Dynamic modeling of multi-layered unconfined and confined aquifers in the presence of climate change

The Emilia-Romagna region (Italy) hosts extensive agricultural and industrial activity, and densely populated urban centers. Groundwater largely supplies the resulting high water demand, especially during droughts, whose frequency and intensity is expected to increase. This study estimates the evolution of groundwater conditions in part of Emilia-Romagna, considering climate change and human impacts. The goal is to evaluate the resilience of the regional multi-layered aquifer system to prolonged droughts, and outline potential guidelines for long-term sustainable groundwater management. A numerical groundwater flow model and a random forest algorithm are implemented to compare the performance of a physics-based and a machine learning model in simulating historical and future groundwater levels, and to explore the benefits of their combination. The groundwater model is developed in MODFLOW 6 and the random forest algorithm in R. Input data come from a MODFLOW model covering the entire Emilia-Romagna groundwater system by Arpae (Regional Agency for Prevention, Environment and Energy of Emilia-Romagna) and publicly accessible datasets on the Emilia-Romagna Region and Arpae repositories. Both methods are then applied to analyze scenarios under reduced precipitation and altered pumping, focusing on their combined effects on the regional aquifer system. Results show the aquifer system’s vulnerability to potential future droughts. Increased pumping amplifies precipitation reduction effects, while reduced abstraction can partly mitigate them. Critical hotspots are identified, emphasizing the importance of considering different spatial scales to develop effective mitigation and adaptation strategies. The random forest algorithm provides valuable insights into factors influencing groundwater head distribution, enhancing the groundwater model results interpretation and potential improvement. However, its lack of physical grounding limits its generalization capability. These findings highlight the value of integrating physics-based and machine learning techniques for improving their performance, making a significant contribution to groundwater modelling, which will become increasingly important in the future.