A model for the spatial distribution of snow water equivalent parameterized from the spatial variability of precipitation
- 1Norwegian Water Resources and energy Directorate, P.O. Box 5091, Maj. 0301 Oslo, Norway
- 2Department of Geosciences, University of Oslo, Oslo, Norway
- anow at: Norconsult AS, P.O. Box 626, 1303, Sandvika, Norway
Abstract. Snow is an important and complicated element in hydrological modelling. The traditional catchment hydrological model with its many free calibration parameters, also in snow sub-models, is not a well-suited tool for predicting conditions for which it has not been calibrated. Such conditions include prediction in ungauged basins and assessing hydrological effects of climate change. In this study, a new model for the spatial distribution of snow water equivalent (SWE), parameterized solely from observed spatial variability of precipitation, is compared with the current snow distribution model used in the operational flood forecasting models in Norway. The former model uses a dynamic gamma distribution and is called Snow Distribution_Gamma, (SD_G), whereas the latter model has a fixed, calibrated coefficient of variation, which parameterizes a log-normal model for snow distribution and is called Snow Distribution_Log-Normal (SD_LN). The two models are implemented in the parameter parsimonious rainfall–runoff model Distance Distribution Dynamics (DDD), and their capability for predicting runoff, SWE and snow-covered area (SCA) is tested and compared for 71 Norwegian catchments. The calibration period is 1985–2000 and validation period is 2000–2014. Results show that SD_G better simulates SCA when compared with MODIS satellite-derived snow cover. In addition, SWE is simulated more realistically in that seasonal snow is melted out and the building up of "snow towers" and giving spurious positive trends in SWE, typical for SD_LN, is prevented. The precision of runoff simulations using SD_G is slightly inferior, with a reduction in Nash–Sutcliffe and Kling–Gupta efficiency criterion of 0.01, but it is shown that the high precision in runoff prediction using SD_LN is accompanied with erroneous simulations of SWE.