A Bayesian approach to snow water equivalent reconstruction

Snow water equivalent (SWE) reconstruction methods have previously been used to characterize seasonal SWE accumulation using mass and energy balance models and time series of remotely sensed snow-covered areas (SCA). Recognizing that the spatial signature of the seasonal SWE accumulation is an integration of a series of snowfall events, we have formulated a Bayesian SWE reconstruction that combines time series of remote sensing estimates of SCA with a land surface model to estimate storm-specific snowfall distribution with a retrospective data assimilation scheme. The analysis is identical in form to the ensemble Kalman filter or ensemble Kalman smoother. A snow depletion curve is used to relate SCA and SWE. We perform a series of synthetic tests to assess how much information concerning snowfall accumulation patterns can be extracted from a time series of SCA measurements taken during the ablation season. The posterior estimate of SWE calculated via our method effectively recovers the true SWE: the mean and standard deviation of the SWE estimate error improve by 86% and by 78%, respectively, over the prior for pixels with vegetation fraction less than 90%. The sensitivity of the method to climatic and physiographic variables and input errors is investigated. The technique shows promise for future work in characterizing spatial patterns of snowfall over mountainous regions.