Monitoring household water use in rapidly urbanizing regions is hampered by costly, time-intensive enumeration methods and surveys. We investigate whether publicly available imagery—satellite tiles, Google Street View (GSV) segmentation—and simple geospatial covariates (nightlight intensity, population density) can be utilized to predict household water consumption from the municipal piped network in Hubballi-Dharwad, India. Our analysis is restricted to areas with metered 24/7 water supply where reliable consumption measurements are available, representing approximately 43% of the municipal area. While prior work has typically used single imagery sources, we provide the first systematic comparison of multiple imagery modalities (survey-based, satellite-based, street-view-based, and hybrid approaches) specifically for piped water consumption prediction. We compare four approaches: survey features (benchmark), Convolutional Neural Networks (CNN) embeddings (satellite, GSV, combined), and GSV semantic maps with auxiliary data. Under an ordinal classification framework, GSV segmentation plus remote-sensing covariates achieves 0.55 accuracy (before tuning) for water use, approaching but not fully matching survey-based models (0.59 accuracy) before hyperparameter tuning, with performance gaps remaining for intermediate consumption categories. Error analysis reveals that misclassifications concentrate in systematic patterns: consumption predictions show overestimation bias when visual features suggest affluence despite moderate actual use, while income predictions exhibit boundary confusion between adjacent middle categories. Our findings demonstrate that open-access imagery, coupled with minimal geospatial data, offers a promising alternative to obtaining reliable household piped water consumption estimates using surveys in urban analytics.
Wang et al. (Thu,) studied this question.