Evaluation of Plot-Scale Methods for Assessing and Monitoring Salt Marsh Vegetation Composition and Cover

Abstract

Vegetation is a key component of salt marsh monitoring programs, but different methods can make comparing datasets difficult. We compared data on vegetation composition and cover collected with 3 methods (point-intercept, Braun–Blanquet visual, and floristic quality assessment [FQA]) in 3 Rhode Island salt marshes. No significant differences in plant community composition were found among the methods, and differences in individual species cover in a marsh never exceeded 6% between methods. All methods were highly repeatable, with no differences in data collected by different people. However, FQA was less effective at identifying temporal changes at the plot scale. If data are collected from many plots in a marsh, any of the methods are appropriate, but if plot-scale patterns are of interest, we recommend point-intercept.

Introduction

Tidal salt marshes in various regions are undergoing unprecedented change, often due to impacts from accelerating sea-level rise working alone or synergistically with other stressors (Cahoon et al. 2019, Raposa et al. 2018, Wasson et al. 2017, Watson et al. 2017). Changes include net marsh-area loss, edge erosion, channel widening, proliferating crab populations, marsh surface pond formation, and vegetation shifts indicative of increased tidal flooding (Luk and Zajac 2013, Raposa et al. 2017, Watson et al. 2017, Wilson et al. 2012). The need for assessing and monitoring key indicators of marsh condition, structure, and function is clear, but the methods used to measure such indicators can vary among projects, making comparisons across marshes, regions, and organizations difficult.

Parameters that describe vegetation communities (e.g., composition, species cover, and richness) are an almost-universal component of salt marsh monitoring and assessment programs. Vegetation forms the structural and trophic foundation of salt marshes, trapping organic matter and inorganic particles needed for vertical marsh growth, while providing food and refuge for a variety of resident and nursery species (McKinney et al. 2008, Morris et al. 2002). Vegetation reacts predictably to common stressors, such as sea-level rise, (Raposa et al. 2017), nutrient loading (Wigand et al. 2014), impoundment (Roman et al. 1984), and trophic-level disturbances (Bertness et al. 2014). It is also easy to relate to a wide variety of stakeholder groups; healthy vegetation portrays a properly functioning marsh, while stressed or absent vegetation can signify marshes in trouble and in need of management (Cole Ekberg et al. 2017, Coleman and Kirwan 2018, Wigand et al. 2017).

Salt marsh vegetation is typically dominated by a few stress-tolerant species, and is therefore relatively easy to sample. A long list of possible parameters exists for describing marsh vegetation communities and condition, but species composition, cover, richness, canopy height, and to a lesser extent, stem density, are often selected, in part due to the ease with which they are quantified in the field. Two plot-based methods are the most common ways to assess vegetation composition and cover. The visual estimation approach is categorical. It involves one or more field personnel estimating the relative cover (%) of individual vegetation species (and often other cover classes such as bare ground), grouped into one of several potential cover-class categories (e.g., the 5 Braun–Blanquet classes; Kent and Coker 1992). In contrast, the point-intercept approach is quantitative. It uses a thin rod vertically dropped at multiple points within each plot, and every vegetation species hit by the rod is recorded as a discrete data point; the estimated cover of each species can then be easily calculated (e.g., if a species is hit 10 times out of 50 points, its percent cover is [10 / 50] * 100 = 20%). The visual categorical approach is easy to conduct, and surveys can be conducted rapidly by experienced personnel; however, this method can be prone to error due to observer bias (e.g., Kennedy and Addison 1987). The point-intercept method is considered more objective and better suited for quantitative statistical analysis but can be more labor intensive, especially in plots containing multiple intermixed species or multiple canopy layers, and often misses rare species (Dethier et al. 1993). Some research has compared quantitative versus categorical plant-survey methods, and a general finding is that data from the 2 approaches are often similar and comparable (Damgaard 2014; Poissonet et al. 1972; Smartt et al. 1974, 1976). However, most of these studies compared the 2 approaches over single or short-term sampling periods. Plant-sampling methods also need to be evaluated in the context of longer-term monitoring and compared with alternate rapid assessment methods that are now being implemented in some US state monitoring programs. One such rapid plant survey method has been used to support floristic quality assessment (FQA) in Rhode Island (Kutcher and Forrester 2018). This FQA method is a categorical visual estimation method similar to Braun–Blanquet, but it includes only 3 broad cover classes.

The overall goal of this study is to directly compare 3 field sampling methods (point-intercept, Braun–Blanquet visual, and FQA visual) for estimating community composition and cover of salt marsh vegetation. Specifically, this study aims to (1) directly compare community composition and percent cover of species across the 3 methods, (2) quantify the repeatability of each method with replicated sampling events, (3) compare data collected by different people, and (4) use existing long-term point-intercept data to evaluate the power of each method for detection of temporal changes. Results from our study will help guide efforts to synthesize and compare data from disparate programs using multiple survey methods and help organizations pick appropriate sampling methods to meet their monitoring goals.

Field-site Description

This study was conducted at 3 sites within 2 tidal salt marshes in Rhode Island, where plot-based point-intercept vegetation monitoring was already occurring as part of distinct projects (Fig. 1). The first site, Coggeshall reference marsh, is a 21-ha section of salt marsh located on Prudence Island within the Narragansett Bay National Estuarine Research Reserve (NBNERR). It is a designated NERR-Sentinel Site for examining long-term marsh changes in response to changing tidal water levels (NERRS 2012); vegetation was monitored 3 times at this site from 2000 to 2004 and annually from 2008 to 2017. The second site, Coggeshall restoration marsh, is a 4-ha section of marsh abutting the Coggeshall reference marsh. It was identified as severely waterlogged in 2012 and subsequently underwent channel excavation as part of an experiment to evaluate its effectiveness at improving marsh draining and alleviating waterlogging. Vegetation was monitored annually at this site from 2013 to 2016 as part of this experiment. The third site, Ninigret Marsh, is a 12-ha marsh located in a coastal lagoon along Rhode Island’s south shore. It is the site of a large-scale sediment placement experiment, where nearby dredge sediments were placed on the marsh to increase its elevation. Vegetation has been monitored annually from 2015 to 2018 as part of this experiment. All 3 sites are generally dominated by short-form Spartina alterniflora Loisel (Smooth Cordgrass), with patches of salt meadow species (Spartina patens (Aiton) Muhl. [Saltmeadow Cordgrass], Distichlis spicata (L.) Greene [Marsh Spikegrass], and Juncus gerardii Loisel [Saltmeadow Rush]) at higher elevations and Iva frutescens L. (Jesuit’s Bark) spanning much of the upper marsh border. Bare patches are interspersed throughout much of each marsh. Coggeshall reference and restoration marshes are meso-tidal (range = ~0.75 m); Ninigret is micro-tidal (range = ~0.3 m). All 3 marshes are generally polyhaline (i.e., salinities varying from ~18 to 30 ppt), except for some brackish patches along the upper borders.

Figure 1.

Map of the 3 salt marsh study sites located in Rhode Island.

Methods

Field sampling methods

In August/September 2016, we measured vegetation community composition and percent cover at all 3 marshes using the point-intercept, Braun–Blanquet, and FQA methods. We sampled 21 plots in Coggeshall reference, 24 plots in Coggeshall restoration, and 20 plots in Ninigret using 1-m2 quadrats. In each marsh, plots were established along multiple transects that ran from the marsh/estuary edge to the marsh/upland edge, following protocols outlined in Roman et al. (2001).

For the point-intercept method, we sampled 50 grid points in each plot (7 rows by 7 columns, with 1 random point) and used a 3 mm diameter by 0.9 m long welding rod as the intercept. All vegetation species hit by the rod at every point were recorded (multiple species could be hit each time, resulting in total percent cover estimates that could potentially sum to greater than 100% in a single plot). Other cover types (e.g., stone) were also recorded if hit. Bare ground was only recorded if the rod did not hit any live vegetation at a given point. Conspicuously dead vegetation was recorded as bare, and wrack was recorded as a separate, higher canopy layer (i.e., if wrack was hit, we also recorded anything that the rod hit underneath it). For the categorical methods, a single observer would estimate vegetation cover by visually examining each plot and categorizing each vegetation species or other cover type into 1 of 5 cover classes (<1–5%, 6–25%, 26–50%, 51–75%, >75%) for Braun–Blanquet, or 1 of 3 classes (scarce: <10%, common: 11–60%, dominant: >60%) for FQA.

Data were collected by a single person using all 3 methods in sequence for each plot at each marsh. The order of each method at each plot was randomly assigned to reduce sampling bias that might occur if the methods were always conducted in the same order. To examine variability in data collected by the same person (i.e., to estimate the repeatability of each method), we repeated surveys using each method on a second date at a subset of plots at each marsh (13 total plots for point intercept, 13 for Braun–Blanquet, and 14 for FQA). Finally, to examine variability in data collected in the same plot by different people, data were collected at a subset of 6 plots at Ninigret by 2 different people on separate dates using each method.

Data analysis

We used Stuart’s tau-c (Stuart 1953) to estimate the degree of association among the percent cover measurements of target species/cover types (S. alterniflora, S. patens, D. spicata, I. frutescens, Phragmites australis (Cav.) Trin. Ex Steud [Common Reed], Baccharis halimifolia L. [Eastern Baccharis], and bare ground) collected from the 3 different methods. Stuart’s tau is a nonparametric evaluation of the degree of agreement among ordinal categorical measurements that accounts for differing numbers of categories among the measurements. Using this approach allowed us to evaluate degrees of association with the point-intercept method (which, as a quantitative method, had a much larger number of possible result values than the other methods) as well as between the Braun-Blanquet and FQA methods (using midpoints for each cover class), using a consistent methodology.

We used Whittaker’s percent similarity index to quantify variability in data collected by the same person on 2 separate dates (i.e., within-person variability, or repeatability) and variability in data collected at the same plots by 2 different people. This index includes an adjustment of counts into percentages and allowed us to compare data from an overall plot rather than on a species-specific basis. We calculated the index for each plot using raw point-intercept data and midpoints for each Braun–Blanquet and FQA cover class using the formula:

$$\text{PSc}=100-0.5\left(\Sigma \left|{\text{a}}_{\text{i}}-{\text{b}}_{\text{i}}\right|\right),$$

where a_i and b_i are the percentages of the plot that the given species i represents for measurements a and b, respectively (where “a” corresponds to the initial measurement and “b” to the re-measurement, or “a” to one person and “b” to the second). The calculated PSc score would be 100% if there was total agreement between the 2 plots being compared across all species, and 0% if there was no agreement whatsoever (i.e., the closer the calculated value is to 100%, the greater the similarity). We then used one-way ANOVA to compare mean PSc scores across plots to determine (1) if there were differences in repeatability among methods and (2) if between-person variability differed among the methods.

We used one-way analysis of similarity (ANOSIM; PRIMER v.7.0.13) to statistically compare vegetation community data collected with the 3 different methods. We ran a total of 3 ANOSIM tests (1 for each marsh) to directly compare the methods. Additionally, we used ANOSIM to compare vegetation community data collected with the same method among the 3 marshes (e.g., Coggeshall reference vs Coggeshall restoration vs Ninigret, using only point-intercept data) to determine if each method produced the same general statistical results. All ANOSIM tests were run using a Bray-Curtis resemblance matrix and a square-root data transformation to downweight dominant species. We again used midpoints of each cover class for both visual methods to convert class scores to actual estimates of percent cover.

We also explored how data collected with each sampling method might be used to visualize long-term temporal trends in cover of individual species by analyzing our longest existing time series of point-intercept data (Coggeshall Marsh from 2000 to 2017), focusing on the cover of S. alterniflora, S. patens, and bare ground over time. For each plot each year, we then converted these actual cover data to the corresponding cover class for each of the 2 visual methods used in this study (e.g., if S. alterniflora cover in a plot was 45% from point-intercept, it was coded as class 2 for the FQA and class 3 for Braun–Blanquet). These conversions are of course hypothetical; had FQA and Braun–Blanquet data also been collected in the field during these years, results may have differed. However, given that cover estimates for individual species are very similar among the 3 methods (see below), these conversions are likely realistic. We then plotted cover data from each method to visualize patterns over time. Finally, we examined the ability of each method to detect long-term changes in cover over time at the individual plot scale (again focusing on S. alterniflora, S. patens, and bare ground). We ran linear regression tests using 2000–2017 cover data for each in the 21 Coggeshall reference plots (again, using converted visual data from actual point-intercept data). Bonferroni corrections were applied for each method, resulting in an adjusted alpha of 0.0024.

Results

Comparing species cover between methods

Estimates of individual species cover in each marsh were generally similar among the 3 methods. For all species in all marshes, absolute differences in cover between pairs of methods was always less than ~6%, and less than 3% in 65% of cases (Table 1). Stuart’s tau coefficients comparing cover of target species in plots between point-intercept, Braun–Blanquet, and FQA demonstrate strong associations between method pairs in all cases, with correlations significantly greater than zero for all species for all 3 method pairs (Table 2). Generally, FQA had slightly greater agreement with the point-intercept method than the Braun–Blanquet method did, with a tau greater by an amount varying from 0.02 to 0.10 for each species. The agreement between the point-intercept and FQA methods was comparable to that between the FQA and Braun–Blanquet methods. For all method comparisons, the level of agreement tended to be weaker for species that were found in fewer plots, specifically, B. halimifolia, I. frutescens, and P. australis.

Table 1.

Absolute percent cover of vegetation species in 3 marshes from 3 assessment methods (PI = point-intercept, BB = Braun–Blanquet, and FQA = floristic quality assessment), and the % difference in cover between pairs of methods. Differences in cover are color-coded using conditional formatting in Microsoft Excel so that larger differences are shown in red and smaller differences are green, with intermediate levels in yellow and orange.

¹SPAALT=Spartina alterniflora; SPAPAT=Spartina patens; BARE=bare ground; DISSPA=Distichlis spicata; JUNGER=Juncus gerardii; IVAFRU=Iva frutescens; PHRAUS=Phragmites australis; ELORUS=Eleocharis rostellata; BACHAM=Baccharis halimifolia; SALSPP=Salicornia spp.; LIMCAR=Limonium carolinianum; ATRPAT=Atriplex patula; PLAMAR=Plantago maritima; WRACK=wrack; SYMTEN=Symphyotrichum tenuifolium; UNKNOWN=unknown; MORPEN=Morella pensylvanica; SCHAME=Schoenoplactus americanus; ALTOFF=Althaea officinalis; PLUODO=Pluchea odorata; VACCOR=Vaccinium corymbosum; SOLSEM=Solidago sempervirens.

Table 2.

Comparison of target species percent cover (plots as replicates) between the quantitative point-intercept (PI) and categorical Braun–Blanquet (BB) and FQA methods using Stuart’s tau-c.

Comparison	Cover type	Tau-c	Standard error	P value
PI v BB
	Baccharis halimifolia	0.28	0.09	0.0021
	Bare ground	0.83	0.04	<0.0001
	Distichlis spicata	0.87	0.05	<0.0001
	Iva frutescens	0.45	0.09	<0.0001
	Phragmites australis	0.40	0.09	<0.0001
	Spartina alterniflora	0.85	0.04	<0.0001
	Spartina patens	0.80	0.06	<0.0001
PI v FQA
	Baccharis halimifolia	0.31	0.10	0.0019
	Bare ground	0.86	0.05	<0.0001
	Distichlis spicata	0.93	0.04	<0.0001
	Iva frutescens	0.48	0.10	<0.0001
	Phragmites australis	0.41	0.10	<0.0001
	Spartina alterniflora	0.90	0.03	<0.0001
	Spartina patens	0.89	0.05	<0.0001
BB v FQA
	Baccharis halimifolia	0.31	0.10	0.0019
	Bare ground	0.81	0.06	<0.0001
	Distichlis spicata	0.92	0.04	<0.0001
	Iva frutescens	0.48	0.09	<0.0001
	Phragmites australis	0.41	0.09	<0.0001
	Spartina alterniflora	0.89	0.03	<0.0001
	Spartina patens	0.86	0.06	<0.0001

Repeatability, and variability between people

On average, mean percent similarity between initial and re-measurement data collected by the same person was greater than 90% for all 3 methods, and exceeded 80% at almost every plot (Fig. 2). Percent similarities tended to vary most widely for FQA; this result was due to the smaller number of cover categories for this method, which resulted in more frequent 100% agreement, but larger deviations from 100% when there was a difference in score. However, mean percent similarity was not significantly different among the 3 methods (ANOVA: F = 0.50, P = 0.61). In contrast, percent similarities for data collected by different people were slightly lower than between re-measurements. Mean similarities for data collected by different people varied between 80% (FQA) and 91% (Braun–Blanquet) but overall were also not significantly different among methods (ANOVA: F = 1.82, P = 0.20).

Figure 2.

Variability in vegetation cover data collected by (a) the same person on 2 different dates (i.e., repeatability) and (b) 2 different people on different dates. In each case, bars are shown for each of the 3 methods included in our study (PI = point-intercept, BB = Braun–Blanquet, and FQA = floristic quality assessment). Each bar is the mean and SE of percent similarity (PSc) scores based on data from replicated plots (a: n = 13 for PI, 13 for BB, and 14 for FQA; b: n = 6 for each method).

Comparing marsh community composition among methods

There were no significant differences in vegetation community composition among sampling methods for any marsh (ANOSIM: Coggeshall reference Global R = −0.022, P = 0.90; Coggeshall restoration Global R = −0.031, P = 0.96; Ninigret Global R = −0.037, P = 0.98). Vegetation communities were, however, significantly different among the 3 marshes based on all 3 sampling methods. In each case, pairwise comparisons resulted in the same findings; Coggeshall reference and Coggeshall restoration were always significantly different, as were Coggeshall reference and Ninigret, while Coggeshall restoration and Ninigret were never significantly different (Table 3).

Table 3.

Summary of Analysis of Similarity (ANOSIM) results comparing vegetation community data collected with the 3 different survey methods. The first 3 comparisons directly compared the methods in each individual marsh; the second 3 compared communuties among the 3 marshes separately for each method to see if the methods produced the same general results.

Comparison	Global R	P	Pairwise comparison	Pairwise R	P
Methods within COGG REF	−0.022	0.90
Methods within COGG REST	−0.031	0.96
Methods within NINIGRET	−0.037	0.98
Marshes by point-intercept	0.060	0.02	COGG REF v COGG REST	0.071	0.02
			COGG REF v NINIGRET	0.101	0.02
			COGG REST v NINIGRET	0.003	0.41
Marshes by Braun-Blanquet	0.049	0.04	COGG REF v COGG REST	0.054	0.05
			COGG REF v NINIGRET	0.095	0.02
			COGG REST v NINIGRET	−0.011	0.55
Marshes by FQA	0.056	0.02	COGG REF v COGG REST	0.067	0.04
			COGG REF v NINIGRET	0.108	0.01
			COGG REST v NINIGRET	−0.011	0.53

²N=215,937. Data requested from SINAN, the national syphilis reporting system of the Ministry of Health, Brazil

Utility of each method for change analysis

Time-series point-intercept data from Coggeshall Marsh spanning 2000–2017 show that the percent cover of dominant species at the whole-marsh scale is rapidly changing over time, and the same patterns are clearly apparent after converting these data to corresponding Braun–Blanquet and FQA categorical cover classes (Fig. 3). The patterns from point-intercept and Braun–Blanquet are almost identical, but FQA deviated somewhat in a few instances (e.g., S. alterniflora cover from 2012–2016 and bare ground cover in 2014 and 2017).

Figure 3.

Change in percent cover over time for (a) Spartina alterniflora, (b) Spartina patens, and (c) bare ground from field-collected point-intercept data and Braun–Blanquet and FQA data estimated from the point-intercept value in each plot. Each point is the mean across 21 plots. All 3 methods generally show the same patterns over time (i.e., peak in S. alterniflora cover around 2013, steady decrease in S. patens over time, and rapid increase in bare ground since ~2013).

Additional differences among methods emerged when considering changes in a single cover type over time within individual plots. Pooled across the 3 tested (S. alterniflora, S. patens, and bare ground), a significant change in cover over time was detected by at least 1 method in 11 out of 21 plots in Coggeshall (a change was detected by at least 1 method in 5 plots for S. alterniflora, 8 plots for S. patens, and in 4 plots for bare ground; Table 4). Point-intercept and Braun–Blanquet detected the same change in the same plot 11 times, but this occurred only 5 times for point-intercept and FQA, and 6 times for Braun–Blanquet and FQA. Point-intercept and Braun–Blanquet also detected a similar number of overall changes (13 and 14, respectively), but FQA only detected 7. Thus, FQA was often not in agreement with the other 2 methods, and it detected fewer overall changes over time.

Table 4.

Summary of significant changes in cover over time for S. alterniflora, S. patens and bare ground in 21 individual plots in Coggeshall Marsh using data from three methods. In each case, linear regression was run to detect change using Bonferroni corrections within each species/method (n = 21; adjusted α = 0.0024).

Plot #	Spartina alterniflora			Spartina patens			Bare ground
	PI	BB	FQA	PI	BB	FQA	PI	BB	FQA
1								*
2	*	*	*	*	*	*
3
4				*	*
5
6	*	*		*	*		*
7				*	*	*
8
9									*
10								*
11
12				*	*
13	*	*	*	*	*
14		*	*	*	*	*
15
16
17
18
19
20	*			*	*
21

^*indicates significant change over time at this significance level. PI = point-intercept, BB = Braun-Blanquet, and FQA = floristic quality assessment.

Discussion

Our results show that the quantitative point-intercept and the categorical Braun–Blanquet and FQA methods produced very similar results for static onetime assessments of vegetation composition and cover at the whole marsh and plot scale. They also produced very similar long-term patterns in species cover at the site scale, but differences emerged among methods when analyzing cover over time at the plot scale. In general, all 3 methods are highly repeatable and produce statistically similar results from different field personnel. By all accounts from our study, data from the point-intercept and Braun–Blanquet methods are virtually interchangeable; the FQA is also generally comparable except for detecting change over time at the plot level. However, we caution that care must still be taken when directly comparing data collected with different methods because they are not always in perfect agreement, especially when conducting analyses for an individual plot.

The point-intercept and Braun–Blanquet cover class methods are both commonly used for marsh monitoring and assessment. Many studies demonstrate that data from these 2 approaches are often very comparable, and both have been recommended for use in monitoring (Dethier et al. 1993, Roman et al. 2001). The point-intercept is generally perceived as more objective (Bonham 1989), which theoretically minimizes bias and error that could be introduced from different observers participating in long-term monitoring programs. Others may prefer the Braun–Blanquet method because it is more easily applied by field personnel with varying levels of experience and is generally less time-intensive (Dethier et al. 1993). Our results agree with these earlier studies and show that in almost all situations we tested, data from the 2 methods are very highly comparable (e.g., mean species cover in a marsh never differed by more than ~3.5% between these methods, and was less than 2% different in 90% of cases), and the choice of which method to use may come down to personal preference, available resources, or to specific monitoring goals.

The FQA categories were designed to assign broad, functional cover classes to species identification data collected along long walking transects used in freshwater wetland assessment (Kutcher and Forrester 2018). This method has not previously been used for salt marsh monitoring, but its use is appealing because it represents a potentially very rapid approach for assessing species cover. Braun–Blanquet includes 5 cover classes and it is sometimes difficult or time-consuming for different observers to place the cover of a species into the same class (Leps and Hadicova 1992). With only 3 classes, it should be theoretically easier with FQA to assign the correct class for a given species in a plot, but at a potential cost of reduced accuracy. Although the FQA method was not intended to be used to detect changes across time or space, our results show that the FQA actually performed very similarly to the point-intercept and Braun–Blanquet in most practical analyses, but not when assessing long-term changes in plant cover in individual plots; this result is likely a result of the poor resolution and reduced accuracy in FQA cover estimates for a plot, which reduces statistical power. We did not quantify how long it took to conduct each method at each plot, but in general we feel that it took about the same amount of time to assess a single plot with Braun–Blanquet and FQA (likely in part due to the simple community structure and low richness in our salt marshes). These factors combine to suggest that if a categorical assessment is to be used for monitoring salt marsh vegetation, Braun–Blanquet is a better choice than FQA.

Change detection over time

The ability to detect and quantify long-term changes in salt marsh plant species cover is key for helping identify dominant stressors to marshes (Raposa et al. 2017) and developing appropriate management strategies (Wigand et al. 2017). At the scale of an entire marsh, our results show that the same general patterns in cover over time are clearly apparent regardless of the monitoring method used; each should be equally powerful for visualizing dominant long-term changes. However, in other cases it may be necessary to track changes at the scale of an individual monitoring plot; in plot-based replicated field experiments for example, or for identifying variability in stress response within a marsh. In these cases, we again found generally good agreement between point-intercept and Braun-Blanquet time series data. However, FQA detected fewer overall changes over time than the other 2 methods and it did not agree with the other methods as consistently as the others agreed with each other. From these results, we consider all 3 methods appropriate for detecting change over time at the whole marsh scale, but recommend point-intercept and Braun-Blanquet for detecting temporal change at individual plots.

Accuracy of each method at the plot scale

The limited ability of FQA to detect change over time in a plot may be directly related to its reduced resolution stemming from including only 3 broad cover classes. In fact, Braun–Blanquet has a similar though less extreme issue with resolution when estimating species cover in a plot at a single point in time. In many cases, and in contrast to point-intercept, both categorical methods are unable to evaluate the true cover of a species in a plot. Consider a species whose actual percent cover in a plot is 28%. Depending on how many intercept points are used in the plot, it is at least possible to achieve this exact cover using point-intercept (i.e., if it hits 14 points when 50 intercepts are used, or 28 points if 100 intercepts are used). For Braun–Blanquet, the best estimate will be either 15.5% or 38% (the midpoints of class 2 and 3, respectively), and for FQA, it is likely that this species would be given a score of 2, translating to a midpoint of 35.5% cover. In this example, it is not possible for either of the categorical cover class methods to get closer than ~7–10% of true plant cover. In contrast, if a species has an actual cover that approximates the midpoint for one or both visual methods (e.g., 36%), all 3 methods should perform similarly. However, to increase the chances of accurately estimating the cover of all species in each individual plot, we recommend the point-intercept over the categorical methods due to its higher resolution.

The reduced resolution and accuracy of the categorical methods, especially FQA, may also hinder their ability to detect correlations between vegetation cover and other parameters in a marsh, which is a technique often used in exploratory field surveys. Consider another hypothetical example where vegetation cover and a potential stressor (e.g., crab abundance) are quantified at 20 plots in a marsh. In this example, actual vegetation cover in a plot varies from 15% to 60% and there is a strong negative correlation between cover and crab abundance. The point-intercept and Braun–Blanquet methods again have the potential to detect this relationship, but not FQA, because vegetation cover in all 20 plots would be static at 35.5%.

Quantifying bare ground cover with different methods

The common cover type that varied most among sampling methods was bare ground. Bare ground cover represented 2 of the 4 highest differences in cover between methods (and 4 of the top 11; Table 1) and differed more than S. alterniflora and S. patens when contrasting change detection over time among methods. The amount of bare ground can be a useful indicator of stress to a marsh, particularly from sea-level rise and excessive inundation (Ganju et al. 2017, Raposa et al. 2017), and it should be a core component of monitoring programs concerned with these issues. Why did bare ground cover differ so much between methods? Unlike live vegetation, bare ground cover is potentially much more subjective among observers. Using point-intercept, for example, while the rod either hits a live plant or not, without proper and consistent training, different observers may interpret bare ground in different ways. Some might only record bare ground if the intercept rod does not otherwise hit any live vegetation. Others may record a point as bare if the rod hits bare substrate, regardless of what it encountered above. For visual methods, the amount of bare ground could simply vary depending on the angle that different people observe the plot. Clearly, steps must be taken to minimize potentially large differences in bare ground cover between observers and methods, especially when this cover type will be used to help select management and intervention strategies for degraded marshes.

Study context and limitations

The results of our study have direct applicability to salt marshes in southern New England, and nearby regions with similar vegetation communities (e.g., northern New England, Long Island Sound, and mid-Atlantic). They may also be generally applicable to other regions with different species and more complex assemblages, but comparisons like ours should be repeated elsewhere to assess their general applicability. Our study was also limited because we used only 2 people for conducting surveys, and we assessed only 3 field methods. Both people that conducted the surveys in this study are experienced in collecting vegetation data in southern New England marshes; data collected by less-experienced personnel may differ to a larger degree, even for the more objective point-intercept method (K.B. Raposa, pers. observ.). In addition, other field-based methods could also be chosen for quantifying vegetation composition (e.g., line-intercept methods; Elzinga et al. 1998), while remote-sensing techniques may be more appropriate for mapping and quantifying vegetation composition, distribution, and extent at multiple scales and resolutions (e.g., Campbell et al. 2017, Gilmore et al. 2009, Judd et al. 2007). However, our results clearly provide insight into the effectiveness of 3 practical field methods for assessing marsh vegetation in general, and specifically in marshes spanning the northeastern US.

Conclusions and recommendations

Our overall recommendation is that the quantitative point-intercept method should be used for assessing marsh vegetation when time and resources allow, especially when the focus is on plot-scale analyses. The expanded use of this method will eventually lead to improved data comparability and produce long-term datasets with more accuracy. Even though the point-intercept is considered to be highly objective, we caution that it is still possible for sizable differences in cover to arise among different observers; inter-calibrations among observers should therefore be carried out when possible to allow for better data comparability. We also note that categorical estimation methods are also effective in many situations, and in fact may be more appropriate than point-intercept in some cases (e.g., when vegetation canopies are very tall or for existing programs that already use categorical methods). If a categorical assessment method is to be used, we recommend the Braun–Blanquet over the FQA. Braun–Blanquet has a higher resolution and, from our experience, takes a similar amount of time per plot to use. Our results can be used to help researchers decide which marsh vegetation sampling method is most appropriate for their needs and goals and should therefore ultimately lead to improved data comparability among programs and help advance meta-analyses and large-scale syntheses. Ultimately, however, the choice of sampling method depends on the questions being asked and the resources available.