Seasonal snow is a critical resource for society by providing water for billions, supporting agriculture, clean energy, and tourism, and is an important element within the climate system by influencing the global energy balance. However, accurately quantifying snow mass, particularly in mountainous regions, remains a challenge due to substantial observational and modeling limitations. As such, data assimilation (DA) offers a powerful solution by integrating observations with physically-based models to improve estimates of the snowpack. Previous snow DA studies have employed an Ensemble Kalman Filter (EnKF) to assimilate Sentinel-1 satellite-based snow depth retrievals, demonstrating improved accuracy in modeled snow depth, mass, and streamflow when evaluated against in-situ measurements. In those studies, the uncertainty of the assimilated retrievals was assumed to be static in time and space, likely leading to a suboptimal use of the observational information. Here, we present several advances in snow DA. Using an EnKF, we assimilate novel snow depth retrievals derived from a machine learning product that leverages Sentinel-1 backscatter observations, land cover, and topographic information over the European Alps. We also incorporate a spatiotemporally dynamic observation error, whereby the uncertainty of the assimilated snow depth retrieval varies in space and time with snow depth. The machine learning snow depth retrieval product is assimilated into the Noah-MP land surface model over the entire European Alps at 1 km resolution for the years 2015–2023 and snow depth, snow water equivalent, and snow cover are evaluated against independent in-situ data and satellite observations. This work demonstrates the benefits of machine learning based snow depth retrievals and dynamic observation errors in EnKF-based snow DA.