An efficient statistical approach to multi-site downscaling of daily precipitation series in the context of climate change.

Global Climate Models (GCMs) have been extensively used in many climate change impact studies. However, the coarser resolution of these GCM outputs is not adequate to assess the potential effects of climate change on local scale. Downscaling techniques have thus been proposed to resolve this problem either by dynamical or statistical approaches. The statistical downscaling (SD) methods are widely privileged because of their simplicity of implementation and use. However, many of them ignore the observed spatial dependence between different locations, which significantly affects the impact study results. An improved multi-site SD approach is thus presented in this paper to downscaling of daily precipitation at many sites concurrently. This approach is based on a combination of multiple regression models for rainfall occurrences and amounts and the Singular Value Decomposition technique, which models the stochastic components of these regression models to preserve accurately the space-time statistical properties of the daily precipitation. Furthermore, this method was able to describe adequately the intermittency property of the precipitation processes. The proposed approach has been assessed using 10 rain gauges located in the southwest region of Quebec and southeast region of Ontario in Canada, and climate predictors from the National Centers for Environmental Prediction/National Centre for Atmospheric Research re-analysis data set. The results have indicated the ability of the proposed approach to reproduce accurately multiple observed statistical properties of the precipitation occurrences and amounts, the at-site temporal persistence, the spatial dependence between sites and the temporal variability and spatial intermittency of the precipitation processes. [ABSTRACT FROM AUTHOR]