Access to clean water is crucial for sustainable development, yet challenges persist, particularly in developing nations. In Ethiopia, limited research quantifies urban household water consumption and informs effective interventions. This study addresses this gap by focusing on Adama city. The research questions are: 1) Where are the water sources and how do they support urbanization?; 2) What factors influence residential water consumption?; 3) How do water consumption patterns and reliability vary across neighborhoods?; and 4) What water-sensitive interventions are recommended? The study contributes by providing baseline data on water consumption, examining influential variables in a developing country context, and demonstrating the reliability of machine learning for predicting water consumption, exceeding traditional linear regression models. The overarching aim is to contribute to informed decision-making for improved water management in Adama city and similar contexts.

Existing studies on residential water consumption often focus on limited variables or specific communities, particularly in developing nations. Research in developed countries is more extensive. This study addresses this imbalance by comprehensively investigating various factors, including socioeconomic (family size, housing quality, income, number of rooms, legal status of parcel), climatic (temperature, rainfall), and topographical (elevation, slope) variables. Previous studies have shown mixed results regarding the influence of factors like income and family size on water consumption. This research seeks to provide a more nuanced understanding of these relationships within the specific context of Adama city.

This study employed a mixed-methods approach, combining top-down and bottom-up data collection. Top-down data included city-level water consumption records from Adama City Water Supply and Sewerage Service Enterprise, providing information on water production, consumption, and non-revenue water across sectors. Bottom-up data was gathered via a 100% response rate household questionnaire survey (n=400) which collected information on monthly water consumption (from bills), socioeconomic characteristics, water-saving practices, and geographic location. The survey employed a seven-point Likert scale to assess water conservation behaviors. Spatial data, including topographic features (digital elevation model, slope, aspect, topographic position index, topographic ruggedness index) and climatic data (monthly minimum and maximum temperature, annual rainfall) from the National Metrology Institute of Ethiopia, were integrated using GIS. Socioeconomic data were rasterized using Kriging interpolation. A random forest regression (RFR) model in R software was then used to predict daily household water consumption, using 16 predictor variables (socioeconomic, topographic, climatic). The model was trained and tested using a 90:10 split, and 10-fold cross-validation was repeated 5 times. Model performance was evaluated using metrics including R-squared, RMSE, MAE, and hydroGOF.

Adama city relies primarily on the Awash River for water, with limited groundwater sources due to declining capacity and poor water quality. The water distribution network covers only 45% of the city's master plan, leading to significant unmet water demand (38%). Average daily water consumption is 69 liters per person, below the national standard. The study found that residential areas consume 73.15% of the total water supply. Significant spatial variations in water consumption were observed, with intermediate settlements showing the highest per-household consumption, and the city center having the highest per-capita consumption. The RFR model showed a strong predictive ability (R-squared of 0.77), with household size, housing quality, income, and number of rooms identified as the most important predictors. Formal housing parcels exhibited significantly higher water consumption compared to informal ones. Water conservation practices were variable, with some positive behaviors (e.g., turning off taps), but limited adoption of water-saving devices and alternative water sources. The predicted minimum water consumption per household is 229–455 liters/day, and the maximum is 682–909 liters/day.

The findings highlight the urgent need for comprehensive water management strategies in Adama city. The reliance on a single water source is unsustainable, necessitating the development of alternative sources (e.g., rainwater harvesting) and improved infrastructure. The strong influence of socioeconomic factors underscores the importance of targeted interventions to improve water-saving behaviors in households, especially those in formal settlements with higher consumption. The RFR model's success in predicting water consumption demonstrates the potential of machine learning for informing targeted interventions. The spatial variations in consumption patterns call for localized solutions, tailored to the specific needs of different urban neighborhoods.

This study provides valuable baseline data on water consumption in Adama city and identifies key factors influencing household water use. The application of RFR effectively predicted water consumption and revealed spatial heterogeneity. The findings support recommendations for improving water conservation behaviors, diversifying water sources, promoting a fit-for-purpose water system, and transitioning towards a water-sensitive city. Future research could expand the study to other Ethiopian cities to enhance generalizability and develop more robust water management strategies.

This study focused solely on Adama city, limiting the generalizability of the findings to other contexts. Future research should investigate other urban areas to validate the findings and explore the influence of additional factors. The reliance on self-reported data for water conservation practices might introduce some bias. More objective measures of water use could strengthen future studies.