Summary: | The aim of this document is to provide an overview and some results of the calibration work of the LISFLOOD model [De Roo et al., 2000; 2001] that has been carried out during 2005. LISFLOOD forms the core of the European Flood Alert System (EFAS) and is used for impact studies to evaluate the effect of land use and climate changes on the hydrological behaviour across Europe. It is a spatially distributed, partly physically-based hydrological model embedded within a PCRaster GIS environment. The model simulates river discharges in drainage basins as a function of spatial information on topography, soils and land cover. The accuracy of the model predictions depends on the ability of the model to capture the dominating hydrological processes that transfer precipitation into river runoff at the catchment scale, and on its ability to reproduce historical time series of observed river discharges.A crucial step which contributes significantly to the accuracy of the LISFLOOD forecasts and simulations is the calibration of the model for all European catchments. Owing to the general nature of LISFLOOD, its application to any given river basin requires that certain parameters of conceptual functions be identified for the particular basin. In the process of calibration, the values of unknown model parameters are tuned such that the model matches the observed predictions as closely as possible. During the early stages of the EFAS project, the LISFLOOD model has been crudely calibrated, without taking due account of the spatial variability of the parameters over the different hydrological regimes across Europe. A set of 240 parameters realizations was generated, and for large catchments the parameter set was chosen that best reproduced a time series of observed river discharges at the outlet. For ungauged catchments the parameter set that gave the best prediction in most other catchments was used. The underlying assumption was that the 240 parameter realizations were a representative sample of the feasible parameter space. More recently, several detached national experts have been working on a more detailed calibration of the LISFLOOD model for the Danube and Elbe catchments, typically by manually adjusting the parameters while visually inspecting the agreement between the observed and simulated discharges. However, the subjective and time-consuming nature of the trial-and-error method renders this method unappealing for use at a European scale. The large number of catchments for which the model needs to be calibrated calls for an automatic parameter estimation procedure. Besides shortening the implementation time this will also enhance the reliability of the calibrated parameters due to a more exhaustive exploration of the parameter space.An automatic calibration procedure has been developed for LISFLOOD, based on the Shuffled Complex Evolution Metropolis (SCEM-UA) global optimization algorithm [Vrugt et al., 2003]. The algorithm automatically searches through the space of feasible parameter values and finds the parameter values that produce the best model performance. It also yields a posterior parameter distribution, which reflects the residual uncertainty about the model parameters after taking into account the discharge observations. The posterior distribution forms the basis for making probabilistic flow predictions. To overcome the computational burden the optimization has been implemented using parallel computing.The work done on the calibration in 2005 resulted in one paper in the proceedings of the International Conference on Innovation, advances and implementation of flood forecasting technology in Tromsø [Feyen et al., 2005a], an oral presentation at the American Geophysical Union Fall Meeting in San Francisco [Feyen et al., 2005b], an article submitted to Journal of Hydrology [Feyen et al., 2006a], and a manuscript in preparation [Feyen et al., 2006b]. JRC.H.7 - Land management and natural hazards
|