DOI: 10.22531/muglajsci.1774068 ISSN: 2149-3596

A Soft Computing Framework for System Identification and Predictive Modeling of Rainwater Harvesting from Complex Architectural Geometries

Tevfik Denizhan Müftüoğlu
Show PDF Cite
Global water shortages have become a pressing problem which requires an immediate inclusion of alternative sources of water within all urban planning initiatives. Rainwater harvesting has been identified as one strategy that can be used to provide increased climate resilience and sustainability. There are however significant hydrologic complexities associated with assessing the potential for rainwater harvesting on public structures featuring complex architectural designs such as the multi-domed roof design of mosques that remain largely unexamined in contemporary literature. Traditional hydraulic models assume constant runoff coefficients, uniform rainfall, and ideal surfaces, overlooking seasonality, material differences, and irregular geometry. A soft-computing framework combining XGBoost and SHAP applied to ten mosque domes in Turkey with long-term rainfall and geometric data provides accuracy and interpretability by outperforming the baseline equation (R² = 0.72) and eleven other algorithms, achieving R² = 0.998 and RMSE = 0.40. SHAP identified rainfall, dome area, and seasonality as dominant drivers; roof material and institutional context also contributed. In wet months some domes offset over 90% of ablution-related water use; in dry months yields fell below 2%. This is the first empirical study to evaluate mosque domes as harvesting surfaces using interpretable AI, showing contributions to sustainable water management.

More from our Archive