A Statistical GCM Downscaling Method Based on Multivariate Regression and Multi-Datasets Using in the Huaihe River Basin

To fully utilize the limited data and explore the effects of different forecast factors selection and forecast factor range selection on statistical downscaling, forecast factors were jointly selected from the monthly average datasets of NCEP/NCAR and ECMWF ERA-40, as well as the 74 circulation indices provided by the National Climate Center of China (NCCC). From these factors, 80 were selected as potential downscaling factors. Without assuming that the relationship between GCM and large-scale factors is identical to that between observations and large-scale factors, by assuming the stability of each relationship from now to the future, based on the three basic conditions of statistical downscaling, two distinct relationships are established, which are then integrated using large-scale factors to derive the downscaling equation for GCM. A multivariate regression downscaling method has been developed, which integrates simple correlation analysis and empirical orthogonal function (EOF) for the selection of forecast factors. Taking the Huaihe River Basin as the study area, the downscaling effect of this method was discussed for different scale ranges (uniform factor for the entire basin, one set of factors for each of the upper, middle, and lower reaches of the basin, and site-by-site factors) for the integrated model obtained by integrating GCMs with good simulation results to reduce the interference caused by different model differences. The downscaling model was established using data between 1961-1990 and validated using data between 1991-2000. The results indicate that the statistical downscaling model for the Huaihe River Basin, developed with this regression method, achieves an accuracy rate for monthly total precipitation within the range of 65% to 80%, and for monthly average temperature, it exceeds 90%, thereby validating the model's efficacy as a statistical downscaling approach. Uncertainty analysis reveals that the effectiveness of statistical downscaling is influenced by the selection of downscaling factors, the range of factor effects, and the combination of factors. Precipitation forecasting is more influenced by these factors, the combinations with the highest precipitation efficiency coefficients predominantly result from regional downscaling factors, whereas most of the combinations with the highest correlation coefficients are based on station-specific downscaling factors. The two measurement indicators for evaluating the effectiveness of precipitation downscaling are not easily harmonized, suggesting that the factors influencing precipitation exhibit significant variability. Moreover, precipitation within the basin is influenced by diverse factors, with complex local precipitation characteristics, and in summary it is more suitable to select downscaling factors for precipitation within a smaller range of effects in the Huaihe River Basin, while the range for temperature downscaling factors can be chosen on a basin-wide scale, the efficiency coefficients and correlation coefficients are both the highest, indicating that the factors affecting temperature have a large area of influence in the Huaihe River Basin, and the temperature climate characteristics of the entire basin are relatively consistent. The uncertainty in precipitation downscaling exceeds that of temperature downscaling, making the forecast more difficult.