Interannual stability of grid cell snow depletion curves as estimated from MODIS images

The effect of calibrating spatially distributed snow model parameters using satellite data is evaluated by a cross‐validation approach in a 26,000 km 2 mountainous region in Norway. From 6 years of data and 13 to 15 Moderate Resolution Imaging Spectroradiometer (MODIS) images per melt season, parameters of local snow depletion curves are estimated annually for each grid cell. The estimated values are averaged over 5 years and evaluated by the sixth year, using each year in turn for validation. The parameters are the subgrid snow storage coefficient of variation cv , the premelt snow‐covered fraction A 0 , the premelt snow storage m , and the time sequence of accumulated melt depth { λ }. The likelihood is formulated in terms of the Normalized Difference Snow Index (NDSI), rather than fractional snow‐covered area, in order to avoid highly skewed distributions for values close to 0 or 1. The 5 year averaged values for cv , A 0 , and a bias‐correcting snow storage multiplier m corr were applied in rerunning the precipitation‐runoff model. For 22 subcatchments within the region, validation‐year standard error in snow melt runoff volume was reduced from 21% to 13%. Standard error in NDSI on the grid cell level was reduced from 0.34 to 0.27. The stationarity of individual parameters was also evaluated, comparing the 5 year calibrated values for each of cv , A 0 , and m corr to the validation‐year estimates, after normalizing for the prior information. On average, the calibrated maps for cv , A 0 , and m corr predicted 32%, 46%, and 56%, respectively, of the spatial variance in the validation year's change from prior to posterior estimates.