Automated hydrologic forecasting using open-source sensors: Predicting stream depths across 200,000 km2

Wireless sensor networks support decision-making in diverse environmental contexts. Adoption of these networks has increased dramatically due to technological advances that have increased value while lowering cost. However, real-time information only allows for reactive management. As most interventions take time, predictions across these sensor networks enable better planning and decision making. Prediction models across large water level and discharge sensor networks do exist. However, they have limitations in their accessibility, automaticity, and data requirements. We present an open-source method for automatically generating computationally cheap rainfall-runoff models for any depth or discharge sensor given only its measurements and location. We characterize reliability in a real-world case study across 200,000 km${}^2$, evaluate long-term accuracy, and assess sensitivity to measurement noise and errors in catchment delineation. The method’s accuracy, computational efficiency, and automaticity make it a valuable asset to support operational decision making for diverse stakeholders including bridge inspectors and utilities.