The area studied included 14 states of the New England and Mid-Atlantic regions of the United States and three provinces of Atlantic Canada (here after the region): Maine, New Hampshire, Vermont, New York, Massachusetts, Rhode Island, Connecticut, Pennsylvania, Delaware, New Jersey, Maryland, Ohio, West Virginia, Virginia, New Brunswick, Nova Scotia, Prince Edward Island, and portions of Quebec. The area covers 870,247 km2, supports over 13,500 species of plants, vertebrates, and macro-invertebrates, and has a wide diversity of lithologies and topography.
The boundaries of TNC's terrestrial ecoregions (TNC 2012) were used as a stratifying framework. The ecoregions were developed in conjunction with the USDA Forest Service and are a modification of Keys et al.'s ecoregions (1995). Six ecoregions were fully contained within the area of interest: Central Appalachian, Chesapeake Bay, High Allegheny Plateau, Lower New England, North Atlantic Coast, and Northern Appalachian/Acadian. Six other ecoregions had a portion of their full extent included in the region.
Mapping Geophysical Settings
To map the region's geophysical settings, we compiled bedrock and surficial geology data sets for each state and province at the scale of 1:125,000. We grouped the >200 bedrock types into nine lithogeochemical classes based on genesis, chemistry, and weathering properties (Table 1). Elevation data were taken directly from a 30-m digital elevation model (DEM) (Gesch 2007) and classified into six elevation zones corresponding to major changes in dominant vegetation (Table 1). Geology classes and elevation zones matched those described in Anderson and Ferree (2010).
Table 1. The 30 geophysical settings used as a framework for assessing site resilience to climate change relative to geology classes and elevation zones
| ||Elevation rangeb|
|Lithologya||Coastal 0–6 m||Low 6–244 m||Mid 244–762 m||High 762–1097 m||Alpine >1097 m|
|Mudstone, claystone, siltstone, nonfissile shale, sandstone, breccia, conglomerate, greywake, Arenites, slate, phyllite, pelite, schist, pelitic schist, granofel, quartzite|| || || || || |
|Fissile shale|| || || || || |
|Limestone, dolomite, dolostone, other carbonate-rich clastic rocks, marble|| || || || || |
|Calcareous shale and sandstone, calc-silicate granofel, calcareous schist and phyllite|| || || || || |
|Granite, granodiorite, rhyolite, felsite, pegmatite,(granitic gneiss, charnocktites, migmatites|| || || || || |
|Anorthosite, gabbro, diabase, basalt, diorite, andesite, syenite, trachyte, greenstone, amphibolites, epidiorite, granulite, bostonite, essexite|| || || || || |
|Serpentine, soapstone, pyroxenite, dunite, peridotite, talc schist|| || || || || |
|Unconsolidated sand, gravel, pebble, till|| || || || || |
|Unconsolidated mud, clay, drift, ancient lake deposits|| || || || || |
|Roughly equal mixtures of two|| ||L:GRAN/CALC|| || || |
|geology classes|| ||L:GRAN/COARSE|| || || |
|Very steep slopes at any elevation||N/A||STEEP||STEEP||STEEP||STEEP|
In the GIS, all information was summarized on a grid of 156,581 hexagons on which each hexagon was 405 hectares. This hexagon size allowed us to maintain relatively fine-scale detail while accounting for spatial error in the location of features such as rare species or bedrock outcrops. The hexagon grid fully tessellated the entire region and the individual hexagons (referred to as sites) can readily aggregate to delimit larger areas.
We used a cluster analysis to assess the geophysical similarity between hexagons and group them into geophysical setting types. For each hexagon, we tabulated the type and abundance of each geology class, elevation zone, and landform type (described below). Then we performed a hierarchical cluster analysis in PC-ORD (McCune & Mefford 1999) with the Sorensen similarity index and a flexible beta linkage technique with beta set at 25 to group similar hexagons. We defined the geophysical setting groups by studying the dissimilarity scores and identifying the relatively homogenous clusters.
Estimating Site Resilience
We developed separate estimates of landscape diversity and connectedness for every 30-m cell and then combined these to estimate resilience for each 30-m cell. Subsequently, for each hexagon we calculated the mean and standard deviation of each individual and combined factor based on the 30-m grid cells contained within the hexagon.
Landscape diversity summarized the variety of landforms, elevation range, and density of wetlands in a given search area. The landform variety component was based on a spatially comprehensive landform model that delineated 11 surface features: cliff and steep slope, summit and ridge-top, northeast facing side slope, southwest facing side slope, cove and slope bottom, low hill, low hilltop flat, valley and toe slope, dry flat, wet flat, and water (Fig. 1). The model, an expansion of Conacher and Darymple's (1977) nine-unit land surface model, delimits recognizable landforms as combinations of slope, land position, aspect, and moisture accumulation that correspond to local topographic environments with distinct combinations of moisture, radiant energy, and deposition. Technical methods for mapping landforms were based on Fels and Matson (1996) and are described in detail in Anderson et al. (2012). The model was derived from a 30-m DEM.
Figure 1. The full landform model mapped for Mount Mansfield, Vermont (U.S.A.), showing the estimated number of microclimates (8 within the black circle); the progression from flats to slopes (small maps); and how the full landform model lies across the landscape (large map). The relative size of the 40-ha focal area is shown in the black circle (adapted from Anderson et al. 2012).
Download figure to PowerPoint
We used a focal variety analysis to tabulate the number of landforms within a 40-ha circular area around every 30-m cell. The size of the search area was derived by systematically testing many possible sizes (4–400 ha) to find the one with the highest between-cell variance and thus the maximum discrimination between sites (i.e., too large and all sites had all landforms, too small and all sites had only one landform). For consistency, we used a 40-ha area for all the landscape diversity metrics. Because sites with a larger variety of landforms provide more microclimate options, cells were scored by their landform number from 1 to 11 (Fig. 1). Our assumption was that most plant and vertebrate populations could access this relatively small neighborhood to locate suitable microclimates.
To assess the local elevation range, we used a focal range analysis on the DEM to tabulate the range in elevation within a 40-ha circular search area around each 30 m cell. Cells were scored by their elevation range (1–795 m) and log transformed to approximate a normal distribution.
We combined the results into a single metric and weighted landform variety twice as much as elevation range because the landform model delineates contrasting micro-climates more precisely than elevation. Before combining, we transformed both metrics to standardized normal distributions. The final index was landscape diversity = (2*landform variety + 1*elevation range)/3.
In extremely flat areas, the landscape diversity index could not provide useful discrimination between many equivalent cells. For these areas (<0.5% slope), we added wetland density as a finer-scale indicator of subtle micro-topographic features not captured by the wet flat element in the landform model. We combined the National Wetland Inventory (USFW 2012), National Land Cover Database wetlands (NLCD 2001), and the Northern Appalachian/Acadian wetlands (TNC 2012) into a single data set and used a focal function to calculate the density of surrounding wetlands for every 30-m cell in the region. To account for the flat topography, we used two circular search areas (a 40-ha area and a 400-ha area) and combined the results into a weighted index that gave twice the weight to the 40-ha search area. Wetland density was defined as total area of wetlands divided by the size of the search area, and the index was wetland density = (2*density of 40 ha + 1*density of 400 ha)/2.
We converted the scores to a standard normal distribution. In cells with <5% slope, we added the wetland density scores to the two other landscape diversity metrics as follows: landscape diversity in flats = (2*landform variety +1*elevation range + 1*wetland density)/4.
Local connectedness was designed to estimate the degree of permeability, or conversely the degree of resistance, surrounding each cell in the region. We used a resistant kernel analysis and software created by the UMASS CAPs program to measure connectedness (Compton et al. 2007). The algorithm measures the connectivity of a focal cell to its ecological neighborhood when the cell is viewed as a source of movement radiating out in all directions.
The metric is built on the assumption that the permeability of two adjacent cells increases as their ecological similarity increases and decreases as their similarity decreases. Contrasting elements are scored with resistance weights to reflect differences in structure, composition, degree of development, or use. The theoretical spread of a species or process outward from a focal cell is a function of the resistance values of the neighboring cells and their distance from the focal cell out to a maximum distance of 3 km. The local connectedness score for a cell was equal to the area of spread accounting for resistance divided by the theoretical area of spread if there were no resistance.
Our resistance surface was based on a classified land use map with roads and railroads embedded in the grid (NLCD 2001; Tele Atlas North America, ESRI 2012). We simplified the land cover into six basic elements and assigned resistance weights to each category based on a simplified version of Compton's similarity index, where natural land was given the lowest resistance weight (10) and high intensity developed land was given the highest weight (100). Minor roads were overlaid on the grid and added 10 points of resistance to the cell containing them. We tested the sensitivity of the outcomes to the resistance weights by running the analysis for three test areas and systematically changing the weights. We finalized the weights by reviewing the test results with a team of state-based conservation scientists until we reached agreement. The final weights were as follows (NLCD classes in parenthesis): 10, natural lands and water (evergreen, deciduous, and mixed forest, shrub or scrub, grassland, woody and herbaceous wetland, water); 50, unnatural barrens (barren); 80, agricultural or modified lands (pasture, cultivated); 90, low intensity development (developed open space, low intensity developed); 100, high intensity development (medium intensity developed, high intensity developed, major roads). We aggregated the 30 m resistance surface to a grid of 90-m cells to reduce the considerable processing time before running the resistant kernel algorithm and computing the score for each cell. Cell scores ranged from 0 to 1 and were converted to a standard normal distribution for the region.
We estimated a site resilience score by summing the landscape diversity and local connectedness grids into a single metric. We used standard normalized values and gave equal weight to each factor: estimate of resilience score = (landscape diversity + local connectedness)/2. Our assumption was that the two factors are complementary and mutually reinforcing (e.g., micro-climatic diversity has more value if the area is connected and visa versa). The final output was a 30-m grid of estimated site resilience.
For each hexagon, we calculated the mean and standard deviation of landscape diversity, local connectedness, and site resilience based on the 30-m grid cells contained within the hexagon. We transformed the hexagon scores to standard normal distributions, normalizing the values to three extents: region, through the mean and standard deviation of all hexagons in the region; geophysical settings through the mean and standard deviation of all hexagons of each geophysical setting type; and setting within ecoregions the mean and standard deviation of all geophysical setting types within each ecoregion. For analysis purposes, we considered the mean the range of values from SD –0.5 to 0.5 and high-scoring hexagons those hexagons with mean resilience scores >SD 0.5.
Biodiversity and Secured Lands
Sites with high-quality biodiversity features were compiled from nine ecoregional assessments completed by TNC (2012). For each hexagon, we summarized the type and amount of high-quality biodiversity features (species and communities) within it and the amount of land permanently secured for conservation. We converted biodiversity data sets (points and polygons) to points based on the polygon's centroid and used only occurrences with precise locations. The TNC portfolio of sites contains a selective subset of all features in the region including the locations of 4592 viable populations of rare species and 2170 high-quality examples of representative natural communities. The data have a high degree of consistency because they were reviewed by experts within each ecoregion and assessed with a standard set of criteria. Viability criteria were based on the size, condition, and landscape context of the occurrence. The assessments were performed by teams of scientists, including experts on various taxa, and the final selection of sites was based solely on the quality of the biodiversity feature and a set of distribution and numeric goals. No optimization software was used to select sites. The portfolios represent a set of sites that, if conserved, would collectively protect the full biological diversity of an ecoregion.
To compare the portfolio sites with the resilience scores, we categorized the hexagons into standard deviation groups (SD +2.5, +1.5, +0.5, –0.5, –1.5, and –2.5) based on the mean resilience score of the hexagon. For this step, hexagons were scored with respect to their setting and ecoregion and with respect to all sites in the region and were assigned whichever score was higher. This adjustment, affecting 9% of hexagons, corrected for the fact that some of the highest scoring places in the region were only average for their setting because some settings had such inherently high resilience scores.
Information on land securement was compiled from state, federal, local, and private sources and was standardized across states (details in Anderson and Olivero Sheldon 2011). The data set included only land permanently secured against conversion to development and contained over 9.8 million ha of public and private lands permanently protected by fee or easement. Only land intended for nature conservation (GAP status 1 or 2) or multiple uses (GAP status 3) was included. Data for the TNC portfolio and secured lands were for the U.S. portion of the study only.