Predicting the Optimal Tank Volume of a Rainwater Harvesting System: A Machine Learning Approach Based on Building Characteristics

Rainwater harvesting (RWH) is a promising strategy for addressing water scarcity, and the optimal design of storage tanks is vital for their efficiency. This study develops a predictive model using machine learning algorithms, specifically Random Forest regression optimized with Bayesian methods, to accurately forecast the optimal tank volume for rainwater harvesting systems in Guangzhou, China. The model considers building characteristics such as catchment area, floor area, daily water consumption, and number of occupants as input variables. Four machine learning models (random forest, AdaBoost, light gradient boosting machine, and support vector regression) were evaluated, with random forest demonstrating superior predictive performance. Bayesian optimization further improved model accuracy, resulting in a mean absolute error (MAE) of less than 4.2, a root mean square error (RMSE) of less than 5.8, and a Nash-Sutcliffe efficiency (NSE) greater than 0.87. Notably, catchment area and floor area emerged as the most influential variables affecting tank volume predictions. The predictive performance varied across climate years, highlighting the importance of considering interannual rainfall variability. This study presents a novel approach for optimizing rainwater harvesting system design, providing technical support for maximizing economic returns and water resource management.