The authors wish to express their gratitude to the institutions that provided both financial and scientific support for this study: ADREUC – Association pour le développement des rencontres et des échanges universitaires et culturels; Agence de l’eau RMC; OFB; SMMAR; and SMAC. Special thanks are extended to the scientific, institutional, and local contributors, listed in alphabetical order: A. Boisselier, J. Chabaud, P.-H. Ilhès, C. Magro, H. Mathieu-Subias, F. Paraire, B. Sire, and J. Vidallier. We also acknowledge the faculty researchers (J.-M. Carozza, C. Bilodeau, S. Dahech, P. Passy) and students who participated in the DYNARISK – Dynamiques des milieux et risques master’s field courses during the microchipped cobbles injection and/or prospecting. Our thanks go to G. Fantino (GeoPeka) for his work in molding the artificial cobbles, and to the local residents, A. & D. Baudouy, whose assistance in drilling natural cobbles during the first field survey in 2012 was especially appreciated. Finally, the authors extend their heartfelt thanks to the two anonymous reviewers and the journal’s editor-in-chief for their valuable contributions in enhancing and facilitating the publication of this paper.

1. Introduction

1Over the past thirty years, water management in France has undergone a transformation, largely driven by political and public initiatives. Initially, efforts were focused on sanitation (e.g., reducing pollution) and hydraulic engineering (i.e., mitigating flood risk through channel rectification and recalibration). However, the focus has now shifted towards restoring more natural hydrosystem functions within watersheds (Morandi and Piégay, 2017). This marks a true paradigm shift, spurred by the Water Framework Directive (WFD 2000/60/EC). River restoration has thus become a key tool for improving the ecological and hydromorphological status of watercourses and reducing associated risks (Adam et al., 2007; Habersack and Piégay, 2008; Dufour and Piégay, 2009; Bennett et al., 2011; Clifford, 2012; Souchon, 2012; Campana et al., 2014; Kurth and Schirmer, 2014; Valette et al., 2015; Wohl et al., 2015; Gurnell et al., 2016; Lespez and Germaine, 2016; Rinaldi et al., 2016; Germaine et al., 2019). Applying the concept of restoration in the field faces challenges due to the complexity of river types, particularly their capacity for adjustment, which largely depends on internal energy and material flows within the fluvial system (Brierley and Fryirs, 2005). The relationship between river and energy flow is captured by A. Brookes (1988) as specific stream power (ω). Rivers with a ω value below 35 W/m² have limited capacity for hydromorphological self-adjustment, with virtually no effect below 25 W/m². Rivers with a ω value between 35 and 100 W/m² possess some capacity for adjustment, which becomes much greater when ω reaches or exceeds 100 W/m². The energy available in the river is confronted with the coarse-sediment stock available in the active channel and its watershed (Lane, 1955), and aggradation (i.e., hydromorphological equilibrium) can only be achieved if the volume and size of the incoming bedload (Q_s) exceeds the river’s capacity [f(ω)] to transport it. Conversely, if [f(ω) > Q_s], the riverbed enters a phase of degradation (i.e., disequilibrium).

2The restoration of the left-bank tributaries of the Aude River in the Minervois benefits from favorable conditions (i.e., high-energy rivers with ω > 100 W/m² when the 2-yr flood discharge is surpassed, and a strong capacity for self-adjustment). However, it also presents a paradox: restoring rivers to their more natural state means creating more dynamic, wandering rivers, which can become more hazardous, given the human presence in floodplains. This raises an important question: can the watershed provide the additional sediment load needed to respond to the increased energy in the rivers? The Syndicat mixte des milieux aquatiques et des rivières (SMMAR), along with the Syndicat mixte Aude Centre (SMAC), began addressing this issue early on, starting in 2002, with involvement from Université Paris Cité/UMR 8586 PRODIG (noted UPC/PRODIG in the text) as early as 1999. River restoration in the Minervois illustrates the unique case of rivers that have undergone three key phases (fig. 1): (i) Rivers were braided or wandering since the beginning of the Quaternary; (ii) Since at least the 18^th century, human intervention has altered river courses and channel patterns, making them artificially narrow, confined, and sinuous, primarily for agricultural purposes (e.g., diverting water into canals for irrigation and floodplain cultivation, installing mills). As a result, riverbeds were transformed into water-management tools, but they became wholly inadequate for handling the sediment and water fluxes during flood events. The ecosystem was disrupted, with rivers losing virtually all lateral mobility and space for divagation. Thus, the challenge of restoring the riverbeds in the Aude watershed involves acting on two interdependent factors: hydromorphology and ecology. Hydromorphological efforts aim to: restore rivers’ mobility space (e.g., by removing embankments; Biron et al., 2014), which involves managing bank erosion, bed slope, and the free flow of water, species, and sediments; and reconnect rivers with their flood plains (Serra-Llobet et al., 2022), particularly by managing inundation during flood events at low-risk sites upstream of populated or urban areas. Ecological actions include: spatial management of riparian vegetation and its species (Piégay et al., 2005; Corenblit et al., 2020); developing alluvial forests in coordination with the alluvial aquifer and water tables; and rehabilitating wetlands (Ock et al., 2015), which can only be achieved by increasing the hydromorphological diversity of riverbeds.

3Over the past 25 years, numerous field visits by UPC/PRODIG (including 17 research missions, 24 fieldwork courses training 519 students, 32 master’s theses, and 2 PhD theses) have provided valuable feedback on the hydromorphological restoration carried out by local and regional river syndicates in the left-bank tributaries of the Aude River in the Minervois region. This feedback primarily concerns (i) the time required for riverbeds to adjust to new hydromorphological conditions. The paper begins with the assumption (Hypothesis 1) that the readjustment time for a high-energy river can be relatively short (just a few years) once freed from the embankment structures imposed by centuries of inappropriate river design. Furthermore, the paper examines (ii) the role of floods and regular hydrological events in the hydromorphological functioning of the channel, specifically focusing on hydraulic geometry, channel pattern, bedload transport, and sediment balance within the active channel. Two more hypotheses can be proposed. (Hypothesis 2) The rivers in the Minervois region have experienced a chronic sediment deficit since their modern fluvial change, initiated by three factors: channelization in the 18^th-20^th centuries; disconnection between slopes and channels in the mountains and insufficient sediment supply from tributaries; and the depletion of alluvial stock in watersheds due to the end of the Little Ice Age (characterized by cooler, more hydromorphogenic conditions) and the reforestation of mountain massifs between 1860-1882 and the interwar period, which aimed to combat the overproduction of debris from slope erosion during the ‘épisodes cévenols’ (heavy rainfall events from the Mediterranean; fig. 2). (Hypothesis 3) The restoration work initiated in 2003 by the SMMAR, in conjunction with the SMAC since 2017, aims to restore hydrosedimentary storage spaces (i.e., flood-wave damping) by promoting the reactivation of wandering active channels and widening them upstream of at-risk areas in the piedmont (fig. 1). If the restoration is successful, it should lead to a lasting shift in channel pattern (i.e., fluvial metamorphosis, sensu Schumm, 1977), for which preparations must be made. The scientific findings presented in this paper will assess the effectiveness of the SMMAR/SMAC restoration efforts over the past 20 years. Moreover, they will contribute to a better understanding of the complex spatial and temporal adjustments of high-energy, gravel-bed rivers that have begun to wander again in response to restoration operations aimed at removing bank-erosion protections and flood-defense structures.

2. Study area and hydromorphological-restoration context

2.1. Watersheds with contrasting hydrological conditions and high exposure to flood risk

4The studied watersheds drain three morpho-structural units: from upstream to downstream and from north to south. Firstly, the Montagne Noire, marking the southern terminus of the Massif Central, is predominantly composed of schist and limestone. Next, the Minervois piedmont, characterized by Tertiary molasse formations that result from the erosion of the Montagne Noire. Finally, this piedmont transitions downstream into the Carcassonne corridor, also known as the ‘gouttière de l’Aude’ (Aude’s gutter; fig. 3A). The Carcassonne corridor, situated between the Montagne Noire to the north and the Pyrenees to the south, acts as a connecting link between the Aquitaine Basin to the west and the Mediterranean Basin to the east.

5The studied watersheds, though small in size (surface area < 110 km²), are drained by high-energy, gravel-bed rivers (slope: 4.8-0.7 %; D₅₀: 133-54 mm; ω for 1-yr flood: 848-124 W/m²), which flow regime is temporary during the year (fig. 3B, C). Ordinary flows do not exceed a few cubic meters. However, the watersheds are highly prone to extreme flash floods [maximum peak flow (Q_PF) up to 337 m³/s], due to three groups of controlling factors identified by G. Arnaud-Fassetta and M. Fort (2011):

• predisposing factors (of a natural nature): Watersheds are exposed to barometric depressions rising from the Mediterranean towards the continent. The orientation of hydrographic networks favors flow towards the SE while depressions rise in the opposite direction (NW). The schistose mountains, a major source of debris, contrast with the sedimentary piedmont, where gentle slopes are prone to debris accumulation. Furthermore, the round shape of the upstream watersheds results in extremely short hydrological response times, taking less than 3 h for floodwaters to flow down the entire watershed;

• triggering factors (meteorological and hydrological): Mediterranean episodes, such as those Nov. 1999, Mar. 2011, Feb. 2017, Oct. 2018 (fig. 3C), cause exceptionally intense rainfall triggered by active thunderstorm lines that remain virtually stationary over the watersheds for more than 10 h, resulting in significant cumulative rainfall (> 200-300 mm/24 h). The extremely short response time (2 to 3 h) between the onset of rainfall and the onset of flooding reflects the very high intensity of the rainfall, the brief concentration time of these small watersheds, and the proportion of Hortonian runoff, which leads to rapid increases in peak discharge: between 191 m³/s (4.6 m³.s^-1.km^-2) in Caunes-Minervois and 236 m³/s (2.2 m³.s^-1.km^-2) in La Redorte in the Argent Double River (Arnaud-Fassetta et al., 2002), and between 234 m³/s (5.4 m³.s^-1.km^-2) in Villeneuve-Minervois and 337 m³/s (4.7 m³.s^-1.km^-2) in Villegly in the Clamoux River (Gaume, 2002) during the 60-150-yr flood of Nov. 1999;

• aggravating factors (natural or anthropogenic): Naturally, moderate rainfall in the preceding days or weeks saturates the soil, favoring Hortonian runoff on the day of the flood. In the very narrow, steeply-sloping gorges of the mountain, the energy of the runoff is increased tenfold, reinforcing hydro-sedimentary transfers to the piedmont. However, anthropogenic factors often exacerbate the situation during a flood event. For instance, bridges can create log jams, which increase the flooded areas upstream and subsequently elevate ω (up to 3,426 W/m²) and Q_PF downstream during break-up. Urban sprawl on interfluves (erosional glacis from the end of the Tertiary covered by thin Quaternary alluvium) makes soils impermeable, reinforcing Hortonian runoff.

6The economic and human stakes are high in the floodplains, especially in the piedmont, due to the large number of rural villages near the riverbeds, increased occupation of valley bottoms and interfluves over recent decades, dynamics of riparian and alluvial forest in the riverbeds, extensive vineyards, significant orchards and olive groves, cutting of woods and allotments in the floodplain, and the presence of the Canal du Midi at the watershed outlets, all within a context of declining water resources (Lavie et al., 2018).

2.2. Chronology of river-restoration efforts and research developments

7Research into applied fluvial geomorphology in the Minervois region began following the 60-150-yr flood of Nov. 12-13, 1999. This hydroclimatic disaster in the Aude department resulted in 25 fatalities, 1 missing person, and €400 million in material damage:

8three weeks later (Dec. 1999), a 6-day field mission by UPC/PRODIG evaluated the hydromorphological impacts in the Argent Double watershed, which experienced one of the heaviest recorded rainfalls (458 mm/48 h at the Caunes-Minervois station). The study highlighted recurring geomorphological phenomena such as debris flows in the mountain (Gazel/Rieussec tributary), gullying of molassic slopes, channel widening, and avulsions in the piedmont area. The study also examined issues related to embankment structures and their destruction or damage, as well as the challenges posed by confluences and the problematic, systematic straightening of river sinuosities. Additionally, the problem of logjams and other temporary or permanent obstacles obstructing water flow was raised. The study identified a single, narrow, constricted, and sinuous channel pattern that does not align with the river’s historical wandering behavior. These findings, reported by M. Fort et al. (2001) and G. Arnaud-Fassetta et al. (2002), emphasized the impracticality of replicating the same hydromorphological reconstruction in the Argent Double River (fig. 4A-D);

• in May 2002, the SMMAR was established to manage water and flood risks at the watershed level. In Apr. 2004, a meeting between UPC/PRODIG and SMMAR’s director, J. Chabaud, resulted in a report that provided a map of the flood hazard from the Nov. 12-13, 1999 event in the Argent Double and Rivassel floodplains. The report included a description of the hydromorphological malfunctions observed during the flood and proposed measures to reduce hydrological risk. These measures aimed to restore longitudinal and lateral continuity, expand the freedom space for rivers, and improve channel-floodplain connectivity through the potential flood-wave expansion zones (Arnaud-Fassetta et al., 2004);

• in Jul. 2006, the Commission des Hydrosystèmes Continentaux (Comité National Français de Géographie) organized a field trip to the Aude watershed, where the mapping of the space of good functioning in the Argent Double River was presented (Arnaud-Fassetta et al., 2006). This concept was intended to be later applied to other rivers, including the Orbiel and Clamoux (Arnaud-Fassetta and Fort, 2009). In Oct. 2007, a field meeting took place with M. Dupuis, the SMMAR river technician. He manages the Clamoux watershed using a ‘let the river do its thing’ approach, coupled with land acquisition along the watercourse initiated in 2005. Within just two years, a few inexpensive interventions (under €70,000) enabled the rivers to flow freely and be ready to wander (fig. 4E-F). The Clamoux River model was gradually applied to the Argent Double River by M. Pouillat, the SMMAR technician managing the watershed from 2008 onwards. In Nov. 2009, marking the tenth anniversary of the Nov. 1999 flood, UPC/PRODIG presented its systemic approach to flood-risk reduction in the Aude watershed (Arnaud-Fassetta and Fort, 2011);

• on Mar. 15, 2011, the Clamoux watershed experienced its first ‘morphogenic’ 50-yr flood since 1999, while the Argent Double watershed experienced a 5-yr flood. Research revealed a lack of precise data on bedload transport and sediment balance. In Oct. 2012, UPC/PRODIG injected the first bedload tracer particles equipped with PIT tags (passive transponders) into the Argent Double River, followed by similar injections into the Clamoux River in Nov. 2013. These microchip-equipped pebbles and cobbles (referred to as cobbles hereafter) were monitored annually from 2012 to 2017 for ‘cobble transport distance’ (CTD) in relation to hydrology and riverbed geometry, along with additional cobble injections and surveys (Michler, 2013; Chauvet and Dorval, 2014; Michler, 2014; Chauvet, 2015; Espiaut and Fontanieu, 2016; Gros and Vallin, 2016; Berthelot and Tourade, 2017). This research determined the critical water discharge (Q_c) at which cobbles begin to move: 6 m³/s in the Argent Double River and 14 m³/s in the Clamoux River, indicating that cobbles are mobilized at ordinary high-water levels (Michler et al., 2016).

• in Jan. 2017, the SMAC, a subdivision of the SMMAR, was created to manage rivers such as the Clamoux and Argent Double. This formation preceded the second ‘morphogenic’ 20-yr flood on Feb. 14, 2017, when the Q_PF reached 243 m³/s in the Clamoux River (Malves-Minervois station) and 123 m³/s in the Argent Double River (Les Salices station). The presence of UPC/PRODIG members at the start of the flood enabled precise dynamic hydromorphological mapping (Walentin, 2018). However, there remains a lack of detailed knowledge about slope modifications and sediment volumes eroded, transported, and deposited. In Mar. 2018, a meeting at the SMAC office in Conques-sur-Orbiel brought together key stakeholders, including H. Mathieu-Subias (SMMAR technical director), C. Magro (SMAC president), F. Housset (intervention officer, Agence de l’eau RMC), M. Dupuis (SMMAR river technician for the Clamoux River), G. Brousse (PhD student at UPC/PRODIG), and G. Arnaud-Fassetta (Professor at UPC/PRODIG). During this meeting, UPC/PRODIG provided a summary of the effectiveness of restoration work and proposed improvements for monitoring bedload transport and sediment balance in restored river sections using LiDAR surveys;

• on Oct. 15, 2018, the rivers Clamoux and Argent Double experienced their third ‘morphogenic’ 50-yr flood (Q_PF = 154 m³/s and 169 m³/s, respectively). Unfortunately, this was also the second dramatic hydroclimatic event, resulting in 14 deaths and €270 million in damages in the Aude department (Picot, 2019). In the tributary riverbeds, particularly in the Rieu Sec brook (a tributary of the Orbiel River), sediment balance was severely deficient. Over 50,000 m³ of alluvium were eroded down to bedrock in a 33 km² watershed, with Q_PF reaching 200 m³/s (6.1 m³.s^-1.km^-2; Bellon and Perrine, 2019). The return period of this hydroclimatic event in the Rieu Sec valley was estimated to be 204-296 years based on ¹⁴C dating of exhumed fossil wood. In Sept. 2020, UPC/PRODIG’s first PhD thesis (funded by the OFB and supervised by G. Arnaud-Fassetta and F. Liébault) was defended by G. Brousse (2020), focusing on the effectiveness of restoration work and the resilience of altered torrential rivers. The thesis concluded that while the restoration work was generally effective – reclaiming freedom space for rivers allowed the reactivation of lateral dynamics and habitat diversification – the sediment balance, measured only at the surface level, indicated high cobble mobility and issues with the dispersal of tracer cobbles equipped with PIT tags, making them increasingly difficult to monitor. In 2020-2021, AIT-tags (active transponders), funded by the OFB, were introduced into the cobbles of the rivers Clamoux and Argent Double to improve the monitoring of bedload transport (Theureaux, 2021). In Aug. 2021, a paper promoted the ‘metric grid’ as a method to facilitate the monitoring of riverbank dynamics (Arnaud-Fassetta et al., 2021), complementing the study of sedimentary facies in the channel (Buffington and Montgomery, 1997);

• in Jun. 2022, the final report on the monitoring of tracer cobble movements equipped with PIT tags was completed. This report followed a two-month survey of the rivers Clamoux and Argent Double and their tributaries (Flesch and Yvars, 2022; Gonod and Thas, 2022; Magallon and Pignot, 2022). In Sept. 2023, UPC/PRODIG launched a second PhD thesis (funded by the OFB) focusing on the Aude watershed, led by M. Brun, with supervision from G. Arnaud-Fassetta, D. Corenblit, and G. Melun. In collaboration with the Pôle écohydraulique of OFB Toulouse and the SMMAR/SMAC, the thesis has two primary goals: to measure and monitor river restoration (including hydromorphology, sediment transport, and sediment balance) and to quantify the response of environmental habitability to the riverbed-restoration efforts. In Nov. 2023, the SMAC received a study report on bedload transport (tracer cobbles) and sediment balance (using LiDAR) in river sections restored by the SMMAR (Arnaud-Fassetta and Brun, 2023). This study was part of a partnership between CNRS (UPC/PRODIG) and SMAC/Agence de l’eau RMC.

3. Methods

9Based on observations and measurements from fieldwork and image processing, the developed methods aimed to verify whether the restored sections of the left-bank tributaries of the Aude River function as hydrosedimentary storage areas and flood-energy absorption sites. This was accomplished by quantifying (i) CTD in the watersheds, and (ii) sediment balance in river sections where restoration work had been carried out, to understand the response to new flow and channel-pattern conditions.

3.1. Bedload dynamics

10Between 2012 and 2022, UPC/PRODIG injected and monitored 1618 cobbles equipped with microchips (RFID transponders) in the Argent Double (n = 672) and Clamoux (n = 946) watersheds to create a dynamic map and quantify the CTD. In rivers with contrasting hydrological regimes, which transport most of their cobbles only a few days each year, it is unrealistic to propose CTDs in m/yr. Such a measure would be meaningless and would exaggerate the reality of bedload transport. Therefore, we have expressed the average CTD in terms of the number of days per year, by thresholding discharge values greater than the flow that sets the bedload in motion over the study period. The average daily CTD (in m/d) is calculated by dividing the average CTD (in m) by the number of days (d) on which Q_PF exceeds 6 m³/s in the Argent Double River (between Caunes-Minervois and Peyriac-Minervois) and 14 m³/s in the Clamoux River (between Villeneuve-Minervois and Villegly). These critical values for bedload movement were determined empirically through numerous and repeated field observations, comparing hydrological conditions with cobble distance measurements. The variability of the distances traveled by the cobbles is discussed in relation to hydrological variability, particularly the effects of flooding compared to ordinary high water. Two tracking techniques were used to achieve this: cobbles equipped with either passive integrated transponder (PIT) tags or active integrated transponder (AIT) tags (fig. 5).

3.1.1. Tracer cobbles equipped with PIT tags

11PIT tagging is a widely used process for tracing river cobbles since the 2000s (Lamarre et al., 2005). PIT tags offer an effective tracing solution for understanding alluvial movements in riverbeds (e.g., Rollet et al., 2008; Chapuis et al., 2015; Piégay et al., 2016; Gilet et al., 2018; Graff et al., 2018; Dépret et al., 2023; Liébault et al., 2024b). Robust and durable for around 10 years, their tracking within the cobbles into which they are introduced provides information on areas of erosion, transport, and preferential accumulation along the active channel. The cobble is equipped with a microchip (brand: Oregon; length: 23 mm; diameter: 3.65 mm; weight: 0.6 g), which relays radio wave frequencies (LF 134.2 kHz) when it comes into contact with an electromagnetic field transmitted by the antenna. Since the chip has no energy source of its own, it is considered ‘passive’. Information concerning the chip’s location is received through frequency exchange, varying between high frequencies for remote reception and low frequencies for penetrating matter (useful if the PIT tag is buried). PIT tags are detected by the mobile RFID antenna when it passes within 70 cm of the equipped cobble (fig. 6A, B). When the PIT-tagged cobble is located, its position in the active channel (ordinary-flow channel, ordinary high-water alluvial bar, secondary-flood channel, or flood alluvial bar) is noted, along with the GPS coordinates. Between 2012 and 2017, 1,448 tracer particles equipped with PIT tags were injected into rivers (fig. 3). In the Argent Double watershed, between Oct. 2012 and Feb. 2017, 588 equipped cobbles were distributed over 11 sites as follows: in the Argent Double River (n_PIT = 398; from mountain to piedmont), and three of its tributaries (Gazel/Rieussec, Lavandières, Souc; n_PIT = 190). In the Clamoux watershed, between Nov. 2013 and Feb. 2017, 860 PIT tags were introduced into cobbles spread over 14 sites, including not only the Clamoux River (n_PIT = 616; from mountain top piedmont) but also three of its tributaries (Serremijeanne, Ganganel/Cros, Pémol; n_PIT = 244). The PIT tags were inserted into natural pebbles, cobbles, and boulders. The particle size distribution of the tracer particles was determined from the mean grain-size curve deduced from riverbed particle-size measurements (n = 2,027). Pebbles smaller than 32 mm could only be equipped with PIT tags with difficulty. Therefore, the particle-size classes selected represent 64 % of the riverbed’s grain-size distribution, with care taken to select 7 particle-size classes (Wentworth scale) representative of bedload transport, particularly those around the statistical mode: 32-45 mm and 45-64 mm (coarse pebbles); 64-90 mm, 90-128 mm, 128-180 mm, and 180-256 mm (cobbles); > 256 mm (boulders). At each site, an average of 60 cobbles equipped with PIT tags (Argent Double watershed: 40-75; Clamoux watershed: 13-120) were injected into the channel and onto the alluvial bars (when the cross-section was large and undiked), either in clusters (from 1 x 1 to 4 m, i.e., 1 to 4 m²) or across the entire active channel, depending on the characteristics of the river section. Five field surveys then took place in the Argent Double watershed (Oct. 2012; Mar. 2013; Mar. 2014; Feb. 2017; Mar.-Apr. 2022) and the Clamoux watershed (Nov. 2013; Mar. 2014; Feb. 2016; Feb. 2017; Mar.-Apr. 2022).

3.1.2. Tracer cobbles equipped with AIT tags

12AIT tags enable easier traceability of alluvial movement in river channels (e.g., Cassel et al., 2017; Brousse et al., 2020, 2021). Their detection radius is wider (typically from 5 to 20 m; depth up to 2 m), and data can be read and visualized directly on a field computer via dedicated software (ELOG). The AIT tags were inserted into artificial pebbles and cobbles made by GeoPeka (Fantino and Sigaud, 2021) from a resin and dense compound mixture designed to give them a density close to that of natural cobbles. These artificial pebbles and cobbles were molded from samples of natural particles (schist, sandstone, quartz). The size distribution of the artificial particles was based on the average particle-size curve obtained from analyzing 2100 natural particles directly measured in the rivers Argent Double (n = 1,200) and Clamoux (n = 900) (Wolman, 1954; Cuisenier, 2020). Pebbles smaller than 32 mm could not be equipped with AIT tags, and cobbles larger than 180 mm could not be molded. Therefore, the selected particle-size classes represent about 50 % of the riverbed’s grain-size distribution. Care was taken to select classes representative of bedload transport, particularly those close to the statistical mode, with the b-axis ranging from 32 mm to 180 mm, divided into 5 particle-size classes (Wentworth scale): 32-45 mm and 45-64 mm (coarse pebbles); 64-90 mm, 90-128 mm, and 128-180 mm (cobbles). Each AIT tag (brand: ELA; frequency: UHF 433.92 MHz) has a four-digit identifier. The cylindrical chip is relatively large (36 mm in diameter and 10 mm thick, weighing 11 g) and has a shorter lifespan (3 to 4 years, with a transmission frequency of 5 s). CIPAM’s Slender III mobile detection antenna, a 50 cm x 50 cm white slab, is connected to a PC via a USB cable (fig. 6C, D). When an AIT-tagged cobble is located, its position within the active channel (ordinary-flow channel, ordinary high-water alluvial bar, secondary-flood channel, or flood alluvial bar) is noted along with its GPS coordinates. The margin of error in x and y is ± 2.5 m. In March 2021, UPC/PRODIG injected 170 artificial particles equipped with AIT tags (fig. 3). In the Argent Double watershed, 84 tracer particles were injected across 4 sites. From upstream to downstream, the first site (LAVA) is in the Lavandières brook, southwest of Caunes-Minervois, and the other three are in the Argent Double River: at the confluence with the Lavandières brook (CONF), at the entrance to the restored section at La Prade (COUD), and at the foot of the former, unauthorized dump (DECH). Clusters were preferred when the channel was less than 9 m wide (LAVA, CONF, and COUD), and cross-sections were used when the active channel was wider than 9 m (DECH). In the Clamoux watershed, 86 tracer particles were injected across 4 sites. From upstream to downstream, the first site (PEM) is in the Pémol brook, a right-bank tributary of the Clamoux River between Cabrespine and Villeneuve-Minervois, and the other three are in the Clamoux River: at the Villeneuve-Minervois cemetery upstream of the restored section at Les Aygadons (CIM), at the entrance to the restored section at Les Aygadons, 50 m upstream of the bridge (AYM), and within the restored section at Les Aygadons (AYV). Clusters were preferred when the channel was less than 9 m wide (PEM, CIM, AYM), while cross-sections were used when the active channel was wider than 9 m (AYV). Four field surveys were then conducted in both watersheds: Nov. 2021; Mar. 2022; Apr. 2022; Oct. 2022. Additionally, a unique field survey in the river tributaries was carried out in Mar. 2024.

3.2. Sediment balance

13Airborne topographic LiDAR (light detection and ranging) surveys have been used to quantify sediment balances (2016-2021) in the river sections restored by the SMMAR/SMAC, and to understand the benefits compared to unrestored sections. The principle of LiDAR is based on an infrared laser that scans the landscape to measure the distance between the aircraft and the ground, aiming to produce a three-dimensional representation of the landscape with ‘Ultra High Definition’ (centimetric precision). LiDAR images are increasingly used to study changes in riverbeds (e.g., Charlton et al., 2003; Kondolf and Piégay, 2003; Notebaert et al., 2009; Tacon et al., 2011; Brasington et al., 2012; Bizzi et al., 2019; Brousse et al., 2020, 2021; Gilet et al., 2021; Liébault et al., 2024a). At least two LiDAR point clouds, acquired at different dates (monthly or annually), are transformed into high-resolution digital terrain models (DTMs; Lallias-Tacon et al., 2014). Diachronic analysis of these DTMs is used to quantify (in plan, elevation, and volume) the 3D evolution of alluvial forms and formations. Conversion from LiDAR point clouds to DTMs was carried out using open-source software packages QGIS and CloudCompare. The LiDAR point clouds were already classified according to the nature of the points surveyed (soil, vegetation, buildings). The data was then filtered using CloudCompare to retain only points belonging to the ‘soil’ class. Once filtered, CloudCompare was used to produce a 1-m resolution DTM from the ‘soil’ point cloud. The DTMs, once generated, were aligned using the QGIS tool ‘Align Rasters’ to perform diachronic analysis of two DTMs. This analysis involved subtracting the elevations of the most recent DTM from those of the oldest DTM, creating a new raster that highlights the accumulation areas and erosion zones. The QGIS tool ‘Raster Surface Volume’ was used to provide the surface areas and volumes mobilized. The restoration zones managed by the SMMAR/SMAC were the subject of 4 sequential airborne LiDAR surveys: May 9, 2016; Dec. 29, 2019; Oct. 2, 2020; and Aug. 27, 2021 (fig. 6E). The Root Mean Square Error (RMSE), which indicates the absolute error between the real values and the acquired values, range from 0.053 to 0.808 m. Two restoration zones were analyzed in particular: the Argent Double River (La Prade plain) and the Clamoux River (Les Aygadons plain). On each of the LiDAR images, we subdivided 3 sections: upstream of the restored section (a), the restored section (b), and downstream of the restored section (c), to better quantify the benefits of the intervention work in the riverbeds. Each section was quantified in terms of volume (in m³), which is positive (aggradation) or negative (erosion) depending on the hydromorphological dynamics. This volume was related to the surface area of the section (in m²) to obtain comparable values (in m³/m²) for sediment gain or loss at each site.

4. Results

4.1. Mapping and quantification of bedload dynamics (2012-2024)

14We will present the CTDs, highlighting the differences not only between the mountain and the piedmont, but also between the trunk channel and the tributaries, as well as between restored and unrestored river sections (fig. 7).

4.1.1. Argent Double watershed

15The results pertain to cobble-bedload monitoring using PIT-tag technology over the period 2012-2022, and AIT-tag technology over the period 2021-2024.

Montagne Noire

16In the active channel of the Argent Double River, from Oct. 22, 2012 to Mar. 19, 2013 (149 d), the hydrological regime was dominated by ordinary high water, with Q_PF = 12.1 m³/s (< 2-yr flood) reached on Mar. 6, 2013. A downstream decrease in CTD was observed in the ordinary channel, varying from 146 m/d [AR1 site, 73 % return rate (RR)] to 27 m/d (AR4 site, 83 % RR) over 1.5 d, during which ω decreased downstream from 296 to 171 W/m², respectively. From Mar. 19, 2013 to Mar. 15, 2014 (362 d), the hydrological regime remained dominated by ordinary high water, with the Q_PF not exceeding 7.6 m³/s (< 1-yr flood) reached on Apr. 1, 2013. A downstream decrease in CTD was confirmed, varying from 74 m/d (AR1 site, 75 % RR) to 12 m/d (AR4 site, 83 % RR) over 3 d, during which ω decreased downstream from 242 to 134 W/m², respectively. From Mar. 15, 2014 to Feb. 10, 2017 (1,064 d), the ordinary high-water context was slightly more sustained, with Q_PF = 22.6 m³/s (< 2-yr flood) on Jan. 28, 2017. As a result, CTD increased in the ordinary channel, varying downstream from 114 m/d (AR1 site, 15 % RR) to 56 m/d (AR2 site, 61 % RR) over 3 d, during which ω decreased from 479 to 387 W/m², respectively. From Feb. 10 to 25, 2017 (15 d), ordinary high water interspersed with a 20-yr flood, whose Q_PF = 97 m³/s (ω = 1,218-1,468 W/m²) at Citou was reached on Feb. 14, 2017. CTD was 26 m/d (AR1 site, 10 % RR) and 27 m/d (AR2 site, 4.5 % RR) over 10 d. From Feb. 25, 2017 to Apr. 25, 2022 (1,886 d), ordinary high water interspersed with a 50-yr flood, with Q_PF = 105 m³/s (ω = 1,318-1,451 W/m²) at Citou on Oct. 15, 2018. CTD varied from 0 (AR1 site, 2.5 % RR) to 32 m/d (AR2 site, 7.5 % RR) over 11.5 d.

17In the Gazel/Rieussec brook, a right-bank tributary, from Oct. 22, 2012 to Mar. 19, 2013 (149 d), the flow did not exceed a few liters per second on Mar. 6, 2013. CTD in the ordinary channel was 7 m/d (AR3 site, 98 % RR), for ω not exceeding 49 W/m². From Mar. 19, 2013 to Mar. 15, 2014 (362 d), the brook had flow (a few liters per second) over 1.5 d, around Apr. 1, 2013. CTD was 5 m/d (AR3 site, 100 % RR), with ω not exceeding 46 W/m². From Mar. 15, 2014 to Feb. 10, 2017 (1,064 d), the Gazel brook reached a Q_PF of 3 m³/s (< 2-yr flood; ω = 301 W/m²) on Jan. 28, 2017. CTD was 13 m/d (AR3 site, 98 % RR), over 2 d. From Feb. 10, 2017 to Apr. 25, 2022 (1,901 d), the Gazel brook experienced many days without flow, a few ordinary flows, a 20-yr flood and a 50-yr flood, resulting in a Q_PF of 14 m³/s (ω = 371 W/m²) on Feb. 14, 2017. CTD reached 37 m/d (AR3 site, 2.5 % RR), over 4 d.

Minervois piedmont

18In the active channel of the Argent Double River:

• in the upstream part of the piedmont, from Mar. 15, 2013 to Apr. 25, 2022 (3,329 d), the hydrological context was dominated by ordinary high water alternating with a 20-yr flood and a 50-yr flood, with Q_PF = 135 m³/s (ω = 2,119 W/m²) at Caunes-Minervois on Oct. 15, 2018. CTD was 28 m/d (AR5 site, 2 % RR) for 27.5 d. From Mar. 3, 2021 to Nov. 16, 2021 (259 d), the hydrological regime was dominated by a few ordinary flows (1.5 m³/s; ω = 26 W/m²). No movement of cobbles was observed (site CONF; 100 % RR). From Nov. 16, 2021 to Mar. 10, 2022 (115 d), ordinary high water caused the Q_PF to rise to 12 m³/s (< 2-yr flood) on Jan. 11, 2022 (ω = 188 W/m²). CTD was 296 m/d for a whole day (CONF site, 75 % RR). From Mar. 10 to Apr. 15, 2022 (37 d), it was then possible to measure the impact of the 5-yr flood of Mar. 2022 (Q_PF = 50 m³/s at Caunes-Minervois on Mar. 13, 2022) on cobble transit. The CONF injection site is located in a semi-constrained plane-bed section (presence of an old wall edging the active channel), leading to an increase in ω (706 W/m²) of the Argent Double River. As a result, the cobbles covered a remarkable distance, averaging 133 m/d over 8.5 d (CONF site, 75 % RR), with 39 % of the cobbles accumulating in the La Prade restored river section. From Apr. 15 to Oct. 10, 2022 (181 d), the hydrological regime became calm again, with the passage of ordinary high water whose Q_PF did not exceed 8.1 m³/s (< 2-yr flood) on Apr. 24, 2022 (ω = 135 W/m²). CTD in the ordinary channel was 121 m/d over 2 d (CONF site, 30 % RR);

• in the restored river section (fig. 8A), from Oct. 22, 2012 to Mar. 19, 2013 (149 d), the hydrological regime was dominated by ordinary high water, with Q_PF = 12.1 m³/s (< 2-yr flood; ω = 36 W/m²) on Mar. 6, 2013. No cobbles moved (AR6 site, 90 % RR). From Mar. 19, 2013 to Mar. 15, 2014 (362 d), the hydrological regime was dominated by ordinary high water, with Q_PF = 7.6 m³/s (< 1-yr flood; ω = 3 W/m²) on Apr. 1, 2013. CTD was 7 m/d over 3 d (AR6 site, 90 % RR). From Mar. 15, 2014 to Apr. 25, 2022 (2,964 d), ordinary high water interspersed with a 20-yr flood and a 50-yr flood, with Q_PF = 154 m³/s (La Redorte) on Oct. 15, 2018 (ω = 285 W/m²). CTD was 14 m/d over 24.5 d (AR6 site, 7.5 % RR). Cobbles equipped with AIT tags (COUD and DECH sites) will allow us to refine the impact of ordinary high water on bedload transport. The COUD site is located at the beginning to the La Prade river-restored section, where the first work took place in 2016: the single, narrow and dammed channel of the Argent Double River (cross-section: 11 m) now gives way to the start of a slightly wider active channel (cross-section: 21 m). The DECH site is represented as a transect laid out at the foot of the former Caunes-Minervois dump, perpendicular to the flow axis of the Argent Double active channel (cross-section: 63 m). From Mar. 3, 2021 to Nov. 16, 2021 (259 d), the hydrological regime was dominated by a few ordinary flows (1.5 m³/s; ω = 22 W/m²). No movement of cobbles was observed (COUD and DECH sites; 100 % RR). From Nov. 16, 2021 to Mar. 10, 2022 (115 d), ordinary high water caused the Q_PF to rise to 12 m³/s (< 2-yr flood) on Jan. 11, 2022 (ω = 120-129 W/m²). CTD in the ordinary channel was 128 m/d for a whole day (COUD and DECH sites, 45 % RR). From Mar. 10 to Apr. 15, 2022 (37 d), a 5-yr flood reached a Q_PF of 50 m³/s at Caunes-Minervois on Mar. 13, 2022, and ω decreased downstream from the COUD site (ω = 405 W/m²) to the DECH site (ω = 324 W/m²). CTD also decreased downstream, from 164 m/d (COUD site, 80 % RR) to 54 m/d (DECH site, 72 % RR) over 8.5 d. 45 % of cobbles are accumulated in the La Prade river-restored section. More than half of the cobbles are still in place in the alluvial bar deposited by the Feb. 2017-Oct. 2018 floods in the right bank of the Argent Double River. Their remoteness and relative height compared to the bottom of the main channel explain their immobility. In the alluvial bar, the 5-yr flood of Mar. 2022 displaced some cobbles by only a few tens of centimeters. A major flood (of the 2017 or 2018 type) is required to move them. From Apr. 15 to Oct. 10, 2022 (181 d), the hydrological regime was dominated by ordinary high water, with Q_PF = 8.1 m³/s (< 2-yr flood) on Apr. 24, 2022, and decreasing values downstream of ω, from 105 W/m² (COUD site) to 87 W/m² (DECH site). CTD also decreased downstream, from 91 m/d (COUD site, 35 % RR) to 42 m/d (DECH site, 72 % RR) over 2 d;

• in the downstream part of the piedmont, from Oct. 22, 2012 to Mar. 19, 2013 (149 d), the hydrological regime was dominated by ordinary high water, with Q_PF = 12.1 m³/s (< 2-yr flood) on Mar. 6, 2013. Firstly, ω decreased downstream from 142 W/m² (AR7 site) to 98 W/m² (AR8 site). Secondly, CTD, which was zero at AR7 site (98 % RR), increased downstream, with 69 m/d over 1.5 d at AR8 site (50 % RR). From Mar. 19, 2013 to Mar. 15, 2014 (362 d), the hydrological regime was dominated by ordinary high water, with Q_PF = 7.6 m³/s (< 1-yr flood) on 1 Apr. 2013. ω decreased downstream from 97 W/m² (AR7 site) to 70 W/m² (AR8 site), but CTD increased downstream, from 18 m/d (AR7 site, 100 % RR) to 31 m/d (AR8 site, 50 % RR) over 3 d.

19The alluvial contribution of tributaries to the Argent Double River is as follows (fig. 9D, E):

• in the Lavandières brook, a right-bank tributary, no cobbles moved between Mar. 3, 2021 and Oct. 10, 2022, due to a hydrological regime marked by little or no flow (LAVA site, 100 % RR). The initial cluster only shifted a few decimeters, after the passage of a few ordinary flows in Mar. 2022 and Apr. 2022. The immobility of the cobbles was confirmed by a field survey carried out in Mar. 2024. The only significant cobble movement was observed at AF3 site (67 % RR) between Feb. 25, 2017 and Apr. 25, 2022 (1,886 d). This period includes the 50-yr flood (20 m³/s; ω = 405 W/m²), during which the Lavandières brook moved its cobbles by 56 m/d over 2 d. Most of these remain concentrated near the injection site, with the rest stretching downstream towards the confluence;

• in the Souc brook, a left-bank tributary, from 25 Feb. 2017 to 25 Apr. 2022 (1886 d), the 50-yr flood of Oct. 15, 2018 (20 m³/s; ω = 351 W/m²) caused a CTD of 200 m/d over 2 d. The distribution of cobbles is relatively homogeneous in the first third of the stream up to the first ford crossing (57 % of cobbles found), due to a shallower gradient. Only 7 % of cobbles made it over the obstacle, and very few have entered the Argent Double River.

4.1.2. Clamoux watershed

20The analysis periods correspond to 2013-2022 for PIT tags and 2021-2024 for AIT tags.

Montagne Noire

21In the active channel of the Clamoux River, from Nov. 23, 2013 to Mar. 17, 2014 (115 d), the hydrological regime was dominated by ordinary high water, with Q_PF = 16.8 m³/s (< 1.88-yr flood) on Jan. 25, 2014. CTD in the ordinary channel varied from 23 m/d (IS1 site, 93 % RR) to 67 m/d (IS2 site, 85 % RR) over 1.5 d, during which ω decreased downstream, from 769 to 356 W/m², respectively. From Mar. 17, 2014 to Feb. 25, 2017 (1077 d), ordinary high water interspersed with a 20-yr flood, which Q_PF reached 92 m³/s at Laval (IS1 site) and Cabrespine (S1) and 143 m³/s at Villeneuve-Minervois (S2 site) on Feb. 14, 2017. The average CTD was 874 m/d, ranging from 1100 m/d (IS1 and S1 sites, 8 % RR) and 647 m/d (S2 site, 14 % RR) over 4 to 6 d. ω values also decrease downstream, from 1,895 W/m² (IS1 site) to 962 W/m² (S2 site). 1 cobble (b axis: 98 mm) injected in the mountain (S1 site, Cabrespine) on Feb. 15, 2016 was found in the piedmont (Les Aygadons plain) at a distance of 8571 m one year after its injection. Its displacement is primarily attributable to the 83 h (i.e., not quite 3.5 d) of the Feb. 2017 flood (Berthelot and Tourade, 2017), giving it a velocity of 104 m/h (i.e., 2.9 cm/s), which indicates the extent of cobble movement during a 20-yr flood. From Feb. 25, 2017 to Apr. 25, 2022 (1,886 d), ordinary high water interspersed with a 50-yr flood, whose Q_PF of 109 m³/s (IS1 site; ω = 2,139 W/m²) at Laval and 112 m³/s at Cabrespine (IS2 site; ω = 1,187 W/m²) was reached on Oct. 15, 2018. CTD dropped to 8 m/d (5 % RR) over 11.5 d. The Cabrespine flood-expansion zone (IS2 site; cross-section: 20 m) appears to be a sediment storage area, as evidenced by 16 cobbles found in Apr. 2022 (fig. 8B). Downstream, at Villeneuve-Minervois (CIM site), from Mar. 2, 2021 to Nov. 16, 2021 (260 d), the hydrological regime was dominated by a few ordinary flows (1.5 m³/s; ω = 59 W/m²). No movement of cobbles was observed (100 % RR). From Nov. 16, 2021 to Mar. 17, 2022 (122 d), ordinary high water caused the Q_PF to rise to 20 m³/s (< 2-yr flood) on Jan. 11, 2022 (ω = 523 W/m²). CTD was 100 m/d for a whole day (26 % RR), with the cobbles scattered along the main channel without reaching the restored river section of the Les Aygadons plain. From Mar. 17 to Apr. 15, 2022 (30 d), Q_PF did not exceed 14 m³/s (1.5-yr flood; ω = 412 W/m²). CTD in the ordinary channel was 162 m/d over 8.5 d (83 % RR), with 10 % of cobbles entering in the restored river section of the Les Aygadons plain, and 32 % downstream of it. From Apr. 15 to Oct. 12, 2022 (181 d), under similar hydrologic conditions (Q_PF < 14 m³/s; ω = 412 W/m²), the CTD in the ordinary channel was 199 m/d over 2 d (52 % RR).

22The alluvial contribution of tributaries to the Clamoux River is as follows (fig. 9A-C):

• in the Serremijeanne brook, a left-bank tributary, from Feb. 15, 2016 to Feb. 25, 2017 (377 d), the hydrological regime experienced significant runoff for 1 d due to the heavy rains of Feb. 14, 2017. At the P1 site, flow reached 20 m³/s (ω = 2,731 W/m²) and CTD was 245 m/d over 1 d only (20 % RR). From Feb. 25, 2017 to Apr. 25, 2022 (1,916 d), the hydrological regime included 48 hours of significant runoff (30 m³/s; ω = 2,466-2,551 W/m²) due to the heavy rains of Oct. 14-15, 2018. CTD was 27 m/d over 2 d (32 % RR). 55 % of cobbles have not moved since their injection. Only 3 cobbles reached the Clamoux River over this period, despite the Feb. 2017 and Oct. 2018 flood events;

• in the Ganganel/Cros brook, a right-bank tributary, from Feb. 25, 2017 to Apr. 25, 2022 (1,916 d), the hydrological regime experienced 48 hours of significant runoff (25 m³/s; ω = 2,590 W/m²) due to the heavy rains of Oct. 14-15, 2018. CTD was 5 m/d over 2 d (AF2 site, 51 % RR). 91 % of cobbles are still present in the Ganganel/Cros brook in the Cabrespine crossing, and 9 % have reached the Clamoux River, despite the Oct. 2018 flood event;

• in the Pémol brook, a right-bank tributary, from Feb. 15, 2016 to Feb. 25, 2017 (377 d), the hydrological regime saw 1 d of dynamic flow due to the heavy rains of Feb. 14, 2017. At the P2 and P3 sites, flow reached 14 m³/s (ω = 989 W/m²) and 16 m³/s (ω = 848 W/m²), respectively. The average CTD was 30 m/d over 1 d only (RR 21 %). From Feb. 25, 2017 to Apr. 25, 2022 (1,916 d), the hydrological regime included 1.5 days of significant runoff (16 m³/s; ω = 1,224 W/m²) linked to the heavy rains of Oct. 14-15, 2018. CTD was 8 m/d over 1.5 d (P2 site, 31 % RR). Most of the cobbles were deposited immediately upstream of the Pémol/Clamoux confluence, despite the Oct. 2018 flood event. Monitoring of cobbles equipped with AIT tags confirms the very low contribution of Pémol brook to the Clamoux River. No significant movement of cobbles (PEM site, 100 % RR) was observed between Mar. 3, 2021 and Mar. 28, 2024. The initial cluster only stretched by a few decimeters following the passage of ordinary flows in Mar. 2022 and Apr. 2022. The Pémol brook lacked the water or power to move its cobble bedload.

Minervois piedmont

23In the active channel of the Clamoux River:

• in the upstream part of the piedmont, from Nov. 23, 2013 to Mar. 17, 2014 (115 d), the hydrological context was dominated by ordinary high water, with Q_PF = 16.8 m³/s (1.88-yr flood; ω = 681 W/m²) on Jan. 25, 2014. CTD in the ordinary channel was 139 m/d (IS3 site, 76 % RR) for 1.5 d. From Feb. 15, 2016 to Feb. 25, 2017 (377 d), ordinary flows interspersed with a 20-yr flood, with Q_PF = 243 m³/s (ω = 1,907 W/m²) at Villeneuve-Minervois on Feb. 14, 2017. CTD was 75 m/d over 4 d (S3A site, 23 % RR). From Mar. 17, 2014 to Apr. 25, 2022 (2,962 d), the IS3 site recorded high water and two major flood events (2017 20-yr flood and 2018 50-yr flood), with Q_PF = 243 m³/s (ω = 3,426 W/m²) at Villeneuve-Minervois on Feb. 14, 2017. CTD was 31 m/d over 24.5 d (2 % RR). From Mar. 2, 2021 to Nov. 16, 2021(260 d), the hydrological regime was dominated by ordinary flows, with Q_PF = 1.5 m³/s (ω = 24 W/m²). No movement of cobbles was observed (site AYM; 90 % RR). From Nov. 16, 2021 to Mar. 17, 2022 (122 d), ordinary high water caused the Q_PF to rise to 20 m³/s (< 2-yr flood) on Mar. 13, 2022 (ω = 217 W/m²). CTD was 394 m/d for a whole day (AYM site, 50 % RR). 80 % of cobbles preferentially accumulated in the main channel at the entrance to the restored river section of the Les Aygadons plain, 10 % reached it and 10 % moved further downstream. From Mar. 17 to Apr. 15, 2022 (30 d), hydrological conditions were calm, with Q_PF < 14 m³/s (1.5-yr flood; ω = 165 W/m²). CTD in the ordinary channel was 78 m/d over 8.5 d (AYM site, 45 % RR). Only 22 % of cobbles are now in the main channel at the entrance to the restored river section, 33 % have reached it and 45 % are downstream from it. From Apr. 15 to Oct. 12, 2022 (181 d), under the same hydrological conditions (Q_PF < 14 m³/s; ω = 165 W/m²), CTD in the ordinary channel was 6.5 m/d over 2 d (AYM site, 15 % RR);

• in the restored river section (fig. 8C), from Nov. 23, 2013 to Mar. 17, 2014 (115 d), the hydrological regime was dominated by ordinary high water, with Q_PF = 16.8 m³/s (1.88-yr flood; ω = 185 W/m²) on Jan. 25, 2014. CTD in the ordinary channel was 141 m/d over 1.5 d (IS4 site, 76 % RR). From Mar. 17, 2014 to Feb. 25, 2017 (1,077 d), ordinary high water was interspersed with a 20-yr flood, with Q_PF = 243 m³/s (Villeneuve-Minervois) on Feb. 14, 2017 (ω = 1,602 W/m²). CTD is 43 m/d over 4 d (IS4 site, 11 % RR). From Feb. 25, 2017 to Apr. 25, 2022 (1886 d), ordinary high water interspersed with a 50-yr flood, with Q_PF = 169 m³/s (Villeneuve-Minervois) on Oct. 15, 2018 (ω = 928 W/m²). CTD is 29 m/d over 11.5 d (IS4 site, 8 % RR). Cobbles equipped with AIT tags (AYV site) will allow us to refine the impact of ordinary high water on bedload transport. The AYV site is represented by a transect laid across the Clamoux active channel at Les Aygadons, perpendicular to its flow axis (cross-section: 27 m). Cobbles were placed every meter both in the main channel and in the left-bank alluvial bar formed by the Feb. 2017 and Oct. 2018 floods. The top of the alluvial bar is +1.3 m above the bottom of the main channel (cross-section: 8 m). From Mar. 2, 2021 to Nov. 16, 2021(260 d), the hydrological regime was dominated by ordinary flows, with Q_PF = 1.5 m³/s (ω = 19 W/m²). No movement of cobbles was observed (90 % RR). From Nov. 16, 2021 to Mar. 17, 2022 (122 d), ordinary high water caused the Q_PF to rise to 20 m³/s (< 2-yr flood) on Mar. 13, 2022 (ω = 199 W/m²). CTD was 5 m/d for a whole day (80 % RR). Despite the passage of the 2-yr flood, most cobbles did not move in the left-bank alluvial bar, but all moved in the main channel. From Mar. 17 to Apr. 15, 2022 (30 d), hydrological conditions were calm, with Q_PF < 14 m³/s (1.5-yr flood; ω = 151 W/m²). Cobbles located in the main channel (25 %) moved slightly. CTD in the ordinary channel was 0.2 m/d over 8.5 d (80 % RR). The remaining 75 % of cobbles did not move, as they are located in the alluvial bar in the left-bank of the Clamoux River. From Apr. 15 to Oct. 12, 2022 (181 d), under the same hydrological conditions (Q_PF < 14 m³/s; ω = 165 W/m²), CTD in the ordinary channel was 6.5 m/d over 2 d (42 % RR). The minimum value (< 1 m/d) corresponds to cobbles in the alluvial bars from the 2017 and 2018 floods, while the maximum value (316 m/d) corresponds to cobbles transported in the main channel. The Les Aygadons plain appears to be an area of sustainable sediment storage, as cobbles accumulate in the riverbanks bordering the main channel.

4.2. Sediment balance (2016-2021)

24The quantified sediment balances in the rivers Argent Double and Clamoux reveal the dynamics of fluvial forms and sediments in the river sections restored by the SMMAR/SMAC.

4.2.1. Argent Double River

25The La Prade plain serves as the first flood expansion area at the exit of the mountain (fig. 10). Since 2016, when the riparian land was reacquired by the river syndicate and hydraulic structures destroyed by flooding were not rebuilt, the SMMAR has made this a section where the Argent Double River can now overflow and dissipate its energy by wandering:

• between Mar. 30, 2016 and Oct. 2, 2020 [4 yrs and 7 mos (i.e., months) or 1,648 d], the impacts of the 2017 and 2018 floods resulted in the formation of an alluvial mattress characterized by point bars and the infilling of the former 2016 channel. During this period, the sediment balance is positive, amounting to +423 m³, with sector B, identified as the central area of the restored river section, experiencing the greatest sediment accumulation (0.14 m³/m²);

• between Oct. 2, 2020 and Aug. 27, 2021 (10 mos or 330 d), the channel experienced erosion attributed to ordinary high waters and a 2-yr flood (Q_PF < 30 m³/s). In 10 mos, the sediment balance shifted once more into deficit (-3,616 m³). The upstream sectors (A and B) were notably affected (-0.16 and -0.15 m³/m², respectively);

• throughout the 5.5-year observation period (Mar. 30, 2016-Aug. 27, 2021) a discernible hydrosedimentary cycle emerges. Flood events significantly contributed to sediment accumulation, while ordinary flows promoted erosion. In a first stage, over the first 4 to 5 years, the sediment balance remains slightly positive, bolstered by the impacts of floods. However, in a second stage, we can see how rapidly (in less than 1 year) ordinary flows can deplete the sediment benefits gained from floods, with effective bedload transport occurring in the main channel during periods of ordinary high water. Consequently, while floods provide a temporary boost to the sediment balance, the overall balance remains negative throughout the hydrosedimentary cycle characterized by the interplay of ‘floods vs. ordinary flows’, resulting in a cumulative deficit of -2,576 m³. Notably, sector C, located downstream of the restored river section, appears to experience the most severe erosion, recording the greatest negative impact (-0.4 m³/m²).

4.2.2. Clamoux River

26The Les Aygadons plain serves as SMMAR’s flagship site (fig. 11). This site was the first to undergo restoration back in 2005, which involved acquiring riparian land along the river and redesigning certain sections to enable the Clamoux River to reclaim its historical wandering pattern, a shape it had lost due to centuries of enforced channelization. Indeed, the Clamoux River was embanked and meandered until the flood of 1999, and subsequent restoration work turned it into a wandering channel. The research question here is as follows. The geomorphology as much as the ecology of the riverbed depend on the influence of the 2011, 2017 and 2018 floods (do they really promote sediment accumulation?) and of the little-known influence of ordinary high water (what effects do they have on the sediment balance?):

• between Dec. 29, 2019 and Oct. 2, 2020 (9 mos or 279 d), the sediment balance appears to have reached a relatively stable equilibrium, showing a net loss of just -152 m³. This stability can be attributed to the significant sediment accumulation caused by the 2017 and 2018 floods, which occurred 2-3 years prior (fig. 11D-F) and led to the formation of alluvial bars. Notably, Sector B, which represents the middle section of the restored river, exhibited the most favorable sediment balance, measured at -0.001 m³/m²;

• between Oct. 2, 2020 and Aug. 27, 2021 (11 mos or 330 d), the impact of ordinary high water became evident, resulting in significant erosion within the main channel, especially in the concave areas of sector B (fig. 11G). The erosion effects observed in sectors A (-0.04 m³/m²) and B (-0.05 m³/m²) could not be compensated for by the dominant sediment accumulation in sector C (0.03 m³/m²). Upstream sediment accumulation is further hindered by the infilling of the former Clamoux channel (ante Oct. 2020). In the former main channel (‘omega’), the presence of a few deposits (gravelly sands) only in the upstream part of the channel show that it is disconnected most of the time for ordinary high water, bearing in mind that accumulations downstream are linked to slope failures. As a result, the overall sediment balance during this period turned negative, totaling -1,720 m³;

• in summary, between Dec. 29, 2019 and Aug. 27, 2021 (20 mos or 608 d), the overall sediment balance is negative, amounting to -1,872 m³. Sector A (-0.04 m³/m²), and particularly sector B (-0.06 m³/m²), experienced significant impacts due to substantial sediment transport into the main channel. Meanwhile, sector C (0.02 m³/m²) faced challenges in effectively collecting the sediment resulting from upstream erosion within the ordinary channel.

5. Overview and discussion

5.1. Critical analysis of data obtained from tracer cobbles

5.1.1. The key issue of return rates as a preliminary consideration

27The issue of RRs for tracer cobbles equipped with microchips (see Liébault et al., 2024b) is the initial point of discussion:

• the average RR (35 %) is relatively low for cobbles equipped with PIT tags, whose CTDs are tracked over a 10-yr period, and exhibits significant disparities (fig. 7). It is higher in the Argent Double watershed (48 %), whether in the mountain (48 %) or in the piedmont (48 %), due to a fairly good RR on tributaries (72 %), and despite a lower RR in the trunk channel (41 %). The RR is much lower in the Clamoux watershed (22 %), both in the mountain (22 %) and in the piedmont (22 %), with fairly low values in the tributaries (18 %), although this was somewhat balanced by higher values in the trunk channel (23 %). In general, the RR is better in the tributaries (43 %) than in the trunk channel (33 %), where there has been much more cobble movement. RRs tend to decrease over time, with notable declines from 82 % (Oct. 22, 2012) to 2 % (Apr. 25, 2022) in the Argent Double River, and from 83 % (Nov. 23, 2013) to 6 % (Apr. 25, 2022) in the Clamoux River;

• RRs are significantly better for cobbles equipped with AIT tags, whose CTDs were tracked over a 2-yr period. Averaging 75 %, it is slightly higher in the Argent Double watershed (79 %), due to a better RR in the trunk channel (71 %), than in the Clamoux watershed (71 %), where the RR in the trunk channel is only 62 %. In the tributaries, the RR is notably higher (100 %), in stark contrast to the trunk channel (66 %). This disparity highlights the greater mobility of cobbles within the trunk channel. Additionally, RRs tend to decline over time, dropping from 96 % (Nov. 16, 2021) to 56 % (Oct. 12, 2022).

28The significant number of cobbles that were not found, especially during the 5- to 10-year analysis period involving PIT tags, can be attributed to four main factors:

• PIT tags are at the end of their life (10 years). This is possible for PIT tags inserted in trunk-channel cobbles, but the fact that most of cobbles equipped with PIT tags were found in the tributaries tends to exclude (or minimize) this hypothesis;

• PIT tags were damaged during transport. This was made possible by the intensity of the Feb. 2017 and Oct. 2018 floods, with ω at Q_PF ranging from 1,468 W/m² (2017, Argent Double River) to 3,426 W/m² (2018, Clamoux River) in the piedmont. However, this explanation does not apply to the AIT tags as they were installed after these major floods;

• cobbles equipped with PIT tags are buried to depths exceeding 50 cm within the alluvium. This scenario is particularly plausible, considering that the active layer of cobbles mobilized during the floods is estimated to range between 0.5 m (field observations) and 2 m (depth at which the Tertiary molasse bedrock underlying the alluvium outcrops). However, this hypothesis regarding the relationship with the active layer does not account for the absence of some AIT-tagged cobbles, as AIT tags can be detected at depths of up to 2 m in sediment and/or water;

• cobbles equipped with PIT tags were evacuated downstream beyond the survey area, i.e., downstream of the river confluences (Clamoux River/Orbiel River; Argent Double River/Aude River). This is one of the most convincing hypotheses. It is supported by the CTDs measured by AIT tags. Between 2021 and 2022, under relatively calm hydrological conditions (i.e., without significant flood events), the maximum CTDs measured in the Argent Double River ranged from 3,788 to 3,935 m over a 4-mos period, with particularly notable values occurring during a few days of a 5-yr flood. In the Clamoux River, maximum CTDs were recorded between 2,560 and 2,997 m in a 1-mo period, again peaking during a brief 2-yr flood. Additionally, a maximum CTD of 2,545 m was measured over 6 mos of ordinary high water. Given these hydrological conditions, over a 10-yr period, the cobbles in the Argent Double River could travel a distance of 37 to 39 km, while in the Clamoux River, they could cover 25 to 29 km. This distance is significantly greater than the distance from the rivers’ outlets to the most upstream tracer-cobble injection sites (26 km and 22 km, respectively). When considering flood events such as those in 2017 and 2018, where CTDs could increase by a factor of 3.3 (see below), the potential for downstream transport is amplified. For instance, a cobble equipped with a PIT tag was documented to cover 8,571 m from Cabrespine to the Les Aygadons plain in just 83 h, further illustrating the feasibility of tracer cobbles (PIT tags) injected in the piedmont being evacuated from the watershed during two significant flood events. These values, therefore, represent minimum estimates (Liébault et al., 2012), supporting the plausibility of this scenario.

5.1.2. Transport distance: significant variation depending on the type of tracer used

29The previous point raises a critical hypothesis: the CTDs inferred from the analysis of cobbles equipped with PIT tags may actually underrepresent the true extent of cobble movement. This suggests that the CTDs obtained from PIT tags represent minimum values. A comparison of CTDs from cobbles equipped with PIT tags vs. those equipped with AIT tags reveals a notable discrepancy in velocity (fig. 7):

• in the trunk channels of both the rivers Argent Double and Clamoux, cobbles equipped with AIT tags exhibited average migration velocities of 87 m/d over a span of 11.5 d (from Mar. 3, 2021, to Oct. 12, 2022), which is 1.6 times greater than the average velocity of 53 m/d recorded for PIT-tagged cobbles over 8.5 d (between Oct. 22, 2012, and Feb. 10, 2017), during similar hydrological conditions (ordinary high water or less than a 2-yr flood). This difference is even more pronounced for the maximum CTDs covered by tracer cobbles, with AIT-tagged cobbles traveling an average of 3.9 times further (2,997 m between Mar. 17, 2022, and Apr. 15, 2022) compared to PIT-tagged cobbles (779 m between Mar. 15, 2014, and Feb. 10, 2017) during ordinary high water or floods of less than 2 years;

• conversely, in the tributaries, the trend is reversed, as the greatest CTDs were recorded for cobbles equipped with PIT tags (8 m/d over 4.5 d between Oct. 22, 2012, and Feb. 10, 2017). In contrast, the cobbles equipped with AIT tags showed minimal movement during the period from Mar. 3, 2021, to Oct. 12, 2022. This discrepancy can be attributed to hydrological factors, as there were measurable river flows (i.e., morphogenic events) of a few cubic meters during the period tracked by PIT tags, whereas flows were virtually absent during the period recorded for the AIT tags.

30The observed differences in CTD between passive and active transponders can be attributed to three primary factors (the potential difference in density between natural and artificial cobbles can be disregarded, as the artificial cobbles have been molded from resin that has been augmented with heavy minerals, ensuring their density closely matches that of natural cobbles):

• over the 2-yr observation period (2021-2022), which included regular field missions with multi-year intervals, there may have been improved tracking of maximum cobble displacements. In contrast, the analysis over a longer timeframe of 10 years (2012-2022) with extended observation intervals tends to smooth out maximum cobble displacements, potentially underestimating their movement.

• there is a notable difference in the size of the cobbles equipped with the respective microchips. The most common grain-size class (i.e., statistical mode) for cobbles equipped with PIT tags is 60-90 mm, whereas for those equipped with AIT tags, it is 32-45 mm (fig. 12A). This difference in sediment size may account for the faster downstream progression of cobbles equipped with AIT tags, despite both sediment size distributions show that the three granulometric classes (32-45 mm, 45-64 mm, and 64-90 mm) collectively represent a significant portion of the total sediment (between 60 % and 95 %).

• the CTDs may have also increased due to heightened energy levels within the fluvial system, stemming from changes in slope that affect the transport capacity of the rivers (fig. 13). The failure to rebuild weirs damaged by the 1999 flood and subsequent flood events (in 2011, 2017, and 2018), combined with restoration efforts designed to allow rivers to wander and reconnect with their floodplains, have led to a general increase in slope throughout the watershed.

5.2. Downstream decrease in cobble transport distances, grain size, and stream power

5.2.1. Mountain vs. piedmont

31In both watersheds, we observe a decrease in CTDs from the mountain to the piedmont (fig. 7). Over the period 2012-2022, in the Argent Double channel, cobbles equipped with PIT tags show a significant reduction in CTD (-57 %) falling from 49 between 49 m/d in the mountain to 21 m/d in the piedmont. In the Clamoux channel, the reduction in CTD is even more pronounced, decreasing by 82 % from 422 m/d in the mountain to 76 m/d in the upstream piedmont. This downstream decrease in CTD is also reflected in the maximum cobble displacement values. Over the period 2012-2022, maximum CTDs in the Argent Double channel remain relatively consistent between the mountain (1,188 m) and the piedmont (1,333 m). However, in the Clamoux channel, maximum CTDs show a dramatic decrease of 84 %, plummeting from 8,571 m in the mountain to 1,333 m in the piedmont. The downstream decrease in CTD can be attributed to the reduction in energy slope as one moves downstream.

32In the piedmonts of both watersheds, another gradient emerges. In the Argent Double channel, the average CTD of cobbles equipped with PIT tags decreases by 75 % between upstream part of the piedmont (30 m/d) and the restored river section in the La Prade plain (7 m/d), and then increases again by 77 % in the downstream part of the piedmont (30 m/d) over the period from 2012 to 2022. A similar trend is observed in terms of maximum CTDs, which decrease by 71 % between the upstream part of the piedmont (1,333 m) and the restored river section of the La Prade plain (389 m), before increasing again by 13 % towards the downstream part of the piedmont (446 m). In the Clamoux channel, the average CTD of cobbles equipped with PIT tags decreases by 15 % between the upstream part of the piedmont (82 m/d) and the restored river section of the Les Aygadons plain (71 m/d) between 2013 and 2022. Concurrently, maximum CTDs exhibit a similar trend, decreasing by 81 % between the upstream piedmont (1,333 m) and the restored river section of the Les Aygadons plain (735 m).

33The same gradient of transport-distance downstream attenuation is verified in both watersheds for cobbles equipped with AIT tags over the 2021-2022 period. In the Argent Double channel, average CTDs fall by 77 %, from 137 m/d in the upstream piedmont to 32 m/d in the restored river section of the La Prade plain. A similar trend is observed in terms of maximum CTDs, which decrease by 67 % from 3,935 m in the upstream piedmont to 1,285 m in the restored river section of the La Prade plain.

34In the Clamoux channel, CTDs vary even more significantly, with a 91 % reduction from 120 m/d in the upstream piedmont to 11 m/d in the restored river section of the Les Aygadons plain. Concurrently, maximum CTDs decrease by 75 % from 2,560 m in the upstream piedmont to 632 m in the restored river section of the Les Aygadons plain. These observations suggest that the restored river sections in the La Prade and Les Aygadons plains act as areas where cobbles are significantly slowed down. In both watersheds, the decrease in CTDs in the restored river sections confirms the effectiveness of SMMAR’s restoration efforts.

5.2.2. Transport distance vs. grain size

35CTDs are inversely proportional to their size (fig. 12B). This inverse relationship is evident from the data on average cumulative distances covered by cobbles equipped with AIT tags over the period 2021-2022 (589 d, among which 11.5 d > Q_c). CTDs decrease by 27 % between grain-size classes 32-45 mm (146 m/d) and 45-64 mm (107 m/d), then by 75 % between classes 45-64 mm and 64-90 mm (26 m/d), and finally by 14 % between classes 64-90 mm and 90-128 mm (23 m/d). In total, there is an 84 % decrease in CTDs between the finest (32-45 mm) and coarsest (128-180 mm) grain-size classes. A hydrosedimentary threshold appears between the 45-64 mm and 64-90 mm classes. For cobbles equipped with PIT tags, identifying an inter-class trend is more challenging over the periods 2012-2022 (3,473 d, among which 29 d > Q_c) in the Argent Double River and 2013-2022 (3,076 d, among 26 d > Q_c) in the Clamoux River. The only clear trend is an overall 54 % decrease in CTDs, from the finest (32-45 mm; 26 m/d) to coarsest (> 180 mm; 12 m/d) granulometric classes. Reducing CTDs in the piedmont is advantageous for the implementation of river restoration actions. However, the sensitivity of the finest particles to ordinary bedload transport complicates such efforts.

5.2.3. Transport distance vs. flow energy

36Maximum CTDs vary according to river flow energy in both watersheds (fig. 5, 7). In the Argent Double River, a 5-yr flood (Q_PF = 50 m³/s; ω = 706 W/m²) transports cobbles over distances 6.1 times greater than ordinary flows below the 2-yr flood (Q_PF = 8.1 m³/s; ω = 135 W/m²), i.e., resulting in an 84 % increase in CTDs – from 640 m during ordinary flows to 3935 m during a 5-year flood. In the Clamoux River, a 50-yr flood (Q_PF = 243 m³/s; ω = 866 W/m²) transports cobbles over distances 3.3 times greater than ordinary flows below the 2-yr flood (Q_PF = 20 m³/s; ω = 165 W/m²), leading to a 70 % increase in CTDs – from 2,560 m during ordinary flows to 8,571 m during a 50-year flood. The study thus indicates that cobbles are not only mobile during the highest flood peaks but also move during low flows, below the bankfull stage. This observation is corroborated by L. Gilet et al. (2020) using the ‘stream impulse’ method.

5.3. Very short transport times for cobble bedload

37In both watersheds, the time required to set the bedload in motion is very short (fig. 7). Over the observation period (3,476 d), cobbles moved just over 4 d/yr in the trunk channel, accounting for only 1.2 % of the time. In the tributaries, cobbles moved less than 1 d/yr, which is a mere 0.1 % of the time. These findings align with those of P. Allemand et al. (2023), who reported similarly short annual mobilization times for boulders in the gravel-bed rivers of the Guadeloupe archipelago. In the studied watersheds, this short annual mobilization times can be explained by the scarcity of morphogenic flows. For instance, in the Argent Double River, a Q_PF value of 12.1 m³/s – the critical threshold at which bedload transport begins – occurs only 9 d/yr (Michler, 2013). Conversely, during 20-50-yr floods, cobbles move 58 % of the time, and during ordinary high water and 2-5-yr floods, they move 42 % of the time. Ordinary high waters are nearly as effective as floods in triggering bedload transport. This efficacy can be attributed to (i) the rapid increase in runoff energy (ω), which can easily exceed the critical threshold of 35 W/m² during ordinary high-water events, and (ii) the shape and density of the cobbles (fig. 12C, D). In the studied rivers, nearly 50 % of the cobbles consist of shale (density = 2.2) and have a blade or disc shape, making them more susceptible to being picked up by the runoff (Cuisenier, 2020).

5.4. The relative effectiveness of restored river sections

38The SMMAR has optimized nearly every part of the watersheds where there was a wide floodplain, favoring bed widening and reconnections with floodplains to store inundation water and river sediments. Several initiatives have been undertaken to accomplish this, such as refraining from reconstructing hydraulic structures along the channel, removing some existing structures, loosening bank sediments, and purchasing land in the valley bottom through the river syndicate. These efforts aim to facilitate the revitalization of the active channel’s margins by the rivers and ensure they regain their historic lateral mobility space.

5.4.1. Evidence of shorter cobble transport distances in restored river sections

39The effectiveness of the river sections restored by the SMMAR is confirmed by the monitoring of tracer cobbles (fig. 8). In the Argent Double River, 72 % of the cobbles equipped with PIT tags that were injected in 2012 were found in 2022 in the flood expansion zones of Citou and La Prade. These areas prolong the residence time of the cobbles due to slope reduction. The 2021-2022 monitoring of cobbles equipped with AIT tags reveals that CTDs in the La Prade plain (DECH site) are more than 4 times lower (32 m/d over 11.5 d) compared to the upstream part of the piedmont (CONF site; 137 m/d over 11.5 d). Additionally, 57 % of the cobbles equipped with AIT tags were deposited in the downstream part of the La Prade restoration site, where SMMAR’s efforts have widened the wandering active channel. In the Clamoux River, the flood expansion zone (Cabrespine) and the enlargement of the wandering active channel (Les Aygadons plain) serve as long-term accumulation areas for cobbles, whether from upstream or local sites. Of the cobbles equipped with PIT tags injected between 2013-2016, 38 % were found in 2022 within the Cabrespine flood expansion zone and the restored Les Aygadons section, indicating the efficacy of these large, wandering channels in trapping cobbles and extending their residence time compared to sections more constricted by hydraulic structures. The 2021-2022 monitoring of cobbles equipped with AIT tags shows that cobble movement velocity in the restored Les Aygadons section (AYV site) is 10 to 11 times lower (11 m/d) than at the two upstream sites (CIM and AYM sites; 115 and 120 m/d, respectively). SMMAR’s actions, including widening the river cross-sections, have helped to reduce the stream power (fig. 12E). In the Argent Double River, ω values are reduced by 36 % (< 2-yr flood, CONF vs. DECH sites), 54 % (5-yr flood, CONF vs. DECH sites), and 88 % (20-50-yr flood, AR5 vs. AR6 sites) in restored river sections compared to unrestored river sections upstream. In the Clamoux River, ω reductions of 63 % (< 2-yr flood, CIM vs. AYV sites) and 73 % (20-50-yr flood, IS3 vs. IS4 sites) were measured in restored sections compared to unrestored sections upstream. These results reflect the effectiveness of the river-restoration sites, where sediment trapping forms alluvial bars and cobble residence times are extended compared to more constricted river sections (fig. 12H). The primary reason for the successful sediment trapping (alluvial bars) in the restored river sections by the SMMAR (Les Aygadons, La Prade) is the alluvial contribution from flood events (2017-2018).

5.4.2. Excessively rapid cobble migration in the main channel: Sediment storage during flood events vs. erosion from ordinary high water?

40In the rivers Argent Double and Clamoux, there is a clear tendency for cobbles to be preferentially deposited on wide, relatively shallower floodplains. In the sections currently undergoing restoration work, it has been shown that the residence time of cobbles is extended. However, when the cobbles are not retained in the alluvial bars, both the Clamoux River and, even more so, the Argent Double River evacuate them very quickly, regardless of the hydrological context (fig. 7, 12F), even though the CTDs remain proportional to ω and the duration of flood event (Michler et al., 2016). In the Clamoux channel, between Mar. 3, 2021 and Oct. 12, 2022 (589 d), within a fairly calm hydrological context (one 2-yr flood plus a few days of ordinary high water), the cobbles placed in the alluvial bars in the restored river section (AYV site) remained there. In contrast, the main Clamoux channel evacuated 100 % of those in the main channel, and 22 % of the cobbles introduced at the upstream sites (CIM and AYM sites), which are less than 2.5 km away, crossed the restored river section. Over the same period, cobbles equipped with AIT tags traveled up to 4 km (mean: 93 m/d over 11.5 d) in the main channel of the Argent Double River. This rapid movement of cobbles occurred in an ordinary hydrological context (Q_PF < 50 m³/s). The same conclusions can be drawn from the sediment balance results obtained from LiDAR image analysis. Floods are indeed the source of excess sediment (particularly in large sections of the floodplain), while ordinary flows are the source of sediment deficits in the main channel. From this, we can derive a general outline of a morphosedimentary cycle in which floods primarily serve as sediment-storing events in the floodplain. They deposit the majority of the alluvium on the margins of the active channel in the form of gravelly bars. Ordinary (erosive) flows then regularly evacuate sediment from the channel bottom and the internal front of the alluvial bars. Comparing where the cobbles equipped with PIT tags were initially placed in the two studied rivers with their current locations shows a significant increase in cobbles within the axis of the ordinary channels, rising from 23 % to 86 % (fig. 12G). At the same time, cobbles accumulated in the ordinary alluvial bars have been eroded by lateral undermining, decreasing from 51 % to 1 %. Cobbles placed in secondary flood channels were removed. Only the cobbles in the flood alluvial bars remain, dropping from 20 % at the time of implementation to 13 % in 2022. The effectiveness of the restored river sections is therefore moderate for cobbles located in the main channel. These cobbles quickly move into and then very quickly (within a few months) out of the restored, wandering active channels.

41The rapid cobble displacement in the main channel, including in restored sections, contributes to channel incision. On average, there has been a 50 cm incision in all river sections restored in recent years (fig. 13A-D). Channel incision has been accompanied by an increase in energy slope. In the Clamoux River, within the restored river section of Les Aygadons, the slope appears stable between 2019 and 2020. However, it has increased from 1.24 % in 1999 to 1.46 % in 2022 (Gonod and Thas, 2022). Similarly, in the Argent Double River’s restored river section of La Prade, the gradient has increased from 0.88 % in 2016 to 1.08 % in 2021, and further to 1.14 % in 2022. The average slope of the restored river section at Peyriac-Minervois has also increased, from 0.48 % in 2016 to 0.82 % in 2022 (Magallon and Pignot, 2022). The hypothesis proposed to explain the increase in slope (and consequently the transport capacity) is the change in channel pattern brought about by river restoration (fig. 13E). The transition from a sinuous to a wandering pattern has shortened the length of watercourses. Between 2000 and 2024, in the restored river sections, the sinuosity index has fallen from 2.24 to 2.12 in the Clamoux River, and from 1.27 to 1.21 in the Argent Double River. Additionally, the removal of lateral constraints has allowed flood flows a more direct trajectory.

5.4.3. Insufficient alluvial contribution from tributaries

42A sediment imbalance has been observed between the rivers Argent Double and Clamoux and their tributaries (fig. 7, 14B). The tributaries do not contribute sufficiently to the cobble recharge of the trunk channels into which they confluence. However, the situation differs slightly between the rivers Argent Double and Clamoux:

• Argent Double River. In the mountain, the rate of cobble movement (measured with Pit tags) is 8 times slower in the Gazel/Cros tributary (8 m/d over 4.5 d) than in the Argent Double River (67 m/d over 7.5 d) during periods of ordinary hydrology (2012-2017). Therefore, the cobble bedload contributed by the tributary is much lower than the transport capacity of the Argent Double River. During flood periods (2017-2022), the CTD of the Gazel/Cros (37 m/d over 4 d) temporarily exceeds that of the Argent Double River (21 m/d over 21.5 d), enabling the tributary to contribute to recharging the riverbed of the Argent Double River 18 % of the time. In the piedmont, only during a flood conditions (2017-2022), the rate of cobble movement in the Lavandières tributary (56 m/d over 2 d) exceeds that of the Argent Double River (28 m/d over 27.5 d) for 7 % of time. During the calm 2021-2022 hydrological period, it contributed nothing to the Argent Double River. The Souc tributary (200 m/d over 2 d) manages to exceed the cobble velocity (21 m/d over 28.5 d) of the Argent Double River during 7 % of flood events. Between Mar. 2021 and Mar. 2024, the Argent Double River received no cobble inflow (AIT tags) from its tributary, which remained virtually unchanged. Bedload transport in the Argent Double River was ensured by the reworking of cobbles within its own riverbed, particularly in the main channel. Overall, except during flood events, the contribution of cobbles from tributaries is insignificant compared to the transport capacity of the Argent Double River, resulting in incision in the ordinary channel;

• Clamoux River. During periods of ordinary hydrology, the Serremijeanne, Ganganel/Cros, and Pémol tributaries provide a low cobble bedload (measured with PIT tags), far below the transport capacity of the Clamoux River. Over 6 years (2016-2022), the tributaries injected only 7 % of their cobble bedload into the Clamoux River, which is insufficient. While the CTD in the tributaries averages 36 m/d over 2 d in the mountain, the Clamoux River moves its cobbles at much higher velocities: 422 m/d over 19 d in the mountain and 76 m/d over 19 d in the piedmont. Between Mar. 2021 and Mar. 2024, the Clamoux River received no cobbles (AIT tags) from its tributary, which remained virtually unchanged 0.9 km upstream of the Pémol/Clamoux confluence. Sediment transit in the Clamoux River is ensured by the reworking of cobbles within its own riverbed, particularly in the main channel. This explains why the main channel of the Clamoux River has experienced chronic incision over many years. The tributaries contribute too feebly to the sedimentary recharge of the Clamoux River, except during major floods.

43The insufficient alluvial contribution of tributaries raises the question of the need to reactivate sedimentary sources within the watershed to balance sedimentary dynamics between floods and ordinary high-water events.

A: riverbed adjustment time (adapted from Knighton, 1998). After 25 years (T0 = 1999 flood), but especially since the fluvial change (passage from a single, sinuous, embanked riverbed to a wandering active channel) imposed by the restoration actions, the rivers are in time T1, i.e., in the recovery time marked by the influence of floods (sediment supply) and ordinary high waters (erosive), and on the way to time T2, which will be the full adjustment to new flow and bedload transport conditions. 1: Ordinary channel; 2: Ordinary-channel boundary; 3: Alluvial bar (minor flood); 4: Alluvial bar (major flood); 5: Alluvial-bar boundary; 6: Bank erosion resulting from active-channel widening (major flood); 7: Erosive water flow: 8: Flow generating sediment input; 9: Channel embankment; 10: Riparian vegetation; B: morphosedimentary cycle determined by alternating floods (T1a) and ordinary high waters (T1b), and their impact on the sediment balance at the watershed scale, including both trunk channels and tributaries.
A : temps d’ajustement des lits de rivière (adapté de Knighton, 1998). Après 25 ans (T0 = crue de 1999), mais surtout depuis la métamorphose fluviale (passage d’un lit de rivière unique, sinueux et corseté à une bande active divagante) contrôlée par les actions de restauration, les rivières sont au temps T1, c’est-à-dire dans le temps de récupération marqué par l’influence des crues (pourvoyeuses en sédiments) et des hautes eaux ordinaires (érosives), et en route vers le temps T2, qui sera l’ajustement complet aux nouvelles conditions d’écoulement et de transport de la charge de fond. 1 : Chenal ordinaire ; 2 : Limite du chenal ordinaire ; 3 : Banc alluvial (crue mineure) ; 4 : Banc alluvial (crue majeure) ; 5 : Limite des bancs alluviaux ; 6 : Érosion des berges résultant de l’élargissement de la bande active (crue majeure) ; 7 : Écoulement générant une érosion du lit ; 8 : Écoulement générant un apport en alluvions ; 9 : Chenal corseté ; 10 : Ripisylve ; B : cycle morpho-sédimentaire déterminé par l’alternance des crues (T1a) et des hautes eaux ordinaires (T1b) et leur impact sur le bilan sédimentaire à l’échelle du bassin-versant, incluant à la fois les chenaux principaux et leurs affluents. 1 : Apport élevé en sédiments ; 2 : Faible apport en sédiments.

5.5. Keep the watershed in mind: A crucial aspect of successful riverbed restoration

44The results reveal contrasting hydromorphological responses in the restored areas, observable at the river section scale. These differences ultimately reflect the broader functionality at the watershed level.

45One of the problems highlighted in this study is the incision of the ordinary channel, caused by sediment outflows exceeding inflows. This incision results from the overly rapid transit of cobbles in the main channel and insufficient alluvial contributions from tributaries. Combined with a reduction in river length, incision also leads to steeper channel gradients. The rapid transit of sediment does not equate to the rapid transit of water. The peak flows required to mobilize the bedload are relatively low: 6 m³/s in the Argent Double River and 14 m³/s in the Clamoux River, while the ω values are already high in absolute terms, ranging from 22 to 242 W/m² for ordinary high-water events (< 1-yr flood). At this stage, these hydrological and hydraulic factors do not impact flood risk. However, during a major flood (200 < Q_PF < 400 m³/s; 285 < ω < 3,426 W/m²), flows can accelerate sharply, and the increasing slope could become problematic. Therefore, it is crucial to anticipate and understand the ongoing hydromorphological changes to better predict the future of our rivers (see Piégay et al., 2023). One of the major challenges for the future in the Minervois rivers is to monitor energy slopes. In the rivers studied, these slopes are constrained by outcrops of Tertiary molasse rock, which regularly appear in the riverbed, forming hard points that support the longitudinal profile (fig. 1D). While we are confident that today’s river slopes should not change significantly in the future, it would be advantageous to reduce them locally.

46From their past state as sinuous and laterally constrained systems to today’s wandering rivers, these waterways are undergoing rapid hydromorphological adjustments, with fluvial metamorphosis occurring in less than 10 yrs. The floodplain is perpetually at risk, as evidenced by the 60-150-yr flood of Nov. 1999, the 50-yr flood of Mar. 2011, the 20-yr flood of Feb. 2017, and the 50-yr flood of Oct. 2018. Solutions must be found on a watershed scale to address issues related to hydrosedimentary stocks and fluxes (Walling, 1983; Fryirs, 2013; Wohl et al., 2019). This requires a swift understanding and response at multiple levels within the watershed, aiming to restore sediment to the main river channels, retain it longer, and manage hydraulic energy effectively. At a local scale (site), priorities include promoting sinuosity, reactivating sediment sources, and reducing channel slopes. This involves continuous monitoring of riverbank evolution (Arnaud-Fassetta et al., 2021), channel-bank vegetation (Corenblit et al., 2024), sediment facies (Buffington and Montgomery, 1997), and sediment balances in the floodplain. The results of this study indicate that many of the restored sites currently exhibit a sediment deficit. These sites are still in their early years of operation, adjusting to new conditions, and have not experienced flood events since 2018 to deliver sediment to the active channel. The explanatory diagram is as follows (fig. 14A). The era of dammed rivers (T0) has ended, marking the state of the river before 1999. Today (T1), we are in the process of establishing a not-quite-balanced wandering pattern, characterized by sediment loss due to ordinary flows and increasing gradients. In the future (T2), the river is expected to achieve a fully balanced wandering pattern, with its proper functioning depending on the balance between flood events and ordinary hydrology. During transition from T1 to T2, sediment inputs from the upstream basin, tributaries, and banks are necessary, along with prolonged sediment residence. This can be facilitated by riparian vegetation, which divides water flows and traps sediment, thereby creating sinuosity that encourages bank erosion (i.e., sediment supply to the channel) and extends the river’s course, thus reducing gradient. Future LiDAR surveys should allow for the monitoring of these adjustments over their ‘geomorphological time’ (sensu Bravard, 1998). This will help extend some short sequences (e.g., Clamoux River, Orbiel River), analyze recently restored sites where data are currently unavailable, and better understand the rivers’ transition from T1 to T2.

47Sediment sources can be readily reactivated within watersheds, highlighting the importance of continuing to quantify bedload transport using the upstream-downstream approach of the sediment cascade:

• in the sub-basins, particularly the upstream and intermediate parts, steep slopes underlain by schist debris are prone to landslides. In the mountainous part of the Clamoux watershed near Castans, a significant landslide occurred on February 18, 2017. This landslide, measuring 250 m in length, 30 m in width, and 10 m in depth, occurred on slopes of 33 % and was estimated at 75,000 m³ (Berthelot and Tourade, 2017; fig. 15A). Most of the landslide deposits reached the riverbed, contributing locally to the river’s sediment load (fig. 15B). Another landslide occurred in the Argent Double Valley downstream of Villa Marie Andrée, but stabilization measures have been implemented, preventing it from continuing to serve as a sediment source for the river;

• tributaries have formed terraces, which are remnants of former debris flows or hyper-concentrated flows. These terraces can be reactivated by brook channels during floods if the protective walls are removed (fig. 15C). However, it is important to preserve the old RTM (Service de restauration des terrains en montagne) structures, which are heritage assets dating back to the 18^th and 19^th centuries;

• downstream, sub-basin outlets are frequently obstructed by bridges, which creates sediment disconnectivity during floods. As a result, most of the sediment from tributaries never reaches the main channel and instead accumulates as terraces (fig. 15D). To address this issue, fording or adapting bridge sections is recommended (Gurung et al., 2020);

• many tributary riverbeds are clogged with woody vegetation, making their maintenance essential (fig. 15E);

• some sub-basin outlets, such as those at Gazel and Rieussec, could be rectified to prevent downstream constriction due to both structural (possibility of rock excavation) and anthropic (road) factors (fig. 15F);

• in the piedmont, facilitating bank erosion of the trunk channels is important, as there is significant potential for cobbles in the old active-channel deposits. One approach to achieve this is by promoting channel sinuosity and removing the remaining longitudinal structures in areas without critical stakes (fig. 15G). However, weirs should not be systematically removed, despite encouragement from the WFD. They play a fundamental role in maintaining the alluvial floor in Mediterranean rivers and also facilitate overflow, thereby enhancing the connection between the channel and the flood plain. This is particularly crucial in contexts where channel incision is caused by ordinary flows.

48Ultimately, it’s not enough to merely reactivate sedimentary sources; managing cobble trapping in the riverbed is also essential. The role of natural vegetation and transverse woody obstacles should be carefully evaluated, as they can significantly influence the course of the ordinary channel (fig. 15H).

6. Conclusions

49The paper validates the three initial hypotheses:

• riverbed restoration in the Aude department leads to significant hydromorphological readjustments, enabling rivers to regain a wandering channel pattern that existed before their channelization. The fluvial metamorphosis of gravel-bed rivers can be very rapid, occurring in less than 10 years and sometimes within just 5 years. The paper highlights the crucial roles of both floods and ordinary high water in driving fluvial change. Currently, rivers are seeking a new equilibrium that needs to be understood and supported. This is essential to reconcile ‘geomorphological time’ for river adjustment with the imperatives of securing and protecting floodplains against highly morphogenic floods and devastating inundations;

• hydromorphological restoration of Mediterranean rivers involves managing hydrological extremes, not just flood events. Numerous studies have rightly emphasized the crucial role of floods in the erosion balance of watersheds, particularly for suspended loads observed at watershed outlets in the Mediterranean areas (Serrat et al., 2001; Boudet et al., 2017). However, work within the watersheds (this study) shows that floods, such as their role in riverbed grading and shape regeneration, are just as important as ordinary high water in day-to-day bedload management. In studied watersheds prone to a chronic sediment deficit, it is crucial, if not urgent, to mitigate the impact of ordinary flows, which have the energy (ω > 35 W/m²) to cause significant and regular movements of the gravelly bedload, leading to riverbed incision. This focus on ordinary sediment transport, in addition to sediment transport by floods, has prompted a reevaluation of the concept of ‘dominant discharge’ (Wolman and Miller, 1960; Carling, 1988). Adapted to the Mediterranean context, this reevaluation aims to differentiate the effectiveness of flood events from that of ordinary high water. As the paper demonstrates, ordinary flows play a significant role in the active channel as regular gravelly-bedload evacuators out of the watershed. Because sedimentary sources are not very active, it is prudent to give greater consideration to ordinary flows alongside floods in the watershed’s sediment balance. The effect of floods has been shown to be beneficial for active channels, where most of the bedload accumulates as alluvial bars during flood events. The Mediterranean region, and particularly the Aude department, regularly experiences intense rain events that cause severe floods. Climate change is likely to increase their frequency in the future (IPCC data). If taken seriously in flood-risk prevention measures within the watershed, climate change could potentially enhance the active channels by providing more frequent bedload supplies and revitalizing tributaries where ‘geomorphological work’ is compatible with ‘safety measures’;

• the paper highlights the benefits of riverbed restoration. A 5.5-yr morphosedimentary cycle is shown to transition from a state of sediment equilibrium to one of disequilibrium, following a phase of dominant alluvial filling due to floods, followed by effective bedload evacuation by ordinary high waters. Overall, we observe recurrent erosion and overly rapid transit of cobbles in the main channel during periods of ordinary hydrology, including in restoration areas. Furthermore, there is an insufficient contribution of tributaries in the sediment recharge of trunk channels, necessitating the reactivation of sediment sources. Many of the restored sites currently under study are expected to show positive sediment balance in the coming years, thanks to the role of floods, which are essential for sediment supply in the active channels studied. However, it is challenging to allow hydromorphological processes to take full effect without a basic understanding of the river system’s trajectory, both in the future and in light of the past. More than ever, there is a re-emphasis on the importance of a systemic analysis of watersheds and their hydromorphological trajectory, based on studying process-response relationships (Chorley and Kennedy, 1971; Schumm, 1977; Petts and Amoros, 1996). Our role, as managers or researchers, is to provide credible answers in this ongoing adjustment of riverbed processes, especially since we initiate restoration operations that we know will inevitably modify (and indeed should modify) the fundamental variables of the river system. In this regard, the paper shows that the ‘watershed’s systemic complexity’ is not universally transferable: each river presents its own unique challenges and solutions. The era of dominant engineering with pre-designed hydromodeling solutions is now complemented by ‘case-by-case’ approaches based on hydromorphological and ecological leverage and fieldwork.

*Corresponding author. Tel.: +33 (0)1 57 27 72 69. E-mail address: gilles.arnaud-fassetta@u-paris.fr (G. Arnaud-Fassetta)