Constraining coupled hydrological-hydraulic flood model by past storm events and post-event measurements in data-sparse regions

Flood modelling in data-sparse regions have been always limited to empirical, statistical or geomorphic approaches that are suitable to produce regional hazard maps. Such coarse resolution maps are not adapted for basin-scale applications, small to medium sized basins (< 1000 km(2)), especially when detailed estimates of flows and water levels of a particular event is required and hence cannot replace the hydrological/hydraulic modelling. The latter is a challenging task in data-sparse regions characterized by floods of typical duration times of a few hours which offer little opportunity for real-time recording by traditional rain-gauge networks, remote sensing or satellite imaging. Such data sparseness is not always compatible with the resolution, in both space and time, of the hydrological and hydraulic models. We propose a framework for flood modelling using sparse data from a coupled hydrological-hydraulic model constrained by past storm events and post-event measurements in space. The approach is applied to the Awali river basin (301 km(2)), in Lebanon, particularly to simulate the investigated early January 2013 extreme flood event, which is considered one of the largest events in the last three decades. The hydrological model was calibrated and evaluated with 12 past storm events aiming at defining narrow parameter ranges and uncertainty was performed with Monte Carlo simulations for these parameter ranges. The hydraulic model, based on a fine resolution DEM, was simulated using hydrological outflows and validated with 27 post-event measurements in space of high water marks. The resulting outflow values were satisfactory, and uncertainty was reduced when compared with arbitrarily wide parameter ranges. The hydrological model performance was highly variable but for the hydraulic model, 93% of the observed water levels fall within the simulated uncertainty bounds with an RMSE error of 0.26 m. The proposed framework allows mapping the possible inundation and can be compared to other approaches dealing with model complexity and associated performances.