The prediction of river morphology evolution is very complicated, especially in the case of mountain and Piedmont rivers with complex morphologies, steep slopes, and heterogeneous grain sizes. The Lac des Gaves (LDG) reach, located within the Gave de Pau River in the Hautes-Pyrénées department, France, has precisely the complex morphological characteristics mentioned above. This reach has gone through severe sediment extractions for over 50 years, leading to the construction of two weirs for riverbed stabilisation. Two large floods resulted in changes in the LDG's hydromorphological characteristics as it went from a single channel river section to a braided river reach. In this study, a 2D hydromorphological model is developed with the TELEMAC-MASCARET system to reproduce the evolution of the channel following a flood that occurred in 2018. The model's validity is assessed by comparing the simulated topographic evolution to the observed one. The results reveal the challenge to choose well-fitted sediment transport equations and friction laws that would make it possible to reproduce such complex morphology. Although the exact localisation of the multiple channels forming the braided nature of the LDG was challenging to reproduce, our model was able to provide reliable volumetric predictions as it reproduces the filling of the LDG correctly. The influence of the two weirs on the river's current and future morphology is also studied. The aim is to provide decision-makers with more reliable predictions to design suitable restoration measures for the LDG reach.

Flood events can lead to considerable sediment transport that has an influence on flow dynamics. Understanding the interactions between flow dynamics and morphological changes is thus of growing interest in the research community

River restoration for flood prevention purposes is generally related to achieving a sufficient degree of protection through the design of solutions ranging from the installation of physical infrastructures to alternative measures for risk reduction

When the model is well calibrated and sufficiently validated with real field data, the main advantage of modelling is that it is possible to simulate restoration scenarios challenging to implement in the field

The effects of the interactions between hydrodynamics and morphodynamics have proven to be particularly dramatic during the flood of 2013, an almost 100-year return period event that severely impacted the Gave de Pau catchment, especially the Lac des Gaves reach in the Hautes-Pyrénées department in France, named LDG here. This former artificial lake within the Gave de Pau riverbed, delimited by two weirs, has undergone years of sediment extractions. These activities have led to a robust hydromorphological imbalance that is disturbing the watercourse's normal functioning in this area. Since the flood of June 2013, the lake has been almost completely filled with sediment, which may lead to river diversion towards populated areas. Upstream the second weir, the Gave de Pau has precisely the complex morphological properties mentioned above. In this area, the river presents specific aspects of Piedmont rivers, characterised by very heterogeneous grain sizes and a complex braided morphology, which indicates considerable sediment delivery from the upstream catchments. Conversely, downstream the weir, an active channel shrinkage is observed, characteristic of a sediment deficit and a sediment discontinuity that has led to serious ecological damage and navigation problems.

The TELEMAC-MASCARET (

The present work serves to illustrate (1) the ability of a 2D numerical model to reproduce hydromorphological processes in complex river morphology, (2) the performance of different friction laws and sediment transport equations, and (3) how a 2D hydromorphological model can help river managers better understand the dynamics within the LDG reach in order to evaluate the impacts of a given restoration measure on the system and adopt a sustainable and rational management orientation

The paper is organised as follows: Sect.

The Gave de Pau watershed (Fig.

The Gave de Pau catchment and its main upstream sub-catchments: the Gave de Cauterets and the Gave de Gavarnie sub-catchments (

The Gave de Cauterets subcatchment has a compact shape and besides the Cambasque (left bank) and Lutour (right bank) rivers, its other tributaries are very small and they have a typical torrent morphology with very steep slopes. The main valley and its tributary valleys have a south–north orientation. In these conditions, we may expect a simultaneous hydrological functioning of its main tributaries in case of widespread precipitation and thus a rapid concentration of overland flow as soon as we reach the city of Cauterets. From the Spanish border (south), where its main tributaries originate, to its outlet, the drainage slope is rarely below 2 %, and it reaches its maximum in the gorges areas. The mean elevation of the watershed is above 2000 m a.s.l. and two-thirds of its surface is between 1500 and 2500 m a.s.l.

Before its junction with the Gave de Cauterets, the Gave de Gavarnie drains a very wide catchment, whose area is approximately 486 km

As for the Gave de Cauterets and its main tributaries, the Gave de Gavarnie valley has a south–north orientation. Unlike the Gave de Cauterets catchment with its compact morphology, the Gave de Gavarnie watershed is wider and the supplies of its main tributaries are gradually distributed from upstream to downstream. It is thus very probable that these physiographic factors condition an uneven distribution of precipitation at the catchment scale. During rainy events, we may expect contrasted repartitions from one tributary to another. Therefore, different hydrological situations may sometimes lead to the same discharge and volume at the outlet. The main morphometric characteristics of each subcatchment are presented in Table ^{®}, BD ALTI^{®}, Corine Land Cover^{®}, IGN^{®}).

Main characteristics of the Gave de Cauterets and the Gave de Gavarnie subcatchments.

The LDG (Fig.

Morphological changes observed in the LDG reach. Aerial photo in

Like several research

Some examples of damage caused by the flood of June 2013 at different locations and different streams.

During this event, the LDG acted like a sediment trap as it intercepted almost all sediment coming from the upstream catchments (Figs.

Longitudinal evolution of the Lac des Gaves reach following the flood of 2018.

To re-establish the natural flow, reduce flood risks, and restore ecological continuity, river managers are considering lowering or even suppressing the weirs. However, even if these restoration measures seem to be relevant over the long term, many hydromorphological and ecological effects might emerge, such as backward erosion, over-delivery of sediment to the downstream fluvial system, to name a few

A hydromorphological 2D model was developed at the LDG reach scale to understand the different morphological processes within this channel and help river managers make an informed decision on the restoration of this reach. One of the processes on which the modelling efforts will focus is the deposition phenomenon within the LDG as it represents the potential volumes that might be mobilised if the weir-lowering/removal restoration measure is considered.

The system TELEMAC-MASCARET is considered for the numerical simulations. TELEMAC-MASCARET is an open-source software package with numerous modules to compute free surface flows, sediment transport, swell, and water quality

The hydrodynamic module, TELEMAC2D, solves shallow water equations (SWE) simultaneously

The TELEMAC model treats turbulence from a diffusion term. Four options are available and they were all tested in the framework of the study.

Constant viscosity model: the associated coefficient represents molecular viscosity, turbulent viscosity, and dispersion.

Elder model: this takes into account the dispersion by assuming that the vertical profiles of the velocities are logarithmic.

Smagorinsky model: this is generally used for maritime domains with large-scale fluctuation phenomena.

The morphodynamic module is based on the

Two friction laws were considered: the widely known Manning–

The Manning–Strickler friction law can be expressed as follows (Eq.

The morphodynamic module SISYPHE considers several semi-empirical sediment transport equations

The Meyer-Peter–Müller equation is a threshold equation and its original formulation considers a critical Shields parameter equal to 0.047. A sensitivity analysis was performed on this parameter as its value can highly influence sediment transport. The equation is written as follows (Eq.

This formulation is primarily based of laboratory experimentation with uniform and non-uniform sediments. It is one of the most used equations when it comes to studying a river or a laboratory case study with a heterogeneous grain size. This characteristic makes it adapted to the LDG reach. However, the fact that it is only calibrated with laboratory measurements can lead to non-realistic results with in situ input data. In addition, the Meyer-Peter–Müller equation is an excess shear relationship, and its original formulation considers a critical Shields parameter equal to 0.047 as a threshold for characterising the incipient motion of bed grains.

This non-threshold equation results from the work of

The Recking equation was coded in the subroutine “qsform.f” as it was not available among the proposed sediment transport equation in the SISYPHE module. The main advantages of this formulation are that

it considers partial transport,

it has been developed based on field data, which makes it adapted to cross-section-averaged calculations,

it has been validated with a wide data set for different independent watercourses, and

it is adapted to mountain and Piedmont rivers with steep slopes and coarse grain size.

A model was developed at the LDG's reach scale to reproduce the hydrodynamic and morphodynamic processes that occurred during the 10-year return flood of June 2018. In fact, it was the only event for which we had the before and after topo-bathymetric data, necessary to check the model's ability to reproduce the observed bed evolution modifications. The followed methodology considered field data collection for the model's development and performance evaluation, the model generation, the selection of a relevant hydrodynamic model, after which a clear hydrodynamic calibration with a fixed bed to select the riverbed roughness was performed, to finally run the morphodynamic model with the two different bedload transport equations.

The model starts at the junction of the Gave de Gavarnie and the Gave de Cauterets and extends up to the weir of the municipality of Agos-Vidalos (Fig.

A lidar digital elevation model (DEM) surveyed in 2016: the planimetric resolution is 1 m and the

A lidar DEM surveyed in 2019 a few months after the flood of June 2018.

Dredging data (SHEM) provided by the former operators of the weirs: these data give information on the possible bedload fraction that fills the LDG. Unfortunately, no grain-size distribution was available.

Grain-size data, collected on the ground over four sediment bars along the considered river reach. The value of the sediment diameter is directly obtained thanks to these grain-size measurements: the hydro-morphodynamic computations considered only

Hydraulic data representing water levels surveyed during the recession time of the 2018 flood event.

Overview of the considered area for hydromorphological modelling and identification of the different areas of interest.

It is common to use sediment trap dredging data to estimate event-driven sediment transport in mountainous catchments as its measurement can be complicated in such flow conditions

The input discharges were generated by the physically based distributed hydrological model MARINE

The MARINE model is capable of simulating flood hydrographs at any point in the drainage network, which is a real advantage in the accurate approximation of the inputs to the Lac des Gaves system. Thus, three hydrographs were extracted for the big mesh (Fig.

Considered meshes for the hydromorphological modelling. The orange lines represent soft lines corresponding to roads or river banks where we force the mesher to pass through.

We built unstructured triangulated meshes using the software BlueKenue (

An 8 km long unstructured triangulated mesh that covers the whole study area (355 062 elements) was built. The mesh size is 3 m within the watercourse, 2 m in the fishery water intake area, and 100 m in the floodplain.

A finer, 2 km long mesh in the LDG area around the two weirs (201 569 elements) was built. The mesh size for this smaller domain is 1 m in the riverbed, 2 m in the fishery water intake, and 20 m in the floodplain.

The finer mesh covers a much smaller area. Indeed, the computational cost with such a fine mesh on the whole domain would have been too high, so the finer mesh is used to perform a less time-consuming fine analysis of the sediment transport behaviour around the area of interest: the LDG between the two weirs. The obtained results with this small mesh allowed us to pick the best performing parameters for the whole domain with which we only simulated restoration scenarios, resulting in a substantial saving of time. To represent the anthropogenic structures along the river, fixed embankments, weirs and rip-raps were considered as non-erodible (blue in Fig.

For the mesh representing the entire study area, four boundary conditions were defined. Upstream, discharges are set as an input for the Gave de Gavarnie and Gave de Cauterets branches and the Gave d'Azun branch downstream the LDG. The downstream boundary condition is a free surface elevation determined by a rating curve calculated with a weir law (Eq.

As for the sediment transport boundary conditions, we first attempted to prescribe solid discharges estimated thanks to the 2018 hydrograph and the

For the smaller mesh, two boundary conditions were defined. The 2018 flow hydrograph for the Gave de Pau River is set as an input upstream. The downstream boundary condition is a free surface elevation estimated with the same weir law presented above (Eq.

Longitudinal profile of observed and simulated water surfaces corresponding to a discharge of 58.4 m

In a classical way, first hydrodynamic calculations are carried out. To calibrate the hydrodynamics, simulations on a steady state were performed for a discharge of 58.4 m

In the TELEMAC-MASCARET system, two categories of parameters can be adjusted: the numerical parameters (time step, type of solver, and its accuracy) and the physical ones

The hydro-morphodynamic simulations were based on the flood event of 2018 that we assumed to be responsible for the visible morphological changes between the two topographic campaigns of 2016 and 2019. Unfortunately, there have been no topographic campaigns between 2016 and 2019 that would account for the effects of the 2018 flood only. First, the friction coefficient and the turbulence model selected during the hydrodynamic calibration process were used for the first hydro-morphodynamic simulations. Then, other simulations for the two sediment transport (MPM and Recking) and friction (Ferguson and Strickler) equations were performed. The specific parameters of each sediment transport equation (Shields number, MPM coefficient, slope effect, etc.) were tested afterwards to analyse their influence on the performance of the simulations.

The model's qualitative performance evaluation was first done by visually comparing the simulated bed elevation changes maps to the DEM of difference (DoD). This allows for qualitative evaluation of the model's capacity to reproduce the spatial variability of the processes (erosion and deposition) and to locate possible aberrations. Longitudinal profiles and cross sections where significant morphodynamic processes occurred were also compared to acquire a more refined vision of the bed elevation changes at the local scale. However, in braided rivers, as in the Gave de Pau, significant variability is observed between two measurements considering the channel migration phenomena.

The model's calibration requires a considerable amount of computations where different parameters are modified. Rapidly identifying the best-performing model with a cost function can thus be time saving. The selected cost function for this is the Brier skill score (BSS). It was developed initially for the assessment of meteorological model performance and uses a baseline prediction to quantify a model's new prediction skill. Furthermore, over the last decades, many hydromorphological studies have considered it to evaluate the model's skill to simulate the sediment erosion and deposition processes along the whole domain

Classification of BSS values for model performance evaluation

Using a cost function to evaluate a hydromorphological model's performance with a braided morphology can be quite pessimistic. To date, numerical models cannot predict channel migration processes that occur in braided rivers. These phenomena are uncertain and random. A modeller should thus not expect the model to predict channel migration accurately during a flood. Despite these limitations, the choice of a 2D model has been made because it allows for better representation of the hydrodynamics and in particular of the friction taking into account a spatialisation of the water height. Even if the representation of the braiding and of the different flow arms is not the real one, the 2D model has the advantage of a continuity of the dynamics, contrary to the 1D model with interpolation between two profiles and water height projected on the DEM to estimate the extent of the flooded area.

As the issue here is the filling of the LDG and the high amount of sediment that might be delivered to the downstream system if the weirs are levelled, the comparison of the simulated deposited volumes with the field data appears to be a relevant model performance indicator. Field erosion and deposition areas were estimated through topo-bathymetric differencing between two lidar DEMs surveyed in 2016 and 2019 (Fig.

Eroded (red) and deposited (blue) areas in the LDG reach estimated through topo-bathymetric differencing between the two lidar DEMs surveyed in 2016 and 2019. The upstream weir is represented in black and the downstream weir in grey. This figure illustrates the filling of the LDG as almost all its surface represents deposited materials.

We also considered plotting on a histogram the area of bed experiencing morphological changes as performed by

The TELEMAC-MASCARET modelling system can run in parallel mode using domain decomposition and codes based on the Message Passing Interface (MPI) standard. The calibration scenarios have been carried out on a Linux server over 16 processors at the Institut de Mécanique des Fluides de Toulouse (IMFT). The performance of the two friction laws and the two bedload equations was assessed with the three performance indicators previously mentioned: longitudinal profile and cross-section comparison, BSS scores along with the whole domain, and comparison of the deposited volumes.

The simulated results are compared only within the intersection of areas that were emerged during the two lidar campaigns (2016–2019) as this technique does not collect submerged bathymetric data. In general, the model seems to correctly represent the filling tendency of the LDG as a deposition front can clearly be observed, which is coherent with what is observed in the field (Fig.

Comparison of the simulated bed elevation changes with the two bedload equations (MPM and Recking) and the two friction laws (Strickler and Ferguson).

Bed elevation changes around the Beaucens weir.

Comparison of simulated deposited volumes and observed ones using the score

If we compare the obtained results with the Ferguson friction law for the two bedload transport equations, the simulations with the MPM equation tend to predict higher sediment deposition and erosion amounts. As it is a threshold equation, the results below or around the critical shear stress can be poor because of a zero prediction or an overestimation of sediment transport. This equation is considered efficient when

The morphodynamics around the upstream weir (Fig.

Further quantitative investigations were done by comparing longitudinal profiles for both sediment transport equations and friction laws. For the MPM equation, longitudinal profile comparison confirms that the Strickler friction law tends to overestimate bedload deposition within the LDG (Fig.

Simulated longitudinal profile (solid lines) and maximum simulated water surface (dashed lines) comparison for the MPM equation and the two considered friction laws: STRICK for Strickler friction law, FERG for the Ferguson friction equation –

Simulated longitudinal profile (solid lines) and maximum simulated water surface (dashed lines) comparison for the Recking equation and the two considered friction laws: STRICK for Strickler friction law, FERG for the Ferguson friction equation –

The longitudinal profile comparison for both bedload transport equations with the Ferguson friction law (Fig.

The evaluation of the model's performance over the cross sections confirms this statement (Fig.

Simulated longitudinal profile comparison for the Recking and the MPM bedload equations and Ferguson friction law –

Cross-section comparison upstream (orange cross section) and downstream (blue cross section) the Beaucens weir

Finally, this topographic examination questions the classical performance analysis methods for morphodynamic models. Knowing the multiple variabilities in a mountain braided watercourse, performance criteria combined with local altimetric analysis might be too strict and incomplete to assess the ability of the model to reproduce the mobilised sediment volumes over a flood event. As the aim is to give the local elected representatives indications regarding possible sustainable restoration scenarios, a volumetric analysis can provide valuable additional insights as it gives information on the possible volumes that might end up downstream if a weir-lowering/removal solution is considered.

To compare the simulated deposited volumes to the ones observed in the field, the score

The reconstruction of the filling of the lake through different periods has allowed for the collection of interesting data that provide annual trends of material input. These results are derived from an analysis of bathymetric profiles from which the volumes were extracted. For the flood of 2018, a total (bedload and suspension) sediment deposition volume of 81 220 m

As mentioned above, the model only considers bedload transport. Hence, the score calculation was performed on the fraction of sediment estimated to be deposited via a bedload transport process. Generally speaking, the model seems to simulate sediment deposition close to the upper interval limit (16 % of the total deposited volume observed including both suspension and bedload transport). The best results (

The statistical distribution was analysed upstream the LDG (Fig.

Statistical distribution of erosion/deposition. The grey histogram shows observed bed elevation changes and the coloured ones show model predictions with the different bedload and friction equations within the LDG

The model seems to provide reliable results with the Ferguson friction law, so this equation was selected to perform restoration scenario simulations. The two bedload transport equations (MPM and Recking) were considered. The MPM is considered to provide more extreme results that we view as the worst-case scenario, whereas the Recking equation is considered more realistic or even to minimise the transported volumes. Two restoration scenarios were performed using the lidar DEM surveyed in 2019 as the initial topography: business-as-usual (BAU), corresponding to the current situation, where we consider that no restoration measure has been implemented, and weir lowering (WL), implemented through the modification of the bathymetry, for which the downstream weir (Préchac) is lowered by 2 m. To assess the influence of each scenario, two cross sections, upstream and downstream the weir, were analysed (Fig.

Panels

For the BAU scenario, very few morphological changes are observed upstream of the Préchac weir. Downstream, some deposition is observed with the MPM equation for the two different flood scenarios (one with a single 2018-like flood event and one with two consecutive 2018-like flood events). This consolidates the observations made above regarding the fact that this sediment transport equation tends to estimate more important sediment transport volumes than the Recking one for high flow situations. Very few changes are observed for the Recking equation for the BAU scenario. As expected, the WL scenario shows more bed elevation modifications for both sediment transport equations. Upstream the Préchac weir, very few changes are observed for the scenario with a single 2018-like flood event with the MPM equation. However, severe incisions are seen in the scenario with two consecutive 2018-like flood events (up to

The surprising nature of the results is the fact that the reaction of the model with the MPM equation for the scenario with a single 2018-like flood event seems to be very modest, whether it is upstream or downstream the Préchac weir. This might be due to the fact that this is a threshold equation and that the critical shear stress in this area might not be exceeded in order to generate sediment transport and thus morphological modification. Conversely, as the Recking equation is not a threshold one, partial transport is estimated even for small discharges, which can explain the observed morphological changes.

The results highlight the importance of the friction law as it conditions the results of the shear stress calculation and thus bedload transport. Friction laws are equations that usually link flow velocity to depth and roughness

For the two sediment transport equations, the model predicts that all deposition only occurs within the first half of the LDG. This could be due to the fact that many factors are not considered in the model such as the consideration of the whole grain-size distribution for bedload transport. In addition, the downstream part of the LDG is mainly composed of very fine sediment: bed elevation changes within this section are thus, for the most part, probably due to suspended load, not considered in our model. In addition, it is likely that the roughness parameters used by the two considered friction laws (

Each simulation was assessed with the BSS since this score was considered relevant for morphological change evaluation by the recent literature. More than 60 simulations were performed with different model parameter combinations. The best BSS results were obtained for simulations with very few riverbed changes (BSS

Two restoration scenarios were performed around the LDG reach: weir lowering (WL) and business-as-usual (BAU). As expected, the WL scenario showed significant bed elevation changes for both sediment transport equations (MPM and Recking), whereas the BAU scenario predicted very few changes. In any case, even if the simulated bed elevation changes after the weir lowering considerably enhances the ecological situation of the LDG reach by reactivating sediment continuity, allowing for the circulation of anadromous fishes, this scenario might pose serious operational problems for river managers. The upstream incision can, for example, induce

temporary bank erosion that can lead to the loss of portions of agricultural lands;

the propagation of the incision upstream until it meets a blocking point (the Beaucens weir, and then the same problem will be observed);

the lowering of the water table on which farmers depend;

the disconnection of the fishery water intake.

However, our model's simulated over-deposition and incision have some limitations. Indeed, only bedload was considered with coarse sediment and not the total sediment mixture. The suspended load is completely neglected. Still, a significant percentage of the LDG is composed of very fine sediments (silt, clay), especially downstream, which might be mobilised quickly after an action on the downstream weir. This means that a portion of the mobilised sediment will certainly be flushed far downstream and not have time to settle and induce the simulated morphological changes. Deposition would still be observed but to a lesser extent. In addition, our model only considers one homogeneous grain size in the whole domain, which might also explain the over-deposition simulated downstream. One potential improvement for the model would thus be to consider the whole grain-size distribution spatially distributed over the studied domain to have a more realistic view of the impacts of any restoration measure. This can be done with the brand new sediment transport module GAIA developed to handle grain-size issues better. To sum up, our model reproduces realistic tendencies but can still be improved to make better volumetric estimations. One recommendation to decision-makers is to not only consider the downstream weir but to consider both weirs in the restoration project. In addition, in such kinds of complex morphologies, the main advice is to consider an adaptive management strategy with step-by-step monitoring and eventual corrections if needed.

The evolution of river morphology is very complicated to predict, especially in the case of mountain and Piedmont rivers with complex morphologies. River restoration in such terrain can thus be challenging for river managers due to the random nature of riverbed evolution. Reliable hydromorphological numerical modelling combined with good field expertise can be helpful in this case for better river management. Within this framework, our study focused on the development of a 2D hydro-morphodynamic model over the Lac des Gaves reach in the Hautes-Pyrénées, France, with the TELEMAC-MASCARET system. This river reach has precisely the morphological characteristics mentioned above as it is a braided channel with a very heterogeneous grain-size distribution. The aim was to reproduce the bed elevation changes following the 2018 flood event that considerably impacted the channel's morphology to propose relevant and sustainable restoration solutions. Two bedload transport equations, Meyer-Peter–Müller and Recking, were used with the Ferguson and the Strickler friction laws to assess sediment transport processes. Three performance criteria were considered to assess the validity of the developed model: the comparison of longitudinal profiles, the Brier skill score, and the analysis of deposited volumes within the LDG.

The 2D hydro-morphodynamic model performed realistic simulations with the Ferguson friction law for both sediment transport equations (Recking and MPM). These results confirm the necessity of using a friction equation adapted to river reaches with high relative roughness and significant sediment load. The developed model tends to overestimate sediment deposition within the LDG. This might be due to the fact that it is a monodisperse model, considering bedload only with one homogeneous grain size, whereas in reality, finer sediments are also available. These are likely to be “flushed” and travel longer distances before being deposited, which is not simulated here. The simulated bed elevation changes can thus be considered to overestimate that which can actually be seen on the ground. Further improvements regarding these aspects are necessary, knowing the heterogeneity of sediment sizes within the LDG reach. Simulations on the updated sediment transport module GAIA, developed to handle the grain-size distribution issue better, are considered to improve the hydro-morphodynamic model.

Moreover, this study shows that the BSS might not be the right performance criterion to consider for rivers with braided morphologies. These complex configurations remain very difficult to reproduce using 2D models. The BSS score can thus give very pessimistic results, whereas the model correctly reproduces the most important processes (erosion and deposition areas). We recommend considering an integrative approach where the modeller combines multiple assessment criteria such as long profiles and cross-section elevation changes and volumetric estimations to evaluate the model's performance.

Finally, even if our model can still be improved, it provides valuable information on the possible consequences of a restoration scenario to river managers. Many operational issues were raised for the weir-lowering scenario, such as the increase in flood risks downstream or severe erosion upstream that could translate the issue to the upstream weir. Knowing the complexity of river restoration projects in these kinds of complex morphologies, considering an adaptive management strategy with step-by-step monitoring, and eventual corrections might be more appropriate rather than a radical measure. In addition, enhancing the hydro-morphodynamic model after considering the whole grain-size distribution and its actualisation after each morphogenetic event can be used as a decision-making tool that can assist river managers and help them communicate with elected representatives.

The Mascaret code is available at

Data can be accessed by contacting Olivier Frysou: olivier.frysou@plvg.fr.

RY carried out the simulations and the field experiments in collaboration with HR, LC, and OF. RY wrote the article with support from LC and HR. FP helped supervise the project and organise a stakeholder meeting to communicate the results with elected representatives.

The contact author has declared that none of the authors has any competing interests.

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors.

We are grateful to the Pays de Lourdes et des Vallées des Gaves public institution for having given us the opportunity to undertake this work. This research was funded by the three following groups: the Adour-Garonne Water Agency, the government, and the Occitanie Pyrénées Méditerranée region, whom we thank sincerely.

This research has been supported by the Agence de l'Eau Adour-Garonne (grant no. 240651666) and the Région Occitanie Pyrénées-Méditerranée (grant no. 16002759).

This paper was edited by Tom Coulthard and Lina Polvi Sjöberg and reviewed by Damien Kuss, Clément Misset, and Saraswati Thapa.