• Each household’s water end-use model was initialized with user input &literature data. • Hourly water consumption data calibrated the model via Approximate Bayesian Computation. • Models validated on 20% of each home’s data with accurate results. • Shower use gaps highlight areas to improve model accuracy. • Framework offers actionable insights for smarter water conservation. Effective water conservation demands actionable insights into residential end-use patterns. Although previous research has explored water end-use disaggregation from second- or minute-resolution smart meter data, such approaches often involve expensive sensors and data infrastructure. Hourly-resolution data, on the other hand, is more commonly used to track consumption trends but generally lacks the detail needed for end-use disaggregation. In this study, we address this gap by calibrating a stochastic household water end-use (SThWE) model using hourly data. The model incorporates self-reported frequencies and durations of various water-use activities, combined with end-use water intensity estimates from existing literature. To calibrate the model, we apply Approximate Bayesian Computation (ABC), enabling the estimation of posterior distributions for water use intensity, duration, and frequency across multiple end-use categories. For individual household’s SThWE, we used 80% of its data for calibration and 20% for validation. Analysis of validation results from 30 households in Alicante, Spain suggests that the ABC-SThWE framework reliably estimates hourly consumption for individual households in the validation dataset. The ABC-SThWE framework further demonstrates how pairing hourly smart meter data with user-defined and literature-informed parameters can yield valuable insights for smart water demand management and conservation. In the discussion, we explained the discrepancy between ABC-inferred parameters and those derived from flow sensor data for shower use, both at individual and group household levels, suggesting areas to improve SThWE and inference accuracy.
Fang et al. (Sun,) studied this question.