Prediction of urban residential end-use water demands by integrating known and unknown water demand drivers at multiple scales II: Model application and validation

- An application of a stochastic modeling method is presented.
- Probability distributions to describe each probability space is identified.
- The model validation demonstrates a satisfactory predictive capacity.
- This also evidences model's effectiveness to predict water demand at multiple scales.

Detailed prediction of end-use water demand at multiple spatial and temporal scales is essential for planning urban water supply using multiple water sources based on fit-for-purpose criteria. This paper presents the application of a stochastic model to predict urban residential end-use water demands at multiple spatial and temporal scales. The model includes an improved representation of spatial and temporal variability of urban residential water use by considering the effect of a significant number of water demand drivers such as household size, dwelling type, appliance efficiency, availability of water end-uses/appliances at dwellings, presence of children, presence of people at home, diurnal behavioral patterns and temperature. A stochastic approach is used to describe the variability of residential water demand that is not captured by these known explanatory variables. The model is validated against quarterly meter readings and hourly water use data. The validation of household water demand at a quarterly scale with billing data shows Correlation coefficients (R2) ranging between 90% and 96% and Nash-Sutcliffe coefficients ranging between 0.70 and 0.92 for the four seasons analyzed which, verifies the predictive capacity of the model. The model validation also demonstrates the statistical stability of the selected probability distributions used in modeling the unexplained behavior of urban residential water consumers. The hourly scale validation also demonstrates a satisfactory predictive capacity in predicting household water demand. This also evidences the effectiveness of the modeling approach to predict urban residential water demand at multiple temporal scales.