Adaptive water management in the face of uncertainty: Integrating machine learning, groundwater modeling and robust decision making

Abstract

This study examines climate change and water resources management challenges facing water suppliers in drought-prone regions and is particularly relevant to the American West, where agencies balance the management of imported and local water resources across multiple future uncertainties. We apply Robust Decision Making (RDM) to water management planning challenges facing the San Bernardino Valley Municipal Water District (Valley District) and investigate the performance of a machine learning-based representation of two local groundwater basins. To do so, we assess three machine learning methods--Random Forest (RF), Support Vector Machines (SVM) and Artificial Neural Networks (ANN)--and their ability to simulate the output of a high-resolution MODFLOW model. We find that RF produces the most accurate results, and thus we incorporate the RF version of the MODFLOW model into the study’s RDM approach. This constitutes an advancement to the field of decisionmaking under deep uncertainty (DMDU) through a novel application of machine learning that shortens modeling run times and allows for a greater exploration of the uncertainty space, including a broad range of future climate changes and drought conditions. This paper also constitutes an advancement to the field of empirical groundwater modeling by showing that RF is capable of simulating average basin groundwater level changes. Our results also suggest that demand management can significantly reduce vulnerabilities to drought and other climate changes, and we provide recommendations on additional adaptive management options and key signposts to track for the Valley District.