Predicting tidal marsh vegetation for habitat restoration design, monitoring, and Sea Level Rise
Project Abstract
RTK-GPS surveys of tidal marsh vegetation have been completed over the past 20 years, incidental to other work, in the Skagit, Stillaguamish, Snohomish, Nooksack, Nisqually, Dosewallips, and Hamma-hamma deltas, as well as the margins of Padilla Bay. This project uses this data to develop predictive models for vegetation species distributions for Puget Sound tidal marshes. This model could provide guidance to design and adaptively manage tidal marsh restoration and to anticipate and plan for impacts, such as sea-level rise or salinity change.
Problem Statement
Tidal marsh restoration to achieve ecological and social goals requires confidence that ecological engineering actions will be effective. This requires design and planning tools that can accurately predict likely outcomes of restoration actions. A critical element of marsh restoration is vegetation, because vegetation is the foundation of marsh food webs that support fish and wildlife, and vegetation also interacts with hydrodynamic and sedimentary processes to stabilize marshes in the face of sea-level rise, as well as short-term disturbances such as floods and storms. Thus, prediction of the vegetation likely to be supported by a restoration project is needed to appropriately choose restoration sites for desired vegetation, or to develop grading plans to achieve essential marsh plain elevations.
Predictive vegetation models (PVMs) have been developed on a site-scale basis for only a few restoration projects: Fir Island Farms restoration (Skagit Delta), the zis-a-ba and Leque Island (Stillaguamish Delta), and a 3-acre experimental non-native cattail treatment site (Skagit Delta) (Hood 2013). Typically, restoration project planners and designers do not use PVMs, but instead assume their restoration site will look approximately like a nearby reference site, without any particular justification beyond the proximity of the reference site. This leads to coarse expectations and coarse evaluations of success, because of the imprecision of the approach. Finer spatial scale prediction, based on spatially varying processes that affect vegetation composition, will allow more effective restoration project planning, design, and evaluation (monitoring).
Some common practical examples of the utility of PVMs include:
-
A subsided agricultural site is chosen for restoration. Is the subsided elevation too low for vegetation? Will sediment need to be imported to raise the site elevation to allow vegetation development? What should the design elevation be?
-
After channel excavation and dike removal there is an excess of sediment to dispose, but hauling the sediment off-site is expensive. If sediment is regraded on-site to achieve some habitat diversity, e.g., as natural levees along excavated channels, or other slightly higher areas, what new species of vegetation might occupy those areas? Would tidal shrubs be appropriate? Sitka spruce? Is there risk of invasive non-native species colonizing the higher sites?
-
If a problematic non-native species exists on a proposed restoration site, e.g., reed canarygrass, would regrading be effective in controlling it and what species would likely replace it? Shallow excavation of marsh plains in the Columbia River Estuary has been used effectively to replace reed canarygrass with native wapato.
-
If control of a non-native species is desired in existing tidal marshes, e.g., narrow-leaf cattail, which areas in a river delta might have the highest success?
Beyond planning and design, PVMs provide standards to evaluate monitoring results and support adaptive management. For example, is a restoration site bare because it is too subsided or is high sheet flow sheer stress, resulting from insufficient channel excavation, preventing seed recruitment? PVMs for Puget Sound river delta marshes, collectively or individually, will facilitate adoption of PVMs in marsh restoration planning, design, and evaluation.
Hypothesis Statement
Tidal marsh vegetation species distributions can be predicted by a statistical model from independent variables that include marsh elevation, salinity, tide range, and fetch.
Tidal vegetation species distributions are known to be constrained by physical stress (salinity, inundation duration, wave exposure) at their seaward limits and by competition with other vegetation at their landward limits (Vince and Snow 1984; Bertness and Ellison 1987). The physical stresses can be either directly measured (salinity) or indirectly indexed (tidal range x elevation; fetch x elevation). Thus, measures of elevation, salinity, tidal range, and fetch should be predictive of species distributions. In addition to competition, other factors can also affect species distributions, such as physical disturbance (e.g., by logs, which affects competition), herbivory (by geese, ducks, insects), nutrient pollution (especially nitrogen, which affects competition), and sediment texture (e.g., sand vs. silt, which affects drainage and inundation stress, and nutrient retention). The influence of sediment texture may be tested on an opportunistic basis. For example, large newly developing sandbars at the mouth of the North Fork Skagit distributary are rapidly being colonized by willows (Salix spp.) and alder (Alnus rubra), and appear to be growing at an elevation at least 30 cm lower than normal in the Skagit marshes (personal observation). This may be a rare opportunity to contrast the effects of different soil textures on vegetation distributions in tidal marshes. The other additional factors that can affect species distributions, are beyond the scope of the proposed analysis, because they would require considerable additional effort and cost to examine comprehensively. Furthermore, physical stresses are likely the most important predictive variables, while competitive effects can be accounted for, statistically, by their likely correlation with elevation, i.e., competition occurs at the landward (higher elevation) extent of a species’ elevational range, so competitive effects are to some degree included in the factor, elevation. For the purposes of developing a predictive tool, it is not essential to entirely resolve all of the underlying causative mechanistic details of species distributions. Further examination of omitted controls on vegetation distributions can occur in the future, if the PVM developed by the current effort is shown to be seriously deficient, or if patterns of deficiency are observed that appear potentially related to the omitted predictors. Experience to date (e.g., Hood 2013, 2019a, b) suggests that the limited focus on physical predictors is likely to be successful for the purpose of developing a useful predictive tool for restoration planning, design, and monitoring.
Methods
Sampling method
If necessary, data gaps at various sites will be filled using point sampling with an RTK-GPS (3-cm horizontal and vertical resolution) along a series of random transects spanning areas of interest, with each sampling point at least 25 m apart to minimize spatial autocorrelation. At each GPS point the dominant and subdominant vascular plant species will be noted while the RTK-GPS simultaneously acquires the horizontal and vertical location of the point. Relative plant abundance at each point will be determined by visual estimation of aerial cover within a 1-m radius from the sample point (Hood 2013).
Statistical analysis
Maxent, a machine learning approach to modeling species niches and distributions, is open source software from the American Museum of Natural History (https://biodiversityinformatics.amnh.org/open_source/maxent/), and may be used to develop PVMs for each species of interest. A comparison of six common statistical approaches to PVM development has shown that maxent provides the best results (Tarkesh and Jetschke 2012). Non-parametric multiple regression (NPMR) has also been advocated (McCune 2006), is available in accessible software (HyperNiche2), and will be an alternate approach, if necessary. Exploratory statistical analysis and explanatory graphical presentations of species distributions will likely follow earlier published approaches (Hood 2013).
PVMs will first be developed individually for each river delta marsh using a training set of data that will consist of a random selection of 80% of the available vegetation data. The remaining 20% of the available vegetation data will be compared to model predictions to determine model reliability for each river delta marsh. After learning from these experiences, a synoptic PVM will be attempted for all of the study sites collectively, and similarly tested for reliability. Model performance will be assessed using confusion matrices and associated statistical summaries (Allouche et al. 2006).
References
Allouche O, A Tsoar, R Kadmon. 2006. Assessing the accuracy of species distribution models: Prevalence, kappa and the true skill statistic (TSS). Journal of Applied Ecology 43: 1223-1232.
Bertness, M and AM Ellison. 1987. Determinants of pattern in a New England salt marsh plant community. Ecological Monographs 57: 129-147.
Hood WG. 2019a. Fir Island Farms 3rd-year Post-Restoration Tidal Marsh Monitoring Report. Prepared for Washington Department of Fish and Wildlife. Skagit River System Cooperative, LaConner, WA.
Hood WG. 2019b. zis a ba 2nd-year Post-Restoration Tidal Marsh Monitoring Report. Prepared for Stillaguamish Tribe Natural Resources Department. Skagit River System Cooperative, LaConner, WA.
Hood, WG. 2013. Applying and testing a predictive vegetation model to management of the invasive cattail, Typha angustifolia, in an oligohaline tidal marsh reveals priority effects caused by non-stationarity. Wetlands Ecology and Management 21: 229-242.
McCune, B. 2006. Nonparametric Multiplicative Regression for Habitat Modeling. http://www.pcord.com/NPMRintro.pdf.
Tarkesh, M. and G. Jetschke. 2012. Comparison of six correlative models in predictive vegetation mapping on a local scale. Environmental and Ecological Statistics 19: 437-457.
Vince, SW and AA Snow. 1984. Plant zonation in an Alaskan salt marsh. Journal of Ecology 72: 651-667.
Relationships
- broader: Sea otters
- broader: floodplain;river delta
- involves organization: tribal