Predictive ability of climate change with the automated statistical downscaling method in a freeze–thaw agricultural area

Precipitation and temperature in freeze–thaw agricultural area have different patterns under global warming. In this study, a statistical relationship between large-scale changes in climate variables and local weather data was built by applying an automated statistical downscaling (ASD) model in the Sanjiang Plain in China. We evaluated the prediction ability of the ASD model in terms of spatial–temporal changes in freeze–thaw agricultural area, and the temperature and precipitation changes in the twenty-first century under the representative concentration pathway 4.5 (RCP4.5) scenario were estimated. The results revealed that the explained variances in temperature were higher than 0.93 during the calibration and verification periods, which demonstrated good simulation capacity. The R² of precipitation was acceptable due to the randomness and complexity of daily precipitation. Based on the Geophysical Fluid Dynamics Laboratory Earth System Model with the Generalized Ocean Layer Dynamics component (GFDL-CM3), the regional climate simulation provided good predictions. By 2100, the average, maximum and minimum temperatures in this area could increase by 2.0–2.5 °C, 2.5 °C and 2.5–4.0 °C, respectively. In terms of the spatial distribution, temperatures could increase faster in the northern region and slower in the central and southern regions. The warming trends in summer and winter were more significant than those in spring and autumn. There was no significant change in annual precipitation in the twenty-first century (increased approximately 10 mm by 2100). Precipitation decreased obviously in July and August (approximately 0.4 mm/day), and other months showed an increasing trend (approximately 0.5–0.9 mm/day). There will be large spatial variation of precipitation in the future changes. The results could serve as a reference for assessing non-point source pollution and agricultural management.