Search

Machine learning models have recently gained popularity in hydrological modelling at the catchment scale, fuelled by the increasing availability of large-sample data sets and the increasing accessibility of deep learning frameworks, computing environments, and open-source tools. In particular, several large-sample studies at daily and monthly time scales across the globe showed successful applications of the LSTM architecture as a regional model learning of the hydrological behaviour at the catchment scale. Yet, a deeper understanding of how machine learning models close the water balance and how they deal with inter-catchment groundwater flows is needed to move towards better process understanding. We investigate the performance and behaviour of the LSTM architecture at a monthly time step on a large sample French data set coined CHAMEAU – following the CAMELS initiative. To provide additional information to the learning step of the LSTM, we use the parameter sets and fluxes from the conceptual GR2M model that has a dedicated formulation to deal with inter-catchment groundwater flows. We see this study as a contribution towards the development of hybrid hydrological models.

Whether they refer to it as validation, verification, or evaluation, hydrological practitioners regularly need to compute performance metrics to measure the differences between observed and simulated/predicted streamflow time series. While the metrics used are often the same (MAE, NSE, KGE, Brier, CRPS, etc.), the tools used to compute them are seldom the same. In some cases, specific tools are not used and the computation of the metrics are directly hand written in the scripts used to analyse model outputs. In addition, the computation of performance metrics is often accompanied with a variety of pre- and post-processing steps that are rarely documented (e.g. handling of missing data, data transformation, selection of events, uncertainty estimation). This can be error prone and hinder the reproducibility of published results. The sharing of tools computing these performance metrics is likely limited by the variety of programming environments in the hydrological community, and by well-established practices in operational environments that are difficult to modify. In order to enable the sharing between researchers and practitioners and move towards more reproducible hydrological science, we argue that an evaluation tool for streamflow predictions must be polyglot (i.e. that it must be usable in several programming languages) and that it must not only compute the performance metrics themselves, but also the pre- and post-processing steps required to compute them. To this end, we present a new open source, polyglot, and compiled tool for the evaluation of deterministic and probabilistic streamflow predictions. The tool, named “evalhyd”, can be used in Python, in R, and as a command line tool. We will present the concept behind its development and illustrate how it works in practice through examples from operational streamflow predictions in France. We will also discuss further steps and remaining challenges in the evaluation of hydrological model predictions.

When they are used for operational forecasting, hydrological models are almost always combined with some kind of updating procedures. Then a question arises: should the model parameters be calibrated with or without the updating procedures? Calibrating with the updating procedures often improves forecast efficiency, but it can also lead to parameter inconsistency and ultimately to a drop in performance in some cases.In this study, we evaluate the pros and cons of making the parameters of a flood forecasting model vary with lead times. We investigate the dependencies of the model parameters to the lead times and determine where and when this procedure significantly improves forecast quality. A modified version of the GR5H hydrological model is used on 229 French catchments where 10,652 events were selected. The model is run at the hourly time step and combined with a simple updating procedure to produce forecasts at four lead times. The model parameters were estimated from a large screening of the parameter space (3 million runs for each catchment). Results show that the parameters related to fast catchment processes are the most dependant on lead times, indicating the need for more specific parameter estimation methods when modelling catchments prone to flash floods.

Accounting for the variability of hydrological processes and climate conditions between catchments and within catchments remains a challenge in rainfall–runoff modelling. Among the many approaches developed over the past decades, multi-model approaches provide a way to take into account the uncertainty linked to the choice of model structure and its parameter estimates. Semi-distributed approaches make it possible to account explicitly for spatial variability while maintaining a limited level of complexity. However, these two approaches have rarely been used together. Such a combination would allow us to take advantage of both methods. The aim of this work is to answer the following question: What is the possible contribution of a multi-model approach within a variable spatial framework compared to lumped single models for streamflow simulation? To this end, a set of 121 catchments with limited influence in France was assembled, with precipitation, potential evapotranspiration and streamflow data at the hourly time step over the period 1998–2018. The semi-distribution set-up was kept simple by considering a single downstream catchment defined by an outlet, and one or more upstream sub-catchments. The multi-model approach was implemented with 13 rainfall–runoff model structures, three calibration options and two spatial frameworks, for a total of 78 distinct modelling options. A simple average method was used to combine the various simulated streamflow at the outlet of the catchments and sub-catchments. The most efficient lumped model on a given catchment was taken as the benchmark for model evaluation. Overall, the semi-distributed multi-model approach yields better performance than the different lumped models considered individually. The gain is mainly brought about by the multi-model set-up, with the spatial framework providing a benefit on a more occasional basis. These results, based on a large catchment set, evince the benefits of using a multi-model in a variable spatial framework to simulate streamflow.

While hydrological models aim to represent the hydrological behaviour of catchments, many of them have been streamlined on the exclusive basis of streamflow simulation performance, i.e. among the possible parameter sets, the 'optimal' is the one which brings the best simulation of streamflow during the calibration period. However, we sometimes encounter 'optimal' sets which perform well in discharge simulation but yield unrealistic simulations of other fluxes (e.g. actual evaporation fluxes, inter-catchment groundwater fluxes). Previous studies tried to constrain the exploration of parameter space with measurements complementary to river discharge: this application of extra information aims to increase the physical realism of the model compared to discharge-only calibration. In this study, we carry out an original investigation to take advantage of the spatial patterns of the complementary data in order to drive the calibration towards a more spatially consistent solution.We propose here a feasibility test, to constrain the spatial consistency of fluxes of a semi-distributed GR model (GRSD). Our study area is the Somme catchment (6100 km2 ) with 17 internal gauging stations, each of them having more than 15 years of discharge measurement. As a first step, we use the long-term actual evaporation from Budyko-estimation as an extra constraint, which has been widely used for describing spatial patterns of climate. In the second step, we develop a criterion describing the spatial consistency between the pattern of measured and simulated fluxes. By constraining the model with extra information, the model is expected to yield a more consistent simulation of fluxes in comparison with the classical calibration practice. Moreover, we analysed the impact of this additional constraint on the spatial organisation of IGF over the catchment as both the other components in water balance analysis, actual evaporation and discharge, are constrained.

The calibration of hydrological models through the use of automatic algorithms aims at identifying parameter sets that minimize the deviation of simulations from observations (often streamflows). It is a widespread technique that has been the subject of much research in the past. Indeed, the choice of objective function (i.e. the criterion or combination of criteria to optimize) can significantly impact the parameter set values identified as optimal by the algorithm. Besides, the actual goal of the model application (flood or low-flow estimation, for instance) influences the way calibration is undertaken. This article discusses how mathematical transformations, which are sometimes applied to the target variable before calculating the objective function, impact model simulations. Such transformations, for example square root or logarithmic, aim at increasing the weight of errors made in specific ranges of the hydrograph. Typically, a logarithmic transformation tends to increase the fit of streamflows to lower values, compared to no transformation. We show in a catchment set that the impact of these transformations on the obtained time series can sometimes be different from what could be expected. Extreme transformations, such as squared or inverse of squared transformations, lead to models that are specialized for extreme streamflows, but show poor performance outside the range of the targeted streamflows and are less robust. Other transformations, such as the power 0.2, the Box–Cox and the logarithmic transformations, can be qualified as more generalist, and show a good performance for the intermediate range of streamflows, along with an acceptable performance for extreme streamflows.

Water management practices (dam management, uptakes for irrigation, industry and domestic water supply or water sanitation) can strongly influence river flow at the catchment scale. If water withdrawals and releases are important (for example in the case of a large dam), this influence can strongly impact the performance of bucket-type rainfall-runoff models whose structure do not explicitly consider water management practices. However, the influence of water management can also be compensated by the calibration of parameters, which is difficult to perceive when the model is conceptual, as the performance may not necessarily be negatively affected. This work aims at evaluating the effect of these compensations over the parameters of the GR4J bucket-type rainfall-runoff model over an influenced catchment. To do so, the model was applied in a semi-distributed way over the rainfed Sèvre Nantaise catchment in Western France. For this catchment, we used uptakes, sanitation and reservoir management data estimated at a daily time step, as well as a dense network of daily flow measurement stations. Water management practices were explicitly taken into account in the GR4J model through the airGR and airGRiwrm R packages that allow injecting or removing water from the river depending on water management practices. The model was calibrated on the different flow stations both with and without taking water withdrawals and releases into account. The obtained parameter values were compared in terms of reliability, robustness, plausibility and spatial homogeneity.

Semi-distributed hydrological models have the potential to improve the efficiency of flood forecasting chains. Such models take into account the spatial distribution of both meteorological forcings and soil moisture states to predict streamflow along the river network and allow the assimilation of streamflow observations on multiple internal flow gauges. How to update the model states on ungauged upstream sub-catchments remains however a challenge. Indeed, the actual relative contribution of each upstream sub-catchment to the observed streamflow at the outlet cannot be observed but simply estimated by the hydrological model from the simulated upstream streamflow and routing-lag. In this work, we test the following hypotheses:the simultaneous assimilation of streamflow observations at internal gauges should improve streamflow predictions at the main downstream outlet accounting for time lags between flow gauges is needed to efficiently update model states in cases where no streamflow observations are available at internal gauges The analysis is performed with a semi-distributed version of the hourly GR5H model (de Lavenne et al., 2019; Peredo-Ramirez et al., 2021) on a large dataset of French gauged catchments, each one having at least one internal gauged station. Several experiments were set up in gauged and pseudo-ungauged contexts to test both hypotheses. Two updating schemes were used: a particle filter (Piazzi et al., 2021) and a direct insertion method used in the operational flood forecasting model GRP (Furusho et al., 2016). Referencesde Lavenne, A., Andréassian, V., Thirel, G., Ramos, M.-H., & Perrin, C. (2019). A regularization approach to improve the sequential calibration of a semidistributed hydrological model. Water Resources Research, 55, 8821–8839, 2018WR024266.Furusho, C., Perrin, C., Viatgé, J., Lamblin, R., and Andréassian, V. (2016). Collaborative work between operational forecasters and scientists for better flood forecasts, La Houille Blanche, 102:4, 5-10.Peredo-Ramirez, D., Ramos, M.-H., Andréassian, V., and Oudin, L. (2021). Investigating hydrological model versatility to simulate extreme flood events. Hydrological Sciences Journal, in review.Piazzi, G., Thirel, G., Perrin, C., and Delaigue, O. (2021). Sequential data assimilation for streamflow forecasting: assessing the sensitivity to uncertainties and updated variables of a conceptual hydrological model at basin scale. Water Resources Research, 57(4), e2020WR028390.

Accounting for the variability of processes and climate conditions between catchments and within catchments remains a challenge in hydrological modelling. To address this issue, various approaches were developed over the past decades. Among them, multi-model approaches provide a way to quantify and reduce the uncertainty linked to the choice of model structure, and semi-distributed approaches propose a good compromise to account for spatial variability of the processes by dividing the catchment in sub-catchments while maintaining a limited level of complexity. However, these two approaches were barely applied together. The aim of this work is to answer the following question: can we improve the efficiency of hydrological models by implementing a multi-model approach within a semi-distributed framework? In this work, the benchmark considered is a lumped model with a fixed structure.To this end, a large set of 147 catchments in France was assembled, with precipitation, evapotranspiration and flow data at an hourly time step over the 1998-2018 period. The semi-distribution set-up was kept simple by considering a single intermediate catchment between a downstream station and one or more upstream catchments. The multi-model approach was implemented with two versions of the GR model (namely GR4H and GR5H). Within a semi-distributed framework, the two models were either used individually, i.e. applied on all sub-catchments (called GR4H-SD and GR5H-SD respectively), or combined using a simple and a weighted mean.The first step of this work was to check whether past conclusions published in the scientific literature, obtained with lumped multi-models, were the same in a semi-distributed framework. In other words, does the multi-model approach generate better performance than individual models in a semi-distributed context?Another possible combination of the semi-distributed and the multi-model approaches would be to make different choices of model structures or combinations on each sub-catchment. Intuitively, it makes sense to propagate the flow simulated by the best model from upstream to downstream. The second analysis therefore focuses on the following question: is the best upstream model always the most useful downstream?The results and the operational implications of this work will be analyzed in the case of the Rhône basin.

High-intensity precipitation events activate heterogeneous catchment processes and trigger flood events for which hydrological models are known to have limited predictive efficiency. The goal of this study is to find a way to inform the structure of a hydrological model on the time variability of dominant processes by using precipitation intensities as a surrogate for the activation of fast heterogeneous runoff processes. Accurately simulating flood generated by high-intensity rainfall events occurring on dry soil-moisture conditions, during or after the summer period, is a particular challenge (Astagneau et al., 2021), typically in southern France in a Mediterranean context.Starting from the concept developed by Peredo et al. (2021), we made three modelling hypotheses on the explicit dependence of the storages and fluxes of a conceptual hydrological model on rainfall intensity rates. The modifications introduced increasingly modify model functioning as rainfall intensity increases, and are neutral in case of low rainfall intensity. We evaluated these hypotheses over a large set of 229 catchments in France representing a large variety of conditions, and we used a selection of 10,652 flood events to evaluate model performance. We used the GR5H model at the hourly time step as structural basis for model improvement. The relevance of our approach was assessed against several event and catchment characteristics (precipitation intensity and spatial variability, antecedent wetness conditions…) characterising very different flood generating processes.Results showed significant improvements in the simulation of summer floods resulting from high-intensity rainfall events in several catchments of the Mediterranean area, while maintaining a limited level of complexity of the model structure.ReferencesAstagneau, P. C., et al., (2021). When does a parsimonious model fail to simulate floods? Learning from the seasonality of model bias. Hydrological Sciences Journal, 66(8), 1288–1305, https://doi.org/10.1080/02626667.2021.1923720Peredo, D., et al., (2021). Investigating hydrological model versatility to simulate extreme flood events. Hydrological Sciences Journal, In Review.

In many countries, rivers are the primary supply of water. A number of uses are concerned (drinking water, irrigation, hydropower…) and can be strongly affected by water shortages. Therefore, there is a need of early anticipation of low-flow periods to improve water management. This is strengthened by the perspective of having more severe summer low-flows in the context of climate change. Several French institutes (BRGM, EDF, INRAE, Lorraine University and Météo-France) have been collaborating to develop an operational tool for low-flow forecasting, called PREMHYCE (Tilmant et al., 2020). This platform produces forecasts in real time on more than 900 catchments in metropolitan France since 2017, in cooperation with French operational services of water management. PREMHYCE includes five hydrological models which can be calibrated on gauged catchments and which assimilate flow observations. Low-flow forecasts based on ensemble streamflow prediction (ESP) can be issued up to either 15 days ahead, based on scenarios from the European Centre for Medium-Range Weather Forecasts (ECMWF) or 90 days ahead, using historical climatic data as ensembles of future input scenarios. These climatic data (precipitation and temperature) are provided by Météo-France with the daily gridded SAFRAN reanalysis on the 1958-2020 period, which includes a wide range of conditions. Outputs from the different hydrological models are combined into a simple multi-model approach to improve robustness of the forecasts. The tool provides text files and graphical representation of forecasted low-flows, and probability to be under low-flow thresholds provided by users. In 2020, a website has been developed allowing easier access to the forecasts for users. The presentation will show the main characteristics of this operational tool and results on the recent low-flow periods. Keywords: low-flow forecasting, hydrological model, ensemble streamflow prediction, water resource managementReferenceTilmant, F., Nicolle, P., Bourgin, F., Besson, F., Delaigue, O., Etchevers, P., François, D., Le Lay, M., Perrin, C., Rousset, F., Thiéry, D., Magand, C., Leurent, T., Jacob, E., 2020: PREMHYCE: an operational tool for low-flow forecasting, La Houille Blanche (5) 37-44, DOI: 10.1051/lhb/2020043

To facilitate reproducible hydrological research and support model testing and evaluation, several datasets gathering hydroclimatic information on large catchment samples have been released in different regions of the world over the last years. Addor et al. (2017) proposed a large dataset of Catchment Attributes and Meteorology for Large-sample Studies (CAMELS), consisting of 671 catchments in the contiguous United States. Then other datasets were produced in the CAMELS framework in Great Britain (CAMELS-GB, 671 catchments), Chile (CAMELS-CL, 516 catchments), Brazil (CAMELS-BR, 897 catchments) and Australia (CAMELS-AUS, 222 catchments). They consist of catchment hydro-meteorological time series and catchment attributes.This presentation aims at introducing CAMELS-FR, a dataset on French catchments that was set up for hydrological studies. This dataset has been assembled at INRAE, France, based on an automatized processing chain of national data products (Delaigue et al. 2020), among which are the SAFRAN atmospheric reanalysis produced by Météo-France (Vidal et al., 2010) and the national river flow archive (Banque HYDRO), maintained by the French centre for flood forecasting (SCHAPI). CAMELS-FR provides daily catchment-scale hydrometeorological time series (streamflow, solid and liquid precipitation, potential evapotranspiration, temperature, etc.) covering the 1958-2020 period. Catchment characteristics such as land cover, topography (i.e. elevation and slope distributions, drainage density, topographic index, etc.) are also provided, with information about possible regulations upstream, and with some basic information on data quality. Graphical summary sheets provide synthetic information on the main characteristics of each catchment. The performance of several lumped rainfall-model models applied on the CAMELS-FR dataset will be presented, in order to highlight potential uses of this dataset for modeling applications. The CAMELS-FR dataset will be made freely available to the scientific community in partnership with data providers (Météo-France and SCHAPI). Références :Addor, N. et al. (2017). The CAMELS data set: catchment attributes and meteorology for large-sample studies, HESS, 21, 5293-5313, https://doi.org/10.5194/hess-21-5293-2017.Delaigue, O. et al. (2020). Database of watershedscale hydroclimatic observations in France. Université Paris-Saclay, INRAE, HYCAR Research Unit, Hydrology group, Antony, https://webgr.inrae.fr/base-de-donnees.Vidal, J.P. et al. (2010). A 50-Year High-Resolution Atmospheric Reanalysis over France with the Safran System. Int. J. Climatol. 30, 11 (2010): 1627‑44. https://doi.org/10.1002/joc.2003.

Real-time flood forecasting and other hydrological applications in mountainous areas require a good understanding and accounting of snow accumulation and melt. The CemaNeige snow model is currently used with the GRP hydrological model by most regional operational services in France to produce floods forecasts with lead times varying from a few hours to a few days. The snow model is based on a degree-day approach and needs limited inputs (precipitation and air temperature) spatialized on altitude bands over the catchment. It was initially developed on snow-dominated catchments by Valéry et al. (2014) using only streamflow series as calibration information. The model was then adapted by Riboust et al. (2019) to better simulate snow-covered areas as estimated by MODIS satellite images. These data were used as a secondary source of information for parameter calibration. All these developments were made at the daily time step. However, for real-time purposes, the outputs from the snow model are often needed at a finer time step, typically hourly or sub-hourly.Here the transposability and the consistency of the CemaNeige model were studied across a range of time steps, from hourly to daily, on a set of snow-influenced catchments. This follows previous works on the hydrological model to improve its consistency across time scales (see e.g. Viatgé et al., 2019). Several questions were addressed:To which extent are the outputs of the snow model and its parameters consistent across various time steps? Is the current snow model complexity (structure and parameters) sufficient to simulate the snow influence at the catchment scale at sub-daily time steps? Can we expect better results by running the snow model at sub-daily time steps than by disaggregating the outputs of the snow model run at the daily time? The answers to these questions will be presented based on a comprehensive testing scheme and a set of numerical criteria. Perspectives in terms of operational use will be discussed.

International audience; Over the past few decades, several naturalization methods have been developed for removing anthropogenic influences from streamflow time series, to the point that naturalized flows are often considered true natural flows in many studies. However, such trust in a particular naturalization method does not expose the assumptions underlying the method, nor does it quantifies the associated uncertainty. This review provides an overview of streamflow naturalization approaches. The terminology associated with naturalization is discussed, and a classification of naturalization methods according to their data requirements and main assumptions is proposed. A large set of studies developing or applying naturalization methods are reviewed, and the main challenges associated with the methods applied are assessed. To give a more concrete example, a focus is made on studies conducted in France over the last decade, which applied naturalization methods to estimate water extraction limits in rivers.

International audience; In many countries, rivers are the primary supply of water. A number of uses (drinking water, irrigation, hydropower…) can be strongly affected by water shortages. Therefore, there is a need for the early anticipation of low-flow periods to improve water management. This is strengthened by the perspective of having more severe low flows in the context of climate change. Several French institutions have been collaborating over the last years to develop an operational tool for low-flow forecasting, called PREMHYCE. It was tested in real time on more than one hundred catchments in continental France since 2017. PREMHYCE includes five hydrological models, which are calibrated on gauged catchments and which assimilate flow observations. Low-flow forecasts can be issued up to 90 days ahead, based on Ensemble Streamflow Prediction (ESP) using historical climatic data as ensembles of future input scenarios. The performance of the forecasting tool is analysed on the 2017–2018 low flows on 38 catchments for which forecasts are available for all the models. The capacity to anticipate low-flow events shows an average anticipation of 40 days. The short-term forecast (7 days ahead) is good for most of the models.; De nombreux usages de l'eau peuvent être fortement impactés par les pénuries d'eau (eau potable, irrigation, hydroéléctricité...). Il est donc nécessaire d'anticiper les périodes d'étiage afin d'améliorer la gestion de l'eau. Ceci est renforcé par la perspective d'étiages futurs plus sévères dans le contexte du changement climatique. Cinq institutions françaises ont développé un outil opérationnel de prévision des bas débits, PREMHYCE. Il est testé en temps réel sur une centaine de bassins versants de France métropolitaine depuis 2017. PREMHYCE comprend cinq modèles hydrologiques qui peuvent être calés sur des bassins versants jaugés et assimilent les dernières observations de débit. Les prévisions de débits sont émises jusqu'à un horizon de 90 jours selon l'approche Ensemble Streamflow Prediction (ESP) (données météorologiques historiques utilisées comme ensemble de scénarios d'entrée). Ces données météorologiques (précipitations, évapotranspiration et température) sont issues de la réanalyse SAFRAN journalière de Météo-France, sur la période 1958–2018. Les performances de l'outil sont analysées sur les étiages 2017–2018 pour 38 bassins versants sur lesquels les prévisions sont disponibles pour tous les modèles. Ces derniers ont montré des capacités d'anticipation de l'ordre de 40 jours en moyenne. La plupart des modèles présentent une précision satisfaisante pour prévoir les bas débits à courte échéance (j + 7).

The MOSARH21 project evaluated the future impacts of climate change on streamflows from the French tributaries of the River Rhine based on recent climate projections (CMIP5 experiment). The adopted methodology relies on two hydrological models (GRSD and LARSIM), used jointly with an ensemble of climate projections disaggregated at the local scale. The impacts were quantified with several indicators describing rivers regimes, floods and low flows. The evolutions of mean interannual streamflow indicate a slight increase, which could even become important for the radiative scenario RCP 8.5 (the most pessimistic one in terms of climate warming). Flood could intensify in the near future (2021-2050). In the far future (2071-2100), the evolution of floods indicators is more uncertain, as the hydrological projections diverge. Low flows could be decreasing in the near future. Their evolution in the far future is more uncertain, going from drastic decrease to sensible rise depending on the chosen scenario. The results were compared to those from previous impact studies performed on the basin. Given the methodological differences and the obtained results, we can conclude that the MOSARH21 project is consistent with the conclusions of the previous studies.

International audience; Flash-flood events can have catastrophic socio-economic consequences. To reduce their impacts, it is of crucial importance to set up efficient warning systems. Although first operational flash-flood warning systems have recently been implemented, some limitations are clearly identified by end-users: non-exhaustive geographic coverage, limited lead times, warnings based on hazard assessment instead of risk. However, the desirable improvements raise real scientific challenges in various domains. In this context, the PICS (Prevision immediate des impacts des crues soudaines – Flash-flood events impacts nowcasting - 2018-2022) project gathers French scientific teams with varied skills (meteorologists, hydrologists, hydraulicians, economists, social geographers) and operational stakeholders (civil security, local authorities, insurance companies, managers of hydroelectric facilities and transport network). Funded by the French national research agency (ANR), it aims to develop and evaluate pre-operational forecasting chains able to estimate the potential impacts of flash floods with short anticipation lead times (up to 6 hours). These modelling chains include different components. Distributed hydrological models transform the observed and forecasted rainfall into runoff. Hydraulic models translate this runoff into potential flooded areas. Impact models incorporate these results to evaluate the potential for social and economic impacts. The research and operational partners selected four case studies based on various criteria, including the occurrence of human impacts and damages, the availability of validation data. Validation data include discharges recorded at gauging stations, but also more original information collected after each event, such as peak discharges and maximum water levels estimated from flood marks, insurance claims, damages observed on infrastructure (roads, railway…), victims interviews, casualties,etc. This presentation focuses on the methodology used for the involvement of representative potential end-users, leading to fruitfull dialogues and informative outcomes. Some of the first results of the project are also presented.

National audience; Nowadays, flood forecasting is often done at the hourly time step, which is a good compromise between data availability and the necessary resolution to simulate flood events. However, in a number of cases, this fixed time step shows limitations, and shorter (or even larger) time steps are required. Therefore, it is necessary to have hydrological models able to adapt to various time steps. Generally, this adaptation does not limit to a simple change of the length of the time step in the model equations. The time-step dependency must be made explicit in some model parameters, and the model structure must also be adapted to consistently represent processes across time steps. The article presents the adaptation of the GRP hydrological model, which is widely used by flood forecasting services in operational conditions in France, so that it can behave and run consistently at different time steps. The tests were made on a set of 240 French catchments.; La prévision des crues est souvent réalisée au pas de temps horaire, qui est un bon compromis entre disponibilité des données et résolution nécessaire pour bien rendre compte des phénomènes de crue. Cependant, dans un certain nombre de cas, ce pas de temps fixe est limitant, et des pas de temps plus fins (ou éventuellement plus larges) sont mieux adaptés. Il est donc nécessaire d'avoir des modèles hydrologiques capables de s'adapter à un fonctionnement à différents pas de temps. Or, cette adaptation ne se résume généralement pas à un simple changement de la longueur du pas de temps dans les équations du modèle. Elle implique une explicitation de la dépendance possible de certains paramètres du modèle au pas de temps, et également une adaptation de la structure du modèle pour que les processus soient représentés de manière cohérente entre les pas de temps. L'article présente une adaptation du modèle hydrologique GRP, qui est largement utilisé en conditions opérationnelles par les services de prévision des crues français, pour qu'il fonctionne correctement à différents pas de temps. Les tests sont réalisés sur une base de 240 bassins versants français.

To improve the predictive capability of a model, one must identify situations where it fails to provide satisfactory results. We tried to identify the deficiencies of a lumped rainfall-runoff model used for flood simulation (the hourly GR5H-I model) by evaluating it over a large set of 229 French catchments and 11,054 flood events. Evaluating model simulations separately for individual flood events allowed us identifying a seasonal trend: while the model yielded good performance in terms of aggregated statistics, grouping results by season showed clear underestimations of most of the floods occurring in summer. The largest underestimations of flood volumes were identified when high-intensity precipitation events occurred and when the precipitation field was highly spatially variable. Low antecedent soil moisture conditions were also found to be strongly correlated with model bias. Overall, this study pinpoints the need to better account for short-duration processes to improve the GR5H-I model for flood simulation.

International audience; Abstract. Prior to their use under future changing climate conditions, all hydrological models should be thoroughly evaluated regarding their temporal transferability (application in different time periods) and extrapolation capacity (application beyond the range of known past conditions). This note presents a straightforward evaluation framework aimed at detecting potential undesirable climate dependencies in hydrological models: the robustness assessment test (RAT). Although it is conceptually inspired by the classic differential split-sample test of Klemeš (1986), the RAT presents the advantage of being applicable to all types of models, be they calibrated or not (i.e. regionalized or physically based). In this note, we present the RAT, illustrate its application on a set of 21 catchments, verify its applicability hypotheses and compare it to previously published tests. Results show that the RAT is an efficient evaluation approach, passing it successfully can be considered a prerequisite for any hydrological model to be used for climate change impact studies.

Low-flow forecasting can help to improve water management at places where a number of uses can be affected by diminishing water supply from rivers. Several French institutes (INRAE, BRGM, EDF, Lorraine University and Météo-France) have been collaborating to set up an operational platform, called PREMHYCE, for low-flow forecasting at the national scale, in cooperation with operational services. PREMHYCE includes five hydrological models and low-flow forecasts can be issued up to 90 days ahead for more than 800 basins. Several input scenarios are considered: ECMWF 14-days ensemble forecasts, ensemble streamflow prediction (ESP) using historical climatic data, and a no precipitation scenario. Outputs from the different hydrological models are combined into a multi-model approach to improve robustness of the forecasts. The tool provides text files and graphical representation of forecasted low-flows, as well as key low-flow indicators, such as the probabilities of being under low-flow thresholds provided by operational services. The presentation will show the main characteristics of this operational forecast platform, its latest developments and the results on the recent low-flow periods.

International audience

International audience; The Surface Water and Ocean Topography (SWOT) space mission will map surface area and water level changes in lakes at the global scale. Such new data are of great interest to better understand and model lake dynamics as well as to improve water management. In this study, we used the large-scale SWOT simulator developed at the French Space National Center (CNES) to estimate the expected measurement errors of the water level of different water bodies in France. These water bodies include five large reservoirs of the Seine River and numerous small gravel pits located in the Seine alluvial plain of La Bassée upstream of the city of Paris. The results show that the SWOT mission will allow to observe water levels with a precision of a few tens of centimeters (10 cm for the largest water reservoir (Orient), 23 km$^2$), even for the small gravel pits of size of a few hectares (standard deviation error lower than 0.25 m for water bodies larger than 6 ha). The benefit of the temporal sampling for water level monitoring is also highlighted on time series of pseudo-observations based on real measurements perturbed with the simulated noise errors. Then, the added value of these future data for the simulation of lake energy budgets is shown using the FLake lake model through sensitivity experiments. Results show that the SWOT data will help to model the surface temperature of the studied water bodies with a precision better than 0.5 K and the evaporation with an accuracy better than 0.2 mm/day. These large improvements compared to the errors obtained when a constant water level is prescribed (1.2 K and 0.6 mm/day) demonstrate the potential of SWOT for monitoring the lake energy budgets at global scale in addition to the other foreseen applications in operational reservoir management.

International audience; To assess the predictive performance, robustness and generality of watershed-scale hydrological models, we conducted a detailed multi-objective evaluation of two conceptual rainfall-runoff models (the GRX model, based on the GR4J model, and the MRX model, based on the MORDOR model), of differing complexity (with respectively, 5 and 11 free parameters in the rainfall-runoff module, and 4 and 11 free parameters in the snow module). These models were compared on a large sample of 2050 watersheds worldwide. Our results, based on the three components of the Kling-Gupta Efficiency metric (KGE), indicate that both models provide (on average) similar levels of performance in evaluation when calibrated with KGE, for water balance (mean bias lower than 2%), time-series variability (mean variability bias lower than 2%) and temporal correlation (mean correlation around 0.83). Further, both models clearly suffer from lack of robustness when simulating water balance, with a significant increase of the proportion of biased simulations over the evaluation periods (absolute bias lower than 2% in calibration and lower than 20% in evaluation for 80% of the watersheds). Simulation performance depend more on the hydro-meteorological conditions of a given period than on the complexity of the model structure. We also show that long-term aggregate statistics (computed on the overall period) can fail to reveal considerable sub-period variability in model performance, thereby providing inaccurate diagnostic assessment of the predictive model performance. Typically the median absolute bias is lower than 8% in evaluation, but the median maximum bias can be as high as 50% within a subperiod, for both models, when calibrated with KGE.

Skillful streamflow forecasts provide a key support to several water-related applications. Ensemble forecasting systems are gaining a widespread interest, since they allow accounting for different sources of uncertainty. Because of the critical impact of the initial conditions (ICs) on the forecast accuracy, it is essential to improve their estimates via data assimilation (DA). This study aims at assessing the sensitivity of the DA-based estimation of forecast ICs to several sources of uncertainty and to the update of different model states and parameters of a conceptual rainfall-runoff model. The performance of two sequential ensemble-based techniques are compared, namely Ensemble Kalman filter and Particle filter, in terms of both efficiency and temporal persistence of the updating effect through the assimilation of observed discharges at the forecast time. Several experiments specifically address the impact of the meteorological, model state and parameter uncertainties over 232 catchments in France. Results show that the benefit of the DA-based estimation of ICs for forecasting is the largest when focusing on the level of the model routing store, which is the internal state the most correlated to streamflow. While the EnKF-based forecasts outperform the PF-based ones when accounting for the meteorological uncertainty, the representation of the model state uncertainty allows greatly improving the accuracy of the PF-based predictions, with a longer-lasting updating effect (up to 10 days). Conversely, the forecasting skill is undermined when accounting for the parameter uncertainty, due to the change in the hydrological responsiveness through the update of both the production and routing store levels. A further effort is focused on assessing the impact of the spatial resolution of the hydrological model on the predictive accuracy of DA-based streamflow forecasts.

In many countries, rivers are the primary supply of water. A number of uses are concerned (drinking water, irrigation, hydropower…) and they can be strongly affected by water shortages. Therefore, there is a need of early anticipation of low-flow periods to improve water management. This is strengthened by the perspective of having more severe summer low-flows in the context of climate change. Several French institutes (Irstea, BRGM, Météo-France, EDF and Lorraine University) have been collaborating to develop an operational tool for low-flow forecasting, called PREMHYCE. It is tested in real time since 2017, and implemented on 259 catchments in metropolitan France, in cooperation with operational services which provide streamflow observations and use low-flow forecasts from the tool. PREMHYCE includes five hydrological models which can be calibrated on gauged catchments and which assimilate flow observations. Low-flow forecasts can be issued up to 90 days ahead, based on several inputs scenarios: ECMWF 10-days ensemble forecasts, ensemble streamflow prediction (ESP) using historical climatic data as ensembles of future input scenarios, and a no precipitation scenario. Climatic data (precipitation, evapotranspiration and temperature) are provided by Météo-France with the daily gridded SAFRAN reanalysis on the 1959-2019 period, which includes a wide range of conditions. The tool provides text files and graphical representation of forecasted low-flows, and probability to be under low-flow thresholds provided by users. Outputs from the different hydrological models can be combined within a multi-model approach to improve robustness of the forecastsThe presentation will show the main characteristics of this operational tool, the probabilistic evaluation framework, results on the recent low-flow periods, and how feedbacks from end-users can help improving the tool.

Log-transformed discharge is often used to calculate performance criteria to better focus on low flows. This prior transformation limits the heteroscedasticity of model residuals and was largely applied in criteria based on squared residuals, like the Nash–Sutcliffe efficiency (NSE). In the recent years, NSE has been shown to have mathematical limitations and the Kling–Gupta efficiency (KGE) was proposed as an alternative to provide more balance between the expected qualities of a model (namely representing the water balance, flow variability and correlation). As in the case of NSE, several authors used the KGE criterion (or its improved version KGE′) with a prior logarithmic transformation on flows. However, we show that the use of this transformation is not adapted to the case of the KGE (or KGE′) criterion and may lead to several numerical issues, potentially resulting in a biased evaluation of model performance. We present the theoretical underpinning aspects of these issues and concrete modelling examples, showing that KGE′ computed on log-transformed flows should be avoided. Alternatives are discussed.

This book analyzes the spatial concepts of Lithuania and other place names that either competed with the term Lithuania or were of a different taxonomic level in the ‘long’ nineteenth century (1795...

This paper presents a novel method to estimate flow duration curves (FDCs) for ungauged catchments with permanent flow. It is based on a dataset of 521 catchments located throughout France. The method consists in a three-step procedure called the Quantile Solidarity (QS) approach. First, a regression-based model is built to estimate FDCs in ungauged catchments, linking each flow quantile independently to physical descriptors. The second step involves imposing the continuity of the regression-based model parameters along the quantiles (hence quantile solidarity) to obtain a large reduction in the number of parameters used to estimate the FDC. The last step consists in spatially interpolating the model residuals to further improve the performance of the FDC estimation. The QS approach yields a robust and parsimonious FDC estimation without any loss in simulation efficiency and ensures strictly monotonic FDCs.EDITOR A. Castellarin ASSOCIATE EDITOR A. Efstratiadis EDITOR A. Castellarin ASSOCIATE EDITOR A. Efstratiadis

Resume Les changements climatiques projetes a l’horizon du milieu du XXIe siecle pourraient avoir des impacts importants sur la disponibilite en eau en France. L’etude prospective Explore 2070, pilotee par le ministere charge de l’ecologie, a ainsi mis en evidence la necessite de quantifier et d’anticiper ces changements, et de construire des strategies d’adaptation pour limiter leurs consequences negatives sur les hydrosystemes et les activites humaines. Cet article analyse dans quelle mesure ces travaux ont pu contribuer a la sensibilisation des acteurs de l’eau et a la reflexion sur l’adaptation au changement climatique en France.

International audience; Le débit des rivières au cours des mois d’étiage dépend des conditions climatiques passées. Mais la sensibilité des rivières aux sécheresses peut beaucoup varier en fonction d’un ensemble de facteurs naturels (climat, géologie, couvert végétal…) et anthropiques (présence d’ouvrage de régulation). A l’échelle de la France, cartographier la diversité des sensibilités des bassins versants aux facteurs influençant les étiages permet d’une part d’accroître notre connaissance de la géographie des étiages, et d’autre part d’évaluer de manière simple (mais non simpliste) l’impact de scénarios climatiques futurs.Nous nous proposons ici de reprendre la méthodologie d’évaluation de l’élasticité du module de l’écoulement que nous avons proposée (Andréassian et al., 2016) pour l’appliquer au débit mensuel minimum (QMNA). La Figure 1 présente un exemple du graphique illustrant le concept d’élasticité climatique des QMNA, où nous représentons pour l’Aube à Arcis l’anomalie relative du QMNA (〖QMNA〗_n/(QMNA) ̅ -1) en fonction de l’anomalie relative des précipitations annuelles (P_n/P ̅ -1).Notre analyse porte sur un échantillon de 519 bassins versants français pseudo-naturels (c’est-à-dire peu ou pas régulés), auxquels nous ajoutons quelques cas particuliers concernant des bassins versants régulés par des barrages, où nous montrons comment le volume stocké influence l’élasticité du QMNA.

International audience; In many countries, rivers are the primary supply of water. A number of uses are concerned (drinking water, irrigation, hydropower...) and they can be strongly affected by water shortages. Therefore, there is a need of early anticipation of low-flow periods to improve water management. This is strengthened by the perspective of having more severe summer low-flows in the context of climate change. Several French institutes (Irstea, BRGM, Météo-France, EDF and Lorraine University) have been collaborating to develop an operational tool for low-flow forecasting, called PREMHYCE. It is tested in real time on 118 catchments in metropolitan France since 2017, in cooperation with operational services which provide streamflow observations and use low-flow forecasts from the tool. PREMHYCE includes five hydrological models which can be calibrated on gauged catchments and which assimilate flow observations. Low-flow forecasts can be issued up to 90 days ahead, based on ensemble streamflow prediction (ESP) using historical climatic data as ensembles of future input scenarios. These climatic data (precipitation, evapotranspiration and temperature) are provided by Météo-France with the daily gridded SAFRAN reanalysis on the 1958-2018 period, which includes a wide range of conditions. The tool provides text files and graphical representation of forecasted low-flows, and probability to be under low-flow thresholds provided by users. Outputs from the different hydrological models can be combined into a simple multi-model approach to improve robustness of the forecasts. The presentation will show the main characteristics of this operational tool and results on the recent low-flow periods.

An increasing number of flood forecasting services assess and communicate the uncertainty associated with their forecasts. While obtaining reliable forecasts is a key issue, it is a challenging task, especially when forecasting high flows in an extrapolation context, i.e. when the event magnitude is larger than what was observed before. In this study, we present a crash-testing framework that evaluates the quality of hydrological forecasts in an extrapolation context. The experiment set-up is based on (i) a large set of catchments in France, (ii) the GRP rainfall–runoff model designed for flood forecasting and used by the French operational services and (iii) an empirical hydrologic uncertainty processor designed to estimate conditional predictive uncertainty from the hydrological model residuals. The variants of the uncertainty processor used in this study differ in the data transformation they use (log, Box–Cox and log–sinh) to account for heteroscedasticity and the evolution of the other properties of the predictive distribution with the discharge magnitude. Different data subsets were selected based on a preliminary event selection. Various aspects of the probabilistic performance of the variants of the hydrologic uncertainty processor, reliability, sharpness and overall quality were evaluated. Overall, the results highlight the challenge of uncertainty quantification when forecasting high flows. They show a significant drop in reliability when forecasting high flows in an extrapolation context and considerable variability among catchments and across lead times. The increase in statistical treatment complexity did not result in significant improvement, which suggests that a parsimonious and easily understandable data transformation such as the log transformation or the Box–Cox transformation can be a reasonable choice for flood forecasting.

International audience; Consistent hydrological models for multiple time steps are needed to respond to the operational requirement of using a model for different objectives and with varied data sets. In this paper, we propose a rationale for multitime step model development based on the temporal consistency of the fluxes modelled. The new methodology is applied for the temporal downscaling of a lumped rainfall-runoff model (GR4J) from daily to sub-hourly time steps, within a coherent modelling framework. The modelling tests are performed at eight time steps from daily to 6-min, for a set of 240 French catchments, for which 2400 flood events were selected. To identify an improved multi-time step model structure, a two-step approach is followed. First, we propose a model diagnosis to evaluate the internal behaviour of the baseline model across time steps, whereby a significant inconsistency was found in the simulated water fluxes, especially interception. Second, we recommend a prognosis to improve the scale invariance of the modelled water balance by matching the model fluxes across time scales. The new model structure retained for sub-daily time steps is derived from the daily baseline model by refining the representation of the interception process, so that the flux-matching condition is satisfied. A complementary modification of the groundwater exchange function is also resumed following a previous study. The structural changes are motivated by the improvement of model flux consistency across time steps, while also improving model performance. This new dual paradigm for model identification, based on both flux coherence and output accuracy, also results in more robust model parameters across time steps.

International audience; In semidistributed hydrological modeling, sequential calibration usually refers to the calibration of a model by considering not only the flows observed at the outlet of a catchment but also the different gauging points inside the catchment from upstream to downstream. While sequential calibration aims to optimize the performance at these interior gauged points, we show that it generally fails to improve performance at ungauged points. In this paper, we propose a regularization approach for the sequential calibration of semidistributed hydrological models. It consists in adding a priori information on optimal parameter sets for each modeling unit of the semidistributed model. Calibration iterations are then performed by jointly maximizing simulation performance and minimizing drifts from the a priori parameter sets. The combination of these two sources of information is handled by a parameter k to which the method is quite sensitive. The method is applied to 1,305 catchments in France over 30 years. The leave-one-out validation shows that, at locations considered as ungauged, model simulations are significantly improved (over all the catchments, the median KGE criterion is increased from 0.75 to 0.83 and the first quartile from 0.35 to 0.66), while model performance at gauged points is not significantly impacted by the use of the regularization approach. Small catchments benefit most from this calibration strategy. These performances are, however, very similar to the performances obtained with a lumped model based on similar conceptualization.

Publisher's version (útgefin grein), This paper is the outcome of a community initiative to identify major unsolved scientific problems in hydrology motivated by a need for stronger harmonisation of research efforts. The procedure involved a public consultation through online media, followed by two workshops through which a large number of potential science questions were collated, prioritised, and synthesised. In spite of the diversity of the participants (230 scientists in total), the process revealed much about community priorities and the state of our science: a preference for continuity in research questions rather than radical departures or redirections from past and current work. Questions remain focused on the process-based understanding of hydrological variability and causality at all space and time scales. Increased attention to environmental change drives a new emphasis on understanding how change propagates across interfaces within the hydrological system and across disciplinary boundaries. In particular, the expansion of the human footprint raises a new set of questions related to human interactions with nature and water cycle feedbacks in the context of complex water management problems. We hope that this reflection and synthesis of the 23 unsolved problems in hydrology will help guide research efforts for some years to come., We would like to thank the members of the IAHS, EGU, AGU and IAH for supporting this initiative. The LinkedIn group and overall secretariat was hosted by the IAHS, the Splinter meeting by EGU and the Vienna Catchment Science Symposium by the Vienna Doctoral Programme on Water Resource Systems (DK W1219-N28) funded by the Austrian Science Funds (FWF)., "Peer Reviewed"

International audience; Towards the real-time production of predictive intervals around streamflow forecasts in Vigicrues in France. Although quantifying uncertainties in modelling is an old topic,many developments have been made over the past fifteen years in the context of flood forecasting, mainly with the development of probabilistic approaches. In France, within the national network for floodforecasting (Vigicrues), developments have been made to propose a post-processing methodology to quantify predictive uncertainty (OTAMIN), which automatically generates confidence intervals around stream flow forecasts, and a tool that is used by forecasters to expertise these confidence intervals (EAO/EXPRESSO). This paper presents the main principles of this general approach, gives an example of its recent application in operational conditions and discusses its usefulness for the users and the general public. Some perspectives of improvement are discussed.; La quantification des incertitudes en modélisation est un sujet ancien, mais de nombreux développements ont été réalisés au cours des quinze dernières années dans un contexte de prévision des crues, avec l'essor notamment des approches probabilistes. En France, au sein du réseau national de prévision des crues (Vigicrues), cela a conduit à la mise au point d'une méthode automatisée de quantification des incertitudes prédictives (OTAMIN) rentrant dans la catégorie des approches de \post- traitement, qui génère des intervalles de confiance sur les prévisions de débit, et d'un outil d'expertise par les prévisionnistes de ces intervalles (EAO/EXPRESSO). Cet article présente les principes méthodologiques de cette démarche, en donne une illustration d'application opérationnelle récente et discute de son utilité pour des utilisateurs institutionnels et pour le grand public. Quelques perspectives d'amélioration sont avancées.

A partir de 2003, sous l'impulsion du Ministere de l'ecologie, le systeme national d'annonce des crues a progressivement evolue vers un systeme de prevision de crues. Le Service Central d'Hydrometeorologie et d'Appui a la Prevision des Inondations (SCHAPI) a ete cree pour coordonner cette evolution et la modernisation des dispositifs d'anticipation et d'alerte aux crues, en collaboration avec les Services de Prevision des Crues (SPC) a l'echelle des entites hydrographiques regionales. Ce reseau etatique en charge de la prevision des crues s'est tourne vers divers acteurs de la recherche et de l'ingenierie, dont Irstea. La collaboration pluriannuelle avec Irstea a donne naissance au modele de prevision des crues GRP, aujourd'hui largement utilise par les SPC. Cette collaboration perenne revet plusieurs aspects : l'appui a la mise en œuvre de l'outil sur les bassins suivis par le reseau, l'evaluation des performances comparativement a d'autres outils, l'analyse des retours d'experience au travers d'un groupe d'utilisateurs ou encore la formation des utilisateurs. Elle conduit en particulier a traduire en questions de recherche les problemes souleves, et a proposer des solutions concretes et efficaces pour faire de GRP un outil plus performant. L'article decrit la demarche globale de developpement partenarial et en illustre les avantages au travers d'exemples concrets.

During World War I, Martynas Jankus became the most well-known deportee in the Lithuanian-speaking world. This article uses a variety of sources, including his wartime letters, diary, and postwar memoirs, to explore which factors were the most important in enabling him to survive deportation, how representative he is of the larger population of deportees from East Prussia and their experiences, and how his life as a deportee was affected by laws and agreements at the international and domestic levels.

This paper investigates the temporal transposability of hydrological models under contrasted climate conditions and evaluates the added value of using an ensemble of model structures for flow simulation. This is achieved by applying the Differential Split Sample Test procedure to twenty lumped conceptual models on a catchment in the Province of Québec (Canada) and another one in the State of Bavaria (Germany). First, a calibration/validation procedure was applied on four historical non-continuous periods with contrasted climate conditions. Then, model efficiency was quantified individually (for each model) and collectively (for the model ensemble). The individual analysis evaluated model performance and robustness. The ensemble investigation, based on the average of simulated discharges, focused on the twenty-member ensemble and all possible model subsets. Results showed that using a single model may provide hazardous results when the model is to be applied in contrasted conditions. Overall, some models turned out as a good compromise in terms of performance and robustness, but generally not as much as the twenty-model ensemble. Model subsets offered yet improved performance over the twenty-model ensemble, but at the expanse of spatial transposability (i.e. need of site-specific analysis).

International audience; ABSTRACT.-In mountainous regions, near-surface meteorological reanalyses can carry significant uncertainties due to the complex topography and the scarcity of observations. Combining different reanalyses has proved to be efficient in reducing these uncertainties. This technique is applied over a French mountainous basin in the southern Alps, the Durance basin, where two reanalyses exist : SAFRAN and SPAZM. The comparison of the features of these two rea-nalyses shows that they are complementary. The precipitations provided by SPAZM are at a finer spatial resolution than SAFRAN and are based on more observations at high altitudes. However, on the contrary to SAFRAN, SPAZM does not provide the full set of atmospheric variables required by hydrological models based on water and energy balance. SPAZM and SAFRAN were thus combined. Precipitation and temperature from SAFRAN are corrected by monthly temperature and precipitation from SPAZM, and the other variables from SAFRAN are then corrected accordingly. As a result, the meteorological product called DuO was obtained, providing eight atmospheric variables (temperature at 2 m, rainfall, snowfall, incoming solar and infrared radiation, air humidity, wind speed and potential evapotranspiration), at the hourly timescale and at the 1-km 2 resolution. The strong topographical heterogeneity is thus better represented in the upstream part of the Durance catchment. DuO is tested for hydrological modelling, using the Catchment Land Surface Model (CLSM), a land surface model distributed over sub-catchments. The analysis of various output variables (stream-flow, snowpack and evapotranspiration) shows that the hybridization of the two reanalyses benefits from their respective qualities and yields a more realistic representation of the hydrological processes in the CLSM model.; La rareté des mesures et la complexité du terrain dans les régions montagneuses engendrent d'importantes incertitudes dans les réanalyses météorologiques de surface. L'hybridation de plusieurs réanalyses est une technique efficace pour limiter ces incertitudes. Dans le bassin versant de la Durance, deux réanalyses se basant sur des réseaux de mesures et des techniques d'extrapolation différentes coexistent, SAFRAN et SPAZM. Leur comparaison montre qu'elles sont complémentaires car les précipitations de SPAZM à une résolution spatiale plus fine que celle de SAFRAN, se basent sur un plus grand nombre d'observations en haute altitude. Cependant, SPAZM ne fournit pas l'ensemble des variables atmosphériques dont un modèle hydrologique à bilans d'eau et d'énergie a besoin, contrairement à SAFRAN. L'hybridation de ces deux réanalyses a donc été effectuée. Le produit météorologique obtenu, DuO, fournit huit variables atmosphériques (température de l'air à 2 m, pluie, neige, rayonnements incidents, humidité de l'air, vent et évapo-transpiration potentielle) au pas de temps horaire et à la résolution de 1 km 2. DuO a été testé dans un objectif de modé-lisation hydrologique, avec le modèle semi-distribué CLSM (Catchment Land Surface Model). L'analyse des différentes variables de sortie (débits, stock de neige et évapotranspiration) montre que l'hybridation des deux réanalyses, valorisant les qualités respectives de SAFRAN et de SPAZM, permet une représentation plus réaliste des processus hydrologiques au sein du modèle CLSM. Building hybrid near-surface reanalysis adapted to mountainous regions: example of the DuO product over the Durance basin, France