Real-time data assimilation in urban rainfall-runoff models

Abstract

Real-time control of urban rainfall-runoff systems can help limit flooding, and minimise combined sewerage overflow. To improve the ability of runoff models to inform this control decision, a data assimilation methodology is presented where downstream prediction errors are used to update upstream model states at an earlier time step. The methodology led to improved, ‘corrected’ predictions after model re-propagation to the current time, and improved discharge forecasts. Assimilation performance was sensitive to the update lag time, and the presence of control structures in the model, which affect the ability of assimilation procedures to map observation information to state space.