In this study, we quantify the effects of leaf traits and dispersal attributes on species responses to pig soil disturbance at two spatial scales – 0.5 m² patches embedded along 20 m transects within sites – across a gradient of pig density in a Hawaiian montane wet forest using Bayesian mixed models. 

Native and non-native species demonstrated divergent responses, with increasing presence and abundance of non-native species in the understory as soil disturbance within patches and sites increased. Dominant patterns in measured traits tracked the leaf economic spectrum (LES), with non-native species tending toward resource-acquisitive traits. Species with resource-acquisitive traits, regardless of identity, were favored with disturbance and responded positively to light availability in disturbed sites. Models showed species primarily dispersed by wind were more prevalent in disturbed patches and sites than those dispersed by endozoochory, while seed mass had no effect.

From Methods outlined in the publication:

Data Collection

The impacts of pig activity and disturbance vary by scale, and effects on community composition are mediated in part by differences in establishment success due to disturbance within habitat patches. To capture these effects, we used a hierarchical design incorporating pig density at the zone level, our broadest spatial scale, and pig soil disturbance (i.e. physically overturned soil as a result of pig activity) at two consecutively smaller scales: (i) within sites nested within zones (hereafter, Dsite), and (ii) within habitat patches nested within sites (hereafter, Dpatch). Four zones with 1-km radii, roughly equal to conservative estimates of pig home-range size in Hawaiian forests, were established within the study site (Fig. 1A) (Diong, 1982). To capture patterns in understory composition and soil disturbance at the site- and patch-scale, we surveyed 30 sites within each zone using a stratified-random design (1 additional site was accidentally surveyed in Zone A, leading to a total of 121 sites). Sites with < 50% canopy coverage were excluded to target the understory under forested conditions. At each site, we established a 20-m transect oriented along a random azimuth from 0 – 360 degrees to capture site-scale community composition and disturbance. We then uniformly distributed 0.5 m x 0.5 m quadrats (hereafter, patches) at 3-m intervals along the transect starting from the 2-meter mark, for a total of 6 patches per site representing individual habitat patches where plant species establishment might occur. 14 patches were discarded from the analysis due to trees or other obstructions preventing quadrat establishment at locations along the transect, resulting in a total of 712 patch-scale measurements. This nested, hierarchical approach allowed us to untangle the effects of pig disturbance across multiple spatial scales and to test the underlying impacts of disturbance on vegetation community dynamics.

Disturbance and Vegetation Characterization

To quantify pig density within zones, we conducted a camera trap survey during Summer 2019 designed to estimate density following Rowcliffe et al. (2014). We randomly generated 20 potential trapping locations within each zone and deployed cameras at a random subset of 4-6 locations over three 10–14-day trapping periods. The absence of pigs at the fenced zone D was confirmed by camera traps at 7 locations over the final two trapping periods. Camera traps were programmed to collect a set of three pictures for every trigger with a 1-second delay between triggers. Cameras were attached to trees within a 20-m radius from each sampling location in an area where pig activity was evident from signs of trails, rooting activity, and scat to avoid underestimation of population size (Rowcliffe et al., 2008). Cameras were attached ~50 cm above the base of a tree and were oriented to have a capture distance of 5 m and to capture images within an angle of ~45 degrees. To calculate pig density estimates, we fit a random encounter model (REM), which assumes the trapping rate (number of photographs taken per unit time) provides quantitative information about the density of a species (Rowcliffe et al., 2008, 2014). This model estimates population density from random encounters with camera traps while parameterizing covariates that affect trapping rate. We used a bootstrapping approach to integrate sampling (location-specific variability) and non-sampling (i.e., model dependencies) error in our final density estimates (see SM).

To capture site- and patch-specific patterns of pig disturbance, we conservatively defined disturbed area as that with visible signs of mechanical disturbance to surface soil structure from digging, wallowing, and trampling activity. Other ungulates that could be associated with these disturbances were not detected in the camera trap survey and are presumed to be absent. Pig trails, defined here as patches and pathways with clear evidence of pig use but without overturned soil, were excluded. To calculate disturbance extent within patches (Dpatch), we estimated the percent of disturbed area within each patch and assigned estimates to cover categories of 0%, 0–10%, 10–25%, 25–50%, 50–75%, and 75–100%. Estimates are presented as the median value for each category to reduce measurement bias. To capture disturbance at the site scale (Dsite), we independently measured disturbance along each 20-m transect at each site by summing the length (within 0.05 m) of transect segments intersecting disturbed soil as defined above. As nurse logs are important for native woody species establishment in Hawaiian wet forests (Rehm et al., 2021; Santiago, 2000), we quantified nurse log area using the same methods as disturbance (NLpatch in patches, NLsite in sites).

We estimated light availability using hemispherical photographs taken with a 180° fisheye camera lens at three points along each transect, each located between quadrat pairs. Using the Gap Light Analyzer software (Frazer et al., 1999), we calculated canopy openness for each photograph, which we used as a metric for light availability. For patch-scale measurement, we assigned light availability values (hereafter, Lightpatch) to pairs of patches located most closely to where each photograph was taken, and averaged all three measurements for site-scale estimates (hereafter, Lightsite).

Finally, we assessed the composition of understory vascular plants at both the patch- and site-scale. At the patch scale, we identified individuals < 2 m tall and rooted within each quadrat to assess species occurrence (i.e. presence/absence) within habitat patches. At the site-scale, we evaluated patterns of community composition by estimating understory species cover using a point-intercept approach. Vascular plant species intercepted between 0–2 m in height were recorded at intervals of 0.5 m for a total of 41 points per site. Individuals were identified to the species level except for Peperomia, Coprosma, and Adenophorus, which were identified to the genus level because characteristics for species-level identification were typically missing. Flowering plant taxonomy was based on Wagner (1991) with updated nomenclature according to Smith and Brown (2018), and pteridophyte taxonomy was based on Palmer (2002).

Trait data collection

To understand how plant responses to pig activity and disturbance are influenced by differences in plant ecological function, we measured functional traits associated with leaf construction and dispersal. Leaf investment strategies can be identified by covariance in leaf traits linked to maximum photosynthetic rate and growth, such as specific leaf area (SLA), leaf carbon to nitrogen ratio (C:N), leaf thickness, and leaf dry matter content (LDMC) (Díaz et al., 2016; Reich, 2014; Wright et al., 2004). To measure these characteristics, we sampled fully expanded leaves (≥5 per individual) from 6 haphazardly selected individuals of each species where possible, distributed across at least 3 pig-activity zones. Due to the presence of Sphagnum in Kohala (Schomaker, 2017), 3 individuals of each species were selected to be rooted in Sphagnum where possible to account for differences in N concentrations in Sphagnum and non-Sphagnum areas. All leaves were collected, weighed, placed in moist plastic bags, and leaf area and thickness were measured within 48 hours of collection. Leaf area was measured using a LI-COR area meter. Leaves were weighed for fresh mass, dried at 60°C for at least 48 hours or until fully dry, weighed again for dry mass, ground using a Wiley mill, and sent to the University of New Mexico Center for Stable Isotopes (Albuquerque, NM) for %C and %N content using accelerator mass spectrometry. SLA was calculated by dividing leaf area by dry mass, and LDMC was calculated by dividing dry mass by fresh mass. Lycophytes were removed from the analysis due to incongruencies in comparing lycophyte microphylls to megaphylls of other species. Average seed mass for angiosperms was extracted from Yoshinaga et al. (unpub) and published sources from the TRY Plant Traits Database (Kattge et al., 2020) (SM). For species for which data were not available, we averaged available seed mass values among the most closely related species in the genus. Analyses involving seed mass and dispersal were conducted with pteridophytes removed.