Authors: Dane Liljestrand – University of Utah
Title: Snow-Probe Measurements, LiDAR, and Machine Learning for Modeling Snow Distribution in Complex Terrain
Presentation Type: Lightning Talk
Abstract: Seasonal snowpack in montane watersheds, particularly in the Western United States, is a key component of downstream water supply and streamflow, though the accurate representation of such areas is an ongoing effort. We have developed a multi-step, Gaussian-based machine-learning model framework that combines optimized snow-probe depth data with static LiDAR terrain features to estimate catchment-scale snow distribution. The framework guides snow samplers to optimal sampling locations that most physiographically represent the study region where point measurements are largely informative of the broader snow distribution. Using just 10 observations from a small sub-catchment, the model accurately estimates snow depth distribution across a larger region when compared with LiDAR measurements. Model results show performance resilience across sub-catchments of varying spatial, topographical, and vegetative characteristics, with no permanent or costly instrumentation. This method exhibits an approach to optimizing in-situ snow monitoring networks as well as a potential for citizen-scientist data to efficiently provide seamless modeled snow depth across different spatial ranges in snow-covered regions.