Simultaneous optimization of snow and soil sensor locations using machine learning in East River, Colorado

Abstract:

Designing optimal sensing and sampling strategies for snow and soil moisture is critical for developing a predictive understanding of hydrologic processes in montane catchments, where complex terrain and spatial variability of vegetation and soil properties lead to significant spatial and temporal variability. High-resolution measurements of terrain and soil properties (e.g., airborne LIDAR data) enable increasingly granular approaches to optimal monitoring of montane catchments. We developed a multi-step machine learning-based approach that employs LIDAR data to simultaneously optimize sensor placements for both snow and soil moisture in montane catchments. We used data from the NASA Airborne Snow Observatory (ASO) for snow data, and satellite images (Landsat, Planet) for plant phenology. This work extends a previous machine learning-based approach to optimize sensor placements for monitoring snow depth in montane catchments by including additional variables (snow depth, model-simulated soil moisture distributions, surface temperature) to find optimal locations. We evaluate the approach at a watershed in the East River, Colorado.