CR prepress abstract - doi: 10.3354/cr01460
Multisite and multivariate GCM downscaling using a distribution-free shuffle procedure for correlation reconstruction
Zhi Li*, Xiaoping Shi, Jingjing Li
ABSTRACT: Reconstructing the multisite and multivariate correlations of meteorological variables during downscaling of general circulation model (GCM) outputs is extremely important for hydrological variability simulation. To this end, this study proposes a two-step approach to downscale GCM considering the correlations between sites and variables. The first step generates single-site climate change scenarios for daily precipitation, maximum and minimum temperature. This step is open to any downscaling technique since the subsequent correlation reconstruction does not change the statistical parameters. This study employed quantile mapping method for spatial downscaling and the Richardson-type weather generator for temporal downscaling. The second step recorrelates the generated daily and site-based climate change scenarios by a distribution-free shuffle procedure. We used the precipitation, maximum and minimum temperature from a network of 15 weather stations as well as those from one GCM to develop and evaluate the method. Results showed that the proposed method satisfactorily preserved the downscaled statistical parameters, intersite and intervariable correlations. Simultaneously shuffling the data for all stations and all three variables for each month, the proposed approach is easy-to-implement and computation-saving, and provides a rational option for multisite and multivariate GCM downscaling.