Flood inundation model updating using an ensemble Kalman filter and spatially distributed measurements

This paper examines critically the application to a site along the River Crouch, Essex of a river-flow forecasting approach based on a one-dimensional hydraulic flow simulation model updated using real-time data within an ensemble Kalman filtering framework. Given a specified validation location and forecast period the objective of the forecasting model was to estimate water level more accurately with updating than without. The method used to estimate both model state and state uncertainty was evaluated in terms of its forecast accuracy and representation of forecast uncertainty. The ensemble Kalman filter lead to an increase in forecast accuracy of between 50% and 70% depending on location. The hyperparameters of the filter could be calibrated to make estimates of forecast uncertainty at a specific location, where the most data were available. However, the presence of systematic errors in the simulation model and especially measurement data meant that uncertainty estimates were inaccurate at other locations. Although, the major source of uncertainty in this model came from the boundary condition, additional uncertainty within the model domain was required, particularly between channel and floodplain. Changing the temporal sampling rate and spatial density of samples had little effect on the accuracy of forecasts at this site. However, uncertainty was under-estimated when the temporal sampling rate was decreased, indicating that the relative uncertainties prescribed to the simulation model and measurement model were inadequate