Identification of the characteristics of non-stationary spatio-temporal variations of future temperature in the Tibetan Plateau based on a coupled EOF-EEMD method

The climate model provides simulation results in studying the climate change and its consequences. However, its application in a specific relatively small area (compared to global scale) is somehow confined for its lackness in high resolution and poverty in accuracy. Therefore, downscaling and bias correction are necessary to be undertaken to improve the output data from the climate model. Because a single EOF model is difficult to identify the time series change trend, this paper uses EOF and ensemble empirical mode decomposition (EEMD) coupling to accurately identify the statistical characteristics of time series to extract the temporal and spatial variation characteristics of meteorological data. In this study, the monthly mean temperature observation data set of ERA5 from 1970 to 2014 was used. First of all, six climate models and Multi-Model Ensemble (MME) average models of CMIP6 were evaluated and optimized by Taylor diagram, Taylor index, interannual variability assessment index and rank scoring method, and the best data set were chosen for the later treatment. Then, the Delta bias correction method and Normal distribution matching method were used to correct the chosen data. Finally, the temporal and spatial variation characteristics of temperature in the Tibet Plateau from 2015 to 2100 under SSP1-2.6, SSP2-4.5 and SSP5-8.5 scenarios were analyzed. The results show that: (1) Among the six CMIP6 models and MME average models selected in this paper, the EC-Earth3 model has the best performance in simulating temperature. (2) Comparing the results of the EC-Earth3 model after the Delta bias correction with the observation results, the regional averages of the deterministic coefficient (R²) and the Nash efficiency coefficient (NSE) are 0.992 and 0.983, respectively. After the Normal distribution matching method is used to correct, the regional averages of the deterministic coefficient (R²) and the Nash efficiency coefficient (NSE) are 0.990 and 0.978, respectively. Therefore, the Delta bias correction has a better correction effect on the monthly temperature of the model. (3) By coupling the EOF-EEMD method, it is found that the first typical field shows consistent changes in the whole region under the three scenarios, and there are common temperature change sensitive areas and non-sensitive areas under the SSP1-2.6 and SSP2-4.5 scenarios, namely, the northern Tibetan Plateau and the Pamir Plateau. The temperature of the second typical field shows a distribution that gradually decreases (SSP1-2.6, SSP2-4.5) or increases (SSP5-8.5) from the upper reaches of the Zhaqu River to the surrounding areas. Under the SSP1-2.6 scenario, the plateau as a whole is cooling down in the east and warming up in the west. Under the SSP2-4.5 and SSP5-8.5 scenarios, the plateau first warms up in the east and cools down in the west, and then cools down in the east and warms up in the west. This study can provide a reference bias correction method for a more accurate application of climate model data in the Tibet Plateau, and provide key basic information supporting in-depth assessment of the impact of temperature changes on water resources, ecosystems and environment in the Tibet Plateau.