Quantifying the Impact of Observation Operators on Flood Inundation Forecast Quality

Abstract:

The accuracy of flood forecasts can determine the efficacy of the preparedness and response to extreme events. Unfortunately, the uncertainty in model inputs and parameters leads to inherently erroneous flood inundation simulations, thus prohibiting their operational use to inform decision-making during crisis situations. Literature has demonstrated the potential improvement from assimilating flood extent observations into hydraulic models to reduce the uncertainty in the resulting forecasts. In this context, the observation operator chosen to combine the models and observations at each assimilation time step is crucial to the efficacy of the model-data integration. This study, therefore, investigates the impact on the hydraulic inundation forecast quality of two recently proposed likelihood functions for flood extent assimilation using particle filters. The hydraulic model LISFLOOD-FP was implemented for the 2011 flood event in the Clarence Catchment, Australia and the 2007 event in the Severn Catchment, UK, with synthetic flood extents assimilated into the model using two different observation operators. Inflow uncertainties propagated from precipitation forecasts were considered as the primary source of errors in the open loop model ensemble, with the truth model set up using gauge-observed inflows and calibrated channel friction assumed to be perfect or error-free. The impact of assimilating flood extents using each model likelihood function on the channel water levels and floodplain inundation extent and depth was investigated. Preliminary results indicate that the optimum observation operator depends on catchment morphological characteristics and flooded area coverage. The conclusions from this study support the objective and optimum selection of observation operators for flood extent assimilation problems, such that the benefits of this model-data integration exercise can be translated effectively into improved forecast quality.