Improving snow water equivalent maps with machine learning of snow survey and LiDAR measurements

Authors: BROXTON, P.D. - University Of Arizona; VAN LEEUWEN, W.J.D. - University Of Arizona; Biederman, Joel

Submitted to: Water Resources Research
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 4/6/2019
Publication Date: 4/18/2019
Publication URL: https://handle.nal.usda.gov/10113/6471239
Citation: Broxton, P., Van Leeuwen, W., Biederman, J.A. 2019. Improving snow water equivalent maps with machine learning of snow survey and LiDAR measurements. Water Resources Research. 55:3739-3757. https://doi.org/10.1029/2018WR024146.
DOI: https://doi.org/10.1029/2018WR024146

Interpretive Summary: In the western US, a majority of surface water originates from mountain snowmelt. Knowing the quantity of water in the snowpack, known as the snow water equivalent (SWE), is critical for water supply forecasts and management of rivers and streams for water delivery and hydropower. In this study, we develop a new method to estimate SWE by combining aerial remote sensing maps of snow depth with snow density maps generated through machine learning of hundreds of field measurements of snow density. This study finds that snow density can vary by as much 75%, highlighting the importance of considering the spatial variability of snow density when estimating SWE. In addition, we show that using spatially variable maps of snow density can impact watershed-scale SWE estimation up to 20% as compared to using snow density measurements from commonly used snow monitoring stations. The method described in this study will be useful for generating SWE estimates for water supply monitoring, evaluating snow models, and understanding how changing mountain forests might impact SWE.

Technical Abstract: In the semiarid interior western US, where a majority of surface water supply comes from mountain forests, high resolution aerial LiDAR-based surveys are commonly used to study snow. These surveys provide rich information about snow depth, but they are usually not accompanied by spatially explicit measurements of snow density, which leads to uncertainty about the estimation of snow water equivalent (SWE). In this study, we use the aerial LiDAR data along with a very large and precisely geolocated array of field measured snow depths (~4000 observations) and snow densities (~300 observations) as well as artificial neural network (ANN) machine learning of the snow density measurements to create and validate maps of snow depth, snow density, and SWE over two sites along Arizona’s Mogollon Rim in February and March, 2017. These maps show differences between mid-winter and late-winter snow conditions, and daily snow camera (a.k.a. “snowtography”) observations provide further temporal complement to the gridded data. While the spatial variability of snow density is less than that of snow depth, it can vary by as much as 75% for a single LiDAR coverage, highlighting the importance of considering its variability when estimating SWE. The gridded data show that SNOTEL and other point measurements can be unrepresentative of their surrounding environments not only in terms of SWE and snow depth, but also in terms of snow density. In particular, the LiDAR-ANN SWE estimates can be as much as 20% different than if SNOTEL density were used with LiDAR snow depths to estimate SWE.