Snow Water Equivalent Retrieval Using Active and Passive Microwave Observations

This paper implements a newly developed combined active and passive algorithm for the retrieval of snow water equivalent (SWE) by using three‐channel active and two‐channel passive observations. First, passive microwave observations at 19 and 37 GHz are used to determine the scattering albedo of snow. An a priori scattering albedo is obtained by averaging over time series observations. Second, 13.3 GHz is introduced to formulate a three‐channel (9.6, 13.3, and 17.2 GHz) radar algorithm which reduces effects of background scattering from the snow‐soil interface, and improves SWE retrieval. In the algorithm, the bicontinuous dense media radiative transfer (DMRT‐Bic) is used to compute look‐up tables (LUTs) of both radar backscatter and radiometer brightness temperatures (TBs) of the snowpack. To accelerate the retrieval, a parameterized model is derived from LUT by regression training, which links backscatter to the scattering albedo at 9.6 GHz or 13.3 GHz and to SWE. The volume scattering of snow is obtained by subtracting the background scattering from radar observations. SWE is then retrieved through a cost function that is guided by the a priori scattering albedo obtained from the passive microwave observations. The proposed algorithm, along with the active‐only version, is evaluated against the Finnish Nordic Snow Radar Experiment (NoSREx) data set measured in 2009–2013. The combined active‐passive algorithm achieves root mean square errors (RSME) less than 27 mm and correlation coefficients above 0.68 for 2009–2010, RMSE less than 21 mm and correlation above 0.85 for 2010–2011, and RMSE less than 40 mm and correlation above 0.38 for 2012–2013. Key Points Snow water equivalent retrieval using X (9.6 GHz) and upper Ku band (17.2 GHz) radar observations is improved by adding lower Ku‐band (13.3 GHz) data Passive observations are used to obtain scattering albedos, which improves the radar retrieval algorithm performance The resulting combined active and passive algorithm is validated against the Finnish NoSREx dataset