Assessing bank erosion hazards along large rivers in the Anthropocene: a geospatial framework from the St. Lawrence fluvial system

Abstract Over the past decades, riparian land-use changes coupled to the multiplication of river infrastructures have enhanced vulnerability issues for societies and ecosystems located along large rivers. Exposure to geohazards is also changing due to the ongoing climate change, underlining the need for flexible management strategies for riparian environments. In this perspective, GIS-based mapping allows integrating a wide range of environmental data. However, such datasets are often incomplete and not homogeneous over large geographical scales, which can be problematic for the implementation of regional land-use planning strategies. Using the St. Lawrence fluvial system (SLFS) (Québec, Canada) as a case study, this article reports and describes a high-resolution approach to map position, characteristics and erosion susceptibility of natural and artificial riverbanks from a combination of field-based, remote sensing and local knowledge-derived data. This approach allowed identifying erosion-prone sites and highlighting dominant erosion processes and spatially constrain them along the SLFS. The proposed geospatial framework constitutes (1) an initial portrait of the riverscape that will allow an effective implementation of future monitoring and process-based studies; and (2) a first step in supporting land-use planning stakeholders in the selection of appropriate measures to ensure a greater resilience of riparian communities and ecosystems.


Introduction
Erosion and flood hazards constitute major concerns for riparian communities located along large river systems. Their associated risks are often significant because lowland areas bordering rivers are generally densely populated economic corridors. Deficient management strategies associated with poor scientific knowledge and nonintegrated decision-making approaches increase the exposure and vulnerability of riparian populations and sensitive ecosystems, which may result in severe damages to infrastructures, population displacement and riparian ecosystems degradation (Drejza et al. 2011;Rangel-Buitrago et al. 2018;Williams et al. 2018;Best 2019). The setup of integrated management policies at regional scales are now widely accepted and used to face erosion hazards (Kumar et al. 2018;Rangel-Buitrago et al. 2018;Williams et al. 2018).
Large rivers marked by a strong seasonality are particularly exposed to climate change, which impacts both riverbank erosion (Chassiot et al. 2020) and flooding dynamics (Chen and She 2020; Moragoda and Cohen 2020;Rondeau-Genesse 2020). Among these systems, the St. Lawrence River (SL) is one of the largest and most populated ice-affected rivers worldwide (Figure 1(A)), with nearly 50% of today's Qu ebec population concentrated in the SL lowlands (Morneau et al. 2014) (Figure  1(B)). This densely populated SL sector is mainly vulnerable to hydrometeorological hazards such as storm surges and spring floods that enhance erosion mechanisms on riverbanks. For example, recent floods in 2017 and 2019 reached unprecedented highwater levels in the upstream part of the river that caused the degradation of mitigation structures and inhabited lands, leading to the evacuation of residents (CMM 2017) and a cost of 390 million CAD in non-insured damages to the Government of Qu ebec. 1 Following these events along the SL, many questions regarding future large-scale management policies were raised in order to help municipalities to deal with erosion hazards. With ongoing climate change, the issue of sustaining or adapting geohazards management strategies has become even more crucial for ensuring the safety of inhabitants while limiting financial costs (Buffin-B elanger et al. 2015). Currently, a high percentage of the decision-making process for erosion management is based on economic considerations (Cooper and McKenna 2008;Rangel-Buitrago and Anfuso 2015), local non-integrated approaches to develop solutions ) and fears of the stakeholders (Finkl 2016). These deficient strategies usually resulted from a scarcity of regional-scale knowledge and information for large rivers, which limits objective dialogue between stakeholders (Ward et al. 2001;Petts et al. 2006;Habersack et al. 2016;Pi egay et al. 2020). In this perspective, the development of an accurate geospatial database on the state of riverbanks becomes necessary in order to establish a river-specific framework and a baseline for future quantitative studies on erosion-prone sites (Currin et al. 2015;Rogers and Woodroffe 2016;Fraser et al. 2017;Marteau et al. 2017;Best 2019). With such missing information along the St. Lawrence River fluvial system (SLFS; Figure 1(B)), there is a need to setup a largescale and qualitative mapping approach in order to support integrated management Lawrence River and location of the study area, the fluvial system. Average and maximum (in bold) tidal ranges are shown for some municipalities along the river. (C) The non-tidal influenced river section (RS) and (D) the tidal-influenced fluvial estuary section (FES) separated by the tidal limit (yellow dashed line). White arrows indicate flow directions. decisions by establishing (1) the conceptual foundations of erosion hazards; and (2) a standardized riverbank positioning and classification method adapted to this specific hydrosystem.
Using the SLFS as a case study, this article proposes an adaptable GIS-based approach developed to map riverbank positions and properties from the integration of field-based, remote sensing and local knowledge geospatial information. This stepby-step method, which can be adjusted to any large river systems, provides a detailed portrait of a riverscape through maps of high-resolution (1:600 scale) and precision (<1 m). It also allows identifying riverbank erosion hotspots and geomorphological processes at work, while defining their spatial distribution along the river course. The development of such an accurate geospatial framework constitutes (1) an initial portrait of the riverscape allowing an effective implementation of the next monitoring and process-based studies; and (2) a first step in supporting land-use planning stakeholders in the selection of appropriate measures to ensure a greater resilience and an adequate long-term protection of riparian communities and ecosystems.

Regional settings
With a total length of about 1200 km, the SL is one of the largest ice-affected rivers of the world, flowing out of Lake Ontario and draining a catchment covering about 1.3 million km 2 (Figure 1(A)). From upstream to downstream of the Qu ebec part, the SL system is divided into five major hydrographic units, which the river section (RS) and fluvial estuary section (FES) are grouped together to discern the fluvial system from the marine system downstream of Qu ebec City (Figure 1(B)). The boundary between the RS and FES is defined by the upstream limit of tidal processes at the mouth of Lake St. Pierre, in front of the city of Trois-Rivi eres (Matte et al. 2017) (Figure 1(A,D)). The study area also includes the mouths of major tributaries upstream to the first anthropogenic structure encountered (i.e. dam, bridge, etc.; Figure 1(C,D)).
The mean annual discharge in the SLFS increases gradually from Cornwall (7370 m 3 /s) to Qu ebec City (12,200 m 3 /s) (Rondeau et al. 2000;Matte et al. 2017) and flows through the St. Lawrence Platform geological province bordered to the north by the Canadian Shield and to the south by the Appalachians (Lavoie 1994) (Figure  1(B)). The modern SLFS evolves from a complex hydrosystem made of several river channels, branches, archipelagos and shallow fluvial lakes (RS) into a 1 to 5 km-wide estuary where fluvial processes formed bluffs and terraces that extend to Qu ebec City (FES). The modern physiography and surface geology of the SL is mainly inherited from the passage and retreat of the Laurentide Ice Sheet and subsequent marine invasion in the glacio-isostatically depressed basin located in the St. Lawrence Lowlands (Occhietti et al. 2011;Dalton et al. 2020), where a sequence of glacial, paraglacial and postglacial sediments was deposited and uplifted before the onset of modern river processes (Lamarche 2005;Occhietti 2007;Normandeau et al. 2017) (Figure 1(C,D)).
The climate ruling the SLFS is humid continental, with cold winters characterized by freezing temperatures and snowfall. Spring snowmelt, in some cases associated with heavy rains, favor occasional flooding in the lowlands. However, the strong seasonality of the region limits the possibility of compound flood events occurring, with most river discharge peaks in the spring and storm surges in the fall (Ward et al. 2018;Couasnon et al. 2020). The physiography of the SLFS directs these winds into a preferential northeast-southwest axis (Figure 1(C,D)). In addition, freezing temperatures during winter allow river ice to develop and eventually create ice-jams (Morse et al. 2003). The ice-jam hazards are, however, mitigated by the passage of icebreakers on the seaway since the beginning of the twentieth century (Ouellet and Baird 1978).
The SL was a main artery for Indigenous peoples and a gateway for former European settlers in Canada (Dickason 1996;Pintal et al. 2015) and is today a major waterway for global commerce, with about 4500 cargo ships per year (SLSMC 2019). Over the past 400 years, human-induced modifications of the riverscape resulted in deforestation and concreting of riverbanks, construction of ports and flow obstacles such as dams, piers, dikes, locks, barriers and rock armors for water regulation, hydroelectricity and hazards mitigation purposes (Figure 1(A)). Since the mid-twentieth century, the dredging of the riverbed created a seaway favorable for the passage of cargo ships (Morin and Côt e 2003) (Figure 1(C,D)). Meanwhile, agriculture, industrialization and urbanization in the lowlands maintained a constant pressure on the

Methodology
The high diversity of geo-ecosystems in large rivers requires the implementation of a simplified method to allow consistent and comparable results along the river course. The steps leading to the positioning and the classification of the riverbanks, the assessment of erosion hazards and finally to the identification of erosion-prone sites are illustrated by a flowchart in Figure 2 with the SLFS as an example. The approach describes in the following section derives from previous work of similar nature in the SL marine system (Drejza et al. 2011(Drejza et al. , 2019Quintin et al. 2016;Fraser et al. 2017;Sauv e et al. 2020).

Literature review
A literature review (step 1 in Figure 2) including both peer-reviewed articles and non-reviewed literature was undertaken in order to collect information on (1) current land management policies on Qu ebec's riparian environments; (2) former studies undertaken on the SLFS (Argus Groupe-Conseil 1996, Dauphin 2000Rondeau et al. 2000;Dauphin and Lehoux 2004;Morneau et al. 2014;Quintin et al. 2016); (3) methods used to classify riverbanks, assess and monitor erosion dynamics in time and space (Winterbottom and Gilvear 2000;Fortin 2010;Krapesch et al. 2011;Klemas 2013;Currin et al. 2015;Gilvear and Bryant 2016;Rogers and Woodroffe 2016;Cooper and Jackson 2019); (4) erosion hazards management approaches worldwide (Biron et al. 2014;Habersack et al. 2016;Williams et al. 2018) (step 1 in Figure 2); and (5) erosion mechanisms in temperate (Lawler et al. 1997;Grove et al. 2013;Henshaw et al. 2013) and cold (Reid 1985;Boucher et al. 2009;Kessler et al. 2013;Chassiot et al. 2020;Roland et al. 2021) environments. This first step is essential to develop an appropriate erosion hazards conceptual framework and riverbank classification that will be specific to the river physiography and environmental stress factors.

Data collection
In order to delineate riverbanks with great precision, the core of the geospatial approach needed to be based on high-resolution data such as LiDAR from topographic airborne surveys and orthomosaics of digital aerial photographs (step 2 in Figure 2). LiDAR data and orthomosaics are currently the most precise and consistent geospatial information for regional-scale assessment of erosion hazards, while now commonly accessible from governmental agencies or municipalities in inhabited area as large river lowlands (Mallet et al. 2012;Tomset and Layland 2019;Williams et al. 2020). The datasets were then compiled in a geodatabase that also contained approximately 18,000 georeferenced oblique photographs and videos from boat and helicopter surveys conducted between 2017 and 2019. Data collected during these field surveys were mainly used as a primary tool for riverbank segmentation and classification as well as to complement locations where remote sensing data could not provide quality information due to vegetation cover, flooded banks and/or poor image resolution. Boat and field surveys were also essential for documenting geomorphological processes and the state of degradation of artificial riverbanks, which were not always visible on aerial images. In addition, several participative mapping workshops were held with stakeholders to evaluate the needs  Sergy (2008). The yellow line marks the baseline for riverbank delineation, while the orange one refers to complementary lines (slope changes at the crest and the toe of the bluff). Oblique photographs taken from a helicopter showing (C) a coastline acting as the baseline on the crest of an active soft bluff and (D) an inhabited terrace at the toe of a bluff with the baseline on the top of protection structures. and gaps in erosion hazards management and to collect information from field actors (step 2 in Figure 2). The local knowledge-derived data acquired during these meetings provided valuable information for validating the riverbank classification results and for providing crucial information on the timing and intensity of erosion mechanisms at specific sites.

Riverbank positioning
In Anthropocene river systems, the delineation of a channel baseline must consider various artificial riverbanks, while at the same time respecting natural geomorphological limits relevant for monitoring erosion (step 3 in Figure 2) (Pi egay et al. 2020). This task is complicated by the fact that large hydrosystems may present several types of geomorphological boundaries that can evolve progressively along the river course ( Figure 1). The geospatial data allowed an accurate and precise (<1 m) hand delineation of the SLFS channels, a task that neither automated delineation algorithms nor satellite data (i.e. lower resolution data) can currently achieve. In some sectors, the year of acquisition between LiDAR and orthomosaics differs. In such cases, the most recent dataset was used to digitize the riverbank baseline at a scale of 1:600, which constitutes the highest resolution for riverbank mapping in the SLFS to date. Figure 3 illustrates how the hand delineation was performed using two geospatial datasets. Details of the mapping procedure according to the morphology, the influence of tidal processes and the presence or absence of artificial riverbanks are described in the following section.
The criteria used for the FES follow the method outlined in Quintin et al. (2016), Fraser et al. (2017 and Sauv e et al. (2020) for the SL marine system, with the coastline -the upper slope, if present, marking the limit of coastal processes-and the shoreline -the terrestrial limit submerged during upper high tides, above which herbaceous vegetation grows-being considered as the main geomorphological boundaries. Therefore, the coastlines correspond to all types of bluffs; the shorelines correspond to the others, namely beach terraces and tidal marshes. When a coastal segment presented both shorelines and coastlines, the choice between the two was made according to the presence or the absence of vulnerable habitats (i.e. built environment or wetlands). As a result, the first vulnerable riverbank type encountered riverward was considered as the riverbank baseline and the second one was classified as a complementary feature, which made it possible to broaden the environmental context of the area in the geospatial database (see complementary lines in Figure 3). The complementary lines were drawn to a maximum of 200 m behind the riverbank baseline.
For the RS, the bankfull discharge limit defining the minor riverbed was used to map the riverbank baseline, in agreement with previous work on Qu ebec (Boucher et al. 2009;Fortin 2010;Biron et al. 2013Biron et al. , 2014Buffin-B elanger et al. 2015;Rousseau et al. 2018) and worldwide rivers (Lastra et al. 2008;Bizzi and Lerner 2012;Fern andez et al. 2012;Schmitt et al. 2014;Heitmuller et al. 2015). Where the bankfull discharge limit was not clear or fuzzy on remote sensing data -as it is often the case for wetlands-the limit of arborescent vegetation was considered to map the baseline. In addition, the mapping of complementary lines includes, when identified, the crest and toe of bluffs located up to 200 m behind the riverbank baseline as well as limits of herbaceous fluvial marshes in front of it.
In both SLFS sections, field survey and remote sensing data allowed mapping artificial riverbanks that are in some cases located at the base of the riverbank slope. Most of these structures were designed to prevent erosion and/or flooding near built environment. In such cases, the positioning of the baseline was made on the top of artificial structures (Figure 3(D)). Consequently, the baseline shifts in some situations between the crest or the base of the riverbank slope, depending on the presence of artificialities downslope. This concept is illustrated in Figure 3 for a FES site, where the transition from 3A to 3B shows the shift of the baseline from the shoreline in 3C (human habitats settled downslope on a terrace at the toe of a bluff) to the coastline in 3D (crest of the bluff without habitats in front of it).

Segmentation and classification
Following riverbank positioning, a manual segmentation of the baseline was undertaken in order to assign riverbank physical attributes for each individual segment with homogeneous and continuous characteristics over a minimum length of 5 m (step 4 in Figure 2). A summary of SLFS-specific riverbank types is provided in Table 1, with the location of some types depending on the presence or absence of tidal processes. For instance, beach terraces have no equivalent in the RS due to more constant water levels that do not allow their formation. In order to simplify the reading of the results and the mapping, some terms were gathered to create six general classes: soft and rocky riverbanks, embankments, wetlands, beach terraces and canals (Table 1). The classification of artificial riverbanks was derived from field observations, analysis of remote sensing data and feedbacks from collaborators and local stakeholders. In total, 13 types of artificial riverbanks regrouped in six general categories were mapped according to their shape, function and composing material (Table 2).

Development of the erosion index
Riverbank classification was followed by an assessment of erosion susceptibility for each segment, based on geomorphic indicators (GI) of erosion from field reports, photos, videos, local knowledge and remote sensing information (step 5 in Figure 2). Several morphologies inherited from erosion processes were used as GI of active erosion for riverbank segments: slide scars, convex-shaped bank lines, cantilevered structures, uprooted trees on steep slopes, sediment fan or rock debris at the toe of the slope. The state of the vegetation was also taken into consideration, with bare banks or those with a degraded vegetation strip band being considered more prone to erosion.
The state of deterioration of artificial riverbanks, when present, was then reported in the attribute table of the database. This evaluation was made using essentially oblique photographs and videos collected during boat and helicopter surveys. In total, more than 20 fieldtrips together with consultations with local stakeholders and collaborators allowed collecting up-to-date information and validating interpretations made from remote sensing data. Four state of deterioration were defined: (1) good, when no visible damage was observed over 75% of the length of the structure; (2) partially good, when degradation (cracks, collapse, rust, rotting, etc.) was observed over 25% to 50% of the length of the structure; (3) partially bad, when degradation was evidenced over 50% to 75% of the structure; and (4) bad, when damages were observed over 75% of the structure ). Combining the erosion susceptibility and the state of deterioration of the artificialities allowed creating a decision tree (step 5 in Figure 2) towards a 3-level erosion index (EI) ranked from 0 (no erosion) to 2 (severe erosion). Figure 4 illustrates these levels with examples from artificial and natural riverbanks. In some rare cases, sedimentary accretion (Acc) was added on a segment when several stakeholders mentioned this trend (step 5 in Figure 2).

Identification of erosion-prone sites
Once the mapping of the EI was completed, preliminary results presented during workshops with citizens, managers and stakeholders -during which up to 100 participants were consulted-allowed discussing and adjusting mapping results as well as characterizing erosion-prone sites highlighted by the analysis. These participatory mapping workshops and periods of discussions on local knowledge provided better guidance and information to the research team on the riverscape. Combining these inputs from the workshops and the analysis of the riverbank characterization allowed establishing a list of erosion-prone sites (step 6 in Figure 2) with a minimum length of 100 m. In some cases, two continuous segments sharing common characteristics and EI were aggregated into a single site.

Constrains and limits of the study
Despite the application of a rigorous protocol based on high-resolution mapping, the limitations of the methods used for this study should not be overlooked, as the reliability of the remote sensing data depends on several factors, such as (1) their resolution and position accuracy; (2) shading, vegetation cover density and higherthan-normal water levels at the time of shooting; and (3) subjective interpretation by different users (Winterbottom and Gilvear 2000;Provencher and Dubois 2007;Marcus and Fonstad 2008;Fortin 2010;Gilvear and Bryant 2016). In addition, due to the extent of large rivers, the spatial coverage and the resolution of geospatial data can differ from one area to another. Some of regions may have benefited from more recent geospatial data with state-of-the-art techniques providing higher resolution, allowing a better assessment of erosion susceptibility. Similarly, it should be expected that the quality and the quantity of information provided by the different contributors during workshops are greater nearby populated areas. Finally, the erosion dynamics and the processes controlling it are here based on the development of a qualitative EI, whose criteria remain subjective and difficult to homogenize over such large and diverse hydrosystems as the SLFS.

Results
The key physical properties of the SLFS provided by the high-resolution mapping of 3191 km of riverbanks are summarized below. The riverbank baseline is composed of 13,120 individual segments to which have been attributed specific physical and/or anthropogenic characteristics. Maps and main statistics of riverbank classification are presented for each of the two main hydrographic units (i.e. the upstream RS and the downstream FES) and for the entire SLFS. The complete geospatial database is available in open access on the St. Lawrence Global Observatory website (see Data availability).
The riparian environment is marked by the presence of the metropolitan community of Montr eal, the most densely populated area in Qu ebec, and by the various navigation structures associated with the SL seaway. Many banks have been modified and backfilled (44%/1107 km) to allow occupation of the flood-prone lowlands along the river, mostly around the city of Montr eal ( Figure 5(B)). In nearly half of the cases, these anthropogenic modifications on the riparian slope have been completed by the installation of protection structures (48%/613 km) to prevent the erosion hazards, but several bare embankments (31%/392 km), are also present ( Figure 5(C)). Among artificial structures, unprotected embankments and rock armors appear to be more sensitive to erosion, as 32% (124 km) and 38% (137 km) of them, respectively, show an EI > 0 (Figures 5(C) and 6(A)).
Erosion is quite low in the RS since 72% (1829 km) of the segments show an EI ¼ 0 ( Figure 6(A)). Along the banks with an EI ¼ 1 (19%/481 km) or 2 (8%/199 km), two erosion hotspots can be identified: (1) the mouth of the Ottawa River and (2) the archipelagos between east of Montr eal and Lake St. Pierre (Figure 6(A)). These hotspots also correspond to the areas where the banks are mainly natural ( Figure 5(A)). These qualitative observations combined with workshops resulted in the preliminary identification of 186 erosion-prone sites (462 km).

The fluvial estuary section
The FES is characterized by far fewer islands and channels than the upstream RS. As a result, the FES represents only 21% (661 km) of the SLFS riverbanks, but a greater diversity of riverbank types can be observed within this section (Figure 7(A)). Of the 394 km (60%) of natural banks, wetlands (31%/207 km) are the most frequent. They are concentrated in the concave parts of the river course, away from urban centers and in the channel north ofÎle d'Orl eans (Figure 7(B)). Beach terraces (11%/70 km) and soft banks (14%/90 km) are evenly distributed throughout the sector, but the natural rocky banks (11%/75 km) are generally confined between Portneuf and Qu ebec City. The same stretch of the river, particularly on its south riverbank, concentrates high soft/rocky bluffs (15%/102 km; Figure 7(B)).
The FES has a high proportion of artificial segments, accounting for 40% (267 km) of the riverbank delineation, where embankments prevail over 33% (218 km) ( Figure  7(A,B)). Most of the artificialities are concentrated around the metropolitan community of Qu ebec City and Trois-Rivi eres, the two largest cities in the FES. The artificial riverbanks are mostly protected by structures (68%/180 km). Unlike the RS, only 15% (40 km) of the embankments remain unprotected (Figure 7(C)), but much fewer major constructions (i.e. dams, canals and navigation structures) are present, except around the ports of Qu ebec, Trois-Rivi eres and B ecancour (17%/42 km) (Figure 7(C)). As for the RS section, unprotected embankments and rock armors are the two most erosion-prone artificial bank types with EI > 0 on 54% (21 km) and 30% (33 km), respectively (Figures 6(B) and 7(C)). Finally, 18% (49 km) of the artificialities were built on natural banks without the need of backfilling.
Several riparian environments are exposed to erosion, as reflected by an EI ¼ 1 (23%/150 km) or 2 (17%/114 km) in the FES. Two erosion hotspots are identified: the channel north ofÎle d'Orl eans, essentially composed of marshes with soft tidal cliffs, and the Portneuf area, mainly characterized by high unconsolidated or rocky bluffs with no vegetation cover (Figure 6(B)). The latter is also characterized by the presence of some eroding beach terraces and wetlands located at the toe of anthropogenic structures (Figures 6(B) and 7(B)). The mapping and the classification, combined to information from workshops, finally, allowed identifying 65 erosion-prone sites representing 215 km of vulnerable riverbanks.

Synthesis of the St. Lawrence fluvial system
The riverbank baseline of the SLFS reaches a total length of 3191 km, of which 48% (1537 km) have been artificialized (Figure 8). Embankments are the most frequent type of riverbank (42%/1324 km), followed by wetlands (30%/948 km) and natural unconsolidated deposits (19%/618 km) inherited from the last glacial era. Embankments and natural deposits are evenly distributed along the SLFS, whereas wetlands are located around the archipelagos, Lake St. Pierre and along the north channel ofÎle d'Orl eans (Figure 8). Among the artificial riverbanks, 52% (793 km) correspond to protection structures settled on various riverbank types and 28% (431 km) to unprotected embankments (Figure 8(A)). The majority (75%/1148 km) of the artificialities are in good condition (Figure 8(A)), although unprotected embankments and rock armors (rip-raps) appear to be the most exposed to erosive processes, representing 81% (315 km) of the structures showing signs of degradation.
Among the entire SLFS, 70% (2225 km) of the riverbank baseline appear unaffected by erosion (EI ¼ 0), while 30% (944 km) are submitted to active erosion processes (Figure 8), of which 10% (313 km) are considered as severely degraded (EI ¼ 2). Natural riverbanks account for 70% (663 km) of the eroding sites (Figure 8(B)). The most sensitive riverbank types are soft banks and wetlands with 45% (278 km) and 34% (326 km) showing an EI > 0, respectively (Figure 8(B)). Finally, the combined approach of high-resolution mapping and local knowledge information allowed identifying and characterizing 251 erosion-prone sites representing a total of 677 km. Many of these sites are concentrated in four regional erosion hotspots, namely (1) the mouth of the Ottawa River; (2) the archipelagos between Montr eal and Lake St. Pierre; (3) the cliffs near Portneuf; and (4) the tidal marshes in the north channel of Ile d'Orl eans ( Figure 6).

Outlooks for future monitoring studies
Despite the qualitative nature of the results, the geospatial framework provides a regional-scale portrait of an Anthropocene large river with a consistent methodology along 3191 km of riverbanks (Figure 8). The first attempt to digitize the riverbank positions in the SLFS was undertaken by Sergy (2008), but with a too low precision delineation to allow addressing erosion hazards issues ( Figure  3(A,B)). A first large-scale and accurate mapping on GIS support is essential for producing subsequent reliable monitoring data addressing erosion and flood hazards issues, while offering an open access availability for scientists and policymakers (Lastra et al. 2008;Wheaton et al. 2009;Klemas 2013;Biron et al. 2014;Petropoulos et al. 2015;Fraser et al. 2017). Replication of this 2D approach with decadal intervals and new remote sensing data would allow precise changes to be observed and quantified. Conversely, common erosion hazards management approaches on large hydrosystem are generally based on low-resolution data such as satellite imagery or historic aerial photographs (Sergy 2008;Bryant 2016, Pi egay et al. 2020), which limits the accuracy of the monitoring studies. A second analysis of this type would thus make it possible to (1) validate the results of this study (erosion index); (2) characterize the evolution of erosion dynamics; and (3) measure the rates of retreat at the large river extent (Winterbottom and Gilvear 2000;Bizzy et al. 2016;Best 2019).
Two-dimensional approaches are usually suitable for a first large-scale analysis, but when it comes to local erosion-prone sites, process-based approaches with short-term observations are mandatory (Roland et al. 2021). In fact, high accuracy and precision as well as a short time interval (intra-annual) between each survey are often required to determine the best management solutions (Grove et al. 2013;Smeeckaert et al. 2013;Turner et al. 2016;Scarelli et al. 2017;Pi egay et al. 2020;Volpano et al. 2020). For this purpose, it is more appropriate to use easily reproducible 3D methods with high geodetic accuracy such as GNSS, RTK/PPK UAV or terrestrial LiDAR surveys (Klemas 2013;Joyal et al. 2016;Turner et al. 2016;Pi egay et al. 2020;Roland et al. 2021). For example, intra-annual topo-bathymetric surveys would isolate the impacts of episodic events from slower and more gradual retreats and provide insights on the seasonal erosion dynamics (Klemas 2013;Mandlburger et al. 2015;Roland et al. 2021). In addition, the use of monitoring stations that integrate different sensors (pressure, turbidimeter, ADCP, camera, piezometer, etc.) would allow documenting the land-river sediment continuum as well as the role of erosion mechanisms and controlling factors in erosion-prone sites (Zaggia et al. 2017;Scarpa et al. 2019;Roland et al. 2021). Conversely, short-term studies quickly become expensive and time-consuming; therefore, a preliminary identification of erosion-prone sites is important at the river scale. Dialogue between stakeholders can then be initiated to prioritize first actions to be taken at different spatial scales and to better understand land-use management issues on vulnerable sites.

Riverbank erosion in the SLFS hotspots
Riverbank erosion in large rivers results from a wide variety of terrestrial, fluvial and estuarine processes (Couper and Maddock 2001;Henshaw et al. 2013;Chassiot et al. 2020) that are enhanced by climate change and a growing human pressure (Gregory 2006;Best 2019;Goudie 2020;Pi egay et al. 2020;Wohl 2020). Mapping results combined with field surveys and local knowledge-derived data allowed identifying a first overview of those processes with a seasonal perspective along different stretches of the SLFS (Figure 9). By providing a large quantity of geospatial data, the proposed approach can also easily allow highlighting spatial distribution and intensity of geomorphic processes at work along riverbanks ( Figure 10). The conceptualization and classification of this information combined with high-resolution maps is a major step in hazards management as it can also raise awareness among decision-makers (Figures 9 and 10).
The main erosion mechanisms observed in the SLFS erosion hotspots are discussed below. These mechanisms are characterized by a geographical context combining a high anthropogenic pressure as well as seasonal and fetch-limited processes (<50 km) (Houser 2010;Nordstrom and Jackson 2012;Prahalad et al. 2015) (Figures 6, 9 and 10). The diversification and the intensification of human activities in the SLFS over the last century appear to have caused human-driven erosion processes to overcome those of natural origin; a trend that diminishes towards the lower reaches of the FES (Figure 10).

Flow regulation and flood-induced erosion in a built environment
The Moses-Saunders dam located at Cornwall plays a major role in water levels regulation and flow input in the RS, which strongly limits flood probabilities coming from Lake Ontario (Figure 1(C)). Flood-induced erosion from the main headwater is mitigated, but the involvement of tributaries to flood hazards in the SLFS remains. Most of these tributaries are regulated by run-of-the-river structures, through which fluvial parameters are preserved to some extent. However, dam management operations such as turbine regulation or gate opening can create artificial hydrographs with hydropeaking, i.e. discontinuous and artificial flow fluctuations downstream of the dam (Greimel et al. 2018;Schmutz and Moog 2018). For example, the Ottawa River floods in 2017 and 2019 led to the opening of all the gates of the Carillon dam to discharge flows reaching more than four times the annual average value during a sustained period (ORRPB 2019). The turbulent and erosive artificial flows released by the dam during these events severely impacted downstream structures and riverbanks (Figures 7(A) and 11(A)). Nevertheless, the Ottawa River Regulation Planning Board (ORRPB) has stated that water levels at the mouth of the Ottawa River would have been nearly one meter higher and much more destructive if the river watershed had not been regulated during these floods (ORRPB 2019).
Field surveys conducted shortly after these high-water levels allowed reporting that more artificial riverbanks had an EI > 0 compared to natural riverbanks, which appeared to be more resilient to shear stresses with gentler slopes and vegetation cover. Many residences and infrastructures were damaged by the turbulent waters, indicating that the protection structures were inefficient to prevent flood-induced erosion by overflows (Figures 9 and 11(A)). Flood hazards may persist despite the presence of water management and protection structures in the SLFS, particularly at the mouth of the Ottawa River, which is one of the four regional erosion hotspots identified in this study (Figures 7(A) and 11(A)). In fact, these water management structures can falsely create a sense of security for riparian communities, leading them to further develop riverbanks and increase their exposure to flood-induced erosion. Backfilling operations started at the beginning of the Nineteenth century to expand residential areas in the SLFS. These operations were done especially on riparian wetlands and lowlands of the RS, as shown by the 44% (1107 km) of the riverbank baseline now associated to embankment (Jean and L etourneau 2011) (Figure 5(B)). Encroachment on the river has since regularly been carried out, in some cases over distance of up to several tens of meters. These artificial slopes, easily recognizable on LiDAR data, are generally composed of loose material of various grain sizes, ranging from silt to cobbles. In most cases, these slopes are protected by concrete walls or rock armor that are often weakly adapted to flood hazards (Jafarnejad et al. 2017) (Figures 5(C) and 11(A)). Moreover, that non-cohesive material increases the sensitivity of the artificial riverbanks to erosion by overflow, which significantly increase the vulnerability of the built environment and thereby endangering riparian populations (Gilvear and Black 1999).

Exposition to wake action from St. Lawrence seaway
The establishment of the SL seaway in 1959 and the increased popularity of recreational boating significantly exacerbate riverbank erosion within the RS (Department of Public Works 1968;Panasuk 1987;Argus Groupe-Conseil 1991Lehoux et al 1997;Dauphin 2000;Dauphin and Lehoux 2004;Gharbi et al. 2010) (Figure 1(C,D)). Major river adjustments were made between Trois-Rivi eres and Cornwall to adapt the SL channel for the seaway (Argus Groupe-Conseil 1996; Morin and Leclerc 1998;De Koninck 2000;Morin and Côt e 2003;Morse et al. 2003). The combination of all these anthropogenic stressors strongly alters the natural flow, sediment continuum as well as riparian ecosystems of the RS.
According to the riverbank classification in the RS, 37% (298 km) of the natural banks of the archipelagos (excluding Montr eal, Laval, Bizard and Perrot islands) show active erosion (EI > 0), compared to 23% (361 km) for the southern and northern banks of the river (Figure 7(A)). This difference could be mainly related to the proximity of the islands to the seaway, their exposition to recreational boating and to a stabilization of the riverbanks, which are more densely populated than the archipelagos (Figure 11(B)). Several authors have also raised the issue of wake erosion in the multiple archipelagos between Montr eal and Lake St. Pierre (Argus Groupe-Conseil 1991Lehoux et al. 1997;Dauphin 2000;Dauphin and Lehoux 2004;Gharbi et al. 2010), where are located most of the remaining natural banks of the RS ( Figure  5(A)). The wake generated by commercial vessels operating in the seaway could impact the shoreline at a distance of up to 800 m and more severely when initiated at less than 400 m from the shore (Dauphin 2000) (Figure 11(B)). For the SLFS, approximately 18% (584 km) of the riverbanks are within this zone of wake influence (Figure 1(C,D)). For instance, 25% (50 km) of the severely degraded riverbanks (EI ¼ 2) in the RS are located within 800 m of the SL seaway. The ship wake impact is mainly concentrated between Montr eal and Sorel, where 85% of the total erosion attributable to commercial vessels between Cornwall and Qu ebec City occurs. (Dauphin 2000). As a result, riverbanks are eroding under the stress of these anthropogenic changes, threatening the built environment as well as agricultural lands and natural ecosystems. However, since riverbank erosion results from a series of geomorphic and hydrologic processes, wake action is not the sole factor influencing retreat rates, even for ships travelling less than 100 m from the banks. Desiccation of fine materials, freeze-thaw or the absence of vegetation also influence bank retreat rates in this sector (Gaskin et al. 2003).
On natural riverbanks of the RS, qualitative observations indicate that the presence of wetlands, such as swamps, herbaceous marshes and aquatic grass beds, appear to limit the impacts of water level variations and waves of anthropogenic origin ( Figure  11(B)). The classification conducted in the RS shows that riverbank types associated with fluvial marshes in front of the baseline (i.e. 9% with EI > 0) are less sensitive to erosion than other natural banks (i.e. 16%), supporting the conclusions of previous work on the role of fluvial wetlands in riparian protection and resilience (Murgatroyd and Ternan 1983;Thorne 1990;Simon and Collison 2002;Currin et al. 2015). According to Currin et al. (2015), shorelines consisting of marshes or swamp forests have lower rates of erosion than bare ones, particularly in high wave conditions. Even soft banks bordered by a particularly narrow band of vegetation have lower erosion rates than non-vegetated soft banks (Currin et al. 2015). Conversely, the establishment of artificialities in these fragile environments disrupts their ecological integrity, leading to their degradation or complete disappearance (Figure 11(B)). It is necessary to ensure that these sensitive environments are capable of withstanding high-intensity waves that would otherwise not be naturally observed without the presence of the SL seaway.

Mass-wasting on high bluffs
The FES is characterized by the presence of high (>5 m) soft and rocky bluffs with 46% showing an EI > 0 (Figures 6(B) and 7(B)). These cliffs are generally steep and mainly composed of shales and unconsolidated materials sensitive to meteorological alteration by frost and water, which make them more vulnerable to terrestrial erosion processes than other riverbank types in this area (Figure 12(A)). Roland et al. (2021) have highlighted that strong seasonality that promotes freeze/thaw cycles events is a dominant process on bluff recession in cold riparian environments. Within the FES, 83% of the unconsolidated cliff segments are partially or completely vegetated, but field surveys suggest that they can quickly become unstable (Figure 12(A)). Erosion along soft bluffs generally results from the joint action of slow and continuous mechanisms (i.e. freeze/thaw, seepage and desiccation) with rapid and occasional mechanisms such as runoff and mass-wasting processes (Joyal et al. 2016;Chassiot et al. 2020;Volpano et al. 2020;Roland et al. 2021); one striking example is the 2019 landslide that destroyed a marina in Deschaillons-sur-Saint-Laurent, nearby Portneuf (Figure 1(B)). In addition, several slide scars, some of which are now vegetated, remain visible on LiDAR data along bluff segments. This information suggests masswasting is an important process of riverbank erosion within the FES. Moreover, 9% of the riverbank segments correspond to rocky cliffs along the FES (Figure 7(B)). These cliffs consist, for the most part, of friable sedimentary rocks where freeze-thaw cycles are very active, especially on about 20% of these riverbanks where an absence of vegetation is observed. Although their annual retreat rates are usually slow, rapid mass-wasting processes such as rockfalls and skinflows already occurred in the area. However, few studies have described the lithostratigraphy of these bluffs (Besr e and Occhietti 2007; Occhietti 2007) but without addressing them in terms of geohazards.

Fetch-limited storm surges and relative sea level rise
In contrast to the RS wetlands, results show that tidal marshes are the most sensitive riverbank type of the FES, with 58% showing GI of erosion and with continuous segments up to 3.2 km with an erosive tidal cliff (Figures 6(B) and 7(B)). The majority of these segments with a high EI are located in the channel north ofÎle d'Orl eans (Figure 12(B)). In this sector, erosion results from the joint action of storm surges, anthropogenic pressure, ice foot duration, tidal cycles and relative sea level fluctuations (Forbes and Taylor 1994;Argus Groupe-Conseil 1996;Bernatchez and Dubois 2006;Drapeau 2007;Bhiry et al. 2013). The suspension of eroded fine sediments linked with tidal range maximum close to 6 m make the northern arm ofÎle d'Orl eans, especially Cap Tourmente mudflats, the most dynamic area of the FES (S erodes 1980). Located in a fetch-limited environment (<50 km), these marshes are less exposed to high-energy wind, but wind-generated waves remain the primary mechanism of tidal marsh shoreline change (Houser 2010;Nordstrom and Jackson 2012;Prahalad et al. 2015). Low-energy waves could also impact marshes, especially after a storm-induced breach in the vegetation cover (Prahalad et al. 2015). Bhiry et al. (2013) monitored tidal marshes in the FES and have also observed that the height of a marsh tidal cliff greatly influences the rate of retreat, i.e. the higher the slope, the greater the rate of retreat.
Relative sea level rise is a fundamental driver influencing changes in coastal environment (Houser 2010;Nordstrom and Jackson 2012;Prahalad et al. 2015;Vousdoukas, et al. 2020aVousdoukas, et al. , 2020b. The anticipated rise of global sea level will test the resilience of tidal marshes and other coast types. A recent study noticed a relative sea level rise slightly lower than 2 mm/year nearÎle d'Orl eans between 1990 and 2017 (Rondeau-Genesse 2020). Taking into account glacio-isostatic uplift and global sea level rise with the IPCC RCP8.5 scenario, James et al. (2014) predicted a relative sea level rise at Qu ebec City between 20 and 60 cm for 2100. A continuous supply of sediment will therefore be required to keep the coasts in their current position; otherwise, they could migrate landward (Prahalad et al. 2015). As the sediment load of the SL is already low (Milliman and Meade 1983;Rondeau et al. 2000), the addition of anthropogenic disturbances in its sedimentary regime could have negative impacts on the adaptive capacity of these systems. Moreover, the expected landward migration of the riverbank position would also be greatly conditioned by the presence of obstacles (e.g. cliffs, protection structures, etc.) that could block this riparian adjustment (Nordstrom and Jackson 2012;Bernatchez and Quintin 2016). These conditions can lead to the degradation or even the disappearance of tidal marshes (Prahalad et al. 2015;Bernatchez and Quintin 2016) and sandy beaches (Nordstrom and Jackson 2012;Vousdoukas et al. 2020aVousdoukas et al. , 2020b; two essential ecosystems for biodiversity, hazards mitigation, tourism and recreational activities.

Conclusions
The high-resolution GIS-based mapping approach reported in this article allowed identifying and spatially defining erosion hotspots and main mechanisms active in a large river that has been highly impacted by human activities. The analysis based on a high-resolution and multisource geospatial dataset covering a large geographical region coupled with field observations and local knowledge information from stakeholders allowed delineating, segmenting and classifying nearly 3200 km of riverbanks according to (1) their geomorphology; (2) the presence/absence of artificialities with their state of degradation; and (3) assessing their erosion susceptibility. These results represent a significant update from previous regional studies by providing open access GIS data to guide land management and conservation programs along the SLFS as well as a methodological guideline for other large rivers of the world.
Over the entire SLFS, riverbanks are half natural, half artificial. Embankments (42%), wetlands (30%) and soft deposits (19%) are the more represented types of riverbanks. Among artificial riverbanks, 52% are protected by structures such as rock armors or walls, 28% remain unprotected, while 16% represent various navigation structures and canals, mainly located in the upstream section around Montr eal. These structures are generally in good condition as they protect riverbanks from erosion, but 25% of them show signs of degradation. Over the entire SLFS, 70% of the riverbanks have a low erosion susceptibility, 20% show an EI ¼ 1 and 10% present an EI ¼ 2. These results allowed highlighting four regional erosion hotspots and 251 erosion-prone sites where future high-precision monitoring and process-based studies should be undertaken. Riverbank erosion in the SLFS is a result of the combined action of (1) human activities such as hydrological regulation by dams, wake action by ships nearby the seaway, land-use management, backfilling and encroachment on the river course that increase the erosion susceptibility and (2) natural processes such as storm surges, floods, river ice and mass-wasting, some of which are expected to intensify with ongoing climate change and future relative sea level rise.
GIS-based scientific knowledge easily accessible to populations and policymakers is necessary for increasing the resilience of riparian communities and ecosystems to erosion hazards, while allowing improving and implementing appropriate mitigation techniques on riverbank geo-ecosystems. The proposed methodology can be established for regional mapping projects on other large rivers in order to identify erosion-prone sectors and assess hydrogeomorphological changes in both space and time. A suitable approach should then include higher-resolution monitoring (intraand inter-annual) of the land-river continuum with topo-bathymetric surveys and hydrodynamic studies on vulnerable sites. Future projects on riverbank erosion dynamics in large rivers should be oriented at (1) providing a better understanding of the risks associated with anthropogenic processes, such as the impact of wake action, water level management and riverbank artificialization; (2) establishing an open access base of geospatial information and scientific knowledge for land-use planning choices; (3) increasing the resilience of riparian populations and ecosystems in a context of climate change; and (4) identifying sites of interest for conservation or restoration, such as artificial riverbanks with inadequate protection structures.

Acknowledgements
This project was funded by the 'Minist ere de l'Environnement et de la Lutte contre les changements climatiques (MELCC)' (Government of Qu ebec). We thank students and research assistants (Claudine Ouellet and Thibault Labarre) who collaborated actively to this project. We gratefully acknowledge people of Ouranos for their indispensable help in the organization of the workshops and their valuable advices. We also thank the 'Tables de concertation r egionales' (Regional Round Tables) for their support in this project by providing information on the dynamics of the St. Lawrence fluvial system through many helpful discussions.

Funding
Minist ere de l'Environnement de la Lutte contre les changements climatiques (MELCC)' (Government of Qu ebec) This paper summarises the result of a research project funded by the 'Minist ere de l'Environnement de la Lutte contre les changements climatiques (MELCC)' (Government of Qu ebec).