A processing–modeling routine to use SNOTEL hourly data in snowpack dynamic models

Abstract

SNOTEL hourly and daily data are a strategic information about snowpack dynamics in western United States. Hourly data are highly noisy due to, e.g., non-physical temperature-based fluctuations of the signal or gauge under-catch. Noise may hinder, among other factors, the correct evaluation of precipitation events or the measurement of SWE, hence the reconstruction of accumulation and melt run-off timing. This makes hourly data practically useless without a denoising procedure. As this time resolution is widely used in hydrologic applications, here we test SNOTEL hourly data in modeling snowpack dynamics. A one-dimensional model of snow depth, snow water equivalent and bulk snow density has been adopted to this aim. We define an automated processing routine to denoise data-series of snow depth, snow water equivalent, bulk snow density, liquid and solid precipitation. Special attention is paid into distinguishing the different types of precipitation, and processing snow depth data. Since sub-daily physical oscillations in snow depth data are difficult to be separated from instrument noise, a joint processing–modeling procedure has been designed. Forty SNOTEL sites throughout the western United States with multi-year data are considered for testing the procedure. The analysis shows that the model performance, expressed in terms of median values of the Nash–Sutcliffe coefficient, are higher than 0.8 for all the three variables, provided the first year of each dataset is used in the calibration phase.