Demonstrating Physics Guided Machine Learning for Snow Hydrology Estimation in the Western United States

Keywords: snow hydrology, machine learning, physics-guided neural networks
Abstract Type: Virtual Poster Abstract

Abstract

Snow is a critical component of global climate regulation and provides water resources to over one billion people worldwide. Yet current measurement methods and modeling techniques lack the ability to fully capture snow characteristics such as snow water equivalent (SWE), snow depth, and density across variable landscapes. In recent years, the development of physics-informed machine learning methods has demonstrated promise for combining data-driven learning and physical information. However, this capability has not been widely explored within snow hydrology. In this work we demonstrate how machine learning architecture can incorporate knowledge of physical processes to provide more accurate estimations of snow variables. We trained a physically guided neural network (PGNN) at 49 SNOTEL locations spanning a range of snow climates in the western US using 9 years of daily data. The research addressed two questions. In the first, the performance of a PGNN was compared against a plain neural network, a high-quality physical model and a common statistical snow model. The second question investigated how regionally trained PGNNs compared to a westwide model as well as their transferability between multiple snow regions. The results showed that 1) combining physical knowledge and machine learning reduced RMSE values of SWE estimation by 35% compared to a physical model and 51% compared to a neural network, and 2) that regionalization only provided minimal benefits compared with a westwide model. These findings indicate that physically guided machine learning approaches can provide accurate information about a variety of snow characteristics.