Skill of remote sensing snow products for distributed runoff prediction

With increasing availability of remote sensing snow cover products we aim to evaluate the skill of these datasets with regard to hydrological discharge simulation. In this paper ten model variants using different snow cover data (MOD10A1, IMS, AMSR-E SWE, GLOBSNOW SWE and observed in situ snow depth) and two different model structures for snow accumulation and snowmelt switching (based on snow cover data time series or temperature time series) are calibrated with a global optimisation algorithm. The simulated discharge is subjected to five criteria for validation, while the GLUE methodology is used for uncertainty analysis of the ten model variants. The skill of the datasets is tested for the Biebrza River catchment, which has a hydrological regime dominated by snowmelt. The discharge simulations are conducted with the distributed rainfall-runoff model WetSpa. MOD10A1 was the only data source which improved the validation Nash-Sutcliffe (NS) scores in reference to a standard model. However, other evaluation measures indicate that the following data sources performed better than the standard model: MOD10A1, observed snow depth and GLOBSNOW for Kling-Gupta efficiency and for high flows; IMS and MOD10A1 for bias; GLOBSNOW and MOD10A1 for coefficient of determination. MOD10A1 has the highest spatial resolution of all analysed data sources which might contribute to the high skill of this data. The use of the data-based switching model structure generally narrowed the behavioural parameter sets during the uncertainty analysis when compared to the temperature-based switching. However, no clear relation was observed between the prediction confidence interval and the two model structures. It is concluded that the skill of the remote sensing snow cover data for the model is positive, although, strongly varying with the data source used.