A socio-hydrological framework for assessing groundwater sustainability in an irrigated agricultural landscape.

In my research, I use a socio-hydrological framework to examine groundwater and how people interact with it and make decisions related to it. In theory, socio-hydrological modeling frameworks should assess how humans and groundwater co-evolve, interact, and respond under different circumstances and challenges. In practice, however, most modeling frameworks fail to adequately represent the intertwined groundwater and human systems, especially human behavior due to the use of over-simplified models.  

To address this gap, my research combines both natural science and social science methods. I do this by integrating hydrologic models with an agent-based model. As shown in the figure below, it features a social system model (Agent-based Model) for characterizing behaviors of groundwater irrigation and coupled watershed-aquifer models (Soil and Water Assessment Tool (SWAT) and MODFLOW) for modeling surface water and groundwater interactions. The construction of the ABM shall be informed by interviews and surveys to adequately capture and represent human decision-making. 

The Soil and Water Assessment Tool (SWAT)

This framework accounts for the feedback and dynamics within and between the human and the water systems. It also makes it possible to make projections and reconstructions to test different ‘what if’ scenarios. One section of this framework was utilized to study the impacts of cover crops and climate change on water dynamics in our study area. The objective of the study was to improve the understanding of the long-term impacts of cover crops in combination with climate change on agricultural water management in an agricultural system following a hydrological modeling framework. To achieve this objective, the Soil and Water Assessment Tool (SWAT) modeling and water footprint analyses were conducted. 

The SWAT model was developed for the Central Platte River Basin for the purposes of simulating and evaluating the hydrological system in terms of both water quantity and quality. The Soil and Water Assessment Tool (SWAT) model (Version 2012) was used to cover the 27-year period from 1989 to 2015. Key inputs to the SWAT model included weather (i.e., precipitation, solar radiation, temperature, relative humidity, and wind speed) data, topography, soils data, and land cover (Table 1), as well as typical land management practices in the region. All the inputs were downloaded from either state, national or academic agencies. The coordinate system for all the input data was set to North American Datum 1983 (NAD83) Universal Transverse Mercator Zone 14 N.

Table 1: Data and sources

The SWAT model requires a wide range of inputs including spatial data and weather data. Table 1 shows the provided data and respective sources for the case study. It includes climate, agriculture management, as well as remotely sensed spatial (e.g. DEM, soil and land use) information.

In this study, the management practices selected to represent the agricultural conditions were fertilizer application, irrigation, planting operation, harvest and kill operations as well as the incorporation of rye after cash crop (corn and soybean) harvest. Planting and harvesting dates were obtained from literature and communication with local experts. Irrigation and fertilizer application were set up using the automatic option. Therefore, irrigation events were triggered by water stress threshold (soil moisture deficit) and fertilizer was applied based on nutrients stress factor. Auto fertilizer and auto irrigation options of the SWAT model were used based on the assumption that farmers were using good management practices.

Table 2: SWAT model cover crop scenarios

This study assessed 38 General Circulation Models (GCMs) from the Fifth Coupled Model Intercomparison Project (CMIP5; Taylor et al. 2012). This represented the set of GCMs that were available, contained daily data from 1950 to 2100, and included historical and future outputs for both the moderate and high representative concentration pathways (RCP4.5 and RCP8.5, respectively; Moss et al. 2010; Knutti 2014). GCM projections of temperature change were calculated as absolute differences (future - current) while projections of future precipitation are calculated as relative changes of the current period (future/current * 100%). The annual precipitation and temperature changes were calculated for the period between 2030−2050, referenced to 1950−1980.

Table 3: Selected representative GCMs

The SWAT modeling framework was utilized in two studies to help understand how water resources respond to different changes in the natural system. To represent the future climate, downscaled data from five Global Climate Models (GCMs) were utilized in this study. Broadly, we examined the following:

1. Sustainable cropping practices and their impacts on the water balance

The purpose of this study was to use a hydrological modeling framework (Figure 2) to evaluate the long-term impacts of cover crops on water availability under different climate scenarios.

We assessed the impact of introduced winter cover crops on the water flows and fluxes in the watershed under study. We hypothesized that the introduction of an over the winter cover crop with spring termination should influence these variables than the scenario without a cover crop. To address this objective, the SWAT model was set up to assess the effects of cover crops and climate variability on the water flows and fluxes in a regional agricultural watershed by assessing how different water balance components were impacted (Figure 3)

2. Agricultural water management (AWM) and how it affects the resilience of watersheds

Given the knowledge gaps in cover crops and Agricultural Water Management (AWM), it is vital to thoroughly assess the long-term impacts of cover crop installations on water cycles associated with AWM. This study aimed primarily to investigate the impacts of installing cover crops for agricultural systems and their impacts on the resilience of such a system. We examined one of the most utilized cover crop species, Cereal rye (Secale cereale), introduced during the winter period following the harvest of the main crops. This study used hydrological model simulations and water footprint accounting framework  (Figure 4) to investigate the effects of using cover crops as an AWM strategy and rebound effects on water resources sustainability. A combination of the comprehensive simulation using the SWAT model and water footprint sustainability indices was employed in an agricultural watershed (Middle Platte watershed) to perform the evaluations. 

The SWAT model outputs were utilized to assess the impacts of cover crops and climate change on water dynamics. The figures below illustrate the components used in our analysis as well as the procedure:

Groundwater Model - Cooperative Hydrology Study (COHYST) 

This study developed Python-based FloPy scripts to interact with the COHYST MODFLOW groundwater model. This involved several steps to ensure compatibility and accuracy when handling input and output files. The COHYST model, critical for groundwater management in Nebraska, uses the USGS MODFLOW software, which relies on formatted input and output files structured using Fortran-type format descriptors.

To access and utilize the model effectively, FloPy, an open-source Python library for MODFLOW, was employed. The COHYST MODFLOW model input files were loaded into FloPy using functions such as flopy.modflow.Modflow() to read and interpret the model's configuration. This step required careful handling of directory structures and ensuring all required model files (e.g., NAM, DIS, BAS, and NWT files) were accessible.

MODFLOW's reliance on Fortran-type format descriptors for interpreting input/output files necessitated resetting the format in certain cases. Using FloPy's ArrayOutput and related tools, modifications were made to ensure proper alignment of data read/write operations, adhering to the structured formatting standards described in MODFLOW documentation. This step addressed common challenges, such as incorrect alignment of numerical data or truncation issues during file reading or writing.

To ensure the COHYST model was accurately replicated and operational within the FloPy framework, debugging was conducted using FloPy's plotting and visualization tools. Once the COHYST model was successfully integrated, FloPy was used for sensitivity analyses and scenario testing. Test simulations with varying input parameters (e.g., recharge rates, pumping rates) were run, leveraging the COHYST model's calibrated structure.

This approach provided a flexible and efficient means of interacting with the COHYST MODFLOW model while maintaining the integrity of its original Fortran-based data structures. The use of FloPy facilitated visualization and analysis of results, enhancing the model's applicability for groundwater resource management in the region.

Social System Model - Agent-based Modeling in Mesa

The development of the social system model involved a number of steps. To develop the Agent-Based Model (ABM) that represents local irrigation decision-making, we designed and conducted interviews and mail surveys following a Theory of Planned Behavior (TPB) framework. TPB, developed by Ajzen (1991) poses that the intention to perform is dependent upon three constructs: attitude, subjective norms, and perceived behavioral control. The theory’s concise and efficient structure makes it ideal for analyzing decision-making behaviors from information collected from the study area through interactions with stakeholders via interviews and surveys.

A sample (200) of the research subjects was approached for interviews during the development of the survey instrument. Based on their insights, the survey was developed, revised and updated in preparation for the first round of mailing. We used our data and relevant literature (e.g., irrigation pumping records, published survey results) to refine behavioral rules of irrigators and the causal relations that lead to irrigation decisions, which was ultimately used to inform the design of ABM.

Data collection techniques and study population

Our CPNRD study has been approved by the Southern Illinois University - Carbondale Human Subjects Committee (HSC). We used an exploratory, sequential, mixed-methods approach for optimum analysis as well as robust understanding of issues. Therefore, while quantitative data collection and analysis was prioritized, qualitative methods were also used to shape the questions and statements in the final instruments (mail survey and interview script). This approach followed the Ajzen, (2006) recommendations to collect qualitative data about underlying beliefs before developing the final survey. This aids in expanding the understanding of what considerations involving irrigation decision-making are of importance to the target audience. This information was then used to develop the quantitative instrument, a mail survey that was pre-tested. In the subsequent stage of data collection, the survey was designed such that it would help expand the understanding of intentions of conserving groundwater by sampling a large number of people in a statistically representative sample.

Data was collected in several phases; the first qualitative phase of the study aimed to help us garner beliefs regarding groundwater use and conservation among a small sample of farmers in the study area. 200 farmers were randomly selected and invited to participate in interviews/discussions. An interview guide with standardized open-ended questions was utilized to prompt behavioral, normative, and control beliefs related to irrigation decision-making. During this stage, the survey was also pre-tested to determine whether additional variables could be identified that were absent in the initial survey instrument. The outcome of these interviews was used to inform the development of a survey instrument for use in the second phase of the study. The survey instrument designed in accordance with the TPB was used to quantitatively evaluate factors that influence farmers’ intentions to engage in water saving behavior in a larger sample of farmers. The survey also contained demographic questions to provide context for the result analysis.

The second phase included a mail survey of farmers in CPNRD basin and follow-up interviews with willing participants. This research was designed and implemented as a cross-sectional survey to collect data from a sample of the population in the study area. We surveyed farmers during the spring-summer of 2023 in selected counties within the Central Nebraska region (McPherson, Lincoln, Logan, Custer, Dawson, Gopher, Frontier, Phelps, Kearney, Adams, Hamilton, Buffalo, Nance, Platte, Butler, Polk, Hall, Howard, and Merrick). A sample of 1,000 addresses was purchased from a reputable company that specializes in farmers and ranchers’ databases in the US. Of this population, a simple random sample of 800 individuals were selected and addresses were confirmed. To administer the mailed survey, a modified Dillman et al. (2014) tailored design method was used where a pre-notice flyer was shared via social media and contacts in the study area in mid-April 2023. This was followed by a survey packet at the end of April 2023, followed by a reminder postcard in mid-May 2023, with replacement survey packets whenever necessary. In order to encourage participation and improve response rate, this study utilized gift card incentives for participants.

Theoretical framework

The survey design was informed by the Theory of Planned Behavior (TPB) developed in 1991 by Ajzen as a successor to the Theory of Reasoned Action (Fishbein and Ajzen, 1975). TPB has been widely applied to investigate complex relationships between intentions and behaviors/actions. According to TPB, the immediate precursor of action is the intention to act, and the strength of the intention determines the likeliness to perform a behavior, or the willingness to try and exert effort towards the task at hand (Ajzen, 1991; Suh and Hsieh, 2016).

This study incorporated moral norms to an extended TPB. A moral norm is something that mediates a person's attitude before it affects his intention.  This approach was taken in line with Kaiser and Scheuthle (2003) findings where a moral norm was found to be a supplementary predictor of a person's intention to act in a conservational manner (after attitudes, subjective norms, and perceived behavioral control). Additionally, risk perception was also included since both environmental policy and the perceived acceptability of environmental policy are to a large extent based on the perceptions and judgments of risk of various groups of both experts and laypersons (Douglas and Wildavsky, 1982; Roe, 1996).

Survey questions/questionnaire design

To develop the survey, we created questions related to general farm information, basic farmer characteristics, irrigation and conservation practice, moral norms and risk perception. Farmers were asked about irrigation methods, timing, amounts and water sources, decisions on irrigation as well as their opinions on issues pertaining to groundwater availability and use and perceptions on changing climate. In addition, factors that could impact groundwater conservation (e.g., groundwater availability, behavioral norms, social norms, perceived behavioral control, risk perception, moral norms, etc.) were listed in Likert type survey questions with option to choose a level of agreement ranging between strongly agree and strongly disagree or extremely likely to extremely unlikely. Besides these questions, information on factors influencing decision-making was sought to assess the motivation and intention of farmers towards groundwater conservation. Questions including aforementioned factors and additional questions were mainly developed based on reviewed literature, discussions with agriculture extension staff and farmers. Overall, the survey contained 37 questions across 12 pages (Survey questionnaire available in supplementary information).

Limitations

The following limitations were identified for this study:

1. Despite numerous efforts to encourage participation in the study, the survey garnered a low response rate. We tried to remedy this by having an additional round of surveys shared during a harvest day celebration in Fall 2024
2. To obtain producers’ mailing addresses, we procured a list from a marketing database service. Therefore, the study is limited by the use of a convenience sample.
3. The study relied on self-reported data, which may be influenced by individual response bias, participant understanding or interpretation of questions, and introspective ability (Hoskin, 2012).

Data Exploration with PCA (Principal Component Analysis) and Cluster Analysis (CA)

PCA is a dimensionality reduction technique that simplifies large datasets by identifying the most important variables (principal components). It's widely used for visualizing and understanding complex data. High-dimensional data, like survey data, can be challenging to interpret. PCA reduces this complexity, making it easier to spot patterns, clusters, or outliers.

In this project, PCA was utilized for preprocessing the data because it provides a clear visual representation of sample variability, helping identify batch effects or experimental differences. We paired PCA with the k-means clustering technique for deeper insights into our data's structure. We utilized Python's scikit-learn to make PCA implementation simple.

Factors that were constructed based on PCA results were rotated using the varimax method for easier interpretation. Only the factors exceeding an eigenvalue of one were retained for cluster analysis. Cluster analysis helped us to group farmers based on our data according to their (dis)similarity in terms of their attributes represented by selected variables. Cases within a certain cluster should be similar to each other, and cases belonging to different clusters should be dissimilar. The K-means clustering method was utilized in this study. The result was four clusters.

In total, 54 variables were included in the PCA, of which ten principal components with eigenvalues >1 were identified for further analysis. Then, 10 components explained >70 % of the total variability in the dataset were selected. Each component was evaluated to help develop descriptive labels that characterize them according to the variables with which they were most strongly associated.