Search

The French State services in charge of flood forecasting supervise about 22,000 km among the 120,000 km of the French rivers within a warning procedure called Vigilance Crues (http://www.vigicrues.gouv.fr). Some recent dramatic flood events on small watershed not covered by Vigilance Crues highlight the need for a new warning procedure to anticipate violent flash floods that regularly affect rapid river-basins. Thus the concept emerged of an automatic warning service specifically dedicated to local crisis managers. This service will be less elaborated than Vigilance Crues, probably with false alarms and missed events sometimes, but it will deliver a first information. The generation of the warning is based on a simple rainfall-runoff hydrological model developed by Irstea on all French rivers, fed with radar-gauge rainfall grids provided by Meteo-France. Every fifteen minutes, the hydrological model estimates the discharges on the rivers eligible to the service and determine if certain thresholds corresponding to a high or very high flood are likely to be exceeded. The last step of the real-time system is to determine which municipalities are concerned with flood risk and send them an automatic warning by voice call, optionally by sms or email. A specific web interface is available for users to monitor the evolution of the flood risk on maps that are updated every 15 minutes. This new flash flood warning service will be operational early 2017 as a free service for about 8,000 French municipalities.

Occurring at small temporal and spatial scales, flash floods (FF) can cause severe economic damages and human losses. To better anticipate such events and mitigate their impacts, the French Ministry in charge of Ecology has decided to set up a national FF warning system over the French territory. This automated system will be run by the SCHAPI, the French national service in charge of flood forecasting, providing warnings for fast-responding ungauged catchments (area ranging from ~10 to ~1000 km2). It will therefore be complementary to the SCHAPI’s national “vigilance” system which concerns only gauged catchments. The FF warning system to be implemented in 2017 will be based on a discharge-threshold flood warning method called AIGA (Javelle et al. 2014). This method has been experimented in real time in the south of France in the RHYTMME project (http://rhytmme.irstea.fr). It consists in comparing discharges generated by a simple conceptual hourly hydrologic model run at a 1-km2 resolution to reference flood quantiles of different (e.g., 2-, 10- and 50-year) return periods. Therefore the system characterizes in real time the severity of ongoing events by the range of the return period estimated by AIGA at any point along the river network. The hydrologic model ingests operational rainfall radar-gauge products from Météo-France and takes into account the baseflow and the initial soil humidity conditions to better estimate the basin response to rainfall inputs. To meet the requirements of the future FF warning system, the AIGA method has been extended to the whole French territory (except Corsica and overseas French territories). The calibration, regionalization and validation procedures of the hydrologic model were carried out using data for ~700 hydrometric stations from the 2002-2015 period. Performance of the warning system was evaluated with various contingency criteria (e.g., probability of detection and success rate). Furthermore, specific flood events were analysed in more details, by comparing warnings issued for exceeding different critical flood quantiles and their associated timing with field observations. The performance results show that the proposed FF warning system is useful, especially for ungauged sites. The analysis also points out the need to account for the uncertainties in the precipitation inputs and the hydrological modelling, as well as include precipitation forecasts to improve the effective warning lead time.

The dreadful floods of 1999, 2002 and 2003 in South of France have alerted public opinion on the need for a more efficient and a further generalized national flood-forecasting system. This is why in 2003 Irstea and Meteo-France have implemented a new warning method for flash floods, including on small watersheds, using radar rainfall data in real-time: the AIGA method. This modelling method currently provides real-time information on the magnitude of floods, but doesn’t take into account the elements at risk surrounding the river streams. Its benefit for crisis management is therefore limited as it doesn’t give information on the actual flood risk. To improve the relevance of the AIGA method, this paper shows the benefits of the combination of hydrological warnings with an exposure index, to be able to assess the risk of flood-related damage in real time. To complete this aim, this work presents an innovative and easily reproducible method to evaluate exposure to floods over large areas with simple land-use data. For validation purpose, a damage database has been implemented to test the relevance of both AIGA warnings and exposure levels. A case study on the floods of the 3rd October 2015 is presented to test the effectiveness of the combination of hazard and exposure to assess the risk of flood-related damage. This combination seems to give an accurate overview of the streams at risk, where the most important amount of damage has been observed after the flood.

The absence of validation or comparison data for verifying flood mapping methods poses a significant challenge in developing operational hydraulic approaches. This article aims to address this gap by presenting a benchmark dataset for flash flood mapping in the French Mediterranean region. The dataset described in this paper (Nicolle et al., 2024) includes flood hazard maps and simulation results of three actual flash flood events, all computed in steady regime at a 5-meter resolution using a 2D SWE model (neglecting inertia) named Floodos (Davy et al., 2017). Additionally, it includes the input data necessary (Digital Terrain Models, inflow discharges, hydrographic network) for conducting similar simulations with other hydrodynamic modeling approaches, in both steady and unsteady regimes. A comprehensive validation dataset, comprising observed flood extents, high water marks, and rating curves, is also provided, enabling a detailed evaluation of 2D hydraulic simulation results. The simulation results from Floodos, compared against stage-discharge rating curves available at gauging stations, yielded highly encouraging outcomes. The median error (sim. - obs.) was -0.04 m for the 2-year return period and -0.14 m across all simulated return periods, ranging from 2 to 1000 years. [ABSTRACT FROM AUTHOR]

Estimating spatially distributed hydrological parameters in ungauged catchments poses a challenging regionalization problem and requires imposing spatial constraints given the sparsity of discharge data. A possible approach is to search for a transfer function that quantitatively relates physical descriptors to conceptual model parameters. This paper introduces a Hybrid Data Assimilation and Parameter Regionalization (HDA‐PR) approach incorporating learnable regionalization mappings, based on either multi‐linear regression or artificial neural networks (ANNs), into a differentiable hydrological model. This approach demonstrates how two differentiable codes can be linked and their gradients chained, enabling the exploitation of heterogeneous data sets across extensive spatio‐temporal computational domains within a high‐dimensional regionalization context, using accurate adjoint‐based gradients. The inverse problem is tackled with a multi‐gauge calibration cost function accounting for information from multiple observation sites. HDA‐PR was tested on high‐resolution, hourly and kilometric regional modeling of 126 flash‐flood‐prone catchments in the French Mediterranean region. The results highlight a strong regionalization performance of HDA‐PR especially in the most challenging upstream‐to‐downstream extrapolation scenario with ANN, achieving median Nash‐Sutcliffe efficiency (NSE) scores from 0.6 to 0.71 for spatial, temporal, spatio‐temporal validations, and improving NSE by up to 30% on average compared to the baseline model calibrated with lumped parameters. Multiple evaluation metrics based on flood‐oriented hydrological signatures also indicate that the use of an ANN leads to better performances than a multi‐linear regression in a validation context. ANN enables to learn a non‐linear descriptors‐to‐parameters mapping which provides better model controllability than a linear mapping for complex calibration cases. Key Points: Novel approach for regional calibration of a distributed hydrologic model using learnable and non‐linear descriptors‐to‐parameters mappingsOriginal combination of numerical adjoint model and neural network Jacobian: accurate gradients enable high‐dimensional optimizationExtensive case study in flash‐flood‐prone Mediterranean region shows effective regionalization of high‐resolution model with neural network [ABSTRACT FROM AUTHOR]

Premier risque naturel en France, les inondations peuvent entraîner des pertes humaines et provoquer d’importants dégâts matériels. Parmi elles, les crues soudaines causées par des pluies très fortes enregistrées en très peu de temps restent encore difficiles à anticiper et à gérer. La méthode AIGA présentée dans ce article est un outil destiné à mieux identifier ces pluies intenses et anticiper ces crues rapides, notamment quand elles se manifestent au sein de petits bassins versants non équipés de station de mesure. Après une présentation des principes de la méthode, les auteurs s'intéressent à sa mise en oeuvre dans des dispositifs opérationnels de surveillance et d'anticipation des aléas hydrométéorologiques.

Prediction in ungauged basins (PUB), where flow measurements are unavailable, is a critical need in hydrology and has been a focal point of extensive research efforts in this field over the past two decades. From the perspective of deep learning, PUB can be viewed as a scenario where the generalization capability of a pretrained neural network is employed to make predictions on samples that were not included in its training data set. This paper adopts this view and conducts genuine PUB using long short-term memory (LSTM) networks. Unlike PUB approaches based on k-fold training-test technique, where an arbitrary catchment B is treated as gauged in k -1 rounds and as ungauged in one round, our approach ensures that the sample for which the PUB is conducted (the UNGAUGED sample) is completely independent from the sample used to previously train the LSTMs (the GAUGED sample). The UNGAUGED sample includes 379 catchments from five hydrological regimes: Uniform, Mediterranean, Oceanic, Nivo-Pluvial, and Nival. PUB predictions are conducted using LSTMs trained both at the regime level (using only gauged catchments within a specific regime) and at the national level (using all gauged catchments). For benchmarking the performance of LSTM in PUB, four regionalized variants of the GR4J conceptual model are considered: spatial proximity, multi-attribute proximity, regime proximity, and IQ-IP-Tmin proximity, where IQ, IP, and Tmin are the indices defining the five hydrological regimes. To align with the study's fully ungauged context, the IQ index, which is also an input feature for the LSTMs, and the regime classification, crucial for the REGIME LSTMs, are reproduced under ungauged conditions using a regime-informed neural network and an XGBoost multi-class classifier respectively. The results demonstrate the overall superior performance of NATIONAL LSTMs compared to REGIME LSTMs. Among the four regionalization approaches tested for GR4J, the IQ-IP-Tmin proximity approach proves to be the most effective when analyzed on a regime-wise basis. When comparing the best-performing LSTM with the best-performing GR4J model within each regime, LSTMs show superior performance in both the Nival and Mediterranean regimes. [ABSTRACT FROM AUTHOR]

Flash floods have dramatic economic and social consequences, and efficient adaptation policies are required to reduce their impacts, especially in a context of global change. Developing more efficient flash flood forecasting systems can largely contribute to these adaptation requirements. The aim of this study was to assess the ability of a new seamless short range (0–6 h) ensemble quantitative precipitation forecast (QPF) product, called PIAF-EPS and recently developed by Météo-France, to predict flash floods when used as input of an operational hydrological forecasting chain. For this purpose, eight flash flood events that occurred in the French Mediterranean region between 2019 and 2021 were reanalysed, using a similar hydrological modeling chain to the one implemented in the French "Vigicrues-Flash" operational flash flood monitoring system. The hydrological forecasts obtained from PIAF-EPS were compared to the forecasts obtained with different deterministic QPFs from which PIAF-EPS is directly derived (i.e. the AROME-NWC numerical weather prediction model, and the deterministic PIAF product). The verification method applied in this work uses scores calculated on contingency tables, and combines the forecasts issued on each 1 km 2 pixel of the territory. This offers a detailed view of the forecast performances, covering the whole river network and including the small ungauged rivers. The results confirm the added value of the ensemble PIAF-EPS approach for flash flood forecasting, in comparison to the different deterministic scenarios considered. [ABSTRACT FROM AUTHOR]

Reducing uncertainty and improving robustness and spatio-temporal extrapolation capabilities remain key challenges in hydrological modeling especially for flood forecasting over large areas. Parsimonious model structures and effective optimization strategies are crucially needed to tackle the difficult issue of distributed hydrological model calibration from sparse integrative discharge data, that is in general high dimensional inverse problems. This contribution presents the first evaluation of Variational Data Assimilation (VDA), very well suited to this context but still rarely employed in hydrology because of high technicality, and successfully applied here to the spatially distributed calibration of a newly taylored grid-based parsimonious model structure and corresponding adjoint, over a large sample. It is based on the Variational Data Assimilation (VDA) framework of SMASH (Spatially distributed Modelling and ASsimilation for Hydrology) platform, underlying the French national flash flood forecasting system Vigicrues Flash. It proposes an upgraded distributed hourly rainfall-runoff model structure employing GR-based operators, including a non-conservative flux, and its adjoint obtained by automatic differentiation for VDA. The performances of the approach are assessed over annual, seasonal and floods timescales via standard performance metrics and in spatio-temporal validation. The gain of using the proposed non-conservative 6-parameters model structure is highlighted in terms of performance and robustness, compared to a simpler 3-parameters structure. Spatially distributed calibrations lead to a significant gain in terms of reaching high performances in calibration and temporal validation on the catchments sample, with median efficiencies respectively of NSE = 0.88 (resp. 0.85) and NSE = 0.8 (resp. 0.79) over the total time window on period p 2 (resp. p 1). Simulated signatures in temporal validation over 1443 (resp. 1522) flood events on period p 2 (resp. p 1) are quite good with median flood (NSE ; KGE) of (0.63; 0.59) (resp. (0.55; 0.53)). Spatio-temporal validations, i.e. on pseudo ungauged cases, lead to encouraging performances also. Moreover, the influence of certain catchment characteristics on model performance and parametric sensitivity is analyzed. Best performances are obtained for Oceanic and Mediterranean basins whereas it performs less well over Uniform basins with significant influence of multi-frequency hydrogeological processes. Interestingly, regional sensitivity analysis revealed that the non conservative water exchange parameter and the production parameter, impacting the simulated runoff amount, are the most sensitive parameters along with the routing parameter especially for faster responding catchments. This study is a first step in the construction of a flexibe and versatile multi-model and optimization framework with hydbrid methods for regional hydrological modeling with multi-source data assimilation. [ABSTRACT FROM AUTHOR]

In the field of Deep Learning, the long short-term memory (LSTM) networks lie in the category of recurrent neural network (RNN) architectures. The distinctive capability of the LSTM is learning non linear long term dependency structures. This makes the LSTM a good candidate for prediction tasks in non linear time dependent systems such as prediction of runoff in a catchment. In this study, we use a large sample of 740 gauged catchments with very diverse hydro-geo-climatic conditions across France. We present a regime classification based on three hydro-climatic indices to identify and classify catchments with similar hydrological behaviors. We do this because we aim to investigate how regime derived information can be used in training LSTM-based runoff models. The LSTM-based models that we investigate include local models trained on individual catchments as well as regional models trained on a group of catchments. In local training, for each regime, we identify the optimal lookback, i.e. the length of the sequence of past forcing data that the LSTM needs to work through. We then use this length in training regional models that differ in two aspects: 1) hydrological homogeneity of the catchments used in their training, 2) configuration of the static attributes used in their inputs. We examine how each of these aspects contributes to learning of the LSTM in regional training. At every step of this study, we benchmark performances of the LSTM against a conceptual model (GR4J) on both train and unseen data. We show that the optimal lookback is regime dependent and homogeneity of the train catchments in regional training has a more significant contribution to learning of the LSTM than the number of the train catchments. [ABSTRACT FROM AUTHOR]