Our work was conducted under a permit from the University of North Florida Institutional Animal Care and Use Committee (permit #18-005).
Field methods
We performed nighttime spotlight surveys with an outboard motorboat throughout 2019 to determine relative alligator abundance, distribution, and habitat selection. This technique is an established method for estimating relative population sizes in crocodilians across heterogeneous habitat32. However, a limitation of spotlight surveys is the variation in detection probability caused by different environmental conditions or observers29. To control for these effects, we implemented a standardized survey protocol33,34. All surveys covered the first 8 km of nine tributaries within the lower St. Johns River system, starting at the point where each tributary meets the main channel of the river (Fig. 1). We limited our surveys to the first 8 km because some tributaries contained low bridges that blocked boat access after this point. We chose tributaries that were surrounded by different amounts of urban land cover such that our surveys spanned an urbanization gradient from approximately 5 to 80% urban land cover within 1 km of the river’s edge (Fig. 2). GIS analyses also revealed that land use patterns around the St. Johns River are dynamic, with different urban land cover proportions at 0.1, 1, 3, and 5 km from the water’s edge for each tributary (Fig. 2). To reduce temporal bias, we conducted surveys over the span of 1 year and segregated sampling periods into four distinct seasons (winter [Dec–Feb], spring [Mar–May], summer [Jun–Aug], and fall [Sep–Nov]. We surveyed each tributary one time during the middle month of each season, resulting in a total of four surveys per tributary. We surveyed the tributaries in a quasi-random fashion because the tributaries closest to the mouth of the St. Johns River are under significant tidal influence, so we timed surveys of those tributaries during periods of high tide in order to access the full survey area. We only performed surveys when rainfall was absent and wind speeds were below 16 km/h since these factors have been shown to affect alligator detection probability24. Quasi-random sampling over the span of a year was best suited to randomize environmental conditions that affect nighttime spotlight survey counts, such as water level, temperature, moon phase, and moon illumination24,31,35,36.
Map of the tributaries surrounding the lower St. Johns River that were surveyed as part of our study (white areas). From northeast to south: Clapboard Creek, Dunn Creek, Broward River, Trout River, Arlington River, Ortega River, Doctors Lake, Julington Creek, and Black Creek. The map was created with ArcGIS Pro 2.6 (https://arcgis.pro/).
Levels of urban development (FLUCCS code 1000) surrounding the tributaries of the St. Johns River that were surveyed in this study. Land use was quantified using 0.1, 1, 3, and 5 km buffers around each tributary transect.
We began all surveys no earlier than 30 min after sunset and we maintained a constant boat speed of 10–12 km/h. At the start and end of each survey we recorded moon phase, weather conditions, visibility, ambient light, air temperature, water temperature, and salinity. We detected alligator eyeshine primarily using two 1200 lm handheld spotlights, but we also used additional handheld lights (6000 lumens) often throughout the surveys. As soon as we detected eyeshine we approached the alligator at reduced speed. We placed each individual into a size class (30–90 cm [juvenile], 90–180 cm [sub-adult], 180–270 cm [adult], 270–360 cm [large adult], + 360 cm [largest adult]) by estimating the distance between the eyes and the tip of the snout37,38. If an alligator submerged before size estimation could take place, we recorded its length as unknown or simply larger or smaller than 180 cm. At each sighting we recorded global positioning system location using the on-deck boat navigation unit. We measured environmental characteristics at each sighting using a YSI meter (Pro2030; YSI; Yellow Springs, Ohio, USA), a thermometer, and a sky quality meter (SQM; Unihedron; Grimsby, Ontario, Canada).
We recorded information about habitat characteristics for each sighting following previous studies13,26. We first visually characterized habitat in a 10 m radius circle centered on the alligator sighting location (“used habitat”). We recorded the proportion of open water, emergent vegetation, floating vegetation, anthropogenic structure, and dry ground within the circle, as well as the alligator’s distance from shore, vegetation, and anthropogenic structure. We then visually classified the same habitat characteristics in a 20 × 100 m plot centered on the alligator sighting location and stretching along the shoreline (“available habitat”). If an alligator sighting occurred entirely in open water, then we shifted the available habitat plot to the closest shoreline. For each used habitat circle and available habitat plot, we classified the respective shorelines as natural, hardened, or mixed, depending on if the shore was totally vegetated, subject to anthropogenic armoring, or a mixture of the two types, respectively. We also estimated the proportion of shoreline found within these areas that were covered in naturally growing vegetation rather than anthropogenically altered lawns.
Land use classification
We used ArcGIS Pro (ESRI; Redlands, CA, USA) for all spatial data manipulation and visualization. We acquired land use and cover data from the St. Johns River Water Management District (SJRWMD) via the Florida Geographic Data Library. For all analyses we used data from the most recent SJRWMD dataset, which was from 2014.
We split a 100 k definition polygon of the St. Johns River to create smaller units representing each tributary transect. The resulting features consisted of the main portion of each tributary surveyed where lower order streams that were not surveyed were deleted. Because the extent to which alligators respond to land use changes was not known a priori, we buffered the transect polygon feature for each tributary to 0.1, 1, 3, and 5 km to further clip the SJRWMD land cover and use data layer. By creating four buffers for each of the nine tributaries, we generated a total of 36 land cover and use layers.
We classified land use types through the Florida Land Use and Cover Classification System (FLUCCS), as cited in SJRWMD metadata documentation. This hierarchical coding scheme contains four levels, of which we used the highest level (level 1) designation. This particular level classifies land use into nine distinct categories. These categories included urban and built-up; agriculture; upland nonforested; upland forests; water; wetlands; barren land; transportation, communication, and utilities; and special classification. For the purposes of this study, we only included defined terrestrial land use types in statistical analyses. These land use types were urban and built-up (e.g., residential, industrial, and recreational areas), agriculture (e.g., cropland, pastures, aquaculture), upland nonforested (e.g., shrub and brushland), upland forests (e.g., coniferous forests, hardwood forests, tree plantations), wetlands (e.g., freshwater/saltwater marshes, mangrove swamps, wet prairies), barren land (e.g., beaches other than swimming beaches, borrow areas, spoil areas), and transportation, communication and utilities (e.g., highways, electrical power facilities, wastewater treatment facilities). We calculated the proportions of each land use type using each respective land use shape area divided by total shape area.
Statistical analyses
To determine if environmental conditions and/or land use characteristics affect broad scale alligator distribution, we performed multiple analyses using SPSS (IBM; Armonk, NY, USA). We included all alligator sightings in our analyses, but we did not apply population estimate correction equations to the alligator counts because they tend to underestimate population numbers in crocodilians39. Sighting data used in statistical analyses therefore represent relative alligator abundance, not estimates of true alligator population size. We first checked normality for each variable using Kolmogorov–Smirnov and Shapiro–Wilk tests to determine if parametric or nonparametric tests were appropriate. Normality varied greatly across the suite of variables; therefore, Spearman’s rho and Pearson’s correlation coefficient were used when appropriate. We then performed simple linear regression to determine if there were any direct relationships between relative alligator abundance and individual variables. We performed these tests for alligator counts in each tributary by season and for the average number of sightings per tributary across seasons. We also averaged environmental variables for each tributary by season and for the average value per tributary across seasons. We tested for the effect of land use at all four buffer sizes for each tributary, including all relevant terrestrial land use types.
We then performed multiple linear regression analyses in a stepwise manner. This modeling system excluded variables found to be highly correlated with other variables (multicollinear) and retained variables that significantly contributed to the model (P ≤ 0.05). We then performed these tests on modified datasets that did not contain the two most saline tributaries to further validate preliminary findings. When more than one significant model was produced for a given data set, we calculated AICc values to rank models while penalizing model complexity and accounting for our small sample sizes.
To evaluate habitat selection, we compared percent shoreline vegetation and the proportions of habitat characteristics found in the 10 m radius circle to those found in the remaining areas of each respective 20 × 100 m plot using the Wilcoxon signed rank test. When comparisons could be made between two normally distributed groups of data, we used a paired sample t test instead. While comparing used to available habitat data was the basis of the tests, the amount of data per analysis differed between analysis groups. The first group was composed of all habitat selection data across time and space. This “global” dataset was the most robust in terms of sample size but may have been biased by double counting individuals across time. The second group was divided by season, so analyses were performed on all data collected within a season across space. This group removed the bias of double counting individuals but may be affected by variation in the number of sightings per season and tributary.
Source: Ecology - nature.com
