Estuaries and Coasts

, Volume 35, Issue 1, pp 23–46

Integrating Scales of Seagrass Monitoring to Meet Conservation Needs

Authors

    • US Geological SurveyPatuxent Wildlife Research Center
  • Blaine S. Kopp
    • US Geological SurveyPatuxent Wildlife Research Center
    • Kimball Union Academy
  • Bradley J. Peterson
    • School of Marine and Atmospheric SciencesStony Brook University
  • Penelope S. Pooler
    • National Park Service, Northeast Coastal and Barrier NetworkUniversity of Rhode Island Coastal Institute in Kingston
Article

DOI: 10.1007/s12237-011-9410-x

Cite this article as:
Neckles, H.A., Kopp, B.S., Peterson, B.J. et al. Estuaries and Coasts (2012) 35: 23. doi:10.1007/s12237-011-9410-x

Abstract

We evaluated a hierarchical framework for seagrass monitoring in two estuaries in the northeastern USA: Little Pleasant Bay, Massachusetts, and Great South Bay/Moriches Bay, New York. This approach includes three tiers of monitoring that are integrated across spatial scales and sampling intensities. We identified monitoring attributes for determining attainment of conservation objectives to protect seagrass ecosystems from estuarine nutrient enrichment. Existing mapping programs provided large-scale information on seagrass distribution and bed sizes (tier 1 monitoring). We supplemented this with bay-wide, quadrat-based assessments of seagrass percent cover and canopy height at permanent sampling stations following a spatially distributed random design (tier 2 monitoring). Resampling simulations showed that four observations per station were sufficient to minimize bias in estimating mean percent cover on a bay-wide scale, and sample sizes of 55 stations in a 624-ha system and 198 stations in a 9,220-ha system were sufficient to detect absolute temporal increases in seagrass abundance from 25% to 49% cover and from 4% to 12% cover, respectively. We made high-resolution measurements of seagrass condition (percent cover, canopy height, total and reproductive shoot density, biomass, and seagrass depth limit) at a representative index site in each system (tier 3 monitoring). Tier 3 data helped explain system-wide changes. Our results suggest tiered monitoring as an efficient and feasible way to detect and predict changes in seagrass systems relative to multi-scale conservation objectives.

Keywords

SeagrassMonitoringMulti-scaleEelgrassMeasurable attributesSampling design

Introduction

Seagrass meadows are noted for their extremely high productivity and contribution of valuable ecosystem functions and services to the coastal zone (Hemminga and Duarte 2000). Worldwide seagrass declines have been caused by anthropogenic impacts, including direct physical disturbance and indirect effects of watershed development and consequent water quality degradation (Orth et al. 2006a; Waycott et al. 2009). It is currently estimated that at least 29% of the global seagrass habitat has disappeared since the late 1800s, and rates of decline and the total area of seagrass lost have risen dramatically in recent decades: the median rate of seagrass loss worldwide increased from <2% per year before 1960 to 7.7% per year in the 1990s (Waycott et al. 2009). Recognition of these extensive losses has led to a concomitant proliferation of seagrass monitoring efforts around the world (Duarte et al. 2004). Paradoxically, however, seagrass monitoring programs often fail to detect declines until substantial losses have occurred, at which point management interventions may be ineffective or too costly to implement (Duarte 2002). Therefore, the need for improved monitoring approaches is paramount to seagrass conservation.

Given increasing demands for seagrass conservation and limited resources for monitoring, monitoring programs must be as efficient as possible. Several characteristics of environmental monitoring are associated with improved efficiency and effectiveness. First, monitoring programs should be guided by models of system responses to natural forces and management actions, with monitoring efforts targeting the precise information needed to make conservation decisions (Nichols and Williams 2006). The results of such targeted monitoring will help discriminate among hypotheses about stresses on ecosystem integrity, evaluate the effectiveness of management remedies, and identify future management needs (Nichols and Williams 2006). Second, monitoring should be capable of detecting change in the resource with a high degree of power (Urquhart et al. 1998; Elzinga et al. 2001). At a minimum, monitoring must detect resource declines before losses are irreversible; preferably, monitoring results should predict changes that can be prevented by management actions (Dale and Beyeler 2001). Finally, the monitoring design must be feasible to execute in terms of methods of data collection, sampling logistics, and monetary costs (Dale and Beyeler 2001; Kurtz et al. 2001).

The core of efficient and effective environmental monitoring is a set of clear conservation objectives and the attributes measured to evaluate whether these objectives are achieved (Dale and Beyeler 2001; Keeney and Gregory 2005). Seagrass conservation goals may be broadly defined in terms of protecting ecosystem integrity, biodiversity, or critical habitat (ASMFC 1997; Pulich 1999; Borum et al. 2004; Kenworthy et al. 2006), or they may focus on specific seagrass restoration targets (Short et al. 2000; Virnstein and Morris 1996; Orth et al. 2002a), protective water quality criteria (Dennison et al. 1993; Johansson and Greening 2000; Wazniak et al. 2007), or minimizing certain anthropogenic impacts (e.g., dredging: Long et al. 1996; Sheridan 2004; recreational boating: Sargent et al. 1995; Fletcher et al. 2009; commercial fishing: Stephan et al. 2000; Orth et al. 2002b). Explicitly or implicitly, seagrass conservation and management objectives frequently span spatial scales, such as maintaining seagrass areal extent, shoot density, and primary productivity (Virnstein 2000). A major challenge associated with seagrass monitoring thus lies in identifying a suite of monitoring attributes that together are sufficient for evaluating progress toward achieving conservation goals at multiple scales yet are feasible to measure.

Existing seagrass monitoring programs around the world address questions at different scales. Periodic mapping based on remotely sensed data is used widely to provide information on the location, sizes, and structure of seagrass beds over broad geographic areas (e.g., Short and Burdick 1996; Kendrick et al. 2000; Kurz et al. 2000; Handley et al. 2007; Orth et al. 2010), often adhering to standard procedures for the acquisition and interpretation of aerial photographs (Finkbeiner et al. 2001). Higher resolution monitoring of seagrass condition has been implemented at the scale of entire bays and estuaries (e.g., Durako et al. 2002; Fourqurean et al. 2002; Dowty et al. 2005; Virnstein and Morris 1996; Avery 2000; Corbett et al. 2005) and individual seagrass beds (e.g., Neverauskas 1987; Deis 2000; McKenzie et al. 2003; Morris and Virnstein 2004; Short et al. 2006a). Although survey methods depend on program goals and site characteristics, most seagrass monitoring at this resolution incorporates quadrat- or transect-based measurements using direct observations or underwater videography, with sampling locations guided by probability designs to allow valid inferences regarding the entire area of interest. Monitoring such characteristics as percent ground cover, shoot density, and areal biomass at this level provides information relevant to population-scale management objectives.

A hierarchical approach to monitoring can accommodate conservation objectives at a range of scales while maximizing efficiency and forecasting capabilities. A three-tier framework has been proposed for coordinated monitoring of environmental resources at multiple spatiotemporal scales (NSTC 1997; Bricker and Ruggiero 1998). Conceived originally to facilitate integrated monitoring and assessments on a national scale, the framework is equally adaptable to individual ecosystems. Conceptually, the three tiers of monitoring are represented by layers of a triangle, with the spatial extent of monitoring decreasing from the base to the vertex. Tier 1 monitoring, at the base of the triangle, characterizes a few ecosystem properties simultaneously over very large spatial scales, typically using airborne or satellite remote sensing methods. Monitoring at this scale is useful to quantify resource extent and distribution over large regions. Tier 2 monitoring addresses specific environmental issues or ecosystem properties at a higher resolution over large geographic areas, generally using ground-based approaches. Monitoring of a limited number of characteristics can occur at a large number of sites. Data collected at this tier can be used to quantify stressor/response relationships and produce estimates of the ecological condition of resources over broad areas, or the quality of the system as a function of physical, chemical, or biological parameters (cf. Brooks et al. 2004). However, tier 2 data are generally insufficient for developing predictive capabilities. Tier 3 monitoring, at the vertex of the triangle, addresses a greater number of properties at a much smaller number of locations or index sites. Intensive monitoring of drivers of change, ecosystem responses, and ecological processes at tier 3 focuses on diagnosing causal relationships. Critical to this framework is the integration across tiers of monitoring through nested site selection, nested attribute sets, modeling, and research. In this way, changes in resource extent or condition detected at tiers 1 and 2 can drive process-based investigations at tier 3, and statistical and explanatory models built on tier 3 and tier 2 data can be used to interpret and predict resource patterns and condition at larger scales (Bricker and Ruggiero 1998).

Hierarchical monitoring offers an efficient means of detecting and predicting change in seagrass ecosystems within the context of multi-scale conservation goals. Recent advances in monitoring and assessment of inland wetlands have focused on a hierarchy of methods varying in scale of application and intensity of effort (Brooks et al. 2004; Kentula 2007; Reiss and Brown 2007). In contrast, although the need for multi-scale approaches to seagrass monitoring has been promoted (Lee Long et al. 1996; Virnstein 2000) and individual seagrass monitoring efforts have spanned spatial scales (e.g., Lee Long et al. 1996; Pergent-Martini and Pergent 1996; Robbins and Bell 2000; Fourqurean et al. 2001; McDonald et al. 2006; Foden 2007), tiered approaches to seagrass monitoring have not been formalized. Within the hierarchical framework described by Bricker and Ruggiero (1998), monitoring at tiers 1 and 2 can provide information on the extent, distribution, and system-wide condition of seagrass resources. Tier 3 monitoring at sites nested within tiers 1 and 2 can provide detailed information on submeter- and shoot-scale processes and responses to a range of potential stressors. This intensive monitoring is needed to understand the mechanisms and causes of change detected at larger scales. For example, through combined scales of monitoring, Fourqurean et al. (2001) were able to link seagrass distribution with seasonal patterns of water temperature and extrapolate seagrass standing crop to a broad region.

We tested a hierarchical approach to monitoring as a logistically and economically feasible means of evaluating progress toward seagrass conservation objectives within two lagoonal estuaries in the northeastern USA—Pleasant Bay, Massachusetts, and the Long Island South Shore Estuary, New York—that differ considerably in size and complexity. The focus of our study was seagrass beds associated with National Park units in these systems, Cape Cod National Seashore in Massachusetts and Fire Island National Seashore in New York. The densest human population in the USA is found in this northeastern region, where the most pervasive threat to estuarine resources is reduced water quality stemming from watershed development (Roman et al. 2000; Short and Short 2003). Much of the watershed area of National Park coastal ecosystems lies outside protective park boundaries and is subject to intense pressures from residential, agricultural, and urban expansion. Land clearing, fertilizer production and application, discharge of sewage and septic systems, and fossil fuel combustion associated with human population growth have accelerated nitrogen and phosphorus loading to the world’s coastal zone since the 1950s (Nixon 1995; Cloern 2001). Consequently, the fundamental conservation goal for National Park estuaries in this densely populated region is to protect estuarine resources from excessive nutrient loads resulting from watershed development (Kopp and Neckles 2009). Given worldwide evidence relating estuarine nutrient enrichment to seagrass declines (Duarte 1995; Borum 1996), this broad conservation goal includes specific conservation objectives for seagrasses: (1) to maximize seagrass distribution and (2) to maximize seagrass growth and production. These seagrass conservation objectives are explicitly multi-scale in nature.

Seagrass mapping is conducted in both study estuaries by state programs at approximately 5-year intervals (MassDEP 2008; NYS Seagrass Task Force 2009). These existing efforts provide tier 1 monitoring information on large-scale changes in seagrass distribution. Our aim was to develop an approach for integrating monitoring of seagrass condition throughout entire systems (tier 2) and at high-resolution index sites (tier 3) in order to improve abilities for evaluating and anticipating threats to seagrass resources associated with estuarine nutrient enrichment. Importantly, to ensure long-term implementation, the approach must be executable within the limited staffing and funding capacity of park management; thus, our focus was on developing an approach that was both scientifically robust and economically practicable.

The primary mechanism linking increased nutrient loading to seagrass declines is attenuation of light through the water column and at leaf surfaces by fast-growing phytoplankton, epiphytic microalgae, and ephemeral macroalgae (Sand-Jensen and Borum 1991; Duarte 1995). The depth limit of seagrass meadows is inversely related to water column nitrogen concentration (Sand-Jensen and Borum 1991; Borum 1996) and light attenuation (Duarte 1991; Dennison et al. 1993), and seagrass responses to altered light availability associated with reduced water quality will be most evident at the deepest locations of existing beds (Dennison and Alberte 1982). Therefore, we expected that high-resolution monitoring of seagrass attributes along a depth gradient (tier 3) would be helpful in elucidating monitoring data on seagrass condition throughout entire bays (tier 2) and forecasting system-wide changes. Given the prevalence of water quality degradation as a primary threat to global seagrass survival (Short and Wyllie-Echeverria 1996; Waycott et al. 2009), this approach to seagrass monitoring should be broadly applicable.

Study Location

Pleasant Bay, Massachusetts

The Pleasant Bay estuary is the largest coastal embayment on Cape Cod, Massachusetts (Fig. 1). The estuary receives freshwater from northern and western shore tributaries and direct groundwater discharge. A narrow barrier beach forms the eastern shore of Pleasant Bay, and the estuary is connected to the Atlantic Ocean through two tidal inlets at its southern end (Fig. 1). The northernmost inlet formed in April 2007, during the course of our investigation, and has had substantial impact on the hydrodynamics in the estuary. The Pleasant Bay system encompasses approximately 2800 ha with an average depth of 1.8 m below mean low water (MLW) and a mean tidal range of approximately 1.5 m. The estuarine system is characterized by broad intertidal and subtidal flats, a narrow channel, and deeper basins in the lower bay and some tributary ponds. Sediments are primarily medium to very fine sand (Psuty and Silveira 2009).
https://static-content.springer.com/image/art%3A10.1007%2Fs12237-011-9410-x/MediaObjects/12237_2011_9410_Fig1_HTML.gif
Fig. 1

Location of study sites in Little Pleasant Bay, Massachusetts, showing tier 1 data on seagrass coverage in 2001 (MassDEP 2008), the tier 2 sampling frame, and the tier 3 monitoring location (star). “Zones” refer to distance zones used in the tier 2 sampling design

The greatest extent of seagrass in the Pleasant Bay estuary is found in Little Pleasant Bay (LPB), which occupies the estuary’s shallow upper basin (Fig. 1). Eelgrass (Zostera marina L.) is the only seagrass in this system. The eastern half of the eelgrass meadow falls within the boundary of Cape Cod National Seashore and is of primary management interest. The main sources of nitrogen loading to the Pleasant Bay estuary are wastewater, fertilizer, runoff from impervious surfaces, and direct atmospheric deposition (Howes et al. 2006). Although nitrogen loading rates are lower than those of other estuaries in the northeastern USA (Carmichael et al. 2004), they have increased substantially over the past few decades due to changing land use patterns in the watershed (Howes et al. 2006). Eelgrass coverage in the Pleasant Bay system declined by 24% from the 1950s to the present (Howes et al. 2006). Retraction of vegetation from the upper reaches of tributaries during this time period is consistent with concurrent patterns of nutrient enrichment in the system (Howes et al. 2006). Our sampling frame for tier 2 monitoring of seagrass condition included all of LPB and the upper tributaries and was 624 ha in size. In 2001, there were 301 ha of eelgrass within the boundaries of our sampling frame (MassDEP 2008). We established a tier 3 monitoring site in the eastern portion of LPB off the shore of Hog Island in an eelgrass bed that was easily accessible; that spanned the depth range of eelgrass in LPB; and was typical for the system in terms of eelgrass density, morphology, and substrate characteristics (Fig. 1).

Long Island South Shore Estuary, New York

The Long Island South Shore Estuary is a series of interconnected bays bounded by the southern coast of Long Island, New York, and outer barrier islands. Our study was within the Great South Bay (GSB) and Moriches Bay (MB) portion of South Shore Estuary bordered by Fire Island, from Fire Island Inlet on the west to Moriches Inlet on the east (Fig. 2). The GSB/MB system stretches approximately 50 km between the two inlets. Freshwater input is primarily from Long Island tributaries and groundwater seepage. The average depth of GSB is 1.3 m (MLW) and of MB is 0.9 m (Wilson et al. 1991); both basins have tidal channels associated with the inlets, but most of the area is <1 m deep (MLW). The tidal range at the Fire Island Inlet is approximately 1.25 m, but it decreases to 0.45 m at a point equidistant from Fire Island Inlet and Moriches Bay (Wong 1981), and sediments are primarily medium to very fine sand (Psuty and Silveira 2009).
https://static-content.springer.com/image/art%3A10.1007%2Fs12237-011-9410-x/MediaObjects/12237_2011_9410_Fig2_HTML.gif
Fig. 2

Location of study sites in Great South Bay and Moriches Bay, New York, showing tier 1 data on seagrass coverage in 2002 (NYS Seagrass Task Force 2009), the tier 2 sampling frame, and the tier 3 monitoring location (star). “Zones” refer to distance zones used in the tier 2 sampling design

The boundary of Fire Island National Seashore extends from the bayside shoreline of the island about 1.2 km into the estuary. The majority of the seagrass in GSB/MB is found within the park boundary (Fig. 2) and includes both eelgrass and widgeon grass (Ruppia maritima). The most significant sources of nitrogen to the system are runoff from impervious surfaces and wastewater (NYS DOS 2001). There is a gradient of decreasing water quality with distance from the inlets (Weiss et al. 2007). Eelgrass in GSB declined in coverage by 23% from 1967 to 1977, presumably in response to reduced light availability associated with eutrophication (Dennison et al. 1989). Although changing land use practices have resulted in declining water column nitrogen concentrations since the mid-1970s (Gobler et al. 2005; Hinga 2005), harmful brown tides (Aureococcus anophagefferens) have recurred since 1985 (Gobler et al. 2005). Dennison et al. (1989) estimated that the shading effect of brown tide blooms from 1985 to 1988 caused an additional 40% reduction in eelgrass coverage in GSB from pre-bloom conditions. Our sampling frame for tier 2 monitoring extended 1 km outside the seashore boundary (Fig. 2) and was 9,220 ha in size. In 2002 there were 2,760 ha of seagrasses within the limits of our sampling frame (NYS Seagrass Task Force 2009; Fig. 2). We established a tier 3 monitoring station in eastern GSB off of Fire Island, in the portion of the bay with the most continuous mapped seagrass distribution (Fig. 2). As in LPB, the exact location of the tier 3 station was selected to be accessible for monitoring and representative of the seagrass beds throughout GSB/MB in terms of depth range and plant and substrate characteristics.

Materials and Methods

Measurable Attributes

To ensure that monitoring programs target the information needed to make conservation decisions, monitoring attributes must be linked explicitly to conservation objectives (Nichols and Williams 2006). We identified attributes of the seagrass ecosystem for evaluating the attainment of conservation objectives associated with seagrass distribution and seagrass condition (Table 1). The attributes spanned scales of monitoring and incorporated indicators of ecological structure, composition, and function (cf. Dale and Beyeler 2001; Evans and Short 2005) that are sensitive to light limitation and thus relevant to management concerns. For example, the correlation between increased nutrient loading and declines in seagrass extent (Sand-Jensen and Borum 1991; Duarte 1995; Latimer and Rego 2010) and population parameters such as shoot density, percent cover, and biomass (Neckles et al. 1993; Short et al. 1995; Taylor et al. 1995; Nixon et al. 2001; Hauxwell et al. 2003) are well documented. Collectively, the variables we selected are among the most commonly measured characteristics in seagrass monitoring programs (Duarte et al. 2004; Krause-Jensen et al. 2004).
Table 1

Seagrass attributes measured at each tier of monitoring

Conservation objective

Measurable attribute

Tier 1

Tier 2

Tier 3

Maximize seagrass areal distribution

Bed size

x

  
 

Depth limit

  

x

Maximize seagrass growth and reproduction

Percent cover

 

x

x

 

Canopy height

 

x

x

 

Total shoot density

  

x

 

Reproductive shoot density

  

x

 

Biomass

  

x

Tier 1 monitoring is provided by existing seagrass mapping programs; tier 2 and tier 3 attributes were tested in this study

Information on seagrass bed size is already gathered by existing tier 1 mapping programs based on aerial photographs acquired at a scale of 1:20,000; our study focused on testing attributes at tiers 2 and 3. The time needed for seagrass sampling depends greatly on whether in situ observations and plant harvesting are required, with sampling time reduced by avoiding need for observations by SCUBA and for processing large quantities of plant material (cf. Jackson and Nemeth 2007). Therefore, to minimize sampling time and associated costs, attributes monitored at tier 2 were limited to variables that could be measured either from a boat or by using snorkel. Methods for sampling at tier 2 and tier 3 scales of monitoring are described separately below.

Tier 2 Monitoring

Sampling Design

We used a restricted random sampling design (Elzinga et al. 2001) to select sampling stations in LPB and GSB/MB. In general, restricted random designs are characterized by first dividing the area to be monitored into equal sized contiguous segments and then randomly positioning a single sampling location within each segment. This approach ensures good dispersion of sampling points through the entire sampled population while incorporating randomization in sample placement. We used a randomized-tessellation stratified design (Stevens 1997) in which a grid of tessellated hexagons serves as the basis for locating random sampling stations. An alternative method for maintaining a spatially well-balanced random sample is the generalized random-tessellation stratified design (Stevens and Olsen 2004), which has more flexibility for incorporating strata with different sampling intensities.

The LPB and GSB/MB sampling frames were partitioned into tessellated hexagonal cells (Figs. 3 and 4) and a random sampling point (set of latitude and longitude coordinates) was selected within each cell. The hexagonal cells in LPB were each 500 m wide and the cells in GSB/MB 925 m wide. A very small fraction of the original cells could not be monitored because they fell entirely on land. The sample sizes were thus defined by the number of attainable cells in each system: 55 in LPB and 198 in GSB/MB. Sampling points were categorized within equal-width distance zones in each bay that represented varying distances from flushing sources. Distance zones in LPB (n = 3) were 1.5-km bands of decreasing distance from the mouth of the bay (Fig. 1) and in GSB/MB (n = 4) were 11.5-km bands along the axis of Fire Island between Fire Island Inlet to the west and Moriches Inlet to the east (Fig. 2). Sampling points were permanent to maximize the ability to detect changes in condition over time (Elzinga et al. 2001).
https://static-content.springer.com/image/art%3A10.1007%2Fs12237-011-9410-x/MediaObjects/12237_2011_9410_Fig3_HTML.gif
Fig. 3

Grid of tessellated hexagons used as the basis for locating sampling stations in Little Pleasant Bay. One random point was selected as a sampling station within each hexagonal cell

https://static-content.springer.com/image/art%3A10.1007%2Fs12237-011-9410-x/MediaObjects/12237_2011_9410_Fig4_HTML.gif
Fig. 4

Grid of tessellated hexagons used as the basis for locating sampling stations in Great South Bay and Moriches Bay. One random point was selected as a sampling station within each hexagonal cell

We sampled tier 2 attributes (percent cover, canopy height; Table 1) in midsummer at the time of maximum eelgrass leaf biomass in the region (Wilson and Brenowitz 1966; Roman and Able 1988). Sampling was conducted during three consecutive years in LPB (2006–2008) and 2 years in GSB/MB (2007, 2009). During sampling, we navigated to sampling points using GPS with ≤4-m positional accuracy. A sampling station was defined as a 10-m diameter circle around the GPS point, determined by the length of the boat plus the potential error in reaching the precise coordinates. At each station, we collected four subsample observations within quadrats located haphazardly off the four quarters of the boat. In addition, during the 2007 only sampling at LPB, the number of subsamples at six randomly selected test stations was increased from 4 to 12. The minimum distance between the 12 subsamples at each test station was 3 m. These data were used for simulations to test the subsampling intensity necessary to characterize the 10-m diameter sampling station.

Field Measurements

We measured tier 2 attributes once per subsample site during each sampling interval. We estimated seagrass percent cover by species within a 0.25-m2 quadrat by vertical observation. Cover measurements in LPB were made exclusively from the boat using a variety of methods, depending on water depth and surface conditions; at a given station, we used the easiest method that allowed a clear view of the substrate. At shallow sites (approx. <60 cm deep), the substrate was clearly visible directly through the water, and at mid-depth sites (approx. 60 cm–1.25 m), we used a view scope to penetrate the water surface; in these instances, the sampling quadrat was a stainless steel frame attached to a rope. At deeper sites (maximum of about 6 m), we viewed the substrate using an underwater video camera mounted directly above an aluminum sampling frame; the frame was lowered to the substrate on a rope. In GSB/MB, we used the same boat-based methods where visibility allowed, but low water clarity demanded snorkeling for accurate measurements at about 40% of the stations. Cover estimates were standardized based on the photographic reference guide published in Short et al. (2006b) and inter-observer training and calibration (Krause-Jensen et al. 2004). Cover from 5% to 100% was recorded in 5% increments, and the presence of very few plants (<5% cover) was assigned a value of 1% cover. Canopy height (defined as the height of 80% of the leaf material of the dominant species, ignoring the tallest 20% of the leaves; Duarte and Kirkman 2001) was measured on samples of three large shoots collected with a rake on a telescoping handle (LPB) or by hand while snorkeling (majority of stations at GSB/MB). Water depth and time were recorded at each station and corrected to MLW using predicted tide heights.

Statistical Analysis of Status and Trends

We used the inverse distance-weighted (IDW) method of spatial interpolation to reveal bay-wide patterns in seagrass status from measurements of percent cover (Watson and Philip 1985). This is a simple method of deterministic interpolation that estimates the value of unsampled points as the weighted average of values from a given number of surrounding points, giving more weight to closest points. IDW assumes a functional form for the weights based on the inverse of the Euclidean distance, but does not assume data characteristics that are required for probabilistic interpolation methods (e.g., kriging) such as a regionally constant mean and a specified spatial covariance structure (Davis 2002). Given the highly irregular shoreline of LPB, we used an impermeable boundary during IDW processing to control inappropriate influences of data points across separate branches of the system. The impermeable boundary was created as a 100-m buffer around the shoreline, with manual adjustments to further isolate kettle hole salt ponds. Interpolated values were categorized within Braun-Blanquet cover classes for display (Elzinga et al. 2001), modified by combining all values ≤5% cover into a single class. Spatial interpolation was done using Spatial Analyst in ArcGIS 9.1.

Spatial interpolation of cover data yields snapshots of seagrass status, but does not provide statistical comparisons of condition between years. Therefore, we used general linear models to make quantitative comparisons over time. We used a repeated measures analysis of variance to test the null hypothesis that percent cover did not change over time and that patterns of response did not differ among distance zones, treating year and distance from flushing (distance zone) as fixed effects, and treating station location as a random effect (Green 1993; Schabenberger and Pierce 2002). Depth was included as a fixed effect in LPB (less than or greater than 1 m MLW), but was not included in the GSB/MB model due to the fairly homogeneous depths in that system. We used various methods to ensure that the data met the assumptions of the analysis. We examined the potential effect of spatial autocorrelation by developing variograms for within-year data sets and evaluating the fit of candidate spatial covariance structures to the data (Cressie 1993). We then compared the likelihood of within-year models specifying these spatial covariance structures (exponential, Gaussian, linear, spherical, and power) to that of a null model (i.e., with no spatial covariance structure; Cressie 1993; Littell et al. 2006). In each case, the inclusion of a spatial covariance structure failed to improve the fit of the model. We used Mauchley’s criterion to verify the compound symmetry of the covariance structure among repeated measures (Potvin et al. 1990) and residual analysis to confirm the overall aptness of the model (Kutner et al. 2005). Variograms were developed using the VARIOGRAM procedure, and data analysis models were developed using the MIXED procedure of the SAS statistical software package (SAS 9.2). Sample means were reported to the nearest 5% cover to match the degree of precision of the original data.

Sample Size Simulations

Using percent cover data collected in 2007 from the six LPB test stations with 12 subsamples each, we performed resampling simulations to determine the within-station sampling intensity (i.e., the number of subsample observations) necessary to achieve high accuracy in estimating the station mean. For each of the six test stations, we drew sets of 10,000 random samples of different sizes (n = 3–11 subsamples) from the pool of 12 subsamples. We used these simulated samples of different sizes to examine the absolute bias associated with estimates of mean percent cover within a test station, or the difference between the sample mean and the true test station mean. The probability of achieving certain levels of bias at individual test stations for each within-station number of subsamples (n = 3–11) was determined from the frequency distribution of the means of 10,000 samples of size n around the true test station mean (n = 12). The probability of achieving certain levels of overall bias for each within-station number of subsamples was determined from the frequency distribution of 10,000 overall means (where the overall mean = the mean of six station means derived from n = 3–11 subsamples) around the grand mean (n = 72, based on 12 subsample observations at each of six test stations).

We examined the influence of the bay-wide sample sizes in LPB and GSB/MB on detecting change in percent cover by simulating sample sets using increased hexagon cell size, resulting in corresponding decreases in the number of stations sampled throughout the bay. We simulated increases of two, three, and four times the original hexagon size by combining adjacent hexagons within distance zones into larger areas and then randomly selecting one of the original groups of subsample observations (i.e., four subsamples from a single station) to represent the data from the simulated larger hexagon. We created multiple simulated data sets for each “new” hexagon size by using different groups of subsamples. For example, for hexagons twice the original size, we randomly divided the set of original subsample groups into two simulated data sets, each using one half of the sample stations. Similarly, we created three simulated data sets for hexagons three times the original size and four simulated data sets for hexagons four times the original size. Simulated data sets were generated for LPB percent cover data in 2006 and 2007 and GSB/MB in 2007 and 2009. We then applied the same repeated measures analysis of variance model to the simulated data sets that was applied to the full data set within each system and compared the results generated using the decreased sample sizes to those generated using the original sample size.

Tier 3 Monitoring

Sampling Design

We established intensive monitoring sites in areas of fairly homogeneous seagrass cover that were broadly representative of conditions in LPB and GSB/MB in terms of seagrass species composition, depth gradient, and sediment characteristics. For tier 3 monitoring, we followed the spatial design of the global seagrass monitoring program SeagrassNet (Short et al. 2006b), although the attributes reported here are a subset of those included in the SeagrassNet protocol. By adopting the SeagrassNet design, our tier 3 data form part of a global monitoring network and can contribute to worldwide assessments of seagrass change. As prescribed by SeagrassNet, each tier 3 site consisted of three permanent 50-m transects established parallel to the shoreline: one transect was near the shallow edge of the seagrass bed, one at a moderate depth, and one at a deep location of the bed (Table 2). We embedded permanent helical screw anchors in the substrate to mark the ends and the middle of each transect (0-, 25-, and 50-m fixed reference points). Twelve permanent 0.25-m2 quadrats were positioned at randomly selected distances along each transect. Sampling was conducted in midsummer during five consecutive years in LPB (2005–2009) and during 2 years in GSB/MB (2007, 2009). At each sampling interval, the precise quadrat locations were determined by stretching temporary fiber glass measuring tapes between the fixed transect markers.
Table 2

Depth of tier 3 monitoring sites in LPB and GSB/MB

Transect

Depth (meters below MLW)

LPB

GSB/MB

Shallow

0.64

0.76

Mid-depth

0.84

0.97

Deep

1.35

1.34

Field Measurements

We tested all tier 3 monitoring attributes (Table 1) at LPB and all but biomass at GSB/MB. We estimated percent cover, measured canopy height, and counted the number of reproductive shoots by seagrass species within the entire 0.25-m2 quadrats by snorkeling (shallow transects) or SCUBA (deep transects). We determined eelgrass shoot density by direct counts of all shoots rooted within the entire quadrat if percent cover was ≤25% or shoot distribution was highly clumped; if percent cover was >25% and shoots were homogeneously distributed, all shoots within a 0.0625-m2 subquadrat were counted. Low visibility in the mixed-species seagrass bed at GSB/MB prohibited accurate counts of the slender, branched widgeon grass stems, so density measurements at GSB/MB were confined to eelgrass.

At LPB, we used an indirect method to determine aboveground and belowground biomass of eelgrass in the permanent quadrats. Biomass samples were harvested using a 0.0035 m2 (6.7-cm diameter) core tube, the size specified by the SeagrassNet protocol. Samples were collected from an area 0.5–1.0 m away from each permanent quadrat with vegetation of the same percent cover and canopy height as in the quadrat. Samples were rinsed with fresh water and stored under refrigeration for processing within 48 h. During processing, only plants with intact meristems were retained. Shoots were cleaned of debris, dead plant material, and epiphytes and sorted into living leaf (including sheaths) and root + rhizome fractions. The number of plants with intact meristems in the biomass sample was recorded and plant material was then dried to constant weight at 60°C and weighed. The average leaf and root + rhizome weight per shoot was determined for each biomass sample and then was applied to shoot density in the respective quadrat to derive the permanent quadrat biomass per unit area (Duarte and Kirkman 2001).

We compared this indirect method of determining biomass in the permanent quadrats with direct measurement by collecting paired test samples at LPB in 2006 (12 paired samples along the mid-depth transect only) and 2007 (five to six paired samples along each of the three transects). We identified sampling locations using a 1-m2 frame divided into 16 quadrats (0.0625 m2). For each sample pair, we placed the frame 1 m away from the transect and randomly selected one 0.0625-m2 test quadrat for sampling. We collected a 0.0035-m2 core from the middle of each 0.0625-m2 test quadrat and then collected all remaining plant material in the quadrat by cutting around the inside of the quadrat to below the root zone. Plant material within both sample fractions (in the central core and remaining in the quadrat) was processed as described previously. Biomass of the 0.0625-m2 test quadrat was then determined indirectly by applying the average weight per shoot from the central core sample to the shoot density of the entire test quadrat (the number of shoots in the core + the number of shoots remaining in the quadrat) and directly (the measured biomass of shoots in the core + the measured biomass of shoots remaining in the quadrat).

We recorded the position of the deep edge of the continuous seagrass meadow as the distance seaward from fixed reference points of the deep transect. We used underwater video with GPS overlay to locate the deep edge of the seagrass bed. The edge of the seagrass bed was defined operationally as the farthest point where shoots were ≤1 m apart (Short et al. 2006b). An underwater video camera was towed behind a boat along offshore trajectories perpendicular to the deep transect beginning at the 0-, 25-, and 50-m transect marks. In LPB, the continuous seafloor image was recorded on a digital videotape stamped with GPS coordinates, depth of the camera, date, time, speed, and heading, and the deep edge of the seagrass bed was identified by viewing the video transect images on a computer monitor. In GSB/MB, the edge of the bed was identified on the video monitor in the field and verified by snorkeling.

Statistical Analysis

Annual quadrat data from PB and GSB/MB were analyzed using a repeated measures analysis of variance with fixed effects of year and depth and random effects of quadrat; this model tested the null hypotheses that seagrass attributes did not change over time or depth and that patterns of response were similar across the depth gradient. We ensured the appropriateness of the statistical models as described for tier 2 trend data, and data were transformed where necessary to correct for non-constancy of error variance. The relationship between indirect and direct methods of biomass determination tested at LPB was quantified with simple linear regression, and a predictive relationship between seagrass biomass, percent cover, and canopy height within the permanent quadrats was developed with multiple regression analysis. Residual plots were used to verify the aptness of all regression models (Kutner et al. 2005).

Results

Tier 2 Monitoring

Little Pleasant Bay

The tier 2 survey of LPB was completed annually in 2–3 days by two to three people. System-wide monitoring of eelgrass condition revealed a change in percent cover following formation of the new inlet. In 2006, the upper tributaries were unvegetated or sparsely vegetated, and eelgrass in the lower bay ranged mostly from 6% to 50% cover (Fig. 5). In 2007 and 2008, the upper tributaries remained sparsely vegetated, but eelgrass in lower LPB ranged mostly from 6% to 75% cover (Fig. 5); thus, after the new inlet formed in April 2007, there was a shift in distribution of observations toward higher cover classes. Trend analysis showed a significant increase in mean percent cover from 2006 to 2007 within the southernmost portion of LPB (Tables 3 and 4, distance zone 3; p < 0.0001). There were no changes in eelgrass cover from 2007 to 2008 (Table 4; within-zone p ranged from 0.44 to 1.0). Depth category had no effect on cover (p = 0.52) and was removed from the model. Average eelgrass canopy height ranged from about 40 cm in areas <1 m deep to 75 cm in deeper areas and was consistent among years.
https://static-content.springer.com/image/art%3A10.1007%2Fs12237-011-9410-x/MediaObjects/12237_2011_9410_Fig5_HTML.gif
Fig. 5

Percent cover of eelgrass (Z. marina) in Little Pleasant Bay, 2006–2008. Seagrass cover was measured at 55 sampling stations using tier 2 monitoring methods and was estimated at unsampled locations using inverse distance weighted spatial interpolation

Table 3

Repeated measures analysis of variance of eelgrass (Z. marina) percent cover measured during tier 2 monitoring in Little Pleasant Bay, 2006–2008

Fixed effect

df

F

p

Year

2, 104

3.67

0.029

Distance

2, 52

14.40

<0.001

Year × Distance

4, 104

5.07

0.001

Table 4

Mean (SE) percent cover of eelgrass (Z. marina) within distance zones of Little Pleasant Bay measured during tier 2 monitoring, 2006–2008

Distance zone

Year

2006

2007

2008

1

0 (0)a

0 (0)a

0 (0)a

2

10 (6)a

10 (6)a

10 (6)a

3

25 (6)a

50 (9)b

45 (8)b

Distance zones are 1.5-km-wide horizontal bands numbered from north to south. Letters refer to comparisons among years within distance zones; values with like letters within a row are not significantly different (p > 0.05). Following statistical analysis, sample means were rounded to the nearest 5% cover to match the degree of precision of the original data

Great South Bay/Moriches Bay

It required 10–12 days for three people to complete the tier 2 survey in GSB/MB. In 2007, eelgrass was densest in the western end of GSB near Fire Island inlet, whereas widgeon grass occurred nearly exclusively in the eastern end of GSB (Fig. 6). In 2009, eelgrass had expanded within eastern GSB (Fig. 6), and both species showed a shift to higher cover classes. The mean percent cover of both species increased significantly within the eastern section of GSB between years (Tables 5 and 6, distance zone 3; eelgrass, p = 0.02; widgeon grass, p < 0.0001), and the mean percent cover of eelgrass also increased significantly within the westernmost section of the bay (Tables 5 and 6, distance zone 1; p < 0.0001). There were no other changes in seagrass cover between years (Table 6; within-zone p ranged from 0.18 to 0.85). Average canopy height of eelgrass-dominated beds was between about 15 and 25 cm in GSB and was 35 cm in MB, whereas average canopy height where widgeon grass dominated was <10 cm; patterns were consistent between years.
https://static-content.springer.com/image/art%3A10.1007%2Fs12237-011-9410-x/MediaObjects/12237_2011_9410_Fig6_HTML.gif
Fig. 6

Percent cover of eelgrass (Z. marina) and widgeon grass (R. maritima) in Great South Bay and Moriches Bay, 2007 and 2009. Seagrass cover was measured at 198 sampling stations using tier 2 monitoring methods and was estimated at unsampled locations using inverse distance weighted spatial interpolation

Table 5

Repeated measures analysis of variance of eelgrass and widgeon grass percent cover measured during tier 2 monitoring in Great South Bay/Moriches Bay, 2007 and 2009

Fixed effect

df

Eelgrass

Widgeon grass

F

p

F

p

Year

1, 194

21.67

0.000

2.48

0.117

Distance

3, 194

10.93

0.000

15.66

0.000

Year × Distance

3, 194

4.51

0.004

5.97

0.001

Table 6

Mean (SE) percent cover of eelgrass and widgeon grass within distance zones of Great South Bay/Moriches Bay measured during tier 2 monitoring, 2007 and 2009

Distance zone

Eelgrass

Widgeon grass

2007

2009

2007

2009

1

15 (3)a

30 (5)b

1 (1)a

1 (1)a

2

1 (1)a

5 (3)a

5 (2)a

5 (2)a

3

5 (2)a

10 (4)b

10 (3)a

20 (4)b

4

5 (3)a

5 (3)a

1 (1)a

0 (0)a

Distance zones are 11.5-km bands along the axis of Fire Island between Fire Island Inlet and Moriches Inlet numbered from west to east. Letters refer to comparisons between years for each species within distance zones; values of the same species with like letters within a row are not significantly different (p > 0.05). To match the degree of precision of the original data, following statistical analysis sample means >2.5% cover were rounded to the nearest 5% and positive means <2.5% were reported as 1

Sample Size Simulations

The number of subsamples, or observations per station, required to achieve a low bias in estimating mean percent cover at a single station varied across the six LPB test stations. At stations with fairly low within-station variability, three to four subsamples yielded a high probability (>80%) that the expected value of the estimator for within-station mean percent cover was within 10 percentage points of the true mean (true mean percent cover ±10 points), regardless of whether mean percent cover was high (Table 7, test stations 2 and 5) or low (Table 7, test station 3). At individual test stations with greater within-station heterogeneity, six to eight subsamples were required to achieve a similar absolute bias (Table 7, test stations 1, 4, and 6). However, a relatively low subsampling effort was still sufficient to achieve a low overall bias in estimating the mean percent cover across all test stations: four subsamples yielded an 80% probability that the expected value of the estimator for overall mean percent cover was within 5 percentage points of the true mean (true mean percent cover ±5 points) and a >99% probability that the expected value of the estimator was within 10 percentage points of the true mean (true mean percent cover ±10 points).
Table 7

Probability of achieving an absolute bias <10 using different numbers of subsamples to estimate mean percent eelgrass cover at tier 2 test stations in Little Pleasant Bay

Station ID

True mean (n = 12)

Standard deviation

Probability

No. of subsamples

3

4

5

6

7

8

9

10

11

1

52.5

23.8

0.617

0.661

0.766

0.840

0.918

0.956

0.990

1

1

2

82.5

17.8

0.705

0.807

0.912

0.938

0.987

0.996

1

1

1

3

4.2

6.3

1

1

1

1

1

1

1

1

1

4

73.8

32.8

0.335

0.455

0.587

0.656

0.738

0.865

0.948

0.984

1

5

100

0

1

1

1

1

1

1

1

1

1

6

76.3

31.8

0.343

0.511

0.643

0.722

0.724

0.903

0.952

0.985

1

The ability to detect temporal changes in seagrass cover within entire sampling frames with decreasing numbers of stations depended on the magnitude of change that occurred. For eelgrass in LPB, analysis of the two simulated data sets including half the original number of sampling stations (n = 26 vs. original n = 55) yielded the same pattern of response as did analysis of the original data set: percent cover increased significantly from 2006 to 2007 within the southernmost distance class (individual simulations: p = 0.001, p = 0.008). The results were inconsistent, however, as the sample size was reduced further. At one third the original number of sampling stations (n = 16), a significant increase in percent cover in the southernmost distance class was detected in two out of three simulations (p = 0.003, p = 0.003, p = 0.20), and at one fourth the original number (n = 13), this increase was detected in two out of four simulations (p < 0.0001, p = 0.004, p = 0.18, p = 0.59). In GSB/MB, the ability to detect temporal changes with reduced sample sizes varied between species. For eelgrass, a significant increase in percent cover from 2007 to 2009 was observed in the westernmost section of GSB (distance zone 1) for both simulations using half the original number of sampling stations (n = 98 vs. original n = 198; individual simulations: p = 0.0005, p = 0.001) and all three using one third the original number (n = 65; p = 0.0009, p = 0.003, p = 0.04). At one fourth the original number of sampling stations (n = 48), this increase was detected in three out of four simulations (p = 0.0003, p = 0.002, p = 0.006, p = 0.75). In the eastern section of GSB, where the absolute increase in eelgrass cover between years was not as great (Table 6, distance zone 3), the change from 2007 to 2009 was marginally to non-significant in simulations using half the original sampling stations (p = 0.06, p = 0.17). A significant increase was detected in one out of three simulations using one third of the original number of sampling stations (p = 0.02, p = 0.14, p = 0.91) and in one out of four simulations using one fourth the original number (p = 0.03, p = 0.10, p = 0.41, p = 0.92). The effect of reducing sample size on the ability to detect an increase in widgeon grass between years in the eastern section of GSB (distance zone 3) varied across simulations, but the significance of the test was always <0.10 (simulations at one half the original sample size: p = 0.0006, p = 0.09; at one third the original sample size: p = 0.01, p = 0.02, p = 0.09; and at one fourth the original sample size: p = 0.02, p = 0.03, p = 0.09, p = 0.09).

Tier 3 Monitoring

Little Pleasant Bay

Annual monitoring of permanent quadrats at the LPB tier 3 site was completed in 1 day by four people. We observed considerable spatial and temporal variability in eelgrass characteristics at this scale (Fig. 7 and Table 8). Within the eelgrass bed, canopy height increased consistently from the shallow to the deep transect (p < 0.0001; Fig. 7b), but the patterns of other attributes across the depth gradient varied among sampling intervals. Therefore, our analysis of monitoring results focused primarily on temporal rather than spatial differences in tier 3 attributes. Results of testing the null hypothesis of no temporal changes in eelgrass attributes within transects are expressed at a statistical threshold of p = 0.05 (Table 8). We also performed all tests using a critical value of p = 0.10 and found the interpretations of the results to be consistent with the more conservative criterion.
https://static-content.springer.com/image/art%3A10.1007%2Fs12237-011-9410-x/MediaObjects/12237_2011_9410_Fig7_HTML.gif
Fig. 7

Shoot density (a), canopy height (b), percent cover (c), and total biomass (d) of eelgrass (Z. marina) at tier 3 monitoring site in Little Pleasant Bay, 2005–2009 (mean ± SE)

Table 8

Differences in tier 3 eelgrass attributes over time in Little Pleasant Bay as determined by repeated measures analysis of variance, 2005–2009

Attribute

Year

2005

2006

2007

2008

2009

Densitya

 Shallow

a

b

c

d

e

 Mid-depth

a

b

ac

d

c

 Deep

a

bc

ab

c

a

Canopy height

 Shallow

a

b

b

bc

c

 Mid-depth

a

b

c

cd

d

 Deep

a

bc

c

d

ab

Percent coverb

 Shallow

a

b

bc

c

d

 Mid-depth

ab

c

b

ab

a

 Deep

a

a

b

c

d

Total biomass

 Shallow

a

b

b

bc

ac

 Mid-depth

a

b

a

c

c

 Deep

a

a

a

b

a

Multiple comparisons are across years within depths. Like letters within a row indicate lack of difference between mean attributes (p > 0.05)

aAnalysis of square root-transformed data

bAnalysis of arcsin transformed data

Along the shallow transect, all eelgrass attributes declined dramatically between 2005 and 2006 as a result of a severe storm in early summer 2006, as evidenced by direct observation of deep piles of uprooted plants on the shoreline immediately following the storm. By 2007, shoot density had already rebounded significantly through tremendous production of lateral shoots, and by 2009 the mean shoot density along the shallow transect was 57% of the high value recorded in 2005 (Fig. 7a). The mean proportion of reproductive shoots was consistently very low (<0.3% of total density). Percent cover showed a pattern similar to density, with a significant increase over 2006 by 2008 (Fig. 7c and Table 8). Canopy height and total biomass along the shallow transect recovered from the storm impact more slowly; it was not until 2009 that the mean values for these attributes increased significantly over 2006 (Fig. 7b, d and Table 8). The ratio of aboveground to belowground biomass was low and stable over time (mean of 2.3 ± 0.2 SE).

Eelgrass attributes along the mid-depth transect also declined in 2006, but the decrease from 2005 was less than that observed along the shallow transect and the subsequent recovery from the storm impact was more rapid (Fig. 7 and Table 8). By 2007, shoot density, canopy height, percent cover, and biomass had already increased significantly from 2006 levels, and these increases were generally maintained through 2009 (Fig. 7a–d and Table 8). The mean proportion of reproductive shoots along the mid-depth transect was always <0.6% and the aboveground-to-belowground biomass ratio was stable over time (mean of 4.3 ± 0.6 SE).

In contrast to eelgrass along the shallow and mid-depth transects, eelgrass attributes along the deep transect were fairly stable over the course of the study. From 2005 to 2007, there were relatively small annual differences in shoot density, canopy height, and percent cover, but these did not translate to any biomass differences (Fig. 7d and Table 8). The mean proportion of reproductive shoots was <0.4% during this interval. Canopy height and percent cover increased substantially in 2008, leading to a similarly large increase in biomass (Fig. 7d and Table 8) with a significantly greater investment in aboveground tissue (mean 2008 aboveground-to-belowground biomass ratio of 15.3 ± 3.4 SE vs. annual means from 3.2 to 9.0 in other years). The mean proportion of reproductive shoots along the deep transect was 1.5% (±1.0 SE) in 2008 and 8.8% (±2.9 SE) in 2009. Shoot density, canopy height, percent cover, and biomass decreased on the deep transect from 2008 to 2009 (Fig. 7a–d and Table 8) so that at the end of the study, eelgrass biomass was similar between the shallow and deep transects (p = 0.55; Fig. 7d).

Tier 3 monitoring revealed a shift in the location of the outer edge of the eelgrass bed over time (Table 9). In 2004 and 2005, along all three offshore trajectories, the eelgrass bed was limited by the central channel in LPB. Distance-to-edge data were not acquired in 2006. In 2007, the eelgrass along the trajectories beginning at the 0- and 25-m transect markers extended continuously through the channel to the opposite shore of LPB, resulting in a near doubling of the distance from the fixed reference points to the edge of the seagrass bed. Along the trajectory beginning at the 50-m transect marker, however, the eelgrass bed remained limited by the channel, which was slightly deeper in that location. This pattern persisted through 2009.
Table 9

Distance seaward (meters) from fixed reference points on the deep transect of the tier 3 monitoring stations in LPB and GSB/MB to the edge of the seagrass bed

Site

Transect reference point (m)

Distance (m)

2004

2005

2007

2008

2009

LPB

0

466

502

1,000

1,000

1,000

25

494

655

993

993

993

50

207

192

203

nd

215

GSB/MB

0

129

152

25

138

159

50

133

165

Reference points are the beginning, middle, and end of the 50-m transect. Monitoring at GSB/MB occurred during 2007 and 2009 only

nd no data

Great South Bay/Moriches Bay

Monitoring of the permanent quadrats at the GSB/MB tier 3 site was completed in 1 day by four people. An arcsine transformation was applied to percent cover data to stabilize sample variances. The shallow and mid-depth transects were dominated by widgeon grass (Fig. 8); vegetated quadrats contained either widgeon grass exclusively or a mixed community with a small proportion of eelgrass. Percent cover and density of eelgrass were very low along these transects, with little to no change from 2007 to 2009 (shallow transect: cover, p = 0.05; density, p = 0.22; mid-depth transect: cover, p = 0.38; density, p = 0.65). During this interval, widgeon grass cover increased substantially along the shallow transect (p < 0.0001) to a mean of about 90% cover and was stable along the mid-depth transect (p > 0.27). Widgeon grass reproductive shoot density decreased from a mean of 35 shoots/m2 (±10 SE) in 2007 to a mean of 17 shoots/m2 (±3 SE) in 2009 (p = 0.008). There was an average of nine widgeon grass reproductive shoots per square meter (±4 SE) along the mid-depth transect, with no difference between years (p = 0.41). The seagrass community along the deep transect was dominated by eelgrass. Eelgrass percent cover (p < 0.0001) and seagrass canopy height (p = 0.04) declined along the deep transect between years (Fig. 8), while widgeon grass cover declined to zero (Fig. 8). There were no reproductive shoots of either species. Average eelgrass density along the deep transect declined from 101 shoots/m2 (±18 SE) in 2007 to 47 shoots/m2 (±13 SE) in 2009 (p < 0.001). However, the seagrass bed expanded seaward between 21 and 32 m along trajectories from the deep transect fixed reference points (Table 9).
https://static-content.springer.com/image/art%3A10.1007%2Fs12237-011-9410-x/MediaObjects/12237_2011_9410_Fig8_HTML.gif
Fig. 8

Percent cover of eelgrass (Z. marina) (a), percent cover of widgeon grass (R. maritima) (b), and canopy height of the seagrass community (c) at tier 3 monitoring site in Great South Bay/Moriches Bay, 2007 and 2009 (mean ± SE)

Biomass Predictions

There was a strong relationship between direct and indirect methods of measuring eelgrass biomass in the LPB test quadrats. Total biomass harvested from 0.0625-m2 test quadrats in LPB was predicted quite accurately from biomass estimated from shoot density in the test quadrats and the average total weight per shoot within the central 0.0035-m2 core (R2 = 0.90, p < 0.0001; Fig. 9). The relationship was stronger for aboveground (R2 = 0.91, p < 0.0001) than belowground (R2 = 0.71, p < 0.0001) biomass components.
https://static-content.springer.com/image/art%3A10.1007%2Fs12237-011-9410-x/MediaObjects/12237_2011_9410_Fig9_HTML.gif
Fig. 9

Relationship between direct (y-axis) and indirect (x-axis) methods of eelgrass (Z. marina) biomass determination in Little Pleasant Bay test quadrats. Points represent total biomass

Multiple linear regression analysis showed that aboveground eelgrass biomass in the LPB tier 3 permanent quadrats depended usually on both percent cover and canopy height (Table 10). In 2006, the canopy height of the vegetation persisting following the severe storm disturbance ranged from 14 to 67 cm and did not explain any variation in aboveground biomass among quadrats beyond that already explained by percent cover (p = 0.69; Table 10). However, as eelgrass recovered in 2007 and 2008, the range in canopy height in the permanent quadrats increased substantially (8 to 91 cm in 2007, 9 to 110 cm in 2008), and the unique contribution of canopy height to explaining the variation in aboveground biomass was highly significant (p < 0.01; Table 10).
Table 10

Regression coefficients (SE) relating eelgrass aboveground biomass to population parameters measured during tier 3 monitoring in Little Pleasant Bay

Variable

2006

2007

2008

Constant

1.497 (1.078)

0.201 (0.573)

0.858 (0.763)

Percent cover

0.092* (0.026)

0.044** (0.010)

0.082** (0.014)

Canopy height

0.018 (0.043)

0.074* (0.019)

0.062* (0.022)

R2

0.71

0.81

0.84

No. observations

25

29

36

Dependent variable is square root of aboveground biomass measured indirectly in permanent 0.25-m2 quadrats; explanatory variables are untransformed

*p < 0.01; **p < 0.001

Discussion

Feasibility of Monitoring

Environmental monitoring is recognized globally as critical to environmental decision making (Griffith 1998; OECD 2003). The success of monitoring in influencing conservation depends on many facets of program development and execution, including planning, statistical design, selection of attributes to measure, data collection and synthesis, and information transfer (Elzinga et al. 2001; Reid 2001; Fancy et al. 2009). Ultimately, the overall feasibility of implementation is related to each of these monitoring elements. In this study, we examined whether tiered seagrass monitoring using time-saving sampling methods enhances feasibility while maintaining rigor. Our use of permanent sampling stations within a hierarchical framework maximized monitoring efficiency and improved the potential to produce statistically meaningful results.

Sampling effort is determined by the number and types of attributes incorporated in a monitoring plan, the methods used to characterize each attribute, and the number of samples collected. We limited tier 2 sampling to only two indicators of seagrass condition (Table 1), each of which is easily measured. Resampling simulations at LPB showed that four subsamples of percent cover per station were sufficient to avoid effects of within-station bias (difference between the mean of the subsamples and the true station mean) on detecting bay-wide change, which served to limit the time on station. The complexity of our analytical model for detecting system-wide trends, incorporating both repeated measures and distance zones, prohibited an accurate calculation of the model’s statistical power or the probability of detecting a true change in percent cover of seagrass (cf. Heidelbaugh and Nelson 1996). However, resampling simulations using reduced sample sizes showed indirectly that our sample size in LPB (55 stations in a 624-ha sampling frame) yielded more than sufficient power to detect an absolute change in mean cover from 25% to 49% (distance zone 3). Within the larger, more complex seagrass system in GSB/MB, our sample size (198 stations in a 9,220-ha sampling frame) was more than sufficient to detect an absolute change in mean cover from 14% to 31% (eelgrass in distance zone 1), and even sampling only half the number of stations, we would still expect to detect an absolute change in cover from 4% to 12% (eelgrass in distance zone 3) and from 10% to 19% (widgeon grass in distance zone 3) at least 50% of the time. Thus, we infer that our sampling design, with an average of 500 m (LPB) to 925 m (GSB/MB) between stations, was appropriate and sufficient in these systems. Our sampling design and attributes measured for tier 2 monitoring are very similar to those of the long-term Fisheries Habitat Assessment Program in Florida Bay, USA (Durako et al. 2002; Hall et al. 2007), with the addition of permanent sampling locations for trend detection. The efficiency afforded by repeated measurements at permanent locations minimized the number of samples, and the consequent field effort, required to detect a certain magnitude of change in percent cover on a bay-wide scale (cf. Elzinga et al. 2001).

We based our hexagon sizes and associated sample sizes for tier 2 monitoring on logistical and statistical considerations. We initially determined the maximum number of stations we could visit in LPB and GSB/MB within a reasonable time frame and then compared the relative area that would be sampled in each location (0.0009% of the total area in LPB, 0.0002% of total area in GSB/MB) to existing seagrass assessments with similar spatial designs (approx. 1 × 10−5% to 9 × 10−5% of total area was sampled in South Florida seagrass studies; Fourqurean et al. 2001, 2003). Although the use of power analysis to plan sample sizes is complicated by the combined incorporation of both permanent sampling stations and distance strata in an analytical model, our study may offer general guidance for future seagrass monitoring efforts using permanent, spatially distributed sampling stations. When permanent sampling units are being compared over two time periods using an unstratified sampling design, it is fairly simple to calculate the number of samples required to detect a specified level of change with a known degree of certainty (statistical power) and probability of falsely concluding a change occurred (probability of committing a type I error, or α). This calculation depends on an estimate of the standard deviation of the differences in observations at individual permanent stations between years (sdiff; Elzinga et al. 2001). Within the distance zones in LPB and GSB/MB where the most eelgrass occurred (Table 4, LPB zone 3; Table 6, GSB/MB zone 1), sdiff was remarkably consistent: LPB zone 3 2006–2007, sdiff = 31; LPB zone 3 2007–2008, sdiff = 29; GSB/MB zone 1 2007–2009, sdiff = 31. Given a mean eelgrass cover of 50% and a sdiff of 30, from 55 to 94 samples would be required to detect a 20% change in cover using permanent quadrats, depending on the statistical power (0.80 or 0.90) and α (0.05 or 0.10) desired (Table 11). The required number of samples is directly related to the power of the test and inversely related to the magnitude of difference detected and the probability of falsely concluding a change occurred (Table 11). Although pilot sampling is advised to determine the sdiff in a new area, the consistency of sdiff among portions of different estuaries and different years in our study suggests that our data may provide a springboard for initial planning of similar monitoring programs in seagrass systems without strong environmental gradients that might influence response (i.e., without strata). If environmental gradients do exist, however, at large sample sizes (greater than approx. n = 30), more statistical precision and power will be gained through designs that reduce the variation among stations as much as possible, e.g., by incorporating strata as appropriate, than by simply adding sampling stations (cf. Elzinga et al. 2001).
Table 11

Number of samples required to detect a specified change in eelgrass cover over two time periods using an unstratified sampling design and permanent sampling stations

Power

α

Minimum detectable relative change from 50% cover (absolute change in percentage points)

10% (5)

20% (10)

30% (15)

50% (25)

0.80

0.05

282

71

31

11

0.10

221

55

25

9

0.90

0.05

378

94

42

15

0.10

307

77

34

12

Calculations assume a standard deviation of the difference in percent cover between years of 30, an initial eelgrass cover of 50%, and a null hypothesis of no change. Power = probability of detecting a real change; α = probability of falsely concluding that a change occurred. The specified percent changes from 50% cover = absolute changes in cover within parentheses

Our use of direct observation for tier 2 monitoring incorporated standard field methods and readily available equipment such as quadrats and boats. However, other approaches for monitoring seagrass condition at meter-scale resolution throughout entire systems exist. Low-altitude aerial photographs taken from planes (Short and Burdick 1996; Neckles et al. 2005), blimps, or kites (Krause-Jensen et al. 2004) can be used to quantify very small seagrass patches, and various hydroacoustic techniques for seafloor characterization have been applied to measuring seagrass cover, canopy height, and biomass (e.g., Sabol et al. 2002; Komatsu et al. 2003; Riegl et al. 2005; Warren and Peterson 2007). These methods rely on specialized equipment for data acquisition and specialized skills for image and data processing, and require additional field efforts for ground verification and obtaining calibration samples; therefore, depending on the resources available for seagrass monitoring, these approaches may not be as cost-effective as direct observations. However, they offer the means to acquire very fine-scale measurements over very large areas and could be integrated into a hierarchical seagrass monitoring program if resources permit.

The efficiency of our tier 2 data collection was related in large part to the relative ease of field sampling. We intentionally restricted monitoring attributes to seagrass characteristics that could be measured from a boat or by snorkeling: percent cover and canopy height (Table 1). Sampling these attributes was simplified by the generally shallow depths of the lagoonal estuaries we studied. By using an underwater camera, we were able to view the substrate clearly from the boat for percent cover observations at even the deepest sampling stations (about 6 m). We were able to collect seagrasses from our deepest vegetated sites (about 3 m) for canopy height measurement using a rake on a telescoping handle, but collection devices such as benthic grabs or corers (e.g., Raz-Guzman and Grizzle 2001) would permit sampling in deeper water. However, in very deep or exposed seagrass systems where remote sampling for percent cover and canopy height is impracticable and SCUBA is required, the time and cost for monitoring these attributes will increase accordingly.

Tier 3 monitoring yields more detailed information on seagrass condition while providing insight into possible causes of change observed at larger scales. Seagrass shoot density, areal biomass, and patterns of biomass allocation are sensitive to perturbation and offer valuable insight into population processes and primary production associated with bed expansion or retraction (Durako 1994; Kirkman 1996; Duarte and Kirkman 2001), but direct shoot counts and biomass harvesting are very labor-intensive. Tiered monitoring restricted these time-consuming measurements to a small geographic area and the use of permanent quadrats improved the ability to detect change with a limited number of samples; both of these factors limited the sampling effort required at this scale.

Employing time-saving methodology for tier 3 biomass determination further increased the feasibility of monitoring. Various nondestructive methods of estimating seagrass biomass from easily measured variables exist (Duarte and Kirkman 2001). For example, models to predict seagrass biomass from ranked standing crop categories (Mellors 1991; Mumby et al. 1997), aerial photographic cover classes (Moore et al. 2000), and percent leaf cover measured directly (Orth and Moore 1988) or by photographic analysis (Lee 1997) have been described. Because these methods rely on traditional destructive samples to calibrate biomass regressions, the time saved through the use of nondestructive sampling is minimal until the total number of samples in the survey is considerably greater than the number of calibration samples harvested (Mumby et al. 1997). In addition, nondestructive methods may require substantial cross-calibration among observers (Mellors 1991), and estimation of aboveground biomass in very dense beds (Lee 1997) or belowground biomass in general (Mellors 1991) may be too imprecise to be useful. By harvesting very small samples (approx. 7-cm diameter cores), we limited sample processing time while measuring both aboveground and belowground shoot biomass. Subsequent indirect determination of areal biomass from average shoot weight and shoot density avoided inter-observer bias and was a reasonable proxy for labor-intensive harvest of quadrats sized appropriately for seagrasses at densities found in LPB (0.0625- to 0.25-m2 quadrats; Duarte and Kirkman 2001). The small core size minimized both the effort required for sample processing and the impact of harvests to National Park sites. An important caveat regarding this indirect method is the need to collect biomass cores from an area in which the plants are the same size as those rooted in the permanent quadrat used for density measurements. Our paired indirect and direct measurements of biomass (Fig. 9) emanated from small (0.0625 m2) areas, which minimized the variation in eelgrass canopy height between the harvested core and the larger quadrat. During actual implementation, biomass cores were collected from areas 0.5–1.0 m away from the permanent quadrats. Plant size can vary considerably within this distance, so biomass sample locations must be selected carefully. We found that the sampling time added by comparing canopy height measurements between permanent quadrats and potential harvest locations was minimal.

Overall, implementation of tier 2 and tier 3 monitoring at the sampling intensities we tested required a moderate investment of personnel. Annual monitoring of both tiers combined required a total of 3 to 13 field days for a maximum of four people, depending on the size of the system, although most sampling was conducted by two to three people. Limiting the resources required to complete a sampling event increases the likelihood of long-term implementation. Ultimately, conservation decisions are both local and regional in scale, and monitoring tools capable of evaluating broad regional trends are needed to support larger scale management and policy decisions (Urquhart et al. 1998). This is particularly true for managing threats to seagrasses from estuarine nutrient enrichment, which typically involve multiple watersheds and localities. The relative efficiency of tiered monitoring would make it feasible to implement throughout a regional network.

Relevance of a Tiered Approach

We found targeted monitoring within a hierarchical framework to have distinct advantages for understanding seagrass status and trends at multiple scales. Sampling at each tier was optimized to measure seagrass attributes linked to the stressor of primary concern, estuarine nutrient enrichment, and integration of information across tiers offered a comprehensive, yet efficient, assessment of seagrass extent and ecological condition. At both LPB and GSB/MB, tier 1 mapping programs provided information on long-term changes in seagrass distribution on bay-wide scales. Nesting tier 2 and tier 3 monitoring within bays with known seagrass distribution supplemented seagrass maps with higher resolution information that is relevant to multi-scale conservation objectives (Table 1) while being useful for interpreting mechanisms of change.

The monitoring data from LPB and GSB/MB indicated an improvement in the condition of seagrass resources during the study period. The increase in percent cover of eelgrass we observed in LPB following formation of the new inlet (tier 2 data, Fig. 5 and Table 4) coincided with an increase in tidal flushing (Kelley and Ramsey 2008). Tier 3 monitoring showed an increase in the seagrass depth limit following the new inlet formation, which is indicative of enhanced light penetration through the water column (Duarte 1991; Dennison et al. 1993) and helps explain the changes observed at tier 2. Tier 3 monitoring also revealed a very high capacity for lateral shoot production in this population, which would contribute to increased cover and bed expansion following an improvement in the light climate. In GSB/MB, the system-wide increases in eelgrass and widgeon grass cover that occurred from 2007 to 2009 (tier 2 data, Fig. 6 and Table 6) bracketed a 2008 brown tide bloom (A. anophagefferens) that was the most intense ever recorded locally (Gobler 2008). Two hypotheses were suggested to explain the unexpected increase in seagrass cover following the bloom: (1) an increase in sexual recruitment following a disturbance caused by extreme light limitation (cf. Duarte et al. 2006; Orth et al. 2006b) or (2) an increase in overall light availability despite the brown tide event. Tier 3 monitoring revealed either a decrease (widgeon grass) or no change (eelgrass) in sexual reproduction between 2007 and 2009; however, the considerable increase in widgeon grass cover at shallow depths and the significant bed expansion both suggest an increase in light availability between years. Subsequent application of the Wave Exposure Model (Malhotra and Fonseca 2007) showed a reduction in wave energy during 2008, which could have reduced sediment resuspension in this shallow coastal lagoon sufficiently to compensate for increased algal light attenuation. In both LPB and GSB/MB, tier 3 monitoring also highlighted patch-scale variability that can contribute to system-wide changes.

The addition of water column light attenuation to the suite of attributes measured at tier 3 would improve the ability to diagnose causal relationships. The SeagrassNet protocol includes 2-week intervals of continuous light measurement at two depths during quarterly sampling events (Short et al. 2006b). We have subsequently initiated deployment of continuously recording light sensors with automatic wipers at two depths for 4 weeks surrounding annual tier 3 monitoring (Kopp and Neckles 2009).

We established one tier 3 site per estuary as a compromise between information gain and monitoring feasibility. If time and resources permit, however, monitoring results from additional tier 3 sites can help elucidate larger scale trends. In particular, if tier 2 sampling is stratified by gradients in habitat characteristics, it would be useful to locate a high-resolution index site within multiple tier 2 strata if possible. Although information from our single tier 3 site in distance zone 3 of GSB/MB (Fig. 2) helps explain the overall changes in seagrass abundance in that system, supplemental high-resolution data from distance zone 1 are needed to interpret the mechanisms underlying the relative increase in dominance of eelgrass in the western end of GSB near Fire Island Inlet (Fig. 6 and Tables 5 and 6).

In areas with substantial depth gradients such as LPB, eelgrass shoot morphology and bed structure vary considerably with depth, from dense short shoots in shallow water to long, relatively sparser shoots in deeper areas (Fig. 7). In areas of more uniform eelgrass morphology, biomass can be predicted from percent cover alone with high accuracy (Orth and Moore 1988), but regressions of tier 3 data from LPB showed that biomass in this location is generally dependent on both the percent cover and the canopy height of the vegetation (Table 10). Because these metrics are measured through tier 2 monitoring, it is possible to apply the biomass models developed through intensive tier 3 monitoring to system-wide data collected at tier 2 to predict aboveground biomass distribution on a bay-wide scale (cf. Fourqurean et al. 2001). Such integration of data across scales contributes additional information for evaluating ecosystem condition relative to conservation objectives for maximizing seagrass growth (Table 1).

The tiered framework (Bricker and Ruggiero 1998) we applied to seagrass monitoring is similar in some ways to a three-level process for comprehensive wetland monitoring and assessment that has been recently evaluated (Brooks et al. 2004; Wardrop et al. 2007c). The wetland assessment hierarchy includes level 1 or landscape assessments using remote sensing data to characterize wetlands at the watershed scale (Brooks et al. 2004; Reiss and Brown 2007), level 2 or rapid assessments using semiquantitative metrics and best professional judgment to classify wetland condition based on field observations (Fennessy et al. 2007), and level 3 or intensive assessments of biological, physical, and chemical characteristics based on quantitative field measurements (Reiss and Brown 2007; Wardrop et al. 2007b). Estimates of wetland condition are produced at each level by comparing assessment results to reference standards, and as the intensity of measurement increases from level 1 to level 3, so too does the confidence in the results. Whereas the tiered framework we adopted relies on nesting of sampling areas and indicators across tiers and consistent application of all scales of monitoring, the wetland assessment model uses unique indicators at each level to produce information of varying spatial resolutions on the same location, with the number of levels dictated by project needs; conceptually, the wetland assessment model is more of a layer cake (with the same geographic area at each scale of measurement) than a triangle (with geographic area decreasing in extent from tier 1 to tier 3). For small seagrass systems, it may be possible to expand the tier 3 sampling frame to provide detailed information on a larger area; conversely, for expansive wetland systems, it may prove beneficial to limit intensive measurements to a subset of the rapid assessment area. Common to both hierarchical approaches, however, integration of information across scales can be used to calibrate ecosystem condition categories, predict condition at local or regional levels, and focus management targets (c.f. Bricker and Ruggiero 1998; CRMSW 2000; Brooks et al. 2004; Reiss and Brown 2007; Wardrop et al. 2007c; Whigham et al. 2007).

Monitoring results are most useful for resource managers if monitoring attributes are linked explicitly to conservation objectives (Nichols and Williams 2006) and are highly relevant to management endpoints (Wardrop et al. 2007a). The monitoring attributes in this study were selected for evaluating whether specific seagrass conservation objectives related to goals of protecting estuarine resources from nutrient enrichment are being achieved. The efficiency of tiered monitoring will facilitate generating information on system-wide seagrass condition and high-resolution mechanisms of change to guide conservation decisions.

Acknowledgments

Funding for this study was provided by the US Geological Survey, Park Oriented Biological Support Program and the National Park Service, Northeast Coastal and Barrier Network. We thank Fred Short for instruction in the SeagrassNet approach; Fred Short, Jeffrey Gaeckle, and David Rivers for help in establishing our tier 3 monitoring site in LPB according to the SeagrassNet spatial design; Holly Bayley, Lena Curtis, Steve Dwyer, Bev Johnson, Carrie Phillips, Lotte Rivers, Steve Smith, Adam Thime, and Jesse Wheeler for field assistance in LPB; and Brooke Rodgers and Jamie Brisbin for field assistance in GSB/MB. We also thank Andrew Gilbert and Dennis Skidds for expert GIS assistance and tutelage. We are grateful to Carrie Phillips and Megan Tyrrell of Cape Cod National Seashore and Michael Bilecki of Fire Island National Seashore for logistical support, and we appreciate the cooperation of Dawson Farber, Harbormaster and Shellfish Constable for the town of Orleans, MA, in maintaining a long-term tier 3 monitoring site in LPB. This manuscript was greatly improved by the comments of Nancy Rybicki, Melisa Wong, and two anonymous reviewers. Use of trade, product, or firm names does not imply endorsement by the US Government.

Copyright information

© Coastal and Estuarine Research Federation (outside the USA) 2011