A hybrid ensemble method to reduce redundancy and improve precipitation forecasting in climate models

Abstract

The persistent changes in hydroclimatic variables have fluctuated the extreme weather patterns, including drought, heat waves, and floods. The global climate models (GCMs) are fundamental and useful tools in projecting the changes in the occurrence of climate events. The present study attempts to overcome this challenge by introducing a new ensemble methodology, known as the grouped analysis-based hybrid weighted ensemble (GABHWE), which is capable of more effectively integrating multivariate and better predicting models. GABHWE integrated two approaches in weighting aspects: (i) hierarchical clustering to set similar GCMs to minimize redundancy and (ii) dynamic time fluctuating weighting, built on exponential transformation of point-wise differences between observed and simulated outcomes. During 1961-2014, the technique was implemented for CMIP6 GCMs across the Tibetan Plateau. The proposed method (GABHWE) is compared with simple model averaging (SMA), as well as Bayesian model averaging (BMA), based on the normalized root mean square error (NRMSE) and the normalized relative absolute error (NRAE). The validated outcomes demonstrate that GABHWE attains lower median NRMSE (0.37149) and NRAE (1.82263) than SMA (0.38254 and 1.86462) and BMA (NRMSE = 0.47983 and NRAE = 0.98487), confirming its improved accuracy and robustness.