Habitat restoration is commonly used to recover ecosystem services, but due to resource constraints, post-project monitoring often fails to fully evaluate the recovery of important ecosystem functions. Metric-based indicators use simple-to-measure variables to assess ecosystem health and function, thereby providing a time- and cost-effective method to improve monitoring. We used a tidal marsh dataset to develop metric-based indicators of ecosystem recovery. In 2021 and 2022, we surveyed eight restored/created and three natural reference tidal marshes in the northern Gulf of Mexico to assess recovery of ecosystem attributes [e.g., above- and below-ground biomass, soil organic matter (SOM), and sediment total carbon (C) and total nitrogen (N)]. To determine what combinations of variables best predicted recovery, we split our data into model training and testing datasets, used backward model selection, and then created and tested a metric-based indicator of ecosystem recovery. Recovery of plant above- and below-ground biomass and sediment structure (i.e., SOM, C, and N)—important measures of wetland carbon sink capacity and biogeochemical functioning—could be predicted through a combination of simpler-to-measure variables, such as time since restoration, percent plant cover, and sediment bulk density. The indicator constructed from these relationships was highly effective in predicting the development of ecosystem attributes (r = 0.85, p < 0.001). This indicator approach provides an effective but simple method to assess the recovery of ecosystem attributes in tidal marshes, and it can be used to develop similar indicators in other ecosystems. By overcoming resource constraints of post-project monitoring, metric-based indicators of ecosystem recovery may serve as a key strategy to improve restoration outcomes.

Study Sites

We conducted our study in the summer of 2021 and 2022 in three references (i.e., natural) and eight restored/created tidal marshes along the Mississippi-Alabama coast of the northern Gulf of Mexico, USA. In Mississippi, sites included two restored marshes (Deer Island-1 and Deer Island-2) located on Biloxi Bay and one restored marsh (Greenwood Island) on the Mississippi Sound. One natural (Fowl River Natural) and three created (Fowl River CON-0, CON-1, and CON-2) marshes were clustered along the West Fowl River on the western shore of Mobile Bay in Alabama. Helen Wood Park, a restored brackish marsh near the mouth of Dog River, was also located on the western shore of Mobile Bay. Two natural marshes (Pelican Point and Weeks Bay Natural) were in Weeks Bay on the eastern shore of Mobile Bay in Alabama. The final restored marsh was along Perdido Bay in Perdido Beach, Alabama.

The eight restored marshes varied in age (7-34 years since restoration), salinity, and vegetation and were developed using three distinct restoration strategies—habitat conversion, living shoreline, and beneficial use. Fowl River CON-0, CON-1, and CON-2 were created by converting coastal pine savanna to tidal marsh in 1987-1988 to mitigate impacts from a coal and grain facility by harvesting pine trees, excavating topsoil to an elevation of +0.3 m (NAVD88) relative to mean sea level, digging canals to create hydrological connectivity, and planting with Spartina alterniflora and Juncus roemerianus. The areas sampled in these three restored sites are now dominated by J. roemerianus, with a narrow fringe of S. alterniflora along the shoreline of the tidal canal. Helen Wood Park and Perdido Beach are living shorelines that are under 20 years old and were planted with and currently dominated by, S. alterniflora and J. roemerianus, respectively. Finally, Deer Island 1, Deer Island 2, and Greenwood Island were created 7-19 years ago by the Mississippi Department of Marine Resources by re-establishing the marsh platform through the beneficial use of dredged sediments and allowing S. alterniflora or S. patens to naturally colonize the sites.

Each of the restored sites was paired apriori with one of the three reference sites based on a combination of factors, including geographic proximity (preferably within the same watershed), site salinity, and vegetation. When marshes in close proximity were not a good match in terms of other site characteristics, we paired sites with similar vegetation, as with Deer Island 1 and Weeks Bay Natural. When designating these site pairings, we also considered the original restoration objectives and management targets, as they could affect the potential for ecosystem attribute recovery. For example, we opted to pair Fowl River Natural with Fowl River CON-0, CON-1, and CON-2, as opposed to other sites that also had similar vegetation and salinity, because it served as the original reference for these constructed marshes.

Data Collection

In May 2021, we randomly established three sampling locations within the dominant plant zone of each site. We collected data during peak growing season (July-August) in 2021 and 2022 from 1 m² quadrats, which were positioned on opposite sides between years to minimize disturbance and avoid resampling the same area. We surveyed areal plant cover within each 1 m² plot (n = 3 plots/marsh) to characterize plant community composition. All plants were identified to species level; vouchers for plants that could not be identified on-site were collected from outside the plots for subsequent identification. Once cover was estimated, standing aboveground biomass was collected from a randomly placed 0.25 m² subplot within the 1 m² plots. Plants were clipped at the sediment surface, bagged, and transported to the lab, where they were sorted into live components by species and total dead and then dried at 60 °C to a constant mass (g m^-2). We collected a sediment core (7.9 cm diameter x 20 cm deep) from within the clipped subplot to assess belowground biomass. This core was sub-sectioned in the field into two, 10-cm sections. In the lab, subsections were rinsed over a sieve (1 mm), and biomass was sorted by size class (i.e., fine, coarse, rhizome) and dried at 60 °C to a constant mass (g m^-2 by depth interval). Roots were not sorted by species; therefore, samples represent community belowground biomass. We did not harvest above- and below-ground biomass at the Perdido Beach living shoreline site, as is it small and we wanted to minimize disturbance. For the indicator, we only included biomass data from the upper 0-10 cm because we were unable to physically core the full 20 cm interval at several of the restored sites.

To quantify sediment attributes [i.e., soil organic matter (SOM), sediment total carbon (C), and nitrogen (N)], we collected duplicate sediment cores (5 cm diameter x 20 cm depth) from each plot. Cores were sub-sectioned in the field into four, 5-cm sections. In the lab, subsections were dried at 60 °C to a constant mass to calculate the bulk density (g cm^-3). Homogenized subsamples were combusted in a muffle furnace at 550 °C for six hours to determine % SOM by loss-on-ignition. Another set of subsamples was analyzed for C and N content at the Alabama Stable Isotope Laboratory. For data analysis, we obtained mean sediment attribute values for bulk density, SOM, C, and N from each plot per year. As with belowground biomass, we only included data from the upper 0-10 cm (i.e., 0-5 and 5-10 cm intervals) because we were unable to physically core the full 20 cm interval at several of the restored sites. To characterize pH and salinity, commonly measured site characteristics that can influence or reflect variation in plant community structure, productivity, and biogeochemistry, porewater was extracted from each plot at 15 cm below the sediment surface using sipper tubes. We measured pH and salinity using a Thermo Orion Star pH meter (model A211) and a YSI conductivity and salinity instrument (model 3100), respectively.

Model Selection Procedure

We first classified response variables as either proxies or ecosystem attributes. Proxies, like site age, plant percent cover, bulk density, pH, and salinity, represent simpler-to-measure components that, alone or in combination with other proxies, can explain or predict some broader, often harder-to-measure, attribute. In this case, proxies represented variables that can be measured at lower costs than many attributes and from relatively small areas to minimize disturbance to, and negative impacts of research activities on, recently restored habitats. Ecosystem attributes represent factors and functions critical to succession from young (e.g., restored) to mature (e.g., references) ecosystems, and that may be either driven by or correlated with proxies. For tidal wetlands, examples include above- and below-ground standing biomass, SOM, and sediment C and N, which should increase with increasing time since restoration. Specifically, we expect plant biomass, SOM, and sediment C and N to increase with increasing site age, as plant cover increases, and as bulk density decreases, reflecting the accumulation of plant biomass, root penetration into the sediments, and greater inputs of plant-derived OM, C, and N to the sediments. Additionally, because plant structure and productivity have been shown to be influenced by, or correspond to, changes in porewater chemistry, variation in site characteristics like salinity or pH may also be useful predictors of ecosystem attributes. Thus, proxies represent the easier, cheaper, and less destructive variables to measure that are also related to, or predictive of, other more difficult, expensive, or destructive variables to measure (i.e., attributes).  

We paired each restored marsh with a reference based on a combination of geographic proximity, plant community composition, and salinity, as described above. Using the site pairings, we calculated response ratios for each plot (n = 3) in every restored marsh using the following equation:

RR_A = (A_R/A_N)

where RR_A is the response ratio of an ecosystem attribute in a restored site relative to its reference, A_R is the value of the attribute calculated at the plot level in the restored marsh and is the value of the attribute calculated at the paired reference marsh. This ratio indicates the recovery trajectory of a given attribute for a restored site relative to its reference. A response ratio < 1 indicates that the restored marsh is underperforming relative to its reference, while a response ratio ≥ 1 indicates that the restored marsh is at least equivalent to the reference. We set the maximum response ratio value to 1, which gave all attributes a shared range of values and facilitated indicator development (described below). In a few instances, restored marshes did outperform their reference (RR > 1), as detailed in the published dataset.

Data from 2021 and 2022 were used to develop two separate datasets (training and testing). The training dataset included two randomly selected plots per site from both years (66% of the total data, n = 30). The testing dataset included the remaining plot from each site per year (n =15) and was withheld from indicator development for subsequent validation (described below). Specifically, we used least squares regressions with backward model selection, with p-value as the stepping method criteria, to determine what combinations of proxies (i.e., explanatory covariates) were most strongly correlated with restoration response ratios for the ecosystem attributes. The top model for each attribute was then used to generate metrics for the indicator.

Indicator Creation and Validation

To create metrics for ecosystem attributes, we followed previously established approaches for IBIs using calculated metrics relating proxies to attribute recovery. First, we created scatterplots relating each proxy (x-axis) to its response ratio (y-axis). Second, we split our gradient of interest (y-axis) into three even groups. Third, we “binned” points along the x-axis intentionally so that most points within each bin fell within the group. Lastly, we assigned a “score” to each bin, the bounds of which constituted the scoring range for each metric. Specifically, we used the absolute value of each proxy’s standardized coefficients to determine the proportional contribution of each proxy to the overall metric. This proportion was used as the total possible score value for a given proxy within the metric, with a maximum score of 1 for a given metric. For example, if a metric included both site age and percent plant cover as proxies and the absolute values of their standardized coefficients within the top model were 0.75 and 0.25, respectively, then the maximum score value for site age within the metric would be 0.75 and the maximum score value for percent plant cover within the metric would be 0.25, for a possible maximum metric of 1. This process was repeated for all proxies in each top model for every ecosystem attribute to create the indicator. To predict recovery, the indicator was then applied to the testing dataset, and recovery scores were compared to the combined restoration ratios to determine effectiveness using Pearson correlations. We opted to use univariate models for this process because they are simpler to construct and interpret, thereby facilitating use by diverse audiences with varying levels of expertise, and they allow for a more straightforward binning approach, which provides a simple scoring system to describe the relative recovery of a restored site.