Article de recherche / Research Article

Prévision hydrométéorologique opérationnelle à EDF-DTG – Progrès récents et état des lieux en 2018

Operational hydrometeorological forecast at EDF-DTG – recent progress and current situation in 2018

¹ EDF-DTG, 14, rue Paul-Mesplé, 31100 Toulouse, France
² EDF-DTG, 21, avenue de l'Europe, BP 41, 38040 Grenoble cedex 9, France

^* Correspondance : laetitia.moulin@edf.fr

Reçu : 14 Février 2018
Accepté : 8 Janvier 2019

Résumé

À la fois pour aider au passage des crues mais aussi pour optimiser sa production hydroélectrique, EDF s'est dotée dans les années 1970 de centres hydrométéorologiques internes afin d'assurer une surveillance des débits et de réaliser des prévisions hydrométéorologiques. Depuis leur création, ces centres ont vu leur activité progresser régulièrement, intégrant au fil des ans de nouveaux enjeux, profitant des avancées méthodologiques réalisées, mettant en œuvre de nouvelles organisations tout en faisant évoluer les supports de diffusion et les informations communiquées. La chaîne de prévision opérationnelle est présentée, avec une attention particulière portée sur les traitements automatiques et sur la place de l'expertise humaine dans les prévisions probabilistes journalières de débits et les prévisions déterministes horaires. Enfin, une revue des principales pistes de préparation de l'avenir est réalisée.

Abstract

To help managing the floods but also to optimize its hydroelectric production, EDF set up internal hydrometeorological centers in the 1970s to monitor flows and produce hydrometeorological forecasts. Since their creation, these centers have seen their activity evolve, integrating new challenges over the years, taking advantage of methodological advances made, implementing new organizations while developing new distribution supports and improving communication. The first step of the operational forecast chain is the analysis of observation data, either climatologic or hydrologic. Then, after a study of the meteorological context, allowing the forecasters to evaluate the uncertainties, they build a precipitation and air temperature scenario to feed the hydrological model. Through this rainfall-runoff model, a streamflow forecast is obtained, that can still be expertized before diffusion to the end-user. Probabilistic daily forecasts at 14 days, statistically calibrated, are automatically elaborated. Then, these are refined by deformation of distribution's synthetic variables. Very useful for various uses, these forecasts allow, a minima, an anticipated detection of a flood event risk of occurrence. In most cases, daily streamflow forecast is sufficient for reservoir volume management, but in case of a rapid evolution, the hourly forecast can be helpful. Thus, the generalization of hourly flow forecast is a major stake for EDF. It results in: (i) the design of an hydrological model adapted to this time step, whose free parameters are then calibrated on more than 200 catchments; and (ii) an enriched communication to represent uncertainties on flood dynamic, volumes, gradients and maximal values of forecasted streamflow. The hourly forecasts are then widely diffused to the end-users. At the same time, in 2008, a precise communication protocol, based on probabilities, has been conceived by EDF to warn in anticipation the dam managers of a potential threshold overrun. Thus, the expressions “Fort Risque” (“greater risk”) and “Risque” (“risk”) were used, communicated and explained, and are today an integral part of the diffused information. So, the hydrometeorological forecast is in permanent evolution at EDF. The next challenges are (i) the realization of expertized probabilistic hourly forecasts and (ii) the production of a more significant number of forecasts (currently 30,000 forecasts a year) answering the end-user expectations, mainly in flood period.