Statistical Downscaling of Rainfall using Large-Scale Predictors

Yan Lu, P.V. Mandapaka, C.C. Kuo, T. Y. Gan, Xiaosheng Qin

A warmer climate is expected to lead to more serious natural disasters, such as heavy storms, prolonged droughts and frequent floods. For a high-density urban region, the flash flood problem may become worse due to possible increasing frequency and magnitude of short-duration rainfalls in the future. A Global Circulation Model (GCM) is a powerful tool to assess the climate change impact. However, the resolution of a GCM output is generally too coarse to be applicable to small regions directly. Two types of approaches, dynamical and statistical downscaling, could be used for bridging the gap between GCM and local climate information. Compared with dynamical downscaling, the statistical approach is more flexible and computationally less intensive. In addition, statistical downscaling tools may be sensitive to the resolution of large-scale predictors. In this study, two downscaling approaches are compared. The first is to use a statistical method (Automatic Statistical Downscaling, ASD) directly to downscale large-scale predictors (i.e. ERA-Interim Reanalysis data) to local rainfall. The second is to combine a dynamical (i.e. MM5) and a statistical method (ASD) to generate the station-level data. The study site is the City of Edmonton and the resolutions of large-scale GCM predictors and dynamical model output are about 150 km and 27 km, respectively. The results show that the downscaled results based on predictors from MM5 is better than that from ERA-Interim, in terms of both accuracy and uncertainty range.