Deriving Snow Depth from ICESat-2 Lidar Multiple Scattering Measurements: Understanding the Performance

The first two parts of this series of studies suggest that snow depth can be derived from the first-, second-, and third-order moments of the lidar backscattering pathlength distribution through the application of diffusion theory and Monte Carlo lidar radiative transfer simulations (Part I, Hu et al. 2022). The retrieval of snow depth is then applied to the ICESat-2 lidar measurements over Arctic sea ice and land surfaces in the Northern Hemisphere. The ground track measurements of snow depth overall showed consistent results over the arctic when co-located with IceBridge airborne snow radar and over terrestrial land when compared to University of Arizona (UA) and Canadian Meteorological Centre (CMC) gridded daily snow products (Part II, Lu et al. 2022). However, examining individual tracks reveals large differences between UA and ICESat-2 snow products, particularly over regions with complex terrain.

To further understand these differences and the performance of the retrieval over a domain in mountainous area in the western U.S. to represent complex terrain during the snow season (December-April). Next, we use a variety of terrestrial datasets to investigate differences between the ICESat-2 derived snow depth and the in situ measurement – derived UA product, including co-located slope and elevation data from the USGS LANDFIRE dataset, MODIS LAI and Vegetation metrics, GEDI L3B canopy heights, and ICESat-2’s ATL08 vegetation metrics. Additionally, we evaluate the performance based on the time in the snow season and in terms of snow density. Results suggest that the decrease in the performance of the retrieval for ICESat-2 snow depth was strongest in areas of higher elevation, slope, and canopy heights. In terms of seasonality, the retrieval performs well early in the snow season but degrades late into the snow season with a tendency to underestimate snow depth. Similar conclusions can be drawn in terms of snow density where larger snow densities resulted in a larger bias in the retrieval. Our results continue to demonstrate the reliability of the ICESat-2 snow depth retrieval and its overall global applicability that was shown in Parts 1 and II of study, but also suggest there are potential weaknesses in the retrieval that we identified and can further improve on in future studies.