Regional models for distributed flash-flood nowcasting : towards an estimation of potential impacts and damages

Flash floods monitoring systems developed up to now generally enable a real-time assessment of the potential flash-floods magnitudes based on highly distributed hydrological models and weather radar records. The approach presented here aims to go one step ahead by offering a direct assessment of the potential impacts of flash floods on inhabited areas. This approach is based on an a priori analysis of the considered area in order (1) to evaluate based on a semi-automatic hydraulic approach (Cartino method) the potentially flooded areas for different discharge levels, and (2) to identify the associated buildings and/or population at risk based on geographic databases. This preliminary analysis enables to build a simplified impact model (discharge-impact curve) for each river reach, which can be used to directly estimate the importance of potentially affected assets based on the outputs of a distributed rainfall-runoff model. This article presents a first case study conducted in the Gard region (south eastern France). The first validation results are presented in terms of (1) accuracy of the delineation of the flooded areas estimated based on the Cartino method and using a high resolution DTM, and (2) relevance and usefulness of the impact model obtained. The impacts estimated at the event scale will now be evaluated in a near future based on insurance claim data provided by CCR (Caisse Centrale de Réassurrance).


Introduction
Hydro-meteorological forecasting is an essential component of real-time flood management.It provides crucial information to crisis managers to anticipate, locate and quantify the floods which will hit the areas at risk.In the particular case of flash floods which may affect watersheds of limited extent spread over the territory, suitable forecasting systems are still currently under development over the world.The first proposed methods were focussed at gauged stream sections where rainfallrunoff models could be calibrated [1,2].The most recent developments aim at providing forecasts also at ungauged locations and often rely on highly distributed hydrological models and on radar based QPEs or rainfall nowcasts as input information [3,4,5,6,7].Such models provide indications of possible flood magnitudes, but are still rarely designed to directly account for the possible associated impacts which is the information needed by the crisis managers in real time to take appropriate decisions such as allocation of rescue means.The translation of flood magnitude into local impacts requires, indeed, a detailed knowledge of the flood extent related to the forecasted discharges as well as the exposure and vulnerability of the considered areas.This may be well known at a very local scale (local authorities, inhabitants of flood prone areas,..), but much more difficult to assess and incorporate at the much larger scale at which hydrometeorological forecasting systems are generally implemented.Considering that these systems are designed to monitor a very large number of small rivers spread over large territories, a large number of simultaneous alarms may be generated in case of a significant rainfall event.Thus, providing directly information on local vulnerabilities and associated possible impacts would probably be helpful for coordination managers to have a faster evaluation of the situation and to focus their actions on the most problematic situations.
The approach developed in this paper aims at a direct evaluation of flash flood impacts on inhabited areas based on a complete hydro-meteorological, hydraulic and impact assessment simulation chain.The question of flash flood impact prediction has already been addressed in some previous works which were up to now mainly focused on road inundation risks [5].The methodology developed herein is based on a comprehensive analysis of the study area in order to build an impact model for each river reach incorporated in the hydro-meteorological simulation chain.This analysis is based on simplified 1-D hydraulic simulations to evaluate the extent of the flooded areas for different discharge levels.Land use databases are then used to evaluate the number of  ( 2016) buildings in the estimated flooded area.Based on these precomputed values, a relation between the discharge and the amount of affected buildings is adjusted for each considered river reach to be used for the real-time forecasts.
Even if the approach developed may appear relatively straightforward, its application on a very detailed stream network including small watersheds prone to flash floods, may be too complex to enable an application at a large scale (more than 100.000km²).Therefore, the challenge has been here to define a simplified and automatic procedure for the elaboration of the impact model, with the objective to limit its implementation time to a reasonable level and to limit also the needs for manual corrections.This objective of simplification has been considered as a priority, placed before the accuracy of the computation of flooded areas and associated possible impacts.In other words, a decrease of the quality of the impact model has been accepted to guaranty its applicability at a large scale including a very detailed stream network.
This article presents the proposed method and its application on a test case study.The accuracy of the impact model, is tested against data of recent floods having recently affected the considered area.The article is organised as follows: the first section presents the application case study and the datasets, the next section presents the methodology developed for the fast computation of flood extents and the associated impact model.This section also describes the procedure used for the evaluation the results.The results are presented in section 4 and discussed in section 5, in which the future perspectives in terms of validation are also presented.Lastly, section 6 presents the conclusions of this work.

Figure 2 .
Figure 2. River network considered in the APSFR of Alès (5 km² upstream catchment surface), coverage of reference flood maps and position of available stream gauges.

Figure 3 .Figure 4 .Figure 5 .EFigure 6 .Figure 7 .EFigure 8 .Figure 9 .Figure 10 .5DOIFigure 11 .
Figure 3. Overall principle of the computation of flood maps based on CartinoPC software: a) input information (position of river streams and approximate extent of flood area), b) position of profiles for the 1-D hydraulic model, c) computation of water levels (1-D hydraulic model), c) map of flooded areas and water depths obtained after post-treatment..

Figure 12 .DOI
Figure 12.Ratios between number of claims and number of policies included in the CCR database within (and outside) estimated flooded areas.