Statistical Downscaling of Future Temperature in Tianshan Mountains Based on Batch Gradient Descent Nonlinear Regression

Abstract Temperature is a key variable in alpine regions such as the Tianshan Mountains, which affects the population and economic development in lowlands by regulating the melting of snow and glaciers. The climate change over the Tianshan Mountains may have a strong climate impact. This study develops a new downscaling scheme including global climate models and DEM to simulate a high‐resolution temperature data set (90 m) in the Tianshan Mountains from 2021 to 2050. First, at low resolution, we developed a batch gradient descent‐based nonlinear regression downscaling model to simulate the relationship between CMIP5 and CMIP6 temperature and explanatory variables. Then, we input high‐resolution explanatory variables into the model to obtain downscaled (90 m) grid temperature data sets. At 30 meteorological stations, the R 2 of the observations and simulations is above 0.80 under RCP4.5 scenario and above 0.60 under SSP245 scenario. The RRMSE and MARE are below 0.53, and the slope of the linear functions of the two are close to 1. The simulation results show that Tianshan will accelerate warming in the next 30 years. Under the RCP4.5 scenario, the temperature will rise the fastest in autumn with 2°C/10a, and decrease in spring with 1.4°C/10a. Under the SSP245 scenario, the temperature will rise significantly in spring and winter, with 1°C/10a. Spatially, high mountains will warm faster than plains and basins. The insights gained in this research have the potential to inform climate forecasting efforts, providing a foundation for predicting runoff dynamics, assessing regional water resources, and estimating net primary productivity.