Fig 1.
Study sites off the coasts of New York (Rockaway Peninsula, Fire Island), New Jersey (Pullen and Long Beach islands), and Virginia (Cedar Island) USA, where we modeled piping plover habitat before and after Hurricane Sandy using a Bayesian network approach.
Fig 2.
Configuration of nodes and edges in the Plover Habitat Bayesian network used to predictively map habitat for piping plovers.
Prior probability distributions for this network were derived from data collected at piping plover nests and random points at sites in 2014 and 2015 using the iPlover data collection application. The network as shown in this figure illustrates prior probability distributions for instances where nests were present (i.e., characteristics associated with habitat).
Table 1.
Landscape variables considered as nodes in the Plover Habitat Bayesian network used to model habitat for piping plovers along the coasts of New York, New Jersey, and Virginia.
Variables were associated with data collected at piping plover nest and random points through the smartphone application iPlover and were also individually mapped for our study areas. Continuous variables were discretized into bins shown in Fig 2.
Table 2.
Predicted habitat, according to the Plover Habitat Bayesian network1, for piping plover nesting before Hurricane Sandy, immediately following Sandy, and ca. 2 years post-Sandy at study areas in New York, New Jersey, and Virginia, USA.
For reference, we indicate the number of nests2 that were established in what the model predicted to be habitat, non-habitat, or uncertain landcover (where a suitability designation could not be made). Sites are listed in decreasing level of anthropogenic development.
Table 3.
Changes in piping plover habitat availability1 for landcover conditions present before Hurricane Sandy, immediately following Sandy, and ca. 2 years post-Sandy at study areas in New York, New Jersey, and Virginia USA.
For reference, we indicate the number of nests that were established in habitat where landcover became habitat, became non-habitat, did not change in suitability, or became uncertain in suitability level between the pre-Sandy to post-Sandy (nests: 2013 breeding season) and between the post-Sandy and ca. 2-year post-Sandy (nests: 2014/2015 breeding season) study periods. Sites are listed in decreasing level of human development.
Fig 3.
(a) Changes in the area of piping plover habitat, as predicted by the Plover Habitat Bayesian network, before Hurricane Sandy, immediately after the storm, and ca. 2 years after the storm at study sites in New York (Rockaway Peninsula, Fire Island), New Jersey (Long Beach and Pullen islands), and Virginia (Cedar Island). Sites are shown in order of proportion of human development on the study site, from Long Beach Island (most developed) to Cedar Island (undeveloped), and bars are divided as the area of habitat found along developed portions of the coastline (i.e., where housing communities, recreational infrastructure, or other human structures directly abut the shoreline; light portions of bars) and habitat found along undeveloped portions of the coastline (dark portions of bars). We also show piping plover (b) population size and (c) average productivity at each study site to illustrate how population dynamics changed with habitat amount after the storm. Population data were collected by outside partners as part of unrelated research and monitoring efforts [67–69].
Fig 4.
Examples of habitat change before Hurricane Sandy, immediately following Sandy, and ca. 2 years post-Sandy on the Fire Island, New York, study area1.
Maps (a)-(c) show habitat in the year corresponding with the aerial imagery used for analysis, with piping plover nest points corresponding to the breeding season for which habitat was available2. Map (d) shows changes in habitat that ostensibly resulted from Hurricane Sandy, while map (e) shows changes in habitat since the hurricane occurred3. 1In these maps, the habitat designation was predicted according to the Plover Habitat Bayesian network. Landcover that was very likely habitat had a ≥ 0.90 probability of being habitat, landcover likely habitat had a probability of 0.66–0.90, landcover as likely as not habitat had a probability of 0.3–0.66, and landcover unlikely habitat had a probability ≤ 0.33. We further differentiated the ‘as likely as not habitat’ category to show (1) areas of true model uncertainty (p = 0.5), where the combination of habitat variables in a given landscape pixel was either not present in the supplemented iPlover dataset or where there was missing information in the ‘case’ presented to the BN for analysis in that pixel and (2) areas where landcover may be of marginal suitability (0.50 < p < 0.66; 0.33 < p < 0.50). 2Nest locations contributed by [68]. 3In these maps, habitat change was predicted by the Plover Habitat Bayesian network, where (d) considers change from the pre- to the post-Sandy study periods and (e) considers change from the post- to the 2 years post-Sandy study periods. Landcover that ‘became habitat’ transformed from as likely as not habitat or non-habitat to habitat; landcover that ‘became uncertain’ transformed from habitat to as likely as not habitat; and landcover that ‘became non-habitat’ transformed from habitat to non-habitat. Landcover that ‘did not change in suitability’ was either habitat, as likely as not habitat, or non-habitat in both study periods considered OR transformed from as likely as not habitat to non-habitat (or vice versa) between study periods.
Fig 5.
Examples of habitat change before Hurricane Sandy, immediately following Sandy, and ca. 2 years post-Sandy within the Rockaway Peninsula, New York, study area.
Maps (a)-(c) show habitat in the year corresponding with the aerial imagery used for analysis, with piping plover nest points corresponding to the breeding season for which habitat was available1. Map (d) shows changes in habitat that ostensibly resulted from Hurricane Sandy, while map (e) shows changes in habitat since the hurricane occurred2. 1Nest points contributed by [68] 2See footnotes accompanying Fig 4.
Fig 6.
Examples of habitat change before Hurricane Sandy, immediately following Sandy, and ca. 2 years post-Sandy in a portion of the Little Beach Unit of the Edwin B. Forsythe National Wildlife Refuge on Pullen Island, New Jersey.
Maps (a)-(c) show habitat in the year corresponding with the aerial imagery used for analysis, with piping plover nest points corresponding to the breeding season for which habitat was available1. Map (d) shows changes in habitat that ostensibly resulted from Hurricane Sandy, while map (e) shows changes in habitat since the hurricane occurred2. 1Nest points contributed by [67]. 2See footnotes accompanying Fig 4.
Fig 7.
Examples of habitat change before Hurricane Sandy, immediately following Sandy, and ca. 2 years post-Sandy in the Holgate Unit of the Edwin B. Forsythe National Wildlife Refuge on Long Beach Island, New Jersey.
Maps (a)-(c) show habitat in the year corresponding with the aerial imagery used for analysis, with piping plover nest points corresponding to the breeding season for which habitat was available1. Map (d) shows changes in habitat that ostensibly resulted from Hurricane Sandy, while map (e) shows changes in habitat since the hurricane occurred2. 1Nest points contributed by [67]. 2See footnotes accompanying Fig 4.
Fig 8.
Examples of habitat change before Hurricane Sandy, immediately following Sandy, and ca. 2 years post-Sandy on Cedar Island, Virginia.
Maps (a)-(c) show habitat in the year corresponding with the aerial imagery used for analysis, with piping plover nest points corresponding to the breeding season for which habitat was available1. Map (d) shows changes in habitat that ostensibly resulted from Hurricane Sandy, while map (e) shows changes in habitat since the hurricane occurred2. 1Nest points contributed by [69]. 2See footnotes accompanying Fig 4.
Fig 9.
(a) Area of piping plover nesting habitat, according to predictions by the Plover Habitat Bayesian network, that was created immediately after Hurricane Sandy (i.e., newly created habitat between the pre- and post-Sandy periods). Bars showing habitat area are divided based on the geomorphic settings (defined in [34]) that were present after Hurricane Sandy according to October/November 2012 aerial photography and lidar. Because habitat continued to increase on Long Beach Island (LBI) and Pullen Island between the post- and ca. 2 years post-Sandy periods, we also include the amount and proportions of habitat created during that second time period for those sites (labeled with ‘12–14’ to denote the second period of habitat gain). Geomorphic settings for ‘12–14’ bars reflect settings available in the ca. post-Sandy period (June 2014). We combined the ‘beach’ and ‘backshore’ settings here for display purposes. Colors comprising bars in (a) are defined by colors of setting names in (b), and sites are listed from most to least developed.
Fig 10.
Proposed relationship between the density of human development across a barrier island and the amount of early successional habitat created by a single low intensity (blue line) or high intensity (red line) storm event.
In this relationship, no single storm event will have a major impact on the amount of early successional habitat on low elevation, undeveloped islands because smaller intensity storms maintain maximum amounts of habitat through time. A single storm will also have a minor impact on habitat on islands with a high density of human development because shoreline modifications prevent overwash and encourage erosion, minimizing the ‘opportunity space’ for new habitats to be created. On high elevation, undeveloped islands (dashed lines) or on islands with moderate levels of development (middle of x-axis), less frequent high-intensity storms are needed to create overwash and new habitats; however, undeveloped stretches of shoreline provide opportunities for habitat to be created. As a result, a large storm like Hurricane Sandy can have a major impact in the amount of habitat available.