Collaborative Research: Quantifying Watershed Dynamics in Snow-Dominated Mountainous Karst Watersheds Using Hybrid Physically Based and Deep Learning Models

Karst aquifers form in regions underlain by highly soluble rock formations, such as limestone, and serve as the primary drinking water source for about a quarter of the world's population. These aquifers are characterized by complex groundwater recharge, storage, and flow patterns in sinkholes, pores, fractures, and conduits. In many mountainous areas of the western U.S. and worldwide that host karst aquifers, most of the annual precipitation falls in the winter as snow. In these snow-dominated karst watersheds, snowmelt recharges aquifers that sustain streamflow in summer when precipitation is scarce and water demand is high. These watersheds are sensitive to year-to-year variations and long-term trends in precipitation and temperature. This creates challenges for sustainable water resource management particularly when a quantitative understanding of mountainous karst watershed response to climate variability is lacking. Such knowledge gaps exist due to complex recharge and discharge processes that occur because of topographical and geological heterogeneities inherent in these watersheds. This project will overcome these limitations and provide a sound scientific basis for improved water resources management. Funding will support both graduate and undergraduate research at multiple universities. Through outreach and educational activities, the project will also engage local stakeholders, the general public, and K-12 students.

The overarching goal of the proposed research is to understand and predict hydrologic responses of snow-dominated mountainous karst aquifers. The three-year project will integrate a spatially distributed, physically based snowmelt model with a data-driven, deep learning model that represents the highly complex karst aquifer system. Field observational and geochemical data sets (including streamflow, and ions and isotopes in stream and spring water) will be collected at various spatial and time scales to identify recharge and discharge characteristics, while also testing the predictive capability and physical representativeness of the deep learning model. Specifically, the project will (1) quantify the spatiotemporal groundwater discharge and streamflow response to snowmelt/rainfall events with varying intensity and duration, (2) determine how interannual climate variability and watershed physical properties influence hydrologic behavior, and (3) test the combined physically based and data-driven modeling approach in different locations and climate conditions. The outcomes will lead to improved understanding of how snow-dominated mountainous karst watersheds respond to climate variability and provide insight into the robustness of the modeling approach for forecasting or transferability to other regions.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.