Electronic sensors to monitor functionality and usage trends of rural water infrastructure in Plateau State, Nigeria

Internet-connected sensor technologies have recently been used to monitor water service infrastructure in remote settings. In this study, 397 groundwater pumps were observed in Plateau State, Nigeria over 12 months in 2021. Two hundred of these sites were instrumented with remotely reporting electronic sensors, including 100 hand-pump sensors, 50 electrical pump sensors, and 50 cistern water-level sensors. Every two months, phone calls and site visits were used to collect a ground-truth of pump functionality: whether the pump was capable of delivering water, regardless of actual use. Our study examined: (1) What are the operating characteristics and trends of these different kinds of water pumps?; (2) Can water-point functionality be predicted with electronic sensors?; and (3) Does the instrumented water-point sample accurately represent average water-system functionality across the region? An automated classifier generated functional/non-functional diagnostics for instrumented pumps on a weekly basis. Classifier diagnostics were compared to ground-truth data, showing an overall accuracy of 91.7% (96.1% for hand-pumps, 63.9% for cisterns, and 93.2% for electrical boreholes), with high fleet-wide sensitivity in correctly identifying a functional pump (94.4%), but poor overall specificity in correctly identifying a non-functional, broken pump (25.0%). This discrepancy is attributable to the sensors' difficulty in distinguishing between a broken pump and an unused pump. Varied patterns were seen in pump usage as a function of rainfall, with hand-pump use decreasing significantly, electrical pump usage decreasing to a lesser degree, and cistern use increasing in response to local rainfall. A comparison of the 200 instrumented to 197 non-instrumented sites showed statistically similar repair and failure rates. The high overall accuracy of the sensor-diagnostic system-and the demonstration that sensor-instrumented sample sites can represent population-level breakdown and repair frequencies-suggests this technology's utility in supporting sample-based monitoring of overall water pump functionality and water volume delivery. However, the poor performance of the system in distinguishing between broken and unused pumps will limit its ability to trigger repair activities at individual pumps.