Downscaling and correction of regional climate models outputs with a hybrid geostatistical approach

This paper presents an approach to downscaling of climate models based on a combination of Generalised-additive-models and geostatistics. The paper aims at increasing the usefulness of Climate Models by creating data sets with a spatio-temporal resolution appropriate for applications of environmental models at the management scale. Simulations of climate change available from global and regional climate models require downscaling and bias correction for hydrological or ecological applications. The paper assess the effectiveness of the approach applied to monthly means of temperature and rainfall, to predict soil wetness conditions. The main focus was on reconstructing the spatial structure of the landscape. The climate model data were firstly downscaled using a geo-statistical approach combining generalised additive models (GAMs) with kriging, making use of the covariates to reproduce the spatial pattern. The downscaled climate model data were corrected for bias using interpolated meteorological ground station data (1961-1999). In general the downscaling approach provided lower RMSE and closer reproduction of the variogram structure. The use of the bias corrected downscaled data improved the results of the model used to predict soil wetness, increasing the validation metrics. The approach is completely implemented in open-source software, in particular GRASS-GIS and R.