Study Species
This study focuses on species in the primarily epiphytic genera Elaphoglossum (ferns, fam. Dryopteridaceae) and Peperomia (dicots, fam. Piperaceae), which have high diversity in tropical montane forests (around 700 and 1700 species described worldwide, respectively). We collected data on all plants found within these genera during field work, but restricted analyses to 67 species that could reliably be identified in the field for detailed distribution surveys. We selected 15 species (see criteria below) for study in reciprocal transplant experiments.
Field Sites
Study area context: Species distribution data were collected by the authors across elevational and climatic gradients on three mountains in Costa Rica and Panama: Monteverde, Volcan Barva, and Volcan Baru (Fig. 1, Fig. S1). Monteverde is one of the best-studied cloud forests in the world, where the impacts of climate change have been demonstrated for decades. Documented abiotic changes have included an increase in precipitation variability, longer and more frequent drought periods, and higher nighttime temperatures (29); all these trends have continued since they were first reported in 1999. Biological changes include upward range shifts of birds and reptiles, amphibian extinctions and extirpations, and reptile and amphibian population declines (29).
Climate data and profiles of field sites: We collected climate data at all sites in Monteverde for time periods ranging from 6 months to 3 years. LogTag data loggers at all sites recorded hourly temperature and relative humidity (RH), while Decagon data loggers at transplant sites recorded temperature, RH, precipitation, and leaf wetness (LWS). Precipitation data were not usable because precipitation gauges were frequently clogged with debris. LWS is measured by a plastic leaf-shaped sensor that records the number of minutes per hour that the sensor surface is wet. Sensors were secured at approximately a 30-degree angle from horizontal to mimic a natural leaf position. “Dry days” occurred when the LWS sensor was dry for the entire day (i.e. recorded 0 minutes wet for every hour in that day), while “wet days” occurred when the LWS sensor was wet for the entire day (recorded 60 minutes wet for every hour in that day). Number of dry days and wet days were calculated as the total number of days from over 11 months of continuous LWS data (6/12/2015-5/24/2016).
Sensors frequently fail in these extremely humid environments, so there are gaps in the climate data. Comparisons between sites are calculated from time periods with continuous data from all sites. Lapse rates provided are calculated from a 6-month period (September 2014-March 2015) when all ten sites had simultaneously functioning LogTag sensors. Temperature and moisture profiles for transplant sites are calculated from an 11-month period (June 2015-May 2016) when both temperature and LWS sensors were working for all six sites (Figure S2).
We report an adiabatic lapse rate of 5.33°C per 1000m elevation in Monteverde and use this for calculations in this study. Lapse rates reported from field climate measurements on Volcan Barva range from 5.1°C (2) to 5.5°C per 1000m (30).
1. Transplant Experiment
Species Selection
Fifteen species of Elaphoglossum and Peperomia are reported in the transplant experiment (Figs. S6, S7). Two additional species were used in the experiment, but had such low sample size that no statistical power could be deduced, so they are excluded from analysis. From the species that were detected at Monteverde during the surveys (see below), we selected species for the transplant experiments if they met the following criteria: could be found growing naturally in the forest understory (not canopy specialists), small enough to be transplanted without serious disturbance (see Methods below), and abundant enough for several dozen individuals to be collected from a single site while leaving sufficient individuals at the site to be monitored as unmanipulated forest controls.
Experimental Design and Methods
All transplant experiments were conducted at Monteverde. Reciprocal transplants were performed across two distinct climatic gradients: a precipitation-only gradient, encompassing four sites at similar elevation (1380-1480m), but very different moisture regimes; and a Pacific slope elevational gradient, with three sites from 1100-1600m encompassing gradients in both temperature and precipitation (Fig. S2). Across each transect, transplants were performed reciprocally between all sites along that transect. We define a “direction” of transplant movement as movement from a unique origin site to a unique destination site (e.g., each arrow in Fig. S2c).
Wooden scaffolds were constructed to be the substrate on which the epiphytes would be moved and persist through the duration of the experiment (Fig. S3). Scaffolds were constructed from 1x2” planks of untreated laurel wood. Scaffold architecture allowed for epiphytes to be place on either horizontal or vertical beams, or at beam intersections, replicating the variety of natural orientations in which they were found growing. Only epiphytes growing in the forest understory (on low trunks or branches, or surviving and thriving on fallen branches) were collected. Transplanted epiphytes were tied to the scaffold with small strips of nylon fabric, a method commonly used in orchid cultivation. When collected epiphytes were growing rooted in a substrate of canopy soil, a piece of that substrate surrounding the plant’s rhizome or root system was cut out and transplanted along with the plant.
Each unique transplant direction (from one site to another) was realized with four replicate scaffolds, which were approximately identical in species composition and number of individuals per species, but not the physical layout of individuals/species on the scaffold; e.g., epiphytes were placed randomly on each scaffold while still accounting for individual architecture (whether a plant was growing on a horizontal or vertical branch), and individuals were spaced spread apart relatively evenly across scaffold beams so that they would not be directly competing for light or space. Ideal sample size was 20 individuals per species per direction, with 5 individuals on each replicate scaffold. In some cases, not enough individuals could be located in nearby forest, so a lower sample size was used. Sample sizes per species per transplant direction are given in table S4.
Each site was divided into four replicate “microsites” placed along a 100m transect in a flat area of the site. This was done to account for the possibility of subtle microclimatic variation, as well as to offset the risk of losing some scaffolds due to tree and branch falls or other natural causes. Microsite location was selected using a random number generator from 0-100, and placed at that distance along the transect. The four replicate scaffolds coming from each unique origin site were distributed among the four microsites. Thus, at each unique destination site, each microsite had identical scaffolds in terms of scaffold origin and composition. Among the 96 total scaffolds used in the experiment, three were lost before the end of the 36-month experiment due to being crushed by fallen branches. All three lost scaffolds were at site 5; two at the same microsite (from origin sites 1 and 5) after 30 months, and the third at a different microsite but also from origin site 5 (e.g. the experimental control), after 24 months. Interval censoring of survival data (see Analysis) allows the data from these scaffolds to remain incorporated into survival curves prior to their destruction, and their removal from the experiment does not factor into calculated mortality estimates.
Transplants were initiated during the rainy season in July-August 2014, when epiphytes were collected and tied onto scaffolds. All scaffolds were left to acclimate for 3-8 weeks between plant setup and movement, and scaffolds were moved to destination sites from September 1-8, 2014. Scaffolds were moved by being carried on the shoulders of two people, who hiked through the forest along trails between sites. Sites 1, 2, and 3 had to be accessed by road as well as trails, so all scaffolds moved to or from those sites were transported by car (for 5-20 minutes), with great care taken to not touch, disturb, or bounce the plants. Sites 4, 5, and 6 were connected by continuous trails.
Two types of control treatments were utilized in this experiment. First, experimental controls were treated exactly as all other transplants, i.e., were tied onto scaffolds and carried through the forest for a comparable amount of time to other transplants (30-60 minutes), but were then deposited at the site of origin. Second, unmanipulated or “forest” controls were plants tagged and monitored in the forest understory without ever being touched. The average survival of forest controls was 40% across all species and sites, which was lower than experimental controls (52%), and forest control survival was idiosyncratic and varied widely by species (Fig. S12). We observed that forest control mortality tended to occur for different reasons than mortality in experimental plants, such as their substrate (branches or trunks) breaking or being disturbed, being crushed by fallen organic material, or simply disappearing from their substrate without a trace. In contrast, experimental plants tended to die in place and could be observed before and after mortality, often appearing to dessicate or rot slowly over the course of multiple observations. From monitoring these control plants, we learned that epiphyte mortality in the forest understory is naturally high due to disturbance events, and thus highly idiosyncratic. For this reason, and because forest control survival was lower than experimental control survival, we do not adjust our experimental survival rates according to control survival for each species, as this would reflect idiosyncratic disturbance events rather than climate-induced mortality. We nonetheless present the forest control data (Fig. S12) to illustrate that the survival of experimental controls was higher than that of forest controls, and thus the experimental transplant procedure itself is unlikely to have inflated mortality rates.
Data on the survival, reproduction, growth, and herbivory of the transplants were collected quarterly for three years, starting immediately after transplant (September 2014), for the duration of the experiment (until September 2017). For every plant at every time point, a photo was taken of the plant, and survival and the number of reproductive structures were recorded. Peperomia produce spike-like infloresences/infructesences with many microscopic flowers and fruits, so the number of infloresence spikes per plant were counted. Elaphoglossum produce dimorphic fertile leaves that are morphologically and developmentally distinct from the sterile leaves, so the number of fertile leaves per plant were counted. Growth and herbivory were recorded for a subset of all plants: one individual per species per scaffold was randomly selected and then repeatedly measured. In the event that the individual being monitored died, a new individual of that species was selected from that scaffold and monitored for as long as it was alive; this process was continued unless all individuals of a particular species on a particular scaffold had died. Growth was measured by counting all leaves on each plant. Herbivory was measured by counting all the leaves with visible herbivory damage and estimating the amount of leaf tissue lost on a scale of 1-5, such that 1 corresponded to between 0-20% of tissue lost (average 10%), 2 means 21-40% (average 30%) of tissue, etc. The overall herbivory score for that plant was then calculated as the total fraction of leaf tissue lost for that plant:
H = (# of leaves with herbivory)*(0.2*herbivory score-0.1)/(total # leaves)
Statistics and Analysis
Transplant survival data: Although species-specific transplant survival is visualized using only the final survival values (after 36 months) for simplicity (Fig. 2, Fig. S6,7), significant differences in survival between groups were determined using survival curve analysis. Survival was analyzed with non-parametric likelihood estimators (NPMLE), more commonly known as Kaplan-Meier (KM) curves, for interval-censored survival data using the R package “interval.” For each transplant species, KM curves were constructed for survival at each destination site. The impact of destination site on survival was assessed for each species using an Asymptotic Logrank trend test (31).
Herbivory data: Herbivory results were analyzed using ANOVA after a log(data + 1) transformation of herbivory values. Two separate metrics of herbivory were assessed: baseline herbivory at the beginning of the experiment, and change in herbivory after transplant. These were both calculated for each individual plant with herbivory measurements. Baseline herbivory was the initial herbivory measurement at the start of the experiment. The post-transplant herbivory rate was averaged across all measurements from 1-3 years post-transplant. The change in herbivory was the post-transplant rate minus the baseline rate. For plants that we began monitoring during the course of the experiment (due to replacing a deceased individual), we assigned them a baseline herbivory value of the average of that species at their same origin site.
2. Elevational range surveys
Survey Design & Methods
Species occurrence data were collected between 2013 and 2016 on the mountains of Monteverde and Volcán Barva in Costa Rica, and Volcán Barú in Panama (Fig. 1, Fig. S1). Sites were established across elevational gradients on all three mountains, and thoroughly surveyed in order to provide genuine presence-absence data, meaning we were certain that undetected species were not present at the site. Standardized survey methodology at each site consisted of six hours of “timed searches,” in which we walked in a zigzag pattern through the understory, stayed within 50m.a.s.l. of the target elevation, and for each of twelve separate 30-minute windows, recorded all species encountered and the total number of individuals per species. Species-Accumulation Curves (SACs) were created using EstimateS software (32-34), using each 30-min window as a sample, and actual diversity was estimated from the asymptote of the curve in order to confirm that the number of species detected at each site was approximately equal to the estimate of true total diversity for the site. Fig. S13 shows, as an example, the SACs for all sites on Baru. In Monteverde, much more extensive occurrence surveys were carried out from 2013-2016, using alternative methods including tree climbing, ground plots, and targeted surveys of fallen trees. These alternative methods were used to confirm that timed walks were the most effective method of detecting the true diversity of a site. In all cases, 6-hour timed walks found the greatest number of species and took the least amount of time among the alternative methods. The total number of species detected from 6-hour timed walks plus additional incidental records from the site, i.e., species observed outside of the timed walks (usually 1-3 additional species) equaled the estimated diversity calculated by EstimateS for nearly all sites, giving us confidence that our species list for each site was comprehensive for our study groups. Thus, the timed walk method was chosen to most effectively survey sites on the other two mountains, enabling surveys over much larger elevational gradients in a feasible time period. In addition to the timed walks, “incidental records” were collected whenever species of interest were encountered on these mountains outside the survey sites, thus contributing to elevational range data. All mountains were surveyed intensively for a period of 1-3 years, with multiple trips up and down the elevational transects, collecting incidental records on many trails other than just those containing the survey sites. Latitude, longitude and elevation of all sites and incidental records were recorded with a Garmin GPS with accuracy <20m.
Statistics and Analysis
To confirm that surveys accurately captured the full diversity at all sites, Species-Accumulation Curves were created and analyzed for each survey site. Analytical methods are described above.
To calculate species’ range sizes along each elevational transect, species ranges were both interpolated and extrapolated as in Colwell et al. (2008) (2). Extrapolation means that a species’ range limit was assigned as the midpoint between its last recorded presence and its first recorded absence. Interpolation means that if a species’ range extent spanned an elevation along a given transect, e.g. it was detected above and below, but it was not detected at that elevation, it was considered present at that elevation.
To calculate extinction risk, we used species’ elevational ranges as measured in the field as a proxy for their climatic tolerances, since the transplant experiment suggested that these species cannot survive much outside their native range, and especially at lower elevations. Ranges were used to extrapolate climatic tolerances in two alternate ways: based on a species’ aggregate range across all three mountains, or based on each population’s elevational range on a single mountain. In the aggregate scenario, each population of a given species across the three mountains was considered to have the same climatic tolerance, spanning the breadth of elevations at which we recorded it in the field across all populations. Thus, the species’ aggregate ranges yield a larger estimate of their climatic tolerance. In the individual population scenario, the populations on each mountain are considered to be restricted to the elevations where they occur on that mountain.
We used the elevational lapse rate we measured at Monteverde (5.33°C per 1000m elevation) to estimate how much of an elevational shift would be needed to maintain a specific temperature profile with a given amount of climate warming. We used these elevational shifts to estimate how much the lower-elevational boundary of a species range would shift upward. Under a “no dispersal” scenario, we assumed the upper boundary of species distributions would remain static, leading to range contractions. Under a “full dispersal” scenario, we assumed upper boundaries would shift upward by an equal amount as lower boundaries, unless upper boundaries were limited by the total elevational extent of a mountain.
Two alternate extinction measures are calculated: Population extirpations and species extinctions. In the former, we measure the number of individual populations expected to lose all their range on the mountain where they occur. For species extinctions, a species is only considered extinct (in this region) if all of its populations on all the mountains where it occurs lose all of their range.
Finally, we calculated the average breadth of elevational range remaining for those species and populations that have not been extirpated. Even while populations remain extant, range loss puts them at higher risk of extinction due to stochasticity, so a decline in average range size indicates greater vulnerability of the assemblage to future loss. Extant species and populations will lose elevational range either due to range contraction from the lower boundary without corresponding expansion (in the no dispersal scenario), or when they abut the tops of their local mountains (in the dispersal scenario).