A geostatistical spatially varying coefficient model for mean annual runoff that incorporates process-based simulations and short records

Abstract. We present a Bayesian geostatistical model for mean annual runoff that incorporates simulations from process-based hydrological model. The are treated as covariate and the regression coefficient is modeled spatial field. This way relationship between (simulations model) response variable (observed runoff) can vary in study area. A preprocessing step including short records modeling also suggested. thus obtain exploit several data sources. By using state-of-the-art statistical methods, fast inference achieved. evaluated by estimating period 1981–2010 127 catchments Norway based on observations 411 catchments. Simulations HBV 1×1 km grid used input. found average proposed approach outperformed purely (HBV) when predicting ungauged partially gauged reduction RMSE compared to was 20 % 58 catchments, where latter due step. For framework method with 10 method. however, methods performed equally well or slightly better than combination approach. In general, we expect outperform geostatistics areas availability low moderate.