Understanding the variability of large-scale statistical downscaling methods under different climate regimes

Abstract

Large-scale downscaling plays an important role in assessing global impacts on hydrological sphere due to climate changes. In such downscaling efforts, it is essential to consider the various climate regimes. Although previous studies have indirectly suggested that the accuracy of downscaling might differ among climate regimes, research that systematically understands or quantifies the variability of this accuracy remains scarce. This study addresses this gap by systematically quantifying the performance of five different large-scale downscaling methods across various climate regimes in the context of downscaling hydroclimatic indicators. Our findings indicate that large-scale downscaling yields the highest accuracy on average when applied to temperature, precipitation, and runoff in tropical, arid, and temperate climate regimes, respectively, while showing poor accuracy in polar regimes for all variables. The maximum difference of normalized root mean squared errors for hydroclimate indicators is 69 % across climate zones, and the spatial distribution of downscaling accuracy aligns with spatial distribution of climate zones. The variation of downscaling accuracy is particularly significant in temperature, precipitation, and seasonal runoff indicators. Furthermore, linkages between accuracy of climate and hydrological indicators differ by climate zones. The underlying reasons for the different accuracy of downscaling are spatially different accuracy of global climate models (GCMs) and interaction of downscaling structure and climate regimes. This study articulated the source of spatially different accuracy/uncertainties for large-scale downscaling that have never been addressed before. The findings of this study provide valuable support in selecting appropriate downscaling methods, ultimately enhancing the spatial reliability and accuracy of large-scale downscaling methods.