Enhancing extreme temperature projections using a hybrid MGWR – Copula approach

Extreme temperature events in Asia are intensifying, necessitating advanced predictive models that account for spatial heterogeneity and complex dependencies among climate variables. This study integrates Multiscale Geographically Weighted Regression (MGWR) and Copula Regression to enhance the accuracy of future extreme temperature projections. By incorporating General Circulation Models (GCMs), this study assesses the evolution of maximum temperatures under changing dependency structures among key climate variables, particularly geopotential height, precipitation, humidity, and wind speed. Our results demonstrate that the hybrid MGWR-Copula model significantly outperforms conventional machine learning approaches, such as Random Forest and Support Vector Machines, in capturing non-linear dependencies and spatial variations. Compared to global regression models, our approach provides higher predictive accuracy, particularly in regions with complex terrain like South Korea and Japan. Furthermore, projections under different Shared Socioeconomic Pathways (SSP2-4.5 and SSP5-8.5) indicate notable increases in extreme temperatures, with high-emission scenarios leading to greater variability and forecast uncertainty. This study presents a robust framework for climate modeling, improving our ability to predict extreme temperatures and informing climate adaptation strategies. By integrating spatial regression, dependence modeling, and machine learning, this offers critical insights for climate risk assessment and policy development in vulnerable regions.