Improving Predictions and Management of Hydrological Extremes through Climate Services

Transcription

1 Vol. 18, EGU , 2016 Author(s) CC Attribution 3.0 License. Improving Predictions and Management of Hydrological Extremes through Climate Services Bart van den Hurk (1), Janet Wijngaard (1), Florian Pappenberger (2), Laurens Bouwer (3), Albrecht Weerts (3), Carlo Buontempo (4), Ralf Doescher (5), Maria Manez (6), Maria-Helena Ramos (7), Cedric Hananel (8), Ertug Ercin (9), Johannes Hunink (10), Bastian Klein (11), Laurent Pouget (12), and Philip Ward (13) (1) KNMI, Atmospheric Research, De Bilt, Netherlands (2) ECMWF, Reading, UK, (3) Deltares, Delft, Netherlands, (4) Metoffice, Exeter, UK, (5) SMHI, Norrkoping, Sweden, (6) Helmholtz Zentrum Geesthacht, Germany, (7) IRSTEA, Antony, France, (8) Arctik, Belgium, (9) Water Footprint Network, Netherlands, (10) Future Water, Spain, (11) Federal Institute of Hydrology (BfG), Germany, (12) CETAqua, Spain, (13) VU Amsterdam, Netherlands The EU Roadmap on Climate Services can be seen as a result of convergence between the society s call for actionable research, and the climate research community providing tailored data, information and knowledge. However, although weather and climate have clearly distinct definitions, a strong link between weather and climate services exists that is not explored extensively. Stakeholders being interviewed in the context of the Roadmap consider climate as a far distant long term feature that is difficult to consider in present-day decision taking, which is dominated by daily experience with handling extreme events. It is argued that this experience is a rich source of inspiration to increase society s resilience to an unknown future. A newly started European research project, IMPREX, is built on the notion that experience in managing current day weather extremes is the best learning school to anticipate consequences of future climate. This paper illustrates possible ways to increase the link between information and services addressing weather and climate time scales by discussing the underlying concepts of IMPREX and its expected outcome.

2 Vol. 18, EGU , 2016 Looking beyond general metrics for model evaluation - lessons from an international model intercomparison study Laurène Bouaziz (1), Tanja de Boer-Euser (2), Claudia Brauer (3), Gilles Drogue (4), Fabrizio Fenicia (5), Benjamin Grelier (4), Jan de Niel (6), Jiri Nossent (7,8), Fernando Pereira (7), Hubert Savenije (2), Guillaume Thirel (9), and Patrick Willems (6) (1) Deltares, Hydrology, Delft, Netherlands (Laurene.Bouaziz@deltares.nl), (2) Water Resources Section, Faculty of Civil Engineering and Geosciences, Delft University of Technology, P.O. Box 5048, NL-2600 GA Delft, The Netherlands, (3) Hydrology and Quantitative Water Management Group, Wageningen University, The Netherlands, (4) Laboratoire LOTERR, Université de Lorraine, Metz, France, (5) Eawag, Dübendorf, Switzerland, (6) Hydraulics division, Department of Civil Engineering, KU Leuven, Kasteelpark Arenberg 40, BE-3001 Leuven, Belgium, (7) Flanders Hydraulics Research, Antwerp, Belgium, (8) Vrije Universiteit Brussel (VUB), Department of Hydrology and Hydraulic Engineering, Brussel, Belgium, (9) Irstea, Hydrology Research Group, Antony, France International collaboration between institutes and universities is a promising way to reach consensus on hydrological model development. Education, experience and expert knowledge of the hydrological community have resulted in the development of a great variety of model concepts, calibration methods and analysis techniques. Although comparison studies are very valuable for international cooperation, they do often not lead to very clear new insights regarding the relevance of the modelled processes. We hypothesise that this is partly caused by model complexity and the used comparison methods, which focus on a good overall performance instead of focusing on specific events. We propose an approach that focuses on the evaluation of specific events. Eight international research groups calibrated their model for the Ourthe catchment in Belgium (1607 km2) and carried out a validation in time for the Ourthe (i.e. on two different periods, one of them on a blind mode for the modellers) and a validation in space for nested and neighbouring catchments of the Meuse in a completely blind mode. For each model, the same protocol was followed and an ensemble of best performing parameter sets was selected. Signatures were first used to assess model performances in the different catchments during validation. Comparison of the models was then followed by evaluation of selected events, which include: low flows, high flows and the transition from low to high flows. While the models show rather similar performances based on general metrics (i.e. Nash-Sutcliffe Efficiency), clear differences can be observed for specific events. While most models are able to simulate high flows well, large differences are observed during low flows and in the ability to capture the first peaks after drier months. The transferability of model parameters to neighbouring and nested catchments is assessed as an additional measure in the model evaluation. This suggested approach helps to select, among competing model alternatives, the most suitable model for a specific purpose.

3 Vol. 18, EGU , 2016 Willingness-to-pay for a probabilistic flood forecast: a risk-based decision-making game Louise Arnal (1,2), Maria-Helena Ramos (3), Erin Coughlan (4), Hannah L. Cloke (1,5), Elisabeth Stephens (1), Fredrik Wetterhall (2), Schalk-Jan van Andel (6), Florian Pappenberger (2,7) (1) Department of Geography and Environmental Science, University of Reading, Reading, United Kingdom, (2) ECMWF, Forecast Department, Reading, United Kingdom, (3) IRSTEA, Catchment Hydrology Research Group, UR HBAN, Antony, France, (4) Red Cross/Red Crescent Climate Centre, The Hague, the Netherlands, (5) Department of Meteorology, University of Reading, Reading, United Kingdom, (6) UNESCO-IHE Institute for Water Education, Delft, the Netherlands, (7) School of Geographical Sciences, University of Bristol, Bristol, United Kingdom Forecast uncertainty is a twofold issue, as it constitutes both an added value and a challenge for the forecaster and the user of the forecasts. Many authors have demonstrated the added (economic) value of probabilistic forecasts over deterministic forecasts for a diversity of activities in the water sector (e.g. flood protection, hydroelectric power management and navigation). However, the richness of the information is also a source of challenges for operational uses, due partially to the difficulty to transform the probability of occurrence of an event into a binary decision. The setup and the results of a risk-based decision-making experiment, designed as a game on the topic of flood protection mitigation, called How much are you prepared to pay for a forecast?, will be presented. The game was played at several workshops in 2015, including during this session at the EGU conference in 2015, and a total of 129 worksheets were collected and analysed. The aim of this experiment was to contribute to the understanding of the role of probabilistic forecasts in decision-making processes and their perceived value by decision-makers. Based on the participants willingness-to-pay for a forecast, the results of the game showed that the value (or the usefulness) of a forecast depends on several factors, including the way users perceive the quality of their forecasts and link it to the perception of their own performances as decision-makers. Balancing avoided costs and the cost (or the benefit) of having forecasts available for making decisions is not straightforward, even in a simplified game situation, and is a topic that deserves more attention from the hydrological forecasting community in the future.

4 Vol. 18, EGU , 2016 Bias correcting precipitation forecasts for extended-range skilful seasonal streamflow predictions Louise Crochemore (1), Maria-Helena Ramos (1), Florian Pappenberger (2,3) (1) Irstea, Hydrology Research Group, Antony, France, (2) ECMWF, European Centre for Medium-Range Weather Forecasts, Shinfield Park, Reading, RG2 9AX, UK, (3) School of Geographical Sciences, University of Bristol, University Road, Bristol, BS8 1SS, UK Meteorological centres make sustained efforts to provide seasonal forecasts that are increasingly skilful, which has the potential to also benefit streamflow forecasting. Seasonal streamflow forecasts can help to take anticipatory measures for a range of applications, such as water supply or hydropower reservoir operation and drought risk management. This study assesses the skill of seasonal precipitation and streamflow forecasts in France in order to provide insights into the way bias correcting seasonal precipitation forecasts can contribute to maintain skill of seasonal flow predictions at extended lead times. First, we evaluate the skill of raw (i.e. without bias correction) seasonal precipitation ensemble forecasts for streamflow forecasting in sixteen French catchments. A lumped daily hydrological model is applied at the catchment scale to transform precipitation into streamflow. A reference prediction system based on historic observed precipitation and watershed initial conditions at the time of forecast (i.e. ESP method) is used as benchmark. In a second step, we apply eight variants of bias correction approaches to the precipitation forecasts prior to generating the flow forecasts. The approaches were based on the linear scaling and the distribution mapping methods. The skill of the ensemble forecasts is assessed in reliability, sharpness, accuracy, and overall performance. The results show that, in most catchments, raw seasonal precipitation and streamflow forecasts are often more skilful than the conventional ESP method in terms of sharpness. However, reliability is an attribute that is not significantly improved. Forecast skill is generally improved when applying bias correction. Two bias correction methods showed the best performance for the studied catchments, with, however, each method being more successful in improving specific attributes of forecast quality: the simple linear scaling of monthly values contributed mainly to increase forecast sharpness and accuracy, while the empirical distribution mapping of daily values was successful in improving forecast reliability.

5 Vol. 18, EGU , 2016 Hydrological Forecasting Practices in Brazil Fernando Fan (1), Rodrigo Paiva (1), Walter Collischonn (1), and Maria-Helena Ramos (2) (1) Institute for Hydraulic Research, Federal University of Rio Grande do Sul, Porto Alegre, Brazil (2) Irstea, Hydrosystems and Bioprocesses Research Unit, 1, rue Pierre Gilles de Gennes, F , Antony, France This work brings a review on current hydrological and flood forecasting practices in Brazil, including the main forecasts applications, the different kinds of techniques that are currently being employed and the institutions involved on forecasts generation. A brief overview of Brazil is provided, including aspects related to its geography, climate, hydrology and flood hazards. A general discussion about the Brazilian practices on hydrological short and medium range forecasting is presented. Detailed examples of some hydrological forecasting systems that are operational or in a research/pre-operational phase using the large scale hydrological model MGB-IPH are also presented. Finally, some suggestions are given about how the forecasting practices in Brazil can be understood nowadays, and what are the perspectives for the future.

6 Vol. 18, EGU , 2016 Hydrologic ensemble prediction: enhancing science, operation and application through HEPEX Fredrik Wetterhall (1), Maria-Helena Ramos (2), Qj Wang (3), and Andy Wood (4) (1) European Centre for Medium Range Weather Forecasts, Reading, United Kingdom (2) Irstea, Hydrology Research Group, Antony, France, (3) CSIRO Land and Water, Clayton, Australia, (4) National Center for Atmospheric Research, Boulder, USA HEPEX (Hydrologic Ensemble Prediction Experiment) was established in March 2004 at a workshop hosted by the European Centre for Medium Range Weather Forecasts (ECMWF), co-sponsored by the US National Weather Service (NWS) and the European Commission (EC). HEPEX and the community it represents has over its more than 10 years of existence continuously worked to promote and advance the science of hydrologic ensemble prediction as well as operational systems and water management applications. Through workshops and conference sessions, HEPEX has connected the research community, forecasters and forecast users and facilitated the exchange of ideas, data, methods and experience. In particular, the establishment of an online blog portal has greatly enhanced community interaction and knowledge sharing ( HEPEX has now a strong and active community of nearly 400 researchers and practitioners around the world. In this poster, we present an overview of recent and planned HEPEX activities, and highlight opportunities to further progress ensemble prediction science, operation and application.

7 Vol. 18, EGU , 2016 Future discharge of the French tributaries of the Rhine: a semi-distributed multi-model approach using CMIP5 projections Guillaume Thirel (1), Alban de Lavenne (1), Jean-Pierre Wagner (2), Charles Perrin (1), Kai Gerlinger (3), Gilles Drogue (4), and Benjamin Renard (5) (1) IRSTEA, Hydrology Research Group (HBAN), Antony, France (guillaume.thirel@irstea.fr), (2) DREAL Lorraine, Metz, France, (3) HYDRON, Karlsruhe, Germany, (4) Laboratoire LOTERR, Université de Lorraine, Metz, France, (5) Irstea, HHLY, Lyon, France Several projects studied the impact of climate change on the Rhine basin during the past years, using the CMIP3 projections (see Explore2070, FLOW MS, RheinBlick2050 or VULNAR), either on the French or German sides. These studies showed the likely decrease of low flows and a high uncertainty regarding the evolution of high flows. This may have tremendous impacts on several aspects related to discharge, including pollution, flood protection, irrigation, rivers ecosystems and drinking water. While focusing on the same basin (or part of it), many differences including the climate scenarios and models, the hydrological models and the study periods used for these projects make the outcomes of these projects difficult to compare rigorously. Therefore the MOSARH21 (stands for MOselle-SArre-RHine discharge in the 21st century) was built to update and homogenise discharge projections for the French tributaries of the Rhine basin. Two types of models were used: the physically-oriented LARSIM model, which is widely used in Germany and was used in one of the previous projects (FLOW MS), and the semi-distributed conceptual GRSD model tested on French catchments for various objectives. Through the use of these two hydrological models and multiple sets of parameters obtained by various calibrations runs, the structural and parametric uncertainties in the hydrological projections were quantified, as they tend to be neglected in climate change impact studies. The focus of the impact analysis is put on low flows, high flows and regime. Although this study considers only French tributaries of the Rhine, it will foster further cooperation on transboundary basins across Europe, and should contribute to propose better bases for the future definition of adaptation strategies between riverine countries.

8 Vol. 18, EGU , 2016 The game of making decisions under uncertainty: How sure must one be? Micha Werner (1,2), Jan Verkade (2), Fredrik Wetterhall (3), Schalk-Jan van Andel (1), and Maria-Helena Ramos (4) (1) UNESCO-IHE Institute of Water Education, Water Engineering, Delft, Netherlands (2) Deltares, PO Box 177, 2600MH Delft, The Netherlands, (3) ECMWF, European Centre for Medium-Range Weather Forecasts, Shinfield Park, Reading, UK, (4) Irstea, Hydrology Research Group, Antony, France Probabilistic hydrometeorological forecasting is now widely accepted to be more skillful than deterministic forecasts, and is increasingly being integrated into operational practice. Provided they are reliable and unbiased, probabilistic forecasts have the advantage that they give decision maker not only the forecast value, but also the uncertainty associated to that prediction. Though that information provides more insight, it does now leave the forecaster/decision maker with the challenge of deciding at what level of probability of a threshold being exceeded the decision to act should be taken. According to the cost-loss theory, that probability should be related to the impact of the threshold being exceeded. However, it is not entirely clear how easy it is for decision makers to follow that rule, even when the impact of a threshold being exceeded, and the actions to choose from are known. To continue the tradition in the Ensemble Hydrometeorological Forecast session, we will address the challenge of making decisions based on probabilistic forecasts through a game to be played with the audience. We will explore how decisions made differ depending on the known impacts of the forecasted events. Participants will be divided into a number of groups with differing levels of impact, and will be faced with a number of forecast situations. They will be asked to make decisions and record the consequence of those decisions. A discussion of the differences in the decisions made will be presented at the end of the game, with a fuller analysis later posted on the HEPEX web site blog (

9 Vol. 18, EGU , 2016 A space-time geostatistical framework for ensemble nowcasting using rainfall radar fields and gauge data Angelica Caseri (1), Maria Helena Ramos (2), Pierre Javelle (1), and Etienne Leblois (3) (1) Irstea, Aix-en-Provence, France (angelica.caseri@irstea.fr; pierre.javelle@irstea.fr), (2) Irstea, Antony, France (maria-helena.ramos@irstea.fr), (3) Irstea, Lyon, France (etienne.leblois@irstea.fr) Floods are responsible for a major part of the total damage caused by natural disasters. Nowcasting systems providing public alerts to flash floods are very important to prevent damages from extreme events and reduce their socio-economic impacts. The major challenge of these systems is to capture high-risk situations in advance, with good accuracy in the intensity, location and timing of future intense precipitation events. Flash flood forecasting has been studied by several authors in different affected areas. The majority of the studies combines rain gauge data with radar imagery advection to improve prediction for the next few hours. Outputs of Numerical Weather Prediction (NWP) models have also been increasingly used to predict ensembles of extreme precipitation events that might trigger flash floods. One of the challenges of the use of NWP for ensemble nowcasting is to successfully generate ensemble forecasts of precipitation in a short time calculation period to enable the production of flood forecasts with sufficient advance to issue flash flood alerts. In this study, we investigate an alternative space-time geostatistical framework to generate multiple scenarios of future rainfall for flash floods nowcasting. The approach is based on conditional simulation and an advection method applied within the Turning Bands Method (TBM). Ensemble forecasts of precipitation fields are generated based on space-time properties given by radar images and precipitation data collected from rain gauges during the development of the rainfall event. The results show that the approach developed can be an interesting alternative to capture precipitation uncertainties in location and intensity and generate ensemble forecasts of rainfall that can be useful to improve alerts for flash floods, especially in small areas.

10 Vol. 18, EGU , 2016 Evaluation of a semi-distributed model through an assessment of the spatial coherence of Intercatchment Groundwater Flows Alban de Lavenne, Guillaume Thirel, Vazken Andréassian, Charles Perrin, and Maria-Helena Ramos Irstea, HBAN, Antony, France (alban.de-lavenne@irstea.fr) Semi-distributed hydrological models aim to provide useful information to understand and manage the spatial distribution of water resources. However, their evaluation is often limited to independent and single evaluations at each sub-catchment within larger catchments. This enables to qualify model performance at different points, but does not provide a coherent assessment of the overall spatial consistency of the model. To cope with these methodological deficiencies, we propose a two-step strategy. First, we apply a sequential spatial calibration procedure to define spatially consistent model parameters. Secondly, we evaluate the hydrological simulations using variables that involve some dependency between sub-catchments to evaluate the overall coherence of model outputs. In this study, we particularly choose to look at the simulated Intercatchment Groundwater Flows (IGF). The idea is that the water that is lost in one place should be recovered somewhere else within the catchment to guarantee a spatially coherent water balance in time. The model used is a recently developed daily semi-distributed model, which is based on a spatial distribution of the lumped GR5J model. The model has five parameters for each sub-catchments and a streamflow velocity parameter for flow routing between them. It implements two reservoirs, one for production and one for routing, and estimates IGF according to the level of the second in a way that catchment can release water to IGF during high flows and receive water through IGF during low flows. The calibration of the model is performed from upstream to downstream, making an efficient use of spatially distributed streamflow measurements. To take model uncertainty into account, we implemented three variants of the original model structure, each one computing in a different way the IGF in each sub-catchment. The study is applied on over 1000 catchments in France. By exploring a wide area and a variability of hydrometeorological conditions, we aim to detect IGF even between catchments which can be quite distant from one another.

11 Vol. 18, EGU , 2016 airgr: a suite of lumped hydrological models in an R-package Laurent Coron (1,2), Charles Perrin (1), Olivier Delaigue (1), Vazken Andréassian (1), and Guillaume Thirel (1) (1) IRSTEA, Hydrology Research Group (HBAN), Antony, France (guillaume.thirel@irstea.fr), (2) Now at EDF DTG, Toulouse, France Lumped hydrological models are useful and convenient tools for research, engineering and educational purposes. They propose catchment-scale representations of the precipitation-discharge relationship. Thanks to their limited data requirements, they can be easily implemented and run. With such models, it is possible to simulate a number of hydrological key processes over the catchment with limited structural and parametric complexity, typically evapotranspiration, runoff, underground losses, etc. The Hydrology Group at Irstea (Antony) has been developing a suite of rainfall-runoff models over the past 30 years with the main objectives of designing models as efficient as possible in terms of streamflow simulation, applicable to a wide range of catchments and having low data requirements. This resulted in a suite of models running at different time steps (from hourly to annual) applicable for various issues including water balance estimation, forecasting, simulation of impacts and scenario testing. Recently, Irstea has developed an easy-to-use R-package (R Core Team, 2015), called airgr, to make these models widely available. It includes: - the water balance annual GR1A (Mouehli et al., 2006), - the monthly GR2M (Mouehli, 2003) models, - three versions of the daily model, namely GR4J (Perrin et al., 2003), GR5J (Le Moine, 2008) and GR6J (Pushpalatha et al., 2011), - the hourly GR4H model (Mathevet, 2005), - a degree-day snow module CemaNeige (Valéry et al., 2014). The airgr package has been designed to facilitate the use by non-expert users and allow the addition of evaluation criteria, models or calibration algorithms selected by the end-user. Each model core is coded in FORTRAN to ensure low computational time. The other package functions (i.e. mainly the calibration algorithm and the efficiency criteria) are coded in R. The package is already used for educational purposes. The presentation will detail the main functionalities of the package and present a case study application. References: - Le Moine, N. (2008), Le bassin versant de surface vu par le souterrain : une voie d amélioration des performances et du réalisme des modèles pluie-débit?, PhD thesis (in French), UPMC, Paris, France. - Mathevet, T. (2005), Quels modèles pluie-débit globaux pour le pas de temps horaire? Développement empirique et comparaison de modèles sur un large échantillon de bassins versants, PhD thesis (in French), ENGREF - Cemagref (Antony), Paris, France. - Mouelhi S. (2003), Vers une chaîne cohérente de modèles pluie-débit conceptuels globaux aux pas de temps pluriannuel, annuel, mensuel et journalier, PhD thesis (in French), ENGREF - Cemagref Antony, Paris, France. - Mouelhi, S., C. Michel, C. Perrin and V. Andréassian (2006), Stepwise development of a two-parameter monthly water balance model, Journal of Hydrology, 318(1-4), , doi: /j.jhydrol Perrin, C., C. Michel and V. Andréassian (2003), Improvement of a parsimonious model for streamflow simulation, Journal of Hydrology, 279(1-4), , doi: /s (03) Pushpalatha, R., C. Perrin, N. Le Moine, T. Mathevet and V. Andréassian (2011), A downward structural sensitivity analysis of hydrological models to improve low-flow simulation, Journal of Hydrology, 411(1-2), 66-76, doi: /j.jhydrol R Core Team (2015). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL - Valéry, A., V. Andréassian and C. Perrin (2014), "As simple as possible but not simpler": What is useful in a temperature-based snow-accounting routine? Part 2 - Sensitivity analysis of the Cemaneige snow accounting routine on 380 catchments, Journal of Hydrology, doi: /j.jhydrol

12 Vol. 18, EGU , 2016 Modelling the snowmelt and the snow water equivalent by creating a simplified energy balance conceptual snow model Philippe Riboust (1,2), Guillaume Thirel (2), Nicolas Le Moine (1), and Pierre Ribstein (1) (1) Sorbonne Universités, UPMC Univ Paris 06, CNRS, EPHE, UMR 7619 Metis, 4 place Jussieu, PARIS, FRANCE, (2) Hydrosystems and Bioprocesses Research Unit (HBAN), Irstea, 1, rue Pierre-Gilles de Gennes, CS 10030, Antony Cedex, France A better knowledge of the accumulated snow on the watersheds will help flood forecasting centres and hydro-power companies to predict the amount of water released during spring snowmelt. Since precipitations gauges are sparse at high elevations and integrative measurements of the snow accumulated on watershed surface are hard to obtain, using snow models is an adequate way to estimate snow water equivalent (SWE) on watersheds. In addition to short term prediction, simulating accurately SWE with snow models should have many advantages. Validating the snow module on both SWE and snowmelt should give a more reliable model for climate change studies or regionalization for ungauged watersheds. The aim of this study is to create a new snow module, which has a structure that allows the use of measured snow data for calibration or assimilation. Energy balance modelling seems to be the logical choice for designing a model in which internal variables, such as SWE, could be compared to observations. Physical models are complex, needing high computational resources and many different types of inputs that are not widely measured at meteorological stations. At the opposite, simple conceptual degree-day models offer to simulate snowmelt using only temperature and precipitation as inputs with fast computing. Its major drawback is to be empirical, i.e. not taking into account all of the processes of the energy balance, which makes this kind of model more difficult to use when willing to compare SWE to observed measurements. In order to reach our objectives, we created a snow model structured by a simplified energy balance where each of the processes is empirically parameterized in order to be calculated using only temperature, precipitation and cloud cover variables. This model s structure is similar to the one created by M.T. Walter (2005), where parameterizations from the literature were used to compute all of the processes of the energy balance. The conductive fluxes into the snowpack were modelled by using analytical solutions to the heat equation taking phase change into account. This approach has the advantage to use few forcing variables and to take into account all the processes of the energy balance. Indeed, the simulations should be quick enough to allow, for example, ensemble prediction or simulation of numerous basins, more easily than physical snow models. The snow module formulation has been completed and is in its validation phase using data from the experimental station of Col de Porte, Alpes, France. Data from the US SNOTEL product will be used in order to test the model structure on a larger scale and to test diverse calibration procedures, since the aim is to use it on a basin scale for discharge modelling purposes.

13 Vol. 18, EGU , 2016 Consistency of internal fluxes in a hydrological model running at multiple time steps Andrea Ficchi, Charles Perrin, and Vazken Andréassian Irstea, HBAN - Hydrosystèmes et bioprocédés, Antony, France (andrea.ficchi@irstea.fr) Improving hydrological models remains a difficult task and many ways can be explored, among which one can find the improvement of spatial representation, the search for more robust parametrization, the better formulation of some processes or the modification of model structures by trial-and-error procedure. Several past works indicate that model parameters and structure can be dependent on the modelling time step, and there is thus some rationale in investigating how a model behaves across various modelling time steps, to find solutions for improvements. Here we analyse the impact of data time step on the consistency of the internal fluxes of a rainfall-runoff model run at various time steps, by using a large data set of 240 catchments. To this end, fine time step hydro-climatic information at sub-hourly resolution is used as input of a parsimonious rainfall-runoff model (GR) that is run at eight different model time steps (from 6 minutes to one day). The initial structure of the tested model (i.e. the baseline) corresponds to the daily model GR4J (Perrin et al., 2003), adapted to be run at variable sub-daily time steps. The modelled fluxes considered are interception, actual evapotranspiration and intercatchment groundwater flows. Observations of these fluxes are not available, but the comparison of modelled fluxes at multiple time steps gives additional information for model identification. The joint analysis of flow simulation performance and consistency of internal fluxes at different time steps provides guidance to the identification of the model components that should be improved. Our analysis indicates that the baseline model structure is to be modified at sub-daily time steps to warrant the consistency and realism of the modelled fluxes. For the baseline model improvement, particular attention is devoted to the interception model component, whose output flux showed the strongest sensitivity to modelling time step. The dependency of the optimal model complexity on time step is also analysed. References: Perrin, C., Michel, C., Andréassian, V., Improvement of a parsimonious model for streamflow simulation. Journal of Hydrology, 279(1-4): DOI: /S (03)