Using a linear unmixing method to improve passive microwave snow depth retrieval

Abstract

Microwave radiometer measurements are used to extract snowpack information. However, the coarse spatial resolution of satellite brightness temperature observations hinders snow depth retrieval algorithm development and other applications. Measurements therefore require better spatial resolution to improve snow depth retrieval with enhanced resolution and accuracy. In this paper, a linear unmixing method is presented to downscale brightness temperatures from FY-3B MWRI data to improve snow depth retrieval. Contributions to brightness temperatures originating from different land surfaces can be identified with high-resolution land-cover images, land surface temperature products, and an antenna gain function. This produces an underdetermined equation set that can be used in an over-specified system, assuming uniform emissivity for each land-cover type within a small localized region. The over-specified equation set then can be solved using a constrained linear least-square method. Finally, brightness temperatures of each land-cover type derived from contaminated mixed pixels are applied to estimate snow depth. Simulation results of three numerical experiments validate that the unmixing algorithm is capable of separating the signals of land-cover types from mixed pixels. The unmixing method is then applied to FY3B-MWRI measurements for snow depth retrieval in Xinjiang Province. The resulting snow depths are compared to weather station observations from January to February 2011. The results show that the snow depth predictions from downscaling brightness temperatures perform better than the original, with higher correlation and lower root mean square errors. This work is an important contribution to accurate snow depth retrieval and benefits future research, with applications in hydrology, meteorology and snow disaster prevention.