Estuaries and Coasts DOI 10.1007/s12237-017-0293-3
MANAGEMENT APPLICATIONS
Development of a Multimetric Index for Integrated Assessment of Salt Marsh Ecosystem Condition Jessica L. Nagel 1 & Hilary A. Neckles 2 & Glenn R. Guntenspergen 1 & Erika N. Rocks 3 & Donald R. Schoolmaster Jr 4 & James B. Grace 4 & Dennis Skidds 5 & Sara Stevens 5
Received: 14 February 2017 / Revised: 7 July 2017 / Accepted: 11 July 2017 # Coastal and Estuarine Research Federation (outside the USA) 2017
Abstract Tools for assessing and communicating salt marsh condition are essential to guide decisions aimed at maintaining or restoring ecosystem integrity and services. Multimetric indices (MMIs) are increasingly used to provide integrated assessments of ecosystem condition. We employed a theorybased approach that considers the multivariate relationship of metrics with human disturbance to construct a salt marsh MMI for five National Parks in the northeastern USA. We quantified the degree of human disturbance for each marsh using the first principal component score from a principal components analysis of physical, chemical, and land use stressors. We then applied a metric selection algorithm to different combinations of about 45 vegetation and nekton metrics (e.g., species abundance, species richness, and ecological and functional classifications) derived from multi-year monitoring
data. While MMIs derived from nekton or vegetation metrics alone were strongly correlated with human disturbance (r values from −0.80 to −0.93), an MMI derived from both vegetation and nekton metrics yielded an exceptionally strong correlation with disturbance (r = −0.96). Individual MMIs included from one to five metrics. The metric-assembly algorithm yielded parsimonious MMIs that exhibit the greatest possible correlations with disturbance in a way that is objective, efficient, and reproducible.
Communicated by Cathleen Wigand
Coastal wetlands occur at the interface between uplands and oceans within a narrow range of elevation influenced by tidal characteristics. Salt marshes occupy as much as 45,000 km2 worldwide (Greenberg et al. 2006). These coastal wetlands provide essential livelihood services to millions of people, as well as critical regulating services such as maintenance of water quality, protection from storms and erosion, and carbon sequestration valued at US $10,000 per hectare per year (Barbier et al. 2011; Hopkinson et al. 2012; Möller et al. 2014). However, as much as half of global coastal wetlands have been lost due to human conversion for agriculture and other anthropogenic changes in land use (Pendleton et al. 2012), and climate change, declining water quality, and changes in sediment delivery rates associated with human activity continue to affect the world’s remaining wetlands (Kirwan and Megonigal 2013). In the USA, direct human modification of coastal wetlands has led to extensive wetland degradation (Kennish 2001; Bertness et al. 2002) through activities such as ditching and drainage of the marsh platform; reduction of tidal exchange by
Electronic supplementary material The online version of this article (doi:10.1007/s12237-017-0293-3) contains supplementary material, which is available to authorized users. * Hilary A. Neckles
[email protected]
1
Patuxent Wildlife Research Center, U.S. Geological Survey, 12100 Beech Forest Rd, Laurel, MD 20708, USA
2
Patuxent Wildlife Research Center, U.S. Geological Survey, 196 Whitten Road, Augusta, ME 04330, USA
3
National Park Service, Northeast Coastal and Barrier Network, 54 Elm Street, Woodstock, VT 05091, USA
4
Wetland and Aquatic Research Center, U.S. Geological Survey, 700 Cajundome Blvd, Lafayette, LA 70506, USA
5
National Park Service, Northeast Coastal and Barrier Network, University of Rhode Island Coastal Institute in Kingston, 1 Greenhouse Rd., Kingston, RI 02881, USA
Keywords Multimetric index . Salt marsh . Ecosystem assessment . Ecological indicators . Coastal wetlands
Introduction
Estuaries and Coasts
dikes, roads, and water control structures; disposal of dredge spoil; introduction of invasive species; and discharge of nitrogen, phosphorus, oil, and other contaminants (Roman et al. 2000). The impacts of such disturbances on salt marsh integrity and their ecosystem services are far reaching (Daiber 1986; Wolfe 1996; Adamowicz and Roman 2005). Salt marsh management and conservation decisions are increasingly focused on maintaining and improving ecological integrity (MEA 2005). As a consequence, approaches to monitor and assess ecosystem condition have gained widespread attention (e.g., Simenstad and Thom 1996; Short et al. 2000; Wigand et al. 2001; Neckles et al. 2002, 2015). Considerable research has been devoted to development of both quantitative and functional approaches to assess ecosystem integrity and track ecosystem condition over time. In recent decades, multimetric indices (MMIs) have been used to integrate numerous characteristics of complex ecosystems into onedimensional indices of ecosystem condition (Karr 1999; Hering et al. 2006). By combining multiple indicators of human disturbance into a composite score, an MMI may be more sensitive to disturbance than any of its individual component metrics (Lee et al. 2004; Schoolmaster et al. 2012). MMIs for wetland ecosystem assessment have been based on specific biota, such as fish (Simon 1998), zooplankton (Lougheed and Chow-Fraser 2002), and plants (Lopez and Fennessy 2002; Mack 2004), and they have also been based on combinations of taxa (plants, invertebrates, and fish; Wilcox et al. 2002; James-Pirri et al. 2014) or ecosystem components (plants, various fauna, and physical environmental features; Wigand et al. 2011; Langman et al. 2012; Miller et al. 2016). As development of wetland MMIs has proliferated, they have been applied to increasingly larger scales. MMIs have been developed worldwide for regional assessments of inland (e.g., Ortega et al. 2004; Boix et al. 2005; Miller et al. 2006; Mereta et al. 2013) and coastal wetlands (e.g., Wigand et al. 2011; Staszak and Armitage 2013; Nestlerode et al. 2014) and are integral to reporting results of the National Wetland Condition Assessment (NWCA) in the conterminous USA (USEPA 2016). Traditionally, MMI development has involved exploratory analysis of ecosystem attributes or metrics measured at sites representing a range of conditions, from a reference standard to sites that are subject to high levels of anthropogenic disturbance (Karr 1999; Karr and Chu 2000). Expert knowledge or various ordination techniques may be used to reduce the number of candidate metrics, organize them into classes, and rank their importance in explaining the variation among sites and metrics then may be selected for inclusion in the MMI based on the strength of bivariate relationships between disturbance and individual responses. However, interactions among ecosystem stressors, environmental covariates, and ecological responses may introduce unexplained variability into metric data sets, reducing the sensitivity of individual metrics to perceived disturbance factors and obscuring determination of the
Bbest^ metrics to include in the MMI (Schoolmaster et al. 2012, 2013). In such cases, Schoolmaster et al. (2012) demonstrated that evaluating how metrics respond to disturbance as a group can improve the potential to generate a maximally sensitive MMI. This may be particularly relevant for salt marsh ecosystems, where relationships among agents of change, physical and chemical environmental stressors, and ecological responses are inherently complex. We applied a multivariate, algorithmic approach for MMI development to salt marsh ecosystems in the northeastern USA. As described originally by Schoolmaster et al. (2012, 2013), this is an empirical method of MMI construction from a set of candidate metrics, in which the unique contribution of each metric toward explaining the variation in human disturbance is evaluated within the context of all candidates. Through an objective, transparent, and reproducible process for metric selection, an MMI is produced that is statistically robust and sensitive given the set of available data. This method of MMI construction was tested previously on freshwater wetlands (Schoolmaster et al. 2013). We applied the algorithmic approach to salt marshes within US National Park Service units distributed between Massachusetts and Maryland. Our study shows the broad utility of this approach for developing robust assessment tools for salt marsh ecosystems.
Methods Salt Marsh Monitoring Data Salt marsh monitoring data were collected from 33 marsh study units (MSUs) in five National Parks within the NPS Northeast Coastal and Barrier Network (NCBN) along the northeastern coast of the USA (Fig. 1 and Table 1). MSUs were established at Cape Cod National Seashore (NS), Sagamore Hill National Historic Site (NHS), Fire Island NS, Gateway National Recreation Area (NRA), and Assateague Island NS. Individual MSUs were delineated to encompass the contiguous marsh area bounded by geographic features, such as tidal creeks, salt marsh-upland borders, grid ditches, shoreline, and other distinguishable features, using GIS. For all parks except Cape Cod NS, all areas of salt marsh within the park were identified, all potential MSUs were delineated, and the final sample of MSUs at each park was then randomly selected from this population. At Cape Cod NS, we used existing MSUs that had been selected as representative of the park’s salt marsh resources. MSU size at most parks ranged from 5 to 15 ha, but at Cape Cod NS, the MSUs were generally larger (up to 168 ha). MMI development was based on marsh vegetation and nekton (i.e., fish and free-swimming crustaceans) monitoring data collected from 2008 to 2013 as part of the network-wide vital signs monitoring program (Stevens et al. 2005).
Estuaries and Coasts Fig. 1 Location of parks sampled within the Northeast Coastal and Barrier Network. Black triangles = MSUs with vegetation and nekton data. White circles = MSUs with vegetation data only
Monitoring occurred biennially in most MSUs, yielding 3 years of data during this time period; however, for some MSUs, only 2 years of data were available (Table 1). Vegetation data were collected in all 33 MSUs and nekton data were collected at only 24 of the MSUs (Table 1). To allow comparisons of MMIs developed from different sized datasets (24 vs 33 MSUs) and metric types (vegetation-only vs vegetation + nekton), we generated four separate datasets: vegetation data alone from 33 MSUs (V33); vegetation and nekton data combined from 24 MSUs (VN24); vegetation data alone from 24 MSUs (V24); and nekton data alone from 24 MSUs (N24). In addition, salt marsh monitoring data were collected irregularly between 2003 and 2007 in five of the MSUs located in three parks as part of a pilot program (Cape Cod NS, Gateway NRA, and Sagamore Hill NHS, Table 1). We applied the MMIs to the data collected during the pilot program to evaluate changes in salt marsh condition as revealed by different MMIs.
The percent cover of marsh vegetation and other cover types (e.g., water, bare ground, wrack, and litter) was measured once during the peak of the growing season (early July through early September) per sampling year. From 2008 to 2013, cover was measured using the BraunBlanquet method (Kent and Coker 1994) within approximately 50 randomized square-meter (1 m × 1 m) quadrats along 10 randomized transects per MSU. Percent cover of each species was estimated visually within cover classes (absent, 66% of the total values within a metric) were omitted from further analysis (Stoddard et al. 2008; Schoolmaster et al. 2013). In addition, metrics that duplicated another metric were removed, as they would have contributed essentially identical information to the MMI. We identified duplicate metrics by high values of inter-metric correlation (r > 0.95 or r < −0.95; Schoolmaster et al. 2013). For each duplicate pair, we retained the metric that was most interpretable in the context of salt marsh ecosystem condition. Following these metric-reduction steps, the V33 data set contained 27 vegetation metrics (Table 2), the V24 data set contained 28 vegetation metrics (Table 2), the N24 data set contained 18 nekton metrics (Table 3), and the VN24 data set contained 27 vegetation (Table 2) and 18 nekton metrics (Table 3). Human Disturbance Index We derived a Human Disturbance Index (HDI) to use for selecting response metrics for inclusion in the MMI. We used existing aerial imagery and the 2006 National Land Cover Database (NLCD; Fry et al. 2011) to quantify human impacts to each MSU in terms of physical, chemical, and land use stressors (Table 4). Categorical landscape metrics and classes were adopted from the New England Rapid Assessment Method for assessing salt marsh condition (Carullo et al. 2007) and aerial photographs of marshes representing each category were used as a guide in classification. The metrics were scored by a trained technician using the most recently available orthophotography in ArcMap™ 10.2 in consultation with park staff familiar with the MSUs. Continuous metrics quantifying the amount of developed land cover within 150and 1-km buffers of each MSU were calculated from NLCD data using ArcGIS™ 10.2. The percent of developed land area in each buffer zone was derived by aggregating developed land cover classes in the NLCD (Developed Open Space, Developed Low Intensity, Developed Medium Intensity, Developed High Intensity, and Barren Land) into a single category. Values were then relativized to the size of the MSU by multiplying by (buffer area/MSU area) (Table 4). Only one MSU (CACO_10) had measurable agricultural land in the buffer zone (relativized percent of agricultural land in the 150-m buffer = 0.8%); thus, this land cover was not included in the HDI formulations.
Estuaries and Coasts Table 2 Candidate vegetation metrics from NCBN parks for incorporation into salt marsh MMIs
Table 3 Candidate nekton metrics from NCBN parks for incorporation into salt marsh MMIs
Vegetation communities
Nekton communities
Percent cover of low marsh species Percent cover of high marsh species
Relative abundance of resident fish species Relative abundance of resident shrimp species
Percent cover of brackish border species
Relative abundance transient fish species Total density (no. m−2) Fish density (no. fish m−2) Decapod density (no. decapods m−2)a Species richness (total no. spp. within each sample averaged over MSU)
Percent cover of salt marsh border species Percent cover of panne, pool, and creek species Percent cover of native species Percent cover of high salinity-tolerant species Species richness (total no. spp. within each sample averaged over MSU)
Native species richness (total no. native spp. within MSU) Introduced species richness (total no. introduced spp. within MSU)
Native species richness (total no. native spp. within each sample averaged over MSU) Native species richness (total no. native spp. within MSU) Introduced species richness (total no. introduced spp. within each sample averaged over MSU)a Introduced species richness (total no. introduced spp. within MSU)a Percent native species richness (no. native spp. in MSU/total no. spp. in MSU) Percent introduced species richness (no. introduced spp. in MSU/total no. spp. in MSU)b High salinity-tolerant species richness (no. high salinity-tolerant spp. within MSU) High salinity-tolerant species richness (no. high salinity-tolerant spp. in sample/total no. spp. in sample) Percent high salinity-tolerant species (no. high salinity-tolerant spp. in MSU/total no. spp. in MSU) Percent high salinity-tolerant species (no. high salinity-tolerant spp. in sample/total no. spp. in sample) Vegetation species Percent frequency Spartina alterniflora Percent cover Spartina patens Percent frequency S. patens Percent cover Distichlis spicata Percent frequency D. spicata Percent cover Salicornia spp. Percent frequency Salicornia spp. Percent frequency Iva frutescens Percent cover Scirpus and Schoenoplectus spp.b Percent frequency Scirpus and Schoenoplectus spp.b Percent cover Phragmites australis Percent frequency P. australis a
Metric used only in the dataset consisting of vegetation data alone from 33 MSUs (V33)
b
Metric used only in the dataset from the 24-MSU subset including both vegetation and nekton data (V24, VN24)
Categorical and continuous disturbance metrics were converted to an ordinal scale following scoring criteria outlined in the New England Rapid Assessment Method (Carullo et al. 2007). Categorical metrics were assigned a rank score in which 0 represented undisturbed conditions and 6 represented
Resident fish species richness (total no. resident fish spp. within MSU) Resident fish species richness (total no. resident fish spp. within each sample averaged over MSU) Percent native species richness (no. native spp. in MSU/total no. spp. in MSU) Percent native species richness (no. native spp. in sample/total no. spp. in sample) Percent resident fish species richness (no. resident fish spp. in MSU/total no. spp. in MSU) Nekton species Relative abundance of Fundulus spp. Relative abundance of Cyprinodon variegatus Relative abundance of Menidia spp. Length of Fundulus heteroclitus a
Decapod = crabs and shrimp
the most disturbed condition. Continuous variables were converted to ranked values using three break-points in the distribution: values ≤ (mean − ½SD) were assigned 0; values > (mean − ½SD) and ≤ mean were assigned 2; values > mean and ≤ (mean + ½SD) were assigned 4; and values > (mean + ½SD) were assigned 6. A Principal Component Analysis (PCA) of the seven ranked disturbance metrics was used to derive two HDIs: HDI 24 corresponded to the metric datasets using 24 MSUs (VN24, V24, N24) and HDI 33 corresponded to the dataset using 33 MSUs (V33). A Spearman rank correlation matrix of the disturbance metrics was used as input for the PCA and we evaluated the Principal Components (PCs) by examining the eigenvectors, scree test, variance explained, and comprehensibility (Kachigan 1982). The first Principal Component score (PC-1) for each site was used as the basis of the HDI score (Falcone et al. 2010) as PC-1 accounted for the largest part of the total variance inherent in the data. The PC-1 scores were transformed to HDI by first reordering if necessary so that more positive values represented higher human disturbance and then rescaling to the HDI range of 0 (minimal human disturbance) to 100 (highest human disturbance). This type of disturbance index has been used successfully for watershed
Estuaries and Coasts Table 4 Index
Land use metrics used to calculate the Human Disturbance
Metric
Classes/rank
Ditch density
Extent of ditching and No (0), low (2), draining of the marsh moderate (4), severe unit (6) No (0), yes (6) Potential restriction of normal flow and tidal range by features such as undersized culverts or bridges, causeways, dikes, etc.
Tidal restriction
Definition
Degree of hydrologic Well flushed (0), connection with marine moderately flushed waters (3), poorly flushed (6) Extent of salt marsh areas Fill/fragmentation No (0), low (2), filled or fragmented (i.e. moderate (4), severe were once whole (6) systems). Relativized % Continuous converted % disturbed landa in 150-m Disturbed_ to ranked variable buffer * (area of 150m buffer/area of MSU) Continuous converted % disturbed landa in 1-km Relativized % to ranked variable buffer * (area of Disturbed_ buffer/area of MSU) 1km Tidal flushing
Point source discharges of pollutants
No (0), low (2), Outfalls, drains emptying moderate (4), severe into marsh (6)
a
Aggregated NLCD classes: Developed Open Space, Developed Low Intensity, Developed Medium Intensity, Developed High Intensity, Barren Land, and Agricultural
and wetland assessment in the Great Lakes Basin (Brazner et al. 2007; Danz et al. 2007), in New England (Wigand et al. 2011), and throughout the western USA (Falcone et al. 2010).
MMI Construction We applied the MMI construction algorithm to the four sets of candidate metrics derived from the 2008–2013 salt marsh monitoring data (V33, V24, VN24, N24; Table 2; Table 3). Because the metrics have different units, all metrics were rescaled to unitless measures with similar ranges using a continuous scoring method (Blocksom 2003): mscaled ¼
m −L U −L
where m = metric, L = 2.5 percentile values of m, and U = 97.5 percentile values of m. Values of m that were U were set to L or U, respectively. Metrics that were positively correlated with HDI were reflected about their midpoints to ensure that the MMI would be negatively correlated with disturbance. MMI assembly then followed a stepwise metric selection
process to maximize the likelihood of predicting the observed HDIs. Details of the MMI assembly algorithm are provided by Schoolmaster et al. (2012, 2013). In general, each candidate metric was used as a starting point, m1. The initial ml was added site-by-site to each of the remaining metrics, mj, to find the combination m1 + mj yielding the strongest negative correlation with HDI. This metric combination was then added to each of the remaining metrics, mj, to find the new combination +mj yielding the strongest negative correlation with HDI. This metric selection process was repeated until the addition of another mj failed to improve the ability to predict HDI (likelihood ratio test, p > 0.05). This entire assembly process was repeated using each metric as the starting point m1, thereby yielding as many candidate MMIs as there were candidate metrics. Each candidate MMIj thus represented a model, HDI = β0 + β1 MMIj, with k parameters (effectively, the number of metrics in MMIj + 2). The Akaike Information Criterion (AIC) of each model was then calculated as the basis for comparing MMIs and selecting those with the highest capacity to predict disturbance. For each candidate MMI, the difference between its AIC and that of the MMI with the lowest AIC was determined (i.e., ΔAIC). All MMIs with ΔAIC