Comparisons of statistical downscaling methods for air temperature over the Qilian Mountains

Abstract

Air temperature is an important indicator of climate change, as well as for understanding changes in hydrology, ecology, and other natural systems. However, meteorological stations that provide reliable temperature observations are usually sparse in areas of complex terrain, thus limiting our ability to quantify high spatial resolution temperature variability in these regions. Here, we apply three statistical downscaling methods to daily air temperature output from the sixth Coupled Model Intercomparison Project (CMIP6), validated with 22 meteorological stations over the Qilian Mountains. Based on different downscaling methods, we find RMSE and MAE are reduced as much as 59–66%, with the ratio of RMSE and MAE to the annual average temperature of stations decreasing from 147.9 to 61.0% and 143.3 to 64.7%, respectively, depending on the method. Compared to the original data, annual temperature based on the best downscaling methods differed by −2.85±3.61°C during the historical 1850–2014 period and, for the 2015–2100 projections, by 2.13±3.30°C for SSP1-2.6, −2.13±3.29°C for SSP2-4.5, −2.11±3.24°C for SSP3-7.0, and −2.12±3.23°C for SSP5-8.5. The downscaled annual air temperatures show a warming trend ranging 0.15–0.22°C/10 years for the historical experiment, 0.08–0.14°C/10 years for SSP1-2.6, 0.24–0.35°C/10 years for SSP2-4.5, 0.43–0.63°C/10 years for SSP3-7.0, and 0.52–0.76°C/10 years for SSP5-8.5 in the Qilian Mountains. These results indicate that the accuracy of the downscaled temperatures is improved compared to the original data. However, we also find that, compared with the downscaled data, the original projections have been overestimated.