Assessing the use of satellite products from multispectral unmixing and differential interferometry for Altay snow water equivalent estimation

Abstract. This work evaluates the application of fractional snow covered area (fSCA) derived from Harmonized Landsat Sentinel (HLS) multi-spectral images and snow products derived from Sentinel-1 and LuTan-1 interferometric Synthetic Aperture Radar (InSAR) pairs, in generating snow water equivalent (SWE) at the Kelan watershed in Altay, northwestern China. The newly retrieved HLS fSCA is first assimilated using a previously-developed snow reanalysis framework to generate daily SWE estimates at a spatial resolution of ~ 150 m over Water Years (WYs) 2018 – 2025. The snow reanalysis dataset is verified against an in situ SWE site, in situ snow depth sites and Unmanned Aerial Vehicle (UAV) Lidar swaths. The 8-year average peak SWE volume estimated from the fSCA constrained dataset is ~ 0.76 km³, 35% of which is distributed at elevations above 1500 m. Based on the snow reanalysis dataset, SWE increases by more than 0.02-0.03 m (corresponding to a 2π wrapping cycle in C band) over a 12-day period at over ~20% of the pixel-time samples in most of the years, likely to be affected by phase wrapping in C-band InSAR retrievals. The likelihood is significantly reduced for L-band retrievals over a 12-day measurement period. Residual phase ambiguity remains in the Sentinel-1 ΔSWE retrievals, with more than 10% of observations requiring at least one 2π cycle to match the reanalysis reference. Verification at the in situ site shows that even a single ΔSWE retrieval affected by phase wrapping can reduce the accuracy of SWE estimates using the current snow reanalysis framework. After resolving the phase wrapping in Sentinel-1 ΔSWE retrievals at the in situ SWE site, SWE generated from adjusted ΔSWE assimilation shows comparable performance to SWE from HLS fSCA assimilation in WYs 2020 and 2021. In WY 2024, however, LuTan-1 ΔSWE assimilation yields a higher unbiased root mean square difference despite negligible phase-wrapping issues, because ΔSWE retrievals are only available on low-accumulation days. The overall limited effectiveness of InSAR ΔSWE assimilation may be partly related to the scarcity of high-quality InSAR observations compared to fSCA observations, Sentinel-1’s relatively long revisit interval, and a data-assimilation approach not ideally suited to ΔSWE observations. Future work should develop assimilation methods that explicitly account for phase-cycle uncertainty, allowing a more rigorous assessment of how well C-band retrievals constrain modeled SWE. Assimilation performance should also be evaluated with high-quality retrievals from upcoming low-frequency (L- and P-band) InSAR missions.