You are on page 1of 16

Agricultural Water Management 256 (2021) 107088

Contents lists available at ScienceDirect

Agricultural Water Management


journal homepage: www.elsevier.com/locate/agwat

Smart water management approach for resource allocation in High-Scale


irrigation systems
Andrés-F. Jiménez a, b, *, Pedro-F. Cárdenas a, Fabián Jiménez c
a
Universidad Nacional de Colombia – Sede Bogotá, School of Engineering, Department of Mechanical and Mechatronics Engineering, un-Robot, Carrera 45 N◦ 26-85 -
Uriel Gutiérrez Building, Bogotá D.C. 111321, Colombia
b
Universidad de los Llanos, School of Basic Sciences and Engineering, Dynamic Systems Research Group, Km. 12 Vía Puerto López, Villavicencio, Colombia
c
Universidad Pedagógica y Tecnológica de Colombia, Avenida Central del Norte 39-115, PBX: (57+8) 7405626, Tunja, Boyacá, Colombia

A R T I C L E I N F O A B S T R A C T

Handling editor: Xiying Zhang Water is a common resource shared by many agents or farmers in irrigation districts, where water management
is, therefore, a very complex problem if the target is the equitable distribution of resources or the efficient use of
Keywords: water. This paper aims to present the design and implementation of an Irrigation Agent-Based Model (IABM) for
Agent-Based Model the distribution of water in an irrigation district. In the IABM proposed, each field or farm can be considered as a
Irrigation scheduling
micro-level and a region or irrigation district as a macro-level. The most approximated irrigation scheduling at
Negotiation
each micro-level allows determining the water needs at the macro-level, as a reference for the negotiation of
Precision irrigation
Water sustainability water distribution. A Geographical Information System (GIS) was used to manage georeferenced field infor­
mation. Field capacity and permanent wilting point values, initial soil moisture, crop types, crop coefficients,
root depth, sowing dates and weather data were incorporated into the IABM model for determining irrigation
prescriptions. When water available in the irrigation district is insufficient to supply water requirements of all the
irrigation fields, the IABM allows water distribution among them according to the cultivated field sizes, irrigation
priorities, phenological states and the behavior of neighbors. When water resources are distributable, the pro­
posed negotiation algorithm ensures that fields will use irrigation prescriptions defined with technical and
agronomic criteria, thus avoiding the waste of water resource. The IABM developed meets the satisfaction
requirement of the greatest number of agents by avoiding as much as possible that the soil water stress increases,
using 25% of permissible level of moisture depletion respect to the total available water in each field.

1. Introduction called agents, use reactive or rational behavior to determine actions in


the surrounding environment (Jimenez et al., 2020b). An agent is an
Efficient water management in agricultural crops is of paramount autonomous entity, which obtains information from the environment
importance for sustainable development and food security (Mango et al., using a perception system, makes decisions through strategies and
2018). Currently, climate change and water scarcity in some parts of the carries out actions to modify the environment using actuators (Russell
world urge the search for solutions that allow the optimization of agri­ and Norvig, 2016).
cultural processes, thus obtaining higher yields with better use of re­ Irrigation districts are complex systems. Farmers can be considered
sources and reduction of production costs (Ojha et al., 2015). Precision as adaptive agents, who have autonomy, intelligence and relative
irrigation consists of sustainable water management, which involves knowledge about their environment (crop fields) (An et al., 2005). These
applying water to crops to meet specific plant needs (Adeyemi et al., agents can update their knowledge and adjust their behaviors to the
2018). With the development of modern technologies, a considerable environment change. Farmers are interconnected through social re­
transformation has been observed in many areas, especially in precision lationships with irrigation management authorities and also with their
agriculture applications. Artificial Intelligence (AI) has acquired a major neighbors. The interactions between farmers could lead to patterns at
role in daily lives, thus improving our perceptions and abilities to modify the aggregate level as conflict resolution through negotiation, coordi­
the environment around us (Talaviya et al., 2020). In AI, the systems, nation, cooperation and competition in the distribution of water

* Correspondence to: Department of Mathematics and Physics. Unillanos, km. 12 Vía Puerto López, 500017 Villavicencio (Meta), Colombia.
E-mail addresses: ajimenez@unillanos.edu.co (A.-F. Jiménez), pfcardenash@unal.edu.co (P.-F. Cárdenas), fabian.jimenez02@uptc.edu.co (F. Jiménez).

https://doi.org/10.1016/j.agwat.2021.107088
Received 2 October 2020; Received in revised form 9 July 2021; Accepted 18 July 2021
Available online 28 July 2021
0378-3774/© 2021 Elsevier B.V. All rights reserved.
A.-F. Jiménez et al. Agricultural Water Management 256 (2021) 107088

Fig. 1. Large-scale irrigation and drainage district of Chicamocha and Firavitoba (UsoChicamocha). Irrigation units: (1) Holanda, (2) Surba, (3) Pantano de Vargas,
(4) Ayalas, (5) Duitama, (6) Cuche, (7) San Rafael, (8) Las Vueltas, (9) Tibasosa, (10) Ministerio and (11) Monquirá. UsoChicamocha Central Station.

resources (Cai and Xiong, 2017). Multi-Agent Systems (MAS) are performance measurement in the design of intelligent agents. Addi­
developed by the interaction of several agents that exchange informa­ tionally, the use of crop characteristics and neighbor behaviors for
tion between them and generally cooperate for solving problems. The negotiation and water distribution introduces proactive and social be­
role of an agent in MAS consists of its nature, interfaces, responsibilities, haviors in the irrigation agents of the ABM proposed in this paper, which
functions and behaviors. Agent-Based Modeling (ABM) is a type of MAS, can be implemented with real devices in crop fields (Jimenez et al.,
whose aim is to use rules of the local agents in the micro-level for finding 2020b). For more details about Multi-Agent Systems and their applica­
a collective behavior of several agents in a macro-level using modeling, tions to irrigation management, please refer to the previous work by
analysis and simulation (Grashey-Jansen, 2014). Jimenez et al. (2020a).
Researches have used ABM to model relationships between the micro This paper shows the design and implementation of an Irrigation
and macro levels and to infer or make decisions based on different AI Agent-Based Model IABM for the Large-Scale Irrigation and Drainage
techniques or expert knowledge. Emergence is a term associated to an District of Chicamocha and Firavitoba (UsoChicamocha) in Boyacá –
unexpected behavior that appears during the algorithms execution and Colombia, where water is a common resource shared by many agents.
was not defined by the programmer in the agents design specification. This district is currently comprised of eleven irrigation units that benefit
The ABM models capture these emerging phenomena and obtain a about 4000 ha of 6087 potentials. Irrigation waters are taken from the
complete overview of the system. ABM applications have been used in Chicamocha riverbed and carried to the fields through pumping stations.
crop modeling irrigation management systems (Matthews, 2006), water Since the implementation of the district in 1998, the bulb onion culti­
management policies in irrigation systems (Rege et al., 2015), resource vation has emerged as an alternative of commercial agriculture under
negotiation processes (Janssen and Baggio, 2016), emulation of irriga­ irrigation, looking for the agricultural reconversion of lands considered
tion water resource monitoring (Zhao et al., 2011) and simulation of for use in livestock. The establishment of productive systems with the
large-scale irrigation models, primarily by flood irrigation (Guyennon use of irrigation has caused difficulties and shortcomings mainly due to
et al., 2016). Competition for the use of water is increasing globally and the lack of knowledge of the biophysical and technological components
the development of models that allow simulating views of the competing associated with the use of water in the exploitation of horticultural
parties are necessary to negotiate water resources and mitigate conflicts. crops, especially the bulb onion, whose planting area in the district has
ABM models have also been used as a conflict resolution strategy for the reached more than 2000 ha per year (1000 per semester cycle). The
management of irrigation resources between stakeholders. Akhbari and specific problem of the UsoChicamocha irrigation district is the physical
Grigg (2013) explain the development of an ABM that takes into account and hydraulic infrastructure, which is considered adequate for the dis­
the interactions among the parties and how they can be encouraged to tribution of land irrigation, but becomes inefficient due to the non-
cooperate in the game to work toward a solution. Cai and Xiong (2017) application of technical criteria with concepts that indicate when, how
implemented an ABM simulation for studying cooperation formation much and how to irrigate (Cely, 2010).
when irrigation is used. The stakeholder first becomes a potential A characteristic of the proposed IABM is the management of local
participant when the cost of cooperation it needs to cover is not higher information in the cultivated fields: farmers are simulated as intelligent
than the amount it can afford or is willing to pay. On top of this, the agents with the capability of determining water requirements based on
propensity that the stakeholder participates is heavily affected by its the soil water balance equation and field data as soil textures and soil
personal characteristics and neighborhood effects. Finally, Belaqziz hydrologic parameters. Additionally, the IABM system is implemented
et al. (2014) implemented an ABM simulation focused on gravity irri­ with social capabilities represented by cooperative and non-cooperative
gation systems with an evolutionary strategy algorithm for the distri­ behaviors among neighbors agents and the ability to negotiate the dis­
bution in rate and time of water resources in a pilot region located in tribution of resources when there is not enough water to maintain all the
Morocco. Several researches have applied multi-agent models to study cultivated plots at field capacity. After the negotiation procedure, the
water distribution in cultivated regions with several stakeholders, but agents apply irrigation to their crops according to new prescriptions that
none have used neither crop phenological characteristics as a relevant seek to optimize the distribution of irrigation in the study region where
negotiation parameter nor agent satisfaction levels as irrigation the system was implemented. The contribution of this work is an Agent-

2
A.-F. Jiménez et al. Agricultural Water Management 256 (2021) 107088

Based modelling framework specifically designed for the management of

0.3
0.9
water resources in an irrigation district, including the representation of

165
19
collective decision-making procedures with the integration of real
geographical and weather data, python™ algorithms and NetLogo. This
framework allows the negotiation of water resources in an irrigation

0.4
1.2
18

165
district, taking into account the emerging behavior from the micro
(farmers) to macro (irrigation district) levels for the efficient manage­
ment of freshwater in agriculture, as one of the most important resources

0.6
1.5
for the humanity subsistence.

17

165
Maturation
2. Materials and methods

0.7
1.8
16

165
This research was developed in the large-scale irrigation and
drainage district of Chicamocha and Firavitoba (UsoChicamocha),
located in Boyacá, Colombia. The division of the territory belonging to

0.7
2.0
the UsoChicamocha irrigation district was defined by the Territorial

15

165
Arrangement Planning (TAP) of the Department of Boyacá, Colombia.
This governmental body divided the district into eleven territorial sec­
tors adjacent to the riverbed of the Chicamocha River to guide and

0.6
2.0
14

157
manage the physical development of the territory and the use of the
land. These territorial sectors defined in the TAP coincide with the
established fields in the irrigation units, without considering the spatial

Crop coefficient, crop evapotranspiration (ETc), and root depth of the bulb onion crop in the area of the San Rafael UsoChicamocha irrigation unit.

0.8
2.6
heterogeneities of the soils. However, climatic conditions of these irri­

13

148
gation units are homogeneous given their spatial proximity, tempera­
ture, relative humidity, rainfall and altitude level. Therefore, the district
has eleven irrigation units with one pumping station each: Holanda,

0.7
1.9
12

139
Surba, Pantano de Vargas, Ayalas, Duitama, Cuche, San Rafael, Las
Vueltas, Tibasosa, Ministerio and Monquirá (Fig. 1).
The water sources are La Copa and La Playa dams, and water is

0.7
2.2
11

130
distributed throughout the Chicamocha riverbed. Each pumping station
is responsible for supplying water to several fields. The experimental
fields used in this study corresponded to the San Rafael irrigation unit

0.9
2.6
that has an area of 8451.3 km2 and is located in 5.789569◦ N,

10

121
72.988214◦ W. Fig. 1 shows the cultivated fields supplied by the San
Rafael pumping station, located in the municipality of Nobsa, Boyacá, Bulbification
Colombia. The area comprised by the San Rafael irrigation unit has an
0.8
3.0
9

112
annual average of solar brightness of 1852hours and an average tem­

Note: rd is root depth (mm) ETc is crop evapotranspiration (mm∙day− 1 ) and kc is crop coefficient.
perature of 14.05◦ C, with a high qualification for the sowing of agri­
cultural species (Weather Spark, 2020). There is a humidity deficit of
0.7
2.3

485 mm/year derived from an annual average of rainfall and evapora­


8

103

tion rates of 798 mm/year and 1283 mm/year, respectively. The envi­
ronmental climate corresponds to dry-cold, with a bimodal regime and a
water deficit in most of the months, with a concentration of rains be­
0.7
2.5
7

94

tween of April and May in the first semester and October to November in
the second, according to 10-year data (Cely, 2010).
0.5
1.6
85
6

2.1. Crop characterization - bulb onion

The bulb onion crop (Allium cepa, Hybrid Yellow Granex) is the
0.5
1.4
5

76

principal product planted in UsoChicamocha. For this study, parameters


and characteristics of this crop were defined using the results of the
project entitled: “Determination of irrigation and drainage parameters
0.4
1.3
4

67

for the cultivation of bulb onion (Allium cepa) in soils of the Alto Chi­
camocha Irrigation District - Boyacá – Colombia (Cod.
1c10937819630)”, supported by the Administrative Department of
0.4
1.3

Science, Technology and Innovation of Colombia - Colciencias and


58
3

finished in 2009. According to this research, three phenological phases


Leaves-development

were defined for the development of this crop.


0.5
1.3

Phase I. Leaves development: It is characterized by a minimal


2

49

accumulation of phytomass and no very noticeable changes in the


Source: (Cely, 2010).

phenotype. This phase begins from transplantation to the beginning of


0.4
1.3

bulbification. Plants develop 58% of the total leaves, acquire 48% of its
40
1

maximum height, and 45% of the maximum number of cataphiles and


63% of the maximum length of the root are formed. These values are
Table 1

Week
Stage

significant for a single-phase, which corresponds to 40% of the total


ETc
Kc

rd

cultivation cycle. In this phenological state, the plant presents a gradual

3
A.-F. Jiménez et al. Agricultural Water Management 256 (2021) 107088

Table 2 Once ETr has been calculated, a crop coefficient value (Kc ) must be
Data description. multiplied by ETr for adjusting the value to the studied crop in its
Data name Time/ Time span Data source different development stages (Subedi and Chávez, 2015). The crop
space water requirements are expressed in millimeters, either per day, month
resolution or vegetative cycle. The definition of the phenological scale and Kc for
Meteorological Day/point 11–01–2019 GrowWeather station the study area are fundamental parameters to carry out the irrigation
variables to with Weather Wizard III programming that is required to be applied at each stage of the crop. The
08–01–2020 console (Davis crop coefficient describes the variations in the amount of water that
Instruments, Hayward,
CA, USA)
plants extract from the soil as they develop, from sowing to harvest, as it
Meteorological Day/point 07–01–2020 Weather Station Pro2 depends on the anatomical, morphological and physiological plant
variables to Plus- Datalogger characteristics (Spiliotopoulos and Loukas, 2019). Due to the charac­
08–01–2020 WeatherLink IP (Davis teristics of the bulb onion crop and the variations in the incidence of ETr
Instruments, Hayward,
during the different growth stages, Kc changes from planting to harvest.
CA, USA)
Interpolated maps: 1:25000 2009 Administrative According to Cely (2010), onion plants have abundant thin roots
Permanent Wilting Department of Science, measuring around 40 mm at the transplanting time. From the third week
point (PWP); Field Technology and after transplanting, the length of the roots begins to increase and
Capacity (FC); Innovation of Colombia consequently their weight. Subsequently, a significant increase in the
Saturation Point - Colciencias (Project:
(SP); clay, silt and 110937819630)
number of roots is evident, becoming abundant. Without interrupting its
sand percentages; growth in length, from week 8 onwards a thickening of the roots occurs.
hydraulic Around week 15, at the beginning of the maturation phase, the roots
conductivity and stop growing, reaching approximately 165 mm in length and maintain a
apparent density.
constant dry weight, until the end of the cultivation cycle. Table 1 shows
Polygon layer- 1:25000 2020 UsoChicamocha
Irrigation area map Irrigation District site-specific values of Kc , effective root depth (rd ) and ETc for the fields
Root depth (rd ) Parameter 2010 Cely (2010). located in the San Rafael irrigation unit of UsoChicamocha and defined
Crop Coefficient (kc ) Parameter 2010 Cely (2010). for bulb onion (Allium cepa, Hybrid Yellow Granex).
Management Parameter Allen et al. (1998).
allowable depletion
2.2. Field data collection
(MAD)

Table 2 describes the data sources used in this study. Daily meteo­
and constant increase in its growth and number of cells of meristematic rological data of precipitation, pressure, maximum temperature, mini­
tissues, besides several small and thin leaves appear, as well as very fine mum temperature, wind speed, radiation, evapotranspiration and
roots and cataphiles. relative humidity were collected from a Davis GrowWeather station with
Phase II. Bulbification: This stage is characterized by an increase in Weather Wizard III console (Davis Instruments, Hayward, CA, USA)
most of the physiological parameters, in which a representative growth located in a field of the San Rafael irrigation unit (5.787021◦ N,
is observed. At the beginning of this phase, the thickening of the base of − 72.973932◦ ), from November 1, 2019 to August 01, 2020. Precipita­
the leaves begins, the number of cataphiles increases and leaves tion and reference evapotranspiration (ETr ) data were used as input
continue to appear. The leaf area increases because its length increases parameters for the hydric balance equation. Rain was measured and ETr
in width and length and a thickening process occurs in the limb and, was computed on hourly time step throughout the 24 h period and
therefore, in the pseudostem. In the fullness of the phase, the equatorial summed up to a daily time step. The effective precipitation was calcu­
diameter of the onion bulbs increases to exceed 65% of its final diam­ lated through the methodology stated by Adnan and Khan (2009), where
eter, the fresh and dry bulb phytomass show a drastic change and the a kp coefficient is obtained from the relationship between evapotrans­
leaves remain erect in the fullness and end of the phase. The total dry piration and the observed precipitation by using Eq. (1). Then, the kp
phytomass increases because all the organs are presenting significant coefficient is multiplied by the observed precipitation and the result is
changes in length, thickness and stored products of photosynthesis. equal to the effective precipitation, as expressed in Eq. (2).
Phase III. Ripening: In this last phase, which goes until the harvest or
0.001143∗ET 0.5208∗ET
(1)
r r
maturity of the crop, the highest rate of accumulation of dry and fresh kp = 1.011 ∗ e− rain − 1.011 ∗ e− rain

phytomass of the bulb stands out. The bulb completes its filling and the
rate of appearance of cataphiles decreases, as they are thickening to Raineff = Rain ∗ kp (2)
consolidate the final formation of the bulb. The leaf part loses impor­
tance, since the foliar area and the number of leaves remain constant and where kp (dimensionless) is the precipitation coefficient, ETr (mm) is
low. At the end of this phase, the leaves die after the plants are over­ the reference evapotranspiration, rain(mm) is the precipitation and
whelmed, which basically consists in drying the bulb neck and as a Raineff (mm) is the effective precipitation.
result, the plant falls due to the weight of its leaves. It is observed that Taking as reference the semi-detailed soil study at scale 1:25000 of
the water consumption of the crop is higher in the second stage of UsoChicamocha (Cely, 2010), a total of 100 sites were geo-referenced
development, in addition to a shorter duration in days. This corresponds and used for defining textural, physical and hydraulic parameters
to the phase where the greatest accumulations of biomass occur and the maps. The geographical distribution of these sites represents the contrast
organs develop, that is, the critical period of the onion bulb cultivation of the soil pattern and climatic distribution of the area occupied by bulb
matches the formation of the bulb, a stage in which it consumes 47% of onion in the UsoChicamocha district. In this study, soil maps were
its water needs. prepared with spatial indicators, including textures, apparent density,
The water requirements in a crop are directly related to the evapo­ total porosity and its distribution, saturated hydraulic conductivity and
transpiration (ET), corresponding to the water that the plant uses to soil moisture points (saturation, FC and PWP). These maps were ob­
carry out its physiological functions, including that which the plant tained by ordinary Kriging interpolation in Quantum GIS (OSGeo, Bea­
expels through the leaves (transpiration), that which is lost from the soil verton, OR, USA) and SAGA (Dept. of Physical Geography, University of
where the crop is (evaporation) and the water intercepted by the foliage Hamburg, GER). The procedure consisted of a structural analysis that
(Allen et al., 1998). ETr is the calculated reference ET for grass (mm). evaluated data correlation and dependency through a variogram, and a
predictive analysis using ordinary Kriging (Arétouyap et al., 2015).

4
A.-F. Jiménez et al. Agricultural Water Management 256 (2021) 107088

Table 3
Variograms parameters and performance values of kriging interpolations.
Parameter Clay Silt Sand Density Saturation Point Field Capacity Permanent Wilting Point

Nugget 320 280 100 0.02 28 28 60


Still 450 360 210 0.06 75 88 160
Range 0.016 0.016 0.024 0.026 0.024 0.024 0.024
RMSE 0.63 0.48 039 0.005 0.19 0.21 0.29
MAE 0.46 0.32 0.27 0.003 0.13 0.15 019

(a) (b)
Fig. 2. Interpolated maps. (a) Field capacity and (b) Permanent wilting point.

Currently, Kriging is the best interpolation technique because it is un­ for Field Capacity and Permanent Wilting Point.
biased. Nevertheless, it requires data to be correlated and dependent. For the management of spatial data, a polygon layer was developed
This structural analysis is conducted by means of the variogram. A with all the fields belonging to the San Rafael irrigation unit (Fig. 2).
variogram is a tool that is used to describe the spatial continuity of a Zone Statistics tool in QGIS was used to extract the values of the inter­
phenomenon (Caridad and Jury, 2013). The variogram provides a polation maps (raster data) to the polygon layer. These values corre­
measure of spatial correlation by describing how the sample data are spond to the average amount of pixels of the interpolation maps that
related to distance and direction (Kaluzny et al., 1998). In order to apply belong to each field polygon of each spatial indicator. Finally, the geo­
the Kriging procedure, the variogram should be fitted to a simplified spatial information was organized in a vector file (in shape format *.shp)
model. Parameters that describe the variogram with sufficient detail are and a *.dbf file with fields of latitude and longitude of polygon centroid,
nugget, sill and range. The spherical model is the most common model in field area, sand, loam and clay percentages, apparent density, saturated
geostatistics and it was selected and implemented using Eq. (3). hydraulic conductivity and soil moisture points (saturation, FC and
⎧ ⎫ PWP). This vector layer corresponds to the spatial parameters that are
⎪ 0, h = 0 ⎪


⎪ ( )


⎪ constant in the study region and represent the invariant aspects of the
⎪ ( )3 ⎪
⎨ h h ⎬ environment in the proposed Agent-Based Model.
γ(h) = C0 + C 1.5 ∗ − 0.5 ∗ , 0<h≤a (3)

⎪ a a ⎪

⎪ ⎪





⎭ 2.3. Irrigation prescription
C0 + C, h > a
A precision irrigation prescription implies watering in such a way
where γ(h) is the variogram, h is the offset, C0 is the nugget variance,
that it represents the necessary humidity to obtain optimum yields. The
C0 +C is the Sill (scale), C is the partial sill (variance), and a is the
total available water (TAW) is the quantity of water that can be stored in
effective range (Arétouyap et al., 2015). Semivariograms were used for
the soil and used by crops (Barradas et al., 2012). TAW is defined by
understanding the spatial dependence for each spatial indicator. For
using the field capacity (FC) and the permanent wilting point (PWP)
each indicator, nugget, sill and range were defined for the spherical
parameters (Eq. (4)).
model and the results of the Kriging ordinary interpolation were eval­
uated using the Mean Absolute Error (MAE) and the Root Mean Square %VWCFC − %VWCPWP
TAW = ∗ Rd (4)
Error (RMSE), according to Table 3. Fig. 2 shows the interpolated maps 100

5
A.-F. Jiménez et al. Agricultural Water Management 256 (2021) 107088

Fig. 3. Model of the Class-Level organization structure for the San Rafael PumpStation-IABM.

where %VWCFC is the volumetric water content at FC, %VWCPWP is the root zone is assumed to be lost through RO or P (Andales et al., 2011).
volumetric water content at PWP and Rd is the effective root depth in The irrigation method efficiency (Ef) (60% for surface irrigation, 75%
mm. The FC is related to the quantity of water that a soil can retain after for sprinkler irrigation and 90% for drip irrigation) is added in the cal­
saturation caused by rain or irrigation activities, with free drainage and culus in Eq. (6).
without losses by evaporation, until the soil moisture stabilizes. The 100
PWP is the point under which a plant cannot extract water from the soil DT = D∗ (6)
Ef
and cannot recover from the loss of water, even if water is applied again
(Ghorbani et al., 2017). where DT is the total irrigation (mm), D is the net irrigation (mm) and Ef
In the IABM, irrigation scheduling must be evaluated for each field is the field application efficiency (%). Finally, the irrigation time is
belonging to the San Rafael irrigation unit of UsoChicamocha. The determined using Eq. (7).
management of irrigation prescriptions in the IABM is used to keep each
DT
field within a soil moisture range defined by the percentage of permis­ Irrigation Time (IT) = (7)
Irrigation application rate
sible level of moisture depletion (MAD). If the soil is sufficiently wet, it
supplies water fast enough to meet the atmospheric demand of the crop The MAD is used as a threshold value for defining the irrigation
and keep water uptake equal to ETc. The knowledge of the initial soil timing associated to the soil water depletion from the FC to the current
water content in the effective root zone, ETc , TAW, rain and irrigation VWC. To enable a crop to extract water from the root zone without
amount on the specific day are necessary for applying the soil water suffering water stress, only the Readily Available Water (RAW) of the
balance equation. Irrigation is required when ETc exceeds the water Total Available Water (TAW) is allowed to be depleted as shown in Eq.
supply. The net irrigation (D) is the difference between FC and the (8) (Al-Kaisi et al., 2009).
current soil volumetric water content. In the field, deep percolation
MAD
P(mm) and runoff RO(mm) are difficult to estimate. Assuming that Irr RAW = TAW × (8)
100
and Raineff readily infiltrate into the soil and discarding RO and P, the
net irrigation is calculated using Eq. (5). The RAW is used as a guide for deciding when to irrigate,
( ) whenever D ≥ RAW. This condition avoids water stress in the crop
D = D− 1 − Irr − Raineff + ETc if D is negative, then set it to 0.0 (5) and does not depend on the capabilities of the irrigation system.

where D (mm) is the soil water deficit or net irrigation requirement; 2.4. Agent modeling language and framework
D− 1 (mm) is the soil water deficit on the previous day; Irr (mm) is the
irrigation amount for the current day; Raineff (mm) is the effective The multi-agent programmable modeling environment NetLogo was
precipitation for the current day and ETc (mm) is the crop evapotrans­ used for the IABM development (Jaxa-Rozen et al., 2019). Netlogo al­
piration or crop water requirement for the current day. The unit mm lows the implementation of the FIPA specifications (Foundation for
means cubic meters of water per square meter of soil. The term D is set Intelligent Physical Agent). The framework uses two extra libraries: one
equal to zero if its value becomes negative when precipitation and/or for Belief, Desires and Intentions agent architecture (BDI), and one for
irrigation exceed D− 1 plus ETc . This means that water added to the soil Agent Communication Language (ACL) that allows the development of
exceeds field capacity within the plant root zone. Any water excess in the goal-oriented agents that communicate using FIPA-ACL messages

6
A.-F. Jiménez et al. Agricultural Water Management 256 (2021) 107088

Fig. 4. Multi-agent irrigation system sequence diagram.

(Sakellariou et al., 2008). NetLogo was selected because it allows the use society abstractions. An organization is a specialized environment type
of geographical information layers, and connectivity with hardware in the IABM. The IABM organization is specified in the model of the
devices and algorithms developed in Python3™ (Wilensky and Rand, Class-Level organization structure (Fig. 3). For the external perspective,
2015). the organization represents an autonomous entity. The organization
Turtles, patches, links and the observer are different types of agents behavior consists, on the one hand, in the emergent properties and
or entities in NetLogo. These agents are able to carry out their own ac­ behavior of their agents, their mutual relationships, observations and
tivities and follow instructions at the same time. Turtles are the "beings" interactions, and on the other hand, in the features and behavior of the
that move around the “world” (the grid of patches). The world is two- entity as a whole. For the internal perspective, organization units are
dimensional and it is divided into a grid of patches. Each patch is a types of environments specified by structures, interactions, roles, con­
squared region of "terrain" on which the turtles can move. The patches straints, rules, among other elements (Cervenka et al., 2005). Addi­
have coordinates and cannot move, but they may have similar properties tionally, agents act not only in isolation but also in cooperation or
to those of turtles. The links are the agents that connect two turtles. coordination with other agents. The resources in the IABM are the water
Finally, the observer has no place in the “terrain”, as it is looking at the and the information stored in the databases.
world of turtles and patches and controls the experiment. Each patch Fig. 3 shows a class diagram depicting the generic organization
and turtle can have their own user-defined variables, thus allowing the structure of the San Rafael Pump Station in the UsoChicamocha irriga­
simulation of complex environments. tion district. The organization comprises a psa agent (of the Pump­
According to the ABM methodology proposed by Wilensky and Rand Station entity role type), a wsua agent (of the WaterSupply entity role
(2015), the first phase is the multi-agent framework design, which takes type), a wsa agent (of the WeatherStation entity role type) and n fa
into account the spatial distribution, interactions and relationships that agents (of the Field entity role type). The fa agents are peers to each
can exist between the agents. The IABM proposed consists of the inte­ other (the cooperate connector), and subordinates to the psa agent (the
gration of the supervising-planning subsystem and the customer sub­ manage connector). The psa agent is superordinate to the wsua agent
system. The supervising-planning subsystem (SPS) is responsible for (the manage connector). Each fa agent of the San Rafael Pump Station
monitoring the irrigation agent status, definition of prescription maps organization comprises a sds agent (of the SmartDecisionSystem entity
using field data, mapping using global information and coordination of role type), a ps agent (of the PerceptionSystem entity role type) and an
conflict resolution between agents. In the SPS, the Pump Station Agent iae agent (of the IrrigationActivationEmulation entity role type). The
(psa), the Weather Station Agent (wsa) and the Water Supply Agent number of fa agents (n) depends on the number of simulated fields.
(wsua) interact. The customer subsystem is constituted by Field Agents
(faj) in charge of the irrigation prescription and the application of water 2.4.1. Sequence diagram
resources, where j is the number of the field corresponding to the fa. The sequence diagram is used for modeling the interactions between
These types of agents accomplish together a fully functional ABM with agents. With this representation, it is possible to illustrate how the
capabilities of autonomy, coordination, deliberation, conflict resolution different agents of the IABM system interact with each other to carry out
and decision making. the irrigation management and the order in which these interactions are
The Agent Modeling Language (AML) is used for specifying the agent performed. For the San Rafael irrigation pump station crop fields were
roles, behavioral aspects and interactions between agents (Cervenka defined according to the UsoChicamocha cartographic map. A lifeline
et al., 2005). Entities in AML correspond to agents, resources, environ­ represents an individual participant in the sequence diagram and it is
ments or organizations, designated to explicitly represent various (MAS) shown using a rectangle for its head, followed by a vertical dashed line

7
A.-F. Jiménez et al. Agricultural Water Management 256 (2021) 107088

that represents the lifetime of the participant. Participants in the pro­ irrigation area, using Eq. (15).
posed IABM are the psa, wsua, wsa, fa1, …, fan (Fig. 4). Synchronous ∑
messages between lifelines were defined. The sender waits for the NTAW(NC) = AWj (NC) + TSW (14)
receiver to process the message and returns an answer before carrying
on with another message. Synchronous messages between lifelines were NTAW(NC)
AWj (NC) = ∑ × FAj (NC) (15)
defined. The sender waits for the receiver to process the message and FAj (NC)
returns before carrying on with another message. In each irrigation At the end of this procedure there is a new cooperative (C) and non-
event, the pump station agent requests from the wsua the total supply cooperative (NC) assignation to all the field agents. If an agent j has
available water (TSAW). Each fa requests rain and reference evapo­ AWj ≥ AWRj , it is cooperative (C), otherwise the agent is non-
transpiration data to the wsa. All the fas plan irrigation and determine cooperative (NC). Eqs. (12)–(15) are repeated until the quantity of
the amount of water they required from wsua. At this stage the fas do TSW = 0 or the number of non-cooperative agents is equal to zero. The
not actually perform irrigation, and only send these values to the psa. last case occurs if the resources are enough to satisfy the needs of all the
The psa verifies that the TSAW is sufficient for the water demand. The fields. In this case, there is no negotiation, all the agents can use the
psa verifies that the TSAW is sufficient for the water demand. If this is irrigation prescription defined, and agents are cooperative. Those as­
not the case, the psa performs the conflict resolution protocol, and signments are used for defining the proportion of neighbors of agent j
finally, sends the final irrigation scheduling to the fas who carry out the with cooperative (C) and non-cooperative (NC) behaviors. Utility func­
irrigation application. tions were defined using Eqs. (16)–(19), according to the model devel­
The model was adjusted according to crop water requirements, soil oped by Edwards et al. (2005), which is inspired in the sociologic
parameters and the characteristics of the San Rafael irrigation unit of diffusion model for water domains. The decision of an agent to change
UsoChicamocha. The ABM implemented in NetLogo acquires informa­ its cooperative or non-cooperative behavior in the negotiation process
tion about the environment using the vector dataset stored in a * .shp relies on a trade-off between social pressure and personal interest. Social
file. Afterwards, each fa determines its irrigation prescription using its pressure is related to the behavior of neighboring agents, while the
local information, such as soil characteristics, crops and sowing date, personal interest is related to the goal of the agent to apply the necessary
and climatic parameters, such as rainfall and reference evapotranspi­ water according to the phenological state of the plants in order to ensure
ration. The Total Water Required (TWR) amount is defined as the sum of the best production in its field.
all the individual agent water requirements (AWR). It is calculated with
the irrigation prescriptions for each field in the region covered by the ψ j (C→C) = a × Pj (C) + I (16)
San Rafael irrigation unit, using Eq. (9). ( ) )
ψ j (C→NC) = 1 − a × (1 − Pj (C) (17)

n
TWR = AWRj (9)
j=1 ψ j (NC→C) = c × Pj (C) + I (18)
( ) )
where j represents each field monitored by an agent and AWRj is the ψ j (NC→NC) = 1 − c × (1 − Pj (C) (19)
water requirement defined by the agent in charge of the field j. The total
supply available water TSAW for the fields belonging to the San Rafael where ψ j (C→C) is the utility function defined for maintaining the
irrigation unit is the total available water that can be used in an irri­ cooperation status of the agent j, ψ j (C→NC) is the utility function
gation event. The TSAW is set by the wsua agent. The available water defined when an agent j changes from cooperation status to non-
per field area AWj is defined using each field area FAj and Eq. (10). cooperation status, ψ j (NC→C) is the utility function defined when
TSAW an agent j changes from non-cooperation status to cooperation status
AWj = ∑n × FAj (10)
j FAj
and ψ j (NC→NC) is the utility function defined for maintaining the
non-cooperation status of agent j. Pj (C) is the proportion of neighbors of
The supply and demand relationships are determined by subtracting
agent j of cooperative behavior (C). The proportion of neighbors for each
the TWR from the TSAW (Eq. (11)).
behavior was calculated using NetLogo, by defining the number of
ODR = TSAW − TWR (11) neighbors in a radius r from the centroid of each polygon feature of the
shapefile that corresponds to the location of each agent. When agents
On the one hand, if ODR < 0, there is a conflict resolution protocol
remain in their current behavior (C→C) or (NC→NC), a = 1 − c = 0.7.
based on the negotiation of water resources between agents, which
On the contrary, when agents change to the opposite behavior (C→NC)
represents cooperation between them. On the other hand, if ODR > 0,
or (NC→C), 1 − a = c = 0.3, according to Akhbari and Grigg (2013). I
there is no need for negotiation between agents, which means that there
is an operator related to the environment and the field agents, and serves
are no cooperation events and field agents receive authorization for
as a modification factor that is, a function of water requirements ac­
applying their initial prescriptions. Initially, in the negotiation process,
cording to the phenological status of crops. The maximum value of the
if an agent j has AWj ≥ AWRj it is cooperative (C), otherwise the agent is
utility function is 1 and the maximum value of I is 0.3. This value was
non-cooperative (NC). The water distribution protocol begins with the
obtained by solving Eq. (16), when all the neighbors are collaborative
assignation of AWj equal to AWRj for cooperative agents. For each
Pj (C) = 1 and the maximum value of ψ j (C→C) = 1= 0.7 × 1 + I, as it
cooperative agent j, there is a saved water amount (SWj (C)) defined
is the highest utility value that can be obtained using Eqs. (16)–(19). The
using Eq. (12).
proposed modification factor function decreases linearly from the
SWj (C) = AWj (C) − AWRj (C) and AWRj (C) = AWj (C) (12) beginning of the crop-development phase (0.3) to the beginning of the
mid-season or bulbification phase for bulb onion (0.0). Once the bul­

TSW = SWj (C) (13) bification phase is finished, the utility function increases linearly until
the end of the late season or ripening phase for bulb onion (0.3). Ac­
The total saved water amount TSW is defined by the sum of all the cording to this function, the priority irrigation is performed in the bul­
SWj (C) according to Eq. (13). The new total available water for non- bification phase, as it is the critical period in bulb onion cultivation,
cooperative agents (NTAW(NC)) is obtained by adding TSW to the hence there will be no pressure on the agent to cooperate and the agent
total available water of the non-cooperative agents using Eq. (14) and it might only be influenced by its neighbors.
is distributed between the non-cooperative agents according to each The linear behavior is established in the transitions between the

8
A.-F. Jiménez et al. Agricultural Water Management 256 (2021) 107088

Fig. 5. Relationship between (a) crop coefficient (FAO, 2020) and (b) proposed modification factor ℑ.

phases of leaves-development to bulbification and bulbification to defined by the sum of all the Ij (C) according to Eq. (22).
ripening, as it is the simplest function to represent the changes, which is ∑
inversely proportional to the behavior of the crop coefficient during the TI = Ij (C) (22)
plant development of a crop, according to FAO (2020) (Fig. 5). The New Total Available Water for the Non-Cooperative agents
The utility function ψ j corresponds to the one with the highest value, (NTAW(NC)) is obtained by adding TI to the total available water of the
taking into account whether the agent was initially cooperative (Eqs. non-cooperative agents using Eq. (23) and it is distributed between the
(16) and (17)) or non-cooperative (Eqs. (18) and (19)). A study on onion non-cooperative agents according to each irrigation area using Eq. (24).
irrigation in Southwest Texas found that reducing the amount of applied ∑
water to 75% of the recommended level resulted in a moderate reduc­ NTAW(NC) = AWj (NC) + TI (23)
tion in marketable yield. In addition, this condition showed a substantial
reduction in the number of larger bulbs sizes (Leskovar et al., 2012). The NTAW(NC)
AWj (NC) = ∑ × FAj (NC) (24)
recommended agricultural management to produce better shape index FAj (NC)
of onion bulbs, is to irrigate onion plants using irrigation level between
80% and 100% of water requirements (Mubarak and Hamdan, 2018). If At this stage, the negotiation process is developed only with the non-
the agent is cooperative, in the proposed method, RAWj (C)/2 is the cooperative agents. AWRj (NC) − AWj (NC) is the maximum water de­
mand for the non-cooperative agent j. By applying the I factor for non-
maximum water amount that it is willing to compromise in the negoti­
ation process, that corresponds to maintain the soil at 87.5% of water cooperative agents and calculating the fas utilities ψ j , the demand
requirements. By applying the I factor and calculating the fas utilities modification amount (Ij (NC)) is defined using Eq. (25).
ψ j , the demand modification amount for the cooperative agent (Ij (C)) is ( ) {( ( ) ( )) ( ) (
AWRj NC − AWj NC × 1− ψ j ;∀j,AWRj NC > AWj NC
) ( )
defined using Eq. (20). Ij NC = ( ) ( )
0; ∀j,AWRj NC ≤ AWj NC
/ )
Ij (C) = RAWj (C) 2 × (1 − ψ j (20) (25)

Then, the new demand of agent j (NAWRj (C)) is calculated using Eq. Then, the new demand of agent j (NAWRj (NC)) is calculated using
(21). Eq. (26).
{ ( ) { ( )
0; ∀j, AWRj NC = 0
NAWRj (C) =
0; ∀j, AWRj (C) = 0
/ (21) NAWRj NC = (26)
RAWj (C) 2 + Ij (C); ∀j, AWRj (C) > 0 AWj (NC) + Ij (NC); ∀j, AWRj (NC) ∕
=0

The total offer modification amount for the cooperative agents (TI) is The non-cooperative total new demand TND(NC) is calculated by

Fig. 6. Influence of cooperative neighbors Pj (C) (%)


and modification factor ℑ on agent j: (a) agent utility
(ψ j ) vs. Pj (C) (%) for cooperative agent behavior, (b)
ψ j vs. Pj (C) (%) for non-cooperative agent behavior,
(c) changes in demand modification amount for
cooperative agent behavior according to their utility
with AWRj = AW j = RAW = 115.9 m3 (3.591mm)
and (d) changes in demand modification amount for
non-cooperative agent behavior according to their
utility, AWRj > AW j , where AWRj = 115.9 m3
(3.591 mm) and AWj = 64.5m3 (2 mm). Agent water
requirement (AWRj ) and field available water
(AWj ). PWP = 36.60%, FC = 52.56%, effective root
depth = 90 mm and area of 32289.87m2 .

9
A.-F. Jiménez et al. Agricultural Water Management 256 (2021) 107088

Fig. 7. NetLogo interface - IABM-WNeg.

adding all the NAWRj (NC) according to Eq. (27). PWP = 36.60%, FC = 52.56%, with an effective root depth of rd =
∑ 90 mm and an area of 32289.87m2 , the RAW corresponded to ((FC −
TND(NC) = NAWRj (NC) (27)
PWP)/100) ∗ rd ∗ 0.25 = 3.591mm, equivalent to 115.9m3 for the total
While NTAW(NC) < TND(NC), there is an iterative process of water area. Fig. 6a shows the behavior of cooperative agents and Fig. 6b the
reduction waj (m3 ) according to a water amount ρ (mm) and the field area behavior of non-cooperative agents. These figures show the influence of
(m2 ) by field agent, until TSAW ≥ TND for all the cooperative and non- the modification factor in decision-making. In Fig. 6a, if the percentage
cooperative agents using Eqs. (28) and (29). of cooperating neighbors is less than 30%, the cooperative agent
changes its behavior to non-cooperative due to the influence of its non-
waj (NC) = FAj (NC) ∗ ρ (28) cooperative neighbors. If the modifying factor increases, the agent is
more permissive and does not change to the non-cooperative state.
NAWRj (NC) = NAWRj (NC) − waj (NC) (29) Furthermore, if the modification factor is high, the crop is not in the
While NTAW > TND, there is an iterative process of water distribu­ bulbification phase and the agent remains as cooperative. As the per­
tion of NTAW − TND between all the cooperative and non-cooperative centage of cooperating neighbors increases, the agent will remain in its
agents according to each irrigation area, where if NAWRj > AWRj , cooperative status despite the change in the values of the modification
then NAWRj = AWRj . According to this algorithm, the non-cooperative factor.
field agents demand more water. However, this new demand could be In Fig. 6b, if the percentage of cooperative neighbors is less than
less than the initially requested and will depend on the behavior of the 70%, the non-cooperative agent remains with this behavior. If the
neighbors and how they cooperate or not in the distribution of the modification factor increases, the agent is more permissive and,
resource. At the end of the negotiation stage, the total supply available depending on the value of this factor, it changes to cooperative behavior.
water TSAW amount has to be more than or equal to the total new de­ The utility of the cooperative agent is generally greater than the utility of
mand TND. In addition, after the negotiation, if NAWRj < AWj , then the non-cooperative agent when the percentage of cooperative neigh­
bors increases. In case of a reduction in the available water, the agent
NAWRj = AWj . Finally, the irrigation agents can define irrigation
that initially presented cooperative behavior has the capability of
events, but they cannot apply water to their fields without the coordi­
sharing a water amount, while the non-cooperative agent requests
nator agent permission.
additional water for satisfying its needs. This behavior is evidenced in
Fig. 6c and 6d for water amount requested after negotiation of the
3. Results and discussion
cooperative and non-cooperative agents, respectively. In Fig. 6c the field
agent had AWRj = AWj = RAW = 115.9 m3 (3.591mm) then, it was
The IABM field agents obtain data from their fields, define irrigation
initially cooperative.
prescriptions and in the supervising-planning subsystem (SPS) the con­
With the negotiation process, each cooperative agent allows the
flict resolution protocol is developed for the distribution of water re­
reduction of water with respect to its initial irrigation prescription, ac­
sources, according to the total available water. Fig. 6a and 6b show the
cording to the modification factor I (Fig. 6c). If I is low, the crop is in
behavior of the utility function for the field agents according to changes
the bulbification phase and the agent allows less water reduction with
in the modification factor with respect to the percentage of cooperative
respect to I = 0.3. In Fig. 6d, the field agent has AWRj > AWj , where
neighbors. The pilot crop for this behavior analysis was located at
5.787980◦ North latitude and 72.974122◦ longitude in the municipality AWRj = 115.9 m3 (3.591 mm) and AWj = 64.5m3 (2 mm), then, it was
of Nobsa, Boyacá, Colombia. According to the soil sample in this field, initially non-cooperative. Additionally, if I is low, the crop is in the
bulbification phase and the agent requests more water with respect to

10
A.-F. Jiménez et al. Agricultural Water Management 256 (2021) 107088

Fig. 8. Maps obtained with IABM-WNeg: (a) Vector file -.shp polygons, (b) permanent wilting point, (c) field capacity, (d) potential evapotranspiration, (e) crop
coefficient, (f) weeks after transplanting, (g) crop evapotranspiration, (h) initial irrigation and (i) initial soil moisture content.

I = 0.3, for Pj (C) higher than 40%. An aspect to consider in future re­
searches is the evaluation of other modification functions and their
response to the differences between the required water and the available
water.

3.1. IABM-Wneg

The software application of the Irrigation Agent Based Model for


Water Negotiation (IABM-WNeg) was developed for studying the
behavior of spatio-temporal variables for water management in the
Usochicamocha irrigation district. IABM-WNeg integrates geographic
information systems (GIS), site-specific technical irrigation prescriptions
and agent-based modeling tools. The negotiation protocol is performed
by obtaining a new agent irrigation prescription per field and finally
implementing a satisfaction algorithm.
Fig. 7 shows the graphical user interface (GUI) of the IABM-WNeg
software application. The GUI is divided into the following seven sec­
tions. First, the Weather Station allows defining and modifying the pa­
rameters of rain and reference evapotranspiration. Second, DataSetMaps
allows loading and viewing the vector file data maps. Third, the Calcu­
lated Maps section enables the calculation of irrigation parameters maps
according to the real information on the fields. Fourth, Conflict resolution
allows the calculation of cooperative and non-cooperative behaviors,
negotiation and satisfaction factor dependent on the soil water stress.
Fifth, Day Simulation allows running all the procedures for obtaining the Fig. 9. Irrigation prescription map obtained with IABM-WNeg.
irrigation prescription after negotiation by one simulation day. This
option is called using a Python™ script that improves the data obtention

11
A.-F. Jiménez et al. Agricultural Water Management 256 (2021) 107088

Table 4
Simulation results with six total supply available water amounts. MTWR = 40030 m3 .
Percentage of MTWR 0.25 10 30 50 70 100

TSAW(m ) 3 100 4000 12000 20000 28000 40000


TWR (m3 ) 41892.2 26128.2 26050.4 26609.0 26243.7 26509.9
TIPAN 99.9 3995.4 11996.9 19993.7 26243.7 26509.9
SL Fields (%) 0 24 90 100 100 100

Note: Maximum Total Water Required (MTWR), Total Supply Available Water (TSAW), Total Water Required (TWR), Total Irrigation Prescription After Negotiation
(TIPAN), Satisfied Level (SL). 890 fields.

for analysis and plotting. Sixth, Temporal Simulation allows running the that the satisfied level occurs when fields have no water stress. The
simulation for several days and plot the temporal behavior of the tank satisfaction level scale from 0 to 1 is used as a visualization mechanism
level, irrigation prescriptions before and after negotiation, and the in the software application of the Irrigation Agent Based Model for
percentage of satisfaction response of agents. Seventh, the Environment is Water Negotiation (IABM-WNeg).
the region where the agents are located and where their spatial behavior Table 4 shows the model capability for distributing water supply
is displayed. It contains the pump station agent (psa), the weather sta­ amounts among the fields belonging to the San Rafael pumping station. The
tion agent (wsa), the water supply agent (wsua) and the field agents dissatisfaction appears when the TSAW is less than 40% of the MTWR,
(faj). Fig. 8 shows the maps that can be obtained using the IABM-Wneg which means that water prescribed per field is insufficient to maintain
application. With the information collected and mapped, the irrigation crops in the best conditions, but at least there is a data-driven criterion for
rescription for the day studied can be calculated, thus obtaining the the distribution of resources according to areas and crop needs. For 0.25%
irrigation prescription map in Fig. 9. of MTWR, the simulation was run with an initial soil moisture between
PWP +(FC − PWP)/2(%) and PWP +(FC − PWP)/2 – λ(%) for each
field. The constant value λ was selected at a value of 3% for obtaining
3.2. One day irrigation negotiation - performance
random values in each field and extract a chart with the necessary char­
acteristics for the explanation and analysis of the data, thus ensuring
The model performance for multiple agents was evaluated using
various levels of dissatisfaction. Fig. 10a shows that, in this case, all crops
several amounts of TSAW. The Maximum Total Water Required (MTWR)
presented water stress and that the color scale allows knowing their
was defined by adding the Readily Available Water (RAW) of all the
severity. The satisfaction level 0 corresponds to the permanent wilting
fields in the study region using FC and PWP. According to Cely (2010),
point.
for the cultivation of bulb onion grown in the UsoChicamocha region,
Fig. 11 shows the relationships between agent water requirements
the MAD corresponds to 25% and the maximum effective root depth is
(AWR), the initial available water (AW1 ), the available water before the
0.165m. For the San Rafael irrigation unit, the MTWR calculated was
negotiation process (AW) and the irrigation prescription after negotia­
40030.422 m3 by adding all the RAW amounts and taking into account
tion (IPAN), per field area using 890 agents in the simulation. When
the cultivated areas and a root depth of 10 cm. Table 4 shows the results
water resources were insufficient, each agent received the acceptance
of the simulation for six amounts of the total supply available water
for applying irrigation at a very low value, which was rarely greater than
TSAW selected within the MTWR calculated. Fig. 10 shows the coop­
the amount of AW for each one (Fig. 11a and 11b). Irrigation pre­
erative and non-cooperative behavior irrigation prescriptions after
scriptions for the fields in this case were defined according to the
negotiation and satisfaction level of agents for the five smallest TSAW
cultivated area, but without satisfying the irrigation needs of most fields,
values of Table 4, because AWRj < AWj for TSAW values greater than
as represented by the high difference between the AWR and the IPAN for
28,000, in all the fields. It can be observed that the number of collab­
each field.
orative agents increases with the available water and it is enough for all
Fig. 11c with 30% of TSAW shows that the negotiation algorithm
the fields. The non-collaborative agents appear because the water
became more evident when the total amount of water available
availability for these fields was insufficient to satisfy the needs of the
increased, which allowed the reduction of the prescriptions in some
crops before the negotiation process.
fields and the addition of these resources to other fields trying to satisfy
When soil water is extracted by evapotranspiration, the depletion
their crops needs. Fig. 11d shows that with 50% of available water, there
increases and stress will be induced when D becomes equal to RAW. The
was more distribution of the resource through negotiation decisions
water stress coefficient (ks ) is calculated using Eq. (30) (Kokkotos et al.,
than through allocation by the division of water by areas, as there was
2020).
enough water resource to negotiate. Several agents did not obtain the
( )
TAWj − Dj − IPANj amount of water that they initially prescribed, but many of them
ksj = (30) received more water than the initially available for their fields. It was
TAWj − RAWj
also observed that when the TSAW increased and was sufficient to
where RAW is the Readily Available Water (mm), TAW is the Total supply the demand of the crops, there was no negotiation process and
Available Water (mm), D is the soil water deficit or net irrigation these fields received the values prescribed by the IABM (Fig. 11e). One
requirement (mm), IPAN is the irrigation prescription applied after of the relevant aspects of the proposed model is that, despite the water
negotiation (mm) and j is the agent identification number corresponding resource is much greater than the prescriptions, the negotiation process
to the same field number. According to Fig. 10, if ks ≥ 1 (satisfied level), ensures that the fields will use irrigation prescriptions defined with
there is no soil water stress and the available water is within the 25% of technical and agronomic criteria, thus avoiding the waste of the water
TAW (purple color); but if 0< ks < 1 (dissatisfaction levels), there is a resource. On the other hand, when water resources are very small, the
stress level in the corresponding field (blue to red colors). If there is IABM allows equitable distribution by field size and irrigation priority
water stress, it is assumed that the agent is unsatisfied because the crop according to phenological states.
could have problems in its normal development. If there is water scarcity Taking into account the results of Figs. 10 and 11, it can be stated
and water resources are extremely low, it is impossible to satisfy the that the IABM developed meets the requirements of seeking the satisfied
needs of all the fields and there are higher levels of dissatisfaction due to level of the greatest number of agents by avoiding as much as possible
values of less than one in the water stress. Additionally, the satisfaction that the water stress coefficient decreases from 25% of the TAW (ks < 1),
level also increases with the irrigation water available. Results highlight considering crop coefficients and root depths. Figs. 10 and 11 show that

12
A.-F. Jiménez et al. Agricultural Water Management 256 (2021) 107088

13 (caption on next page)


A.-F. Jiménez et al. Agricultural Water Management 256 (2021) 107088

Fig. 10. Simulation results for an irrigation prescription day on the San Rafael pumping station: Cooperative and non-cooperative behavior (Left), irrigation pre­
scriptions after negotiation (center), and satisfaction level of agents (Right). (a) TSAW = 100m3 (0.25% of MTWR), (b) TSAW = 0.40 × 104 m3 (10%
of MTWR), (c) TSAW = 1.20 × 104 m3 (30% of MTWR), (d) TSAW = 2.00 × 104 m3 (50% of MTWR) and (e) TSAW = 2.80 × 104 m3 (70% of MTWR).
TSAW is the total supply available water and MTWR is the maximum total water required. (For interpretation of the references to colour in this figure, the reader is
referred to the web version of this article.)

Fig. 11. Relationship between agent water requirements AWR, initial available water per field area (AW1 ), available water per field area before the negotiation
process (AW), and irrigation prescription after negotiation IPAN for each agent in the simulation. Water amounts per agent (Left); AWR and IAN vs. AW (Right). (a)
TSAW = 100 m3 (0.25% of MTWR), (b) TSAW = 0.40 × 104 m3 (10% of MTWR), (c) TSAW = 1.20 × 104 m3 (30% of MTWR), (d) TSAW = 2.00 ×
104 m3 (50% of MTWR) and (e) TSAW = 2.80 × 104 m3 (70% of MTWR). TSAW is the total supply available water and MTWR is the maximum total
water required.

14
A.-F. Jiménez et al. Agricultural Water Management 256 (2021) 107088

when the TSAW increases, the number of field agents with cooperative CRediT authorship contribution statement
behavior also increases. When TSAW is large enough, there is no need of
the negotiation procedure, field agents are collaborative and 100% Andrés-F Jiménez, Pedro-F Cárdenas and Fabián Jiménez con­
satisfaction will be achieved. In the negotiation algorithm, the principal ducted Conceptualization, Methodology and original draft preparation.
constraint is that the sum of final prescriptions for each field after Andrés-F Jiménez and Fabián Jiménez performed Writing – review &
negotiation must not exceed the limit for the amount of water available editing and Visualization.
throughout the region.

4. Conclusions Declaration of Competing Interest

In this article, we showed the particular applied case of the San The authors declare that they have no known competing financial
Rafael irrigation unit of the UsoChicamocha irrigation district. The type interests or personal relationships that could have appeared to influence
of information available about resources and the environment in the the work reported in this paper.
Irrigation Agent Based Model (IABM) implemented plays a key role not
only in the evolution of each field agent model (micro-level) but also in Acknowledgements
the interaction of multiple agents (macro-level). For the indicators
tested, the difference seems to be associated with the availability of A.F. Jiménez expresses his gratitude to the Department of Boyacá
water for supplying all the irrigation needs and the decision mechanisms and Minciencias – Colombia for the support through the scholarship
of each agent in the fields. program No. 733 - 2015 for the Ph.D. program at Universidad Nacional
The IABM developed makes water allocation decisions based on crop de Colombia and to Universidad de los Llanos. Pedro Cardenas expresses
coefficients, root depths, the soil water balance equation and weather his gratitude to MinCiencias - Colombia for the abroad Ph.D. scholarship
data. This tool is especially useful during water scarcity when farmers – 2007. F.R. Jiménez expresses his gratitude to Universidad Pedagogica
are required to make water reduction in the farm, but also when the y Tecnologica de Colombia for allowing the development of this work as
regional waste of water is a priority for sustainable resources manage­ a result of a research proposal for the teaching promotion.
ment. Based on the simulations results, it can be observed that the
proposed model was able to save water even when water reduction is in References
place.
The results presented in this paper are very specific for several fields Adeyemi, O., Grove, I., Peets, S., Domun, Y., Norton, T., 2018. Dynamic neural network
modelling of soil moisture content for predictive irrigation scheduling. Sensors 18
with bulb onion cultivation and the soil and environmental character­
(10), 3408. https://doi.org/10.3390/s18103408.
istics of the study region. But the model is easy to implement with mixed Adnan, S., Khan, A., 2009. Effective rainfall for irrigated agriculture plains of Pakistan.
crops and other climatic conditions. For this purpose, soil sampling is Pak. J. Meteorol. 6 (11), 61–72.
Akhbari, M., Grigg, N., 2013. A framework for an agent-based model to manage water
very important for determining hydraulic parameters such as field ca­
resources conflicts. Water Resour. Manag. 27 (11), 4039–4052. https://doi.org/
pacity and permanent wilting point. Another principal technical aspect 10.1007/s11269-013-0394-0.
is the characterization of crops to understand their phenological needs Allen, R., Pereira, L., Raes, D., Smith, M., 1998. Crop evapotranspiration: guidelines for
and their irrigation requirements during the growing season. computing crop water requirements. In: United Nations FAO, Irrig. Drain.. FAO,
Rome, Italy.
For future work, the research aims to extend this model to several An, L., Linderman, M., Qi, J., Shortridge, A., Liu, J., 2005. Exploring complexity in a
irrigation units of UsoChicamocha. The calculation of water pre­ human–environment system: an agent-based spatial model for multidisciplinary and
scriptions and soil water stress values will be potentiated by the use of multiscale integration. Ann. Am. Assoc. Geogr. 95 (1), 54–79. https://doi.org/
10.1111/j.1467-8306.2005.00450.x.
crop models and the integration of this system with a Web application. Andales, A., Chávez J., Bauder T., Broner I., 2011. Irrigation scheduling: the water
Each agent can accurately formulate its crop irrigation prescriptions, but balance approach, Service 544 in action; no. 4.707.
in water scarcity cases it is quite possible that there is an excess of water Al-Kaisi, M.M., Broner, I., Andales, A.A. , 2009. Crop water use and growth stages, Fact
sheet (Colorado State University. Extension), Crop series; no. 4.715.
that can be optimized. Arétouyap, Z., Nouck, N., Nouayou, R., Méli’i, J., Kemgang Ghomsi, F., Piepi Toko, A.,
From the moment of transplantation in each field, daily water bal­ Asfahani, J., 2015. Influence of the variogram model on an interpolative survey
ance monitoring can be carried out in the IABM simulation, which al­ using Kriging Technique. J. Earth Sci. Clim. Change 6 (316), 2. https://doi.org/
10.4172/2157-7617.1000316.
lows monitoring the behavior of water requirements in each field of the
Barradas, J., Matula, S., Dolezal, F., 2012. Decision support system-fertigation simulator
irrigation district. The IABM allows the determination of an agent (DSS-FS) for design and optimization of sprinkler and drip irrigation systems.
satisfaction level based on the soil water stress coefficient, which is a Comput. Electron. Agric. 86, 111–119. https://doi.org/10.1016/j.
compag.2012.02.015.
performance indication in the water distribution of the studied fields, as
Belaqziz, S., Mangiarotti, S., Le Page, M., Khabba, S., Er-Raki, S., Agouti, T., Jarlan, L.,
it is associated with the permissible soil moisture limits so that crops do 2014. Irrigation scheduling of a classical gravity network based on the Covariance
not suffer damage. Matrix Adaptation–Evolutionary Strategy algorithm. Comput. Electron. Agric. 102,
64–72. https://doi.org/10.1016/j.compag.2014.01.006.
Cai, J., Xiong, H., 2017. An agent-based simulation of cooperation in the use of irrigation
Funding systems. Complex Adapt. Syst. Model. 5 (1), 9. https://doi.org/10.1186/s40294-
017-0047-x.
This research was funded by Minciencias and the Department of Caridad, R., Jury, M., 2013. Spatial and temporal analysis of climate change in
Hispañola. Theor. Appl. Clim. 113, 213–224. https://doi.org/10.1007/s00704-012-
Boyacá - Colombia through the Ph.D. scholarship program No. 733 – 0781-0.
2015. Thesis name: Intelligent Agent-Based model to support irrigation Cely, G., 2010. Determinación de parámetros de riego para el cultivo cebolla de bulbo en
management in agricultural crops – Cod: 46620 – Universidad Nacional el distrito de riego del alto chicamocha, Escuela de Posgrados, (Master Disertation).
de Colombia, Colombia. This research also was funded by the Uni­
〈https://repositorio.unal.edu.co/handle/unal/70468〉.
Cervenka, R., Trencansky, I., Calisti, M., 2005. Modeling social aspects of multi-agent
versidad de los Llanos (DGI) - Colombia, with the project: Agent-Based systems: the AML approach. In International Workshop on Agent-Oriented Software
Model (ABM) for the distributed allocation of resources in continuous Engineering. Springer, Berlin, Heidelberg, pp. 28–39.
Edwards, M., Ferrand, N., Goreaud, F., Huet, S., 2005. The relevance of aggregating a
dynamic systems, with application to irrigation distribution in agricul­
water consumption model cannot be disconnected from the choice of information
tural crops. available on the resource. Simul. Model Pract. Theory 13 (4), 287–307. https://doi.
org/10.1016/j.simpat.2004.11.008.

15
A.-F. Jiménez et al. Agricultural Water Management 256 (2021) 107088

FAO, 2020. Food and Agriculture Organization of the United Nations, Onion. (Last Matthews, R., 2006. The People and Landscape Model (PALM): towards full integration
accessed 02/08/2020). 〈http://www.fao.org/land-water/databases-and-software of human decision-making and biophysical simulation models. Ecol. Model. 194 (4),
/crop-information/onion/en/〉. 329–343. https://doi.org/10.1016/j.ecolmodel.2005.10.032.
Ghorbani, M.A., Shamshirband, S., Haghi, D.Z., Azani, A., Bonakdari, H., Ebtehaj, I., Mubarak, I., Hamdan, A., 2018. Onion crop response to different irrigation and N-
2017. Application of firefly algorithm-based support vector machines for prediction fertilizer levels in dry Mediterranean region. Adv. Hortic. Sci. 32 (4), 495–501.
of field capacity and permanent wilting point. Soil Tillage Res. 172, 32–38. https:// https://doi.org/10.13128/ahs-21934.
doi.org/10.1016/j.still.2017.04.009. Ojha, T., Misra, S., Singh, N., 2015. Wireless sensor networks for agriculture: The state-
Grashey-Jansen, S., 2014. Optimizing irrigation efficiency through the consideration of of-the-art in practice and future challenges. Comput. Electron. Agric. Vol. 118,
soil hydrological properties–examples and simulation approaches. Erdkunde 68, 66–84. https://doi.org/10.1016/j.compag.2015.08.011.
33–48. https://doi.org/10.2307/24365168. Rege, S., Gutiérrez, T., Marvuglia, A., Rege, E., Stilmant, D., 2015. Modelling price
Guyennon, N., Romano, E., Portoghese, I., 2016. Long-term climate sensitivity of an discovery in an agent based model for agriculture in Luxembourg. Comput. Econ.
integrated water supply system: the role of irrigation. Sci. Total. Environ. 565, Finance. Springer, Cham, pp. 91–112. https://doi.org/10.1007/978- 3-319-99624-0_
68–81. https://doi.org/10.1016/j.scitotenv.2016.04.157. 5.
Janssen, M.A., Baggio, J.A., 2016. Using agent-based models to compare behavioral Russell, S., Norvig, P., 2016. Artificial Intelligence: A Modern Approach. Pearson
theories on experimental data: application for irrigation games. J. Environ. Psychol. Education Limited, Malaysia.
46, 106–115. https://doi.org/10.1016/j.jenvp.2016.04.003. Sakellariou, I., Kefalas, P., Stamatopoulou, I., 2008. Enhancing NetLogo to simulate BDI
Jaxa-Rozen, M., Kwakkel, J.H., Bloemendal, M., 2019. A coupled simulation architecture communicating agents. Hellenic Conference on Artificial Intelligence. Springer,
for agent- based/geohydrological modelling with NetLogo and MODFLOW. Environ. Berlin, Heidelberg, pp. 263–275.
Modell. Softw. 115, 19–37. https://doi.org/10.1016/j.envsoft.2019.01.020. Spiliotopoulos, M., Loukas, A., 2019. Hybrid methodology for the estimation of crop
Jimenez, A.F., Cardenas, P.F., Canales, A., Jimenez, F., Portacio, A., 2020a. A survey on coefficients based on satellite imagery and ground-based measurements. Water 11
intelligent agents and multi-agents for irrigation scheduling. Comput. Electron. (7), 1364. https://doi.org/10.3390/w11071364.
Agric. 176, 105474 https://doi.org/10.1016/j.compag.2020.105474. Subedi, A., Chávez, J., 2015. Crop evapotranspiration (ET) estimation models: a review
Jimenez, A.F., Cardenas, P.F., Jimenez, F., Canales, A., López, A., 2020b. A cyber- and discussion of the applicability and limitations of ET methods. J. Agric. Sci. 7 (6),
physical intelligent agent for irrigation scheduling in horticultural crops. Comput. 50. https://doi.org/10.5539/jas.v7n6p50.
Electron. Agric. 178, 105777 https://doi.org/10.1016/j.compag.2020.105777. Talaviya, T., Shah, D., Patel, N., Yagnik, H., Shah, M., 2020. Implementation of artificial
Kaluzny, S., Vega, S., Cardoso, T., Shelly, A., 1998. Analyzing geostatistical data. S+ intelligence in agriculture for optimisation of irrigation and application of pesticides
SpatialStats. Springer, New York, NY. https://doi.org/10.1007/978-1-4615-7826-0_ and herbicides. Artif. Intell. Agric. https://doi.org/10.1016/j.aiia.2020.04.002.
4. Weather Spark. 2020. Average Weather in Nobsa Boyacá, Colombia. 〈https://weatherspa
Kokkotos, E., Zotos, A., Patakas, A., 2020. Evaluation of water stress coefficient Ks in rk.com/y/25267/Average-Weather-in-Nobsa-Colombia-Year-Round〉. (Last accessed
different olive orchards. Agronomy 10 (10), 1594. https://doi.org/10.3390/ 07/20/2020).
agronomy10101594. Wilensky, U., Rand, W., 2015. An Introduction to Agent-Based Modeling: Modeling
Leskovar, D.I., Agehara, S., Yoo, K., Pascual-Seva, N., 2012. Crop coefficient-based deficit Natural, Social, and Engineered Complex Systems with NetLogo. Mit Press.
irrigation and planting density for onion: growth, yield, and bulb quality. Zhao, T., Ding, R., Man, Z., 2011. Long-range monitoring system of irrigated area
HortScience 47 (1), 31–37. https://doi.org/10.21273/HORTSCI.47.1.31. wateruse based on Multi-Agent, in: 2011 International Conference on Mechatronic
Mango, N., Makate, C., Tamene, L., Mponela, P., Ndengu, G., 2018. Adoption of small- Science, Electric Engineering and Comput. (MEC). Jilin, China, IEEE, pp. 580–583.
scale irrigation farming as a climate-smart agriculture practice and its influence on 〈https://doi.org/10.1109/MEC.2011.6025531〉.
household income in the chinyanja triangle. South. Afr. Land 7 (2), 49. https://doi.
org/10.3390/land7020049.

16

You might also like