Une méthode pour anticiper les mises en alerte de crues sur la rivière Thoré (France) Method for anticipating flood warnings for the Thoré river (Southern France)

Résumé

Cet article présente les conclusions d'une recherche visant l'amélioration de l'annonce des crues, et son application à la rivière Thoré, dans le contexte du système d'alerte français. On y exploite les informations météorologiques contenues couramment dans les bulletins d'alerte aux précipitations [BAP] émis par Météo-France, dans le but d'aider les prévisionnistes du Service d'annonce de crues [SAC] à anticiper l'atteinte de la cote d'alerte sur une rivière. Le travail présenté fait partie d'une approche visant à munir les SAC d'outils prévisionnels fonctionnant en temps réel et aptes à prévenir d'une mise en alerte probable. L'approche préconisée conduit à une utilisation directe des informations contenues dans les BAP reçus des services de météorologie dans le processus de surveillance des crues. C'est au moyen de courbes d'intensité-durée-temps d'alerte [IDTA], préalablement établies pour des prévisions de pluies uniformément réparties, et de courbes d'intensité-superficie-temps d'alerte [ISTA] pour les prévisions relatives à des cellules orageuses localisées, que l'approche proposée est développée.

Mots-clés:

• Crues,
• prévision,
• alerte,
• inondation,
• précipitation,
• France,
• modèles neuronaux,
• modèle hydrologique,
• crues rapides

Abstract

This work was designed to contribute to the improvement of flood forecasting, in the context of the French alert system. We propose that the meteorological information contained in the French weather forecast bulletin (Bulletins d'alerte aux précipitations; BAP), produced by Météo-France (French meteorological organization), should be utilized in order to aid the forecasters of the French flood forecasting agencies (Services d'annonce de crues; SAC) to anticipate the timing of an alert associated with an increase in the water level of a river. The goal was to develop an approach to provide the SAC with a real-time operational forecasting tool in order to improve the evaluation of a probable Flood Alert decision. This approach integrates the information contained in the BAP received from Météo-France into the existing flood control process with the use of Duration-Intensity-Warning Time (durée-intensité-temps d'alerte, IDTA) curves for uniform rainfall forecasting, and Intensity-Area-Warning Time curves (intensité-superficie-temps d'alerte, ISTA) for localized storm cells.

The rainfall parameters considered were the intensity (I, mm/h), the duration (D, h), and the area of the watershed affected by the rainfall (S, km²). These parameters are related by the equation V=I x D x S, where V is the volume of rain (hm3). The parameter directly related to the Flood Alert decision is the warning time (T_alerte), measured in hours. It is defined as the time from the beginning of the rainfall to the time when the flow at the watershed outlet reaches the alert flow (Q_alerte in m³ /s), regardless of the maximum discharge (Q_max). Although Qmax may be an important indicator of the magnitude of the upcoming event, the chief concern is the Flood Alert decision, and therefore, the time to alert parameter (T_alert) is of primary importance.

The proposed approach involves creating a graphical connection of a series of rainfall intensity values (I) as a function of a range of rainfall (D) duration values with time to alert (T_alert) curves, which represent the I-D couples. As a result, a SAC forecast agent that receives a BAP indicating the quantitative precipitation forecast in a precise region for a defined period will be able to evaluate the time after the start of the rainfall that the alert flow (Q_alert) will be reached, simply by referring to the IDTA and/or ISTA curves. If an alert is foreseen within a certain delay, the flood forecast agent can wait to receive improved forecasts before making the decision whether to start the flood alert procedures or not.

The construction of the IDTA and ISTA curves requires numerous simulations in order to cover a wide variety of intensity-duration and intensity-area of rainfall couples for which the alert flow (Q_alert) will be reached at the watershed outlet, and therefore the time corresponding to this discharge can be estimated. The simulations were performed through the use of a combination of a deterministic distributed parameter hydrological model and a hydraulic one-dimensional hydrograph transfer model. The neuronal models of the Generalized Regression Network (GRNN) type were also used. This allowed for the extraction of and/or interpolation between values in the database containing the parameters intensity, duration, area, and the hydrographs that resulted from the simulations done with the first two models. The interest in using the GRNN model is to cover a large range of values for all of the parameters considered, without having to simulate all cases, therefore reducing the potential computation time.

We developed this forecasting approach on the basis of a specific case related to the extreme flooding that occurred in southern France on November 1999. More precisely, our case study concerns the mountainous region in the upstream area of the Thoré watershed, in the Tarn Department. The simulation scenarios were 1) uniformly distributed rainfall on a watershed of 208 km² ; 2) storm cells of 9, 36, 64 and 144 km² located in the watershed center; and 3) a storm cell located in various zones of the watershed.

The main observation of the simulation results was that the Talert was constant for a rainfall of intensity I, as long as the duration was longer than the T_alert (i.e., provided it was still raining after Q_alert was attained at the outlet). On the other hand, if the rain stops before Q_alert is attained, T_alert is delayed. T_alert increases as a function of the duration of the rainfall, for a constant I. This is true for both uniform and localized rainfall.

The IDTA and ISTA curves were developed on the basis of several simplifying hypotheses and should be improved in order to increase their precision and flexibility. Therefore this approach can be amended by taking into account the following factors:

- infiltration (when the laws defining it are established);

- the initial conditions: since the results of the simulations for this study are valid for constant initial conditions of Qini=20 m³/s, it would be pertinent to include a correction factor to adjust the results (T_alert) for the real initial conditions such as the actual Qini and the actual soil humidity;

- the spatial variability of the storm cells; and

- the combination of uniformly distributed rainfall and localized storm cells.Nevertheless, we evaluated the use of the forecasting approach with the IDTA and ISTA curves referring to the November 1999 events. The contribution of these curves in the Flood Alert decision process was assessed with a fictitious scenario defined by the issued BAP related to this event. Understanding the simplifying hypotheses discussed above, we conclude that the flood alert on the Thoré River watershed could have been advanced up to seven hours and thirty minutes from the actual time it was issued. In a fast or flash flood event, this range of anticipation could have a considerable impact.

Keywords:

• Flood risk,
• rate of damages,
• 2D modeling,
• Geographical Information System,
• risk management,
• flood risk mapping,
• submersion depth