Noninvasive camera data and spatial capture–recapture models reveal strong temporal variation in fawn survival

Abstract

In south Florida, white-tailed deer (Odocoileus virginianus) are the primary prey of the endangered Florida panther (Puma concolor coryi). Deer populations in some regions of south Florida have declined in recent years, and the role of fawn survival and recruitment in these declines is unknown. Determining known-fate survival of fawns is challenging, requires invasive and costly methods, and often has a limited geographic scope. We deployed 180 cameras throughout the Florida Panther National Wildlife Refuge and the Big Cypress National Preserve to understand how environmental variables influence fawn survival. We identified 271 fawns from 12,715 photographs in 2015 and 2016. We utilized a noninvasive sampling method coupled with a spatial capture–recapture model to estimate the number of fawns born, the spatial distribution of birth locations, and the number of fawns that survived to recruitment (180 days old) during two fawning seasons. We found strong evidence of temporal variation in survival, but little evidence of spatial variation. Within the 10,941-ha study area, we estimated that 305 (95% CI: 245–385) fawns were born in 2015 and 278 (212–381) fawns were born in 2016. In 2015, 36% (110) of the estimated 305 fawns survived to 180 days. However, in 2016, only 13% (36) of the estimated 278 fawns survived to 180 days. The large difference in recruitment between years was likely driven by record flooding in 2016. Our data suggest that extreme weather events, coupled with high adult mortality, likely contributed to recent deer population decline in south Florida through reduced fawn recruitment. Unlike studies of known-fate fawn survival that require labor-intensive and invasive capture of both adults and neonates, our approach relies exclusively on camera data, which makes it possible to conduct studies over broad spatiotemporal scales in challenging environments to illuminate the drivers of variation in juvenile survival.

INTRODUCTION

For many long-lived wildlife species, population growth rates are strongly influenced by variation in neonate and juvenile survival and recruitment (Ciuti et al., 2015; Knoerr et al., 2022; Robinson, Desimone, et al., 2014; Sæther et al., 2013). Estimating survival and recruitment of neonates often requires costly, labor-intensive, and invasive techniques (Duquette et al., 2014; Mills et al., 2008). For many species, deploying telemetry devices requires locating neonates within a few hours of birth, and the transmitters must be designed to allow for rapid growth (Diefenbach et al., 2003; Duquette et al., 2014). Neonates may be too small to carry telemetry devices with sufficient battery power to monitor survival to the desired age. Camera-based noninvasive methods offer a solution to these challenges and can be used in species whose juveniles have unique natural markings, such as spot or stripe patterns (Royle et al., 2014).

The white-tailed deer (Odocoileus virginianus; hereafter, deer) is a culturally and economically important game species in North America (Conover, 2011). Information about local abundance and productivity is needed to assess population viability and inform management. However, studies of neonate deer survival are often characterized by small sample sizes and short durations due to the technological limitations of VHF collars and the high costs of GPS collars, limiting the information available to wildlife managers. Fawn mortality rates range from 9.3% to 90% (DeYoung, 2011), with variation primarily attributed to differing predation rates (Nelson et al., 2015; Shuman et al., 2017). However, other variables such as the quality and quantity of vegetation, fire history, and anthropogenic disturbance can also influence fawn survival (Berger, 2007; Nelson et al., 2015; Shuman et al., 2017; Tollefson et al., 2011).

Although regional environmental variables can impact fawn production and survival, it is often assumed that population-level recruitment rates of many temperate ungulates are stable over time (McCullough, 1987). However, recent studies have demonstrated that annual recruitment rates fluctuate due to changes in habitat quality, predator composition, seasonal mast, and weather patterns (Ciuti et al., 2015; Gulsby et al., 2015; Tollefson et al., 2011). Fawn recruitment may be able to compensate for low annual adult survival in some populations; therefore, management requires accurate estimates of fawn production, survival, and the variability in these parameters (Chitwood, Lashley, Kilgo, Moorman, et al., 2015; Peters et al., 2020; Robinson, Diefenbach, et al., 2014).

Monitoring data and hunter harvest records indicate that white-tailed deer populations have declined recently in some portions of south Florida, raising concerns about the viability of these deer populations (Garrison et al., 2011). Meanwhile, the Florida panther (Puma concolor coryi) population increased from 20–30 individuals in the early 1990s to 120–230 twenty years later (FWC, 2017; Johnson et al., 2010). Florida panther predation is the largest source of mortality for adult deer within the panther's range (Bled et al., 2022), and the adult deer survival rate is lower than it was in the 1990s (Bled et al., 2022; Land, 1991). Although adult female survival is often the most influential vital rate in deer population growth, low fawn recruitment can further constrain growth in low-density herds (Chitwood, Lashley, Kilgo, Moorman, et al., 2015), such as those of south Florida. Therefore, ensuring the sustainability of the low-density deer population in south Florida depends on adequate replacement through fawn recruitment.

Spatial capture–recapture (SCR) methods have been developed to estimate survival and recruitment for many species while accounting for variation in individual detection probability (Chandler et al., 2018; Gardner et al., 2010; Royle et al., 2014). Pairing SCR methods with natural and individually identifiable markings makes it possible to estimate survival and abundance over large spatial and temporal extents, without the costly and labor-intensive capture methods associated with telemetry studies (Royle et al., 2009; Thornton & Pekins, 2015).

Using noninvasive SCR techniques, we examined white-tailed deer fawn abundance and survival in south Florida. We monitored fawns using data from passive trail cameras over a larger spatial range and to an older age than is typical for known-fate fawn survival studies, which are often reliant on VHF collar data. We estimated abundance, survival and recruitment to 180 days, birth density, and detection probability for deer fawns in south Florida. We modeled the effect of environmental variables, such as vegetation type and surface water, on spatial and temporal variation in fawn abundance, survival, and recruitment. Our goals were to estimate fawn recruitment in south Florida to aid in population management, to evaluate the influence of environmental predictors on fawn survival, and to demonstrate the applicability of our noninvasive SCR model.

METHODS Study area

The Big Cypress Basin in south Florida is a low-lying subtropical savannah with distinct wet and dry seasons. Mean annual rainfall was 145 cm during the years of our study, with 60%–80% of the rain falling between May and September (PRISM Climate Group, Oregon State University). Dominant vegetation types are determined by topography, hydroperiod, and soil type and include pine flatwoods, cypress forests, hardwood hammocks, marshes, and prairies (Duever et al., 1986). The 24-year average for surface water depth measured at low-lying marsh gauges during the mid-dry season before the study was 35.2 cm. During fawning in the first year of data collection, the mean surface water depth at marsh gauges was 33.7 cm. However, the second fawning season of the study was one of the wettest dry seasons on record and had a mean surface water depth of 71.1 cm at the same gauges (GoHydrology.org).

We worked on the Florida Panther National Wildlife Refuge (FPNWR) (the Refuge) and in two management units of the Big Cypress National Preserve (BCNP): North Addition Lands and Bear Island (Figure 1). The Refuge is closed to public access and does not permit hunting but has an extensive network of off-road vehicle trails and roads used for management activities. The North Addition Lands are open to the public for hiking and hunting, but motorized vehicles are prohibited. Bear Island allows hunting and permits public access to an extensive network of off-road vehicle and hiking trails.

FIGURE 1. Locations of 180 trail cameras in the Florida Panther National Wildlife Refuge (FP Refuge) and the Big Cypress National Preserve's Bear Island and North Addition Lands units in Florida, USA. Camera data were collected from January 1 to October 1, 2015 and 2016.

In south Florida, deer experience unique stressors from seasonal flooding, low-quality forage, and a diverse predator community, including the endangered Florida panther (FWC, 2017; Johnson et al., 2010). South Florida deer are also prey for alligators (Alligator mississippiensis), black bears (Ursus americanus), bobcats (Lynx rufus), coyotes (Canis latrans), and invasive Burmese pythons (Python bivittatus; Boback et al., 2016; Maehr et al., 1990; Maehr, 1997; McCown & Scheick, 2007). Severe fluctuations in surface water levels in south Florida can cause changes in deer movement, survival, fawn production, and fawn recruitment (Garrison et al., 2011; MacDonald-Beyers & Labisky, 2005). South Florida deer also exhibit lower productivity than many deer populations in temperate North America, with 1.18–1.26 fetuses/pregnant doe in Florida compared with 1.5–2.1 fetuses/pregnant doe in other southeastern deer herds (Fleming et al., 1994; McCown, 1991; Richter & Labisky, 1985). Fawning season for south Florida deer is much less synchronized than in more temperate regions and has been documented from November to March (Fleming et al., 1994), with most fawning timed to match the peak of the dry season in mid-February (Garrison & Gedir, 2006; Richter & Labisky, 1985).

Camera deployment

We deployed 180 passive white-flash motion-sensor camera traps (HCO Outdoor Products, Norcross, GA, USA) to cover gradients of hydrology and human pressure in the study area. We placed one array of 60 cameras in each study unit (Figure 1). In each array, 40 cameras were placed on established roads or trails and spaced approximately 700 m between camera traps. The other 20 cameras in each array were placed off-trail, approximately 250 m from a trail and a paired on-trail camera (Crawford et al., 2019; Margenau et al., 2022). Cameras were placed approximately 0.3 m off the ground and adjusted as needed to avoid wet-season inundation. All cameras were programmed to record single images with no delay between subsequent captures. We visited all camera locations monthly to download data.

Ecological variables

We derived vegetation cover type from the Florida Natural Areas Inventory Cooperative Land Cover (FNAI CLC) 3.2 site-level data at 10-m resolution (FWC/FNAI, 2016). We reclassified the FNAI CLC cover types present in the study area into four broad categories to capture the variation in seasonal hydroperiod, cover availability, canopy closure, and forage production that may have an impact on fawn survival. The vegetation composition of the three 30-km² camera arrays consisted of 29.3% cypress forest, 29.1% pine flatwoods, 35.5% open canopy (marsh, wet, and dry prairie), and 5.5% hardwood hammock (Table 1). The 180 cameras within those three camera arrays were placed with 21.1% of the cameras in cypress, 42.8% in pine flatwoods, 31.1% in open canopy, and 5% in hardwood hammock.

TABLE 1 Summary statistics of predictor variables used in spatial capture–recapture models of fawn birth location, fawn survival, and daily detection rate in south Florida in 2015 and 2016.

Environmental variables	North Addition Lands	Bear Island	FPNWR	Total (10,941 ha)
Percent cover
Cypress	30.7	8.4	50.3	29.3
Flatwoods	37.9	18.4	31.2	29.1
Open canopy	30.8	64.2	11.2	35.5
Hammock	0.6	9.0	7.3	5.5
Human photos per day per camera
2015	0.39 (0–206)	0.43 (0–120)	0.32 (0–94)	0.38 (0–206)
2016	0.56 (0–142)	0.43 (0–65)	0.22 (0–170)	0.40 (0–170)
Water depth at camera (m)
2015	−0.22 (−1.09, 0.72)	−0.15 (−1.24, 0.72)	−0.25 (−1.24, 0.72)	−0.21 (−1.24, 0.72)
2016	0.31 (−0.30, 0.86)	0.38 (−0.45, 0.86)	0.27 (−0.45, 0.86)	0.32 (−0.45, 0.86)

Note: Camera covariates are presented as the mean, with the range in parentheses.

Abbreviation: FPNWR, Florida Panther National Wildlife Refuge.

We quantified daily surface water depth as a spatially and temporally variable covariate using the FNAI cover types and water gauges in marshes across BCNP. We averaged the daily mean surface water depth from three hydrologic depth gauges maintained by the Everglades Depth Estimation Network project and the US Geological Survey. We used a constant correction value for the elevation in each cover type to inform the mean daily inundation of surface water at the point location of each camera trap (R. Sobczak, National Park Service, personal communication).

Data collection and preparation

We identified fawns in photos using their spot patterns, which were distinctive from birth until approximately 7 months old in our study area. We identified unique spot patterns using images from both sides of each fawn, as well as photos from directly behind or in front of the fawn (Chandler et al., 2018). As does and fawns often spent several minutes traveling or foraging in front of cameras, we were able to capture both flanks of most fawns. To avoid mismatching or double-counting fawns that could not be confirmed on both sides, we compared each single-sided fawn detection to all double-sided known fawns in that grid and year. Matches were identified as the same individual. Then, we retained either all-left-sides or all-right-sides for any unmatched single-sided fawns in each grid and year. Each retained single-sided fawn was assigned a unique ID, and any single-sided detections from the opposite side in the same grid and year were discarded to avoid giving two unique IDs to two single-sided detections of the same individual.

We estimated birth date ranges for all fawns based on morphological characteristics. Body mass, chest girth, body length, hind foot length, and head length are useful to determine age of fawns when measuring hoof growth is not possible (Sams et al., 1996). We considered the body size of fawn relative to the doe, head size and shape, brightness of spots, and length of hind foot relative to body size to determine minimum and maximum birth dates. Two skilled observers with experience in capturing and aging fawns independently viewed all images of each unique fawn. Fawns detected very young (<10 days) and surviving many months provided a baseline for determining the birth date ranges of fawns that were detected less frequently. Both observers created a window of possible birth dates for each identified fawn by estimating the minimum and maximum age of the fawn in each detection (Appendix S1: Figures S1 and S2), then using the dates of the detections to calculate the earliest and latest birth date for each fawn. Discrepancies between the independent birth date windows were low (<10% different) and were resolved by combining the earliest and latest birth date from the two observers to create a conservative birth window for each individual.

Fawns often remained in front of a camera for multiple minutes. We removed nonindependent detections if a fawn had been previously detected at the same camera within 1 h. We then created spatially referenced capture histories indicating the independent detections of each fawn at each camera during the study (Appendix S1: Figures S3 and S4). We used all independent detections from January 1 to October 1 in both 2015 and 2016, as we did not detect any new individuals between October 2 and December 31 in either year.

Model fitting

We modeled the number of fawns born, spatiotemporal variation in birth location, daily survival, and detection rate using an expansion of the recently developed model of Chandler et al. (2018). Chandler et al. (2018) did not model spatial or temporal variation in productivity or individual survival, and they only analyzed one season of data from a small subset of the study area. Thus, our model further builds upon this previous work and allows for additional insights into the factors influencing variation in fawn survival and recruitment. This model is a type of open population SCR model (Efford & Schofield, 2020; Gardner et al., 2010; Glennie et al., 2019) in which birth locations are modeled as an outcome of a spatiotemporal point process. A birth site location in our analysis is defined as the point at which a fawn was born, which we estimated using the camera data (see next paragraph). Spatial variation in the density of birth locations was modeled as a function of vegetation cover type (Table 1). Individual birth date was modeled using a uniform distribution between the earliest and latest birth dates, as estimated for each individual fawn by the independent observers.

Inferences were drawn using a statistical model in which detection probability was treated as a function of the individual's age and the distance between its latent birth location and the location of the camera traps. Because we did not capture neonates at the birth site, we estimated birth site locations using the locations and ages of fawns detected by our camera traps. Specifically, the model assumes that the scale parameter of the detection function increases with age, such that young fawns are unlikely to be detected far from their birth site. Thus, for fawns detected early in life, the model estimates their birth location to be close to the detection location. Estimates of birth locations are much more uncertain for fawns first detected at older ages. However, the parameters of the detection function are estimated using the encounter history data and the age data, and uncertainty is fully accounted for when modeling spatial variation in the density of the birth site locations. The details of the statistical model can be found in Chandler et al. (2018).

Additional detection parameters were modeled using trap-specific, temporally varying covariates and included trail status, daily water depth, and human activity rate (Table 1). We set the probability of detection to 0 on any day that a camera was not operational. Trail status (on-trail/off-trail) was a binary variable. Daily water depth was calculated at the camera level using the correction factor and the daily measurement from the hydrologic depth gauges. Human activity at each camera was the number of human photos detected by each camera on each day throughout the fawn monitoring period.

To define the state space, defined as the region within which we estimated abundance of birth site locations, we created an 800-m buffer around each trail camera and combined the resulting polygons to create a 10,941-ha region around the three camera grids. This buffer was large enough to include the birth locations of all fawns that could have been detected by our cameras. The state space (of birth site locations) is sometimes referred to as the observation window or the area of integration in the point process and SCR literature. We first estimated the detection parameters with a larger buffer, and then reduced it to meet the requirement that fawns born near the edge of the spatial region had a negligible detection probability.

We used Bayesian inference with a custom Gibbs sampler written in R (R Core Team, 2022) to fit one model for each study year. We right-censored fawn survival at 180 days old to account for spot loss. Because twinning is rare in our study area (Richter & Labisky, 1985), we modeled all birth locations as statistically independent events, conditional on the covariate effects. Each year's model contained the four vegetation cover types as spatial covariates in both the birth and survival submodels, age as a temporal covariate in the survival submodel, and all three camera-level covariates in the detection submodel. We modeled individual lifetimes using a discrete-time survival model. Daily survival probability was a function of individual-specific, spatially and temporally varying covariates, including individual age and vegetation cover type. We used Normal(0, 25) priors and Normal(0, 2.25) priors for covariates of the density and survival submodels, respectively. Flat priors were used for the covariates of the detection model. We used M = 400 for data augmentation. Each year's model was run with three chains for 42,000 iterations each, discarding the first 2000 iterations as burn-in and thinning the posterior chains by 10. Convergence was assessed by using Gelman–Rubin diagnostic statistics (all R-hat values were <1.1) and visual inspections. Model estimates are reported as posterior means along with 95% CIs.

RESULTS

We obtained 12,923 photos of fawns from January 1 to October 1 of each year (Table 2). We captured images of both sides of the fawn for >90% of individual fawns. We removed 208 (<2%) unusable photos due to low photo quality (N = 107) or because they contained a detection of a single-sided individual (N = 101) that was not retained (either all-left-sides or all-right-sides in each site and year) after the matching and identification process described in the methods. This resulted in a final data set of 12,715 fawn photos.

TABLE 2 Camera trap data collected from two seasons in south Florida.

Camera data details	2015	2016	Total
Trap days operational (%)	44,364 of 49,320 (90.0)	48,559 of 49,500 (98.1)	92,923 of 98,864 (94.0)
Fawn photos collected	7044	5879	12,923
Fawn photos retained (%)	6914 (98.2)	5801 (98.7)	12,715 (98.4)
Individuals identified	147	124	271
Cameras that detected fawns	137 of 180	130 of 180
Independent detections	1097	1007	2104
Max detections for one fawn	37	41
Mean (range) independent detections/fawn	7.55 (1–37)	8.12 (1–41)
Mean (range) no. traps where each fawn was detected	1.64 (1–5)	1.65 (1–5)
No. photos with more than one fawn	0	8
Observer-estimated birth window	27 (8–32) days	23 (4–34) days

Note: All images of spotted fawns collected on camera traps were identified to a unique ID and used to create independent detection histories. Values given within parentheses are either percentages or ranges, and this is denoted in the details of each variable name.

We identified 271 unique fawns, 147 in 2015 and 124 in 2016. The average lengths of the observer-assigned birth windows were 27 days in 2015 and 23 days in 2016 (Table 2). Model-predicted peak birth dates were January 31 in 2015 (95% CI: ±3.3 days) and January 28 in 2016 (±3.8 days; Appendix S1: Figures S1 and S2). There were 305 (95% CI: 245–385) fawns born in 2015 and 278 (212–381) fawns born in 2016, and fawn abundance peaked in mid-February (Figure 2; Appendix S1: Tables S1 and S2). The density of birth site locations was highest in hammock and lowest in cypress in 2015, but did not differ by habitat type in 2016 (Figures 3 and 4; Appendix S1: Tables S1 and S2). More fawns were born in the FPNWR study site than in either of the BCNP study sites (Appendix S1: Figures S3 and S4).

FIGURE 2. The number and density of fawns and recruits alive throughout the 10,941-ha study area in south Florida during 2015 and 2016. Recruits are fawns that survived to ≥180 days.

FIGURE 3. Estimated density of fawn birth locations in four vegetation cover types in 2015 and 2016. Estimates are posterior means and 95% CIs.

FIGURE 4. Estimated density surface of fawn birth locations in fawns per square kilometer throughout the 10,941-ha study area, which consists of the North Addition Lands, Bear Island, and Florida Panther National Wildlife Refuge (Florida Panther NWR) in south Florida. Data are from 147 fawns detected on three camera arrays between January 1 and October 1, 2015. Crosses indicate the locations of 180 trail cameras, with 60 cameras in each array. The estimated density of birth locations did not vary by habitat in 2016, so we only show the density surface for 2015 here.

Fawn survival differed between the 2 years of our study. In 2015 and 2016, fawn survival to 30 days was 81% and 69%, respectively (Figures 2 and 5). Of the 305 fawns born in 2015, 110 (36%) survived to 180 days, but only 36 of 278 fawns (13%) survived to 180 days in 2016 (Figures 2 and 5). There was little evidence from our model of an effect of age or habitat type on daily survival rates for either year (Appendix S1: Tables S1 and S2).

FIGURE 5. Fawn survivorship curves from camera data on 271 uniquely identified individuals from January 1 to October 1, 2015 and 2016. Estimates are posterior means and 95% CIs.

The detection submodel indicated that none of the camera-level covariates affected fawn detection rates in 2015. However, in 2016, fawn detection rate decreased with increasing human activity and increased with increasing surface water depth (Figure 6; Appendix S1: Tables S1 and S2). The fawn detection rate was higher on-trail than off-trail in 2016, but there was no association with trails in 2015 (Appendix S1: Tables S1 and S2).

FIGURE 6. The effects of two continuous variables on fawn daily detection rate in south Florida between January 1 and October 1, 2016. Estimates are posterior means and 95% CIs.

Daily surface water levels at marsh gauges from January 1 to April 1, 2015, around the peak birth window, averaged 33.7 cm (range 7–47 cm; Figure 7). However, daily surface water levels reached a mean of 71.1 cm (range 60–85 cm) between January 1 and April 1, 2016, far exceeding the level found to put fawns at risk (Figure 7).

FIGURE 7. Surface water depth during the fawning season in 2015 and 2016. The gray polygon indicates the depth of surface water that has been previously shown to depress fawn recruitment.

DISCUSSION

Understanding the factors influencing recruitment is important, but notoriously difficult, for species with variable juvenile survival (DeYoung, 2011). We applied and expanded a recently developed noninvasive SCR method to the study of white-tailed deer fawn survival and recruitment in south Florida. Our study represents the first multi-year estimation of fawn survival and recruitment across the Big Cypress Basin since the 1990s (Land, 1991) and the first published application of combining noninvasive methods and Bayesian SCR models to estimate multi-year fawn survival.

Although we found little evidence of spatial variation in birth locations or survival, we observed large interannual variation in fawn survival, where survival to recruitment in 2016 was less than half of survival to recruitment in 2015. This large difference may have been the result of an extremely wet fawn-rearing season in 2016. Starting with the last few decades of human alteration of natural flooding cycles through draining wetlands, rerouting floodwaters, and establishing canals (Duever et al., 1986; Light & Dineen, 1994), an increasingly intense cycle of floods and droughts in the Big Cypress Basin has impacted deer survival and movement (Garrison et al., 2011; MacDonald-Beyers & Labisky, 2005). Additionally, ongoing climate change is exacerbating the intensity of these flooding and drought cycles and the frequency of extreme climatic events, which have significant influences on individual deer movement patterns (Abernathy et al., 2019). Previous research documented a 10-fold decrease in apparent fawn recruitment during a year of abnormally high rainfall (MacDonald-Beyers & Labisky, 2005). Water depths greater than 45–50 cm have been linked to reduced fawn recruitment and reduced adult survival, movement, and productivity (Fleming et al., 1994; Garrison et al., 2011; MacDonald-Beyers & Labisky, 2005). In our study, daily surface water depth during the spring fawning season was twice as deep in 2016 as it was in 2015, and surface water depth in 2016 averaged over 70 cm daily. This extremely wet spring in 2016, during which ~75% of the year's fawns were born (Appendix S1: Figures S1 and S2), may have impacted fawn survival through direct mechanisms, such as drowning, or indirect mechanisms, such as inadequate access to forage or increased susceptibility to predation. The link between increased surface water and decreased fawn survival was also suggested in a previous multiyear study in the Bear Island unit of Big Cypress National Preserve (Land, 1991).

Even in the absence of extreme weather events, fawn production is lower in south Florida than in many other areas of the white-tailed deer's range and is attributed to low-quality nutrition (Duever et al., 1986; Harlow & Jones, 1965). Fecundity in south Florida deer is 1.2 fetuses/pregnant female compared with 1.5–2 fetuses/pregnant female in other parts of the southeastern United States (Fleming et al., 1994; Jones et al., 2010; McCown, 1991; Richter & Labisky, 1985). Using that fecundity rate, our survival estimate of 36% in 2015 would equate to an expected fawn recruitment rate of 43.2% (1.2 × 36%), which may be high enough to replace adult losses in the population (Chitwood, Lashley, Kilgo, Moorman, et al., 2015; Peters et al., 2020). However, our estimate of 13% fawn survival in 2016 translates to a recruitment rate of 15.5%, which is unsustainable for population persistence in the presence of high adult mortality (Bled et al., 2022).

Fawn survival to 30 days was 81% in 2015 and 69% in 2016, and we found little influence of age on the daily survival rate, unlike in other fawn survival studies. Most fawn studies in the southeastern United States show a steep initial decline in survival, where only one third of fawns survive their first 30–60 days (Nelson et al., 2015; Shuman et al., 2017). This high mortality rate then sharply decreases as fawns reach a size that reduces their risk of predation, and daily survival rate for remaining individuals is high. However, we conversely detected an initially high survival rate, which steadily declined over time, with 57% and 33% of fawns surviving to 120 days and, ultimately, 36% and 13% of fawns surviving to 180 days, in 2015 and 2016, respectively (Figure 5).

The gradual but continual decline in fawn survival in our study may be driven by the unique structure of the predator community in south Florida. Coyotes and black bears are often implicated in high mortality rates of newborn fawns in the southeastern United States (Chitwood, Lashley, Kilgo, & Pollock, 2015; Nelson et al., 2015; Shuman et al., 2017; Thornton et al., 2004). However, coyotes are at low densities in south Florida (D. Shindle, USFWS, personal communication). Similarly, bear density in Big Cypress is lower (at 0.132 bears/km²; Humm et al., 2017) than in some areas where bears (at extremely high densities of 0.66 bears/km²) have been found to depredate a substantial portion of fawns (Hooker, 2010; Shuman et al., 2017). Black bears are not effective at catching fawns greater than three weeks old, and most south Florida fawns are already 2–4 months old when bears become active after winter dormancy. Additionally, data indicate that black bears in south Florida rarely prey on deer (Maehr, 1997). This low predation pressure from coyotes and bears on newborn fawns likely shapes the initially high survival we observed within their first 30 days of life. Bobcats and Florida panthers are effective in predating both adult and juvenile deer (Garrison & Gedir, 2006; Maehr, 1997; Maehr et al., 1990; Nelson et al., 2015; Shuman et al., 2017), but prior data indicate that deer make up less than 5% of bobcat diet in our study site (Maehr, 1997). Alligators and Burmese pythons may also prey on both adults and fawns (Boback et al., 2016; Land, 1991; MacDonald-Beyers & Labisky, 2005), especially in flooded habitats, although Burmese pythons had not yet expanded into our site during the study. Thus, predation by Florida panthers may be the dominant form of fawn mortality during the three- to six-month age range, when many other fawn studies report a marked decrease in mortality and release from predation.

The contrast between our study's gradual survivorship curve and the sharp initial decline seen in the survivorship curves of other fawn studies may also be influenced by sampling methodology. Although research indicates there is low risk of marking-induced abandonment (Powell et al., 2005), there may be nonabandonment effects from radio-instrumentation of neonate deer, such as reduced fitness, weight loss, collar injury, or higher predation risk due to higher visibility or inability to flee effectively. This may cause the neonate mortality rate in fawn-collaring studies to be overestimated at young ages. Conversely, we may have been unable to sufficiently sample fawns within the first week of life, when mobility is limited and mortality is high. Although we did detect many fawns within their first few weeks, we likely failed to detect some individuals that died within a few days of birth. If many fawns died before they were detected by cameras, our estimates of the frequency and distribution of births could be biased, but our estimates of recruitment (i.e., number of fawns recruited to the juvenile [i.e., >6 months old] age class) would not be. Simultaneous use of SCR techniques with a traditional fawn-collaring study would clarify if the differences we observed in fawn survival are real or were influenced by study design.

The nutritional quality of forage selected by pregnant and lactating females directly impacts fawn growth and survival (Tollefson et al., 2011), and the selection of either concealment cover or open sightlines during fawning and rearing prioritizes neonatal protection from either cursorial predators or ambush predators (Cherry et al., 2017, 2018). However, we found no evidence that fawn daily survival probability varied among vegetation types, although previous research found that selection of vegetation types differed between adult females that lived and died during the fawning season (Abernathy et al., 2022).

Forage quality and availability of concealment cover also likely drive patterns of fawn birth locations, although the density of birth locations varied little among vegetation types in 2016. In 2015, the birth density was highest in hardwood hammocks, which provide dense cover, and lowest in cypress, which is more frequently inundated with surface water and provides less concealment cover for fawns. Pine flatwoods often have dense cabbage palm (Sabal palmetto) and saw palmetto (Serenoa repens) understories, which can provide both great hiding cover for young fawns and good forage due to the influence of frequent fire on nutritional composition of vegetation in this community. The open sightlines in open canopy habitats may offer reduced fawn predation risk as fawns age and become more mobile. However, the scale at which we examined these habitat covariates may not have been well-suited to identify the effects of the nutritional or concealment benefits that habitat can offer to does and fawns.

Deer in heavily hunted regions alter their movement patterns to avoid interaction with roads and humans (Bonnot et al., 2013; Crawford et al., 2019). The fawn detection rate in our study decreased with increasing human activity, suggesting that adult females and fawns avoided human disturbance (Figure 6). However, in 2016, more fawns were detected on-trail than off-trail. Although these two results may seem contradictory, roads can provide a human shield of reduced predation risk (Berger, 2007), as well as provide easier travel corridors across flooded habitats. Does with fawns often alter their diel activity patterns as an antipredator defense (Crawford et al., 2021), and in our study area, adult females increased their activity on roads and trails during diurnal periods when panthers were less active (Crawford et al., 2019). Together, these findings indicate that adult females and their fawns in south Florida alter their spatiotemporal use of roads and human-dominated areas to avoid the high-risk times for both panther predation and human activity (Bonnot et al., 2013; Crawford et al., 2019). Further, the increased activity of fawns on trails in 2016 was likely driven by use of drier elevated roadways during flood conditions.

Using this noninvasive camera methodology, we were able to monitor fawn survival and recruitment across a large spatial region. Collaring neonates typically requires first capturing pregnant females to deploy radio collars and vaginal implant transmitters, followed by intensive field efforts to capture fawns at the birth site (Chitwood, Lashley, Kilgo, & Pollock, 2015; Shuman et al., 2017). However, prior attempts at capturing and collaring neonates in this system were unsuccessful due to limited road access and low deer density (Land, 1991). Since fawning in south Florida ranges over 6 months (Appendix S1: Figures S1 and S2) and much of our study site consisted of wetlands with minimal road access, it would have been extremely difficult to locate and collar a sufficient number of neonate fawns for this study. Therefore, our camera-based methodology allowed us to estimate fawn survival and recruitment in a system that would be prohibitively challenging and costly for traditional fawn monitoring methods. To further evaluate and refine this camera-based methodology, future studies should estimate recruitment with our model and compare its efficacy to traditional fawn survival estimation through a concurrent fawn-collaring study.

Our observed fawn survival and recruitment estimates suggest that extreme weather events can cause strong temporal variation in demography. Consistently low fawn survival over multiple years, when paired with low adult survival, may be a cause for concern for the viability of deer in the Big Cypress Basin (Bled et al., 2022). Additional sources of fawn mortality, such as the expansion of Burmese pythons into the northern Big Cypress Basin and ongoing hydrologic shifts driven by climate change, may continue to reduce neonate survival and influence other demographic parameters in this system. Our research provides critical data to fill a previous gap in our understanding of fawn survival in south Florida's complex ecological system. The complexities of this system necessitate future population viability analyses and informed management decisions that account for the annual variation in population dynamic parameters driven by ecological fluctuations.

AUTHOR CONTRIBUTIONS

Michael J. Cherry, L. Mike Conner, Elina P. Garrison, Karl V. Miller, and Richard B. Chandler conceived the ideas and designed methodology; Kristin N. Engebretsen collected the data; Kristin N. Engebretsen and Richard B. Chandler analyzed the data; Kristin N. Engebretsen and Richard B. Chandler led the writing of the manuscript. All authors contributed critically to the drafts and gave final approval for publication.

ACKNOWLEDGMENTS

Funding and technical support for this research were provided by FWC, UGA, USFWS, and NPS. The authors thank D. Crawford, B. Kelly, L. Margenau, W. Gurley, C. Kupar, M. H. Reuter, G. Aubin, M. Jorge, K. Piecora, H. Ellsworth, and H. Abernathy for collection of field data and A. Beers for helpful feedback on the manuscript.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

DATA AVAILABILITY STATEMENT

Data and novel code (Engebretsen & Chandler, 2023) are available from Dryad: https://doi.org/10.5061/dryad.3ffbg79n1.