We focused on a stretch of shoreline located in the Rachel Carson National Estuarine Research Reserve along Taylor Creek in Carteret County, North Carolina ( Figure 1 ). This area was identified as a region experiencing high rates of erosion by the reserve staff. Observed erosion is likely due to longshore current velocities caused by the 0.86m tidal range (NOAA, 2025) and the small fetch and high levels of boat traffic in Taylor Creek. To reduce shoreline erosion and trap sediment, as well as protect and restore salt marsh vegetation, engineered intertidal oyster reefs were constructed using a biodegradable substrate (Oyster Catcher™) in July 2020. Oyster Catcher™ reefs are constructed by Sandbar Oyster Company using plant-fiber cloths and a mineral-based hardening agent. If the reef is not colonized by oysters, the material is expected to biodegrade (Lindquist and Cessna, 2018).

We conducted this study on three engineered oyster reefs and one control site with no oyster reef. Two of the oyster reefs measured 12m*2m*0.3m (length x width x height) and one measured 12m*2m*0.2m ( Figure 1D ). All sites, including the control site (CTRL), were 12m in length. Hereinafter, the oyster reefs with a height of 0.3m above the sediment surface will be referred to as high-relief (HR) reefs, and the oyster reef with a height of 0.2m above the sediment surface will be referred to as a low-relief (LR) reef. These two heights were chosen because they are in the range of elevations that are optimal for oyster growth in the Rachel Carson National Estuarine Research Reserve (Fodrie et al., 2014; Ridge et al., 2015). Taylor Creek is dredged for boat passage, and it was dredged during the study period in March of 2022. All three reef sites lacked low marsh but all sites including the control site had a landward high marsh consisting mostly of Juncus roemerianus (black needlerush). The control site also had a section of low marsh with Spartina alterniflora (also known as Sporobolus alterniflorus) as the dominant vegetation type.

We collected wave energy, sediment composition, oyster abundance and height, reef surface complexity, and slope data to characterize the study sites ( Table 1 , see Supplementary Material for methods). The average wave height was similar across all sites, with an average wave height of 0.92cm in 2021 (±0.55cm, p =0.08, Table 1 ). Along the marsh edge, the ratio of mud to sand was similar in 2021 and 2023 ( Supplementary Figure 1 ). The average mud percentage across sites with engineered oyster reefs was 2.95% (± 0.99%) while the average mud percentage at the control site was 6.78% (± 1.48%; Table 1 ). Oyster abundances and lengths were higher at sites 1 (HR) and 3 (HR) when compared to site 4 (LR) (p<0.01; Supplementary Figure 2 ). We calculated a steeper drop off in elevation seaward of the oyster reefs at sites 1 (HR), 2 (CTRL), and 4 (LR) 13°, 11° and 16°, respectively) while site 3 (HR) is characterized by a more gradual decline in elevation (2°) ( Table 1 ; Supplementary Figure 4 ). However, these bathymetric data were collected prior to oyster reef installation, and the authors qualitatively observed the seaward slope increase in steepness at sites 1 (HR), 2 (CTRL), and 4 (LR) after oyster reef installation. Water levels during the study period (March 2, 2022 – April 18, 2023) were slightly lower (1.3cm) than expected based on trends from the previous 10 years (March 2, 2012 – March 1, 2022).

We surveyed Ground Control Points (GCPs) using RTK-GNSS to align drone imagery with real-world XYZ coordinates. To increase accuracy, we created a survey-grade benchmark using the Trimble Spectra Precision SP80 system, which has a high-precision static post-processed accuracy of 3mm horizontally and 3.5mm vertically (Root Mean Squared Error [RMSE]) (SP80 manual, 2019). For benchmarking, we collected data continuously in 1-second intervals for a minimum of 2 hours. The ground-level GCPs consisted of orange squares with a black border and central X hammered flush to the ground with a surveyor’s nail. In 2022, we also included elevated GCPs, consisting of a rebar staked into the ground with an orange square cap placed securely on top, and in 2023, our elevated GCPs consisted of raised PVC pipes driven into the ground with blue disks screwed to the top. We placed the elevated GCPs seaward of the oyster reefs at heights comparable to the reefs to enhance georeferencing around the oyster reefs. We recorded a total of 4 elevated and 18 ground-level GCPs in 2022, and in 2023, we recorded 6 elevated and 11 ground-level GCPs ( Supplementary Table 1 ). The SP80 receiver was placed over the center of each GCP, recording 1-second intervals at 60 fixed epochs. The overall (XYZ) benchmark RMSE for each year is shown in Table 2 . To estimate total positional uncertainty, we conservatively added the benchmark RMSE (base station error), horizontal rover precision, and vertical rover precision for each survey, as shown in Table 2 . This resulted in total RTK-GNSS uncertainties of 0.021 m in 2022 and 0.022 m in 2023.

After surveying GCPs using RTK-GNSS, we conducted drone flights using a DJI Phantom 4 Pro with a 1” CMOS Red-Green-Blue sensor. Federal Aviation Administration Part 107 certified remote pilots operated all flights. Wind speeds were monitored to ensure they were below 15 mph for flight safety and to minimize motion blur in the images. Optimal flight times relative to the predicted Mean Lower Low Water (MLLW) tide were determined using the Beaufort, Taylor Creek, NC station (ID: 8656518) (NOAA) ( Supplementary Figure 5 ). Flights were scheduled near MLLW to maximize topographic exposure; however, environmental conditions could not be fully standardized due to tidal and weather constraints. Variations in predicted low tide levels and cloud cover were unavoidable, but surveys were conducted around the same time of year: March 2, 2022 and April 18, 2023 – to maintain temporal consistency. Flight plans were created using DJI Ground Station Pro with consistent settings across years: 15 m altitude, gimbal pitch angle of -60°, a front overlap ratio of 75%, side overlap ratio of 80% to reduce error (Nesbit and Hugenholtz, 2019). Each survey included two crosshatch flights, one at 30°course angle and another at 90°. We used an oblique gimbal pitch angle because when combined with a crosshatch pattern flight, image geometry is enhanced which reduces systematic error in the DEMs (Nesbit and Hugenholtz, 2019). To help minimize motion blur, photos were captured while hovering at each waypoint. Flights covered 0.26 ha at 4.7 m/s, capturing a total of 182 waypoints across both years. Drone flights were conducted for pre-construction site assessment, but these were not included due to differing global navigation systems used which led to a higher overall RMSE than that of the data included in this study.

We conducted Kruskal-Wallis tests and Dunn’s tests to determine if net volumetric changes varied significantly between site or uncertainty threshold used. We used one-way ANOVAs and Tukey’s post-hoc test to determine if EPRs were significantly different between sites or uncertainty thresholds.

Marsh edge retreat was observed across all sites ( Table 3 ). The magnitude of marsh edge retreat was not significantly different across sites. At site 1 (HR), site 2 (CTRL), site 3 (LR), and site 4 (HR), the marsh edge retreated by 0.29 ± 0.42 m/year, 0.35 ± 0.32 m/year, 0.48 ± 0.24 m/year, and 0.38 ± 0.39m/year, respectively. There was no significant difference between average marsh edge retreat values calculated using different uncertainty thresholds (no transects removed, transects within the EPR error removed (±5.60cm), and transects within the 95% CI of the EPR error removed (±10.97cm) (p = 0.9998). Across all sites, the marsh retreated an average of 0.38 ± 0.34 m/year.

Using the _min LOD uncertainty threshold (uncertainty ±4.91 cm), all sites except site 4 (LR) experienced net volumetric losses ( Table 4 ). Site 2 (CTRL) experienced the greatest average net volumetric loss (-6.91 cm/year), followed by site 1 (HR) (-5.65 cm/year). Net volumetric changes were significantly different between each site (p<0.01). Site 3 (HR) experienced the most average deposition of sediment (0.17 cm/year), but not enough to offset losses. Site 4 (LR) experienced no net volumetric change (0.00 cm/year). Site 2 (CTRL) had the greatest percent area of erosion at 98% while site 4 (LR) eroded at only 4.2% of the total area.

By observing the DoD, we can see that the greatest loss of elevation at site 1 (HR) occurred along the marsh edge (0.1-0.2cm/year), whereas the highest concentration of elevation loss at site 3 occurred on the west side of the reef (0.06-0.1 cm/year) as shown in the map ( Figure 2 ). Site 3 (HR) had some increase in elevation just landward of the reef (0.06-0.1 cm/year). Site 2 (CTRL) had a loss in elevation across the entire site (0.1-0.2 cm/year), and site 4 (LR) held a relatively constant elevation from 2022 to 2023.

Applying different levels of uncertainty to our DoD resulted in variable estimates of net volumetric change, erosion, and deposition rates at each site ( Table 5 ). Our annualized uncertainty thresholds for M C σ , LE95, _min LOD, and CI95 were ±3.79cm, ± 7.44cm, ± 4.91cm, and ±9.63cm, respectively. The estimates of net volumetric change using different uncertainty thresholds were significantly different (p<0.01). Using the most conservative uncertainty threshold (CI95) resulted in the lowest estimates of net volumetric change, erosion, and deposition. This is most notable at site 2 (CTRL) where applying the M C σ uncertainty resulted in an average erosion rate of -6.99cm³/year but applying the CI95 uncertainty resulted in an erosion rate of only -0.31 cm³/year.

The level of uncertainty used in DoD impacts how much detected change is retained or filtered out, as shown in Figure 3 . The DoD without uncertainty filtering (demonstrated using site 1 and shown in Figure 3A ) captures the most elevation change in red but includes differences within the DEM error margin, potentially overestimating volumetric change (Rumsby et al., 2008). Applying M C σ ( Figure 3B ), _min LOD ( Figure 3D ), and LE95 ( Figure 3C ) progressively removes areas with higher uncertainty. However, CI95 ( Figure 3E ) applies the strictest threshold, filtering out subtle changes. Wheaton et al. (2010) note that higher uncertainty thresholds such as CI95 result in more conservative estimates of erosion and deposition volumes. While this approach reduces false positives, it may also exclude small but ecologically meaningful changes in elevation. This could be problematic in short-term monitoring where detecting fine-scale sediment changes is needed to understand shoreline and habitat change (Duffy et al., 2018).

The marsh edge eroded at all sites regardless of reef presence. This is likely due to tidal currents caused by the 0.86m tidal range in Taylor Creek (NOAA, 2025), marsh drainage canals located between site 1 and 2 and between sites 2 and 3, and potentially due to elevation drop-off seaward of the oyster reefs ( Supplementary Figure 1 ). Although elevation has stayed consistent at site 4 (LR), the marsh edge retreated substantially, indicating that the oyster reef at this site is not protecting the marsh edge. However, it is possible that sediment losses from the eroding marsh edge are being trapped by the oyster reef, resulting in no net change in elevation at the site. The reef at site 4 was constructed at a lower height than the reefs at site 1 and 3, which could explain its ability to trap sediments more efficiently. In contrast, sites 1 and 3 displayed a net loss of elevation from 2022 to 2023. Some of the elevation losses could have been a product of initial sediment accretion and then redistribution qualitatively observed immediately post-construction of the reefs from 2020 to 2021 (R. Gittman, personal observation), however, due to logistical constraints, we were unable to quantify changes within the first year using the methods described in this study. Both sites 1 and 3 displayed elevation loss near the eastern side of the marsh edge which could be due to tidal current patterns. The elevation loss on the eastern side of site 3 could also be due to the marsh drainage canal between sites 2 and 3. However, site 1 experienced elevation loss directly landward of the oyster reef while site 3 did not. This could be due to differences in seaward bathymetry as the elevation decline seaward of site 1 is much steeper than the elevation decline at site 3 and/or due to scouring caused by interactions between the oyster reef and tidal currents. Site 1 also experienced high rates of marsh edge retreat and elevation loss along the marsh edge in the western region. This is likely due to a combination of factors including the tidal current velocities and the marsh drainage canal located just west of the marsh edge. It is also possible that the close proximity of the oyster reefs to each other influenced hydrodynamic patterns around the sites and caused erosion. It is important to note that this study lacks pre-restoration monitoring which limits our ability to conclude that marsh and sediment changes are due to reef installation. Future studies should incorporate pre-restoration monitoring to determine the influence of reef installation on erosional patterns. Our results indicate that topographic changes can occur even 2–3 years post installation of shore-protection structures. This highlights the need for long-term monitoring of shore stabilization projects to evaluate performance (see Baker and Gittman, 2024).

Our measured marsh edge retreat measurements in the control and engineered oyster reefs are similar to other values reported in North Carolina at unprotected marshes behind living shorelines. Polk and Eulie, 2018 measured marsh edge retreat at rates of -0.16 ± 0.82 m/year and -0.37 ± 0.16m/year after installation of living shorelines and -0.55 ± 0.82 m/year and -0.67 ± 0.16 at control sites in Carteret County, NC. In this study, horizontal marsh edge movement results using different error thresholds were not statistically significant from each other (p = 0.9998). To provide a more conservative estimate of marsh edge movement, we recommend filtering out transects within the 95% CI of the measure of uncertainty by converting these values to zeros. However, in regions with smaller rates of marsh edge movement, it may be beneficial to apply a less conservative uncertainty threshold to detect areas of marsh edge retreat.

Few studies have monitored elevation changes near living shorelines. However, Smith et al., 2018 monitored elevation landward and seaward of unprotected marshes and living shorelines and found that the elevation at unprotected marshes remained consistent across a 2-year period, while elevation landward of living shorelines increased. This is in contrast to our results which indicated that elevation declined at all sites except our low-relief oyster reef site. Differences in site conditions, including wave energy, alongshore currents, slope, and vegetation coverage, as well as living shoreline designs and substrates likely contribute to differences in observed shoreline changes (see Polk and Eulie, 2018).

Though time-intensive, creating DEMs and orthomosaics with sub-centimeter resolutions using drones and SfM can assist with evaluating NbS (Kumar et al., 2021). Most NbS projects are funded for short-term periods (2–5 years) (Kumar et al., 2021, O’Leary et al., 2023). As such, to accurately evaluate the success of NbS, it is necessary to detect small-scale changes that may occur over small time-scales. High-resolution products can also assist practitioners in detecting spatial patterns of elevation change during ambient conditions or after storm events. For example, DEMs and orthomosaics may reveal erosion hot spots which is useful information for implementing adaptive management. Detecting changes at a centimeter level allows managers to implement NbS before erosion reaches meter-level magnitudes. High-resolution data is also necessary for comparing the coastal protection properties of different NbS as differences in protection may be minimal but ecologically or economically important.

Despite the availability of more sophisticated error-handling techniques, DoD remains widely used due to its efficiency. More advanced methods such as Monte Carlo simulation on DEMs (Oksanen & Sarjakoski, 2005) and point clouds (Cooper et al., 2021) provide improved uncertainty handling. However, these computationally intensive methods require specialized skills, which might make them less practical for certain routine monitoring efforts. It is best practice to consider high levels of uncertainty filtering (e.g., CI95) to extend confidence in detected changes. However, current limitations in vertical accuracy (i.e., at the centimeter level) make their application challenging in short-term monitoring. In this study, the RTK-GNSS benchmarks had overall XYZ uncertainties of 0.021 and 0.022 m at each site (see Table 2 ), and this error was appropriately carried into the SfM data processing. The inherent limitations of current GNSS technology mean that uncertainty at the centimeter level persists throughout the dataset. This makes it difficult to detect sub-centimeter-scale changes over short monitoring periods with high confidence. Stricter thresholds are beneficial for long-term studies to isolate the most significant changes. However, applying these thresholds over short time scales, where volumetric change is already minimal, filtered out nearly all detectable erosion and deposition in this short-term study. Future research should evaluate whether uncertainty thresholds disproportionately filter out subtle, spatially structured changes associated with real geomorphic or ecological processes, particularly in short-term monitoring contexts.

To contextualize changes occurring around living shorelines, it is important to report site characteristics including wave energy, sediment composition, shoreline slope, oyster reef surface complexity, and oyster abundances (Bredes et al., 2024; Tweel et al., 2025; Baggett et al., 2015). Environmental wave energy can influence the ability of living shorelines to trap sediment (Palinkas et al., 2022). High wave energy may result in continued erosion of the shoreline while low energy can result in sediment trapping. However, wave energy must be high enough to transport sediment landward of the living shoreline (Davis et al., 2015; Jackson et al., 2002). Analyzing sediment grain size and shoreline slope can help to contextualize patterns of erosion and accretion (Palinkas et al., 2023; Polk and Eulie, 2018). Fine sediments often indicate low wave energy and sediment accretion while larger grain sizes often indicate high wave energy and erosion (Visher, 1967). Lastly, metrics of oyster reef complexity influence flow thereby influencing sedimentation rates (Styles, 2015). Oyster reefs can reduce incoming wave energy which allows lower rates of erosion and higher rates of sediment accretion (Wiberg et al., 2019).

In this study, we characterized sites by collecting data on recommended metrics including wave heights, sediment composition, landward and seaward shoreline slope, reef surface complexity, and oyster abundances and lengths (Bredes et al., 2024; Baggett et al., 2015). Due to the short duration of our study period, we were unable to make conclusions about how these characteristics may be influencing marsh edge movement and volumetric changes. We did not collect data on tidal current velocities. To assess the performance of shoreline stabilization features, it is critical to understand the processes influencing erosion and accretion patterns. Because of this, future living shoreline studies should monitor for longer durations and incorporate tidal current velocity to monitor seaward slopes (Baker and Gittman, 2024; Wang et al., 2023; Bredes et al., 2024).

As demonstrated in this study, drone imagery and SfM photogrammetry can provide high resolution data for monitoring the coastal protection properties of living shorelines. However, there are several challenges associated with drones and SfM techniques. For example, the tidal levels of our flights were not exactly the same across years which resulted in higher water levels in 2022. Though this area was filtered out of analyses, it meant we were unable to measure erosion or deposition in these areas (see areas with no elevation change coloring in Figure 2 ). Because our site was located in a nature preserve accessible to the public, we were unable to place permanent benchmarks to verify elevation across years at the same location. In environments less accessible by the public, fixed benchmarks (e.g., survey nails or an existing structure) can provide a valuable control for repeat elevation and vertical accuracy checks.

Overall, this study demonstrates that drones and SfM techniques are valuable tools for assessing the performance of different living shoreline structures when appropriate uncertainty thresholds and site conditions are considered. Managers and practitioners should consider the benefits and drawbacks of using different levels of uncertainty when monitoring vertical and horizontal changes, as more conservative thresholds (e.g., CI95) reduce the risk of interpreting noise as real change but may also filter out subtle yet ecologically meaningful changes in short-term monitoring. When installing new living shoreline structures, it is important to consider site characteristics when selecting substrates and designs. For example, while elevation remained stable at the low-relief reef site, marsh edge retreat occurred across all sites, demonstrating the complexity of site-specific responses. More long-term studies are needed to evaluate the use of novel living shoreline structures in varying environmental conditions, and future research should incorporate hydrodynamic data such as tidal currents and seaward slope to provide context for understanding the patterns observed in the drone-derived data.

In addition to post-construction monitoring, drone-based surveys offer powerful tools for pre-construction assessments by enabling high-resolution mapping of elevation, slope, and land cover to inform site selection and design. Pre-construction data are critical for avoiding impacts to ecologically valuable habitats and for establishing robust baselines for change detection. However, as Baker and Gittman (2024) highlighted, most living shoreline projects lack co-funded monitoring at the planning stage, limiting the ability to rigorously evaluate ecological and geomorphic outcomes. This study reflects that challenge, as no pre-restoration data with comparable horizontal resolution and vertical accuracy were available to detect fine-scale changes. This limitation underscores the need to prioritize high-resolution baseline surveys, particularly with drone-based methods. Future efforts should prioritize co-funded pre-restoration drone surveys with in-situ observations to support more rigorous before–and–after comparisons and better link observed outcomes to site characteristics and intervention types.