Modeling interconnectedness of people (social) and nature (ecological) can offer valuable understanding about the dynamics in the midst of social and environmental processes. Within this paper, a novel framework based on Multi-Agent System Simulation (MASS) is introduced that use python-based codes coupling NETLOGO agent-based model platform with a calibrated MODFLOW groundwater model that uses SWAT watershed model by considering the uncertainty associated with farmers' productivity. Then the various policy instruments implemented in the MASS (free-access and consistent tax and quota on groundwater utilization) are compared with the Centralized Optimal Model (COM). In the COM streamflow constraints are imposed by a central planner with impeccable foresight that use MATLAB-based codes coupling the MATLAB with a calibrated groundwater model of MODFLOW that uses SWAT watershed model. This comparison that is one of the most important goals of this paper is based on their environmental and economic impacts. The environmental and economic impacts are measured using two main indicators: the violation of streamflow (VSF) and the average annual benefit (AAB). The results indicate that simulation with agents that are more realist, heterogeneous, shortsighted, and self-interested agents (MASS) perform poorly under consistently applied policies in comparison with COM.

  • Improving agricultural water management requires modeling interconnected social-ecological systems.

  • A comprehensive physical model is essential for thoroughly examining the effectiveness of policy instruments.

  • Policy instrument effectiveness is notably influenced by variations in agent heterogeneity.

Abbreviation

Explanation

AAB

average annual benefit

ABM

agent-based modeling

CDF

cumulative distribution function

COM

centralized optimal model

CPR

common pool resource

HRU

hydrologic response unit

IWRM

integrated water resources management

km

kilometer

MASS

multi-agent system simulation

PSO

particle swarm optimization

SES

social-ecological system

VSF

violation of streamflow

The excessive use of groundwater is often believed to be caused by the failure of the common pool resource (CPR) and can lead to a variety of negative impacts, the decrease in the groundwater level, depletion of streamflow, escalated expenses related to pumping, deterioration of quality of water, damage to water-based ecosystems, and irreversible subsidence of land (Bartolino & Cunningham 2003). The excessive use of CPR and the resulting social and environmental effects can be attributed to the ascent in the global population, escalating levels of consumption, and the widespread use of sophisticated technologies (Grimble 1999; Dietz et al. 2003).

To enhance groundwater management, it is necessary to explicitly model water management systems including humans to uncover not intended and unforeseen feedback (Khan et al. 2017; Pande & Sivapalan 2017; di Baldassarre et al. 2019; Alam et al. 2022). Tools have been developed by integrating human activities with hydrological processes in models that aim to provide solutions for integrated water resources management (IWRM) (Mariño & Simonović 2001; Harou et al. 2009; Pulido-Velazquez et al. 2016) and analyzing water policies (Mulligan et al. 2014; Wu et al. 2015). When stakeholders' behavior is the focus, the approach of an agent-based modeling (ABM) is commonly used to model the consumption of domestic water within intricate water systems (Galán et al. 2009; Ma et al. 2012; Zhang et al. 2023a, 2023b), management of groundwater (Mulligan et al. 2014; Castilla-Rho et al. 2015; Ohab-Yazdi & Ahmadi 2018; Jabbari et al. 2022; Nouri et al. 2022; Aghazadeh et al. 2024; Canales et al. 2024), management of the allocation of water (Yang et al. 2009, 2012), incidents of water pollution (Zechman 2011; Shafiee & Zechman 2013), and urban flood management (Zhang et al. 2023a, 2023b; Nazemi et al. 2024). ABM is favored for its ability to explicitly depict the influencing components in the process of decision-making (Soman et al. 2008). In general, ABM is a modeling technique that simulates the behavior of autonomous and heterogeneous agents and their interactions and comprises four components: (1) agents, which are individual water users; (2) the environment, like a river basin or underground water reservoir, alongside the engineered infrastructure where in agents are situated; (3) management strategies or regulations; and (4) system-level performance metrics, encompassing economic impacts likewise environmental impacts (Zhao et al. 2013). ABMs are becoming more and more popular as a supplement to traditional analytical methods for studying environmental issues. They enable modeling the results at a systemic level that arise from the actions of individual existences. ABMs are particularly useful for studying social-ecological systems (SESs) because they provide a bottom-up perspective. SESs exhibit complex adaptive systems motivated by social and biophysical subsystems that interact with each other but are relatively distinct from each other (Ostrom 2009). ABM first emerged in fields other than water resources management (Schelling 1971; Holland & Sigmund 1995; Axelrod 1997). After that it was adopted for use in water resources management (Yang et al. 2009, 2012; Reeves & Zellner 2010; Holtz & Pahl-Wostl 2012).

To prevent the economic and social impacts that result from excessive use of groundwater, it is desirable to implement strategies for managing that promote the shift toward a more economically efficient utilization of groundwater. To tackle the issue, some of the approaches that can be adopted include implementing quotas, pricing groundwater, and establishing markets to attain effective allocation of groundwater based on economic considerations (Koundouri 2004). Every one of the mentioned policy instruments compels the water user to identify each element of the external consequences related to groundwater extraction. Although numerous research works have explored the effectiveness of varied policy instruments, only a limited number have undertaken this endeavor employing a physically heterogeneous illustration of a basin alongside a spatially disaggregated model (not a single-cell aquifer model) depicting farmers' water utilization behavior (Brozović et al. 2010). Models that consider spatial heterogeneity in groundwater extraction offer a more thorough comprehension of the economic aspects related to groundwater utilization (Mulligan et al. 2014). Quota and tax on water use policy instruments are widely employed to mitigate the poor allocation and excessive utilization of groundwater as a CPR (Koundouri 2004).

This paper investigates the different policy instruments employing a grounded hydrology numerical model that is completely distributed, linked to a distributed model of farmers' water usage behavior influenced by economic factors within an agent-based framework. The study takes into account the heterogeneity related to the productivity of farmers and the physical characteristics of the aquifer in its framework, which is a distinct advantage over the frequently utilized single-cell models in studies of this nature. This work is unique in that it employs a quasi-distributed SWAT watershed model and a numerical model of groundwater flow processes in conjunction with a model representing the dispersed behavior of farmers (agents) in utilizing water that is coupled with Python. These models, when used together, provide a more extensive and authentic illustration of groundwater utilization in a basin, allowing us to evaluate the efficacy of various policies for managing groundwater. This study evaluates three policy instruments: (1) free access with unrestricted behavior of farmers, (2) a consistent tax on water usage, which operates on incentive-based principles, and (3) a consistent quota on water usage, which lacks incentives unless trading is permitted. Every one of these policy instruments is incorporated into an ABM, that is a decentralized model of farmer water utilization behavior. In this paper, the agent-based model will be referred to as multi-agent system simulation (MASS) as it provides a better designation. MASS is used to model the potential behavior of farmers based on their financial factors and the groundwater conditions specific to the local area, as well as the interaction between them. Within MASS, every agent strives to maximize their profits without considering the impacts on profits of other agents, the watercourses, and the underground aquifer. This serves as a typical illustration of market failure in the realm of groundwater CPR, where users prioritize their individual profits over potential anticipated individual and external financial burdens, encompassing the effects on another agent and natural resources depletion. In this paper, an innovative framework utilizing MASS is presented. This framework employs Python-based scripts to integrate the NETLOGO ABM platform with a calibrated MODFLOW groundwater model, which incorporates the SWAT watershed model. One of the other unique aspects of this paper is that it considers the uncertainty associated with farmers' productivity.

The effectiveness of the policy instruments is assessed by comparing them to an ideal strategy for utilizing groundwater referred to as the centralized optimal model (COM) in this paper. The COM is a groundwater use strategy that employs a centralized planner's top-down method to maximize the total agents' profits (or net societal advantages) while adhering to the constraints of streamflow. This is in contrast to the MASS models, which do not have streamflow constraints. The COM is an illustration of potential possibilities accomplished by a central planner possessing complete foresight, allocating groundwater usage across both time and space in a manner that maximizes economic benefits and consistently adheres to streamflow constraints. It functions as a benchmark and ideal against which other measures (free access, tax, and quota MASS) can be evaluated. It represents the upper limit for groundwater management policies intended to sustain streamflow.

This paper is structured in five sections. The methodology is described in Section two, which includes the hydrologic model, the COM and the MASS that is a model of farmers' water usage behavior. Section three provides a detailed presentation of the results. A discussion on these findings is presented in Section four. Finally, Section five provides a summary of the entire study and its significance in the context of groundwater management.

Case study

The Ardabil plain aquifer (Figure 1) is situated in the northwestern region of Iran, in the Ardabil province. The plain is located between 38,000–38,300 N and 48,000–48,400 E and spans an area of approximately 990 km². The area undergoes enjoyable summers and comparatively extended winters, experiencing approximately 300 mm of average annual precipitation. The plain is encircled by heights that form a section of the Alborz Mountains. In the western part of the plain, there are exposed outcrops of conglomerate along with volcanic ashes, tuff, and lahars. These rocks originate from the Sabalan Mountain and affect aquifer recharge due to abundant springs. In the southwestern Ardabil plain, there are rock formations composed of conglomerate, along with sandstone, marl, freshwater pumice, tuff, limestone, and lahar, dating back to the Neogene era. These formations exert minimal impact on the recharge of the aquifer. In the northeastern part, there are rock units such as cherty limestone, cherty dolomite, sandstone, and conglomerate, with thin layers of gypsum, while in the southern, eastern, and northern regions of the plain, there are trachy basalt, megaporphyric trachy andesite, olivine basalt, volcanic breccias, tuff, and sandy tuff. These formations exhibit a moderate potential in terms of aquifer recharge. The Ardabil plain took shape as a result of Quaternary alluvial deposits formed through the erosion of the adjacent mountains. Through geophysical examinations of drilling records and pumping test data, it has been determined that the aquifer reaches a greatest thickness of approximately 220 m and primarily consists of sand, gravel, and a minor proportion of clay. The aquifer exhibits a transmissivity range of 50–2,200 m²/day, with specific yield values spanning a diverse range of 0.021–0.14. Groundwater flow generally moves from diverse directions and converges toward the northwestern area of the plain (Kord et al. 2013).
Figure 1

Location of Ardabil plain in Iran.

Figure 1

Location of Ardabil plain in Iran.

Close modal

Modeling framework

This paper compares the economic and ecological (environmental) impacts of various policy instruments using a MASS (as a decentralized model) and a COM (as a centralized model). Figure 2 shows the modeling framework adopted. At first, in order to create a hydrologic simulation model, coupled of two softwares (SWAT-MODFLOW) are established. Then to create COM, coupled of two softwares (MODFLOW-MATLAB) and to create MASS, coupled of two softwares (MODFLOW-NETLOGO) are used. Policy instruments used in MASS include water use free access, taxes, and quotas. The water use tax and quota are consistently imposed on all agents. The MASS and COM (decision models) are linked to the groundwater model using pumping rates, influencing the groundwater model and leading to revised groundwater levels and streamflow, thereby influencing subsequent decisions related to pumping.
Figure 2

Modeling framework adopted.

Figure 2

Modeling framework adopted.

Close modal

Hydrologic simulation model

In order to simulate Ardabil plain aquifer, a couple of two models are set up (Figure 2). This necessitates the effective incorporation of operational models that not only accurately depict all pertinent processes within the hydrological system in a physically purposeful manner but are also sufficiently straightforward to facilitate extensive basin-wide applications. Understanding the regional component of the hydrological cycle is crucial for evaluating how land use and climate influence the temporal and spatial patterns of the recharge of groundwater. The SWAT model is opted for this particular objective. Although the quasi-distributed SWAT model can accurately simulate the spatiotemporal distribution of the recharge rates of groundwater (at HRUs), its groundwater module is consolidated. As a result, groundwater characteristics spread across an area (like storage coefficients and hydraulic conductivities) cannot be depicted, and the method is considerably restricted in portraying the spatial variations of the levels of groundwater and the dynamics of groundwater flow. The offered modeling framework combines the MODFLOW fully distributed groundwater model (McDonald & Harbaugh 1988) with the SWAT watershed model. In this method, the outputs of the SWAT model serve as inputs for MODFLOW. The model employs monthly stress periods and comprises a single layer with 504 grid cells, each measuring 1.75 km × 1.75 km in size. MODFLOW at every time step computes the hydraulic head for every grid cell by employing a mathematical representation of groundwater flow using a finite difference method. The equation considers hydraulic parameters (like specific yield, transmissivity, and hydraulic conductivity) that vary spatially, along with spatially variable boundary conditions and other spatially variable volumetric fluxes (like drains, recharge, and pumping wells). The calibration of the model is based on historical elevations of the water table at various points across the plain and is done with PEST software. The average recharge for each grid cell is determined using SWAT (Kamali & Niksokhan 2017a, 2017b; Norouzi Khatiri et al. 2020). Data sources used in the model setup such as the sources of hydrogeological parameters are received from the Ardabil Regional Water Company. Figure 3 shows the transmissivity and specific yield maps of aquifer.
Figure 3

The transmissivity and specific yield maps of aquifer.

Figure 3

The transmissivity and specific yield maps of aquifer.

Close modal

Centralized optimal model (COM)

To create COM, a coupled of two models (MODFLOW-MATLAB) is used (Figure 2; Kamali & Niksokhan 2017a, 2017b). The COM is solved using the PSO algorithm in MATLAB. The COM is designed to maximize the total profits of agents across the basin while adhering to streamflow constraints and higher limits on extraction during the period of simulation. A unified objective function encompasses decision variables for every agent. The COM's pumping decisions are made with the consideration of every conceivable situation and choose the most advantageous (optimal) pumping allocation. The management periods of the optimization model are 1 year, yielding a total of 810 (27 agents, 3 crops, and 10 years) decision variables for this model. The formulation is presented in the following equation.

Objective function:
(1)
Subject to:
(2)
(3)
Qa,c,s is the decision variable of flow rate (L3/T) for the site(s) of well a during the season of pumping s implemented to crop c; N is the overall number of crops; G is the overall number of agents; E is the overall number of seasons of pumping; pc is the price at which crop c is marketed; ya,c is the crop yield (in tons per acre) for agent a; D is the period of pumping throughout the season of growing (T); wa,c is the volume of water used per unit area (L) for each agent a and crop c; havg is the mean overall lifting distance (L) for all agents at the start of the simulation model; is the density of water (F/L3); pe is the price of electricity (Rial/(P-T)); e is the pumping efficiency; fc is the operational cost of farming (Rial/L2) for crop c; Qu,a is the upper bound for the decision variable of the flow rate; Sb,t is the streamflow at the ending of stress period t for stream cell b (MODFLOW employs monthly stress periods, distinct from the season of pumping index s); and is the streamflow lower bound.

Multi-agent system simulation (MASS)

To create MASS, coupled of two models MODFLOW-NETLOGO are used (Figure 2). NETLOGO is a programming language designed for multi-agent systems for designing, implementing, and analyzing ABMs that has evolved into a popular choice for this goal because of its engaged user community and ease of use. It was authored by Uri Wilensky in 1999 and is open-source. This platform is ideal for modeling complex systems that evolve across the duration. It allows modelers to guide hundreds or thousands of autonomous ‘agents’ carrying out simultaneously. This feature enables investigating the links between the individual behaviors at the micro-level and the overall patterns that arise from their interactions at the macro-level. The tool is mainly coded in Scala and Java, and it provides a variety of methods and functions that facilitate the quick creation of ABMs with spatially clear features.

Various investigations have concentrated on integrating models of ecological (environmental) systems with MASS (e.g. Bithell & Brasington 2009; Reeves & Zellner 2010; Kelly et al. 2013). The coupling of MASS and physical simulations is particularly pertinent in the context of groundwater resources. Given their widespread exploitation, limited availability, and the intricate interactions among heterogeneous users involved in their management, MASS emerges as a suitable modeling choice. Moreover, monitoring groundwater resources in the subsurface is frequently challenging. Hence, numerical modeling can offer valuable insights for effective resource management.

However, the integration of MASS and models of groundwater might pose specific dilemmas attributed to the extended temporal scales and typically lengthy duration of model execution. Castilla-Rho et al. (2015) outline four considerations associated with traditional methods for modeling interconnected groundwater systems: constraints posed by basic aggregated models, technological intricacies arising from the utilization of interconnected software applications and libraries for data exchange, a deficiency in scenario development flexibility, and the lack of practicality analyzing the sensitivity of various models. To overcome the mentioned constraints, they present an engaging environment that straightly incorporates equations of groundwater flow into the widely used NETLOGO agent-based platform. Considering the continuing pattern toward more intricate human–environmental systems, ABMs (Sun et al. 2016), that consequently pose greater challenges in design, this approach offers the benefit of creating an environment designed for user convenience for the development and application of ABMs of groundwater management. However, the use of NETLOGO has constraints in the context of groundwater management issues in cases where existing geohydrological models are accessible for direct reuse or where detailed geohydrological modeling is necessary.

To address this void, the paper presents a basic interconnected simulation framework that facilitates the integration of the NETLOGO platform with MODFLOW utilizing the Python programming language. This method preserves NETLOGO's ease of use while enabling users to take into account intricate hydrological processes or establish a direct interface between the ABM and pre-existing MODFLOW. This method allows users to create MASS using the user-friendly NETLOGO platform while taking advantage of the complete functionalities offered by MODFLOW. The main core and important part of the codes used in NETLOGO are shown in Figure 4.
Figure 4

The main core and important part of codes used in NETLOGO.

Figure 4

The main core and important part of codes used in NETLOGO.

Close modal
In this analysis, a group of farmers in a sub-basin is represented by an agent. The choices made by the agent encompass crop selection, water usage, and in an indirect manner, cultivated land with irrigation. To enable comparison, similar agents are employed in both the COM and MASS. The number of agents is 27 that serve as the variable for decision-making. Every agent represents a group of farmers responsible for managing the pumping activities at one or more of 240 cells within the Ardabil plain aquifer simulation model. Figure 5 displays the positions of the agents and their numbers, while Table 1 shows the number of agents and the area under cultivation. This approach is taken in order to (1) shorten the computational length of every model iteration and (2) more accurately depict the variability in farm that is often observed in real world (Schaible 2004; Hoppe 2010).
Table 1

Agent's number and area under cultivation

AgentArea under cultivation (ha)AgentArea under cultivation (ha)AgentArea under cultivation (ha)
4,270 10 1,843 19 1,664 
3,912 11 856 20 2,412 
1,716 12 1,479 21 1,563 
2,788 13 1,327 22 1,353 
1,177 14 1,738 23 1,840 
959 15 1,325 24 1,732 
646 16 2,938 25 2,013 
1,338 17 1,435 26 2,035 
2,241 18 908 27 1,959 
AgentArea under cultivation (ha)AgentArea under cultivation (ha)AgentArea under cultivation (ha)
4,270 10 1,843 19 1,664 
3,912 11 856 20 2,412 
1,716 12 1,479 21 1,563 
2,788 13 1,327 22 1,353 
1,177 14 1,738 23 1,840 
959 15 1,325 24 1,732 
646 16 2,938 25 2,013 
1,338 17 1,435 26 2,035 
2,241 18 908 27 1,959 
Figure 5

Agent's number and position.

Figure 5

Agent's number and position.

Close modal

In the MASS and COM, the agents are presented with a selection of three crops: wheat, barley, and potato. These three crops account for a significant portion of agricultural output in the Ardabil Plain, although alfalfa and similar crops are widely favored in that particular area. The financial information regarding costs and revenues is extracted from the local investigation. The operational cost of the farm, denoted as fc, corresponds to the cost of a particular crop, expressed in million rials per acre for every season of cultivation. For wheat, barley, and potato, fc is 150,000,000, 80,000,000, and 1,000,000,000 rials per acre. pc is the crop selling price in rials per kilogram. Regarding wheat, barley, and potato, pc is 125,000, 120,000, and 150,000 rials per kilogram in 1401. The operational expenses and selling price remain consistent throughout the simulated modeling duration. This data is obtained from local investigation from local farmers.

Two parameters that impact the productivity of the agents encompass crop yield (ya,c) and crop irrigation requirements (wa,c). To prevent the uncertainty associated with the productivity parameters from unduly affecting the model results, a total of 10 groups of parameter measurements have been established. Ten runs are conducted for the MASS and COM model types using varied groups of agent productivity parameters. In order to create a basin setting that is more realistic and captures a spectrum of productivity of agent productivity, spanning from highly efficient agents to less efficient ones, the parameters for each agent within every set are subjected to randomization. The aim is to depict in a simplified manner the common variability among agents resulting from experience, embracing technology, and other skills or environmental circumstances. The value of each parameter remains unchanged over the period of the simulation. Due to the prolonged computational processing time and significant storage needs, the decision was made to create only 10 sets of productivity parameters. An overview of the agent parameters within each set is presented in Table 2.

Table 2

Parameters related to agent productivity

Parameter setCrop yield (ton/ac)
Crop irrigation requirement (mm)
Wheat
Barley
Potato
Wheat
Barley
Potato
Avg.Std. dev.Avg.Std. dev.Avg.Std. dev.Avg.Std. dev.Avg.Std. dev.Avg.Std. dev.
33 250 62 195 56 460 85 
33 250 63 195 57 460 86 
6.5 4.5 34 255 62 200 56 470 85 
6.5 4.5 34 255 63 200 57 470 86 
35 260 62 205 56 480 85 
35 260 63 205 57 480 86 
7.5 5.5 36 265 62 210 56 490 85 
7.5 5.5 36 265 63 210 57 490 86 
37 270 62 215 56 500 85 
10 37 270 63 215 57 500 86 
Historical 35 260 62 205 56 480 85 
Parameter setCrop yield (ton/ac)
Crop irrigation requirement (mm)
Wheat
Barley
Potato
Wheat
Barley
Potato
Avg.Std. dev.Avg.Std. dev.Avg.Std. dev.Avg.Std. dev.Avg.Std. dev.Avg.Std. dev.
33 250 62 195 56 460 85 
33 250 63 195 57 460 86 
6.5 4.5 34 255 62 200 56 470 85 
6.5 4.5 34 255 63 200 57 470 86 
35 260 62 205 56 480 85 
35 260 63 205 57 480 86 
7.5 5.5 36 265 62 210 56 490 85 
7.5 5.5 36 265 63 210 57 490 86 
37 270 62 215 56 500 85 
10 37 270 63 215 57 500 86 
Historical 35 260 62 205 56 480 85 

Crop yield (ya,c) for every agent is extracted from the local investigation from local farmers. The crop data provides the mean yield in each year for every part, specified as the overall production for each crop in tons per acre. The values of ya,c are given to each agent through an initial development process an empirical CDF (cumulative distribution function) that is divided into linear parts and adjusted to the crop yield data across time and space, establishing a correlation between yield and relative frequency. Each agent is subsequently allocated the relative frequency values by choosing random values within the range of 0–1 of a uniform distribution. The agents' assigned relative frequency values are employed to select the corresponding yield value through the inverse of the empirical CDF.

Crop irrigation requirements (wa,c) of every agent are extracted from the local investigation from local farmers. The local investigation of the agent's crop yield values and crop irrigation requirements values and normal distribution are depicted in Figure 6. Table 3 shows crop irrigations, harvest, and implant months. A dash in Table 3 represents the months of irrigation.
Table 3

Crop irrigations, harvest, and planting months

CropPlantingHarvestIrrigation month
123456789101112
Wheat October August    – – –    –   
Barley October July    – –     –   
Potato March September     – – – –     
CropPlantingHarvestIrrigation month
123456789101112
Wheat October August    – – –    –   
Barley October July    – –     –   
Potato March September     – – – –     
Figure 6

Histograms and normal distribution of historical crop yield and irrigation requirements.

Figure 6

Histograms and normal distribution of historical crop yield and irrigation requirements.

Close modal

A summary of each modeling framework, MASS and COM, is presented in Table 4. The overall cost for every cultivation season for every agent is calculated by adding the operational costs, costs associated with pumping, and any applicable taxes. The pumping cost is determined by the groundwater depth, the water weight, the price of electricity, and the quantity of water pumped. The income generated by the calculation of every agent involves multiplying the price (at which a product is sold, pc) with the total production for the cultivation season. The cumulative output during the cultivation season depends on two parameters of productivity ya,c and wa,c and the amount of pumped water. By deducting the overall cost from the overall revenue, the overall profit of each agent is calculated. The tax payment is regarded as a positive contribution to the administration and is manifested in the ultimate analysis by adding up the overall profits of the agent and the overall taxes remitted to assess the monetary outcome in each model iteration, or the overall gains of society.

Table 4

Summary of formulation of management

MASSCOM
Objective function Decentralized – Implemented for each year on each agent Centralized – Implemented for all years on all agents 
Optimize (maximize) Profit of single agent Total of profits of agents 
Optimization management period 1 year (repeated for 10 years) 10 years 
Environment simulation software SWAT-MODFLOW SWAT-MODFLOW 
Used software NETLOGO MATLAB 
Optimization algorithm PSO PSO 
Decision variables 3 (3 crops per agent, 1 year, 1 agent) 810 (3 crops per agent, 10 years, 27 agents) 
First constraint Upper bound of pumping (Qa,sUpper bound of pumping (Qa,s
Second constraint Water usage quota Lower bound of streamflow 
MASSCOM
Objective function Decentralized – Implemented for each year on each agent Centralized – Implemented for all years on all agents 
Optimize (maximize) Profit of single agent Total of profits of agents 
Optimization management period 1 year (repeated for 10 years) 10 years 
Environment simulation software SWAT-MODFLOW SWAT-MODFLOW 
Used software NETLOGO MATLAB 
Optimization algorithm PSO PSO 
Decision variables 3 (3 crops per agent, 1 year, 1 agent) 810 (3 crops per agent, 10 years, 27 agents) 
First constraint Upper bound of pumping (Qa,sUpper bound of pumping (Qa,s
Second constraint Water usage quota Lower bound of streamflow 

The function describing the production supposes that the agent will modify the count of cultivated acres with irrigation based on the water pumping decision. To illustrate this point, suppose an agent has 1,000 acres of land area and wheat requires 300 mm of water and because of restrictions the agent reaches a decision to use 150 mm throughout the season of cultivation. The model proceeds under the assumption that the agent will implement reducing the 150 mm of water to half (150/300) for 500 acres so that 500 acres will receive 300 mm of water for wheat irrigation, while the remaining 500 acres will not be irrigated.

MASS aims to maximize the profit of a single agent. The agent's profits are optimized while adhering to pumping upper bounds (dictated by available crop irrigation requirements and land area) and, if applicable, water usage quotas. For this model, an objective function with a decentralized structure is employed for each agent at the beginning of every season of cultivation. At the beginning of each season of growth, the groundwater model computes new groundwater depth values, relying on the previous year's groundwater utilization. Thus, the choices made by the agents regarding water usage encompass their influence on alterations in the groundwater level. The formulation for management, applicable to each agent and each year, is presented in the equation for the water utilization behavior model of the farmer below.

Objective function:
(4)
Subject to:
(5)
(6)
where
(7)
is the agent's maximum available land area (L2); quota is the water use quota applied to all agents in volume per area (L); is the overall lifting distance (L) at the start of the season of pumping s for agent a; and TAX is the tax in rials for each unit of water utilized (Rial/L3).

The pumping decision variable is established annually for each season of cultivation. This means that the decision on how much to pump is made once a year, and it applies to the entire cultivation season. The cost of pumping for each year is determined by the initial groundwater depth at the beginning of the cultivation season. It is assumed that the loss of head because of pipe friction is insignificant and the efficiency of pumping is 70%. It is taken for granted that the return flows applicable to both COM and MASS model types constitute 20% of the water volume extracted. After each agent's yearly optimization, the storage capacity is examined to ensure sufficient available water for supplying the flow rate set by the optimization model.

In the event that there exists an inadequate water supply in the proximity of their well, agents refrain from pumping in that particular cultivation season. Equation (4) shows that fixed costs are anticipated to be considerably less than variable costs. Therefore, we did not include any fixed costs in this analysis.

Policy instruments

The MASS formulation includes water use quotas and taxes to assess alternative, practical policy instruments. The water usage tax and quota are consistently imposed on all agents. The tax is incorporated into the objective function as a levy on complete water utilization, calculated in million rials per m3, falling within the range of 1–11 million rials per m3. Due to the significant differences in the productivity of agents, some agents are more likely to be affected by a tax increase than others. Agents with lower productivity are anticipated to decrease their rate of pumping or stop pumping at a reduced tax rate compared with agents with higher productivity. The tax rates employed in this examination are elevated and are probably not feasible from a political standpoint. The reason for the high rates of tax applied in this study is that the profits from the crops significantly surpass the total costs in the solution of free-access MASS. Therefore, to encourage agents to irrigate a smaller area and diminish water consumption, a high tax rate is required. This is a fact that needs to be acknowledged if the strategy to curtail water usage and, consequently, influence the streams relies solely on a uniform tax. If the revenue of tax is redistributed among the agents in a quantity that is approximately equivalent to the loss of welfare resulting from the tax, then a consistent tax could become a more feasible alternative.

The utilization of water quota is added to the framework for optimizing water usage in the model as limitations on the amount of water applied, which can range from 100 to 600 mm per season of cultivating. The depth is determined by calculating the entire volume of water pumped throughout the season of cultivation and dividing it (measured in cubic meters) by the area of land that can be irrigated (measured in acres) and then expressing the result to millimeters.

The water usage quota imposes a consistent decrease in water usage for each agent, while considering the quantity of land owned by each agent. The number of irrigated acres owned by each agent stays consistent throughout the planning period.

Various policy instruments were assessed based on their impacts on the economic and environmental consequences, then were contrasted to the COM and the free-access MASS. The environmental and economic impacts of a management plan are measured using two main indicators: the violation of streamflow and the average annual benefit (AAB). The violation of streamflow is the proportion of instances (percentage) where the simulated streamflow falls below the specified streamflow objectives that match the streamflow limitation values implemented in the COM. The AAB is calculated by adding the average annual profit across the entire basin and, if relevant, the average annual tax revenue generated across the entire basin.

The economic viability and environmental impact of the COM and MASS (with free access, the specific tax, and quotas) are shown in Figure 7. Each data point in the figure corresponds to the average across all iterations of the model (10 times running) conducted for every scenario that employs all variations of agent productivity parameters, and the average outcome is displayed. This figure illustrates the tradeoff between different types of models and compares the policy instruments that are utilized within the COM and MASS.
Figure 7

The average annual benefit versus the violation of streamflow for the average result of 10 runs. The horizontal axis values for COM and MASS under different conditions (free access, quota, tax (a), and tax (b)) indicate the average annual benefit of the agents. MASS with tax (a) represents the average annual benefit of the agents plus the average annual tax revenue and tax (b) represents the average annual benefit of the agents. Each data point shows the average outcome derived from 10 distinct agent productivity parameter sets. The data points for MASS with quota and MASS with tax reflect various quota or tax values, resulting in multiple points for each model type. Lines connect these data points in ascending or descending order based on the quota or tax values.

Figure 7

The average annual benefit versus the violation of streamflow for the average result of 10 runs. The horizontal axis values for COM and MASS under different conditions (free access, quota, tax (a), and tax (b)) indicate the average annual benefit of the agents. MASS with tax (a) represents the average annual benefit of the agents plus the average annual tax revenue and tax (b) represents the average annual benefit of the agents. Each data point shows the average outcome derived from 10 distinct agent productivity parameter sets. The data points for MASS with quota and MASS with tax reflect various quota or tax values, resulting in multiple points for each model type. Lines connect these data points in ascending or descending order based on the quota or tax values.

Close modal

According to Figure 7, MASS with free access (green square) yields the highest AAB for agents (around 151 trillion rials). However, it has a significant impact on streamflow, with approximately 90% of the violation of streamflow. This outcome mentions that the approach of free access for self-interested and shortsighted agents lacks sustainability in streamflow. The groundwater depletion reserves have resulted in a considerable rise in the violation of streamflow, and the rising costs associated with pumping over a duration have not prevented this depletion. According to this figure, COM (red square) yields around 120.3 trillion rials the average annual benefit for agents. However, it has no impact on streamflow, with 0% of the violation of streamflow. COM suggests that if there is a complete understanding and centralized management of all pumping activities, the limitation on streamflow can achieve contentment, and comparatively great benefits can be obtained (while not reaching the same level as those observed in the free-access approach). Although it is probably impractical, the optimal solution sets the highest attainable outcome that could be accomplished via a policy related to groundwater customized to mitigate the effects on streamflow.

The consistent tax and consistent quota policy instruments demonstrate that it is challenging to achieve the optimal solution due to the limitations of incomplete foresight and the basin heterogeneity. In both cases, when a tax or quota is put into effect, there are significant reductions in average annual benefit or a high violation of streamflow. Neither policy instrument implemented in the MASS can match the performance of an optimal pumping plan that delivers both high economic feasibility (around 120.3 trillion rials) and low streamflow impacts (0% violation of streamflow). In the case of the MASS with quota and the MASS with tax, almost every agent is required to cease pumping to achieve zero violation of streamflow, resulting in extremely low average annual benefit. As it is challenging to minimize the violation of streamflow without significant economic consequences, it seems that a non-zero violation of streamflow must be accepted. The results suggest that the violation of streamflow 20% can be attained with acceptable economic advantages using the MASS with tax and likewise MASS with quota policy instruments. The yellow and orange lines illustrated the results of MASS with tax outcomes, signifying a spectrum influenced by the extent of tax revenue allocation for the advantage of the agents. The orange line in the figure represents the scenario where all revenue from taxes is utilized to the advantage of the agents, while the yellow line demonstrates the scenario in which none of the agents derive benefits from the utilization of tax revenue. Values on the horizontal axis of the orange line represents the average annual benefit of the agents plus the average annual tax revenue and the yellow line represent the average annual benefit of the agents. If the distribution of tax earnings is not efficient (yellow line), the advantages of the agents are substantially lower than those of the optimal solutions and free access. If the allocation of tax revenue is redistributed (orange line), the agents experience notable benefits even in situations where the violation of streamflow is low. In both scenarios, the agents show a minimal reaction in pumping behavior to reduced rates of tax (points on each line located in the figure's upper-rightmost section).

The red line in the figure represents the MASS with quota solutions, which falls between the yellow and orange lines linked to the tax policy instruments. Therefore, the effectiveness of the tax is heavily reliant on the revenue of tax redistribution policy. The proximity of the line of quota to the orange line suggests that the quota may be more effective than most potential redistribution policies.

Figure 8 shows the outcomes of 10 separate runs, each using a different set of agent productivity parameters (refer to Table 5). For each agent, every parameter set has different crop irrigation requirements and crop yield. In other word, in this figure, a sensitivity plot is presented. These parameters are adjusted to account for the uncertainty of the productivity of agents, encompassing fluctuations in the condition of soils, farmer skill, and crop fertilizers. This figure confirms that the overall patterns per every policy depicted in Figure 7 persist. The dashed black lines signify that variations in productivity parameters do not significantly impact the solutions of MASS with quotas.
Table 5

Model results by type and parameter set of productivity

Parameter setMetricCOMMASS
Free accessCap (mm)
 Tax (a) (million rials per m3)
 Tax (b) (million rials per m3)
60050040030020010013579111357911
AAB 115.3 146.9 145.3 138.5 130.6 123.2 112.6 43.9 146.8 143.6 141.6 130.4 114.5 69.8 146.8 99.1 62.3 26.4 22.2 19.9 
VSF 89 88 79 69 58 39 89 87 82 58 28 89 87 82 58 28 
AAB 102.5 135.3 133.9 128.7 120.3 113.1 104.0 39.8 135.2 133.3 131.5 118.9 99.1 61.1 135.2 88.5 52.5 14.9 12.5 11.5 
VSF 85 85 77 66 55 37 85 83 79 52 18 85 83 79 52 18 
AAB 129.8 158.3 156.4 150.1 140.8 133.6 119.7 47.1 158.1 153.3 150.3 138.1 121.7 72.1 158.2 108.4 71.2 34.2 26.2 22.5 
VSF 92 91 82 71 61 40 92 90 85 61 32 92 90 85 61 32 
AAB 136.3 165.2 163.4 156.0 146.4 139.0 124.5 48.8 165.0 162.2 159.9 150.8 134.5 79.9 165.0 117.7 80.6 46.9 36.4 29.5 
VSF 94 92 83 72 62 41 94 91 86 64 36 94 91 86 64 36 
AAB 131.8 164.4 161.0 153.6 144.9 137.7 123.5 48.6 164.1 161.4 159.0 147.2 133.5 78.2 164.1 116.7 79.7 43.2 33.7 28.5 
VSF 94 91 82 72 62 41 94 92 88 62 37 94 92 88 62 37 
AAB 103.6 132.3 128.9 123.7 115.3 108.1 98.9 35.2 131.3 129.6 125.8 116.1 89.1 53.7 131.3 85.2 46.3 12.3 7.5 3.9 
VSF 84 84 76 65 54 36 84 83 80 52 12 84 83 80 52 12 
AAB 109.5 138.8 136.5 130.7 122.2 115.9 105.8 41.6 138.7 134.8 131.4 117.5 101.5 65.2 138.7 90.4 52.4 23.8 18.4 15.3 
VSF 86 85 77 66 56 37 86 85 81 49 19 86 85 81 49 19 
AAB 137.6 168.5 166.9 159.6 150.3 142.3 128.3 53.7 168.4 164.5 161.2 146.1 122.4 86.8 168.4 119.6 82.3 42.1 38.4 36.7 
VSF 95 94 85 74 63 42 95 92 87 58 21 95 92 87 58 21 
AAB 122.4 154.7 149.8 142.7 134.6 126.9 114.8 45.1 154.6 153.0 151.6 136.9 118.6 71.7 154.6 108.1 72.3 33.2 25.3 21.5 
VSF 91 90 81 71 60 40 91 89 85 55 26 91 89 85 55 26 
10 AAB 114.7 146.0 143.5 136.5 128.6 121.2 110.6 43.6 146.0 143.0 141.1 132.5 113.5 65.8 146.0 98.6 61.9 28.4 20.2 16.5 
VSF 88 87 78 68 57 38 88 87 85 62 29 88 87 85 62 29 
Average AAB 120.3 151.0 148.6 142.0 133.4 126.1 114.3 44.7 150.8 147.9 145.3 133.5 114.8 70.4 150.8 103.2 66.2 30.5 24.1 20.6 
VSF 90 89 80 69 59 39 90 88 84 57 26 90 88 84 57 26 
Parameter setMetricCOMMASS
Free accessCap (mm)
 Tax (a) (million rials per m3)
 Tax (b) (million rials per m3)
60050040030020010013579111357911
AAB 115.3 146.9 145.3 138.5 130.6 123.2 112.6 43.9 146.8 143.6 141.6 130.4 114.5 69.8 146.8 99.1 62.3 26.4 22.2 19.9 
VSF 89 88 79 69 58 39 89 87 82 58 28 89 87 82 58 28 
AAB 102.5 135.3 133.9 128.7 120.3 113.1 104.0 39.8 135.2 133.3 131.5 118.9 99.1 61.1 135.2 88.5 52.5 14.9 12.5 11.5 
VSF 85 85 77 66 55 37 85 83 79 52 18 85 83 79 52 18 
AAB 129.8 158.3 156.4 150.1 140.8 133.6 119.7 47.1 158.1 153.3 150.3 138.1 121.7 72.1 158.2 108.4 71.2 34.2 26.2 22.5 
VSF 92 91 82 71 61 40 92 90 85 61 32 92 90 85 61 32 
AAB 136.3 165.2 163.4 156.0 146.4 139.0 124.5 48.8 165.0 162.2 159.9 150.8 134.5 79.9 165.0 117.7 80.6 46.9 36.4 29.5 
VSF 94 92 83 72 62 41 94 91 86 64 36 94 91 86 64 36 
AAB 131.8 164.4 161.0 153.6 144.9 137.7 123.5 48.6 164.1 161.4 159.0 147.2 133.5 78.2 164.1 116.7 79.7 43.2 33.7 28.5 
VSF 94 91 82 72 62 41 94 92 88 62 37 94 92 88 62 37 
AAB 103.6 132.3 128.9 123.7 115.3 108.1 98.9 35.2 131.3 129.6 125.8 116.1 89.1 53.7 131.3 85.2 46.3 12.3 7.5 3.9 
VSF 84 84 76 65 54 36 84 83 80 52 12 84 83 80 52 12 
AAB 109.5 138.8 136.5 130.7 122.2 115.9 105.8 41.6 138.7 134.8 131.4 117.5 101.5 65.2 138.7 90.4 52.4 23.8 18.4 15.3 
VSF 86 85 77 66 56 37 86 85 81 49 19 86 85 81 49 19 
AAB 137.6 168.5 166.9 159.6 150.3 142.3 128.3 53.7 168.4 164.5 161.2 146.1 122.4 86.8 168.4 119.6 82.3 42.1 38.4 36.7 
VSF 95 94 85 74 63 42 95 92 87 58 21 95 92 87 58 21 
AAB 122.4 154.7 149.8 142.7 134.6 126.9 114.8 45.1 154.6 153.0 151.6 136.9 118.6 71.7 154.6 108.1 72.3 33.2 25.3 21.5 
VSF 91 90 81 71 60 40 91 89 85 55 26 91 89 85 55 26 
10 AAB 114.7 146.0 143.5 136.5 128.6 121.2 110.6 43.6 146.0 143.0 141.1 132.5 113.5 65.8 146.0 98.6 61.9 28.4 20.2 16.5 
VSF 88 87 78 68 57 38 88 87 85 62 29 88 87 85 62 29 
Average AAB 120.3 151.0 148.6 142.0 133.4 126.1 114.3 44.7 150.8 147.9 145.3 133.5 114.8 70.4 150.8 103.2 66.2 30.5 24.1 20.6 
VSF 90 89 80 69 59 39 90 88 84 57 26 90 88 84 57 26 

AAB, the average annual benefit; VSF, the violation of streamflow.

Figure 8

The average annual benefit versus the violation of streamflow. The horizontal axis values for COM and MASS under different conditions (free access, quota, tax (a), and tax (b)) indicate the average annual benefit of the agents. MASS with tax (a) represents the average annual benefit of the agents plus the average annual tax revenue and tax (b) represents the average annual benefit of the agents. The data points for MASS with quota and MASS with tax reflect various quota or tax values, resulting in multiple points for each model type. Lines connect these data points in ascending or descending order based on the quota or tax values. Each model type was executed 10 times using 10 different agent productivity parameter sets.

Figure 8

The average annual benefit versus the violation of streamflow. The horizontal axis values for COM and MASS under different conditions (free access, quota, tax (a), and tax (b)) indicate the average annual benefit of the agents. MASS with tax (a) represents the average annual benefit of the agents plus the average annual tax revenue and tax (b) represents the average annual benefit of the agents. The data points for MASS with quota and MASS with tax reflect various quota or tax values, resulting in multiple points for each model type. Lines connect these data points in ascending or descending order based on the quota or tax values. Each model type was executed 10 times using 10 different agent productivity parameter sets.

Close modal

Specifically, when agent productivity changes, the violation of streamflow remains relatively stable, while the benefits for the agents undergo alterations, though not to a degree sufficient to significantly impact the instrument's comparative effectiveness. This is because, in all MASS with quota scenarios, every agent pumps to some extent since the quota limits the amount of each user's pumped water. The pumping would cease only when the associated costs of pumping reached an excessively high level. This results in greater fairness in comparison to the tax outcomes. The scenarios involving MASS with tax lead to a shift in water usage toward the agents with higher productivity, since numerous less-productive ones discontinue water usage when the tax reaches excessive levels. The quota policy option is more resilient when considering the heterogeneity of agents.

Table 5 presents the outcomes of model runs of MASS with quota and MASS with tax in comparison to the COM and free-access MASS. This table illustrates the impact of distinct different configurations of parameters for each model type similar to Figure 7. Generally, the distinctions among the models are more significant than the uncertainty span depicted through the sets of parameters, though there is a certain commonality.

Figure 9 enables a comparison of the economic and environmental impact metrics for whole runs across the formulation of each model (MASS with quota, MASS with tax, free-access MASS, and COM). The boxplots of each figure indicate the ranges of agent productivity parameter sets.
Figure 9

The average annual benefit, the violation of streamflow, and the total basin-wide water use boxplots.

Figure 9

The average annual benefit, the violation of streamflow, and the total basin-wide water use boxplots.

Close modal

The first row enables a comparison of COM and free-access MASS. Two crucial observations can be identified: (1) The free access of self-interested agents is detrimental to the environment and (2) COM can achieve zero the violation of streamflow with only a minor reduction in agent profits. The second row displays the results of MASS with quota. Reducing the quota to mitigate streamflow depletion leads to substantial and probably unaffordable economic consequences. As the quota decreases, there is a minor reduction in the violation of streamflow up to approximately 300 mm, after which a more significant decrease in the violation of streamflow occurs. A quota as low as this one leads to a substantial reduction in the basin's production. One significant difference between quota and tax policies is that the impact of variations in productivity on streamflow violation variability is lessened with a quota policy, which is an anticipated outcome. This is noteworthy because it significantly diminishes the uncertainty linked to the impact of the quota on streamflow. In all runs, a quota of a little more than 100 mm on maximum water usage is necessary to decrease the violation of streamflow to 20% or less. To match the COM's average yearly basin-wide profit of 120.3 trillion rials, a quota of a little more than 200 mm or more is necessary. However, this results in a little more than 39% violation of streamflow. The third row displays the results of MASS with tax (a). Increasing the tax from 1 to 5 million rials per m3 does not result in any advantages for reducing the violation of streamflow and agent profits. The violation of streamflow experiences a significant decline when the tax ranges from 5 to 11 million rials per m3 because many agents stopping pumping. A tax exceeding a little more than 9 million rials per m3 is necessary to decrease the violation of streamflow to below 20%. To achieve the same, average annual benefit as the COM, a tax of a little less than 9 million rials per m3 or less is necessary. However, this results in a minimum violation of streamflow of a little more than 26%. The fourth row displays the results of MASS with tax (b). It is not feasible to maintain low violation of streamflow without decreasing the profits of the agent using this formulation. Increasing the tax from 1 to 5 million rials per m3 does not result in any advantages for reducing the violation of streamflow. Instead, it only leads to a decrease in agent profits. The violation of streamflow experiences a significant decline when the tax ranges from 5 to 11 million rials per m3 due to the presence of numerous agents stop pumping. In this scenario, only the most skilled agents utilize water. A tax exceeding a little more than 9 million rials per m3 is necessary to decrease the violation of streamflow to below 20%. At this tax rate, the average annual benefit is not substantially declined. To achieve the same average annual benefit as the COM, a tax of a little less than 3 million rials per m3 or less is necessary. However, this outcome at a minimum the violation of streamflow of a little more than 88%. The violation of streamflow is most significantly influenced by the variability in farmer productivity.

Figure 10 displays the results of average annual tax benefits MASS with Tax (a) and Tax (b). If the tax rate exceeds 5 million rials per m3, the tax benefits surpass the agents' earnings. This suggests that the tax is quite substantial and possibly is not practical in the absence of a forceful reallocation of tax benefits. This indicates that if the tax rate is below 5 million rials per m3, the violation of streamflow is equivalent to the level of free access and the average annual benefit are also similar. A tax rate as low as this would only lead to a slight decrease in overall water usage and may not be practical to implement.
Figure 10

The average annual tax benefits versus MASS with tax boxplots. The yellow boxplots show the average annual benefit of MASS with Tax (a), the blue boxplots show the average annual benefit of MASS with Tax (b) and the red boxplots show the average annual tax benefits. The sum of the average annual tax benefits and the average annual benefit of MASS with Tax (b) equals the average annual benefit of MASS with Tax (a).

Figure 10

The average annual tax benefits versus MASS with tax boxplots. The yellow boxplots show the average annual benefit of MASS with Tax (a), the blue boxplots show the average annual benefit of MASS with Tax (b) and the red boxplots show the average annual tax benefits. The sum of the average annual tax benefits and the average annual benefit of MASS with Tax (b) equals the average annual benefit of MASS with Tax (a).

Close modal

The literature suggests that the consistent water usage quota and consistent water usage tax possess certain benefits (Pistón et al. 2009; Valle-García et al. 2024). However, the method of modeling used in this study, which combined a groundwater model that considers physical properties and includes heterogeneous agents with varying characteristics, shows that these approaches are more complex than previously thought when applied in practice. The paper suggests that the quota is more resilient and effective than tax. It's worth noting that these findings do not account for precipitation fluctuations, which might be the scenario where the tax operates more effectively. For example, in periods of prolonged aridity, agents could extract water up to the point of economic feasibility, unlike the imposed quota, which would restrict water use regardless of the climate condition. In summary, in the event that variability of climate was to be offered, the tax might become a more attractive alternative. The tax system, when not redistributed, generates a higher return in contrast to the quota, which leads to an equivalent reduction in streamflow. The study indicates that it is crucial to evaluate the appropriate policy prescription for a basin by examining the aquifer's physical depiction and the farmers' heterogeneity.

This paper does not consider numerous other feasible and noteworthy forms of groundwater management approaches such as subsidies for efficient irrigation technologies, a temporary prohibition on new well construction, exchangeable permits, geographically diverse quotas/taxes, adaptable policies, and voluntary limitations. According to most literature, a quota and trade program that provides incentives for managing groundwater use could lead to higher economic returns for irrigation in contrast to the current policy of a quota without trade provisions (Thompson et al. 2009). The text suggests that it would be a logical next step to evaluate each of these alternative policy configurations within the framework established in this paper, considering the outcome of examining data. The text shows that the literature agrees with our findings on two key points regarding the reallocation of tax revenues to the agents (Mulligan et al. 2014). First, the reallocation of a substantial portion of the revenues of tax must be redistributed back to the agents to render the tax choice feasible. Second, the text suggests that the quota system is probably the favored policy choice compared with the tax system. The reason for this is achieving homogeneous tax rates necessary to sufficiently mitigate the impacts on streamflow is likely politically unattainable due to their substantial size. As previously mentioned, introducing a varied tax could be adopted to enhance the feasibility of the tax system.

The study has some limitations that should be taken into account when interpreting the results. These limitations include: (1) presuming that the inputs and parameters of the model remain consistent over time, including productivity parameters, annual climate conditions, operational costs for farmers, and prices at which crops are sold; (2) the precision of numerical solvers for MODFLOW and the optimization algorithms; (3) disregarding the agents fixed costs; (4) employing a function of crop production that is linear; and (5) utilizing streamflow goals that are not connected to the genuine goals within the plain.

This study used a physically based model to evaluate four distinct strategies for the utilization of groundwater in the Ardabil plain aquifer. The COM was employed as a reference point for assessing mechanisms of policy in a world where complete understanding and anticipation are not available. The evaluation of the strategy involved using a behavior model of water utilization of heterogeneous farmers, coupled with a flow process model of heterogeneous groundwater. The policy assessment examined the consistent tax and the consistent quota. The outcomes indicated that these policies were not successful in comparison to the COM in a scenario where agents acted shortsightedly and neglected their impact on streamflow. Each management strategy was evaluated using two key metrics: the average annual benefit, which represents economic benefits, and the violation of streamflow, which represents environmental impact.

A tax of greater than 9 million rials per m3 is needed to decrease the violation of streamflow to fewer than 20%. However, a tax of such size caused the closure of almost all the agents. In evaluating the benefits (sum of the agent profits and tax), this turned into a more feasible choice. A water quota of a little more than 100 mm or less is needed to decrease the violation of streamflow to fewer than 20%. However, this led to a considerable decline in the average annual benefit and significantly decreased the basin's crop production. The study found that the following results were observed: (1) To examine policy instruments implemented in a groundwater plain, an intricate physical model was required. (2) The free-access MASS policy instrument, which assumed shortsighted and self-interested agents, led to notable reductions in streamflow and proved to be unsustainable. (3) The consistent tax policy instrument proved to be somewhat ineffective and necessitated the implementation of a redistribution program for its execution. (4) The consistent quota policy instrument performed more effectively than the tax when not redistributed; however, it needed a quota to such a minimal extent that agents were unable to satisfy their crop requirements for the extent of land dedicated to crops under their ownership. (5) The quota is a stronger and more resilient policy choice for decreasing the violation of streamflow than the tax.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.

Alam
M. F.
,
McClain
M.
,
Sikka
A.
&
Pande
S.
2022
Understanding human–water feedbacks of interventions in agricultural systems with agent based models: A review
.
Environmental Research Letters
17
(
10
),
103003
.
Axelrod
R.
1997
The Complexity of Cooperation: Agent-Based Models of Competition and Collaboration
.
Princeton University Press
,
Princeton, NJ
.
Bartolino
J. R.
&
Cunningham
W. L.
2003
Ground-Water Depletion Across the Nation (No. 103-03)
.
Brozović
N.
,
Sunding
D. L.
&
Zilberman
D.
2010
On the spatial nature of the groundwater pumping externality
.
Resource and Energy Economics
32
(
2
),
154
164
.
Canales
M.
,
Castilla-Rho
J.
,
Rojas
R.
,
Vicuña
S.
&
Ball
J.
2024
Agent-based models of groundwater systems: A review of an emerging approach to simulate the interactions between groundwater and society
.
Environmental Modelling & Software
175
,
105980
.
Castilla-Rho
J. C.
,
Mariethoz
G.
,
Rojas
R.
,
Andersen
M. S.
&
Kelly
B. F.
2015
An agent-based platform for simulating complex human–aquifer interactions in managed groundwater systems
.
Environmental Modelling & Software
73
,
305
323
.
Di Baldassarre
G.
,
Murugesu
S.
,
Maria
R.
,
Christophe
C.
,
Margaret
G.
,
Heidi
K.
,
Megan
K.
,
Mondino
E.
,
Johanna
M.
,
Saket
P.
&
Sanderson
M. R.
2019
Socio-hydrology: Scientific challenges in addressing a societal grand challenge
.
Water Resources Research
55
(
8
),
6327
6355
.
Dietz
T.
,
Ostrom
E.
&
Stern
P. C.
2003
The struggle to govern the commons
.
Science
302
(
5652
),
1907
1912
.
Galán
J. M.
,
López-Paredes
A.
&
Del Olmo
R.
2009
An agent-based model for domestic water management in Valladolid metropolitan area
.
Water Resources Research
45
(
5
),
W05401
.
Harou
J. J.
,
Pulido-Velazquez
M.
,
Rosenberg
D. E.
,
Medellín-Azuara
J.
,
Lund
J. R.
&
Howitt
R. E.
2009
Hydro-economic models: Concepts, design, applications, and future prospects
.
Journal of Hydrology
375
(
3–4
),
627
643
.
Holland
J. H.
&
Sigmund
K.
1995
Hidden order: How adaptation builds complexity
.
Nature
378
(
6556
),
453
453
.
Holtz
G.
&
Pahl-Wostl
C.
2012
An agent-based model of groundwater over-exploitation in the Upper Guadiana, Spain
.
Regional Environmental Change
12
(
1
),
95
121
.
Hoppe
R. A.
2010
Small Farms in the United States: Persistence Under Pressure
.
Diane Publishing, United States Department of Agriculture, Economic Research Service
.
Jabbari
R.
,
Zarghami
M.
,
Anbari
M. J.
&
Nadiri
A. A.
2022
Development of agent-based model with the aim of discussion groundwater resources management policies; case study of Damghan Aquifer
.
Iran-Water Resources Research
17
(
4
),
131
143
.
Kamali
A.
&
Niksokhan
M. H.
2017a
Development of a model for calculation of sustainability index of groundwater resources
.
Iranian Journal of Ecohydrology
4
(
4
),
1071
1087
.
Kelly
R. A.
,
Jakeman
A. J.
,
Barreteau
O.
,
Borsuk
M. E.
,
ElSawah
S.
,
Hamilton
S. H.
,
Henriksen
H. J.
,
Kuikka
S.
,
Maier
H. R.
,
Rizzoli
A. E.
&
Van Delden
H.
2013
Selecting among five common modelling approaches for integrated environmental assessment and management
.
Environmental Modelling & Software
47
,
159
181
.
Khan
H. F.
,
Yang
Y. C.
,
Xie
H.
&
Ringler
C.
2017
A coupled modeling framework for sustainable watershed management in transboundary river basins
.
Hydrology and Earth System Sciences
21
(
12
),
6275
6288
.
Koundouri
P.
2004
Current issues in the economics of groundwater resource management
.
Journal of Economic Surveys
18
(
5
),
703
740
.
Kord
M.
,
Asghari-Moghaddam
A.
&
Nakhaeei
M.
2013
Investigation of hydrogeological characteristics of Ardabil Plain aquifer, Northwest of Iran
.
ISESCO Journal of Science and Technology
9
(
15
),
63
69
.
Mariño
M. A.
&
Simonović
S. P.
2001
Integrated Water Resources Management (No. 272)
.
International Association of Hydrological Sciences, University of California
,
Davis, California, USA
.
Ma
Y.
,
Shen
Z.
,
Kawakami
M.
,
Suzuki
K.
&
Long
Y.
2012
An Agent-Based Approach to Support Decision-Making of Total Amount Control for Household Water Consumption. Geospatial Techniques in Urban Planning, p. 107
.
McDonald
M. G.
&
Harbaugh
A. W.
1988
A Modular Three-Dimensional Finite-Difference Ground-Water Flow Model
.
US Geological Survey
.
Mulligan
K. B.
,
Brown
C.
,
Yang
Y. C. E.
&
Ahlfeld
D. P.
2014
Assessing groundwater policy with coupled economic-groundwater hydrologic modeling
.
Water Resources Research
50
(
3
),
2257
2275
.
Nazemi
A. R.
,
Dolatshahi
M.
&
Kerachian
R.
2024
A decentralized multi-agent framework for urban flood management
.
Sustainable Cities and Society
106
,
105328
.
Norouzi Khatiri
K.
,
Niksokhan
M. H.
,
Sarang
A.
&
Kamali
A.
2020
Coupled simulation-optimization model for the management of groundwater resources by considering uncertainty and conflict resolution
.
Water Resources Management
34
(
11
),
3585
3608
.
Nouri
A.
,
Saghafian
B.
,
Bazargan-Lari
M. R.
,
Delavar
M.
&
Hassanjabbar
A.
2022
Impact of penalty policy on farmers’ overexploitation based on agent-based modeling framework
.
Journal of Water Resources Planning and Management
148
(
5
),
04022015
.
Pande
S.
&
Sivapalan
M.
2017
Progress in socio-hydrology: A meta-analysis of challenges and opportunities
.
Wiley Interdisciplinary Reviews: Water
4
(
4
),
e1193
.
Pistón
J. M.
,
Kolberg
S.
,
Berbel
J.
&
Giannoccaro
G.
2009
Comparative Analysis of Water Saving Policies in Agriculture: Pricing Versus Quotas
.
Pulido-Velazquez
M.
,
Marques
G. F.
,
Harou
J. J.
&
Lund
J. R.
2016
Hydroeconomic models as decision support tools for conjunctive management of surface and groundwater. Integrated groundwater management: Concepts, Approaches and Challenges, (Jakeman, A. J., Barreteau, O., Hunt, R. J., Rinaudo, J.-D., Ross, A., eds.), Springer International Publishing: Cham, Switzerland, 2016, pp. 693–710
.
Schaible
G. D.
2004
Irrigation, Water Conservation, and Farm Size in the Western United States
.
The Economics of Food, Farming, Natural Resources, and Rural America
,
Amber Waves
.
Schelling
T. C.
1971
Dynamic models of segregation
.
Journal of Mathematical Sociology
1
(
2
),
143
186
.
Soman
S.
,
Misgna
G.
,
Kraft
S. E.
,
Lant
C.
&
Beaulieu
J. R.
2008
An Agent-Based Model of Multifunctional Agricultural Landscape Using Genetic Algorithms
.
Thompson
C. L.
,
Supalla
R. J.
,
Martin
D. L.
&
McMullen
B. P.
2009
Evidence supporting cap and trade as a groundwater policy option for reducing irrigation consumptive use
.
JAWRA Journal of the American Water Resources Association
45
(
6
),
1508
1518
.
Valle-García
Á.
,
Gutiérrez-Martín
C.
&
Montilla-López
N. M.
2024
Water pricing and quotas: A quantitative analysis from a private and social perspective
.
Water Resources Management
38
,
1
20
.
Wilensky
U.
1999
NetLogo
.
Wu
B.
,
Zheng
Y.
,
Wu
X.
,
Tian
Y.
,
Han
F.
,
Liu
J.
&
Zheng
C.
2015
Optimizing water resources management in large river basins with integrated surface water-groundwater modeling: A surrogate-based approach
.
Water Resources Research
51
(
4
),
2153
2173
.
Yang
Y. C. E.
,
Cai
X.
&
Stipanović
D. M.
2009
A decentralized optimization algorithm for multiagent system-based watershed management
.
Water Resources Research
45
(
8
),
W08430
.
Yang
Y. C. E.
,
Zhao
J.
&
Cai
X.
2012
Decentralized optimization method for water allocation management in the Yellow River Basin
.
Journal of Water Resources Planning and Management
138
(
4
),
313
325
.
This is an Open Access article distributed under the terms of the Creative Commons Attribution Licence (CC BY 4.0), which permits copying, adaptation and redistribution, provided the original work is properly cited (http://creativecommons.org/licenses/by/4.0/).