Flood modelling in Tunisia: On the suitability of a large-scale hydrological model for flood forecasting at basin scale

Maghreb countries, like the rest of the Mediterranean region, are vulnerable to flood events which often cause disastrous damages and a large number of fatalities. In Europe, this problematic has been addressed by the implementation of the Copernicus European Flood Awareness System (EFAS) that, together with the national and regional flooding schemes, provide a robust tool for flood forecasting. Nevertheless, Maghreb countries do not have such national or regional flooding schemes and, although EFAS covers their northern territories, its forecast capability for these regions is limited as its hydrological model (LISFLOOD) remains uncalibrated due to data unavailability. As data become available, daily river discharge data of 21 Tunisian basins from 1980 to 2018 was used to implement and compare different flood modelling strategies including LISFLOOD and simpler models such as GR4J and IHACRES, which were calibrated for each basin separately. The LISFLOOD model was first implemented with its default parametrization to the 21 basins considered using both, the ERA5 dataset, and observed precipitation data from rain-gauges. Although the use of observations slightly increases the model performance, the performances achieved are substantially lower than with simpler calibrated hydrological models (i.e. GR4J and IHACRES); whereas these simpler models generally present KGE values over 0.4, just four out of the 21 catchments have positive KGE values when discharge is simulated with LISFLOOD.

The model sensitivity to six of its main parameters (Xinanjiang, preferential flow, upper groundwater time constant, lower groundwater time constant, percolation and Manning’s coefficient) was assessed through the application of the Latin hypercube sampling (LHS) scheme. The LHS was used to generate 1000 near-random samples of LISFLOOD parameters sets, to investigate the model sensitivity to these parameters within the 21 basins. This process was repeated constraining the parameter range progressively in order to achieve an optimal parameter set for each catchment, as well as an additional parametrization that could be used in all catchments while resulting into satisfactory performances. Additionally, a Sobol sensitivity analysis was conducted to further investigate the sensitivity of the parameters mentioned above. This analysis revealed that, for extreme discharge values, for extreme discharge values, the most sensitive parameters are the Upper and Lower groundwater time constants and the exponent in Xinanjiang equation for the soil infiltration capacity. Different calibration and validation experiments were carried out with different objective functions, in order to identify the best parameters sets suitable for flood modelling at regional scale.