Limitations on Seasonal Snowmelt Forecast Accuracy

Abstract

Forecasting of seasonal snowmelt runoff is an important exception to the otherwise limited accuracy of long‐term runoff forecasts. Conceptual models of the snow accumulation and ablation and precipitation‐runoff processes have been advocated as a means of introducing physical understanding of the dominant hydrologic processes into seasonal snowmelt runoff processes. However, the paucity of information and extreme spatial variability of meteorological processes and watershed characteristics in mountainous watersheds often result in large simulation errors. The analysis reported here relates monthly simulation error statistics to maximum achievable seasonal runoff forecast accuracy, presuming perfect knowledge of future (gage) precipitation and temperature, but without updating of model parameters or states. In some cases, this accuracy limit is less than that actually achievable using simpler models. The results suggest a preference for simple models if the coefficient of variation of monthly simulation error cannot be reduced below about 10–15%.