Amenity counts significantly improve water consumption predictions.

Area of Science:

• Environmental Science
• Urban Planning
• Data Science

Background:

• Accurate urban water demand forecasting is crucial for managing resources.
• Traditional methods struggle with dynamic human mobility patterns.
• Mobile phone data is difficult to obtain and interpret for location-specific movement.

Purpose of the Study:

• To develop a framework for predicting urban water consumption using open data sources.
• To evaluate the effectiveness of crowd-sourced amenity data as a proxy for human mobility.
• To generalize the prediction model for both serviced and underserved urban areas.

Main Methods:

• Utilized monthly water consumption data from 1790 district metering areas (DMAs) in Metro Manila (Jan 2018 - Jul 2021).
• Incorporated features including geography, population, domestic consumption ratio, and amenity data.
• Compared the predictive performance of various machine learning algorithms, focusing on Gradient Boosting Trees.

Main Results:

• Amenity data reduced prediction error (Mean Absolute Error) by up to 5.73% (1,440 m3/month) compared to population and topology alone.
• Water consumption predictions during the pandemic improved by nearly 16% (1,447 m3/month) with the inclusion of amenity data.
• Gradient Boosting Trees demonstrated the best performance with the selected features.

Conclusions:

• Crowd-sourced amenity data serves as a robust proxy for human mobility in water demand prediction.
• The developed model is resilient to disruptions in human movement, such as lockdowns.
• Amenity data significantly enhances the accuracy and reliability of urban water consumption forecasts.