Amenity counts significantly improve water consumption predictions

Abstract

Anticipating the increase in water demand in an urban area requires us to properly understand daily human movement driven by population size, land use, and amenity types among others. Mobility data from phones can capture human movement, but not only is this hard to obtain, but it also does not tell where the population is going. Previous studies have shown that amenity types can be used to predict people's movement patterns; thus, we propose using crowd-sourced amenity data and other open data sources as reasonable proxies for human mobility. Here we present a framework for predicting water consumption in areas with established service water connections and generalize it to underserved areas. Our work used features such as geography, population, and domestic consumption ratio and compared the prediction performance of various machine learning algorithms. We used 44 months of monthly water consumption data from January 2018 to July 2021, aggregated across 1790 district metering areas (DMAs) in the east service zone of Metro Manila. Results show that amenity counts reduce the mean absolute error (MAE) of predictions by 1,440 m3/month or as much as 5.73% compared to just using population and topology features. Predicted consumption during the pandemic also improved by as much as 1,447 m3/month or nearly 16% compared to just using population and topology features. We find that Gradient Boosting Trees are the best models to handle the data and feature set used in this work. Finally, the developed model is robust to disruptions in human mobility, such as lockdowns, indicating that amenities are sufficient to predict water consumption.