Future precipitation patterns: investigating the IDF curve shifts under CMIP6 pathways

Abstract

Climate change has altered rainfall patterns, leading to urban flooding in Peshawar City. This study develops intensity–duration–frequency (IDF) curves to assess rainfall intensities for various return periods and durations. The methodology involves downscaling and bias correction of general circulation model (GCM) data, followed by feature selection using XGBoost and Extra Tree to rank nine GCMs. The top three models were used as input for four machine learning (ML) algorithms – random forest, regression tree, gradient boosting, and AdaBoost – for multi-model ensemble estimation. The models’ performance was evaluated using mean squared error, mean absolute error, root mean squared error, Nash–Sutcliffe efficiency (NSE), and Willmott's index (WI), with AdaBoost outperforming others. Bias-corrected and ensemble-modeled data were used to develop IDF curves employing normal, lognormal, and Gumbel distributions under shared socioeconomic pathways (SSPs) 245 and 585. Rainfall intensities were estimated for return periods of 2, 10, 25, 50, 75, and 100 years. This study enhances the IDF curve development by integrating advanced bias reduction and ML techniques, providing crucial insights into future rainfall patterns. The findings contribute to urban flood risk management and climate resilience planning for Peshawar City.