Assessment of the impact of socio-economic policies on groundwater consumption using a multi-agent-based modeling approach

Worldwide, human intervention through land use changes and various management practices has led to significant changes in the hydrological regime throughout history. The hydrological regime, which was previously controlled mainly by factors such as climate, land use patterns, and topography, is now increasingly influenced by social and economic factors. Therefore, in addition to investigating hydraulic factors, it is necessary to investigate humans affecting the watershed (Kuil et al., 2019). From a mathematical modeling perspective, it is difficult to incorporate all the behavioral aspects of human and water system interactions, as well as their coevolution (Akhbari & Grigg, 2013). Various techniques have been developed to integrate these systems, but it remains a challenge. One popular approach is to use agent-based modeling (ABM), which is well-suited for simulating complex systems. ABM involves creating computational models that simulate the behavior of autonomous agents to understand the behavior of a larger system. As such, ABM can facilitate shared vision approaches in this area (Akhbari & Grigg, 2015). Overall, the ABM approach focuses directly on individual objects, their behavior, and interactions. As a result, ABM can be considered a step toward understanding and managing the complexity of modern social systems. An ABM with a neighborhood theory was employed by Akhbari & Grigg (2013) to simulate the behavior of farmers in California's Sacramento-San Joaquin delta. Their model demonstrated how humans interacted with each other and how collaboration can be fostered to find solutions. Farhadi et al. (2016) proposed an integrated approach of simulating groundwater using MODFLOW and an ABM framework to study stakeholder decisions for sustainable groundwater management. The results showed that using integrated approaches for water resources management allows decision-makers to evaluate the performance of different management scenarios and make more practical decisions.

Also, Neisi et al. (2020) utilized the Protection Motivation Theory to assess the management behavior of farmers in mitigating the risks of drought in the downstream of the Karkheh Dam. The findings of this study suggest that managerial policies should enable farmers to make decisions regarding their own risk management and have access to various tools and strategies. Aghaie et al. (2020) presented an agent-based groundwater market model to analyze the economic and hydrological effects of different policies on water trading in the Rafsanjan plain in Iran. They showed that precise definitions of surveillance and enforcement policies can lead to the emergence of social norms to prevent farmers' violations and promote efficient markets. Du et al. (2017) proposed an integrated hydrological modeling framework with an ABM to predict farmers' decision-making on water usage. They demonstrated that physical conditions of farmers, such as the distance from rivers and the status of groundwater in the region, significantly affect the interaction between human activities and hydrological systems.

An ABM has been proposed by Nouri et al. (2022) to examine the impact of penalty enforcement on curbing unauthorized groundwater extraction by farmers. The agricultural sector's behavior was simulated on two levels: individual farming agents and group farming agents (unions). The simulation demonstrated that the implementation of a penalty of 0.12$ per m³ resulted in a significant reduction in excessive annual exploitation, from 79 million m³ to zero. Bahrami et al. (2022) developed an ABM to optimize the use of agricultural water resource reservoirs, taking into account the interactions between farmers and policy-makers. The model calculated the release of water from a hypothetical reservoir according to the standard operating policy, and then evaluated the allocation of water to farmers, as well as their ability to adapt to water scarcity through changes in cultivation patterns, area, and irrigation methods. The study found that the level of risk tolerance was the most significant factor affecting farmers' responses, which could be managed to increase their decision-making flexibility. Also, Javan Salehi & Shourian (2023) crafted a socio-hydrological model merging the Soil and Water Assessment Tool (SWAT) and MODFLOW to appraise human–water systems. Through the application of the Value-Belief-Norm Theory, they identify elements influencing farmers' water usage habits. Outcomes suggest that climate change may steer financially disadvantaged farmers toward profitable yet water-demanding crops, potentially reducing Lake Urmia's water inflow. The proposal advocates for policy adjustments to address these effects, offering guidance for future strategies and policy modifications aimed at revitalizing Lake Urmia amid the changing economic and psychological aspects for farmers.

In general, several researchers have asserted that ABM is among the most effective techniques for analyzing and assessing complex adaptive systems. However, there is limited research on the extension of tightly coupled integrated modeling, which includes a comprehensive physically based hydrological model and a model based on stakeholder interactions. The Eshtehard aquifer has experienced a significant decline in its aquifer due to inappropriate utilization of groundwater resources in recent years. As a result, sustainable management of this vital resource, while considering economic, social, and environmental factors, is crucial. So, this study attempted to develop a framework for the operation and management of the Eshtehard aquifer that takes into account stakeholder behavior and feedback. The main goal of this study is to preserve the water resources of the aquifer while satisfying the requirements of the stakeholders, without causing social and economic conflict.

The SWAT is a process-based model at the watershed scale that employs semi-distributed parameters in a continuous-time hydrologic model. The model predicts the impact of land management practices on water, sediment, and nutrient yields during watershed simulation (Sabzzadeh & Shourian, 2020). To merge multiple spatial environmental data, the SWAT system is integrated with a Geographical Information System. The SWAT model of the Eshtehard watershed is established using various data sources, including the Digital Elevation Model with 30m cell size, topographic data (slope and drainage network), land use, soil type, meteorological data (minimum and maximum temperature, precipitation, relative humidity, solar radiation, and wind speed), groundwater data, and crop cultivation data. Data related to the three main irrigated crops, namely WWHT, WBAR, and COT, were obtained from authorized reports by national agencies and incorporated into the model. The model developed for the Eshtehard plain consists of 21 sub-watersheds and 167 HRUs, with agriculture products being cultivated in seven sub-watersheds. The model was executed in monthly time steps for the period of 2000–2018. Four meteorological stations and two hydrometric stations, inclusive of their geographic coordinates and time-series data, have been integrated into the SWAT model. It is pivotal to highlight that the ‘Najm-abad’ hydrometric station serves the purpose of determining the water inflow values into the basin, whereas the ‘Asefoddoleh’ hydrometric station is specifically designated to ascertain the basin's outflow values within the modeling framework. Utilizing land use maps acquired from the Ministry of Energy of Iran and the global soil map, categorization of land use and soil composition across diverse segments of the Eshtehard plain has been established. This model accommodates the inclusion of intricate elements, such as scheduling, irrigation quantities, water sourcing, temporal aspects, fertilizer types and quantities, and planting and harvesting schedules, among others, tailored for each agricultural product. This functionality greatly facilitates the process of comprehensive agricultural planning.

To simulate groundwater resources in the region, the MODFLOW model was used in this study. MODFLOW is a three-dimensional simulation model that uses the finite difference method to simulate groundwater flow (McDonald & Harbaugh, 2003). In this study, the Eshtehard aquifer was divided into 600 cells (20 rows and 30 columns). To model the Eshtehard aquifer, 14 observation wells and 123 extraction wells were used. In addition, the steady model was run from October 2010, and its unsteady development was modeled until September 2018 in 96 monthly time steps. Other relevant groundwater data such as geological information, hydraulic conductivity, specific yield, bedrock data, and monthly water table elevations were collected from the reports approved by the Ministry of Energy of Iran (2014).

Visual representations of the groundwater recharge values derived from the model's output were displayed in Figure 3(b), accompanied by the calibrated hydraulic conductivity coefficients showcased in Figure 3(c). The validation outcomes for the specific yield coefficient are graphically depicted in Figure 3(d).

Farmers' circumstances and attributes greatly impact their thought processes and decision-making abilities. The behavior and innovative tendencies of influential individuals are significantly shaped by their age (Javan Salehi & Shourian, 2023). Older farmers generally show less inclination toward embracing change and taking risks, while younger farmers tend to display a higher inclination for risk-taking and adapting to change. Hence, a clear inverse relationship exists between farmers' age and their risk aversion. This study defined farmers' ages through a random distribution function within the range of 30–70 years, factoring in the aggregation of farmers across seven key agricultural factors. It is important to note that, to maintain consistency in individual conditions affecting decision-making, the farmers' ages remained constant throughout the modeling process (Bahrami et al., 2022). Also, another influential aspect impacting farmers' conduct is their economic status and reliance on agriculture. Those heavily reliant on agriculture for their livelihoods are generally less open to change and taking risks. Conversely, individuals with lesser dependence on agriculture tend to exhibit a higher inclination for taking risks and adapting to changes (Pouladi et al., 2020). Therefore, there is an inverse relationship between relying on agricultural income and farmers' risk aversion. In this study, the reliance of income on agriculture was defined using a random distribution function constrained within a range of 30–100%, ensuring that all farmers have a minimum income dependence of 30% on agriculture across the seven agricultural factors. Overall, in decision-making within human societies, including among farmers, the level of risk aversion significantly influences choices. Factors such as awareness, environmental concerns, economic conditions, and access to water resources, both in recent and historical periods, contribute to this behavior. These factors need to be considered in a simulation model of farmers' behavior. The parameter of risk aversion itself relies on fixed factors like age range and income dependence on agriculture, determining its calculation. Risk aversion demonstrates an inverse correlation with both these factors: as age and income dependence on agriculture increase, a farmer's willingness to take risks decreases. For instance, a 30-year-old farmer with a 30% income dependence on agriculture portrays the highest risk aversion, denoted as 1, whereas a 70-year-old farmer with a 100% income dependence on agriculture exhibits the lowest risk aversion, denoted as 0 (Bahrami et al., 2022). So, the primary factor that influences their decision-making is their level of risk tolerance (Risk_i), which is determined and computed based on two indices: the farmer's age (Age_i) and their income dependence on agriculture (IncDep_i).

Agents make assumptions about the availability of water for the next year, and then they decide on the pattern of cultivation for the upcoming year. The decision-making process of these agents is aimed at maximizing individual and group profits, and for this purpose, they also use their own situations and the situations of their neighbors in recent years and in the past. In the first three years of modeling, agents have the same and constant pattern of cultivation to create the necessary historical memory. It should be noted that this initial pattern is based on actual data used at the plain level, including cultivation of cotton, wheat, and barley with a combination of 20, 40, and 40% for all agents. The decision-making behavior model of farmers consists of four steps performed in order and ultimately ending with the cultivation of the following year.

In Equations (6) and (7), α and β denote the degree of risk-taking and deviation in agricultural consumption, respectively. L₁, L₂, and L₃ are the decision-making thresholds for agents, while index i refers to the farmer agent number, index y denotes the year, and index k is a counter.

Moreover, β denotes the deviation value, i represents the agent number, and y represents the year number. The resulting figure is in cubic meters per hectare and signifies the optimal pumpable water quantity for the next year perceived by the farmer.

Actually, the educational awareness parameter represents the agents' inclination to bring water extraction levels closer to permissible values, which increases with the implementation of education and heightened environmental awareness among farmers. In the absence of education among farmers, this parameter remains at 1, exerting no influence on agents' decision-making. Upon executing educational programs, if an agent's extraction in the previous water year exceeds the permissible amount, the environmental awareness parameter falls below 1, reducing its impact on the agent's inclination toward the permissible level in the new water year. Conversely, if an agent's extraction in the previous water year is below the permissible amount, the environmental awareness parameter exceeds 1, allowing for an increase in the extraction level in the new water year.

The ultimate stage involves selecting a suitable cultivation pattern based on the computed FarmerSub value for each farmer and taking into account the water demands of the three primary agricultural products in the region: cotton, wheat, and barley, which are 8,060, 3,460, and 2,930 m³ per hectare, respectively (Ministry of Energy of Iran, 2014). Selecting a cultivation pattern in proportion to the FarmerSub value is achieved by dividing the feasible numerical range into seven stages, considering the water demands of cotton, wheat, and barley, and determining the percentage of cultivation for each agricultural product based on the corresponding stage of the estimated FarmerSub value. If the estimated FarmerSub value is lower, the cultivation pattern is inclined toward using barley and less toward cotton. Conversely, if the estimated FarmerSub value is higher, the cultivation pattern leans toward using cotton and less toward barley. For average values of FarmerSub, the cultivation pattern tends toward using wheat. Once the cropping pattern is determined, the SWAT cropping pattern is adjusted using the LUP.dat file, and the model is executed for the following year. This process continues in the same manner until the desired time period is completed.

This study is divided into two sections. The first section focuses on calibration and validation analysis of the proposed ABM-SWAT-MODFLOW model, aiming to adjust the parameters and validate the model. The second section involves designing five different scenarios to evaluate the impacts of socio-economic changes on water resources.

In the first step, the proposed SWAT-MODFLOW model needs to be calibrated to provide proper performance in simulating integrated surface and groundwater. In this study, the calibration of the integrated model was performed in three steps. In the first and second steps, separate calibration and validation were performed for the SWAT and MODFLOW models. Then, in the third step, simultaneous calibration of the SWAT-MODFLOW model was done using the information from the first and second steps. After performing the calibration process for the SWAT model, 10 acceptable answer sets with NS coefficients greater than 0.65 were extracted.

Then, by running each answer set in the SWAT model, the recharge values to the aquifer and irrigation volume to the MODFLOW model were entered, and the values of hydraulic conductivity and hydraulic anisotropy coefficient parameters for the steady state and hydraulic conductivity, and specific yield values for the unsteady state were recalculated using the PEST method in the MODFLOW .The results obtained from the comparison between observed and calculated water levels and the grouping of answers that provide the best solution for both models were used as the final answer. It should be mentioned that the SWAT model was calibrated using data from 2004 to 2015, and validation was carried out using data from 2016 to 2018. In addition, the model was warmed up using data from 2002 to 2004.

For further investigation, the values obtained for evaluation criteria of root mean square error (RMSE) and mean absolute error (MAE) for the average water head calculated by the model and observed data in two steady and unsteady states are presented in Table 2. As the presented results show, the MODFLOW model was able to simulate the trend of groundwater head. Overall, the results obtained for the proposed SWAT-MODFLOW model in this study are satisfactory.

Upon developing the ABM, the initial step involves verifying it by modifying the risk-taking levels of agricultural agents (L₁, L₂, and L₃) to facilitate their decision-making process regarding altering the cropping pattern for the upcoming irrigation season. To achieve this, the final risk-taking level values were attained by a trial-and-error method (Du et al., 2017) and by calculating NS and RMSE coefficients for the annual total groundwater withdrawal and aquifer level (Table 3).

Following, Table 4 shows the annual total water withdrawals of farmers in the basin from the aquifer under two scenarios: current conditions (without changing the cropping pattern) and with the annual implementation of the cropping pattern change using the ABM.

Based on the outcomes presented in Table 4, it can be inferred that the ABM fits well with the present state of basin operations, and the selected decision-making boundaries are precise. It can be concluded that the proposed SWAT-MODFLOW-ABM model can effectively depict the consumption behavior of farmers in the Eshtehard plain in terms of groundwater withdrawal. It is worth noting that, where the cropping pattern remains unchanged, the cultivation area for each crop remains constant, thereby impacting the irrigation scheme. Consequently, with full water demand met from the aquifer, water extraction stays consistent and repetitive. Conversely, when changes are introduced to the cropping pattern and the influence of behavioral simulation (ABM model) is applied, the cultivation area, irrigation, and annual groundwater extraction exhibit variations.

A government entity is tasked with overseeing and managing water resource utilization, aiming to guide farmers in cultivating crops with lower water demands by employing various strategies across different scenarios. The policy of ‘Education and increasing farmers’ awareness’ about authorized water resource usage and groundwater preservation holds the potential to effectively conserve water resources by instilling environmental concerns. To implement this policy, the government body can conduct educational programs focusing on informing farmers about the consequences of depleting water resources and urging compliance with defined water extraction limits for the Eshtehard Plain. This policy, when applied mathematically, brings farmers' water extraction closer to permissible limits. The ‘Alternative employment’ policy involves creating suitable platforms and training farmers to initiate home-based businesses like handicrafts, reducing households' reliance on agricultural income and the financial need for profits from farming. This reduced dependency enables farmers to transition from high-profit, high-water-demanding crops like cotton to less profitable and less-water-intensive crops such as wheat and barley, thereby alleviating pressure on the aquifer. For this study, a 20% reduction in farmers' economic reliance on agricultural income is assumed, leading to a corresponding 20% reduction in figures and an increased willingness among farmers to adjust their cropping patterns without significantly affecting overall profitability due to non-agricultural income. The ‘Water pricing’ policy aims to reduce the allure of high-water-consuming products like cotton, which require more water and are more profitable. By diminishing their profitability, this policy guides farmers toward less-water-intensive and less profitable crops like wheat and barley. Water pricing, guided by the Water Resource Pricing Council of the Ministry of Energy and based on relevant time-series data, results in reduced profitability for high-water-demand crops. The implementation of water purchase costs further reduces the profitability of these crops, encouraging farmers to consider cultivating less-water-intensive alternatives.

In this section, the proposed ABM-SWAT-MODFLOW model is employed to create five scenarios (a combination of the aforementioned approaches as shown in Table 5) aimed at assessing the socio-economic impacts on farmers' activities and the state of water resources.

Based on the outcomes presented in Table 6, SC2, SC4, and SC5 scenarios have exhibited the most favorable effect on water consumption, resulting in an increase in the groundwater head at the end of the modeling period. Nonetheless, SC2 scenario's profitability performance was unsatisfactory due to the water purchasing cost incurred by farmers. In addition, in terms of profitability, while the agents' profitability varies in different years depending on the weather conditions, especially precipitation, SC1, SC3, and SC4 scenarios have demonstrated satisfactory performance. Furthermore, water conservation has also been achieved in the SC4 scenario. It should be noted that the SC4 scenario is costly for government agents and requires investment. According to the model results, it can be claimed that the ‘environmental education and awareness-raising’ by creating environmental concerns increases the agents' inclination toward low-water-intensive farming patterns and leads to the revival of groundwater resources while it also leads to a decrease in their income.

The ‘alternative employment’ approach also reduces households' dependency on income from agriculture and consequently reduces the financial need for profitable farming, allowing farmers to cultivate less profitable and less-water-intensive crops such as wheat and barley instead of high-profit and high-water-intensive crops like cotton, resulting in less pressure on the aquifer. However, the negative point for both approaches is their reliance on public budget and investment. In the water pricing approach, given that the more profitable crop in the region (cotton) requires more water, reducing the profit of high-water-consuming crops reduces their attractiveness and drives farmers toward less profitable crops such as wheat and barley. This approach leads to significant income for the government but reduces the farmers' profits significantly.

Taking into account the overall results, it can be summarized that the SC4 scenario has shown the best performance in improving the hydrological–social conditions of the region, resulting in a 13.9% decrease in total groundwater withdrawal over a study period. The average groundwater head in the western half of the region has increased by 1.4 m, and in the most critical piezometer, it has increased by 2.3 m. Meanwhile, it has also caused a 3.3% increase in farmers' total income in the area. However, implementing this scenario requires investment and expenses from the governmental party. On the other hand, the SC5 scenario has also performed well in improving the hydrological–social conditions of the region, resulting in a 14.5% decrease in total groundwater withdrawal over an 8-year period. Furthermore, the average groundwater head in the western half has increased by 1.3 m and has increased by 2.2 m in the most critical piezometer, leading to a 7.9% decrease in the overall income of farmers in the region. Implementing this scenario is not cost-effective for government organizations. Another noteworthy result is that the agents' performance is highly influenced by their fixed parameters, especially agents' risk-taking levels, which are affected by their age, level of economic dependence on agriculture, determining the type of agent response, and the level of their response to different conditions.

The present study aims to model the hydrological and social systems of water resources based on surface and groundwater simulation models, as well as behavioral simulations of water resource managers. One of the main objectives of this research is to investigate the effectiveness of governmental approaches to controlling water extraction by farmers while preserving their economic interests. The studied area is the Eshtehard aquifer in the Alborz province, Iran. Nowadays, uncontrolled water extraction from the Eshtehard aquifer has caused a serious decrease in water levels and land subsidence. Therefore, in this study, the hydrological model of the Eshtehard plain was first simulated by linking the SWAT and the groundwater model (MODFLOW). Then, an agent-based behavioral simulation model (ABM) was built for the stakeholders of the water system, including a governmental agent and seven agricultural agents who have independent decision-making power and the ability to change their cultivation patterns. Subsequently, by integrating the ABM, SWAT, and MODFLOW, an integrated model (ABM-SWAT-MODFLOW) was developed and employed to investigate three policies of ‘Water pricing’, ‘Alternative employment’, and ‘Environmental education and awareness raise’ through five scenarios.

Based on results, the implementation of both ‘Environmental education and awareness raise’ and ‘Alternative employment’ policies in the SC4 scenario leads to a satisfactory situation in enhancing the hydrological–social condition in the region, provided the government has the required financial budget for investment. This policy can result in a total reduction of 13.9% in groundwater extraction over an 8-year period. Furthermore, the average groundwater level has risen by 1.4 m in the western region, and the most crucial piezometer has experienced a 2.3 m increase, leading to a 3.3% rise in the total income of farmers in the area. If a government agent does not have the necessary financial budget for investment, implementing the three approaches of ‘water pricing’, ‘alternative employment’, and ‘environmental education and awareness-raising’ simultaneously (in the SC5 scenario) can be feasible. This approach leads to the best hydrological–social effectiveness, resulting in a 14.5% reduction in total groundwater extraction over an 8-year period. Furthermore, the average groundwater head has increased by 1.3 m in the western half and 2.2 m in the most critical piezometer. This scenario leads to a 7.9% decrease in the total income of farmers in the region. Overall, the results of this study can be useful for future planning and policy-making, estimating the impact of implementing management scenarios on the future conditions of water resources and the economic status of farmers.

All authors contributed to the study's conception and design. Material preparation, data collection, and analysis were performed by RA and checked by MS and MJS. The first draft of the manuscript was written by RA and edited by MJS. MS edited and approved the final manuscript.

Data cannot be made publicly available; readers should contact the corresponding author for details.

The authors declare there is no conflict.