2.1 Study area

The Elm Sike (0.33 km²) is a steep tributary of the Kelsocleuch Burn (6.5 km²) within the Bowmont Water catchment (87 km²), a sub-catchment of the River Tweed in the Scottish Borders (Figure 1; Table 2). Land use consists of rough sheep grazing with bracken (Pteridium), rushes (Juncus sp.) and Molinia grasses, the dominant vegetation types. Due to deforestation, the Bowmont catchment has been essentially treeless since 200 BC (Tipping, 1994). The lithology consists of Early Devonian Andesitic lavas that are overlain by glacial till and alluvium in the valley bottoms. Main soil types in the Elm Sike catchment are brown rankers and brown earths towards the outlet (UK Soil Observatory, accessed May 2024).

Notable floods in September 2008 (Annual Exceedance Probability, AEP of ≥1.42%), July 2009 and September 2012 (AEP of 5%–10%) resulted in considerable channel movement and damage to road infrastructure and property throughout the Bowmont valley (Addy & Wilkinson, 2016). As part of a natural flood management project (Cheviot Futures – Tweed Forum), LBs were added to the lower Elm Sike within a fenced area planted with deciduous tree species in September 2013. This area was selected as it is unproductive for farming. The treated study reach is steep and moderately confined with a 1 m high bedrock knick-point that forms a waterfall (Figure 1c). Towards the outlet, confinement and slope reduce over an alluvial fan.

2.2 Leaky barrier aims

Two aims were set for the LBs. Firstly, the main aim was to increase floodwater storage and roughness to attenuate and delay flood peaks. As there was a perception that the oversupply of sediment to downstream reaches was increasing flood risk and damaging farmland, the second aim was to trap and store sediment. No specific criteria for performance were set, and no consultants were employed for the design. Each LB consisted of two horizontal coniferous timbers securely attached by metal wire to posts embedded into the banks (Figure 1d). A lower gap (mean height: 0.33 m) between the timber underside and the streambed was made due to concerns about the passage of Atlantic salmon (Salmo salar) and brown trout (Salmo trutta; Figure 1d). On average, each structure was 2.8 m wide and 0.7 m high above the streambed, blocking 29% of the bankfull channel cross-section area (Supplementary Materials, Table S1; Figure 1c, d and e). The structures were constructed by hand in two stages. In September 2013, 10 flow restrictors (S1 to S10) were placed near the outlet of the catchment (Figure 1). In February 2014, a further six structures of the same design (S11 to S16) were added upstream, and S2 was lengthened.

The 16 structures created a total backwater storage volume of 51.68 ± 12.11 m³ with 10 structures having backwaters that intersected adjacent upstream structures (Table S1). Between September 2013 and February 2015, geomorphic changes adjacent to the structures and elsewhere were limited (Addy & Wilkinson, 2017); over January 2014 to February 2015, the reach-wide erosion and deposition volumes were 0.86 m³ and 0.45 m³, respectively. Thus, it is assumed that the geomorphic changes induced by the LBs mainly relate to the period of February 2015 to May 2018.

3.1 Hydrology

Three non-vented pressure transducers were used to monitor stream depth continuously at 5-minute intervals over the reach treated with LBs and the upstream control reach from July 2015 until June 2018 (Figure 1c). Two In-situ Rugged Troll 100 s (R3 and R2; accuracy: ±0.05% of range) and a single Van Welt DI 501 Mini Diver (R1; accuracy: ±0.5 cm) were used. Each sensor was suspended inside a PVC pipe that was attached to a post on the bank (R1 and R2) and to the upstream side of structure S3 (R3). At R3, the stage is influenced by both the backwater influence of the structure and the cumulative effects of the structures upstream. Data was downloaded from each sensor on a monthly basis and corrected to account for atmospheric pressure using a nearby barometric pressure sensor (Rugged Baro Troll 100; accuracy: ±0.05% of range). During each visit, depth was also manually measured as a check on the reliability of sensor observations. Corrections were made to ensure depth was relative to a common datum for each sensor in order to compensate for changing bed levels as a result of deposition or erosion.

3.2 Topographical surveys and geomorphic change analysis

Topographical surveys of the streambed and banks were undertaken with a Leica Geosystems 1200 differential GPS (Global Positioning System, accuracy: ± 0.02 m in plan and c. ± 0.03 m in elevation) in combination with a Leica Geosystems GPS 500 base station on an annual basis between February 2015 and May 2018. This encompassed three periods: 2015 to 2016, 2016 to 2017 and 2017 and 2018. More survey points were surveyed within active channel bed areas (>11.5 pts/m²) given the higher topographical variability compared to out-of-bank areas (>2.7 pts/m²; Supplementary Materials, Table S2). All survey points were post-processed to the same datum (Ordnance Survey British National Grid). The geometry (width, length, crest height and lower gap height) of each structure was surveyed following construction. No changes in structure geometry through movement were observed thereafter, and thus the basic structure dimensions are assumed to have remained from the beginning. In addition, the reference heights of each water level recorder were surveyed to convert depth data into Water Surface Elevations (WSEs) referenced to the British National Grid datum, and post-flood high water marks (vegetation debris) left on structures and vegetation were surveyed in the 2016 and 2017 surveys. No high-water marks were identified in the 2018 survey. Due to the time gap between floods and surveys, the flood peak inundation extents and elevations are uncertain.

ArcGIS (Version 10.5.1) was used to produce and analyse terrain models. For each survey following screening to remove dubious survey points, 0.1 m resolution Digital Elevation Models (DEMs) were produced using linear interpolation of triangular irregular network (TIN) models. This method has been shown to be a reliable for representing fluvial topography (Heritage et al., 2009). To determine morphological change and account for uncertainty in the DEMs, the Geomorphic Change Detection (GCD, Version 7; Wheaton et al., 2010) extension for ArcGIS was used to produce DEMs of difference (DoDs) at the 95% confidence interval for each period. A spatially variable rules-based Fuzzy-Inference System (FIS) that used survey point density and slope raster inputs was used to exclude dubious changes. The FIS allowed uncertainty levels to be tailored according to the variability of these inputs. DoD maps of genuine elevation change and metrics on volumetric change were then derived and validated against visual observations (Figure S1). CES (River Conveyance software | HR Wallingford) was used to produce metrics on channel cross-section area to investigate the local morphological effect of the LBs.

3.3 Hydraulic modelling

3.4 Event stage analysis

The hydrological analysis focused on time aspects of the observed stage hydrographs. Travel time between the treated and untreated reaches and changes in rising and falling stage limbs for selected flow events at each station were assessed. Studies into the changing time characteristics in response to NBS have previously been shown to be valuable and potentially circumvent the problems posed by unreliable rating curves for peak discharge characterisation (Black et al., 2021). Flow peak travel speeds (celerity) in m/s for the treated and untreated reach were then calculated by dividing the stream centreline distance measured in ArcGIS by travel time.

Analysis was undertaken in two stages. Firstly, only flow events associated with rainfall or snowmelt within a 24-hour period that resulted in a peak water depth at R1 of ≥0.1 m were selected. Thirty-six events were initially selected for analysis. This included multi-peaked events where the highest peak level was used in subsequent analysis. It also included events where peak stage timings at the downstream stations of each segment (R2 for the upper untreated segment and R1 for the treated segment) preceded the peak times for the upstream station of each segment. These observations, although likely genuine, were excluded from the subsequent analysis. Flow events during the period from 18th November 2015 to 9th March 2016 were also excluded due to a 15-minute interval. Secondly, to enable analysis of hydrograph rising and recession limb rates, the following criteria were then applied for the event hydrograph at each station: (1) rising limb start points were defined at the point at which stage began rising and (2) end of event stage was defined as values that were 125% of the event starting stage (Scheliga et al., 2019), or stage that equalled the starting event value or at the time of the next flow event (Glendell et al., 2014). The revised list consisted of 22 events for the subsequent analysis. Mann–Whitney U-tests suitable for non-normal distributed data conducted in IBM SPSS (Version 22) software were applied to test for statistical significance differences between locations and treatments.

3.5 Statistical analysis of geomorphic responses

To assess which topographical, hydraulic and structural geometry controls influenced geomorphic responses, a multivariable analysis was performed. Backwater net volumetric change and structure lower gap height change for each structure over the three years of monitoring (N = 48) were the two dependent variables considered. Twelve potential independent controls based on field and GIS measurements alongside hydraulic model outputs for each structure were considered as covariates (Table S4). All variables were tested for normality and natural log transformed where required. Ratio variables were centred and scaled. A two-stage approach was then adopted in the multivariable analysis. Firstly, a Multiple Linear Regression (MLR) was undertaken. A best subsets model selection method using Corrected Akaike Information Criterion (AICc) was used to select the most parsimonious models. Secondly, to account for the effect of repeated observations from each LB location, a mixed model analysis was then applied. Structure ID as a factorial variable was included as a random effect, and the covariates as main effects, where p < 0.05 determined by the MLR were used as the fixed effects parameters. A Restricted Maximum Likelihood (REML) method was applied. Model residuals were checked for normality using histograms and Kolmogorov–Smirnov tests. All statistical analyses were undertaken in IBM SPSS (Version 22) software.

4.1 Hydrological responses

Over the three years, high flows were more frequent during the winter months compared to the summer months (Figure 2). Compared to winter high flows, stream responses to high rainfall were often muted during the summer months, presumably due to drier antecedent conditions. Eighteen high flow events that exceeded the lower gap height and intersected with, but did not overtop, LB S3 occurred (Figure 2). This indicates LBs elsewhere were frequently engaged with flows, but despite this interaction, at the reach scale, no statistically significant reduction in flow peak travel speed was observed (Mann–Whitney U-test, p > 0.05; Figure 3). In nine events out of twenty-two, flow peak travel speeds were slower in the treated reach (e.g. Figure 4). Investigation of stage dynamics of rising and falling limbs in response to LB presence also showed a limited effect with no statistically significant difference observed between treatments (Figure 5; Mann–Whitney U-test, p > 0.05).

4.2 Model calibration and validation results

Hydraulic model calibration was achieved through adjustment of Manning's n values and individual representation of the LBs. Peak water levels were predicted to within 2 to 18.7%, model bias ranged from −18.7 to 8.9% and normalised RMSE ranged from 24 to 14% (Table 3; Figure S3). For overall model evaluation, NSE values were good (>0.70) for calibration events C2 and C3, respectively, but poor for event C1, likely due to morphological change during the event at S3, resulting in a change in stage unaccounted for in the model. Model validation produced broadly similar results to the calibration (Table 3; Figure S4). Comparison of model predictions with observed high trash marks for events C1 and C2 showed a larger difference in WSE at the reach scale of 0.16 and 0.13 m, respectively, and adequate agreement in the wetted extent (Figure 6a and b).

Considering the steep and turbulent nature of the reach, the calibrated models were acceptable for predicting the peak flood inundation extents, velocities (Figure S5) and shear stress (Figure S6) distributions for each monitoring period. It is, however, recognised that reliable hydraulic representation is potentially undermined by model equifinality; different combinations of LB and hydraulic roughness parameters could have resulted in the same predictions. However, in the absence of more observations, it was not possible to evaluate this further.

4.3 Hydraulic effects of structures

Hydraulic modelling of 50% AEP (Figure 7) and 5% AEP flow events (Figure 8), allows an assessment of the direct effects of LBs over time. For a 50% AEP flow, total inundation extent only slightly changed due to the presence of the LBs for a given year (1–2.5% increase in inundation; Figure 7a, b). Backwater velocities in the presence of structures were reduced by 1.9–4.3% (Figure 7b, c). Mean wake velocities downstream of the LBs were reduced by 4.4–4.6% and compared to backwaters, were lower by 1.2–6.5% (Figure 7c). In comparison to a larger 5% AEP flow, the presence of structures increased inundation extent again only slightly (3.4–4.6%; Figure 8a), but mean backwater velocities were reduced more by 6.3–11.4% (Figure 8b, c). Mean wake velocities were reduced by 1.6–6.8% in the presence of structures (Figure 8c) and compared to backwaters were higher by 12.6–18.1%, indicating accelerated underflow due to the constriction effect of the LBs (Figure 8c).

4.4 Morphological responses

4.4.1 Overall morphological change

Between 2015 and 2018, a net depositional response of 3.49 ± 0.36 m³ was observed (Figure 9b) following 12 geomorphically significant flow events (Table 4). Deposition responses were most marked within the backwaters of S5, S6 and S7 with erosion responses distributed sporadically throughout the reach (Figure 9a). As a result of channel incision, the lower gap height increased at eight structures but was completely lost due to channel choking at S6 and S7 (Figure 10c, d). Reflecting the reach-scale net deposition response, at 10 of the structures, backwater capacity was reduced (Figure 10a, b), and the total initial capacity (51.68 ± 12.11 m³) decreased by 2.75 m³ (5.33%) to 48.9 ± 9.85 m³.

TABLE 4. Summary of total rainfall, peak flood characteristics and geomorphically significant flows.

Period	Total rainfall (mm)a	R2 peak event water depth (m)	Peak event discharge (m3/s)b	Peak event annual exceedance probability (%)	N peaks > τc D50c	Duration > τc D50 (min)
11/02/2015–23/02/2016	1254.4	0.357 (C1, 05/01/2016)	0.45	10	5	5540
23/02/2016–08/06/2017	1251.33	0.390 (22/11/2016)	0.70	3.7	3	2655
08/06/2017–30/05/2018	934.59	0.318 (C3, 03/01/2018)	0.44	10.8	4	2370

• ^a Rainfall based on Sourhope AWS. See Figure 1b for location.
• ^b Based on HEC-RAS model predictions 5 m downstream of upstream boundary.
• ^c τ_{c D50} critical shear stress based on the Shields equation: $$$ {\tau}_c=\left({\rho}_s-\rho \right){\tau}^{\ast }{D}_{50}g $$$ where ρ_s is the sediment density (2650 kg/m³), ρ is the water density (1000 kg/m³), τ* is the Shields parameter assumed to be 0.045, a value relevant to coarse bedded channels with a mixture of sizes (Church, 2006), D₅₀ is the median bed sediment size (m) based on 2016 Wolman (1954) walk pebble count observations of the active channel and g is the gravitational acceleration (9.81 m/s²). τ_{c D50} (36.25 N/m²) equivalent to 0.21 m water depth at R2 based on cross-section averaged shear stress modelled in CES, is assumed to be constant in all years and representative of the whole reach.

4.5 Hydraulic effects of geomorphic responses

Based on hydraulic modelling, comparison between years showed changes in inundation extent were larger due to morphological changes than the direct hydraulic effect of LBs. Channel deposition most notably in the lower reach (S5-S7) and at the upstream sheep screen, between 2015 and 2016 (Figure 9a), resulted in an increase in inundation extent for the 50% AEP flow of 13.4% (Figure 7a) and smaller increase of 5.6% for the 5% AEP flow (Figure 8a). By reducing in-channel water depths, the deposition also reduced mean backwater velocities notably for the 50% AEP flow (−10%; Figure 7b) with a smaller reduction for the 5% AEP flow (−4.9%; Figure 8b). Mean velocities downstream of the LBs were reduced by 5.6% and 7.3% for 50% and 5% AEP flows, respectively, again reflecting in-channel deposition.

Over 2016 to 2017, inundation extent decreased by 20% for the 50% AEP flow (Figure 7a) but by a smaller decrease of 9% for the 5% AEP flow (Figure 8a. Backwater velocities increased by 13% for the 50% AEP flow (Figure 7a) and 10% for the 5% AEP flow (Figure 8b). Velocities downstream of the structures increased over this period by 8.4% and 8.1% for 50% and 5% AEP flows, respectively. The changes in inundation and velocities from 2016 to 2017 reflected localised channel incision throughout the reach (Figure 9a) that increased conveyance capacity. However, the wider inundation and reduced velocities persisted at S5, S6 and S7.

4.6 Controls on geomorphic responses

Mixed model analysis following initial MLR analysis to reduce the number of covariates (Table S4 and Table S5), showed that none were statistically significant for explaining backwater net volumetric change (Table 5). This indicates differences in sediment backwater deposition between locations could not be explained by the covariates. In contrast, for gap height change, the same covariates selected in the MLR (Table S6) remained statistically significant in the mixed model results, with no change in the overall results (Table 5) as the variance component for location was 0.

5.1 Hydrological and hydraulic effects of leaky barriers

The LBs had no statistically significant effect on flood peak travel speeds and hydrograph rate of stage rise or fall based on 22 selected high flow events over 3 years. Although direct observation of peak discharge attenuation was not possible, this suggests a limited flood risk mitigation impact for both frequent (i.e. >50% AEP) and larger (i.e. <50% AEP) flows. Thus, in relation to the first goal of delaying and attenuating flood peaks, the project was not successful.

Modelling showed the direct hydraulic effects of the LBs were also limited, but the magnitude of the effects is dependent on the size of the event. Across all years, the presence of LBs had a limited impact on backwater velocities, downstream velocities and inundation extents during 50% AEP events, but effects were greater during 5% AEP events due to more engagement of flow with the structures.

The magnitude of hydraulic effects generated by LBs depends on both the structure geometry and placement locations within a reach. Previous flume-based studies have highlighted the importance of the channel cross-sectional blockage ratio (Davidson & Eaton, 2013; Gippel, Finlayson, & O'Neill, 1996) and wood volume or the jam accumulation factor, C_A and the gap height (Follett & Hankin, 2022; Muhawenimana et al., 2021) for governing backwater afflux and hydrodynamic drag. Maximising channel blockage with larger volumes of wood to generate higher C_A and narrower lower gaps, create higher backwater depths relative to unobstructed channel depth, translating into reduced outflow peak magnitudes and increased time delays (Follett & Hankin, 2022). Structure spacing may have also affected the hydraulic performance of the LBs in the current study. Firstly, theshortened backwaters truncated by adjacent structures as occurred in ten cases, limited storage capacity. Secondly, the effective resistance generated by channel-spanning log jams is maximised when they are uniformly spaced to prevent backwater truncation by upstream jams (Follett & Wohl, 2024). The close proximity of ten LBs on the Elm Sike may have, in turn, further reduced the potential to attenuate and delay flood peaks.

The direction and magnitude of hydrological impacts are also dependent on the location within a catchment. At the intervention site, catchment area, channel gradient and availability of floodplain inundation zones are important factors that can determine impact at a catchment outlet (Dixon et al., 2016; Hankin et al., 2021; Quinn et al., 2022). Catchment scale hydrological modelling has shown that the discharge attenuation impacts of ELJs were both negative and positive, but only detectable at channel gradients lower than 0.005 m/m (Dixon et al., 2016). Modelling shows how backwater extent and depth in response to LBs depends on channel gradient Follett & Hankin (2022); of three sites with gradients of up to 0.021 m/m, the greatest outflow delay was observed for the site with the lowest gradient of 0.0019 m/m due to greater backwater length and storage. Nonetheless, in a Pennine stream with a higher channel gradient to the Elm Sike, where eight LBs were added, peak discharge was reduced on average by 10% for events with an AEP of ≤100% (van Leeuwen et al., 2024; Table 1). This suggests that appropriately designed and spaced LBs with lower gaps can mitigate flood peaks in steep catchments at scales less than ~1 km^2, but only smaller, frequent events.

Case studies of low gradient streams with associated floodplains suggest greater potential for storage and, in turn, discharge attenuation. Keys et al. (2018) showed using hydraulic modelling that inundation depth and area increased by 33% and 34%, respectively, due to the addition of three large wood structures in a low gradient reach (Table 1). This resulted in a decrease in maximum thalweg velocities of 10% and an 8% reduction in discharge for a flood that occurs multiple times per year. Norbury et al. (2021) showed at a catchment scale of 0.72 km², the installation of five willowed engineered log jams that crossed a floodplain created ~3,000 m³ of potential storage (Table 1) in contrast to 51.68 ± 12.11 m³ provided by the Elm Sike LBs . This, on average, reduced a range of peak flows by 27.3%, emphasising the importance of sufficient storage volume relative to catchment area to ensure effectiveness.

5.2 Geomorphic responses

Over three years, a reach scale net depositional response of 3.49 ± 0.36 m³ was observed, and the total LB backwater storage capacity decreased by 5.3% to 48.9 ± 9.85 m³. However, annual differences in geomorphic response related to the variable flood regime occurred. The greatest change occurred over 2015–2016, consistent with previous studies that show longer duration floods are associated with the greatest morphological change (Leenman et al., 2023). During this period, floods up to 10% AEP occurred in January 2016 that exceeded τꚍ_{c D50,} resulting in the most extensive and highest magnitude erosion and deposition with a small net erosional volumetric change measured. However, thereafter, morphological changes were smaller and net depositional. The decline in morphological responses over time indicates system recovery as the reach responded to a declining duration of geomorphically significant flows (cf. Milan, 2012). The changing nature of the reach gains or losses of sediment observed over annual timescales, highlights the variable morphological response in such high-energy streams that can alter the functions of LBs.

Despite the morphological changes and high flows documented, all structures remained in place over the three years, reflecting the anchoring used, which was sufficient for the size and power of the stream. LBs in energetic streams are prone to displacement and failure (Lo et al., 2022). Consideration of structure stability is important to ensure the structures function and persist as planned. It is also important for mitigating against the risks of flash flooding generated by sudden structure failure (Hankin et al., 2020).

Geomorphic responses adjacent to the LBs at the end of three years of monitoring were variable. Due to channel incision, the lower gap height increased at eight structures and at ten of the structures backwater capacity was reduced due to deposition. However, cumulative channel choking and overbank deposition leading to a progressive loss of backwater capacity that persisted over the three years, was concentrated only at S5, S6 and S7. This suggests only three of the structures were successful in addressing the second project goal of capturing and retaining sediment.

Mixed model analysis showed that backwater shear stress, channel cross-section area and channel slope were statistically significant explanatory variables that were inversely related to gap height change. Previous research has shown that scour beneath and immediately downstream of LBs is related to flow acceleration and shear stress caused by the restriction on flow (Muhawenimana et al., 2021, 2023; Spreitzer, Tunnicliffe, & Friedrich, 2021). These observations are broadly consistent with the aforementioned controls on the Elm Sike. Bed grain size, unit discharge and wood volume are also important controls on bed scour associated with channel-spanning wood structures (Schalko et al., 2019); analysis of these other potential controls may yield further insight into bed scour variability.

In contrast to lower gap height change, no topographical, hydraulic or structure geometry factors could explain backwater net volumetric responses. This was surprising as previous research has shown the importance of factors such as lower gap height (Lo et al., 2022) and degree of channel blockage (Davidson & Eaton, 2013; Gippel, Finlayson, & O'Neill, 1996) as key controls on backwater ponding and in turn morphological responses to in-channel large wood. Previously, the cascade effect of multiple LBs causing sediment starvation has been suggested to result in reduced sediment deposition at downstream structures (Lo et al., 2022). However, perhaps due to the limited blockage effect of the upper structures, this response was not observable as distance downstream was not a statistically significant covariate in either the MLR or mixed model analyses. Moreover, variable rates of bedload sediment supplied to, and generated within the reach, which cannot be accounted for and less geomorphically significant events following the highly active 2015–2016 period, may have limited the potential for significant backwater effects. This, in turn, could have limited the detection of any significant explanatory controls on backwater deposition.

Lo et al. (2022) found that the greatest and most persistent sediment storage was associated with two LBs that had lower gap heights of ≤0.3 m (Table 1; Figure 11). At one structure after two years, deposition of 2.77 m³ was observed. In comparison to the Elm Sike, the maximum storage for any structure over a year was 0.97 m³, and the threshold lower gap height for significant sediment storage was around ≤0.4 m. Both studies suggest a maximum lower gap of ~0.3–0.4 m or lower is required to capture coarse sediment, or as a minimum to ensure passage of flow.

5.3 Effects of geomorphic change on hydraulics

Hydraulic modelling showed that appreciable geomorphic change forced by the structures where it occurred had a greater indirect effect on channel hydraulics than the direct effect of the LBs. Channel choking (S5, S6 and S7) increased inundation extent and reduced velocities, most notably for 50% AEP flows, effects that persisted throughout the three years. Depending on the aims of a particular LB project, loss of backwater capacity and lower gap through deposition could be negative in terms of water storage capacity directly offered by LBs and may require remedial sediment management to restore capacity. However, loss of channel capacity may be offset by the improved floodplain hydrological connectivity (Spreitzer, Tunnicliffe, & Friedrich, 2021). Improved connectivity can result in significant hydro-ecological changes (Brummer et al., 2006) and, through temporary storage or flow deceleration over hydraulically rough overbank areas, potential attenuation of peak flows.

Changes in channel geometry underneath the structures observed also have hydraulic implications and emphasise the temporally variable nature of LB functioning. Overall mean lower gap height was reduced, and a complete loss of lower gaps occurred in certain years for structures S5, S6 and S7. Such changes in lower gap height could have implications for underflow velocities, drainage rates of temporary storage areas (Roberts et al., 2024), the magnitude of backwater rise (Muhawenimana et al., 2023) and in turn attenuation of flows. Further analysis of these factors could give further insights into the temporal variability of LB hydraulic functioning.

5.4 Implications for leaky barrier design, placement and maintenance

Whilst the findings are location-specific, they have potential implications for the implementation of LBs for flood risk mitigation and river restoration goals. Empirical evidence shows peak flow attenuation is possible in steep channel settings (van Leeuwen et al., 2024) but the limited delay responses observed, and findings from other studies (cf. Black et al., 2021; Hankin et al., 2021; Norbury et al., 2021), suggests low gradient (<0.01 m/m) headwater channels coupled with floodplains that offer expandable storage should be prioritised for LB addition. Structures should also be sized to maximise C_A and uniformly spaced with no intersection of backwaters by adjacent upstream LBs to maximise both storage and flow resistance.

In addition, the design of structures and placement in a catchment also needs to consider the peak flow of concern to target, the storage capacity required and the retention time to ensure effectiveness (Chappell & Beven, 2024). These aspects were not considered in relation to the Elm Sike LBs. Although catchment specificity needs to be considered, designing NBS features that provide 1000 m³/km² coupled with 2 hours of retention time on either side of the peak has been suggested to attenuate 1 year recurrence interval floods in managed grasslands (Chappell & Beven, 2024). Based on a synthesis of offline pond studies, a storage capacity of at least 2000 m³/km² is required to reduce peak flows by >10% for events with an AEP of up to 1% (Roberts et al., 2023). By contrast, the Elm Sike LBs delivered a maximum water storage capacity of 156.6 m³/km². LB hydraulic modelling tools (Follett & Beven, 2023; Geertsema et al., 2020) enable a priori, the optimisation of LB design and placement within a catchment.

Morphological changes initiated by LBs need to be considered during design and anticipated to inform any management actions following construction, depending on project goals. For example, by applying the lower gap threshold of ≤0.4 m, LBs could be located to capture coarse sediment to reduce or delay its transport downstream, but sediment removal may be needed to maintain full backwater capacity. Furthermore, such a threshold could be applied to improve stream-floodplain hydrological connectivity, which may be a goal of both floodplain restoration and flood management initiatives. However, more research is required to elucidate the controls on backwater deposition in relation to variations of catchment size, sediment supply regime and LB geometry.

This study showed variable changes in lower gap height that could also have implications for fish movement. Structure porosity rather than velocities has been observed in a controlled flume environment to have a greater effect on fish passage (Müller et al., 2021). Deposition resulting in smaller lower gaps could result in accelerated flows that are less passable by fish and other biota, presenting a barrier during high flow events. In such cases, management actions may be required to restore the lower gap height to ensure passage.