Spatial Analysis of Potential Nesting Habitat for Florida Sandhill Cranes

The Florida sandhill crane Antigone canadensis pratensis is a subspecies of sandhill cranes, which is found throughout central and southern Florida, Georgia's Okefenokee Swamp, and in Mississippi (Bennett and Bennett 1992). The population in Florida, which is designated by the state as threatened, was estimated to have declined to about 4,500 birds by the mid-2000s from much larger numbers in previous decades (Nesbitt and Hatchitt 2008). Current estimates for Florida are thought to be between 4,000 and 5,000 birds, although these numbers are based on an analysis of habitat availability by the Florida Fish and Wildlife Conservation Commission (2020). Given dramatic changes in land use-land cover in Florida over the past decade and a half (Kautz et al. 2007; Hernández et al. 2012), which pose threats to sandhill cranes, there is an urgent need to map and quantify remaining Florida sandhill crane habitat (Florida Fish and Wildlife Conservation Commission 2013).

Habitat suitability index (HSI) models offer one approach for mapping and assessing habitat conditions for wildlife species (Roloff and Kernohan 1999). These models generally assess suitability on a continuous scale from 0.0 (completely unsuitable) to 1.0 (optimal). Contemporary HSI models are implemented using Geographic Information System (GIS) with spatial data layers representing habitat or other environmental features (e.g., Latif et al. 2018). Downs et al. (2008) developed a HSI model for greater sandhill cranes Antigone canadensis tabida in Ohio that identifies potential nesting sites (emergent marsh) located in close proximity to abundant wetlands, cropland, and pastures or grasslands. Florida sandhill cranes have similar habitat requirements to other subspecies of sandhill cranes (e.g., U.S. Fish and Wildlife Service 1991; Baker et al. 1995; Miller and Barzen 2016), so we adapted the existing HSI model for greater sandhill cranes based on the literature for Florida sandhill cranes and applied it to identify and assess the suitability of potential nesting sites across Florida.

Downs et al. (2008) also proposed an approach to calculate the maximum number of breeding pairs that can simultaneously occupy nests given that they both must be of some minimum quality and spaced some minimal distance apart, sometimes referred to as spatial or landscape carrying capacity (Lui and Wang 2010; Donovan et al. 2012; Glenn et al. 2017; Chapman and Byron 2018). A minimum HSI threshold is used to identify nesting habitat, while observations of nesting density are used to specify an appropriate minimum distance of separation. Once these values are specified, the maximum number of nests that can be occupied at once can be formulated as the anticovering location problem (ACLP). The ACLP is a combinatorial optimization problem that can be solved using linear programming techniques (Moon and Chaudhry 1984). This process involves writing a set of linear equations as a text file and then using specialized software to solve them. We used the ACLP to quantify the maximum number of occupied nests at two spatial scales—each county and the state as a whole, leveraging high performance computing as necessary to obtain solutions for the larger problems. This formal habitat suitability assessment will allow managers and decision makers to identify important habitat areas for crane conservation, determine possible limiting habitat features across the landscape, identify locations to target population monitoring efforts, and update estimates of potential breeding population size.

Florida sandhill cranes nest in emergent palustrine wetlands and roost in shallow wetlands (Layne 1981, 1983; Walkinshaw 1982; Dwyer 1990; Dwyer and Tanner 1992), often selecting sites free from human disturbance (Dwyer and Tanner 1992). They forage and loaf in shallow wetlands (Nesbitt and Williams 1990; Bennett 1992; Toland 1999), crop fields (Bennett 1992; Toland 1999; Nesbitt and Hatchitt 2008), pastures, prairies, and other grasslands (Nesbitt and Williams 1990). Open lands dominated by grass and other herbaceous vegetation are particularly important to cranes during the brood-rearing and prefledging stage because they support abundant insects and other invertebrates for foraging (Stys 1997). Downs et al.'s (2008) HSI model identifies optimal nesting sites for sandhill cranes based on the availability of these habitat features in close proximity to one another. Habitat is assessed based on five suitability variables (SVs): potential nesting habitat, immediate nesting area, brooding area (includes hatching through prefledging), wetland composition, and upland composition. Values for these SVs are then combined into an equation to compute a final HSI value. Here, we used this existing model but add an additional variable, road proximity, to reflect human disturbance and documented road effects on cranes in Florida (Folk et al. 2001; Table 1).

In practice, SVs are calculated for all individual raster, or grid, cells in a map simultaneously using a GIS and overlaid to compute and map the overall HSI values (Figure 1). Each raster cells is considered as a potential nest site. Potential nesting habitat, SV1, evaluates whether a raster cell in a map contains nesting habitat. The cell is assigned a value of 1.0 if it is classified as emergent wetland and 0.0 otherwise. Immediate nesting area, SV2, measures the availability of nesting habitat immediately surrounding the raster cell. It is measured as the percent of habitat within 120 m of the cell that is classified as emergent wetland. This SV reflects wetland size, where larger wetlands proportionally provide more nesting habitat than smaller ones, up to 4.5 ha. For example, if a cell is surrounded by 75% wetland, SV2 is equal to 0.75. SV3, brooding area, is measured as the abundance of prairie, pasture, grassland, or other herbaceous open land found within 450 m of the potential nest site. If those types comprise ≥20% of the area, then SV3 is equal to 1.0. If the amount is lower, the cell receives a value proportionally lower, because the computed percentage is divided by 20%. For example, if those habitats comprise 10% of the area, SV3 is equal to 0.5. Wetland composition, SV4, is based on the abundance of wetlands of all types within 780 m of the potential nest site. This SV assigns larger values to areas with more abundant freshwater wetlands. All freshwater wetland types were included because they can support foraging, roosting, and other activities and provide additional protective cover from disturbance further away from the nest. If the percentage of wetlands is >15%, then SV4 is equal to 1.0. If it is lower, the cell receives a value proportionally smaller. For example, if the percentage of wetlands is 6%, SV4 is equal to 0.4. Upland composition, SV5, is based on the abundance of grassland, prairie, pasture, cropland, or other herbaceous open land within 780 m of the potential nest site. Similar to SV4, if the percentage of suitable uplands is >15%, then SV5 is equal to 1.0. If it is lower, the cell receives a value proportionally smaller. For example, if the percentage of uplands is 7.5%, SV5 is equal to 0.5. Road proximity, SV6, is based on the distance between the potential nest site and nearest road. If the distance is >100 m, SV6 is equal to 1.0. Shorter distances receive proportionally smaller values. For example, if the distance is 90 m, then SV6 is equal to 0.9.

The six SVs are combined into an equation to compute overall HSI values. Multiplying SV1 by the remainder of the equation ensures that only cells containing nesting habitat receive HSI values greater than zero. Also, SV2 multiplied to ensure that large wetlands receive the largest values, while smaller wetlands reduce the overall HSI value proportionally to their size. Next, the product of SV1 and SV2 is multiplied by a weighted average of SV3, SV4, and SV5, where SV3 is weighted by 2 to reflect the importance of adequate brooding habitat for successful reproduction. Finally, this intermediate value is multiplied by SV6, which has a similar effect of proportionally reducing the overall HSI values for sites near roads. The HSI equation is structured in such a way to ensure overall HSI values are constrained between 0 and 1. Any cell not classified as nesting habitat is calculated as 0.0, regardless of the other SVs. Optimal habitat (HSI = 1.0) is only identified when all individual SVs are also equal to 1.0. Values between 0 and 1 represent a range of marginal- to moderate- to high-quality habitat. For example, using sample values from the previous paragraph describing the model, the overall HSI value for that cell would be as follows: 1 × 0.75 × {[(0.5×2) + 0.4 + 0.5]/4} × 0.9 = 0.32. In this case, habitat is marginal because of suboptimal amounts of wetlands, uplands, and herbaceous open lands.

We applied this updated model to Florida using 10-m-resolution land-cover data from the Florida Natural Areas Inventory published in 2018 to calculate SV1–SV5. We used road data from the U.S. Census Bureau's TIGER/Line database for 2018 to calculate SV6. We calculated the index using standard tools in ArcGIS v. 10.5 (ESRI, Inc., Redlands, CA; Table 1).

Once we identified nesting habitat for Florida sandhill cranes, we calculated the number of breeding territories that can be supported given a specified spatial distance between neighboring nests by using the ACLP. The ACLP determines the maximum number of nest sites that can be occupied such that no two sites are within a specified distance of one another. Although the ACLP has multiple formulations, the simplest formulation is described below.

“Where”

• K = maximum number of occupied nest sites,

• i = index of potential nest sites,

• d_ij = distance between potential nest site i and another site j,

• R = minimum distance required between occupied nests.

The objective function (1) maximizes the total number of potential nest sites selected to be occupied at one time. This simply represents a sum of all decision variables x_i, which identify whether a potential nest site is selected by the model (i.e., x_i = 1) or not (i.e., x_i = 0). Constraints (2) prohibits locating any two nests within the restriction distance, R, of one another. These constraints work by ensuring if a nest site x_i is selected by the model, then any nearby sites x_j cannot be selected at the same time because the sum of their values must be ≤1. Constraints (3) impose binary integer restrictions on the decision variables x_i, forcing their values to be either 0 or 1. In practice, such problems are formulated as linear programs (typically a text files containing the complete sets of linear equations) and solved using commercial, open source solvers, or custom algorithms (see Tong and Murray 2012). In practice, they attempt to solve the objective function to optimality—or as close as possible. Output records the value of the objective, values for all of the decision variables, and optimality gap (percentage the solution is guaranteed to be within the optimal solution).

We formulated the ACLP to estimate the maximum number of occupied nests in each county in Florida. We modelled maximum numbers for the state separately, because summing the county values would overestimate totals due to boundary effects. For reference, a county map is included (Figure S1, Supplemental Material). The ACLP is classified as an nondeterministic polynomial-time complete combinatorial optimization problem, which means it can be very challenging to solve for large problems (Dimitrijević et al. 2013), we used an aggregation approach to reduce the problem size. Rather than solving the problem using 10-m raster cells as a potential nest sites, we reduced the problem by aggregating them to a grid of hexagons with a 100-m spacing between centroids (Downs et al. 2020). If a hexagon contained nesting habitat based on a minimum HSI value threshold, then we considered its centroid as a potential nest location. In practice, this means that the hexagon centroids represent centers of possible nesting territories, with those centers located no more than ∼50 m from nesting habitat. This is a relatively negligible distance considering Florida sandhill crane home ranges average 447 ha (Nesbitt and Williams 1990). We computed maximum numbers of occupied nests for each county, as well as the state overall, using four minimum HSI value thresholds: 0.3, 0.5, 0.7, and 1.0. These values represent a range of habitat quality from marginal to moderate to high quality to optimal, respectively. We applied a restriction distance of 1,000 m for all scenarios because that value approximates the average nesting density of 1.9 birds/km² for pairs in Florida (Bennett 1989). Using these values and inputs, we formulated the ACLP for each scenario in three steps: (1) identify potential nesting sites based on the specified minimum HSI threshold and compute the distance between them; (2) use a custom Python script to read the centroid identifiers and distances from a database and write the appropriate ACLP equations in lp format; and (3) solve the lp files using IBM C-PLEX v. 12.9.0, a commercial optimization solver. Sample input and output files are available in Data S1–S4 (Supplemental Material). For all scenarios, we initially ran the solver for a maximum of 5,000 s on a Windows-based personal computer with 32GB of RAM. If the problem did not solve to optimality in that amount of time, we then ran the problems on a Linux machine with an AMD Threadripper 16-core processor with 128GB RAM.

Mapping SV1 reveals that emergent marshes available for nesting Florida sandhill cranes are widespread across the state, although more dominant in the southern and central portions of the state (Figure 2a,b). The average value of 0.07 suggests that emergent marshes cover 7% of the state (Table 2). Herbaceous open lands for brooding are more abundant in central Florida, extending into the panhandle, and noticeably sparse in the south (Figure 2c). Other than the binary SV1 layer, this variable had the lowest layer average, suggesting it is the most limiting. Available upland habitats, including a variety of agricultural cover types, are abundant across the state, although sparse in the south and some parts of the panhandle (Figure 2d). Wetlands are abundant state-wide (Figure 2e). Road effects generally follow the distribution of cities in the state, with significant roadless areas found in rural areas (Figure 2f). When combined, overall HSI values reflect the spatial intersection of these habitat features. Although local variability is somewhat masked at the state-wide scale, optimal nesting habitat is concentrated in central Florida, with marginal quality habitat also abundant in the south (Figure 3). Nesting habitat is very sparse in the panhandle and other parts of northern Florida.

Optimal solutions for maximum numbers of occupied nests were found within 5,000 s for all counties and the state using minimum HSI thresholds of 1.0 and 0.7. For a HSI threshold of 0.5, optimal solutions were found on the initial runs for all but Okeechobee County and the state. These were solved to optimality on subsequent runs on the higher performance computer. For the 0.3 HSI threshold, optimal solutions were found on the initial runs for all but DeSoto, Glades, Hendry, Highlands, Okeechobee, Osceola, and Sarasota counties, but they also were solved to optimality on subsequent runs in <1 wk of running time. However, the state-wide problem was terminated after 1 mo of running time with an optimality gap of 0.78%.

Estimates of maximum occupied nests were largest for Polk, Osceola, Okeechobee, Hendry, and Highlands counties, all of which are located in the central part of the state (Table 3). For all counties, as the minimum HSI value is reduced, maximum nesting numbers increase substantially. For example, depending on the threshold, estimates for Pasco County ranged from 39 nests (HSI = 1) to 154 (HSI ≥ 0.7) to 246 (HSI ≥ 0.5) to 423 (HSI ≥ 0.3; Figure 4). State-wide estimates displayed a similar pattern ranging from 1,794 nests (HSI = 1) to 5,540 (HSI ≥ 0.7) to 8,723 (HSI ≥ 0.5) to 14,530 (HSI ≥ 0.3). On average, county estimates were five times as large when comparing the 0.5 HSI threshold with 1.0, indicating the results are sensitive to the threshold specified (Figure 5). This is expected because greater availability of habitat would be expected to accommodate larger numbers of nests. However, increases were not uniform between counties because of differences in both the quality and spatial distribution of nesting habitat.

Our HSI map displays the spatial distribution of Florida sandhill crane nesting habitat, identifying specific wetlands that may potentially support breeding pairs. Additionally, mapping individual SVs highlights habitats that may be limiting population growth and range expansion. Few cranes have been documented in the panhandle and northeastern Florida, which can be explained largely by the lack of emergent marshes and limited brooding habitat. In the south, emergent marshes are abundant, but both brooding habitat and other uplands are sparsely distributed. However, central Florida supports all requisite habitat types, often in close proximity to one another, supporting optimal habitat conditions for nesting in some locations. The most limiting factor appears to be herbaceous open lands for brooding, which reduces overall suitability in many local areas where emergent marshes, other wetlands, and cropland are abundant, even within central Florida.

Although the HSI map identifies potential nesting habitat, simply computing the area of habitat would not be a reliable indicator of potential population size because it would not reflect the required spacing between neighboring territories. Conversely, estimates of maximum numbers of occupied nests obtained by solving the ACLP do reflect the spatial distribution of habitat and should be a reliable indicator of potential population size. However, as expected, estimates are sensitive to both the minimum HSI threshold and specified restriction distance. The distance has little effect at optimal HSI values but is substantial at lower thresholds. Despite this trend, the 1,000-m spacing reflects the average observed in the state, so values for those scenarios are the most reasonable to report. However, greater sandhill cranes have been observed nesting at much greater densities in Wisconsin (Barzen et al. 2016), so these estimates might be conservative. In terms of HSI threshold, a conservative approach is to use a minimum value of 0.7, which represents high quality. This ensures that potential nest sites are near to significant amounts of herbaceous open lands needed for brooding. Therefore, 5,540 pairs represents a conservative estimate of the number of breeding pairs of Florida sandhill cranes that can be supported in Florida, with greater numbers possible if cranes will nest in marginal- to moderate-quality habitats or at greater densities than previously observed. In more optimistic scenarios, >15,000 pairs could be supported, which is a much larger value than that currently estimated by the Florida Fish and Wildlife Conservation Commission (2020).

A potential limitation is that GIS and optimization techniques are data- and processing-time intensive and require some level of training. Previous efforts to compute maximum numbers of occupied nests for greater sandhill cranes were completed only for relatively small, isolated management areas (Downs et al. 2008). However, our results demonstrate these problems are solvable on a state-wide scale with modest high performance computing resources. Only one of the state-wide scenarios did not solve to optimality, though the optimality gap was <1%, which is a relatively small amount for this problem. For larger problems, alternative formulations of the ACLP can be used to improve solvability (Murray and Kim 2008). Future work might compare the availability of nesting habitat in Florida with other regions, or repeat the analysis over different time periods to evaluate changes to nesting habitat availability.

Combined, the HSI map and nest occupancy estimates offer wildlife managers tools to identify where to find a species and predict potential population sizes. Given the Florida sandhill crane is currently state-listed as threatened, these outputs provide useful resources to guide a variety of research and management activities. For example, these maps might be used to guide population surveys, the results of which might be used to update breeding pair estimates if they are observed nesting at greater densities. Aerial surveys in high-suitability nesting habitat or road surveys along upland habitats near nesting habitat could be conducted to locate breeding pairs (Hereford and Dedrickson 2018). Additionally, the HSI maps might also be used to identify crane habitats in need of protection or guide restoration efforts. For example, it may be possible to target grassland restoration efforts in areas where herbaceous open land is the only habitat missing. This activity that might be used to increase population sizes across the Florida landscape, especially in the south where wetlands are abundant.

The authors would like to thank the associate editors and anonymous reviewers for their thoughtful comments that greatly improved this manuscript.

Any use of trade, product, website, or firm names is for descriptive purposes only and does not imply endorsement by the U.S. Government.

Please note: The Journal of Fish and Wildlife Management is not responsible for the content or functionality of any supplemental material. Queries should be directed to the corresponding author for the article.

Data S1. Python program used to generate linear programming (lp) files for the anticovering location problem (ACLP). The ACLP can be formulated to calculate the maximum number of breeding bird pairs that can simultaneously occupy nests given that they both must be of some minimum quality and spaced some minimal distance apart. The program requires a csv file that records the distances between potential nest sites that are located less than that distance from one another. It outputs a text file in lp format that can be used with a commercial optimization solver. The user must specify the file path for both the input csv file and the output lp file in the program and the total number of potential sites for the problem. The program provided was used to calculate the spatial carrying capacity for Florida sandhill cranes Antigone canadensis pratensis in Alachua County, Florida, in 2018.

Found at DOI: https://doi.org/10.3996/092019-JFWM-077.S1 (1 KB PY).

Data S2. Sample csv file recording distances between potential nest sites. The csv file requires three columns: the identifier of the potential nesting site (“INPUT_FID”), the identifier for the nearby potential nesting site (“NEAR_FID”), and the distance between them (“DISTANCE”). The first column in the cvs file records the record number but is not used by the program. The sample csv file provided was used as an input for the Python program in S1, representing potential nesting sites in Alachua County, Florida, in 2018.

Found at DOI: https://doi.org/10.3996/092019-JFWM-077.S2 (8.52 MB CSV).

Data S3. A sample lp file generated from Python program S1 using csv file S2. This lp file records equations used to solve the anticovering location problem, including the objectives (“Maximize”), constraints (“SUBJECT TO”), and binary decision variables (“BINARIES”). This lp file was used with csv file S2 to calculate the maximum number of nesting pairs for Alachua County, Florida, in 2018using IBM C-PLEX.

Found at DOI: https://doi.org/10.3996/092019-JFWM-077.S3 (5.07 MB TXT).

Data S4. Shapefile representing locations of potential nest sites for sample problem from S1 to S3. The GIS layer provides a spatial reference for mapping the locations of potential nest sites in the sample problem of calculating the maximum number of occupied Florida sandhill crane nests in Alachua County, Florida, in 2018.

Found at DOI: https://doi.org/10.3996/092019-JFWM-077.S4 (65 KB ZIP).

Figure S1. Reference map of counties in Florida, in 2018.

Found at DOI: https://doi.org/10.3996/092019-JFWM-077.S5 (617 KB TIF).

Reference S1.Florida Fish and Wildlife Conservation Commission. 2013. A species action plan for the Florida sandhill crane. Tallahassee: Florida Fish and Wildlife Conservation Commission.

Found at DOI: https://doi.org/10.3996/092019-JFWM-077.S6 (2.04 MB PDF).

Reference S2.Latif QS, Saab VA, Haas JR, Dudley JG. 2018. FIRE-BIRD: a GIS-based toolset for applying habitat suitability models to inform land management planning. General Technical Report RMRS-GTR-391. Fort Collins, Colorado: U.S. Department of Agriculture, Forest Service, Rocky Mountain Research Station.

Found at DOI: https://doi.org/10.3996/092019-JFWM-077.S7 (14.75 MB PDF).