Quantifying Uncertainty Sources in Simulated Brightness Temperatures Generated with Snow Model Output from Season Long Snowpack Simulations Using the Noah and Noah-MP Land Surface Models

Snow is one of the world’s most important substances, yet it remains stubbornly difficult to observe and simulate. Satellite remote sensing platforms and advances in data assimilation and modeling techniques have markedly improved snowpack characterization and prediction. However there are still substantial unresolved problems. In particular, passive microwave snow water equivalent (SWE) retrievals are subject to large uncertainties due in part to the fact that they are derived from weakly constrained semi-empirical formulas, whose performance is regionally dependent. Data assimilation algorithms that ingest empirically derived snow state variables from remote-sensing platforms are exposed to this uncertainty. An increasingly popular data assimilation strategy is to use physically based radiative transfer models to produce simulated brightness temperatures using snow model inputs so that observed satellite observed brightness temperatures can be directly assimilated into numerical models, thereby bypassing the need for any derived products. However, the simulated brightness temperatures are also subject to potentially large uncertainties originating from uncertainty in a) meteorological inputs (e.g, precipitation), b) deficiencies of the snow model, and c) deficiencies of the radiative transfer.

In this study, numerical simulations are used to quantify uncertainty in simulated brightness temperatures as they relate to meteorological inputs, snow model physics, and the representation of snow within the radiative transfer model. Season long numerical simulations of a snowpack over the Red River region of North Dakota are performed. Two different land surface models: Noah and Noah-MP, and two different meteorological forcings: NLDAS and WRF output, are compared. The dense media radiative transfer multilayer model (DMRT-ML) is used to simulate the brightness temperatures. To investigate the impacts of how snow is represented in the DMRT-ML, a simple offline diagnostic multi-layer snow model is used to simulate the snow stratification and microstructure evolution within Noah and Noah-MP. The results are evaluated against AMSR-E satellite data, at 19 and 37 gHz bands, and the key sources of uncertainty are discussed.