Loss-compensated snowfall scaling using large-area lidar observations improves snow depth simulations over multiple years

Abstract. In many regions, reliable runoff forecasts over spring and summer depend on accurate predictions of snowmelt volume and timing, which in turn require snow models that realistically represent peak snow amounts as well as their spatial distribution across the basin. Observation-based snowfall scaling methods can reduce biases in the snowfall input data and improve the representation of spatial variability in the modelled snowpack. These methods adjust the forcing data of snow models so that the simulations match snow depth observations, e.g. from high-resolution lidar scans. Our results show that existing snowfall scaling methods cannot reliably incorporate lidar observations over two large study regions in South Norway. In particular, existing methods fail to produce correct results in areas where snow has melted before the lidar observation. To overcome this limitation, we introduce the loss-compensated scaling method, which accounts for melt and sublimation losses that occur prior to the lidar observation. Simulations with this method reliably reproduce the assimilated lidar snow depth observations. Furthermore, we demonstrate that, with this method, information from a single lidar scan can substantially improve model performance even in years without such observations, reducing the simulated bias in snow depth by at least 67 % and the RMSE by at least 50 %. Thus, our improved method provides a reliable way to integrate lidar scans into physics-based snow models, enhancing model performance over multiple years, even in large study areas where low-lying regions start to melt before the basin reaches peak snow accumulation.