Ethics statement

No permits were required of our research team to access sites on the publicly-owned, federally-managed property where the study occurred, as the study was collaboratively conducted with the land manager. Field studies did not involve any state- or federally-protected species.

Study system and species

U.S. Army Garrison Fort Bragg (35°8’21"N, 78°59’57"W) spans approximately 73,468 ha in the Sand Hills ecoregion (hereafter Sandhills) of south-central North Carolina [44]. The Sandhills are a matrix of xeric uplands and mesic/hydric lowlands (i.e. wetlands). The xeric uplands have a savanna physiognomy that is maintained by fire and dominated by a longleaf pine (Pinus palustris Mill.) canopy [45]. Growing season (April–September) prescribed fires are scheduled on a 3-year rotation to approximate the mean historic fire return interval [46]. While the understory of the uplands typically burns during prescribed fires, fires penetrate wetlands much less frequently [37, 47, 48]. Fire kills aboveground herbaceous and shrub biomass and significantly reduces the size and fecundity of woody resprouts for the first several years post-fire [49–51].

The majority of L. subcoriacea populations within North Carolina are known from Fort Bragg and have been observed in four wetland vegetation communities [32, 47, 52]: Streamhead Pocosins, Sandhill Streamhead Swamps, Streamhead Atlantic White Cedar Forests, and Sandhill Seeps. The four communities differ in fire frequency, hydrology, canopy cover, understory cover, and other environmental factors [32]. Streamhead Pocosins have a relatively open canopy dominated by Pinus serotina Michaux and a dense understory dominated by a variety of evergreen shrub species (Ilex coriacea (Pursh) Chapman and I. glabra (L.) Gray are the dominant fleshy-fruited species). Sandhill Streamhead Swamps have a canopy composed of Nyssa biflora Walter, Acer rubrum L., and Liriodendron tulipifera L., with an understory of evergreen shrub species. In general, the canopy of Sandhill Streamhead Swamps is more closed and the understory less dense, relative to Streamhead Pocosins. Streamhead Atlantic White Cedar Forests have the highest canopy cover of the four vegetation communities, with the canopy containing at least 50% Atlantic white cedar (Chamaecyparis thyoides L.) coverage. Finally, Sandhill Seeps typically occur midslope where erosion has exposed the clay subsoils, leading to semipermanently saturated conditions at the surface. Sandhill Seeps are generally smaller in area, have a relatively open overstory and midstory, and support a larger herbaceous component relative to the other three vegetation communities. Sandhill Streamhead Swamps and Streamhead Atlantic White Cedar Forests have low relative fire frequency, Streamhead Pocosins have medium relative fire frequency, and Sandhill Seeps have high fire frequency [32].

The single-seeded drupes of L. subcoriacea mature simultaneously in late July and early August (S1 Fig), and although not documented in the species, are most likely avian-dispersed based on fruit traits [53–55] and observations for congeners [34, 56]. Although L. subcoriacea has fruits/seeds of similar size to L. benzoin (ovate seeds are 7.02 ± 0.29 mm, n = 101 [37] and 7.02 ± 0.11 mm, n = 50 [57] long, respectively), L. subcoriacea individuals produce fewer fruits than L. benzoin (80 ± 178, n = 290 [37] and 164 ± 104, n = 11 [58], respectively). Fruit production in both species likely varies as a function of individual size and habitat conditions that affect productivity (e.g., light, water and nutrient availability) [59], but for L. subcoriacea habitat effects are obscured by high variability in individual fruit production [37].

Potential seed dispersers and predators

In a separate effort, we developed post-breeding avian occupancy models for Fort Bragg based on survey data collected during mid-August 2017 and 2018 [60] and coinciding with our seed dispersal and predation studies (next section). Occupancy models accounted for imperfect detection [61], as implemented in the R (R Development Core Team, 2021) package unmarked using the double-observer approach [62]. We documented 22 seasonally frugivorous species (as described by numerous publications; e.g.,[33, 58]) during the survey, and generated spatial occupancy estimates at a 30 x 30 m resolution for 20 of these species. Many of these species are expected to consume L. subcoriacea fruits and disperse seeds, given that bill gape widths exceed the size of L. subcoriacea fruits/seeds. One exception is the Northern Cardinal (Cardinalis cardinalis L.), which is primarily granivorous during the non-breeding season [63]. For five avian species documented visiting fruiting L. subcoriacea (see next section), we compared potential differences in the magnitude of their roles as seed dispersers or predators by calculating their mean occupancy at 88 georeferenced female plants across 69 L. subcoriacea populations [32, 37]. We assessed differences in these mean occupancy estimates among species pairs with bootstrapped confidence intervals. In addition to the Northern Cardinal, as many as ten different rodent species are likely to predate L. subcoriacea seeds in our study system post-dispersal [e.g., 64, 65].

Estimating seed dispersal and pre-dispersal predation with fruit counts and seed traps

Over three years (2017–2019), we used seed traps to estimate the number and proportion of L. subcoriacea fruits that were dispersed, fell from the maternal plant, or predated pre-dispersal. We constructed seed traps having two 1 x 1 x 0.08 m wooden interlocking frames, elevated on four 1 m wooden legs. We covered the upper frame with galvanized hardware cloth (1.26 cm² mesh) to allow fallen fruits to pass and be intercepted by stainless steel mesh window screen covering the lower frame. This design ensured that fallen fruits could not be removed by animals or precipitation. We placed a single seed trap beneath 26 of 88 (29.5%) annually monitored female individuals [37]. We chose individuals and subjectively positioned traps based on the efficacy of trap placement, which was influenced by shrub branching structure (horizontal reach ≥ 1 m), approximately level topography, and number of fruits. Traps encompassed approximately 15–25% of the individual canopies. We positioned traps in mid-July of each year, before L. subcoriacea fruits begin to ripen (S1 Fig). While deploying the traps, we counted the number of ripe and unripe fruits on each individual (107.7 ± 231.9; 12–1500) and the subset directly above the trap (57.9 ± 98.6; 4–624) that would be intercepted if they were to fall from the plant instead of being removed by a disperser or seed predator. We revisited individuals every 3–4 days to 1) recount the remaining ripe and unripe fruits on individuals and above the seed traps; 2) recover and count whole fallen fruits, regurgitated whole seeds, and remnants of predated fruits and seeds intercepted by the traps; and 3) record fecal evidence of rodent visitors. We revisited seed traps for 30 days, or until no fruits remained on the shrub, whichever came first.

For each individual plant (n = 26; 42 total observations across 3 years), we estimated 1) dispersal as the number and proportion of fruits counted above the seed trap that were removed, but neither recovered as whole fruits, regurgitated whole seeds, nor as fruit/seed fragments in the trap; 2) pre-dispersal predation as the number and proportion of fruits counted above the seed trap that were found as fruit/seed fragments in the trap; and 3) dispersal failure as the number and proportion of fruits counted above the seed trap that were found as whole fruits, or regurgitated whole seeds within the trap. We acknowledge that predation of fruits/seeds counted above seed traps could have taken place elsewhere, resulting in underestimation. Additionally, it is possible that fruits/seeds removed from neighboring plants may have been deposited in the seed traps beneath our focal plants, which would cause pre-dispersal predation and dispersal failure to be overestimated.

Each year we also placed ≥ 1 Reconyx HyperFire PC800 infrared camera (Holmen, WI, USA) at a subset of the L. subcoriacea individuals where seed traps were positioned and at several additional fruiting individuals without seed traps to collect observations of diurnal and nocturnal endothermic visitors that may function as seed predators or dispersers. The dense vegetation within communities occupied by L. subcoriacea, particularly Streamhead Pocosins where most populations occur, severely limits observations of frugivores and seed predators. At distances ≥ 2 m, it typically becomes impossible to clearly see individual L. subcoriacea. Therefore, we chose individuals based on our ability to position the camera(s) to have an unobstructed line(s) of sight on fruit-bearing branches. We mounted cameras at 1.5 m above the ground on tripods located ≤ 2 m from target individuals, set them on motion detection photographic mode, and left them in place for the entire period during which fruit counts were collected and seed traps were monitored. Cameras were positioned for a total of 264,000 (25 cameras on 20 individuals), 63,360 (12 cameras on 10 individuals), and 25,344 hrs (8 cameras on 6 individuals), in 2017, 2018, and 2019, respectively. We visually reviewed images recorded by the cameras for the presence of seed predators and frugivores, documenting species identity when possible. Observations were only used to identify visitors at fruiting individuals and not used in any quantitative analyses.

Estimating post-dispersal seed predation with seed depots

In 2017 and 2019, we conducted 10-day seed removal experiments from August 1–14, coinciding with L. subcoriacea seed dispersal (S1 Fig). Using ArcGIS (ArcPro v. 2.2, ESRI, Redlands, CA, USA), we generated 5,000 random points within Streamhead Pocosins, Sandhill Streamhead Swamps, and Sandhill Seeps vegetation communities. Streamhead Atlantic White Cedar Forests were not included due to their rarity and limited distribution on Fort Bragg. We delineated the locations of the vegetation communities using the best available information, which was a map of historical (before 1750) vegetation [66]. Intensive use of prescribed fire and hardwood suppression on Fort Bragg during the past 30 years has resulted in a landscape that more closely resembles the pyroclimax vegetation states characterized in this historical map than the 20th century fire suppressed state. We organized a subset of the random points, which were separated by > 1 km, into routes (two in 2017; four in 2019) composed of approximately 15 points each that could be visited within a single day. Routes were located in the western and northwestern portion of the installation, where most L. subcoriacea populations occur. Although the location of route points was approximated by the location of the random points, the actual location was haphazardly chosen at the time of depot deployment to match one of the three vegetation communities. Points were located 20–50 m from the firebreaks (single lane trails) used to traverse the routes.

We installed a pair of seed removal depots separated by 1 m at 21 and 60 points during mid-July 2017 and 2019, respectively. We used two types of depots: 1) closed, allowing invertebrate seed predators access; and 2) open, allowing both invertebrate and small mammalian seed predators access. We constructed depots from inverted translucent plastic buckets (12.7 cm in height and 21.9 cm in diameter) with two 8 cm x 15 cm openings cut into opposite sides. We covered the openings of closed depots with galvanized hardware cloth (1.26 cm² mesh) and left the openings of open depots unobstructed. We affixed mesh window screen to the inverted opening of both depot types to contain seeds, while preventing rainfall accumulation. Similarly designed seed depots have been successfully used in other seed predation studies conducted in longleaf pine ecosystems [e.g., 64, 67]. We deployed pairs of empty open and closed depots at points 10–14 days prior to seed presentation. On 4 successive days we travelled each of the routes and loaded depots at each point with 10 L. benzoin seeds that had been purchased from Sheffield’s Seed Company (Locke, NY, USA), vacuum sealed in a plastic bag, and heated to 100°C in a water bath for 10 minutes to kill the embryos. We used the seeds of L. benzoin due to a lack of L. subcoriacea seeds and assumed that the seed removal observed for this surrogate congener would be the same as for L. subcoriacea. The two species were only recently taxonomically separated and have similar sized seeds [68]. Each year several individual depots (6 in 2017; 11 in 2019) or depot pairs were compromised by fires, flooding, and animal damage, and removed from the dataset.

Intrinsic and extrinsic drivers

We estimated individual height, individual crop size, and conspecific fruiting neighborhood using demographic data collected from females during 2017–2019 [37]. We calculated fruiting neighborhood by summing the individual crop sizes of all fruiting L. subcoriacea within 5, 10, and 30 m radius nested buffers of each seed trap. We did not include any other fruiting species occurring within buffers in the neighborhood estimates. Although the heterospecific fruiting neighborhood is also known have positive and negative effects on dispersal [23–25], difficulty moving through the dense shrubs in L. subcoriacea habitats precluded us from making accurate counts of fruits.

We estimated time since the last fire (TSLF) based on a previously-developed methodology for estimating fire occurrence [51]. Briefly, we used Landsat satellite imagery (30 x 30 m resolution) and several imagery indices, coupled with a Random Forest classifier, to identify areas that had burned on Fort Bragg. We classified all pixels from Landsat images from 1991–2019, as either burned, or not on an annual basis. We then estimated the mean TSLF within a 30 m radius buffer around each seed trap and depot location.

We estimated canopy and understory cover using Airborne Light Detection and Ranging (LiDAR) data acquired from several flights flown December 20–27, 2012. We first converted the LiDAR dataset into a 25 m² raster image. We then summarized the number of aboveground (AG) and bare earth (BE) returns points within each cell and estimated canopy cover using the formula . We then classified the AG return points by height (< 0.25 m = ground, 0.25–5 m = low vegetation) and calculated understory cover using the formula . Finally, we calculated the mean canopy and understory cover (proportion) within a 30 m radius of each seed trap and depot location. Although prescribed fires may have temporarily reduced understory cover during the 5–7 years between LiDAR data collection and our study, both the upland herbaceous vegetation (≤ 1 growing season) and the woody wetland vegetation (≥ 3 years) rapidly recovers to pre-burn size [49]. Across all three years of our study mean TSLF was ≥ 3 years for 90.5% of seed trap and 84% of depot locations. The majority of these seed trap (75%) and depot (59%) locations where TSLF was < 3 years were in Sandhill Seeps, which are generally small and isolated within the upland savanna matrix.

In 2019, we collected substrate cover data at each depot pair within two 1 m² microplots centered on the depots and oriented to prevent spatial overlap. We collected data on the percentage cover of litter, fine woody debris (< 10 cm diameter), coarse woody debris (> 10 cm diameter), bare ground, herbaceous vegetation cover (< 1 m), and woody vegetation cover (< 1 m) observed in each microplot and recorded the mean of the paired plots.

Data analyses

For our analyses, we used estimates of the number and proportion of seeds dispersed and predated that were based on the initial and final counts. We removed one individual with more than 400 fruits from the neighborhood estimates because of its outlier status. We explored the drivers for correlations and found minimal evidence of collinearity (|r| > 0.70 [69]; S1 and S2 Tables). We fit univariate, mixed-effects generalized linear regression models to investigate the effects of each intrinsic and extrinsic driver on the number and proportion of seeds dispersed and predated pre- and post-dispersal using the lme4 package [70]. Seed trap and depot IDs were included as a random effect and habitat as a fixed effect. For proportion data, we used a binomial error structure with a logit link. For count data, we used a Poisson error structure with a log link with a random intercept. We included a quadratic term within models to account for possible non-linear effects of drivers. Full models included the driver and the quadratic term. We evaluated the effect of the potential drivers on model explanatory power by evaluating the full model against the reduced (without quadratic) model using a type II likelihood ratio χ² test (base::anova; [71]) [72, 73]. If there was no difference in explanatory power between the full and reduced model, we did not include the quadratic term. As needed, we evaluated the explanatory power between the reduced model and null (intercept only) model using the same procedure. We performed all statistical analyses using the statistical platform R version 4.1 [74], with the script used to analyze the data available from the authors. Our threshold of statistical significance was α = 0.05. For all significant models, we calculated a pseudo R² [75].

Pre-dispersal seed predation and observed seed predators

Across all years and individuals the mean proportion of pre-dispersal seed predation was 0.28 ± 0.06 SD. Most individuals experienced low pre-dispersal seed predation, but a few experienced high predation (range = 0.0–1.0) (S2 Fig). There was a negative quadratic effect of understory cover on the proportion of seeds predated ( = 4.79, P = 0.03), with individuals in low and high understory having relatively low seed predation compared to those in intermediate understory (Fig 1). No other intrinsic or extrinsic drivers had a significant effect on the number or proportion of seeds predated pre-dispersal (Table 2). Data and additional details about regression results can be found in S1 and S2 Files.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Relationship between pre-dispersal predation of Lindera subcoriacea seeds (proportion) and understory cover (scaled and centered).

Points are individual observations, the line is from a mixed-effects generalized linear regression model with a quadratic effect (full model), and shaded region is the 95% confidence interval.

https://doi.org/10.1371/journal.pone.0283810.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Results testing for the significance of intrinsic and extrinsic drivers on treatment (pre-dispersal predation) and deposition (post-dispersal predation) subcomponents of seed dispersal effectiveness quality (number and proportion).

Model comparisons evaluated with type II likelihood ratio χ² tests are identified by letter codes after each driver for the number and proportion of seeds, respectively. P values in bold indicate a significant effect (α = 0.05).

https://doi.org/10.1371/journal.pone.0283810.t002

Camera traps recorded one granivorous avian species at fruiting L. subcoriacea, the Northern Cardinal [63]. Within the grid cells occupied by 88 L. subcoriacea individuals across 69 populations on Fort Bragg the mean occupancy of the Northern Cardinal was 0.42 ± 0.11 [60]. Cameras also documented small nocturnal mammalian granivores likely in the genus Peromyscus within the canopies of fruiting L. subcoriacea. Fecal evidence of rodent visitors was observed in 45% of the seed traps where pre-dispersal predation was documented over the three years. In contrast, fecal evidence of rodent visitors was only observed in two seed traps where no pre-dispersal predation was documented.

Seed dispersal and observed seed dispersers

Across all years and individuals, the mean proportion of dispersed seeds was 0.69 ± 0.25. Of the seeds that were not predated pre-dispersal, a high mean proportion (0.96 ± 0.03) were estimated to be dispersed (S3 Fig), with the rest falling below maternal plants. A positive linear effect of individual crop size ( = 97.7, P < 0.001) and a negative quadratic effect of height ( = 32.2, P < 0.001) were documented for the number (Figs 2 and 3), but not the proportion of seeds dispersed (Table 3). There also was a positive linear effect of TSLF ( = 28.8, P < 0.001) on the number of seeds dispersed, and a positive quadratic effect of TSLF ( = 13.96, P < 0.001) on the proportion of seeds dispersed. In addition, there were positive linear effects of fruiting neighborhood on the number and proportion of seeds dispersed; effects were observed at the 5 and 10 m scales for the number of seeds dispersed and at all three scales for the proportion of seeds dispersed (Table 3; Fig 4). No other intrinsic or extrinsic drivers had a significant effect on the number or proportion of seeds dispersed (Table 3).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Relationship between the number of Lindera subcoriacea seeds dispersed and individual crop size.

Points are individual observations, the line is from a mixed-effects generalized linear regression model (no quadratic term; reduced model), and the shaded region is the 95% confidence interval.

https://doi.org/10.1371/journal.pone.0283810.g002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Relationship between the number of Lindera subcoriacea seeds dispersed and individual height (scaled).

Points are individual observations, the line is from a mixed-effects generalized linear regression model with a quadratic effect (full model), and the shaded region is the 95% confidence interval.

https://doi.org/10.1371/journal.pone.0283810.g003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4.

Relationship between proportion of Lindera subcoriacea seeds dispersed and fruiting neighborhoods (scaled and centered) within (A) 5 m, (B) 10 m, and (C) 30 m radius nested buffers. Points are individual observations, the lines are from mixed-effects generalized linear regression models (reduced models), and the shaded regions are 95% confidence intervals.

https://doi.org/10.1371/journal.pone.0283810.g004

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Results testing for the significance of intrinsic and extrinsic drivers on the quantity component (number and proportion) of seed dispersal effectiveness.

Model comparisons evaluated with type II likelihood ratio χ² tests are identified by letter codes after each driver for the number and proportion of seeds, respectively. P values in bold indicate a significant effect (α = 0.05).

https://doi.org/10.1371/journal.pone.0283810.t003

Frugivores documented with camera traps at fruiting L. subcoriacea included the American Robin (Turdus migratorius L.), Grey Catbird (Dumetella carolinensis L.), Red-headed Woodpecker (Melanerpes erythrocephalus L.), and White-eyed Vireo (Vireo griseus Boddaert). Given that these four species all have bill gape widths larger than the fruits/seeds of L. subcoriacea, they are interpreted to be dispersers. The mean occupancies estimated for the American Robin, Red-headed Woodpecker, and White-eyed Vireo within L. subcoriacea populations across Fort Bragg were 0.35 ± 0.14, 0.23 ± 0.10, and 0.13 ± 0.05, respectively. Bootstrapped confidence intervals for the differences in occupancy estimates among species pairs indicated that frugivore occupancies differed, and were lower than Northern Cardinal occupancy (S3 Table). We could not include Grey Catbird in this comparison because insufficient observations (< 10) during avian surveys prevented us from generating an occupancy map for this species.

Post-dispersal seed predation

For the seed depot data, there was a difference between the closed and open depots ( = 116, P < 0.001), with open depots having a greater mean proportion of seeds removed (0.65 ± 0.42 vs 0.18 ± 0.28). We removed the closed depots from all further analyses, given the apparent dominant role of rodent granivores in post-dispersal predation. There was a significant difference between a model that included fine woody debris and the null model ( = 3.91, P = 0.048); with higher proportions of seeds predated where the percentage in the substrate was high. No other extrinsic drivers had a significant effect on the proportion of seeds predated collectively by rodents and invertebrates post-dispersal (Table 2). Data and additional details about regression results can be found in S2 and S3 Files.

Pre-dispersal predation

Studies of pre-dispersal seed predation are relatively uncommon compared to studies of post-dispersal predation, and typically focus on insect rather than vertebrate seed predators [e.g., 78]. However, Bell and Clark [79] collected seeds in seed traps in North Carolina over a ~15-year period, examined them for vertebrate seed predation, and documented 18% and 51% pre-dispersal predation of Cornus florida L. and Nyssa sylvatica Marshall seeds, respectively. Thus, the 28% of L. subcoriacea seeds that are predated by vertebrates prior to dispersal is within the range documented for these two other drupe-bearing, large-seeded species in the southeastern USA.

Both avian and mammalian seed predators may be attracted to the seeds of Lindera spp., which are known to have high lipid and crude protein content [33, 80–82]. However, our methods did not allow us to unequivocally distinguish the magnitude of predation attributable to these two vertebrate guilds. We observed the Northern Cardinal and small arboreal rodents (Peromyscus spp.), both well-documented seed predators, at fruiting L. subcoriacea. Johnson et al. [83] reported that Northern Cardinals commonly crushed the fruits and seeds of L. benzoin and Smith et al. [56] similarly reported the species as a pre-dispersal seed predator of L. melissifolia. Given our observations and the high estimated mean occupancy of the Northern Cardinal in L. subcoriacea populations (0.42 ± 0.11) during August, it is likely that the species is also a L. subcoriacea seed predator.

Although pre-dispersal seed predation by small arboreal rodents has not been reported for other Lindera spp. in North America or Asia, our findings for L. subcoriacea are not surprising, but rather a heretofore underappreciated source of seed loss. Granivorous rodents are a common component of small mammal communities in many systems, including longleaf pine ecosystems [84]. In our study system Peromyscus spp. are the most abundant species of rodent granivore, representing 77–94% of captured individuals and reaching densities of ~50 individuals per 60 x 60 m plot in lowland hardwood habitats [65]. Peromyscus spp. exhibit strong preferences for nuts and the seeds of fleshy fruits cleaned of their pulp, such as those deposited post-dispersal, over intact fruits containing seeds [e.g., 85]. However, a positive relationship between Peromyscus spp. arboreal activity and density of fleshy fruit producing trees has also been documented in areas without nut producing species [86], which would be characteristic of the wetland habitats occupied by L. subcoriacea.

The abundance and behavior of both avian and mammalian seed predators are known to vary in response to diverse environmental factors that influence food availability, foraging efficiency, refuge availability, and real and perceived risks of predation [e.g., 87–89]. In the four vegetation communities occupied by L. subcoriacea, light availability and fire frequency are expected to differ [32, 47, 90] and to influence food- and cover-related resources utilized by seed predators [64, 91]. Although we did not identify differences in pre-dispersal seed predation among vegetation communities or in relation to fire, we did document a negative quadratic relationship with percentage understory cover. Seed predation is generally expected to be greater in areas having high vegetation cover [e.g., 38, 92, 93], but for L. subcoriacea the highest proportions of pre-dispersal seed predation occurred at intermediate levels of understory cover. Structural complexity can reduce foraging efficiency and is a potential explanation for the observed reduction in pre-dispersal seed predation at high understory cover [e.g., 88], whereas predation risk may limit the activity of seed predators under low understory cover [87].

Dispersal

If not predated pre-dispersal, the likelihood that L. subcoriacea seeds are dispersed appears to be comparable to dispersal estimates for L. benzoin. We estimated that a mean of 69% of seeds are dispersed away from maternal plants and only 3% of seeds fail to disperse. For L. benzoin, reported dispersal estimates range from 60–90% [33, 34, 94], however neither pre-dispersal predation nor dispersal failure were accounted for in these estimates. Despite rapid and high percentages of removal of fruits by presumed dispersers, L. benzoin, C. florida, and N. sylvatica are also known to suffer dispersal limitation and have small seed shadows [95–97]. Small seed shadows, in addition to source limitation, may also be important for L. subcoriacea, constraining the number of occupied sites and affecting inter-population dynamics [94].

Both plant height and crop size have been found to be intrinsic drivers of seed dispersal [14, 98]. The relationship between dispersal and plant height is not as well-studied as is that for crop size, and where investigated has exhibited variable results [12]. For L. subcoriacea, we documented a negative quadratic relationship between plant height and number of seeds dispersed. Previous work has documented a negative quadratic relationship between height and fruit crop size for the species [37], but no relationship between height and crop size was identified for the individuals included in this study (S2 File). Therefore, height had an effect on seed dispersal independent of any effect of individual crop size. For crop size, a recent meta-analysis found a consistent positive relationship with the number, but not the proportion of seeds dispersed [99]. Our findings align with those of the meta-analysis.

Conspecific and heterospecific fruiting neighborhoods can either negatively influence SDE quantity by generating competition for dispersers [e.g., 21–23], or positively influence dispersal quantity through facilitation [e.g., 24,25]. We documented a positive relationship between the number and proportion of seeds dispersed and L. subcoriacea fruiting neighborhood at nearly all scales examined, suggesting active selection by dispersers. A similar positive conspecific effect has been reported for L. benzoin, with avian frugivores foraging more frequently on clumped than isolated fruit displays [100]. Although conspecific fruiting neighborhood enhances L. subcoriacea SDE quantity, it may also potentially reduce SDE quality via reduced dispersal distances [e.g., 26]. Although we did not quantify the number or identity of other fruit-bearing species within L. subcoriacea fruiting neighborhoods, we do not think that the presence of these species had a large negative effect on L. subcoriacea seed dispersal, as we documented positive linear relationships between dispersal and both individual crop size and conspecific fruiting neighborhoods. Rather, as suggested by the diet complementation hypothesis [101, 102], it is possible that heterospecific fruit-bearing species within neighborhoods facilitated L. subcoriacea seed dispersal. This hypothesis proposes that negative frequency-dependent fruit selection for complementary nutrients [e.g., 103, 104] can increase the SDE of rare plants having fruits of relatively higher value compared to common species because their fruits will be consumed at proportionally higher rates [105]. The fruits of Lindera spp. are known to be a preferred, high quality, food resource for avian frugivores [33, 106].

As is the case for seed predators, the abundance and behavior of avian seed dispersers are also known to vary in response to diverse environmental factors [39, 107, 108]. Although we did not find any effect of vegetation community or understory cover on dispersal, Moore and Willson [34] documented differences in fruit removal rates between forest interior and gap habitats for L. benzoin, however the patterns reversed over the fruiting season. We did document a positive linear relationship between time since last fire (TSLF) and the number of seeds dispersed, and a positive quadratic relationship between TSLF and the proportion of seeds dispersed. Fire reduces cover and food resources for avian frugivores by top-killing the woody stems of fleshy-fruited shrubs and setting back recovery to pre-burn size and fruit production for at least 3–4 years in our study system [37, 49, 91].

All of the avian frugivores documented at fruiting L. subcoriacea individuals by the camera traps are known to consume L. benzoin and/or L. melissifolia fruits [34, 100, 109, 110], and are also likely to disperse L. subcoriacea seeds. These four species are known to swallow whole fruits and regurgitate large seeds [e.g., 83]. Bill gape widths are a limiting factor for consumption of fruits with this handling approach [13]. The bill gape widths of all four of these putative dispersers and the seed predator (i.e. Northern Cardinal) are > 10 mm and exceed the size of L. subcoriacea fruits [83, 111]. Although not documented by our camera traps, it seems likely that many seasonally frugivorous avian species occurring at our study site during August and having adequately large bill gape widths also consume L. subcoriacea fruits and disperse seeds.

Although observations of avian frugivore visitation to fruiting L. subcoriacea are interesting, knowledge of disperser identity provides no explicit information about disperser effectiveness. Nonetheless, some inferences about the likely relative dispersal effectiveness of the four species can be made. For SDE quantity, the number of fruits consumed is expected to increase with avian body mass due to positive relationships between body mass and basal metabolic rate and gut capacity [112]. Therefore, not accounting for any interspecific dietary differences, ordered low to high, White-eyed Vireo, Gray Catbird, Red-headed Woodpecker and American Robin individuals are expected to consume and disperse increasing numbers of fruits/seeds [113]. For SDE quality, we suspect that the Red-headed Woodpecker may be the least effective of the four species, as it commonly inhabits upland savanna communities that would be unsuitable for L. subcoriacea recruitment [114]. In contrast, the Gray Catbird and White-eyed Vireo primarily move within dense vegetation [114]. In our study system, densely vegetated areas are typically wetland communities that are potentially suitable sites for L. subcoriacea recruitment [37]. Although, the American Robin is commonly found in upland savannas during the breeding season, their post-breeding shift to a more frugivorous diet is expected to cause greater use of wetland habitats, where fleshy-fruited plant species are abundant, fruit is more likely to be available, and L. subcoriacea germination and recruitment is possible [47, 91, 115]. Wall et al. [60] estimated post-breeding occupancy maps for these avian species across our study site that align with these natural history-based interpretations of their differential habitat use.

Post-dispersal predation

When the levels of post-dispersal predation (65%) that we documented for surrogate L. benzoin seeds are combined with other L. subcoriacea seed losses, seedling recruitment and population growth rates are potentially limited. The proportion of L. benzoin seeds that are predated post-dispersal in our study system appears to be comparable to available estimates for the endangered L. melissifolia, which also has seeds of similar size and presumably composition [116]. Martins et al. [109] used video cameras to document vertebrate visitors at L. melissifolia seed plots. They documented Northern Cardinals, Peromyscus spp., and gray squirrels (Sciurus carolinensis Gmelin) consuming seeds within plots. Seed removal from their plots varied among sites and years, with the proportions and rates of removal ranging from 1.0 within 7 days to 0.0 over 68 days. They also recorded higher percentages and rates of seed removal in plots with high (50–100%), than low (0–25%) understory cover. For example, they estimated that the percentage of seeds removed ranged from approximately 55–70% after 15 days in high understory cover conditions. Despite differences in seed presentation, these are comparable proportions of removal over the same duration and in similar vegetation communities and understory cover as our seed removal study. Consequently, it is unlikely that our sous vide heat treatment of L. benzoin seeds had any substantial effect on our findings due to potential changes in olfactory cues used by rodent granivores.

We did not find any influence of vegetation community type, understory cover, or TSLF on post-dispersal seed predation over two years. For substrate data collected during 2019, we only found a significant positive effect of fine woody debris on seed predation. Overall, these findings were surprising given the abundant evidence that these extrinsic drivers and their interactions can influence rodent granivory [64, 117, 118] and abundance [65] in our study system and more broadly in longleaf pine ecosystems. We estimated mean TSLF and understory cover within a 30 m radius of depot locations, which may have been too coarse of a resolution to identify relationships for small rodent granivores despite successful application elsewhere [e.g., 119]. With regards to fine woody debris, we are unable to speculate why there might be a positive relationship with seed predation except that perhaps fine woody debris was associated with nearby coarse woody debris or standing snags not recorded in our 1 m² microplots. Both coarse woody debris and snags have been shown to be important habitat features for small rodent granivores in the southeastern USA and to influence foraging behavior [41].

Although direct evidence of in situ seed predation was commonly found in the form of empty endocarps within depots, we also observed cases where seeds were removed with no evidence of consumption. In our analyses, we assumed all seeds removed from the seed depots were consumed, as opposed to being taken to another location and cached (i.e. secondary dispersal, diplochory). This assumption can lead to overestimation of post-dispersal predation. However, secondary dispersal is difficult to quantify and can be affected by many factors including those related to seeds (e.g., size, nutrient composition, secondary metabolites, strength of protective layers, etc.), granivores (e.g., species, sex, individual behavior, physiological state, etc.), and environmental conditions (e.g., season, lunar cycle, predation risk, etc.) [120].

Conservation implications and additional research needs

Our estimates for pre- and post-dispersal predation of L. subcoriacea seeds suggest these two qualitative subcomponents of SDE potentially limit recruitment and population growth rates. This is especially true given the low fecundity, recruitment, and population growth rate documented for the species [37]. They also suggest that active conservation actions may be needed to improve the species conservation status. For example, either in situ seeding into predator exclosures or ex situ propagation and seedling outplanting may be warranted conservation strategies, if demonstrated to have adequate success [121]. In addition to directly promoting the general goals of population resiliency and redundancy [122], these actions could also be implemented to increase L. subcoriacea fruiting neighborhoods at scales that enhance SDE quantity. If either approach is employed, care should be taken to use best conservation practices (e.g., harvesting only a subset of available seeds, carefully evaluating the potential suitability of outplanting sites, monitoring the success of conservation actions, etc.) [e.g., 123, 124].

Although our findings suggest L. subcoriacea seeds are being dispersed by at least several avian species, we do not know with certainty the absolute or relative dispersal effectiveness of the different species in terms of quantity and quality. Given the challenges of observing and tracking birds within the vegetation communities occupied by L. subcoriacea, it might be useful to instead explore SDE by modeling the fruit consumption, gut retention times, movements, and habitat occupancy of the various species to identify the relative numbers of seeds potentially dispersed and the suitability of dispersal locations for L. subcoriacea recruitment [e.g., 125–127]. The dispersal information generated by this sort of modeling effort would also be useful for informing models of L. subcoriacea regional population dynamics that could guide conservation actions such as small population augmentation and population introduction to enhance connectivity [128, 129]. For example, Cipollini et al. [94] examined the importance of long-distance dispersal for patch-specific demography and mean population growth rate of L. benzoin. Like L. benzoin, L. subcoriacea populations are also likely dynamic, with inter-population processes (e.g., immigration and emigration) affecting the viability of populations within sites that vary in suitability over time and space.

Additional studies of L. subcoriacea SDE quality are also needed. For example, nothing is known about the factors that potentially limit seedling establishment such as light availability, drought, flooding, pathogens, or herbivory, which have been either documented or speculated to influence establishment of L. benzoin and/or L. melissifolia [36, 96, 130, 131]. However, low fecundity and high interannual variation in seed production will likely hamper these studies by limiting access to the numbers of seeds needed for robust study designs [37].