Integrating Statistical Distributions with Machine Learning to Model IDF Curve Shifts under Future Climate Pathways

Abubakr Taha Bakheit^1*Ali Aldrees¹Abdeliazim Mustafa¹Assoc. Prof. Dr. Gasim Hayder²Muhammad Babur³Shay Haq⁴

• ¹Prince Sattam bin Abdulaziz University, Al Kharj, Saudi Arabia
• ²Universiti Tenaga Nasional, Kajang, Malaysia
• ³University of Central Punjab, Lahore, Pakistan
• ⁴National University of Sciences and Technology (NUST), Islamabad, Pakistan

Climate change has intensified rainfall variability, increasing urban flooding risks in arid regions like Makkah and Riyadh. This study develops Intensity-Duration-Frequency (IDF) curves to analyze rainfall intensities for various storm durations and return periods, supporting urban planning and water resource management. Historical precipitation data (1950–2020) and future projections from two Shared Socioeconomic Pathway scenarios (2021–2100) were used to construct IDF curves for Makkah and Riyadh to assess precipitation extremes and support hydrological and infrastructure planning. Downscaling and bias correction were applied to five Global Climate Models, followed by feature engineering using CatBoost and LightGBM. Multi-Model Ensemble (MME) predictions were then evaluated using machine learning algorithms, including AdaBoost, CatBoost, and XGBoost, with XGBoost achieving the highest accuracy. For precipitation modeling, Gamma and Log-Pearson 3 distributions were identified as the best fits for observed and projected data in Makkah and Riyadh, respectively, underscoring the importance of selecting appropriate probability distributions to accurately capture precipitation extremes. The study offers a predictive tool in terms of climate resilience of urban areas within arid zones, which strengthens climate projections to aid decision-making.