Fine-Scale Mapping of Coastal Plant Communities in the Northeastern USA

volume 39pages1728(2019)

Abstract

Salt marshes of the northeastern United States are dynamic landscapes where the tidal flooding regime creates patterns of plant zonation based on differences in elevation, salinity, and local hydrology. These patterns of zonation can change quickly due to both natural and anthropogenic stressors, making tidal marshes vulnerable to degradation and loss. We compared several remote sensing techniques to develop a tool that accurately maps high- and low-marsh zonation to use in management and conservation planning for this ecosystem in the northeast USA. We found that random forests (RF) outperformed other classifier tools when applied to the most recent National Agricultural Imagery Program (NAIP) imagery, NAIP derivatives, and elevation data between coastal Maine and Virginia, USA. We then used RF methods to classify plant zonation within a 500-m buffer around coastal marsh delineated in the National Wetland Inventory. We found mean classification accuracies of 94% for high marsh, 76% for low marsh zones, and 90% overall map accuracy. The detailed output is a 3-m resolution continuous map of tidal marsh vegetation communities and cover classes that can be used in habitat modeling of marsh-obligate species or to monitor changes in marsh plant communities over time.

Introduction

Coastal marshes are among the world’s most productive ecosystems and provide significant services to humans across the globe. These marshes serve as a gateway between land and sea for humans and wildlife alike, act as a buffer against coastal storms, and provide critical nutrients to marine food webs (Barbier et al. 2011). Tidal marshes also support and protect biodiversity by providing habitat to marine and estuarine fish, crustacean populations, and migratory birds (Boesch and Turner 1984; Master 1992; Brown et al. 2002).

Within the world’s tidal marsh systems, those located along the Atlantic coast of the United States support the highest number of terrestrial vertebrate specialists described worldwide (Greenberg et al. 2006). This suite of species includes herpetofauna and mammals, but the majority of described vertebrate specialists are birds. Several species are limited completely to these marshes during the breeding season, several of which are in decline (Correll et al. 2017), with extinction predicted for the saltmarsh sparrow within 50 years (Correll et al. 2017; Field et al. 2017a, c).

These declining species nest predominantly within the high-marsh zone, one of several vegetation communities found within coastal marshes. High marsh differs from other marsh areas in elevation, salinity, and frequency of inundation (Bertness and Ellison 1987; Pennings and Callaway 1992; Ewanchuk and Bertness 2004) and is characterized by flooding during spring tides linked to the lunar cycle. In the northeastern United States, the plant species Spartina patens, short-form S. alterniflora, Distichlis spicata, and Juncus gerardii characterize high-marsh zones, which also include Salicornia spp., Glaux maritima, and Solidago sempervirens (Nixon and Oviatt 1973; Bertness 1991; Emery et al. 2001, Ewanchuk and Bertness 2004). Conversely, low marsh is characterized by daily tidal flooding and is a near monoculture of tall form S. alterniflora. The surrounding terrestrial border experiences infrequent inundation by salt water during extreme tides and storms, and is characterized by a more diverse flora that is often dominated by Iva frutescens and Typha spp. (Miller and Egler 1950; Ewanchuk and Bertness 2004). Introduced Phragmites australis (hereafter Phragmites) also occurs within this ecosystem, especially around the borders of disturbed marshes (Chambers et al. 1999; Philipp and Field 2005).

These plant community zones can be quickly altered by both natural and anthropogenic stressors such as sea-level rise, nutrient run-off from adjacent uplands, and the spread of introduced species (Day et al. 2008). The significant increase in sea level during recent decades poses one of the largest threats to these marsh ecosystems. As sea levels encroach on the marshes’ seaward side and upland marsh migration is limited by human-developed coastal infrastructure (Field et al. 2017b) and upland habitats (Field et al. 2016), a “pinching effect” can occur, resulting in marsh loss. Coastal marshes can combat rising sea levels through vertical growth, or accretion (Kirwan et al. 2016), but when the rate of sea-level rise exceeds the rate of accretion, marsh area will decline (Crosby et al. 2016). Rising sea levels can also drive invasion of high marsh areas with flood-tolerant low marsh species (Donnelly and Bertness 2001; Field et al. 2016), causing transition from high to low marsh (Kirwan et al. 2016). This pattern, however, is not ubiquitous to all marshes (Kirwan and Guntenspergen 2010; Wilson et al. 2014). In addition to sea-level rise, extreme storm events that flood the coastline have been shown to permanently alter marsh structure within days (Day et al. 2008) and can have a lasting effect on plant community structure and saltmarsh degradation.

Marsh degradation and rapid change is likely to continue into the future due to the paired effects of climate change and human development. Sea levels are expected to rise substantially between 2013 and 2100 (IPCC 2014), and continuing storm events affecting coastal regions are also predicted. The future distribution of high- and low-marsh habitat therefore remains uncertain (Chu-Agor et al. 2011; Kirwan et al. 2016). It is essential to develop tools to identify coastal marsh plant communities, particularly high marsh, on a biologically relevant timescale to protect existing ecosystem services and to inform the adaptive management of coastal wetlands as habitat for high-marsh specialist species.

The physical and biological characteristics that differentiate high marsh from low marsh and other marsh plant communities are potentially detectible using remotely-sensed multispectral and hyperspectral imagery. Both types of imagery can record wavelengths of light outside of the visible range for humans, with hyperspectral imagery recording reflectance values in much finer detail and precision (hundreds of individual bands recorded) than multispectral imagery (several wide-ranging bands recorded, e.g. red green blue, or RGB, imagery). Several studies have previously demonstrated distinct spectral differences between tidal marsh species using hyperspectral imagery (Rosso et al. 2005; Belluco et al. 2006; Yang 2009). Such imagery, when combined with elevation data, has previously produced high-accuracy classifications of tidal marsh vegetation communities, albeit at smaller spatial scales (e.g. Hladik et al. 2013).

Hyperspectral imagery is costly, however, especially across large landscapes (Adam et al. 2010); Belluco et al. (2006) compared several aerial and satellite sensors with changing spatial and spectral resolution, and although hyperspectral imagery performed slightly better than the multispectral, spatial resolution was the most important factor in classifier performance. Belluco et al. (2006) recommend the use of multispectral satellite imagery for the mapping of marsh vegetation. Beside the visible spectrum (RGB), multispectral imagery should also include infrared (IR) reflectance values to allow differentiation of vegetation types, calculation of vegetation indices (e.g. the Normalized Difference Vegetation Index or NDVI, Rouse et al. 1974) and detection of soil moisture differences (Jin and Sader 2005; Pettorelli et al. 2005), particularly in tidal wetlands (Klemas 2011). The IR spectrum has previously been used as a tool to predict tidal marsh communities both in smaller regions within the northeastern United States (Gilmore et al. 2008; Hoover et al. 2010; Meiman et al. 2012) and elsewhere (Isacch et al. 2006; Liu et al. 2010). An exception to this has been in the classification of invasive Phragmites that often borders tidal marshes. Large-scale classification of this wetland class has met with some success (e.g. Bourgeau-Chavez et al. 2015, Long et al. 2017), however success is limited when using RGB and IR inputs alone (Samiappan et al. 2017). A large-scale effort to map coastal Phragmites in the northeastern US has not yet been attempted.

Due to the large spatial scale at which northeastern tidal marshes occur, publicly-available and low-cost imagery datasets offer the most promising option for repeatedly delineating large swaths of marsh along the coast. Landsat satellite imagery provided by the National Aeronautic and Space Administration (NASA) is publicly available multispectral imagery including RGB and IR bands provided at a 30 × 30 m resolution and is often used for classifying coarse cover types across large landscapes. In the case of tidal marshes, however, the heterogeneity in tidal marsh vegetation often occurs at scales smaller than 30 m pixels, and large-scale classifications of tidal marsh plant communities using this imagery have so far been unfruitful (e.g. Correll 2015). There is thus a clear need for an alternative path to create a regional classification of tidal marsh vegetation.

Recent advances in high-resolution airborne imagery provide new opportunities to develop large-scale classifications of coastal plant communities in the northeast, including Phragmites (e.g. Xie et al. 2015). The National Agricultural Imagery Program (NAIP) from the US Department of Agriculture (USDA 2016) captures 3-band, high-resolution RGB orthophotos during the growing season. Since 2007 most states have added a Near-InfraRed (NIR) band to the image requirements to aid in the accurate classification of vegetative cover. The image resolution is 1 m with 6-m horizontal accuracy and a maximum of 10% cloud cover. The imagery, freely available for governmental agencies and the public, is an affordable alternative to commercial aerial and high-resolution multispectral satellite imagery. Recent applications of NAIP imagery include mapping of tree cover (Davies et al. 2010), forest clearings (Baker et al. 2013), isolated trees (Meneguzzo et al. 2013), land cover classification (Baker et al. 2013) and mining activity (Maxwell et al. 2014).

In this study we compare several remote sensing techniques applied to NAIP imagery, elevation data from the National Elevation Dataset (NED) provided by the US Geological Survey (USGS 2015), and local tidal information records from the National Oceanic and Atmospheric Administration (NOAA 2016) to develop an affordable tool capable of repeated classification of high-marsh zones in tidal marshes in the northeastern United States. We then use the best-performing classifier to categorize marsh vegetation communities with a 3-m resolution from coastal Maine to Virginia, USA.

Methods

Study Site and Community Types

Our marsh-mapping effort encompasses all salt marshes of the Northeast Atlantic coast of the USA, from northern Maine to Virginia. To define our classification extent we applied a 500-m buffer to all coastal, tidal marsh as delineated by the National Wetland Inventory (NWI, USFWS 2010) estuarine emergent wetland (E2EM) layer. The study site is further split into 8 subzones (Fig. 1) to accommodate data management and processing.

Fig. 1
figure1

Geographic extent of our study site from Maine to Virginia, USA, divided into 8 biogeographic regions used in the remote sensing classification of tidal marsh cover types

These coastal marshes vary substantially from north to south. Due to local bathymetric structure, tidal amplitudes in the Gulf of Maine are among the highest in the world (Garrett 1972), while those farther south experience much less variation between high and low tides. Similarly, a preponderance of rocky or highly sloped shorelines in the north limits marshes to small (~10–100 ha) patches, while southern marshes form larger patches of marsh along the coast. Across our study area, however, tidal marsh ecosystems can be reliably separated into six distinct cover types, plus two bordering cover types, which we included in our marsh mapping effort:

  1. 1.

    High marsh: Area flooded during spring tides related to the lunar cycle and dominated by Spartina patens, Distichlis spicata, Juncus gerardii, and short form Spartina alterniflora. Other species include Juncus roemerianus, Scirpus pungens, Scirpus robustus, Limonium nashii, Aster tenuifolius, and Triglochin maritima.

  2. 2.

    Low marsh: Area flooded regularly by daily tides and dominated by tall form Spartina alterniflora.

  3. 3.

    Salt pools/pannes: Depressed, bare areas with sparse vegetation cover and extremely high soil salinities. Generally, pools retain water between high tides while pannes do not.

  4. 4.

    Terrestrial border: Area infrequently flooded by storm and spring tides and can include areas of marsh with fresh/brackish water due to a high water table and/or runoff from impervious surfaces. Typical plant species include Typha angustifolia, Iva frutescens, Baccharis halimifolia, Solidago sempervirens, Scirpus robustus, and Spartina pectinata.

  5. 5.

    Phragmites: The exotic invasive form of Phragmites australis (subspecies australis). This subspecies is of considerable management interest (Saltonstall 2002), especially in marshes with freshwater input, upland development, and/or increased nutrients (Dreyer and Niering 1995; Bertness et al. 2002; Silliman and Bertness 2004).

  6. 6.

    Mudflat: Exposed muddy areas free of vegetation.

  7. 7.

    Open water (bordering cover type): Channels and bays leading to open ocean included within the 500-m buffer.

  8. 8.

    Upland (bordering cover type): All non-marsh terrestrial cover included within the 500-m buffer.

Data Sources

We collected training data for marsh vegetation classes both in the field and remotely using aerial imagery, depending on the cover type. We collected training polygons for high marsh, low marsh, and Phragmites between May and August of 2015 and 2016. Technicians collecting polygon data were collectively trained at the beginning of the season in salt marsh vegetation identification. Phragmites polygons mapped were of the invasive Phragmites australis australis and not of the native North American form Phragmites australis americanus. Tecnnicians used a GEO 7X Trimble GPS (Trimble 2015) without an external antenna for all community delineation. Horizontal accuracy of this unit without the external antenna is estimated at <1 m by the manufacturer.

We used a generalized random tessellation stratified (GRTS) sampling framework designed to sample tidal marsh bird communities across all ownership types (Wiest et al. 2016) to select randomly-located delineation sites for training data across our study area. Technicians navigated to bird survey points and located contiguous patches of high marsh, low marsh, and Phragmites larger than 10 × 10 m as they traversed to each bird survey location. At each patch, they placed a stake flag or other highly visible marker on the ground to indicate the beginning of polygon delineation and then delineated the outer boundary of the patch on foot by walking the outer perimeter with the GEO 7X. We collected training data for open water, pools and pannes, and mudflat cover classes using manual digitization of 2014–2015 1-m NAIP imagery using ArcGIS 10.3 (ESRI 2016) since these cover classes are easily identifiable in visible wavelength imagery.

We used the most recent digital ortho-photography (RGB and NIR) available from the NAIP collected during the growing season from 2014 or 2015 as imagery predictor data (see Appendix A for acquisition year by state). We resampled raw 1 m NAIP imagery to 3 m resolution to match the spatial scale of the NED, which was used as the digital elevation model (DEM) for this analysis. We also calculated NAIP imagery derivatives using ArcMap 10.3 using the raw band values. We refer to them as ‘pseudo’-vegetation indices because we used the raw band values instead of reflectance values. In total, we used the following data inputs as predictor variables: DEM, Raw NAIP Band 1, Raw NAIP Band 2, Raw NAIP Band 3, Raw NAIP band 4, NDVI, the Normalized Difference Water Index (McFeeters 1996), the Difference Vegetation Index (Richardson and Wiegand 1977), and the first three principle components from a principal components analysis (Fung and Ledrew 1987) of the four NAIP bands, which collectively explained >95% of the variance.

Marsh habitats are often influenced by their elevation and topographic context and therefore can often be successfully mapped using elevation data (e.g. Hladik et al. 2013, Maxwell et al. 2016). Elevational data is particularly helpful in tidal marshes, where topography can drive tidal flooding frequency and thus can influence plant species zonation (Silvestri et al. 2005). We used the NED for all elevation predictor data. The NED is derived from different contributed datasets and then processed by the USGS into a near-continuous DEM at various resolutions across the US. We used 1/9-s (~3 m resolution) data when available for our classification. When no 1/9 arc-second imagery was available, we used 1/3 arc-second data (~10 m resolution). To account for the large differences in tidal inundation across our study area, we collected tidal data for the study area from the closest NOAA tidal gauge station, creating 29 different tidal zones (NOAA 2016). For each of the stations we collected the following tidal datums: HAT, MHHW, MHW, MSL, NAVD88 and MAX (Table 1). We resampled the NED data to an exact 3 m resolution to match the upscaled NAIP imagery and clipped the resulting imagery with the 500 m buffer around all coastal tidal marsh in the NWI. We further clipped the NED by the 29 tidal gauge zones of the study area, and rescaled each zone to the NAVD88 datum using the NOAA tidal amplitude data. To calibrate the DEM across the entire study site we used the Mean High Tide (MHT) divided by Mean Highest High Tide (MHHT) value for each tidal gauge zone as a basis for elevational differences between marsh zone types.

Table 1 Summary and definitions of datums collected by the National Oceanic and Atmospheric Administration (NOAA) and used to classify marsh cover type from Virginia to northern Maine, USA

We classified the NED of each tidal gauge zone based on the tidal amplitude data for that zone. To do this, we rescaled NOAA tidal amplitude data to match the NAVD88 datum used in the elevation dataset, and defined elevation limits for open water, high marsh, low marsh, terrestrial border, and upland class based on flooding history (Fig. 2). We then used these thresholds in conjunction with NAIP imagery reflectance values and derivatives to identify water, high marsh, low marsh, and Phragmites cover types. We used only elevation thresholds to define the terrestrial border and upland cover types. In rare cases in small areas along the coast where there was no NED DEM layer available, we classified the marsh communities without the DEM data input. In these cases, no terrestrial border or upland was defined.

Fig. 2
figure2

Graphic representation of the tidal elevation limits for tidal marsh community classification effort in the northeastern USA based on flooding history are represented by the National Oceanic and Atmospheric Administration (NOAA) vertical datums

Data Analysis

We compared three classification methods using training data collected in 2015 to delineate tidal marsh cover types. We conducted this comparison of classifiers on NAIP imagery, imagery derivatives, and NED elevation data from the center of our study area (Delaware Bay, subregion 6, Fig. 1) to maximize utility of the resulting method to both the north and south. First, we used classification and regression trees (CART), a fast and flexible rule-based classifier where no statistical data distribution is required (Otukei and Blaschke 2010). CART methods are particularly useful when integrating environmental variables with different measurement scales and are robust for large datasets. Post-hoc pruning removes nodes with low explanatory power to reduce overfitting. We used the R package rpart (Therneau et al. 2015) which implements the CART methods described by (Breiman et al. 1984). Secondly, we used random forests (RF), which are collections of decision trees that improve the accuracy and stability of a single decision tree (Breiman 2001). RFs perform well with small training sets, uses a random subset of the training data to calculate the variable importance, and similar to cross-validation the out-of-bag (OOB) error estimates delivers a measure of classification accuracy (Breiman 2001). We used the R package RandomForest (Liaw and Wiener 2002) with Ntree and mtry default values for all RF analyses. Finally, we used support vector machines (SVM), which are non-parametric classifiers that use risk minimization to separate classes defined by the ‘support vectors’, or points that occur closest to the splitting threshold. In general, SVM offers high training performance versus low generalizing errors, but is sensitive to over-fitting, especially with noisy and unbalanced data. We used the svm function in the R package e1071 (Dimitriadou et al. 2006) with the default parameter values and polynomial kernel.

To select the best remote sensing classifier for the final marsh layer, we classified and validated a randomly-selected independent training and validation subset (66% training, 33% validation) of the first year of polygon training data (2015, n = 36 training polygons) from zone 6 and compared classification accuracies across CART, RF, and SVM methods. We also considered specifications and known strengths of the of the classifiers including robustness, sensitivity to over fitting, and performance with relatively small training datasets.

We applied the best performing classifier, RF, to all biogeographic zones from Maine to Virginia using NAIP imagery, imagery derivatives, NED elevation data, and NOAA tidal gauge information. For the final classification, we used the out-of-bag (OOB) error estimates produced by the RF algorithm to measure accuracy of our classification by zone. We then clipped the resulting marsh classification by the DEM-based cover types (upland, terrestrial border), wherever the NED layer was available. Due to variable image quality and/or due to high tide during image acquisition, we sometimes encountered artefacts in the imagery that affected the accuracy of our classification, particularly in Zones 1 and 6. Sun glitter or high mud content in open water sometimes caused misclassification of this cover type as low marsh. We used the RF probability scores for the open water class to better represent the actual water cover and then updated the open water classification in Zones 1 and 6 to improve overall accuracy. All methods and datasets involved in our study are freely available to the public, and our analyses are limited to tools available through ArcGIS, a geographic information system commonly-used by federal, state, and private conservation organizations, or simple open-source Program R code (Appendix B).

Results

We collected a total of 2655 field training polygons across all cover types from Maine to Virginia (Table 2). Of these training polygons, 36 were used in our methods comparison in Zone 6 for high marsh (n = 18), low marsh (n = 14), and Phragmites (n = 4). In this comparison we found that RF methods generally outperformed the other two tools in classification of tidal marsh plant communities (Table 3). Classification of high and low marsh cover types returned accuracy rates ranging between 73% and 88% across all classifiers, with RF producing substantially lower error rates for high marsh and all three classifiers producing similar levels of accuracy for low marsh (Table 3). All classifiers produced low accuracy rates for the Phragmites cover type, ranging from 32 to 55%, but SVM methods were substantially worse when compared with the other two classifiers.

Table 2 Percent classification error for cover classes of tidal marsh vegetation communities from Virginia to Maine, USA
Table 3 Average classification error (%) for three different remote sensing techniques for classifying tidal marsh vegetation communities in Delaware Bay, USA using National Agricultural Imagery Program (NAIP) imagery from 2015

Our final data layer produced through RF methods classified a total of 16,014 km2 of tidal marsh and bordering communities within our defined 500-m buffer at a 3-m resolution (Fig. 3). This layer is now publicly available for public download at https://nalcc.databasin.org/galleries/46d6e771dd6f4fdb8aa5eb46efffffa7.

Fig. 3
figure3

Visualization of a detailed random forest classification of vegetation communities within tidal marsh habitat in the Northeastern USA. Maps shows example portions of the entire coverage in (a) northern Massachusetts and (b) Delaware Bay, New Jersey; entire coverage is available at https://nalcc.databasin.org/galleries/46d6e771dd6f4fdb8aa5eb46efffffa7

Mean classification accuracies varied among cover types, ranging from >99% for open water and mudflat to 25% for Phragmites (Table 2). Within geographic zones, open water, mudflat, and pools/pannes were classified with high (>95%) accuracy in almost all cases. Classification of the three vegetation types varied among cover classes. High marsh was classified with a mean accuracy of 94%, but with clear regional variation. In the regions from New Jersey north and in the inner portions of Chesapeake Bay, accuracy was generally greater than 95%. In contrast, accuracy from Delaware Bay to Virginia was lower with 83–87% of high marsh correctly classified. High marsh was most commonly misclassified as low marsh. Classification of low marsh was generally less accurate than high marsh, with an overall accuracy of 77%. Low marsh was regularly confounded with open water and classification accuracy tended to be higher in the southern zones, especially the inner portions of Delaware and Chesapeake Bays.

Classification of Phragmites showed the greatest variation across regions, and was often confounded with terrestrial border. Overall, classification accuracy for this cover type was 75%, and in coastal Massachusetts accuracy approached 95%. By contrast, overall Phragmites accuracy in coastal New Jersey was 67%, and in our northernmost region (coastal New Hampshire and Maine) accuracy was 20%.

Finally, almost 4000 km2 of the data layer is covered by tidal marsh (i.e., excluding open water or upland, Table 4), with the majority classified as high marsh (36%) followed by low marsh (21%), mudflats (7%), Phragmites (7%), pools and pannes (5%), and terrestrial border (24%). The distribution of cover types varies from north to south with an overall increase in low marsh area and Phragmites to the south. Conversely, the percentage of mudflat cover increases to the north.

Table 4 Total marsh area (km2) and percentage of total marsh for each cover class throughout the northeastern USA, from Virginia to Maine

Discussion

Tidal marshes of the northeastern USA are critical pieces of the coastal landscape, providing key wildlife habitat and ecosystem services to humans (Barbier et al. 2011). Our effort applies well-established methods and data sources for remote sensing of wetlands to classify plant communities within this important ecosystem from Maine to Virginia; the resulting data layer is the first of its kind to classify this marsh ecosystem at such a high spatial resolution (3 m) regionally, with a mean map classification accuracy of 90% and a classification accuracy for high marsh vegetation of 94%.

The high classification accuracies in the high marsh zone make this data layer particularly helpful for use by marsh managers, researchers, and planners. Tracking change in marsh vegetation can be used to understand which marshes are most impacted by sea-level rise, predict changes in flooding risk potential for coastal properties (Arkema et al. 2013), and to identify marshes with low rates of change to protect as important habitat for marsh-obligate species. As sea levels continue to rise and human development continues to alter marsh hydrology and accretion (Day et al. 2008), a method to repeatedly classify cover types of coastal marsh will be integral to measuring the amount and distribution of available habitat and change in habitat over time. The small spatial resolution and high horizontal accuracy of our data layer also allows it to be used across varying spatial extents, from local municipalities to multi-state regions.

The NAIP dataset used in our classification offers a high-resolution, low-cost set of multispectral imagery with a refresh rate of 3 years. This dataset, however, has limitations when used over large areas (see also Meneguzzo et al. 2013), and potential users should carefully consider the pros and cons of this dataset before setting out on a similar classification effort. Variation in the time and day of image acquisition across the NAIP dataset results in different tidal stages, plant phenology, and illumination across images. The NAIP post-hoc color balancing applied to these images by each contractor (7 contractors across our study area) does not correct for differences in illumination and atmosphere in a standardized way. This results in a radiometric imbalance across the spatial extent of the dataset, limiting the use of the NAIP dataset in large-scale classification efforts to vegetative communities with either large differences in reflectance values across the RGB and NIR bands (as is the case with tidal marsh) or those varying across other characteristics measurable across large areas (e.g. elevation in the case of tidal marshes). Further, the temporal resolution of the NAIP dataset (one set of images per year, flown during “greenup” between June and August) is ideal for marsh classification but limits the use of NAIP data to analyses specific to this time of the year. Since the timing of high and low tide changes daily, this temporal resolution also results in imagery taken at different parts of the daily tidal cycle, and low marsh or mudflat areas were likely partially flooded when most imagery was collected.

While these sources of error likely contributed to some noise within our classification effort, multiple contractors collected the imagery included in each analysis zone across different times of tidal inundation, making a quantitative comparison of the combined effects of these shortcomings difficult. Conveniently, the RF methods used in our final classification work well with these described sources of variation with low threat of over-fitting (see comparison of data sources and RF classification in Fig. 4). Classification success (Table 2) assessed through OOB error estimate shows high average accuracies for most cover types, although there is variation depending on biogeographic zone. Additional work to collect independent data and compare it to this layer’s predictions would be a valuable next step to identify ways to improve current methods; while OOB error estimates are a commonly used measure of RF accuracy (Breiman 2001), they do not systematically evaluate classification accuracy outside the training polygons (e.g. Fry et al. 2011). Future studies to independently validate our classification outside of OOB error estimation the will directly strengthen support for the long-term use of this tool in monitoring community change in tidal marshes over time. In particular, additional tests to assess on-the-ground accuracy should involve data collected across the entire region due to the OOB differences we find across our analysis zones.

Fig. 4
figure4

Comparison of (a) a random forest classification of tidal marsh vegetation communities in Scarborough Marsh, Maine with source data from (b) the National Agriculture Imagery Program and (c) the National Elevation Dataset (1/9 arc-second layer)

Although the primary focus of our study was to distinguish between high and low marsh, the importance of invasive Phragmites to many management decisions led us to also consider this cover type. Unfortunately, Phragmites proved particularly difficult to classify, however we have chosen to keep this cover type in the overall classification because accuracy is relatively high (>70%) in the majority of our analysis zones (zones 2–4,6–8). The nature and amount of training data available for this cover type (n = 148 polygons, Table 2) compared to the other cover classes likely contributed to error in our Phragmites classification; previous work using RF methods suggest that training sample balance in imperative between classes for optimal classification results (Belgiu and Dragut 2016). Our results may have also been influenced by the size of our Phragmites training polygons, as this rapidly-growing cover type can change patch size quickly, influencing classifier accuracy (Kettenring et al. 2016) and make predicting across large landscapes particularly difficult. Additionally, upon assessment of known marshes, the classification algorithm for Phragmites was regularly confounded with terrestrial border species not included in the training data, particularly with stands of Typha spp., which is similar in structure to Phragmites. This problem combined with the variable ground elevations at which Phragmites can be found likely further confounded the classification. A strong need remains for development of a method for large-scale classification of this invasive species across latitudes, flooding regimes, and imagery sources for use in monitoring and management along the coast.

Conclusion

Repeated, large-scale classification of coastal vegetation communities is urgently needed to help inform a variety of conservation and management issues related to this rapidly shrinking ecosystem. We present methods for a repeatable classification at a 3-m resolution of distinct cover types within tidal marshes of the northeastern USA to serve 1) as a vegetation community delineation for use in management and conservation decision-making, 2) as a layer for local and regional analyses of this biological community, and 3) as a base layer against which future comparisons can measure land cover change over time. These actions are all integral for the long-term preservation of tidal marshes and the species they support, especially as climate change and other human influences continue to affect this ecosystem.

References

  1. Adam E, Mutanga O, Rugege D (2010) Multispectral and hyperspectral remote sensing for identification and mapping of wetland vegetation: a review. Wetl Ecol Manag 18(3):281–296

    Article  Google Scholar 

  2. Arkema KK, Guannel G, Verutes G, Wood SA, Guerry A, Ruckelshaus M, Kareiva P, Lacayo M, Silver JM (2013) Coastal habitats shield people and property from sea-level rise and storms. Nat Clim Chang 3:913–918

    Article  Google Scholar 

  3. Baker B, Warner T, Conley JF, McNeil BE (2013) Does spatial resolution matter? A multi-scale comparison of object-based and pixel-based methods for detecting change associated with gas well drilling operations. Int J Remote Sens 34(5):1633–1651

    Article  Google Scholar 

  4. Barbier EB, Hacker SD, Kennedy C, Koch EW, Stier AC, Silliman BR (2011) The value of estuarine and coastal ecosystem services. Ecol Monogr 81(2):169–193

    Article  Google Scholar 

  5. Belgiu and Dragut (2016) Random forest in remote sensing: a review of applications and future directions. ISPRS J Photogramm Remote Sens 114:24–31

    Article  Google Scholar 

  6. Belluco E, Camuffo M, Ferrari S, Modenese L, Silvestri S, Marani A, Marani M (2006) Mapping salt-marsh vegetation by multispectral and hyperspectral remote sensing. Remote Sens Environ 105(1):54–67

    Article  Google Scholar 

  7. Bertness MD (1991) Zonation of Spartina Patens and Spartina Alterniflora in New England salt marsh. Ecology 72(1):138–148

    Article  Google Scholar 

  8. Bertness MD, Ellison AM (1987) Determinants of pattern in a New England salt marsh plant community. Ecol Monogr 57(2):129–147

    Article  Google Scholar 

  9. Bertness MD, Ewanchuk PJ, Silliman BR (2002) Anthropogenic modification of New England salt marsh landscapes. Proc Natl Acad Sci U S A 99(3):1395–1398

    CAS  Article  Google Scholar 

  10. Boesch DF, Turner RE (1984) Dependence of fishery species on salt marshes: the role of food and refuge. Estuaries 7(4):460

    Article  Google Scholar 

  11. Bourgeau-Chavez L, Endres S, Battaglia M, Miller ME, Banda E, Laubach Z, Marcaccio J (2015) Development of a bi-national Great Lakes coastal wetland and land use map using three-season PALSAR and Landsat imagery. Remote Sens 7(7):8655–8682

    Article  Google Scholar 

  12. Breiman L (2001) Random forests. Mach Learn 45(1):5–32

    Article  Google Scholar 

  13. Breiman L, Friedman JH, Olshen RA, Stone CJ (1984) Classification and regression trees. The Wadsworth Statistics Probability Series (Vol. 19)

  14. Brown S, Harrington B, Parsons K, Mallory E (2002) Waterbird use of northern Atlantic wetlands protected under the north American wetlands conservation act. Waterbirds 25:106–114

    Google Scholar 

  15. Chambers RM, Meyerson LA, Saltonstall K (1999) Expansion of Phragmites australis into tidal wetlands of North America. Aquat Bot 64(3–4):261–273

    Article  Google Scholar 

  16. Chu-Agor ML, Muñoz-Carpena R, Kiker G, Emanuelsson A, Linkov I (2011) Exploring vulnerability of coastal habitats to sea level rise through global sensitivity and uncertainty analyses. Environ Model Softw 26(5):593–604

    Article  Google Scholar 

  17. Correll MD (2015) The biogeography and conservation of tidal marsh bird communities across a changing landscape. Dissertation, University of Maine, Orono Maine USA

  18. Correll MD, Wiest WA, Hodgman TP, Shriver WG, Elphick CS, McGill BJ, O'Brien K, Olsen BJ (2017) Predictors of specialist avifaunal decline in coastal marshes. Conserv Biol 31(1):172–182

    Article  Google Scholar 

  19. Crosby SC, Sax DF, Palmer ME, Booth HS, Deegan LA, Bertness MD, Leslie HM (2016) Salt marsh persistence is threatened by predicted sea-level rise. Estuar Coast Shelf Sci 181:93–99

    Article  Google Scholar 

  20. Davies KW, Petersen SL, Johnson DD, Davis DB, Madsen MD, Zvirzdin DL, Bates JD (2010) Estimating juniper cover from National Agriculture Imagery Program (NAIP) imagery and evaluating relationships between potential cover and environmental variables. Rangel Ecol Manag 63(6):630–637

    Article  Google Scholar 

  21. Day JW, Christian RR, Boesch DM, Yáñez-Arancibia A, Morris J, Twilley RR, Stevenson C (2008) Consequences of climate change on the ecogeomorphology of coastal wetlands. Estuar Coasts 31(3):477–491

    Article  Google Scholar 

  22. Dimitriadou E, Hornik K, Leisch F, Meyer D (2006) e1071: Misc functions of the Department of Statistics, probability theory group (formerly E1071), TU Wien. R package version 1.6–8. https://CRAN.R-project.org/package=e1071

  23. Donnelly JP, Bertness MD (2001) Rapid shoreward encroachment of salt marsh cordgrass in response to accelerated sea-level rise. Proc Natl Acad Sci 98(25):14218–14223

    CAS  Article  Google Scholar 

  24. Dreyer GD, Niering WA (1995) Tidal marshes of Long Island sound: ecology, history and restoration. Bulletins 34. Connecticut College Digital Commons, New London

  25. Emery NC, Ewanchuk PJ, Bertness MD (2001) Competition and salt-marsh plant zonation: stress tolerators may be dominant competitors. Ecology 82(9):2471–2485

    Article  Google Scholar 

  26. ESRI (2016) ArcGIS desktop: release 10.3. Environmental Systems Research Institute, Redlands

    Google Scholar 

  27. Ewanchuk PJ, Bertness MD (2004) Structure and organization of a northern New England salt marsh plant community. J Ecol 92:72–85

    Article  Google Scholar 

  28. Field CR, Gjerdrum C, Elphick CS (2016) Forest resistance to sea-level rise prevents landward migration of tidal marsh. Biol Conserv 201:363–369

    Article  Google Scholar 

  29. Field CR, Bayard TS, Gjerdrum C, Hill JM, Meiman S, Elphick CS (2017a) High-resolution tide projections reveal extinction threshold in response to sea-level rise. Glob Chang Biol 23(5):2058–2070

    Article  Google Scholar 

  30. Field CR, Dayer AA, Elphick CS (2017b) Landowner behavior can determine the success of conservation strategies for ecosystem migration under sea-level rise. Proc Natl Acad Sci 114:9134–9139

    CAS  Article  Google Scholar 

  31. Field CR, Ruskin KJ, Benvenuti B, Borowske A, Cohen JB, Garey L, Hodgman TP, Kern RA, King E, Kocek AR, Kovach AI, O’Brien KM, Olsen BJ, Pau N, Roberts SG, Shelly E, Shriver WG, Walsh J, Elphick CS (2017c) Quantifying the importance of geographic replication and representativeness when estimating demographic rates, using a coastal species as a case study. Ecography 40:001–010

    Article  Google Scholar 

  32. Fry J, Xian G, Jin S, Dewitz J, Homer CG, Yang L, Wickham JD (2011) Completion of the 2006 National Land Cover Database for the conterminous United States. Photogramm Eng Remote Sens 77:858–566

    Google Scholar 

  33. Fung T, Ledrew E (1987) Application of principal components analysis to change detection. Photogrammetric Enginnering and. Remote Sens 53(12):1649–1658

    Google Scholar 

  34. Garrett C (1972) Tidal resonance in the bay of Fundy and gulf of Maine. Nature 238:441–443

    Article  Google Scholar 

  35. Gilmore MS, Wilson EH, Barrett N, Civco DL, Prisloe S, Hurd JD, Chadwick C (2008) Integrating multi-temporal spectral and structural information to map wetland vegetation in a lower Connecticut River tidal marsh. Remote Sens Environ 112(11):4048–4060

    Article  Google Scholar 

  36. Greenberg R, Maldonado JE, Droege S, McDonald MV (2006) Terrestrial vertebrates of tidal marshes: evolution, ecology, and conservation. Stud Avian Biol 32

  37. Hladik C, Schalles J, Alber M (2013) Salt marsh elevation and habitat mapping using hyperspectral and LIDAR data. Remote Sens Environ 139:318–330

    Article  Google Scholar 

  38. Hoover M, Civco D, Whelchel A (2010) The development of a salt marsh migration tool and its application in Long Island sound. ASPRS 2010 Annual Conference Proceedings. San Diego, CA USA

  39. IPCC (2014) Climate Change 2014: Synthesis Report. Contribution of Working Groups I, II, and III to the FIfth Assessment Report of the Intergovernmental Panel on Climate Change [Core Writing Team, Pachauri RK and Meyer LA (eds)]. IPCC, Geneva, Switzerland

  40. Isacch JP, Costa CSB, Rodriguez-Gallego L, Conde D, Escapa M, Gagliardini D, Iribarne OO (2006) Distribution of saltmarsh plant communities associated with environmental factors along a latitudinal gradient on the south-West Atlantic coast. J Biogeogr 33(5):888–900

    Article  Google Scholar 

  41. Jin S, Sader S (2005) Comparison of time series tasseled cap wetness and the normalized difference moisture index in detecting forest disturbances. Remote Sens Environ 94(3):364–372

    Article  Google Scholar 

  42. Kettenring KM, Mock KE, Zaman B, McKee M (2016) Life on the edge: reproductive mode and rate of invasive Phragmites australis patch expansion. Biol Invasions 18(9):2475–2495

    Article  Google Scholar 

  43. Kirwan ML, Guntenspergen GR (2010) Influence of tidal range on the stability of coastal marshland. J Geophys Res 115(F2):1–11

    Article  Google Scholar 

  44. Kirwan ML, Temmerman S, Skeehan EE, Guntenspergen GR, Faghe S (2016) Overestimation of marsh vulnerability to sea level rise. Nat Clim Chang 6(3):253–260

    Article  Google Scholar 

  45. Klemas V (2011) Remote sensing of wetlands: case studies comparing practical techniques. J Coast Res 27(3):418–427

    Article  Google Scholar 

  46. Liaw A, Wiener M (2002) Classification and regression by randomForest. R News 2(3):18–22

    Google Scholar 

  47. Liu C, Jiang H, Hou Y, Zhang S, Su L, Li X, Wen Z (2010) Habitat changes for breeding waterbirds in Yancheng National Nature Reserve, China: a remote sensing study. Wetlands 30(5):879–888

    Article  Google Scholar 

  48. Long AL, Kettenring KM, Hawkins CP, Neale CM (2017) Distribution and drivers of a widespread, invasive wetland grass, Phragmites australis, in wetlands of the great salt Lake, Utah, USA. Wetlands 37(1):45–57

    Article  Google Scholar 

  49. Master TL (1992) Composition, structure, and dynamics of mixed-species foraging aggregations in a southern New Jersey salt marsh. Colon Waterbirds 15(1):66–74

    Article  Google Scholar 

  50. Maxwell AE, Strager MP, Warner TA, Zégre NP, Yuill CB (2014) Comparison of NAIP orthophotography and RapidEye satellite imagery for mapping of mining and mine reclamation. GISci Remote Sens 51(3):301–320

    Article  Google Scholar 

  51. Maxwell AE, Warner TA, Strager MP (2016) Predicting palustrine wetland probability sing random Forest machine learning and digital elevation data-derived terrain variables. Photogramm Eng Remote Sens 82(6):437–447

    Article  Google Scholar 

  52. McFeeters SK (1996) The use of the normalized difference water index (NDWI) in the delineation of open water features. Int J Remote Sens 17(7):1425–1432

    Article  Google Scholar 

  53. Meiman S, Civco D, Holsinger K, Elphick CS (2012) Comparing habitat models using ground-based and remote sensing data: saltmarsh sparrow presence versus nesting. Wetlands 32(4):725–736

    Article  Google Scholar 

  54. Meneguzzo DM, Liknes GC, Nelson MD (2013) Mapping trees outside forests using high-resolution aerial imagery: a comparison of pixel- and object-based classification approaches. Environ Monit Assess 185(8):6261–6275

    Article  Google Scholar 

  55. Miller W, Egler F (1950) Vegetation of the Wequetequock-Pawcatuck tidal-marshes, Connecticut. Ecol Monogr 20(2):143–172

    Article  Google Scholar 

  56. National Oceanic and Atmospheric Administration (2016) Tides and Currents. Available at: https://tidesandcurrents.noaa.gov. Accessed February 2017

  57. Nixon SW, Oviatt CA (1973) Ecology of a New England salt marsh. Ecol Monogr 43(4):463–498

    Article  Google Scholar 

  58. Otukei JR, Blaschke T (2010) Land cover change assessment using decision trees, support vector machines and maximum likelihood classification algorithms. Int J Appl Earth Obs Geoinf 12(1):27–31

    Article  Google Scholar 

  59. Pennings S, Callaway R (1992) Salt marsh plant zonation: the relative importance of competition and physical factors. Ecology 73(2):681–690

    Article  Google Scholar 

  60. Pettorelli N, Vik JO, Mysterud A, Gaillard JM, Tucker CJ, Stenseth NC (2005) Using the satellite-derived NDVI to assess ecological responses to environmental change. Trends Ecol Evol 20(9):503–510

    Article  Google Scholar 

  61. Philipp KR, Field RT (2005) Phragmites australis expansion in Delaware Bay salt marshes. Ecol Eng 25(3):275–291

    Article  Google Scholar 

  62. Richardson AJ, Wiegand CL (1977) Distinguishing vegetation from soil background information. Photogramm Eng Remote Sens 43(12):1541–1552

    Google Scholar 

  63. Rosso PH, Ustin SL, Hastings A (2005) Mapping marshland vegetation of San Francisco Bay, California, using hyperspectral data. Int J Remote Sens 26(23):5169–5191

    Article  Google Scholar 

  64. Rouse JW, Haas RH, Schell JA (1974) Monitoring the vernal advancement and retrogradation (greenwave effect) of natural vegetation. Progress Report. NASA Goddard Space Flight Center, College Station Texas, USA

  65. Saltonstall K (2002) Cryptic invasion by a non-native genotype of the common reed, Phragmites australis, into North America. Proc Natl Acad Sci U S A 99(4):2445–2449

    CAS  Article  Google Scholar 

  66. Samiappan S, Turnage G, Hathcock L, Casagrande L, Stinson P, Moorhead R (2017) Using unmanned aerial systems for high-resolution remote sensing to map invasive Phragmites australis in coastal wetlands. Int J Remote Sens 38(8–10):2199–2217

    Article  Google Scholar 

  67. Silliman BR, Bertness MD (2004) Shoreline development drives invasion of Phragmites australis and the loss of plant diversity on New England salt marshes. Conserv Biol 18(5):1424–1434

    Article  Google Scholar 

  68. Silvestri S, Defina A, Marani M (2005) Tidal regime, salinity and salt marsh plant zonation. Estuar Coast Shelf Sci 62(1–2):119–130

    CAS  Article  Google Scholar 

  69. Therneau T, Atkinson B, Ripley B, Ripley MB (2015) Rpart: recursive partitioning and regression trees. R Package Version 4.1–10

  70. Trimble (2015) GEO 7X Ground Positioning System. Available at: https://www.trimble.com/mappingGIS/geo-7-series

  71. US Department of Agriculture (2016) National Agriculture Imagery Program accessed through the Geospatial Data Gateway. Available at: http:// datagateway.nrcs.usda.gov. Accessed February 2016

  72. US Fish and Wildlife Service. National Wetland Inventory (2010) Available at: https://www.fws.gov/wetlands/index.html. Accessed January 2016

  73. US Geological Survey (2015) National Elevation Dataset (NED). Available at: https://nationalmap.gov/elevation.html. Accessed January 2016

  74. Wiest WA, Correll MD, Olsen BJ, Elphick CS, Hodgman TP, Curson DR, Shriver WG (2016) Population estimates for tidal marsh birds of high conservation concern in the northeastern USA from a design-based survey. Condor 118(2):274–288

    Article  Google Scholar 

  75. Wilson C, Hughes ZJ, FitzGerald DM, Hopkinson CS, Valentine V, Kolker AS (2014) Saltmarsh pool and tidal creek morphodynamics: dynamic equilibrium of northern latitude saltmarshes? Geomorphology 213:99–115

    Article  Google Scholar 

  76. Xie Y, Zhang A, Welsh W (2015) Mapping wetlands and Phragmites using publically available remotely sensed images. Photogramm Eng Remote Sens 81(1):69–78

    Article  Google Scholar 

  77. Yang J (2009) Mapping salt marsh vegetation by integrating hyperspectral imagery and LiDAR remote sensing. In: Wang Y (ed) Remote sensing of coastal environments. CRC Press, Boca Raton, pp 173–186

    Google Scholar 

Download references

Acknowledgements

This work was made possible through financial support from the North Atlantic Landscape Conservation Cooperative and the United States Fish and Wildlife Service (USFWS) Northeast Region Science Applications (#24), and the National Institute of Food and Agriculture, Hatch Project Number ME0-21710 through the Maine Agricultural & Forest Experiment Station. This is Maine State Agricultural and Forest Experimentation Station Publication # 3590. We would like to thank all Saltmarsh Habitat and Avian Research Program (SHARP) field technicians who collected field training data for this effort, and all participating landowners that allowed access to their properties for surveying. We also thank Janet Leese for countless hours spent digitizing training polygons in the lab. Comments from Erin and Kasey Legaard, D. Rosco, N. Hanson, and the Olsen Lab substantially improved the methods described here.

Author information

Affiliations

Authors

Corresponding author

Correspondence to Maureen D. Correll.

Electronic supplementary material

ESM 1

(DOCX 23 kb)

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Correll, M.D., Hantson, W., Hodgman, T.P. et al. Fine-Scale Mapping of Coastal Plant Communities in the Northeastern USA. Wetlands 39, 17–28 (2019). https://doi.org/10.1007/s13157-018-1028-3

Download citation

Keywords

  • High marsh
  • NAIP
  • Random Forest
  • Remote sensing
  • Spartina
  • Tidal marsh