A Novel Hybrid Approach for Predicting Western Australia’s Seasonal Rainfall Variability

Abstract

AbstractIn this paper, 100 years of uninterrupted rainfall data for 12 rainfall stations (four rainfall stations from each region) in Western Australia were analyzed against respective dominant climate indices, and representative prediction models were developed using ARIMAX, GEP, and a hybrid technique (GEP-ARIMAX). Statistical performance evaluators such as Pearson correlation $$(r)$$ ( r ) , root mean square error $$(RMSE)$$ ( R M S E ) , mean absolute error ($$MAE$$ MAE ), and refined Willmot index of agreement ($${d}_{r}$$ d r ) were used to evaluate the prediction performance of the developed models. These models demonstrated their capability to predict up to four months in advance with Pearson correlation $$(r)$$ ( r ) values ranging from 0.53 to 0.83, 0.75 to 0.85, and 0.87 to 0.95 for ARIMAX, GEP, and hybrid (GEP-ARIMAX) models respectively. While compared, the hybrid (GEP-ARIMAX) model showed superior prediction performance in both calibration and validation periods with Pearson correlation $$(r)$$ ( r ) and refined Willmot index of agreement ($${d}_{r}$$ d r ) values were as high as 0.96 and 0.84 respectively. This paper demonstrated a novel hybrid GEP-ARIMAX model showing significantly good rainfall forecasting capability than conventional linear and non-linear models.