impacted yields. Plotting pest counts and yield could
help determine the impact of high pest populations on yields. This feature would
also help growers determine relationships between crop inputs and help them better
understand how to manage the crop. To facilitate this feature, we developed a
general analysis form, which allows the grower to select any recorded variable for
reporting or graphing.
N.D. Clarke et al. / Computers and Electronics in Agriculture 24 (1999) 195–204 203
4. Conclusions
Working with the Grower committee has been very enlightening and productive
experience. They brought a practical and often different viewpoint compared to the
experts. Growers are more concerned about time commitment and user-friendliness
of KBS as their daily work schedules are already very busy. Growers must be
convinced of the benefits of using KBS, both short and long term. Incorporating
the Grower committee in the development of the HGM resulted in the development
of a KBS that better suited their needs and problems. The committee provided
another level for immediate feedback to new ideas and how the system actually
functioned with the primary user. The end result was development of a more
complete KBS and a system, which ultimately should be adopted faster by the main
user group. The HGM is now at the marketing stage and time will tell how quickly
the system will be adopted. To date, growers and greenhouse industries in Canada
and the United States have purchased the software. At demonstrations of the HGM
to various user groups throughout the world, a lot of interest has been expressed in
the availability of the HGM. The HGM is now being modified to create a version
for the Korean greenhouse industry.
Acknowledgements
References
Agriculture and Agri-Food Canada. 1998. Harrow Greenhouse Manager®. Enermodal Engineering Ltd.,
650 Riverbend Drive, Kitchener, Ont.
Clarke, N.D., Shipp, J.L., Jarvis, W.R., Papadopoulos, A.P., Jewett, T.J., 1994. Integrated management
of greenhouse crops — a conceptual and potentially practical model. HortScience 29, 846 – 849.
Crassweller, R.M., Travis, J.W., Heinemann, P.H., Rajotte, E.G., 1993. The future use and development
of expert system technology in horticulture. HortTechnology 3, 203 – 205.
Edwards-Jones, G., 1992. Knowledge-based systems for pest management: an applications-based review.
Pestic. Sci. 36, 143–153.
Edward-Jones, G., 1993. Knowledge-based systems for crop protection: theory and practice. Crop Prot.
12, 565–578.
Freeman, S.A., Jones, D.D., Field, W.E., 1994. A case study in knowledge system acceptance. In:
Watson, D.G., Zazueta, F.S., Harrison, T.V. (Eds.), Computers in Agriculture 1994, Proceedings of
the 5th International Conference. American Society of Agricultural Engineers, St. Joseph, MI, pp.
838–843.
204 N.D. Clarke et al. / Computers and Electronics in Agriculture 24 (1999) 195–204
Foster, M.A., Berry, J.S., 1993. Software support and maintenance in agriculture and natural resources:
critical issues for long-term success. AI Appl. 7, 39 – 40.
Gould, J.D., Boies, S.J., Lewis, C., 1991. Making useable, useful, productivity-enhancing computer
applications. Commun. ACM 34, 75– 85.
Knowledge Garden, 1991. KnowledgePro® Windows Reference Manual. Knowledge Garden, Inc.,
Stony Brook Technology Center, 12-8 Technology Drive, Setauket, NY. 11733.
Lambert, J.R., 1993. Continuous improvement of software support processes. AI Appl. 7, 45 – 48.
Parham, G.L., Poley, J.K., 1993. Creation and maintenance: the split personality of software develop-
ment. AI Appl. 7, 42–44.
Udink ten Cate, A.J., 1991. Information technology and industrial automation trends in agriculture. In:
Hashimoto, Y., Day, W. (Eds.). Mathematical and control applications in agriculture and horticul-
ture. Proceedings of the International Federation of Automatic Control/International Society for
Horticultural Science Workshop, Pergamon Press, Oxford, UK, pp. 259 – 263.