Using SNOTEL data to run the SNOWPACK model: an analysis of model performance and water transport methods

Abstract

Snow modelling is useful for both avalanche forecasting and water supply prediction. With increasing numbers of winter recreationalists, a growing population, and more variable weather patterns due to climate change, there is a need for improved snow modelling in the United States. One example of a snow model is the SNOWPACK model. SNOWPACK was originally developed in Switzerland as an avalanche forecasting aid and was created in conjunction with a dense array of mountain weather stations. Today SNOWPACK is used, or being tested for use, as an avalanche forecasting tool in North America, Europe, and Japan. In the United States, there is also an array of winter weather stations, although these are generally used for water supply prediction. The weather stations are called SNOw TELemetry (SNOTEL) sites and are run by the Natural Resource Conservation Service's Snow Survey program. Until recently, these stations lacked the necessary inputs to run the SNOWPACK model. With the necessary data now available, we look to answer the questions: Can the SNOWPACK model be effectively run with freely available SNOTEL data, what are the differences between three water transport methods (Bucket, NIED, and Richards), and when should each be used? We develop a novel workflow to run the SNOWPACK model with SNOTEL data for 14 sites across the western United States, addressing our research questions. We find that using SNOTEL data to run the SNOWPACK model produces good agreement with observations. A mean absolute average hourly snow water equivalent error of 2.79 cm during the accumulation phase of the snow season and 6.08 cm of error during the melt season is found. We find that Richard's transport method is most appropriate when internal snow properties are required; otherwise, the Bucket method performs well. Lastly, we find an increase in model error when deeper snowpacks are modelled. In this work, we explore the effects of model parameters and input types (e.g., wind and precipitation data) on model performance. Our work provides a framework for running the SNOWPACK model with SNOTEL data given different applications.