Enhancing forecasts of peak runoff events. Application of a feature engineering approach using X-Band radar data

Abstract

Floods cause significant damage to human life, infrastructure, agriculture, and the economy. Predicting peak runoffs is crucial for hazard assessment, but it's challenging in remote areas like the Andes due to limited hydrometeorological data. To address this, we used remote sensing (RS) data, particularly radar-derived precipitation data. Employing the Random Forest (RF) (Machine Learning technique) in combination with a Feature Engineering (FE) strategy we leverage the RS product. This approach allowed us to incorporate physical knowledge into the RF models, enhancing their forecasting performance. The FE strategy is based on an object-based approach, which derive precipitation characteristics from RS data. These characteristics served as inputs for the models, distinguishing them as 'enhanced models' compared to 'referential models' that incorporate precipitation estimates from all available pixels (1210) for each hour. We utilized hourly precipitation radar and runoff data from 29 peak runoff events in a catchment situated in the Ecuadorian Andes. The enhanced models achieved Nash-Sutcliffe efficiencies ranging from 0.94 to 0.50 for lead times between 1 and 6 hours. A comparative analysis between the enhanced and referential models reveals a remarkable 23% increase in NSE efficiency at the 3-hour lead time, which marks the peak improvement. The enhanced models integrated new data into the RF models, resulting in a more accurate representation of precipitation and its temporal transformation into runoff.