关键词: 长江上游流域, CMIP6气候模式, 深度神经网络, 降尺度

Abstract:

The Coupled Model Intercomparison Project Phase 6 (CMIP6) provides systematic data support for global climate research, and downscaling techniques can address the limitation of its spatial resolution, which is insufficient for climate projections in regions with complex terrain. This study focused on the upper Yangtze River basin, systematically evaluated the simulation capabilities of eight CMIP6 climate models for precipitation and mean temperature during the baseline period (2015−2020) using Taylor diagrams, and constructed a multi−feature fusion downscaling model based on Deep Neural Networks (DNN) to downscale the optimized climate models spatially, and analyzed the variation characteristics of precipitation and mean temperature in the basin under four future scenarios (SSP1−2.6, SSP2−4.5, SSP3−7.0 and SSP5−8.5) in 2050. The results showed that the Ec−Earth3−Veg climate model performed best at simulating precipitation in the upper Yangtze River basin from 2015 to 2020, while the ACCESS−ESM1−5 model performed best at simulating mean temperature over the same period. The multi−feature fusion spatial downscaling model, based on a DNN, increased the horizontal resolution of precipitation in the region from 0.7031°×0.7018° to 0.008°×0.008° for both the 2015−2020 and 2050 periods, and the horizontal resolution of mean temperature from 1.875°×1.250° to 0.018°×0.018°. The average correlation coefficient between the downscaled precipitation and mean temperature results and the observed data was 0.47 and 0.88, respectively. Compared with the baseline period, precipitation in 2050 was projected to increase significantly under the SSP1−2.6 and SSP5−8.5 scenarios, while it was expected to decrease significantly under the SSP2−4.5 and SSP3−7.0 scenarios, with the largest decrease of 15.51% under SSP3−7.0. The mean temperature was projected to rise relative to the baseline period in all scenarios, with greater warming in the higher-emission scenarios. This study provides data to support the scientific allocation of water resources and the precise management of agricultural production in the upper Yangtze River basin. It serves as a scientific reference for risk assessment and response strategies to extreme regional climate events.

Key words: Upper Yangtze River basin, CMIP6 climate model, Deep neural network, Downscaling