OPERATIONAL MONITORING OF SNOW WATER EQUIVALENT IN THE WESTERN U.S. USING MACHINE LEARNING AND REMOTE SENSING

In this study, our group has successfully implemented a comprehensive approach to operationally monitor Snow Water Equivalent (SWE) in the western United States using advanced machine learning and remote sensing techniques. Leveraging the ExtraTreeRegressor model, we integrated multi-source data, including satellite imagery, ground-based measurements, and meteorological data, to generate high-resolution SWE maps. Our methodology incorporates various state-of-the-art techniques to enhance prediction accuracy and reliability. The ExtraTreeRegressor model was selected for its superior ability to handle high-dimensional data and its robustness against overfitting, making it particularly well-suited for the complex and variable nature of snowpack data. Compared to traditional SWE models, our AI-driven approach offers several advantages: improved accuracy due to better capture of non-linear relationships, enhanced generalization across diverse climatic and topographical conditions, and effective integration of heterogeneous data sources.

Advanced technologies and algorithms underpinning our work include the use of satellite data (e.g., MODIS) and meteorology model data like gridMET for continuous and extensive spatial coverage, sophisticated feature engineering incorporating meteorological and topographical features, and the ensemble learning approach of the ExtraTreeRegressor model for reduced variance and improved prediction stability. Cross validation using independent datasets ensures the reliability and robustness of our data products. The resulting SWE maps have been published and are accessible at http://geobrain.csiss.gmu.edu/swe_site/. All data products have undergone verification, demonstrating the robustness and reliability of our approach. The results are very promising, showcasing the potential of our model to provide accurate and timely SWE estimations for effective water resource management and environmental monitoring. Future work includes further refining the maps by removing non-snow pixels, enhancing our monitoring spatial extent, and increasing sampling points in wetland and water body areas to improve overall accuracy in non-mountain regions.