Inverse modelling of snow depths

Abstract

Operational snow forecasting models contain parameters for which site-specific values are often unknown. As an improvement a Bayesian procedure is suggested that estimates, from past observations, site-specific parameters with confidence intervals. It turned out that simultaneous estimation of all parameters was most accurate. From 2.5 years of daily snow depth observations the estimates were for snow albedo 0.94, 0.89, and 0.56, for snow emissivity 0.88, 0.92, and 0.99, and for snow density (g/cm³) 0.14, 0.05, and 0.11 at the German weather stations Wasserkuppe, Erfurt-Weimar, and Artern, respectively. Using estimated site-specific parameters, ex post snow depth forecasts achieved an index of agreement IA = 0.4–0.8 with past observations; IA = 0.3–0.8 for a 51-years period. They outperformed the precision of predictions based on default parameter values (0.1 < IA<0.3). The developed inverse approach is recommended for parameter estimation and snow forecasting at sub-alpine stations with more or less urban impact and for application in education.