Forecast of natural aquifer discharge using a data-driven, statistical approach

Water demand in the Western United States frequently exceeds available supplies, leading to disputes over water rights. Existing knowledge shows that reliable forecasting could help manage these conflicts. However, predicting aquifer discharge months in advance remains a challenge. Previous studies have demonstrated the value of statistical models in forecasting, but few have applied them to aquifer systems. This gap motivated the development of a data-driven approach to forecast aquifer discharge. No prior work had resolved the specific timing and magnitude of discharge for the Eastern Snake Plain Aquifer. The need for predictive tools in water management is well established. This paper contributes a novel statistical model to address this issue.

The aim of this study is to forecast natural aquifer discharge months ahead of peak water demand. The focus is on the Eastern Snake Plain Aquifer in southern Idaho, where water rights conflicts are common. The specific problem is the lack of reliable forecasting tools for aquifer discharge. The motivation is to provide water managers with advanced knowledge to prepare for potential mitigation requirements. The study seeks to evaluate the effectiveness of a statistical model in predicting discharge. The timing of peak water demand is annually in July, so a four-month forecast is critical. The study also aims to identify key predictor variables that influence aquifer discharge. The outcome could support better water management decisions.

The study uses an ARIMA time-series model with exogenous predictors (ARIMAX model) to forecast aquifer discharge. The model is trained on historical data to predict discharge four months in advance. Akaike's information criterion is used to select the optimal model configuration. Predictor variables include streamflow, irrigation diversions, and storage. The model is validated using a subset of data to assess its performance. The Nash-Sutcliffe efficiency is calculated to evaluate model accuracy. Variables are chosen based on their relevance to the aquifer's water budget. The model incorporates both recharge and discharge components of the system.

The ARIMAX model successfully forecasts aquifer discharge four months ahead of peak demand. The model's performance on the validation set achieved a Nash-Sutcliffe efficiency of 0.57. Predictor variables included streamflow, two irrigation diversion variables, and storage. These variables were selected based on their relevance to the aquifer's water budget. The coefficients of variation on regression coefficients were all less than 0.5. This indicates strong predictive power for streamflow and irrigation variables. The model with the highest AIC weight included these key predictors. The results suggest that these variables are reliable indicators of aquifer discharge.

The authors propose that the ARIMAX model can provide reliable forecasts of aquifer discharge four months in advance. The model's performance on the validation set supports its use for water management planning. The inclusion of streamflow and irrigation diversion variables improves forecast accuracy. The model incorporates key components of the aquifer's water budget. The study suggests that these variables are strong predictors of discharge. The authors indicate that the model could help water managers prepare for mitigation requirements. The findings may support better decision-making in water rights conflicts. The study emphasizes the value of statistical modeling in forecasting aquifer discharge.