Coastal wetlands mitigate storm flooding and associated costs in estuaries

We first examined the overall role of vegetated saltmarshes in mitigating coastal flooding at the estuary-scale across storm scenarios. Vegetated saltmarshes reduced both the extent and depth of flooding for all estuaries and storm scenarios considered within our study. Ungrazed, vegetated marshes reduced mean terrestrial flood extent by 34.5% (SD ± 24.1), and grazed marshes by 29.1% (SD ± 20.6), compared with unvegetated mudflats (figure 2(a), supplementary table S2). While the mean relative contribution of marshes to flood reduction slightly decreased with increasing storm intensity (supplementary table S2, figure 2(a)), variability between estuaries decreased markedly. Flood water levels were also considerably reduced by vegetated marshes (figure 2(b)), with mean reductions across the storm scenarios of 43.6% (SD ± 23.9) for fully vegetated marshes, and 35.7% (SD ± 22.3) for grazed marshes, compared to unvegetated scenarios. Within estuaries, the proportion of marshes, as well as the vegetation state (vegetated, grazed, unvegetated), played a crucial role in reducing flooding, with estuaries that have a proportionally higher cover of vegetated saltmarsh mitigating both flooding extent and water depth (figure 2(d), supplementary tables S3 and S4).

Zoom In Zoom Out Reset image size

Accompanying the reduction in flood extents and depths, vegetated marshes reduced the relative economic costs from damage to residential and commercial properties, infrastructure, and agricultural land, compared with the unvegetated scenarios across all storm events (figure 2(c), table S5). However, unlike the depth and relative flood extents, vegetated marshes drove substantially greater savings in relative flood cost as storm magnitude increased. Under the 100 year return level storm events, where the potential for catastrophic flooding was higher, vegetated marshes reduced flood water depth at the terrestrial boundary leading to fewer banks and defences overtopping, mitigating resulting flooding and economic costs. Notwithstanding the general economic savings observed across estuaries, vegetation drove an increase in flood damage in a single estuary, where vegetation slowed the upstream passage of surge (supplementary figure S1) during a large (100 year) storm, enhancing localised flooding in particularly low-lying land near the estuary mouth.

Savings of damage costs driven by vegetation equated to an average saving per estuary of 37.1% ($8.4 M, SE ± $4.6 M) for single 100 year return level storms, compared to 31.6% ($3.3 M, SE ± $1.9 M) and 20.5% ($1.36 M, SE ± $0.7 M) for single 10 year and annual return-level storms respectively (supplementary table S5). Across all storm scenarios, this equated to mean annualised cost reduction of 37.8% ($2.7 M, SE ± $0.4 M) per estuary (supplementary table S6), and a mean flood protection value of $4772 (SE ± $1285) per hectare, per year. Despite inter-estuary differences in marsh value for flood mitigation, mean reductions in flooding cost compared favourably to other valuable ecosystem services (supplementary table S7).

To investigate the mechanisms underpinning marsh mitigation of coastal flooding, we first examined how marshes modify indicators of upstream storm surge propagation within sequential 1 km estuary sections. Modelled differences in channel current and water levels between the vegetated and unvegetated scenarios showed a clear pattern of greater divergence with increasing distance upstream. Cumulative drag from fringing vegetation weakened upstream surge through net reductions of surge-driven flood currents (figures 3(a)–(c), 4(c) and (d), supplementary table S8), limiting propagation of surge to inner-estuary areas. Accordingly, vegetated marshes also led to faster attenuation of storm-surge water levels with increasing proportional distance upstream (figures 3(d)–(f), supplementary table S9, illustrated in figures 4(a) and (b)). At the same time vegetated marshes amplified surge level close to the estuary mouths, as surge was less able to dissipate by moving upstream. Both phenomena—water level suppression upstream and surge amplification towards the mouth—increased with storm intensity, indicating that the relative contribution of marshes to surge mitigation upstream increases with increasing storm magnitude. In line with the reduction in surge current and water levels, vegetated marshes ultimately more strongly reduced flooding with increasing distance upstream: vegetation drove mean reductions in flood extent by 16.7% in the inner estuary areas, compared to 8% in the outer-estuary areas for large storms (figures 3(g)–(i), supplementary table S10). Grazing of the marshes appeared to have very little effect on storm surge attenuation as indicated by both water level and currents.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

We also noted that the effect of vegetation on surge propagation to up-stream areas responded to estuary size, with marsh vegetation reducing mean relative channel water levels at the limit of tidal intrusion (LTI) more in smaller estuaries (figure 5, supplementary table S11). These scale effects suggest that small estuaries benefit more from surge reduction due to the presence of marsh vegetation than larger estuaries. This effect was independent of the proportion of marsh area within estuaries, and effects on upstream surge water levels were observed for all storm scenarios.

Zoom In Zoom Out Reset image size

In addition to the estuary-scale surge attenuation, we observed vegetated marshes also drove localised decreases in flooding extent and depth, with effects particularly pronounced in wave-exposed outer-estuary areas (see figure 4 for illustration) suggesting an important dual role of wave attenuation. To begin further investigating the mechanisms of storm wave attenuation in our study estuaries, we examined how saltmarshes influence wave heights in the estuary channel. Our focal metric was wave height relative to the maximum observed within an estuary sector to control for differences in absolute wave height across estuaries and with distance upstream (see section 4).

Differences in wave heights as a function of vegetation state increased with proportional up-stream distance and storm intensity, with the effect of vegetation far stronger and more consistent for the 1 in 100 year storm scenario than either 1 in 1 or 1 in 10 year scenarios (figures 6(a)–(c), supplementary table S12). These results indicate that, compared with unvegetated scenarios, vegetated saltmarsh reduced up-stream propagation of waves through estuary channels, particularly under the largest storms. However, vegetated marshes also tended to amplify wave heights in channels towards the estuary mouths for all storm scenarios, increasing water depth over shallow estuary mouth structures and in turn allowing higher energy waves to propagate into estuaries [61, 62]. Yet, at these outer-estuary locations, the aforementioned increases in surge height (figures 3(d)–(f)), as well as these now described amplified wave heights, generally did not translate into increases in terrestrial flooding (figures 3(g)–(i)). Instead, marsh vegetation offset increases in surge and wave height within estuary channels via localised transformation of waves and surge travelling landward over the marshes, ultimately decreasing flooding.

Zoom In Zoom Out Reset image size

Indeed, examination of wave and surge transformation along transects perpendicular to the shoreline revealed vegetated marshes—particularly wide marshes characteristic of the lower-reaches of estuaries—reduced relative wave height and flood depth at the terrestrial boundary and these benefits increased with marsh width (figures 6(d)–(f)). Smaller, narrower marshes also provided local flood reduction benefits through direct attenuation of landward wave and surge but these benefits were more variable, depending on the hydrodynamic and topographical context of their location. Unlike the estuary-scale effects of vegetation on water level and currents, grazing consistently and negatively impacted on marsh ability to attenuate incoming waves (figures 6(d)–(f)) leading to greater flooding adjacent to the marshes, particularly for wide marshes. The relative impact of grazing on wave-attenuation appeared to be most pronounced for smaller storms but was also evident for the larger storms.

To investigate the role of saltmarsh vegetation in reducing storm flood risk we examine eight case study estuaries along the coast of Wales, UK (supplementary figure S5) which explicitly differed in their size, morphology, marsh extents and exposure to storm events, reflecting the inherent morphological and environmental variability of estuaries [34, 58–60, 70, 85]. The selected estuaries represented a broad range of characteristics to examine the generality of vegetation effects within estuaries: across large tidal range gradients (from mesotidal estuaries with ranges of ∼3 m, up to large macrotidal >10 m ranges), wave exposure driven by wind fetch (<100 km to >1000 km), and estuary sizes (3 km² to >100 km²) and types. While not exhaustive, the environmental and topographical variation across our case-study estuaries broadly represent the properties of many common small-to medium sized estuaries which are described in the literature [59, 60, 86]. These differences, and differences in prevailing conditions, allowed us to explore the role marshes play irrespective of environmental context, and investigate potential interactions between marsh vegetation and estuary characteristics in moderating flooding. The properties of each estuary are summarised in table 1.

For each of the case-study estuaries we used high-resolution hydrodynamic models to investigate how marsh vegetation state changes estuary hydrodynamics and resulting simulated flooding. We created online coupled Delft3D FLOW and WAVE (SWAN Cycle III 41.31) models [88], incorporating the effect of vegetation using a ridged cylinders approach [89] (Delft/SWAN-VEG) on both waves and flow. This approach has been successfully applied in recent numerical modelling studies which include vegetation [90, 91], and has been found to be more consistent across contexts than fixed Manning's n friction approaches [92]. The hydrodynamic processes within Delft3D were calculated using a 2-dimensional depth-averaged form of the unsteady shallow water equations [93] which has been extensively utilised and validated across a variety of applications and timescales [94–96], including for investigating the role of saltmarsh systems [90, 91].

Models were run on high resolution structured grids. Offshore areas were typically represented as 150 × 50 m grids, and grid sizes became progressively smaller towards the estuary mouth and the upstream areas. Grids within estuary boundaries were high resolution and uniform, with 10 × 10 m cell sizes giving good resolution to resolve hydrodynamics in up-stream river channels and marsh creeks. The model domain extended into the terrestrial zone to 3 m elevation above the height of the terrestrial-estuary boundary to characterise terrestrial flood extents and depths. Models were validated using a combination of tidal gauge (British Oceanographic Data Centre) and HOBO Depth logger (U20L) data deployed in each estuary group, and performed well against observed water levels in estuaries (see supplementary validation section—3.2.4—for additional information).

We used three different vegetation states: an unvegetated reference state, an undisturbed fully vegetated state where marsh platforms were fully populated with climax marsh communities, and a Grazed state where vegetation height was reduced to a uniform 8 cm in line with field observations. Marsh vegetation properties were specified using community weighted means of plant trait data from vegetation surveys carried out as part of the CoastWEB project [31, 90] (supplementary table S15).

We also investigated how storm magnitude may change the relationship between vegetation and flood mitigation, as previous studies have indicated that vegetation may become less effective at attenuating energy with increased water levels from surge during larger storms [47], and larger storms are more frequently associated with significant flood events [23, 97, 98]. We used three storm events with increasing magnitudes; an annually expected 1 in 1 year storm event, a 1 in 10 year storm event, and a 1 in 100 year event. These storm events were constructed and calibrated by fitting observed surge [99], wave (Centre for Environment Fisheries & Aquaculture Science), United Kingdom wave hindcast dataset), wind (Met Office, UK) and river flow data (Natural Resources Wales) using a generalised pareto distribution [100] to determine significant storm conditions corresponding to each of the return periods.

To understand the roles and potential interactions of vegetation and storms, we used a fully crossed factorial design, co-varying the vegetation state and storm magnitude over the 8 estuaries, creating 72 individual scenarios. We suspected that in estuaries, flood risk would be dictated by different processes operating at different spatial scales, and so we analysed depths, flood extents, current velocities and significant wave heights for each estuary and scenario at three different spatial grains; local (transect level), segment (1 km segments) and whole estuary levels.

Previous studies have indicated that over-marsh transformation of waves and surge is the most important pathway for reducing flooding in open-coastline systems [26, 49, 63]. To assess whether this also applies in estuarine environments we employed transect sampling to look at wave and surge transformation across individual marshes to examine the importance of marsh vegetation for preventing flooding from local wave and surge overtopping of banks and defences. Transects were created in QGIS and were mapped onto model output data at 250 m intervals, and sampling points were equally spaced at 10 m intervals from the marsh/channel edge until the terrestrial boundary. At each transect point we measured maximum water level, significant wave height and current velocity. From this data, we then examined the total reduction and proportional reduction in water level and wave height driven by the vegetation from the channel to terrestrial boundary as an indicator of the effectiveness of marshes in reducing local storm flooding.

We also suspected that vegetation may play a wider role in estuaries, creating a cumulative drag effect at the estuary scale which could reduce flood extents further upstream [69]. To investigate whether cumulative drag had a strong effect on flood potential, we decomposed the model outputs into 1 km sections along main estuary channels, and measured averaged peak current velocity, averaged peak water level, and averaged peak significant wave height within the estuary area sections, and flood extents and depths in adjacent terrestrial areas using zonal statistics. These 1 km sections extended from the estuary/coastal boundary, up until the LTI. Because of the high degree of variability in absolute area and topography within estuaries, we calculated flood extents, water levels and depth within each block as a proportion of the maximum observed flood extent, levels or depth respectively within each estuary to look at the relative role of marsh vegetation independent of estuary context. We also applied this to the distance of sections upstream, as the estuaries varied considerably in length, with distance being represented as a proportion of distance of each section upstream from the estuary mouth (0) to the LTI (1). At the whole estuary level, we quantified average peak water level, mean peak current velocity (flood and ebb) and mean peak significant wave height using zonal statistics within the boundaries of the whole estuary. Additional information on model specification is available in the supplementary materials (sections 3.1–3.2).

In addition to examining the hydrodynamic consequences of vegetation, we assessed the economic costs associated with flood events based on the extents and depths of flood waters from the hydrodynamic models. We compared the flood damages experienced in each estuary when marshes have no vegetation to those with full or grazed vegetation, and for different storm return levels (1 in 1, 1 in 10, 1 in 100 year). Our calculations aggregated flood damage estimates for residential, commercial, industrial and agricultural properties, as well as to public buildings and water and electricity utility installations using flood cost estimates for saltwater inundation damages (to building fabric, household inventory and domestic clean-up) from cost tables in the 2018 update of the multi-coloured manual (MCM) [101]. Properties in at-risk areas were identified using OS Mastermap layers [102], assigned a building type (e.g. terraced, detached, retail properties etc) and property age using a systematic visual assessment in Google StreetView©, and segmented by neighbourhood for socioeconomic status (UK Census Data [103]) to calculate economic cost values (in GBP(£)).

We also accounted for losses in agricultural output on flooded farmland, flosses from disruption to travel arising from flooded roads, and from restrictions in outdoor recreation activity resulting from the flooding of parks and countryside paths. Roads were identified using the Ordinance Survey Integrated Transport Network [104] data layer, and assigned a value for the average number of vehicles using them per hour from Department for Transport road traffic statistics data (DfT, 2020 [105]). To calculate economic costs from flooding we applied a 'diversion-value method' [101], whereby vehicles were assumed to have to divert to avoid flooding. For the sake of simplicity, we assumed that diversion to extend the journey by a distance equal to the length of road made impassable by the flood water for a period of 12 h. Travel disruption costs were then calculated by multiplying the costs per kilometre of additional travel—provided in the MCM cost tables [101]—by the number of vehicles affected during the flooding event. Estimates of flood losses arising in agriculture and outdoor recreation activity were also calculated, following established repair and disruption values in the Green Book (Central government guidance on appraisal and evaluation) [106].

We calculated absolute flood damage costs for single storms for each return-level event, as well as an annualised cost based on the net present value [101], and these values (in £GBP) were subsequently converted into $USD (at exchange rate of USD$1.36 to GBP£1; 20 December 2020). As the exposure of assets varied between estuaries, we also recalculated these absolute values as proportional reductions, comparing the cost for each scenario with the maximum observed flood cost for each estuary. This allowed us to compare the relative flood protection value of marshes across estuaries, independently of population density and asset exposure. Further details on the economic analysis are available in the supplementary materials (section 3.3)

Analysis of model outputs was conducted in the R statistical computing environment [107]. To analyse relative extent, water level and wave reduction (relative flood effects) data (proportional data) we employed mixed effects beta regression models using the glmmTMB package [108] with the estuary as a random factor to account for unquantified environmental differences in prevailing conditions between estuaries. We assessed effects of a range of different predictors on flood effects at three different scales: transect level (within marshes), up-stream zone, and whole estuary level. The proportional upstream distance and marsh width predictors were log10 transformed—at the estuary scale and marsh transect levels respectively—to account for non-linearity, and model diagnostics performed to ensure adequate model fit. Results were then visualised using the GGplot2 [109], Coefplot [110] and Visreg [111] packages.