Exploitation of the ensemble-based machine learning strategies to elevate the precision of CORDEX regional simulations in precipitation projection.

Multi-model Ensembles (MMEs) are widely used to reduce uncertainties associated with simulations and projections of GCM/RCM.MMEs combine the results of multiple climate models to produce a more robust and reliable prediction. By considering the range of outputs from different models, MMEs can improve the overall accuracy of climate projections. Therefore, the study focused on the use of some techniques, namely Multivariate Linear Regression (MLR), Weighted Average (WA), and some ML algorithms including Least Square Support Vector Machine (LS-SVM), Random Forest, (RF) and multivariate adaptive regression splines (MARS) to develop MMEs to simulate precipitation patterns over Iran. The regional climate models (RCMs) used in this research were extracted from the South Asia Coordinated Regional Climate Downscaling Experiments (CORDEX-SA) dataset. By comparing the individual RCMs and MMEs developed using the proposed methods, it was found that MMEs improved their capabilities compared to individual RCMs in their ability to simulate precipitation patterns. Furthermore, the study revealed that the MME developed using RF (MME-RF) exhibited more consistent performance across different spatial regions compared to other methods, especially WA technique, which displayed the lowest performance in comparison to other methods. Regarding the projections of seasonal precipitation under RCP4.5 and RCP8.5 scenarios, a potential decrease (roughly 6.5%) in precipitation in the western regions during autumn season, was observed. Whereas, the southern and southeast regions of Iran in particular showed a pronounced wetting tendency during the autumn season. According to the forecasts, the maximum percentage change (PC) of precipitation in these regions is expected to increase by 13.88%. [ABSTRACT FROM AUTHOR]