Object-based ensemble estimation of snow depth and snow water equivalent over multiple months in Sodankylä, Finland

Abstract. Snowpack characteristics such as snow depth and snow water equivalent (SWE) are widely studied in regions prone to heavy snowfall and long winters. These features are measured in the field via manual or automated observations and over larger spatial scales with stand-alone remote sensing methods. However, individually these methods may struggle with accurately assessing snow depth and SWE in local spatial scales of several square kilometers. One method for leveraging the benefits of each individual dataset is to link field-based observations with high-resolution remote sensing imagery and then employ machine learning techniques to estimate snow depth and SWE across a broader geographic region. Here, we combined field-based repeat snow depth and SWE measurements over six instances from December 2022 to April 2023 in Sodankylä, Finland with Light Detection and Ranging (LiDAR) and WorldView-2 (WV-2) data to estimate snow depth, SWE, and snow density over a 10 km2 local scale study area. This was achieved with an object-based machine learning ensemble approach by first upscaling more numerous snow depth field data and then utilizing the estimated local scale snow depth to aid in estimating SWE over the study area. Snow density was then calculated from snow depth and SWE estimates. Snow depth peaked in March, SWE shortly after in early April, and snow density at the end of April. The ensemble-based approach had encouraging success with upscaling snow depth and SWE. Associations were also identified with carbon- and mineral-based forest surface soils, alongside dry and wet peatbogs.