On the reliability of seasonal snow forecasts

Abstract. Reliable information on seasonal snow conditions is important for long-range weather forecasting and climate modeling. The reliability of winter-mean hindcasts of snow water equivalent (SWE) produced by the ECMWF for the period 1993–2022 within the CopERnIcus climate change Service Evolution (CERISE) project is evaluated in this study. In probabilistic forecasting, reliability for a binary event is defined as the consistency between forecast probabilities and observed frequencies. Here, reliability is assessed using two independent SWE datasets (ERA5-Land and ESA Snow-CCI v4) across eight land regions in the Northern Hemisphere excluding mountainous regions. The reliability assessment is performed for two tercile-based binary events representing low- and high snow accumulation winters. Reliability is quantified using a weighted linear regression applied to reliability diagrams and is grouped into five categories from perfect to dangerous. The results show good reliability of the ECMWF seasonal snow hindcasts for both low- and high-snow conditions. The assessment shows sensitivity to the choice of verification dataset, with ERA5-Land yielding slightly higher reliability categories than ESA Snow-CCI. It is found that differences in hindcasts reliability between regions and between verification datasets may be linked to snow variability, model representation, and observational uncertainty.