Groundwater Modeling of the Ogallala Aquifer: Use of Machine Learning for Model Parameterization and Sustainability Assessment

Addressing groundwater depletion problems in heterogeneous aquifer systems is a challenge. The heterogeneous Ogallala Aquifer, a critical source of groundwater in the central United States, has undergone decades of decline in water levels due to pumping. This project aims to build a robust groundwater model to evaluate optimal scenarios for sustainable use of the groundwater resource within a section of the Ogallala aquifer located in the Middle Republican Natural Resources District (MRNRD). This study follows a comprehensive approach involving parameterization, construction, and optimization. The model is parametrized using hydraulic conductivity and recharge values obtained from a random forest-based machine learning model and deep neural network regressor-based model. The predicted hydraulic conductivity efficiently identifies heterogeneous layers and enhances the performance of the groundwater model. Similarly, the predicted recharge parameter identifies spatial variation and finds a diverse range of recharge values. Both parameters are incorporated into a Modflow-NWT-based groundwater model. Calibration uses Parameter ESTimation (PEST) algorithm within GMS 10.3.5 software targeting recharge, hydraulic conductivity, evapotranspiration, and riverbed conductance. Scenario analysis using the calibrated model demonstrates the sensitivity of groundwater level and stream flow to changes in pumping and recharge rates. The model is optimized using Genetic algorithms (GAs) and sensitivity analysis. The findings of the optimized aquifer system recommended a balanced approach of increasing recharge rates by 5.8% and decreasing pumping by 5.5 %. This recommendation is consistent with previous studies that emphasized the importance of enhancing recharge and reducing groundwater withdrawal. Using Genetic algorithms for optimization underscores the potential of advanced techniques in achieving balanced and sustainable groundwater management strategies. We recommend that future works focus on refining these models and exploring additional scenarios to ensure long-term groundwater sustainability in the study area. Advisor: Jesse Korus