1. Introduction
Forests in urbanizing coastal zones are undergoing unprecedented change as urbanization accelerates simultaneously with invasive species spread and climate change. Coastal ecosystems are vulnerable to global climate change as sea levels rise and storm surge events increase [
1,
2,
3,
4]. They are also at risk for increased urbanization and development pressures because many urban centers occur in coastal areas. Seventy-five percent of the world’s largest cities—and 65% of the densest cities—are located in coastal zones [
5], and population growth is projected to increase in coastal cities [
6]. At the same time, cities are hotspots for introduction of non-native species [
7]. Streamside forests of urbanizing coastal regions lie at the nexus of these global changes, yet the extent and potential impact of these combined stressors on riparian urban forests and on their ability to provide valuable water filtration and support for biodiversity is currently unknown. These forests are “squeezed” from all sides by adjacent urbanization on land, rising sea levels and storm surge, and from within by non-native invasive plants (
Figure 1). Understanding relationships between urban riparian forests and the combined effects of urbanization, climate change, and invasive species requires a social-ecological approach.
The structure, composition, and ecosystem functions of forest patches are affected by landscape-scale fragmentation, isolation, and adjacent land use change due to urban land use transformation [
8,
9], and also by parcel-scale human activities [
10,
11]. Management actions affecting these forests range from biodiversity conservation to lumber production, and also include actions based on social perceptions and motivations. Land owners or neighbors may choose to thin or cut forests to remove “messy” vegetation [
12,
13] or improve views of waterways. Riparian vegetation is constricted in urban areas by hardened shorelines, and by housing development situated as near as possible to valued water views. Proximity of the built environment to the water can prevent migration of buffering vegetation inland with sea level rise, leading to loss of needed nutrient capture capacity for urban and suburban runoff, as well as decreased habitat value. Reduction of riparian vegetation and connectivity between streams and floodplains can further exacerbate urbanization impacts on streams by reducing nutrient retention and flood water retention [
14,
15,
16]. Both urban conditions and rising saline waters [
17,
18] affect species composition, which is an underlying determinant of forest health and function.
Predicted global sea level rise and storm surges will increase salinity in coastal urban surface waters, altering adjacent forest function and health. The effects of sea level rise may be compounded further by increased storm surges, salt water intrusion into groundwater and changes in sediment flow [
19]. Sea level rise is a global trend with effects that vary widely on a regional level [
20], and all levels of government are now developing strategies for climate change adaptation. Vegetation along urban and suburban shorelines is increasingly considered to be important for protective function in relation to storm surge, leading coastal cities to increase buffer vegetation for this purpose [
21].
While horizontal stressors from urbanization and sea level rise impact these forests, urban biophysical conditions act as a filter on regional species pools [
22] as non-native species are continually introduced into cities [
23,
24]. Top-down impacts on forest regeneration [
25,
26] and bottom-up alterations to soil processes [
27,
28,
29] due to invasive plants make urban riparian forests especially vulnerable. Cities are centers of species introduction, some of which are or become invasive and may spread from cities into surrounding landscapes [
30]. Fragmentation and loss of habitats due to urban land use transformation create high edge-to-interior ratios in urban forest patches [
31], increasing the permeability of these forests to dispersal of invasive plants. Linear forest strips, such as buffers along waterways and roads, are particularly susceptible [
32]. Invasive plant species can not only reduce native biodiversity and the resources available to wildlife, but in the case of invasive vines in urban and suburban forests, they can increase tree mortality and prevent forest regeneration [
33], often in concert with high abundance of herbivores in urbanized regions where predators have been extirpated [
34].
The goal of this study was to understand the extent of riparian buffer forests vulnerable to urbanization, invasive species, and sea level rise impacts (hereafter, “squeezed forests”) that have the potential to alter ecosystem function and health. We conducted a spatial analysis to model the extent and location of these stressors in urban and suburban environments and identified forested areas of greatest vulnerability to urbanization pressures, sea level rise, and storm surge. Preliminary observations indicated that invasive plants are widespread in these forests, but existing records of invasive plant species abundance were not spatially contiguous. To understand the extent and impact of invasive plant pressure, we conducted field sampling of forest composition, forest structure, and local-scale human impacts.
This research addresses existing information gaps at the intersection of multiple large-scale causes of ecosystem change. We asked the following questions:
What is the spatial extent of forests squeezed by urban land use and potential loss to sea level rise and storm surge? Coastal areas of the eastern United States are affected by increasing urbanization, rising sea levels, and increasing inland impacts of storm surges. Estuarine and riverine forests that serve important buffer functions are threatened by both. We expected forests affected by these combined stressors to be common across the study area, and we expected that these forests would also be impacted by invasive plants.
Does non-native plant invasion vary with differences in landscape context and disturbance? Land use and land cover surrounding forest patches influences their composition in complex ways, including species introductions, isolation of populations in habitats, and local microclimate alteration. We expect plant communities of squeezed forests to exhibit effects of local human impacts, surrounding conditions, and land use and land cover adjacent to the forest patch, such as high rates of invasive plant cover and indicators of human use and human-caused ecological disturbance.
Study Area
The urbanized coastal mid-Atlantic region of the United States sits at the intersection of these major forces of ecological change. This segment of the Eastern Seaboard is home to more than 40 million people in a dense matrix of cities and towns stretching along the Atlantic coast between Virginia and New Jersey, including Philadelphia, Washington, Wilmington, Trenton, and Baltimore (
Figure 2). The region has experienced a long-term trend of population growth. From 2000 to 2010 (the most recent census period), Delaware’s population increased by 14.6% (114,334 people), Virginia’s by 13% (922,509), Maryland’s by 9% (477,066), Pennsylvania’s by 3.4% (421,325), and New Jersey’s by 4.5% (377,544 people) [
35]. Population centers are clustered along the coastline, and urban and suburban development has increased in coastal areas at a faster pace than inland. Urban land use transformation in the United States is projected to result in loss of ca. 118,000 km
2 of forest by 2050, with states in this region already exhibiting the nation’s highest rates of urbanization and one—New Jersey—to become more than 50% urban land [
36].
The mid-Atlantic region also contains two large biologically and economically important estuarine systems: the Chesapeake and Delaware Bays. This study is focused on forest patches within the boundaries of the Chesapeake and Delaware Bay watersheds (
Figure 2 and
Figure 3), which together drain an area of nearly ten million hectares. In the Chesapeake Bay watershed, one-third of urban development land use transformation in recent decades has resulted in forest loss, and the fastest-growing urban areas surrounded by forested land have experienced the most loss of forest to impervious surfaces [
37].
This region is also particularly vulnerable to climate change. On the Delmarva Peninsula, which sits between the Chesapeake Bay and Delaware Bay and is shared by the states of Delaware, Maryland, and Virginia, sea level rise is compounded by land subsidence. The peninsula has been sinking at the rate of 1.3 mm per year for the last 1000 to 2000 years [
38]. Consequently, while global sea levels have risen 4–8 inches in the last century, the peninsula’s relative sea level rise (including land subsidence) was approximately 12 inches and projections are that the effective sea level rise may be up to an additional 2 to 5 feet in the next 100 years [
39].
2. Materials and Methods
To understand the degree to which riparian forests are vulnerable to interactive effects of these combined stressors, we examined riparian forest patches likely to be affected by storm surge and sea level rise along a gradient of city size. We conducted analyses at landscape, forest patch, and plot scales. First, we used geospatial data to estimate the extent of forest patches likely to be affected by multiple stressors under varying scenarios of sea level rise at a regional scale. We then selected and sampled 100 sites that would be affected by storm surge under the most conservative model. In these sites, we examined vegetation and indicators of disturbance at a plot scale.
2.1. Squeeze Modeling
To assess the potential extent of urban riparian forests vulnerable to simultaneous stresses of urbanization, sea level rise, and storm surge in this region, we integrated national land use and land cover data with inundation risk modeling.
2.1.1. Modeling Urbanization and Forest Extent
ESRI ArcGIS
® 10.5.1 software [
40] was used to store and analyze over 350 GB of geospatial data. Using the Watershed Boundary Dataset [
41], we delineated watersheds draining into Delaware Bay and the Chesapeake Bay for geospatial analyses, comprising a total study area of 9,829,678 ha. We used the U.S. National Landcover Database (NLCD) 2011 land cover product [
42] to identify forest patches adjacent to urban development. Selected forest patches were located within 500 m of areas classified by NLCD (2011) as Developed Medium Intensity (impervious surface cover 50–79%, commonly including medium to high density single-family housing units) or Developed High Intensity (impervious surface cover 80–100%, commonly including apartment complexes, row houses and commercial/industrial areas). Minimum forest patch size was set at 0.1 ha (0.25 ac), following [
43]; for context, this is also the size of an average American suburban residential yard. Forest type classes included Deciduous Forest, Evergreen Forest, Mixed Forest, and Woody Wetlands. From this analysis, we derived the number, size, and public v. private ownership of forest patches. We also derived land cover type within a 500 m buffer of sites selected for ground sampling, and proportional and absolute perimeter adjoining land cover types of each forest patch. Additional land cover types included Cultivated Crops, Developed Open Space, Hay/Pasture, Emergent Herbaceous vegetation, Shrub/Scrub vegetation, and Barren Land (NLCD 2011).
2.1.2. Modeling Sea Level Rise and Storm Surge Vulnerability
To model the combined stressors of sea level rise and storm surge, we synthesized four datasets: a Digital Elevation Model (DEM) of the area, storm surge surface models, sea level rise surface models, and the National Hydrography Dataset. To estimate storm surge, we merged Sea, Lake, and Overland Surges from Hurricanes (SLOSH) storm surface models for New York, Delaware Bay, and Chesapeake Bay basins [
44,
45,
46]. Ordinary kriging was used to extend the storm surge surface inland, and 10 m rasters estimating storm surges for category 1, 2, and 4 hurricane events were generated. Sea level rise (SLR) was estimated from a DEM mosaic of the NOAA Coastal Services Center Coastal Inundation Digital Elevation Model [
47] and the US Geological Survey standard DEM 1/9 arc-second product [
48] resampled to 10 m using cubic convolution. Sea level rise was estimated for 0.0 m, 0.6 m, and 1.8 m sea level rise scenarios. By combining these layers, 9 storm surge-sea level rise surfaces were generated (Cat 1, 2, and 4 storms) x (0.0 m, 0.6 m, and 1.8 m SLR).
Initial inundation extent was estimated by subtracting DEM values from storm surge/sea level rise surfaces. This initial output was evaluated for connectivity (inundation) using an 8 sided neighborhood rule and connected cells were extracted. Area water features identified in the National Hydrography Dataset [
49] were dissolved and converted to a 10 m raster. This layer and areas of land not typically inundated derived from the DEM hydrologic break lines were combined to form a mask to extract connected cells. Inundated areas with less than 95% confidence were removed to form the final inundation extents.
We note that the resulting model is a “gentle flooding model” (water rises and covers)—no effort was made to model backpressure effects—and inundation is shown as it would appear during the mean high tides (excludes wind driven tides). We also did not account for erosion/deposition, subsidence/uplift, future changes in hydrodynamics, or preexisting conditions such as soil moisture and river input.
2.2. Field Rapid Assessment of Stressors and Forest Conditions
To determine the extent and intensity of invasive plant pressure on forests vulnerable to urbanization and flooding due to storm surge and sea level rise, we conducted a rapid field assessment. This assessment also documented evidence of human activity and non-human ecological disturbances.
2.2.1. Selection and Location of Sampling Points
We selected 20 municipalities representing the range of municipality sizes occurring in the study region (
Figure 3 and
Table A1) and located 5 points in each municipality (total: 100 sites) for field-based measurement of invasive plant species, vegetation composition and structure, and indicators of impact from human management and use. Selected sites were located on public lands to facilitate access for observation. To identify areas of greatest potential vulnerability to combined stressors, forest patches were selected based on adjacent urbanization and potential inundation modeling. Adjacent urbanization was defined using NLCD 2011 [
42] as described in
Section 2.1.1. above. Sampling points were located within the boundary of the most conservative scenario for storm surge and sea level rise: Category 1 hurricane storm surge with no (zero) sea level rise; these locations would be inundated under all scenarios. Center points of sampling plots were randomly assigned within these patches, including additional points for use when encountering barriers preventing access to a coordinate. Field technicians located sampling points using GPS. Where the location of a coordinate in the field was not entirely inside a forest patch due to error inherent in remote sensing and/or change in patch boundaries, plot center was relocated to the interior of the forest patch.
2.2.2. Field Sampling: Vegetation
Circular plots of 400 m
2 (11.3 m radius) were established around the randomized coordinates. A laser hypsometer was used to determine plot boundary. In each plot, all trees > 2.5 cm diameter were identified to species. Greatest degree of vine coverage on trees, dominant vine species, dominant herbaceous species, and indicators of disturbance were recorded for the entire plot. Degree of vine coverage was based on the Schumaker vine invasion index [
50]. Categories of coverage were: 0: no lianas climbing trees, 1: vines at the base of tree bole, 2: vines covering tree bole, 3: partial canopy coverage, and 4: complete canopy coverage. In a 20 m
2 circular subplot at the center of each plot, ground layer cover of all species was visually estimated using ⅛ radial increments. Within subplots, individual trees of all size classes (including seedlings < 1m in height and saplings > 1m in height and < 2.5 cm DBH) and all shrub stems emerging from the ground were counted and identified. Species were categorized as native or non-native to the mid-Atlantic region following the USDA PLANTS Database [
51]. Non-native species were categorized as invasive if they appeared on a state invasive plant list of one of the states in the study region [
52,
53,
54,
55,
56]. All invasive species were non-native. Field identification followed [
55,
57,
58,
59,
60]; taxonomy follows USDA PLANTS Database [
51].
2.2.3. Field Sampling: Disturbance Indicators
Indicators of human and non-human ecological disturbance were developed from regional and urban assessments of environmental impact and were designed to be compatible with these measures for cross-comparability. These included assessments utilized in rural riparian and forest systems [
61,
62,
63,
64], assessments used by urban land managers [
65,
66,
67,
68], studies examining human impacts on urban forest systems [
10,
69], and indicators of non-human forest disturbance [
70,
71]. Evidence of human activity included vegetation manipulation such as cutting or planting; dumping or accumulation of household waste; digging, shoreline alteration, and earth moving; trails, roads, and trampling; recreational equipment and camping; fencing; and building. Non-human disturbances included canopy gaps due to fallen trees and high levels of herbivory. High abundance of white-tailed deer (
Odocoileus virginicus) in the region can affect forest regeneration [
34,
72,
73,
74,
75], so signs of deer (e.g., tracks, browse, scat) were also recorded.
2.3. Statistical Analysis
Statistical analyses were performed in R version 3.6.1 (R Core Team, 2016). All tests for significance are reported at the α = 0.05 critical value, and, in a few cases, the α < 0.10 critical value is reported to identify potential trends. Pearson correlation analysis was used to assess whether the species richness of invasive vines present correlated with the intensity of vine coverage (trunk and canopy cover) across forest patches. To assess the influence of local human activities on invasive plant species, we performed bivariate linear regression analysis between the abundance of invasive trees, saplings, seedlings, shrubs, or total species and the observable signs of natural and human disturbance within forest plots. To assess human influences on invasive plant abundance across spatial scales, we performed pairwise linear regression analysis between the richness of invasive plants or total number of invasive plant species by growth form (i.e., tree, shrub, vine, forb, graminoid) and (1) the proportion of urban, agricultural, and forest land use/land cover within 500 m of the forest plot center, (2) the proportion of urban, agricultural, and forest land use/land cover adjacent to the forest patch (i.e., along forest perimeter), and (3) the municipal population size for each city. Finally, we assessed the relationship between total municipal population for each city and (1) the richness of invasive plants, (2) the total number of invasive plant species by growth form, and (3) the abundance of invasive trees, saplings, seedlings, or shrubs and the total municipal population for each city using linear regression analysis.
5. Conclusions
Here, we have established the spatial extent of forests currently under broad-scale environmental pressures in a major urban region that spans two large and productive estuarine bays, and revealed the role of invasive plants in these forests. By encompassing three major sources of ecosystem change, we provide new insight into how the interaction of these stressors might affect the function of urban riparian buffers, and potential for change. Our results emphasize the importance of protection and restoration of forests in urban regions and point to areas for future work.
Both urbanization pressures and human-caused ecological disturbances are most appropriately understood as part of a social-ecological system. Here, field data from publicly owned forest patches demonstrate impacts of multiple stressors. This approach could be extended to include forests under private ownership to better understand the influence of fine-scale management decision making. Collaboration with social scientists to identify social drivers of decision making at the scale of individual properties that affect buffer function and perception of buffer management, combined with identification of appropriate conservation and management targets for urban riparian buffer forests, would speed development of new approaches to buffer edge management. The integration of environmental and social benefits with improved buffer condition and function can maximize function for habitat and water quality while meeting the needs of urban communities.
In highly urban regions, remnant, regenerating, and emerging riparian ecosystems are simultaneously subject to the multiple stressors of rising seas, increasingly strong storms, urban development, and invasive species spread. While inland development limits their extent and alters their condition, sea level rise squeezes these forest patches from the shore, and invasive plants cover forest canopies and suppress tree regeneration. Where pressures combine, there is a greater likelihood of loss of habitat and water quality buffer functions but also greater opportunity and potential for effective management to mitigate the loss.