Kalman Filtering in Flood Early Warning Systems

Flood Early Warning Systems (FEWS) are used for predicting water levels and discharges at specific locations along a river network. Based on these predictions the authorities involved may decide to implement flood mitigation measures or to evacuate people and livestock. Hence it is important that accurate predictions for as long as possible a time-horizon are made. In the past several Flood Early Warning Systems (FEWS) were developed by Delft Hydraulics. Examples are the Nile FEWS in Sudan and the Indus FEWS in Pakistan.

The computational backbone of a FEWS usually comprises an ensemble of mathematical models, viz. a rainfall-runoff model (if applicable including snowmelt), reservoir routing models and a hydrodynamic (i.c. SOBEK) river flow routing model. In order to make flood forecasts it is essential that the computed hydraulic/hydrologic conditions meet the actual hydraulic/hydrologic observations. In order to achieve this a FEWS is regularly updated using available measurements. In the hydrodynamic (SOBEK) part of FEWS, Delft Hydraulics applies Kalman Filtering (KF), a data assimilation technique in which the optimal actual hydraulic condition of the hydrodynamic (SOBEK) model is established by statistically weighing computed and observed water levels and discharges.

Figure 1a: The Flood Forecasting Model of the Orlice (tributary of river Elbe) FEWS in the North-eastern part of the Czech Republic.

Figure 1b: The Orlice FEWS in the North-eastern part of the Czech Republic

Figure 2: The Kalman Filtering algorithm as applied to the Orlice (tributary of river Elbe) is seen to lead to adequate predictions

more information

For more information please contact Thieu van Mierlo or Henk Ogink.