Forcing the SURFEX/Crocus snow model with combined hourly meteorological forecasts and gridded observations in southern Norway

Abstract

In Norway, 30% of the annual precipitation falls as snow. Knowledge of the snow reservoir is therefore important for energy production and water resource management. The land surface model SURFEX with the detailed snowpack scheme Crocus (SURFEX/Crocus) has been run with a grid spacing of 1km over an area in southern Norway for 2 years (1 September 2014-31 August 2016). Experiments were carried out using two different forcing data sets: (1) hourly forecasts from the operational weather forecast model AROME MetCoOp (2.5km grid spacing) including post-processed temperature (500m grid spacing) and wind, and (2) gridded hourly observations of temperature and precipitation (1km grid spacing) combined with meteorological forecasts from AROME MetCoOp for the remaining weather variables required by SURFEX/Crocus. We present an evaluation of the modelled snow depth and snow cover in comparison to 30 point observations of snow depth and MODIS satellite images of the snow-covered area. The evaluation focuses on snow accumulation and snowmelt. Both experiments are capable of simulating the snowpack over the two winter seasons, but there is an overestimation of snow depth when using meteorological forecasts from AROME MetCoOp (bias of 20cm and RMSE of 56cm), although the snow-covered area in the melt season is better represented by this experiment. The errors, when using AROME MetCoOp as forcing, accumulate over the snow season. When using gridded observations, the simulation of snow depth is significantly improved (the bias for this experiment is 7cm and RMSE 28cm), but the spatial snow cover distribution is not well captured during the melting season. Underestimation of snow depth at high elevations (due to the low elevation bias in the gridded observation data set) likely causes the snow cover to decrease too soon during the melt season, leading to unrealistically little snow by the end of the season. Our results show that forcing data consisting of post-processed NWP data (observations assimilated into the raw NWP weather predictions) are most promising for snow simulations, when larger regions are evaluated. Post-processed NWP data provide a more representative spatial representation for both high mountains and lowlands, compared to interpolated observations. There is, however, an underestimation of snow ablation in both experiments. This is generally due to the absence of wind-induced erosion of snow in the SURFEX/Crocus model, underestimated snowmelt and biases in the forcing data.