Data-Driven Regression Tree Model for Predicting Residential Water Demand Using Property Characteristics

Water demand management is paramount for water utilities to meet the potable water needs of the population. Utilities can forecast residential water demand based on the traits of the households they serve utilizing monthly water consumption data and property information. This research applies a data-driven approach to analyze monthly water consumption patterns at a spatially aggregated level labeled as system subdivisions and their corresponding aggregated property characteristics to predict future demand. A regression tree model is applied using the average lot area, building area, building age, and the number of bathrooms as input variables and the monthly water demand as the response. The tree model explains an average of 62% of monthly water demand across subdivisions, with the lot, building, and basement areas as the most important predictors followed by the building age. Our model also reports that the building's number of bedrooms and bathrooms is not as important as area measurements when working with monthly water demand, which is reported at spatially aggregated levels. Furthermore, a stronger correlation between monthly water demand and household characteristics such as the number of bedrooms was found in the cooler months of January and April, and a weaker correlation in the warmer months of July and September due to the higher consumption variability. Practitioners and researchers can use the proposed model in water infrastructure long-term planning to account for the socioeconomic effects on water demand using publicly available building characteristics.