Accurate downscaling of Global Circulation Models (GCMs) is critical to effectively assess climate change impacts and inform water resource management strategies. To address this need, this study introduces a hybrid downscaling framework that integrates Variational Mode Decomposition (VMD) with four distinct machine learning algorithms: Random Forest (RF), Extra Trees Regressor (ETR), Extreme Gradient Boosting (XGB), and Random Vector Functional Link (RVFL) to project daily maximum temperature. The methodology is implemented using CMIP6 data for the near-future (NF) and mid-future (MF) periods under the SSP1-2.6 and SSP5-8.5 scenarios. Based on the projected temperature data, key extreme temperature indices—specifically the percentage of warm days (TX90p), the Warm Spell Duration Index (WSDI), and the frequency of heatwaves (HW)—are subsequently derived for a detailed analysis of future climate extremes. The results indicate that the VMD-RVFL model exhibits the highest performance, characterized by the lowest error and highest correlation for projecting temperature. An increase in temperature is expected across all countries, with Turkey and Yemen experiencing the most and least pronounced changes, respectively. Specifically, temperatures in Turkey are projected to rise by more than 4 °C under the SSP5-8.5 scenario by 2084, while Yemen may see an increase of about 1 °C under the SSP1-2.6 scenario by 2054. Additionally, the TX90p, heatwave (HW), and Warm Spell Duration Index (WSDI) are projected to rise in all countries, signaling an increase in the frequency of extreme heat events. Qatar is anticipated to experience the most significant rise in TX90p, reaching up to 46.80 days under the SSP5-8.5 (mid-future) scenario by 2084, followed by Bahrain and Kuwait.
