Snow water equivalent in the Sierra Nevada: Blending snow sensor observations with snowmelt model simulations

We estimate the spatial distribution of daily melt‐season snow water equivalent (SWE) over the Sierra Nevada for March to August, 2000–2012, by two methods: reconstruction by combining remotely sensed snow cover images with a spatially distributed snowmelt model and a blended method in which the reconstruction is combined with in situ snow sensor observations. We validate the methods with 17 snow surveys at six locations with spatial sampling and with the operational snow sensor network. We also compare the methods with NOAA's operational Snow Data Assimilation System (SNODAS). Mean biases of the methods compared to the snow surveys are −0.193 m (reconstruction), 0.001 m (blended), and −0.181 m (SNODAS). Corresponding root‐mean‐square errors are 0.252, 0.205, and 0.254 m. Comparison between blended and snow sensor SWE suggests that the current sensor network inadequately represents SWE in the Sierra Nevada because of the low spatial density of sensors in the lower/higher elevations. Mean correlation with streamflow in 19 Sierra Nevada watersheds is better with reconstructed SWE (r = 0.91) versus blended SWE (r = 0.81), snow sensor SWE (r = 0.85), and SNODAS SWE (r = 0.86). On the other hand, the correlation with blended SWE is generally better than with reconstructed, snow sensor, and SNODAS SWE late in the snowmelt season when snow sensors report zero SWE but snow remains in the higher elevations. Sensitivity tests indicate downwelling longwave radiation, snow albedo, forest density, and turbulent fluxes are potentially important sources of errors/uncertainties in reconstructed SWE, and domain‐mean blended SWE is relatively insensitive to the number of snow sensors blended. Key Points The blended SWE product is more accurate than the NOAA SWE product Satellite‐based snow cover depletion data improves SWE estimation Sensitivity tests suggest four key sources of uncertainties in reconstructed SWE