Predicting Household Water Consumption Using Satellite and Street View Images in Two Indian Cities

TL;DR

It is demonstrated that open-access imagery, coupled with minimal geospatial data, offers a promising alternative to obtaining reliable household water consumption estimates using surveys in urban analytics.

Abstract

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 in Hubballi-Dharwad, India. We compare four approaches: survey features (benchmark), 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 for water use, approaching survey-based models (0.59 accuracy). Error analysis shows high precision at extremes of the household water consumption distribution, but confusion among middle classes is due to overlapping visual proxies. We also compare and contrast our estimates for household water consumption to that of household subjective income. Our findings demonstrate that open-access imagery, coupled with minimal geospatial data, offers a promising alternative to obtaining reliable household water consumption estimates using surveys in urban analytics.

Figures (7)

• Figure 1: Spatial distribution of monthly household income and water consumption. Self-reported monthly income (top) and average monthly water consumption (bottom) are shown for continuous water supply zones in Dharwad (left) and Hubballi (right). Maps were produced using Esri’s ArcGIS Pro 3.1.4.
• Figure 2: Illustrative examples street view images with semantic segmentation features. Pixel-level percentages for each detected feature (e.g., buildings, walls, sidewalks, vegetation, and sky) are shown on the left side of each image.
• Figure 3: Comparative performance of models predicting household income and water consumption using survey, image, and geospatial features. Each bar shows Accuracy or ROC-AUC score across four classifiers (Logistic Regression, Random Forest, XGBoost, and LightGBM) under six feature settings, including survey-based, image embeddings, semantic segmentation, and geospatial augmentation.
• Figure 4: Feature importance scores from the best-performing LightGBM model predicting household income and water consumption levels Left: Top 15 feature importance scores to predict income. Right: Top 15 feature importance scores to predict water consumption.
• Figure 5: Confusion matrix of the best performing model on the validation set. Left: Normalized confusion matrix -- income. Right: Normalized confusion matrix -- water consumption.