Passive Microwave Remote Sensing based High Resolution Snow Depth Mapping for Western Himalayan Zones using Multifactor Modelling Approach

Abstract. Spatiotemporal snow depth (SD) mapping in the Indian Western Himalayan (WH) region is essential in many applications pertaining to hydrology, natural disaster management, climate, etc. In-situ techniques for SD measurement are not sufficient to represent the high spatiotemporal variability of SD in WH. Currently, low-frequency passive microwave (PMW) remote sensing-based algorithms are extensively used to monitor SD at regional and global scales. However, only a limited number of PMW SD estimation studies are carried out for WH till date. In addition, the majority of the available PMW SD models for WH locations are developed using limited data and less parameters, therefore cannot be implemented for the entire region. Further, these models have not considered the auxiliary parameters such as location, topography, snow cover days (SCD) into consideration and have poor accuracy (particularly in deep snow), and coarse spatial resolution.

Considering the high spatiotemporal variability of snow depth characteristics across WH region, region wise multifactor models are developed for the first time to estimate SD at high spatial resolution of 500 m x 500 m for three different WH zones i.e., Lower Himalayan Zone (LHZ), Middle Himalayan Zone (MHZ), and Upper Himalayan Zone (UHZ). Multifrequency brightness temperature (TB) observations from Advanced Microwave Scanning Radiometer 2 (AMSR2), SCDs data, terrain parameters (i.e., elevation, slope and ruggedness), geolocation for the winter period (October to March) during 2012–13 to 2016–17 are used for developing the SD models. Different regression approaches (i.e., linear, logarithmic, reciprocal, and power) are developed and evaluated to find if any of these models can address the heterogeneous association between SD observations and PMW TB. The results indicate the following observations: (a) multifactor model developed using power regression has shown improved accuracy in SD retrievals compared to other regression approaches in all WH zones; (b) spatial variability in SD is highly affected by SCDs, terrain parameters, geolocation parameters; (c) compared to the currently operational AMSR2 SD products, the proposed models have shown better SD estimates in all WH zones particularly when SD > 25 cm; (d) the Root Mean Square Error (RMSE) of multifactor models SD estimates increased with an increase in SCD in all WH zones; The multifactor model of MHZ has shown lesser RMSE (i.e., 27.21 cm) compared to LHZ (32.87 cm) and UHZ (42.81 cm). Overall results indicate that the proposed multifactor SD models have achieved higher accuracy in deep snowpack (i.e., SD >25 cm) of WH when compared to various previously developed SD models.