A Machine Learning Model for Estimating Snow Density and Snow Water Equivalent from Snow Depth and Seasonal Snow Climate Classes

Abstract Direct snow water equivalent (SWE) measurements are time consuming, labor intensive, and often impractical over large areas. However, if the snowpack bulk density is known, SWE can be estimated from snow depth. Snow depth is a more accessible proxy for estimating SWE, as it can be quickly and easily measured with high precision. We propose a machine learning (ML) model that estimates snowpack bulk density from snow depth and other variables that only require the location and date of snow depth acquisitions. The estimated snow density can then be used to calculate SWE from snow depth. Our model was trained on ∼2 million data points from 691 Snowpack Telemetry (SNOTEL) stations in the western United States and evaluated on 534 818 samples from 173 independent SNOTEL stations. Results from the proposed ML model lead to a reduction in root-mean-square error (RMSE) of 24%–48% for snow density estimation and 23%–59% for SWE estimation when compared to statistical models. To evaluate the ML model’s transferability, we tested it using data from the Maine Snow Survey in the northeastern United States. The model provided promising density and SWE estimates without any retraining and showed improved performance after retraining with a small portion of local data. The ML models proposed here offer a practical and broadly applicable tool for estimating SWE in regions with limited snowpack monitoring. Significance Statement This study aims to improve snow water equivalent (SWE) estimates by developing a machine learning (ML) model that uses snow depth (HS) and minimal additional data. SWE is critical for managing water resources but is challenging to measure directly, and remote sensing techniques have yet to prove globally reliable for SWE estimation. HS, however, can be easily measured with high precision using lidar and in situ probes. We propose an ML model to estimate snow density from HS, day of water year, elevation, and snow class. The estimated snow density can be used to calculate SWE when combined with HS. This method holds potential for global applications and provides a convenient tool for SWE estimation in regions with limited snowpack monitoring.