Spatial Bayesian Model for Statistical Downscaling of AOGCM to Minimum and Maximum Daily Temperatures

Abstract

Abstract Atmosphere–ocean general circulation models (AOGCMs) are useful for assessing the state of the climate at large scales. Unfortunately, they are not tractable for the finer-scale applications (e.g., hydrometeorological variables). Downscaling methods allow the transfer of large-scale information to finer scales and they are thus relevant for the assessment of finer-scale variables. Among a wide range of downscaling methods, regression-based approaches are commonly used for downscaling AOGCM data because of their low computational requirements. However, downscaled variables are generally reproduced at gauged weather stations only. Results at the gauged stations can then be interpolated a posteriori at ungauged locations with kriging or other methods. In this paper, a spatial Bayesian model is proposed for the downscaling of coarse-scale atmospheric data (i.e., either reanalysis or AOGCM) to minimum and maximum daily temperatures. This approach uses a Bayesian framework for mixing a prior distribution reflecting the monthly spatial dependence of the temperatures with the daily fluctuations induced by the atmospheric predictors. Local characteristics (i.e., altitude and latitude) are also taken into account in the mean of the prior distribution by using a geographical regression model. The posterior distribution thus reflects both monthly local patterns because of the prior and daily larger-scale fluctuations. Finally, the Bayesian approach also allows for the accounting of estimated parameter uncertainty, making it more stable to poor parameter fitting. The method is applied to the southern part of the province of Quebec, Canada. Results show that the downscaled distributions of the temperatures at gauged sites are in sufficient agreement with the validation dataset compared to a classical regression-based method. The proposed model has also the advantage of directly producing temperature maps.