Multi-model statistical downscaling of spring precipitation simulation and projection in central Asia based on canonical correlation analysis

Using precipitation observations at 30 meteorological stations in central Asia, the European Centre for Medium-Range Weather Forecasts 40-year reanalysis dataset (ERA-40), and eight CMIP5 (Coupled Model Inter-comparison Project Phase 5) climate models, statistical downscaling models were constructed based on BP-CCA (the combination of empirical orthogonal function and canonical correlation analysis). The downscaling ability of multiple-model simulations of spring precipitation was evaluated and future changes in precipitation were projected. The results show that the average correlation coefficient between time coefficients of downscaled spring precipitation and corresponding observations is 0.35, and the highest correlation coefficient is 0.62. Spatial correlations were also improved with an average value of 0.87. The absolute values of domain-averaged relative precipitation errors for most models were reduced by 0.2%-8% after statistical downscaling. As a result of statistical downscaling of multi-model ensemble (SDMME) simulation, the relative error reduced from 64% to 4%, the spatial correlation increased from 0.47 to 0.81, and the RMSE reduced to 0.59. These results demonstrate that the simulation of SDMME is better than that of multi-model ensemble (MME) and the downscaling results of most individual models. The projections of SDMME reveal that under the RCP4.5 (Representative Concentration Pathway 4.5) scenario, the projected domain-averaged precipitation changes for the early (2016-2035), middle (2046-2065) and end (2081-2100) of the 21 century are-5.3%, 3.0% and 17.4%, respectively. In the early 21 century, precipitation shows a decreasing trend in most areas and an increasing trend in southern part of central Asia. Significant increasing trend is predicted to occur mainly in the middle and end periods of the 21 century, with a larger magnitude in the latter. The credibility of SDMME forecast gradually enhances with longer forecast time.