https://orcid.org/0009-0001-1429-8085
Figures
Abstract
Landscape connectivity is often negatively impacted by road networks that fragment habitat and result in genetic and demographic consequences for wildlife. Existing roadway structures like bridges, culverts, and underpasses can facilitate connectivity and reduce the barrier effect of roads by providing less risky areas for animals to cross. Estimating areas of high wildlife movement near roads is beneficial for prioritizing transportation investments for wildlife. We used an omnidirectional circuit theory approach to model the movements of eight terrestrial mammal species across the state of Vermont, a forested region central to the globally important Northern Appalachian ecoregion. We combined expert-derived landscape resistance surfaces with wildlife occurrence data to develop species-specific connectivity models at statewide (23,873 km2, 30 m resolution) and roadway structure (100 m radius around 5,912 structures, 0.5 m resolution) scales. The flow of animal movement across the landscape, depicted as electrical current density, was highest for forest-obligate species along the forested, mid-elevation foothills of the Green Mountains in central Vermont and lowest in the agricultural Champlain Valley; however, for more urban- and agriculture-adapted species, flow was highest in developed areas and lower elevation valleys. Average current density was highest for black bear (Ursus americanus), and lowest for striped skunk (Mephitis mephitis) at the statewide scale and highest for raccoon (Procyon lotor) and lowest for moose (Alces alces) at the finer structure scale. Results at both scales revealed different patterns of expected animal movement that reflect the relative extent of connectivity. We then scored connectivity for each structure across all species by combining both scales using four different methods to capture a range of management interests. Rankings varied greatly depending on the method used, highlighting the need to clearly articulate objectives when scoring structures or other features in a landscape. Resistance, occupancy, and current maps also indicated the broad importance of intact forest for connectivity and may be particularly important for identifying priority regions for protection under Vermont’s Community Resilience and Biodiversity Protection Act that mandates protecting 50% of the state by 2050.
Citation: Drasher CE, Slesar C, Hawkins-Hilke J, Gingras G, Marangelo P, Landau V, et al. (2025) Assessing the connectivity value of roadway structures for terrestrial mammals across the Northern Appalachian forest of Vermont. PLoS One 20(9): e0331493. https://doi.org/10.1371/journal.pone.0331493
Editor: Daniel de Paiva Silva, Instituto Federal de Educacao Ciencia e Tecnologia Goiano - Campus Urutai, BRAZIL
Received: September 8, 2024; Accepted: August 15, 2025; Published: September 4, 2025
Copyright: © 2025 Drasher et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: The data reported in this study can be found either in the paper or at the following links. Species occurrence data can be found at: https://www.sciencebase.gov/catalog/item/66cf79afd34e98e8a924c074 . Vermont roadway structure scores can be found at: https://anr.vermont.gov/maps-and-mapping/biofinder.
Funding: This work was supported by a grant from the Vermont Agency of Transportation (EA# 0001057-332). C. Drasher received support from this grant in the form of a Research Assistantship through the University of Vermont (UVM). V. Landau and P. Marangelo received support through a sub-award from this grant. https://vtrans.vermont.gov/ C. Drasher also received support through a teaching assistantship for the Rubenstein School of Environment and Natural Resources at UVM. https://www.uvm.edu/rsenr Funding sources did not play a role in the study design or decision to publish. Funding sources did not determine the ultimate study design, data collection, or analysis. Two co-authors are employees of the Vermont Agency of Transportation (VTrans) and are included as authors due to their contribution to initial project conceptualization, manuscript preparation, and their consultation/knowledge provided during the research process. This work was completed in response to a need for information that VTrans had.
Competing interests: The authors have declared that no competing interests exist.
Introduction
Wildlife populations often depend on landscape connectivity for persistence [1]. A connected landscape provides access to resources, opportunities for dispersal, and allows for genetic exchange between subpopulations [2,3]. Landscape fragmentation can lead to changes in distribution and create isolated populations, which often results in negative impacts to species [1–4]. Globally, terrestrial mammals experiencing higher levels of fragmentation are consequently at a greater risk of extinction [5].
Road networks are wide-reaching and contribute to landscape fragmentation [6,7]. In the United States (US), there are 6.7 million km of public roads that are traveled over 5 trillion km each year [8]. While road networks provide connectivity for human society, they generally decrease connectivity for wildlife through subdivision of habitat [9,10]. Roads cause direct mortality to wildlife through vehicle collisions, with incidents involving large mammals costing US motorists nearly $8.4 billion annually [11]. Roads cause indirect mortality to wildlife by inhibiting gene flow, potentially resulting in less viable populations [12–14] and evolutionary consequences in some cases [15]. Impacts from roadways also vary by species and taxa, and these impacts sometimes extend beyond the actual road footprint [6]. Understanding and mitigating the effects of roads is especially important in the face of climate change, as species are expected to experience range shifts in response to changes in their natural environment [3,16].
Several approaches have been used to mitigate the negative effects of roads on wildlife. Properly designed roadway structures can facilitate wildlife movement across road networks, increasing the permeability of roads [10,17]. Culverts, bridges, underpasses, and overpasses provide opportunities for wildlife to navigate complex transportation networks and reduce the risk of collision [10,18]. Existing structures like culverts are often abundant along roadways; however, some structures are better equipped to facilitate wildlife passage than others [19]. The physical attributes of roadway structures, such as width, length, substrate, and material, may promote or deter wildlife use, depending on species size and movement preferences [17,20–23]. Transportation and wildlife managers can improve existing structures to accommodate a wider range of species [24], yet these projects are often very costly, and it is beneficial to prioritize investments in wildlife-focused infrastructure improvements to areas of the landscape and road network that maximize wildlife connectivity [23,25–27].
To locate areas along roads where investments will be most impactful for improving landscape connectivity, factors considered may include the broader landscape context influencing species movements, as well as the immediate landcover features surrounding existing roadway structure locations [28]. Wildlife movement and landscape connectivity are often scale-dependent [29]. At the landscape scale, wildlife movements become channeled through specific areas based on landcover composition and configuration, and structures that are located within migration paths, corridors, or pinch-points often experience higher use [21,23]. At a finer scale, the presence of and distance to vegetative cover in the immediate vicinity of a structure can influence use [21,24]. Thorough mitigation planning considers multiple spatial scales when prioritizing structure improvements to maximize the probability of use by wildlife species [28]. For instance, a culvert or bridge located outside of known wildlife corridors in the landscape may not experience high levels of use, even if improvements to the immediate site-level landcover and structure characteristics are implemented [21].
Previous studies have used connectivity modeling to identify locations where species are more likely to cross the road to guide the prioritization of mitigation investments [25,27]. For example, Zeller et al. [27] compared multiple connectivity models to identify road crossing hotspots for black bears (Ursus americanus) in Massachusetts, USA including resource selection functions, least-cost paths, resistant kernels, individual based movement models, and a circuit-theory approach using Circuitscape [30,31]. The circuit-theory approach outperformed all other techniques and has been widely applied to connectivity modeling elsewhere [27,32]. This approach models species movements as the flow of electricity through a circuit [30] and incorporates information on the resistance of landscape features, including roads, to the movement of a given species [30,31,33]. Areas of concentrated electricity flow indicate increased predicted wildlife movement, which can inform conservation planning and decision-making [33]. In the context of roads, areas of concentrated flow near existing roadway structures may suggest locations where mitigation activities would be most beneficial.
Landscape resistance surfaces have been widely used in ecological and circuit-based modeling to predict the movement of animals across landscapes, and species-specific resistance surfaces are more precise in explaining wildlife connectivity than generalized approaches that group multiple species into a single model of landscape resistance [34,35]. Species responses to landscape variables differ based on their ecological needs and movement behaviors [35]. For example, human development variables (e.g., roads, buildings) generally present a high level of resistance to the movements of many forest obligate species; however, these same areas may be of low resistance to more generalist species that are adapted to anthropogenic environments. Additionally, the spatial resolution (or pixel size) of resistance surfaces is important for landscape connectivity analyses as some species may respond to fine-scale landscape features, while the movement of other species may be explained by coarser patterns of landscape composition [34,35].
A useful innovation in circuit theory modeling is the development of the Omniscape approach [33,36]. Omniscape allows electricity to move in all directions (or omnidirectionally) throughout the landscape using a moving window technique and incorporates a ‘source-strength’ input that represents the abundance of a species across the analysis area, providing a starting point for electricity that then moves through the resistance surface. As spatial models of abundance are often challenging to develop, maps of distribution based on occupancy or habitat suitability can serve as a proxy for abundance [36]; Pearman-Gillman et al., in review]. Omnidirectional connectivity models reduce biases associated with manually choosing starting and ending points of electricity and are useful when source locations are unknown or when modeling connectivity for multiple species regionally [37]. Omnidirectional landscape connectivity models may also be ideal for species that do not have more predictable migratory patterns between fixed starting and ending locations in a landscape; for example, some western ungulate species exhibit large-scale migrations between seasonal resources [38], whereas mammals in the eastern US maintain home ranges and do not exhibit such large-scale migrations.
We focused on the US state of Vermont to create multi-scale and species-specific omnidirectional connectivity models for terrestrial mammal species, and quantify connectivity around roads and existing roadway structures (culverts, bridges, underpasses) to guide transportation infrastructure mitigation investments. Vermont wildlife and transportation agencies have identified mitigating the impact of roads on wildlife as a goal in their long-term management plans [39–41]. To help prioritize wildlife-focused transportation infrastructure investments in Vermont, transportation managers identified the need to predict areas of higher probable movement of terrestrial species across the road network. By ranking existing roadway structures by their connectivity value for terrestrial mammal species, wildlife improvements can be implemented at high-priority structures as opportunities arise. Omnidirectional models of wildlife movement can be used to quantify and rank the relative value of roadway structures for maintaining landscape connectivity throughout the state to inform transportation management actions that may improve connectivity in high-movement areas.
The goal of our study was to model wildlife movement in Vermont to assess the connectivity value of existing structures along the road network. Our objectives were to 1) create landscape resistance surfaces for eight terrestrial mammal species based on ecologically relevant landcover variables, 2) model the movement of these species in Vermont at two spatial scales, the landscape scale (statewide) and roadway structure scale (individual structure locations), and 3) quantify connectivity at structure locations and combine multi-scale results in several ways to rank individual roadway structures by their collective connectivity value across focal species. The multi-scale and multi-species assessment of structure connectivity values provides important information for decision-making to improve connectivity across road networks.
Methods
Study area
Vermont is centrally located within the Northern Appalachian/Acadian Ecoregion, which covers 330,000 km2 from New York through New Hampshire, Maine, and the Canadian province of Québec [12]. The Appalachians are considered a globally significant region for biodiversity and part of The Nature Conservancy’s Resilient and Connected Network of lands identified as critically important for tackling climate change and conserving biodiversity [42]. The central location of Vermont in the northern Appalachians makes it a key connector that facilitates species movements regionally to other states and provinces. Vermont contains portions of 6 of 10 connectivity corridors for wildlife identified by the Staying Connected Initiative – a consortium of federal, state, non-profit, private, and academic partners [43]. Moreover, the Green Mountain range in Vermont has been identified as a priority area for connectivity within the broader ecoregion and is vulnerable to future human development [12]. Vermont is also located in a unique area of high to very high climate flow and predicted to be an influential connector facilitating the movement of Appalachian species northward as climate changes [42].
Vermont is a rural state covering 23,873 km2 with approximately 645,000 residents (Fig 1) [45]. It is 78% forested, consisting primarily of northern hardwood forest and spruce-fir forest types [46]. Elevation ranges from 29 m along the shores of Lake Champlain to 1,339 m in alpine zones of the Green Mountains [46]. Development is more concentrated in the lowland riparian basins including the Champlain and Connecticut River valleys and along riparian corridors [46]. Agriculture is prominent in the Champlain Valley, and residential development has increased significantly in recent decades, expanding faster than the state’s population growth [46]. Vermont is bordered by the US states of New York, Massachusetts, and New Hampshire, and shares a northern border with the Canadian province of Québec (Fig 1).
We focused our analysis on 5,912 state-managed transportation structure locations on the Interstate System and State Road System (c).
The Vermont road network includes 11,508 km of paved roads and 13,920 km of unpaved roads, totaling 25,428 km of roadway [47] (Fig 1A). This includes larger interstate highways (I-89, I-91, I-93), multiple state highways, and other class I-IV roads managed by the state and local municipalities (Fig 1B). The public drives an estimated 11.8 billion km annually on this road network [48]. The Vermont Agency of Transportation maintains over 50,000 roadway structures (bridges, arch/box/pipe culverts, other underpasses) of varying size and condition, on state roads and some town roads. Many of these structures facilitate the movement of water beneath the roadway, either for intermittent drainage or to pass traffic over rivers and streams. Road classes were grouped into six categories based on size and traffic volume, with input from the Vermont Agency of Transportation, for use in resistance-based connectivity modeling. We selected 5,912 state-managed roadway structures that were >1 m in diameter for analysis (Fig 1C) [49–53]. Structures above this threshold were identified as priority structures by the transportation agency since many were already large enough to facilitate movement for many terrestrial mammal species or may be more easily retrofitted to better accommodate species movements.
Focal species
We focused on eight terrestrial mammal species: black bear (Ursus americanus), bobcat (Lynx rufus), coyote (Canis latrans), moose (Alces alces), raccoon (Procyon lotor), red fox (Vulpes vulpes), striped skunk (Mephitis mephitis), and white-tailed deer (Odocoileus virginianus). These species hold ecological, economic, and cultural significance in Vermont and the broader region, and are wide-ranging and frequently encounter roadways [22,54]. These species are also target species for wildlife management by the transportation and wildlife agencies in the state and selected with input from agency partners [39,41]. The selected species are frequently involved in road collisions, Species of Greatest Conservation Need, high value species for hunting and other forms of recreation, or priority species for the US Department of Agriculture Rabies Management Program [55]. Predicting the movements of these species across the state may have broader implications for species and land management, wildlife disease monitoring, and road collision mitigation.
Objective 1: Create landscape resistance surfaces
We generated species-specific resistance surfaces at two spatial resolutions using an expert elicitation process (methods reviewed and IRB exemption was determined by the University of Vermont Internal Review Board). We conducted an online survey of wildlife experts that asked them to score landscape resistance for land cover variables for each species following Dickson et al. [56]. We invited 10 regional wildlife professionals with expertise in one or more of the focal species to complete species-specific surveys (1 to 3 experts per species). Experts selected were those actively working with one or more of the focal species professionally as state or federal wildlife biologists and managers in Vermont and the broader New England region. All experts were recruited to participate in the survey between April 1 and August 31, 2021, and viewed a pre-survey video that explained the project background, described resistance surfaces and their use in Omniscape connectivity modeling, and walked through the resistance surface survey process. Experts were also provided with a written project information sheet with the survey and informed that any participation in the written survey implied consent. We developed the survey using the LimeSurvey platform (LimeSurvey GmbH 2005; S1 Fig).
Experts scored resistance values of landcover variables in two datasets for use in each spatial scale of the wildlife movement analysis: 1) Landscape scale: 2016 National Landcover Dataset (NLCD) variables (30 m resolution) [44], and 2) Structure scale: 2016 Vermont Center for Geographic Information (VCGI) lidar dataset variables (0.5 m resolution) [57]. In addition to the fifteen variables scored in the NLCD dataset for the landscape scale analyses, we included the six categories of road classes for scoring, and these road class variables were overlaid onto the NLCD raster after removing road pixels that were classified generally as development in NLCD. For the structure scale VCGI dataset, seven of the variables from the base landcover raster dataset were included for scoring, and the six road class categories were also overlaid onto this raster in place of the generic ‘road’ variable. Additionally, two supplemental VCGI vector datasets for agriculture and wetlands were converted to rasters and added to the VCGI raster dataset. In total, experts scored 21 variables in the landscape scale NLCD dataset and 15 variables in the structure scale VCGI dataset (Table 1).
Resistance scoring was based on a 1–100 scale, with 1 representing the lowest resistance to movement, 100 representing the highest resistance to movement, and NA (designated as 101 by the experts) representing a complete (impermeable) barrier. When multiple experts provided input for a species, resistance values for each variable were averaged together and used to create hypothesized resistance maps at each scale and then used in preliminary Omniscape analyses (see Objective 2 below). We produced draft Omniscape maps of statewide movement for each species at the landscape scale, and draft Omniscape maps of fine-scale movements for each species around 5 test structures at the structure scale.
Following these preliminary analyses, we conducted individual follow up meetings with each expert. Experts were presented with the draft Omniscape maps for their species at each scale and given the opportunity to ask interpretation questions and provide feedback on the preliminary results. Following the interview, experts were given one opportunity to adjust their initial resistance values to improve the accuracy of the maps if needed. Once again, resistance values (including any adjusted resistance values) were averaged together for each species and used to create the final resistance inputs used in the Omniscape analyses at each scale (Table 2).
Objective 2: Model and map landscape connectivity
We used Omniscape [33,36] to model and map connectivity for each focal species across Vermont. Omniscape is a package run through the Julia programming language [36,58]. We modeled species-specific landscape connectivity with Omniscape at two spatial scales. At the landscape scale, we modeled connectivity across the entire state of Vermont at a 30 m resolution. At the structure scale, we modeled connectivity within a 100-m radius around each individual structure at a 0.5 m resolution; this radius and resolution is relevant for transportation management actions within the right-of-way of structures.
Moving window algorithm.
Omniscape uses a moving window algorithm to determine the extent to which electrical current can travel from any source location to ground locations in the landscape, and the size of the window may be specified to constrain the flow of electricity to a biologically reasonable distance [36]. Omniscape centers the moving window on a single center pixel in the landscape, clips the source and resistance layers to the designated moving window size, and all source pixels within the window will emit electricity that then travels through the resistance surface to reach the center (ground) pixel. This process is repeated across the landscape iteratively so that each pixel may act as a ground, and all moving window iterations are summed together to create a single map of cumulative current flow across the study area. The size of this moving window ultimately determines how far electricity may travel from each source pixel identified within the moving window back to the center ground pixel [33,36]. To represent general species movement capabilities, we set the moving window size to the average home range size of males and females for each species (Table 3).
Source-strength and resistance inputs
The source-strength input for the analysis determines where electricity is emitted from in the landscape, and how much electricity is emitted from each pixel in a raster map [36]. The source-strength value within pixels ranges from 0 to 1, where pixels with a value closer to 0 emit less electricity, and those with values closer to 1 emit more electricity, corresponding with the predicted amount of individuals coming out of each location in the landscape [36]. We used regional species-specific occupancy models as the source-strength input for connectivity analyses at both the landscape and structure scales (S2 Table) [59]. These occupancy models were generalized linear mixed effects models developed with a model selection approach that analyzed the effects of 74 variables across the New England landscape, and each model performed well against independent empirical wildlife data [59]. These 30 m resolution occupancy models were used at the same 30 m resolution for use in the landscape scale connectivity models, and resampled to 0.5 m resolution for use in the structure scale models.
Once electrical current is emitted from a source pixel it travels toward a ground pixel, and the resistance input determines the path and strength of electricity flow based on the relative cost of movement through each pixel [33,60]. The final species-specific resistance surfaces created in Objective 1 were used as resistance inputs for species-specific Omniscape analyses at each scale (Table 2).
Analysis area
Both the landscape scale and structure scale analyses focused within the bounds of Vermont. However, to avoid edge effects at the landscape scale, the resistance inputs were buffered by 10 km (a distance greater than the moving window radii for each species) into the neighboring New York, Massachusetts, and New Hampshire. Resistance inputs were also buffered into Québec; however, due to differences in landcover data, the North American Land Change Monitoring System (NALCMS) landcover dataset [61] was used over the northern border (similar categories of NALCMS landcover were grouped to coincide with NLCD landcover variables as needed; Table 1). Source-strength inputs were also buffered into the neighboring states and province; however, occupancy data were unavailable in New York and Québec, and therefore approximate source-strength values were inferred in these areas using a Focal Statistics averaging process in ArcGIS (v. 10.8.1, ESRI, Redlands, California, USA) for each species-specific occupancy input (S3 Fig). Buffering the resistance and source-strength inputs allowed Omniscape to start the analysis outside of Vermont, while results were clipped to the boundary of Vermont. Buffering allowed electricity to flow into and out of Vermont, preventing pooling of electricity at the borders which can result in artificially high current densities in border areas [33,62].
At the structure scale, we used a Vermont-specific high-resolution (0.5 meter) landcover dataset for the resistance input, and therefore could not buffer into neighboring states or province to avoid edge effects. However, due to the small Omniscape analysis area around individual structures (100 m) at this scale, less than 1% of structures had radii that extended over the Vermont border (n = 47). Structures located within this edge effect zone may experience some edge effects.
Computation
Landscape scale analyses were run with Omniscape.jl version 0.5.0 (Julia version 1.5.3) [36,58]. Structure scale analyses were run using the VermontTerrestrialPassageTool.jl, a Julia package developed for this analysis to cycle through all structure locations and run targeted Omniscape.jl analyses within a 100-m radius of each structure [63]. All analyses were run on the Bluemoon cluster, a 161-node computing cluster with 8392 compute cores with up to 4 TB RAM, of the University of Vermont Advanced Computing Center (Burlington, Vermont, USA). Landscape scale analyses ran on one node and averaged 3.7 hours to complete per species (maximum compute time of 13.8 hours), and structure scale analyses ran on one node and averaged 84.3 hours to complete per species (maximum compute time 85.7 hours).
Objective 3: Rank roadway structures
We quantified connectivity around each existing roadway structure location from each species-specific connectivity map for each spatial scale. We then combined species-specific results to rank each structure by overall connectivity value for all species. First, we calculated general current density (CD) metrics for each individual structure location: 1) CDmeanLS representing the sum of species-specific mean current density values within 1 km of each structure from each landscape scale (LS) map; and 2) CDmeanSS representing the sum of species-specific mean current density values within 50 m of each structure from each structure scale (SS) analysis. This resulted in 5,912 CDmeanLS values (sum of species-specific means at each location from LS results) and 5,912 CDmeanSS values (sum of species-specific means at each location from SS results).
We then used four different methods of combining connectivity metrics from each spatial scale into a single ranking for each structure, to rank structures from highest overall connectivity value to lowest for all species combined. We used Spearman’s rank correlation to explore relationships between the rankings generated by each method [64].
- Method 1: Landscape Scale Mean Current Density. This method assessed the overall connectivity value of roadway structures based on the average electrical current density surrounding them from the landscape scale analysis. For each structure, the CDmeanLS was calculated, and all 5,912 structures were ranked from highest CDmeanLS to lowest. Structures with a higher CDmeanLS rank have more electrical current surrounding them within a 1 km radius, representing higher priority connectivity locations. This method aimed to quantify average predicted wildlife connectivity around roadway structures based on the broader movement patterns of the species statewide.
- Method 2: Structure Scale Mean Current Density. This method assessed the value of roadway structures based on the average electrical current density surrounding them from the structure scale analysis. For each structure, the CDmeanSS was calculated, and all 5,912 structures were ranked from highest CDmeanSS to lowest. Structures with a higher CDmeanSS rank had more electrical current surrounding them within a 50-m radius and represent higher priority wildlife movement locations based on finer-scale species movements modeled at a 0.5 m resolution only in the immediate vicinity of structures. This method aimed to quantify average predicted wildlife movement around roadway structure locations based on the immediate composition and configuration of landcover at a finer resolution around structures.
- Method 3: Sum of Landscape Scale and Structure Scale Mean Current Densities. This method assessed the overall connectivity value of roadway structures by summing the average electrical current density surrounding them in each scale of analysis. For each structure, the landscape scale CDmeanLS and the structure scale CDmeanSS were summed to get a multi-scale measure of mean current density (CDmeanLS+meanSS). Structures were ranked from highest to lowest CDmeanLS+meanSS. This rank considered both the broad statewide species movements as well as fine-scale movements around structures.
- • Method 4: Landscape Scale Rarity-Weighted Richness Index. This method accounted for any rarity of species-specific current flow around structures, and ranked structures that may have rare/less flow for one or more species higher compared to other structures. We calculated a modified Rarity-Weighted Richness Index (RWRI) [65] for each structure from species-specific mean current density values within 1 km of each structure from the landscape scale maps. This method evaluated structures based on whether they contained rare species flow around them compared to all other ranked structures. For each site, RWRI was calculated as:
Where hi is the number of sites containing electrical flow for species i, and n representing the total number of species with flow at a given site.
Evaluation of connectivity models
We used an independent set of wildlife data collected by state transportation and wildlife agencies to evaluate roadway structure connectivity scores at each spatial scale [66]. Camera trap data were collected from 2015 to 2021 at 49 roadway structure locations across the state that were predicted to be ideal for wildlife movement based on the expert opinion of wildlife and transportation managers [22]. We calculated the probability of species passing through a structure across camera location in two categories: 1) ungulates (moose, deer), and 2) carnivores (bear, bobcat, coyote, red fox, skunk; raccoon detections were not recorded in this dataset). We then compared these estimates of passage to the landscape scale and structure scale ‘connectivity scores’ at each structure. We assumed that likelihood of passage through structures would correlate to connectivity scores. As sites were considered ‘ideal’ for crossing, we expected passage rates and connectivity scores to be high and roughly equivalent if connectivity results were accurate. Probability and connectivity values were standardized on a scale from 1–100 to make comparisons.
Results
Objective 1: Landscape resistance surfaces
All 10 regional wildlife experts responded to the survey, with some experts (n = 5) completing the survey for multiple species given their expertise. For each species, we used input from 1–3 experts to generate resistance values for landscape variables at both spatial scales. Statewide mean resistance was highest for bobcat (19.33 ± 30.53 SE), and lowest for skunk (8.40 ± 16.30 SE). For other medium-sized mammals, statewide mean resistance was 8.78 ± 16.23 for coyote, 9.10 ± 14.82 for raccoon, and 9.62 ± 13.24 for red fox. For larger mammals, statewide mean resistance was 13.68 ± 23.11 for bear, 9.41 ± 15.87 for moose, and 9.03 ± 21.62 for white-tailed deer. Based on the composition of landcover variables in the Vermont landscape, most species had a mean landscape resistance of 9 on a 1 to 101 scale across the study area. Final landscape resistance values (Table 2), determined by an average across expert values, were used as an input in the Omniscape analyses.
Objective 2: Wildlife connectivity models
We developed 8 species-specific maps showing statewide landscape connectivity at the landscape scale and summed these maps together to visualize statewide connectivity patterns of all species combined (Fig 2). From this combined map, statewide mean electrical current density was 388.35 ± 147.82 SE amps for all species combined. Mean electrical current density was highest for bear (94.91 ± 53.76 SE) and lowest for skunk (8.39 ± 4.67 SE) (Table 4). Flow was generally higher along the spine of the Green Mountains and in areas of lower human population density for more forest-obligate species, such as bear and moose. Bobcat, coyote, and white-tailed deer had intermediate levels of flow statewide, with pinch points of concentrated flow in agricultural areas where electrical current became more constrained by roads and cropland. The more urban- and agricultural-adapted species, raccoon, red fox, and skunk, had highest flow in lower elevation areas (such as the Champlain and Connecticut River valleys) and along corridors of human development (including the US Route-7 corridor and the greater Burlington and Montpelier areas).
Areas of high electrical current density represent areas of more concentrated expected species movement.
At the structure scale, we produced 5,912 maps of wildlife connectivity around individual structure locations for each species, totaling 47,296 structure scale maps across all eight species (example maps shown in S4 Fig). Structure-wide mean electrical current density was 361.55 ± 74.12 SE amps for all species combined and was highest for raccoon (123.13 ± 20.43 SE) and lowest for moose (55.93 ± 36.54 SE; Table 4).
The resulting electrical current maps predicting movement flow were a function of the two input layers: species occupancy probability and landscape resistance. Across all species and at each spatial scale of analysis, landscape resistance was low and species occupancy probability was high in natural landcover classes (forest, wetland, and shrub). Landscape resistance was generally high and species occupancy generally low in landcover influenced by anthropogenic use (road, agriculture, and development). Raccoon and skunk represented the exception, as these species had higher probability of occurrence in developed areas and lower levels of resistance to anthropogenic features, therefore showing higher levels of movement flow through areas of development. The composition of landscape variables greatly influenced electrical flow in the resulting Omniscape connectivity maps. Larger areas of natural forest cover (i.e., Green Mountain National Forest) typically had intermediate levels of electrical flow in the output maps: although there was higher electrical input into these areas from higher species occupancy probability, the electricity was able to ‘spread out’ in larger blocks of low resistance habitat, resulting in intermediate levels of flow. Similarly, smaller strips of natural landcover found in the higher-resistance agricultural and human development areas had high electrical flow, as electricity became constrained into smaller areas of ideal movement habitat.
Objective 3: Roadway structure ranking
Spearman’s rank correlations showed a moderate to strong positive correlation between the rankings generated by the different methods (Table 5). Method 1 (landscape scale ranking) and 4 (landscape scale RWRI) had the strongest correlation (ρ = 0.94), as both methods ranked structures based on landscape scale current only. Methods 1 and 3 also showed a moderately positive correlation (ρ = 0.75), as did Methods 3 and 4 (ρ = 0.75). Methods 1 and 2 had a weaker positive correlation (ρ = 0.25), which was not unexpected since some structures may score high based on broader landscape-level movements of a species but lack adequate landcover for movement at the finer resolution within 50 m of the structure (Fig 3).
Structures were ranked from highest mean current density to lowest in each method. Method 1 considered landscape scale flow, Method 2 considered structure scale flow, Method 3 considered landscape and structure scale flow, and Method 4 used a Rarity-Weighted Richness Index (RWRI) to rank structures.
The Top 100 structures ranked by each method showed variable locations in the landscape as high-priority for wildlife-focused transportation investments (Fig 3). Overlap in sites between ranking methods was generally low and varied from 3 to 69 (mean = 18 ± 8.8 SE; Table 5). Method 1 highlighted structures in more natural areas of landcover, along the Green Mountain range and into the northeastern part of the state, as important for prioritization due to their location in areas of higher statewide movement flow for all species. Method 2 showed the greatest difference in structure locations from all ranking methods, as it only considered site-level characteristics within a small radius of structures and ignored statewide patterns of species movement flow. The Top 100 structures ranked with Method 2 were more well-distributed geographically and also found in areas lacking greater landscape scale current flow. Since Method 3 considered both LS and SS movement flow, the Top 100 structures for this method were still found in areas of higher statewide flow, but also in locations with good site-level characteristics; these structures were more geographically spread out than Method 1. Method 4, in contrast, selected the most geographically constrained Top 100 structures, concentrated in select areas containing high-density current for one or more species with rare current flow; these structures were found mostly in the northern part of the state.
Evaluation of connectivity models
We detected the focal species 2,180 times across the 49 camera sites, which were selected in areas predicted to be ideal for wildlife movement and connectivity [65]. Species presence was observed at 15 sites for coyote, 33 for deer, 3 for moose, 13 for bear, 17 for bobcat, 18 for red fox, and 14 for skunk. Naïve probability of presence was 71.4% for carnivores, 67.3% for ungulates, and 81.6% for all species combined across all sites. Estimated connectivity scores for all species combined at the landscape scale and structure scale, standardized on a 0–100 scale, were 75.3 ± 7.9 SD and 72.1 ± 10.5 SD, respectively.
Discussion
Our analysis demonstrates that multi-scale and multi-species omnidirectional connectivity models can be used to determine probable road crossing locations used by terrestrial mammal species, particularly in relation to existing roadway structures, to help guide transportation infrastructure mitigation efforts. Our connectivity modeling included a variety of species and results showed different predicted movement behaviors for different species, but performed well against an independent set of camera data at select roadway structure locations. Differences in model outputs reflect the variation in movement behavior, sensitivity to habitats, and species distributions. The flow of animal movement across the landscape was highest along the forested, mid-elevation foothills of the Green Mountains in central Vermont, especially for forest-obligate species. In contrast, the raccoon and skunk models showed more concentrated movements in developed areas, such as in urban centers of Burlington and Montpelier Vermont, whereas species such as bear and moose showed much lower levels of predicted movement in highly developed areas.
We developed connectivity models for each species as movement behaviors are species-specific, particularly in relation to roadways [6,35,67]. Species-specific maps allow managers and decision-makers to evaluate structures on a species-by-species basis, as needed, depending on management objectives. For example, in Vermont, decisions about retrofitting existing roadway structures with walking shelves or a solid dirt movement surface may be based on the value of a given structure for certain target species like bobcat or moose. We also combined each species map into an all-species map to provide a singular, more general depiction of connectivity for the entire group of focal species and for a broader structure ranking. This approach could be adjusted to meet different management objectives by applying weighted values to higher-priority species maps before combining results. For example, if transportation managers are primarily concerned with identifying high-priority deer crossing locations over other species, the deer-specific model could be multiplied by a weight before summing all species maps together. Focal species selected for this analysis were based on specific transportation and wildlife agency objectives to mitigate the impacts of roadways on larger terrestrial species that present higher risk to motorists (including bear, moose, and deer), as well as species that hold ecological, economic, and cultural value in the state [39]. While prioritizing infrastructure needs for these species may benefit multiple other taxa with similar movement and habitat needs, it would be beneficial for future work to include additional taxa with different or highly specific movement needs, such as amphibians and reptiles that require both upland and wetland habitat on either side of a roadway crossing, and aquatic organism passage through stream crossings. Previous work in the state resulted in the development of the Vermont Culvert Aquatic Organism Passage Screening Tool, to help identify priority stream crossing projects that would increase aquatic organism passage in the state [68].
Circuit-based connectivity models are widely used to predict areas of movement and functional connectivity for species [27,69]. Our approach integrated landscape resistance and species-specific patterns of occupancy. The inclusion of occupancy served to account for the distribution of species when quantifying connectivity, which may improve structure level assessments of connectivity over using a single input to serve as both the source of animals and landscape resistance. For example, a structure surrounded by high resistance movement habitat may still be more valuable for connectivity if species occupancy (and abundance) is very high near that site. However, a structure that is surrounded by low resistance movement habitat may be less valuable and experience less flow if the species does not occur near the site. Integrating both resistance and occurrence information may produce a more robust model that considers probable locations of a species as well as predicted responses to landscape variables during movements.
Connectivity analyses that incorporate multiple scales or resolutions generally provides more accurate depictions of species movement across a landscape [70]. However, connectivity studies often fail to consider the influence of multiple spatial scales on species responses to landcover [71]. Decreasing the grain size to a finer spatial resolution has also been shown to improve connectivity model performance [72]. In our approach, we integrated connectivity at two spatial scales to quantify and rank connectivity at existing roadway structures. Although it can be challenging to determine the best scales to use for connectivity analysis, we believe that these two scales adequately represented the group of species, fine-scale conditions at structures, and helped reduce biases associated with relying on a single spatial scale. For example, a high connectivity value at the structure scale for a given species could lead to the conclusion that the site is of high value to animal movement without considering the broader landscape scale. If the broader landscape scale reflected generally poor conditions for movement, then the high site level score is misleading. Incorporating the landscape scale information in this case essentially tempers the site level score and presumably results in a more accurate assessment of connectivity at the structure.
Similarly, connectivity modeling at the landscape scale was performed at the home range scale by using the average home range size for a given species as the moving window size. This allowed us to appropriately scale the extent of movement at any given point in the landscape to the typical ranging behavior of the species. However, it is important to recognize that road crossings driven by dispersal movements may occur outside of normal home range movement patterns. Depicting connectivity for dispersing animals could be accomplished by simply increasing the moving window size to an average dispersal distance for a given species [73].
Species-specific resistance surfaces could be improved by parameterizing values with empirical wildlife movement data, such as data from GPS-collared animals or genetic samples, though these data are not always available [74]. Furthermore, incorporating dynamism into circuit-based connectivity models is an emerging field of research that would contribute additional detail to wildlife movement predictions, since movement across roadways can vary depending on season, diel period, or other factors for some species [75,76]. Dynamic connectivity models may provide more detailed predictions of movement, however they can be challenging to build, as this requires gathering extensive data on species movement behaviors and landscape resistance associated with (sometimes rapidly) varying landscape conditions [75].
Quantifying connectivity at features or points of interest in the landscape, poses a challenge for any connectivity modeling analysis. We proposed four methods of ranking connectivity to provide a range of options that reflect different potential management objectives. Each method emphasized a particular objective, including to restore or promote landscape-level connectivity for a species (method 1), restore or promote fine-scale connectivity based on site-level characteristics (method 2), consider both landscape-level and fine-scale species movements for promoting connectivity (method 3), and prioritizing connectivity actions in areas where rarer species movements are represented (method 4). Each of the four ranking methods resulted in different assemblages and rankings of roadway structures, highlighting the need to clearly articulate the specific method used, the benefits, limitations, and assumptions of that method, and the broader objective of scoring structures or other features in a landscape. This is especially important as rankings often form the basis of decision-making around prioritizing investments in wildlife management. In the context of transportation management, rankings often identify structures or locations to prioritize for modifications to facilitate movement across road networks [77,78].
Circuit-based connectivity analysis clearly shows patterns of concentrated electrical (animal) flow and is useful for identifying important pinch-points and high-use movement paths in a landscape [79,80]. However, interpretation of circuit-based outputs can at times be misleading, especially for ‘low flow’ areas. For example, areas with low flow may represent truly poor areas of movement (e.g., an urban area for a forest dependent species), but may also represent areas that are highly conducive to flow, just not concentrated flow (e.g., a large forest block for the same species). This discrepancy creates challenges to interpretation. Classifying electrical current outputs into categories of impeded, diffuse, and intensified/channelized flow has been proposed to more clearly interpret patterns of flow and distinguish between poor low-flow areas (impeded) and good low-flow areas (diffuse) [33,81]. Our study focused on pinch points of concentrated electrical flow immediately near roadways, and within small analysis areas at specific roadway structures, therefore accounting for these low flow areas in larger habitat blocks was not as relevant to our objectives.
Circuit-based connectivity models are also difficult to validate; however empirical detection data from roadkill, telemetry, tracking, and wildlife camera traps have been used to evaluate performance in other studies [36,82,83,84]. Our evaluation of connectivity model results against available independent camera data suggested that predictions of increased connectivity coincided with higher naïve occupancy probabilities of ungulates and carnivores at these locations; however, cameras were only deployed at high-value sites with a limited sample size, and detection was not accounted for in occupancy estimates. These results support the connectivity results and provide some insight into the accuracy of the connectivity model predictions. Including data from sites characterized as low-value for connectivity may improve assessments of model performance, but these data were unavailable for this study. Additionally, the species-specific occurrence models used as model inputs performed well when tested against an independent empirical dataset of research-grade wildlife sightings reported through the iNaturalist community-science database [58]. Although the landscape resistance layer input was not tested against independent data, the validation of the occurrence models and congruence of camera data with areas of high predicted wildlife movement in resulting Omniscape models help support the model predictions and ultimately structure rankings.
Projections suggest that future development will threaten connectivity within Vermont and the broader region as trends show an increase in the construction of residential public roadways of 1,000 km per decade [85]. Transportation networks are wide-reaching, and prioritizing connectivity investments can be a difficult task for state transportation and wildlife management agencies that often work with limited funding to implement improvements [25]. The results and model inputs, such as species occurrence and landscape resistance values, can be applied to future scenarios of landscape change and roadway development to predict how species movements and connectivity may change under future landscape conditions, as was done in Pearman-Gillman et al. [86]. Quantifying landscape connectivity provides important information to develop effective management strategies that mitigate the impacts of transportation infrastructure on wildlife populations. Beyond transportation management, the results may be used to identify locations in Vermont for different conservation objectives, such as land protection or restoration of connectivity through fragmented areas. For example, the Vermont Legislature passed the Community Resilience and Biodiversity Act in 2023 that mandates protecting 30% of Vermont’s total land area by 2030 and 50% by 2050 and emphasizes the importance of connected lands for ecological functioning. The results provide maps that can be used to identify high-value sites for connectivity in support of state efforts to meet these protection goals. Similarly, results could be used to better quantify the connectivity value of wildlife corridors across the broader Northern Appalachians, such as those defined by the Staying Connected Initiative and other regional connectivity projects.
Supporting information
S1 Fig. Expert opinion survey used to gather information on species-specific landscape resistance values.
Survey created using the LimeSurvey GmbH (2005) platform.
https://doi.org/10.1371/journal.pone.0331493.s001
(PDF)
S2 Table. Species occurrence models from Pearman-Gillman et al. (2020) used as source-strength inputs for the connectivity analysis.
Top model parameter estimates shown with standard error and upper (UCI) and lower (LCI) confidence intervals. Models were developed as generalized linear mixed effect models that included random and fixed effects and explored the influence of 74 variables.
https://doi.org/10.1371/journal.pone.0331493.s002
(PDF)
S3 Fig. Averaging process used to buffer species-specific occurrence inputs into buffered areas lacking species occurrence data: New York, USA and Québec, Canada.
https://doi.org/10.1371/journal.pone.0331493.s003
(TIF)
S4 Fig. Example of Omniscape analysis at the structure scale.
The eight species-specific models run within a 100 m radius of transportation structure locations predicted different movement patterns and current densities around the structure. Resulting maps were clipped and mean current density was calculated within a 50 m radius of structures to minimize potential edge effects in the modeling.
https://doi.org/10.1371/journal.pone.0331493.s004
(TIF)
Acknowledgments
The project was a collaboration between the Vermont Agency of Transportation, Vermont Fish and Wildlife Department, The Nature Conservancy of Vermont, and University of Vermont. We thank the wildlife professionals who participated in the expert survey and Vermont Advanced Computing Core for computational support. We also thank M. Clark and A. Jones for guidance on connectivity modeling, A. Ingerson for geospatial support, and J. Brady for input on project design.
References
- 1. Curry CJ, Davis BW, Bertola LD, White PA, Murphy WJ, Derr JN. Spatiotemporal Genetic Diversity of Lions Reveals the Influence of Habitat Fragmentation across Africa. Molecular Biology and Evolution. 2020;38(1):48–57.
- 2. Corlatti L, Hackländer K, Frey-Roos F. Ability of wildlife overpasses to provide connectivity and prevent genetic isolation. Conserv Biol. 2009;23(3):548–56. pmid:19210301
- 3. Rudnick D, Ryan S, Beier P, Cushman S, Dieffenbach F, Epps C. The role of landscape connectivity in planning and implementing conservation and restoration priorities. Issues in Ecology. 2012;16:1–20.
- 4. McCown JW, Kubilis P, Eason TH, Scheick BK. Effect of traffic volume on American black bears in central Florida, USA. Ursus. 2009;20(1):39–46.
- 5. Crooks KR, Burdett CL, Theobald DM, King SRB, Di Marco M, Rondinini C, et al. Quantification of habitat fragmentation reveals extinction risk in terrestrial mammals. Proc Natl Acad Sci U S A. 2017;114(29):7635–40. pmid:28673992
- 6. de Rivera CE, Bliss-Ketchum LL, Lafrenz MD, Hanson AV, McKinney-Wise LE, Rodriguez AH, et al. Visualizing Connectivity for Wildlife in a World Without Roads. Front Environ Sci. 2022;10.
- 7. Ament R, Clevenger A, van der Ree R. Addressing ecological connectivity in the development of roads, railways and canals. IUCN WCPA Technical Report Series. Center for Large Landscape Conservation. 2023.
- 8.
Bureau of Transportation Statistics. National transportation statistics 2021. Washington, D.C., USA: US Department of Transportation. 2021.
- 9. Dodd NL, Gagnon JW. Influence of underpasses and traffic on white‐tailed deer highway permeability. Wildlife Society Bulletin. 2011;35(3):270–81.
- 10. Simpson NO, Stewart KM, Schroeder C, Cox M, Huebner K, Wasley T. Overpasses and underpasses: Effectiveness of crossing structures for migratory ungulates. Jour Wild Mgmt. 2016;80(8):1370–8.
- 11.
Huijser MP, McGowen P, Fuller J, Hardy A, Kociolek A, Clevenger AP. Wildlife-vehicle collision reduction study: Report to Congress. Washington, D.C., United States: US Department of Transportation, Federal Highway Administration. 2008.
- 12.
Trombulak S, Anderson M, Baldwin R, Beazley K, Ray J, Reining C, et al. The Northern Appalachian/Acadian Ecoregion: Priority locations for conservation action. Two Countries One Forest Special Report; 2008.
- 13. Westekemper K, Tiesmeyer A, Steyer K, Nowak C, Signer J, Balkenhol N. Do all roads lead to resistance? State road density is the main impediment to gene flow in a flagship species inhabiting a severely fragmented anthropogenic landscape. Ecol Evol. 2021;11(13):8528–41. pmid:34257914
- 14. Hennessy C, Tsai C, Anderson SJ, Zollner PA, Rhodes OE Jr. What’s stopping you? Variability of interstate highways as barriers for four species of terrestrial rodents. Ecosphere. 2018;9(7).
- 15. Brady SP, Richardson JL. Road ecology: shifting gears toward evolutionary perspectives. Frontiers in Ecol & Environ. 2017;15(2):91–8.
- 16. Sawaya MA, Clevenger AP, Kalinowski ST. Demographic connectivity for ursid populations at wildlife crossing structures in Banff National Park. Conserv Biol. 2013;27(4):721–30. pmid:23758532
- 17. Wang Y, Guan L, Chen J, Kong Y. Influences on mammals frequency of use of small bridges and culverts along the Qinghai–Tibet railway, China. Ecological Research. 2018;33(5):879–87.
- 18. Sheikh ZN, Langbein JE, Ryer K, Rahman MS, Gabler CA, Young JH Jr, et al. Use and effectiveness of wildlife exits designed for ocelots and other mesocarnivores on a south Texas highway. Front Ecol Evol. 2023;11.
- 19. Lu S, Yue Y, Wang Y, Zhang D, Yang B, Yu Z, et al. The Factors Influencing Wildlife to Use Existing Bridges and Culverts in Giant Panda National Park. Diversity. 2023;15(4):487.
- 20. Clevenger AP, Waltho N. Factors Influencing the Effectiveness of Wildlife Underpasses in Banff National Park, Alberta, Canada. Conservation Biology. 2000;14(1):47–56.
- 21. Clevenger AP, Waltho N. Performance indices to identify attributes of highway crossing structures facilitating movement of large mammals. Biological Conservation. 2005;121(3):453–64.
- 22.
Marangelo P. Reducing wildlife mortality on roads in Vermont: Determining relationships between structure attributes and wildlife movement frequency through bridges and culverts to improve related conservation investments. Montpelier, Vermont, United States: Vermont Agency of Transportation. 2019.
- 23. Sawyer H, Rodgers PA, Hart T. Pronghorn and mule deer use of underpasses and overpasses along U.S. Highway 191. Wildlife Society Bulletin. 2016;40(2):211–6.
- 24. Brunen B, Daguet C, Jaeger JAG. What attributes are relevant for drainage culverts to serve as efficient road crossing structures for mammals?. J Environ Manage. 2020;268:110423. pmid:32510423
- 25. Gurrutxaga M, Saura S. Prioritizing highway defragmentation locations for restoring landscape connectivity. Envir Conserv. 2013;41(2):157–64.
- 26. Polak T, Rhodes JR, Jones D, Possingham HP. Optimal planning for mitigating the impacts of roads on wildlife. Journal of Applied Ecology. 2014;51(3):726–34.
- 27. Zeller KA, Wattles DW, Destefano S. Evaluating methods for identifying large mammal road crossing locations: black bears as a case study. Landscape Ecol. 2020;35(8):1799–808.
- 28.
Clevenger AP, Huijser MP. Wildlife crossing structure handbook: Design and evaluation in North America. Washington, D.C., United States: Federal Highway Administration. 2011.
- 29. Rezaei S, Mohammadi A, Malakoutikhah S, Khosravi R. Combining multiscale niche modeling, landscape connectivity, and gap analysis to prioritize habitats for conservation of striped hyaena (Hyaena hyaena). PLoS One. 2022;17(2):e0260807. pmid:35143518
- 30. McRae BH, Dickson BG, Keitt TH, Shah VB. Using circuit theory to model connectivity in ecology, evolution, and conservation. Ecology. 2008;89(10):2712–24. pmid:18959309
- 31. McRae BH, Hall SA, Beier P, Theobald DM. Where to restore ecological connectivity? Detecting barriers and quantifying restoration benefits. PLoS One. 2012;7(12):e52604. pmid:23300719
- 32.
Cushman SA, McRae B, Adriaensen F, Beier P, Shirley M, Zeller K. Biological corridors and connectivity. In: Macdonald DW, Willis KJ, editors. Key Topics in Conservation Biology 2. Hoboken, NJ: Wiley-Blackwell. 2013. p. 384–404.
- 33.
McRae B, Popper K, Jones A, Schindel M, Buttrick S, Hall K, et al. Conserving nature’s stage: Mapping omnidirectional connectivity for resilient terrestrial landscapes in the Pacific Northwest. Portland, Oregon, United States: The Nature Conservancy. 2016.
- 34. Shirk AJ, Schroeder MA, Robb LA, Cushman SA. Empirical validation of landscape resistance models: insights from the Greater Sage-Grouse (Centrocercus urophasianus). Landscape Ecol. 2015;30(10):1837–50.
- 35.
Wade AA, McKelvey KS, Schwartz MK. Resistance-surface-based wildlife conservation connectivity modeling: summary of efforts in the United States and guide for practitioners. Fort Collins, CO: US Department of Agriculture, Forest Service, Rocky Mountain Research Station. 2015.
- 36. Landau V, Shah V, Anantharaman R, Hall K. Omniscape.jl: Software to compute omnidirectional landscape connectivity. JOSS. 2021;6(57):2829.
- 37. Phillips P, Clark MM, Baral S, Koen EL, Bowman J. Comparison of methods for estimating omnidirectional landscape connectivity. Landscape Ecol. 2021;36(6):1647–61.
- 38. Middleton AD, Sawyer H, Merkle JA, Kauffman MJ, Cole EK, Dewey SR, et al. Conserving transboundary wildlife migrations: recent insights from the Greater Yellowstone Ecosystem. Frontiers in Ecol & Environ. 2019;18(2):83–91.
- 39.
Vermont Agency of Transportation. 2040 Vermont long-range transportation plan. Barre, Vermont, United States: Vermont Agency of Transportation. 2018.
- 40.
Vermont Fish and Wildlife Department. Vermont Wildlife Action Plan 2015. Montpelier, Vermont, USA: Vermont Agency of Natural Resources, Fish and Wildlife Department. 2015.
- 41.
Vermont Fish and Wildlife Department Strategic Plan 2017-2021. Montpelier, Vermont, USA: Vermont Agency of Natural Resources, Fish and Wildlife Department. 2018.
- 42. Anderson MG, Clark M, Olivero AP, Barnett AR, Hall KR, Cornett MW, et al. A resilient and connected network of sites to sustain biodiversity under a changing climate. Proc Natl Acad Sci U S A. 2023;120(7):e2204434119. pmid:36745800
- 43.
Hawk R, Miller C, Reining C, Gratton L. Staying connected in the northern Green Mountains: Identifying structural pathways and other areas of high conservation priority. Staying Connected Initiative: Staying Connected Initiative, 2012.
- 44.
Dewitz J. National Land Cover Database (NLCD) 2016 Products (ver. 2.0, July 2020): U.S. Geological Survey Data Release. Sioux Falls, SD: U.S. Geological Survey. 2019.
- 45.
U.S. Census Bureau. State population totals and components of change: 2020-2021. In: Bureau USC, editor. 2021.
- 46.
Austin JM, Alexander C, Marshall E, Hammond FM, Shippee J, Thompson E. Conserving Vermont’s natural heritage: A guide for community-based planning for the conservation of Vermont’s fish, wildlife, and biological diversity. 2nd ed. Montpelier, Vermont, United States: Vermont Agency of Natural Resources, Fish and Wildlife Department. 2013.
- 47.
Vermont Agency of Transportation. Summary mileage statistics. Barre, Vermont, USA: Vermont Agency of Transportation. 2019.
- 48.
Vermont Agency of Transportation. Vermont Agency of Transportation 2021 Fact Book and Annual Report. Barre, Vermont, USA: Vermont Agency of Transportation. 2021.
- 49.
Vermont Agency of Natural Resources. Vermont stream geomorphic assessment bridge and culvert dataset. Montpelier, Vermont, USA: Vermont Agency of Natural Resources. 2009.
- 50.
Vermont Agency of Transportation. VT road centerline. Barre, Vermont, USA: Vermont Agency of Transportation. 2021.
- 51.
Vermont Agency of Transportation. VT long structures – bridges and culverts. Barre, Vermont, USA: Vermont Agency of Transportation. 2020.
- 52.
Vermont Agency of Transportation. VT short structures – bridges and culverts. Barre, Vermont, USA: Vermont Agency of Transportation. 2020.
- 53.
Vermont Agency of Transportation. VTrans small culvert inventory – culverts. Barre, Vermont, USA. 2017.
- 54.
Roman J, Erickson JD. Economics of conservation in Vermont. Burlington, Vermont, USA: Gund Institute for Ecological Economics, Rubenstein School of Environment and Natural Resources. 2015.
- 55. Ma X, Bonaparte S, Corbett P, Orciari LA, Gigante CM, Kirby JD, et al. Rabies surveillance in the United States during 2021. J Am Vet Med Assoc. 2023;261(7):1045–53. pmid:36884381
- 56. Dickson BG, Roemer GW, McRae BH, Rundall JM. Models of regional habitat quality and connectivity for pumas (Puma concolor) in the southwestern United States. PLoS One. 2013;8(12):e81898. pmid:24367495
- 57.
Vermont Center for Geographic Information. Vermont High Resolution Land Cover (Beta Release 2019A). Montpelier, Vermont, USA; 2019.
- 58. Bezanson J, Edelman A, Karpinski S, Shah VB. Julia: A Fresh Approach to Numerical Computing. SIAM Rev. 2017;59(1):65–98.
- 59. Pearman-Gillman SB, Katz JE, Mickey RM, Murdoch JD, Donovan TM. Predicting wildlife distribution patterns in New England USA with expert elicitation techniques. Global Ecology and Conservation. 2020;21:e00853.
- 60. Beier P, Spencer W, Baldwin RF, McRae BH. Toward best practices for developing regional connectivity maps. Conserv Biol. 2011;25(5):879–92. pmid:21797924
- 61.
Canada Centre for Remote Sensing, Canada Centre for Mapping and Earth Observation, Natural Resources Canada, Comisión Nacional para el Conocimiento y Uso de la Biodiversidad, Comisión Nacional Forestal, Instituto Nacional de Estadística y Geografía, et al. 2015 Land Cover of North America at 30 meters. 2.0 ed. Ottawa, Ontario, Canada; 2020.
- 62. Koen EL, Garroway CJ, Wilson PJ, Bowman J. The effect of map boundary on estimates of landscape resistance to animal movement. PLoS One. 2010;5(7):e11785. pmid:20668690
- 63.
Landau VA. Vermont Terrestrial Passage Tool.jl. 2020.
- 64.
Pirie W. Spearman rank correlation coefficient. Encyclopedia of Statistical Sciences. 2nd ed. Hoboken, New Jersey, United States: John Wiley & Sons, Inc. 2006.
- 65. Albuquerque F, Beier P. Rarity-weighted richness: a simple and reliable alternative to integer programming and heuristic algorithms for minimum set and maximum coverage problems in conservation planning. PLoS One. 2015;10(3):e0119905. pmid:25780930
- 66.
Drasher CE, Murdoch JD. Improving roadway conservation investments in Vermont: developing a prioritization screening framework for reducing road wildlife mortality and improving wildlife movement through bridges and culverts. Barre, VT: Vermont Agency of Transportation. 2021.
- 67. Duffett D, D’Amico M, Mulero-Pázmány M, González-Suárez M. Species’ traits as predictors of avoidance towards roads and traffic. Ecological Indicators. 2020;115:106402.
- 68.
Inc. MaM. The Vermont cuvlert aquatic organism passage screening tool. South Burlington, Vermont: Vermont Agency of Natural Resources. 2009.
- 69. Cushman SA, McRae B, Adriaensen F, Beier P, Shirley M, Zeller K. Biological corridors and connectivity. Key Topics in Conservation Biology 2. Wiley. 2013. 384–404.
- 70. Ashrafzadeh MR, Khosravi R, Adibi MA, Taktehrani A, Wan HY, Cushman SA. A multi-scale, multi-species approach for assessing effectiveness of habitat and connectivity conservation for endangered felids. Biological Conservation. 2020;245:108523.
- 71. McGarigal K, Wan HY, Zeller KA, Timm BC, Cushman SA. Multi-scale habitat selection modeling: a review and outlook. Landscape Ecol. 2016;31(6):1161–75.
- 72. Zeller KA, McGarigal K, Cushman SA, Beier P, Vickers TW, Boyce WM. Sensitivity of resource selection and connectivity models to landscape definition. Landscape Ecol. 2017;32(4):835–55.
- 73. Diniz MF, Cushman SA, Machado RB, De Marco Júnior P. Landscape connectivity modeling from the perspective of animal dispersal. Landscape Ecol. 2019;35(1):41–58.
- 74. Zeller KA, Jennings MK, Vickers TW, Ernest HB, Cushman SA, Boyce WM. Are all data types and connectivity models created equal? Validating common connectivity approaches with dispersal data. Diversity and Distributions. 2018;24(7):868–79.
- 75. Zeller K, Lewison R, Fletcher R, Tulbure M, Jennings M. Understanding the Importance of Dynamic Landscape Connectivity. Land. 2020;9(9):303.
- 76. Zeller K, Wattles D, Bauder J, DeStefano S. Forecasting Seasonal Habitat Connectivity in a Developing Landscape. Land. 2020;9(7):233.
- 77. Ford AT, Clevenger AP, Huijser MP, Dibb A. Planning and prioritization strategies for phased highway mitigation using wildlife‐vehicle collision data. Wildlife Biology. 2011;17(3):253–65.
- 78. Ribeiro YGG, Ascensão F, Yogui DR, de Barros Ferraz KMPM, Desbiez ALJ. Prioritizing road mitigation using ecologically based land-use planning. Austral Ecology. 2023;48(4):761–73.
- 79. Dickson BG, Albano CM, Anantharaman R, Beier P, Fargione J, Graves TA, et al. Circuit-theory applications to connectivity science and conservation. Conserv Biol. 2019;33(2):239–49. pmid:30311266
- 80. McClure ML, Dickson BG, Nicholson KL. Modeling connectivity to identify current and future anthropogenic barriers to movement of large carnivores: A case study in the American Southwest. Ecol Evol. 2017;7(11):3762–72. pmid:28616173
- 81. Cameron DR, Schloss CA, Theobald DM, Morrison SA. A framework to select strategies for conserving and restoring habitat connectivity in complex landscapes. Conservat Sci and Prac. 2022;4(6).
- 82. Fullman TJ, Wilson RR, Joly K, Gustine DD, Leonard P, Loya WM. Mapping potential effects of proposed roads on migratory connectivity for a highly mobile herbivore using circuit theory. Ecol Appl. 2021;31(1):e2207. pmid:32632940
- 83. Koen EL, Bowman J, Sadowski C, Walpole AA. Landscape connectivity for wildlife: development and validation of multispecies linkage maps. Methods Ecol Evol. 2014;5(7):626–33.
- 84. Laliberté J, St-Laurent M-H. Validation of functional connectivity modeling: The Achilles’ heel of landscape connectivity mapping. Landscape and Urban Planning. 2020;202:103878.
- 85. Baldwin RF, Trombulak SC, Anderson MG, Woolmer G. Projecting transition probabilities for regular public roads at the ecoregion scale: A Northern Appalachian/Acadian case study. Landscape and Urban Planning. 2007;80(4):404–11.
- 86. Pearman-Gillman SB, Duveneck MJ, Murdoch JD, Donovan TM. Drivers and consequences of alternative landscape futures on wildlife distributions in New England, United States. Frontiers in Ecology and Evolution. 2020;8:1–19