Climate change may induce connectivity loss and mountaintop extinction in Central American forests

The study investigates how climate change may alter the environmental suitability and distribution of Central American forest plant functional types (PFTs). By grouping tree species into trait-based PFTs (wet/dry acquisitive, wet/dry conservative, generalist, montane, coniferous) and projecting suitability under multiple climate scenarios (RCP 2.6, 4.5, 8.5) and GCMs, the research aims to identify likely forest type transitions, connectivity losses, and risks of mountaintop extinction, informing conservation and land-use planning in a biodiversity hotspot.

Study area: Central America between 68°–100°W and 4°–24°N (excluding Caribbean islands), extending into parts of Mexico and northern South America to broaden calibration climates. PFT definition and species selection: Lowland (<1000 m) forests divided into wet and dry; higher elevations into coniferous and broadleaved montane. Wet and dry were further split into acquisitive (fast growth/high assimilation) vs conservative (resource preservation/stress tolerance) strategies. A generalist type captured species spanning wet–dry gradients with trait trade-offs. Four representative tree species per PFT were selected based on literature, trait data, broad ecoregional occurrence, and ≥200 records. Data: Presence records from GBIF, BIEN, and de Sousa et al.; taxonomic harmonization; spatial thinning to one record per environmental grid cell to reduce sampling bias. Environmental predictors at 30-arc-second resolution: CHELSA v1.2 bioclim (present 1979–2013; scenarios from CCSM4, HadGEM2-AO, MPI-ESM-LR for RCPs 2.6, 4.5, 8.5), GMTED2010 topography, SoilGrids1km edaphic variables. Multicollinearity reduced via stepwise VIF (<10), retaining: Tmax warmest month, Tmin coldest month, precipitation seasonality, annual precipitation, soil sand and clay fractions (30 cm), soil pH (30 cm), depth to bedrock, hillslope. Modelling: Stacked species distribution modelling (R package SSDM). Algorithms tested: ANN, CTA, GAM, GLM, MARS, MAXENT, SVM, RF. For each species and algorithm: 100 replicates with 70/30 train/test cross-validation. Evaluation metrics: AUC, Continuous Boyce Index (CBI), and a calibration statistic (except MAXENT for the latter). Minimum threshold 0.7; ANN, CTA, GLM, SVM commonly failed and were excluded; retained GAM, MARS, MAXENT, RF. Projections applied to future climates; non-analog climates minimal; values clamped to training range to avoid extrapolation (interpret with caution in clamped areas). Variance partitioning (ANOVA) across grid cells indicated algorithm choice as dominant uncertainty source, followed by RCP and GCM, with replicates lowest; thus 10 replicates per retained algorithm were carried forward for efficiency. Ensembling and PFT aggregation: For each species, unweighted arithmetic mean across retained algorithms and replicates; PFT suitability derived by summing probabilities of the four member species (“richness maps”). Additionally, binarized stacked SDMs (bS-SDM) using MaxSSS threshold and algorithm agreement were produced; cells with bS-SDM ≥2 counted as PFT presences. Range shift analyses: Density of PFT presences plotted over latitude and altitude to assess latitudinal shifts, upslope movements, and connectivity trends. Fragmentation analysis: Using Riitters et al. classification on dominant PFT maps via a 3×3 moving window, computing Pf (proportion) and Pff (connectivity) to assign classes: interior, patch, transitional, perforated, edge, and area loss. Model evaluation: RF showed highest discrimination and calibration with lowest variance; montane (narrow-niched) had highest AUC; generalists had lower AUC but better calibration; consistent performance ranking patterns across algorithms.

- Latitudinal and altitudinal shifts: Montane and coniferous PFTs show upslope shifts of lower suitability boundaries by ~100 m (RCP2.6), ~200 m (RCP4.5), and ~500 m (RCP8.5), indicating elevated risk of mountaintop extinction. Dry acquisitive and generalist presences shift from north to south; dry conservative densities slightly increase between 8–12°N. Wet forest PFTs trend southward and diverge into multiple suitability centers under RCP4.5 and RCP8.5, implying connectivity loss. Small upslope shifts noted for wet conservative. - Dominant PFT changes: Broad transitions from wet towards generalist or dry forest types across large areas. Generalist PFT area increases twofold (RCP2.6) up to fivefold (RCP8.5) and doubles its interior fraction, often occupying emerging bottlenecks. - Fragmentation and connectivity: Wet conservative PFT may lose up to 78% of suitable dominant area, with remaining areas increasingly fragmented (patches/transitional), nearly doubling fragmented fractions under RCP8.5 compared to present. Wet acquisitive shows moderate total area increases, but connectivity bottlenecks appear along the Caribbean coasts of Honduras and Nicaragua and at the Isthmus of Panama. Dry acquisitive shows trends toward reduced interior and increased patch/transitional/perforated areas; dry conservative fragmentation remains similar. Coniferous areas decrease slightly without major fragmentation class shifts. Montane areas shrink by up to 44%, with slight increases in perforated classes. - Spatial connectivity risks: Suitable areas for wet-adapted PFTs along the Caribbean coast increasingly isolated with RCP strength, threatening the Mesoamerican biological corridor’s north–south connectivity.

The findings indicate that climate change will likely reorganize Central American forest compositions, driving transitions from wet to generalist or dry PFTs and fragmenting wet forest habitats. Divergent latitudinal and altitudinal responses among wet PFTs impair connectivity, compounding human-driven fragmentation and jeopardizing the integrity of the Mesoamerican biological corridor. Upslope shifts for montane and coniferous PFTs, restricted by topographic limits, elevate long-term risks of mountaintop extinctions. Ecologically, wet and montane forests harbor the highest vertebrate richness, so their contraction and fragmentation could disproportionately impact biodiversity already under threat. Socioeconomically, a shift toward drier forest types could reduce carbon sequestration and affect ecotourism and timber production, with drought-adapted species potentially outcompeting economically important pines; pest outbreaks and El Niño-related droughts may exacerbate impacts. Other plant groups (e.g., lianas) could further alter structure and carbon dynamics. Conservation responses should prioritize maintaining and enhancing connectivity through biological corridors, protecting climatically stable and key linkage areas, and expanding protected areas for species at risk of mountaintop extinction. Emerging reforestation trends in the Neotropics could complement these strategies if aligned with connectivity conservation. Overall, trait-based PFT projections provide actionable insights for regional planning and can be extended to other biodiversity hotspots to anticipate connectivity loss and mountaintop extinction risks.

This study projects substantial climate-driven shifts in the suitability of Central American forest PFTs, with transitions from wet to generalist/dry types, fragmentation and connectivity losses in wet forests, and upslope pressures on montane and coniferous types that heighten mountaintop extinction risks. The results highlight urgent needs to safeguard and restore landscape connectivity via biological corridors and to expand protected refugia in vulnerability hotspots. Future research should integrate growth responses and disturbance regimes across biomes, examine interactions with other plant groups (e.g., lianas, palms, grasses), and apply trait-based, ensemble SDM approaches to other hotspots to anticipate and mitigate connectivity loss and extinction risks under climate change.

- SDM-related uncertainties: Potential disequilibrium between occurrences and environment, sampling bias, spatial autocorrelation, and algorithm biases. Algorithm choice was the largest source of projection variance, followed by RCP and GCM. - Calibration and extrapolation: Although models met evaluation thresholds (AUC, CBI, calibration statistic), recalibration did not improve performance. Future projections were clamped to training ranges to avoid extrapolation; predictions in clamped areas may underestimate climate change effects. - Interpretation scope: Projections represent environmental suitability, not realized distributions or timelines of change. Actual range shifts and extinctions depend on additional factors (disturbances, resource availability, CO2 effects, dormancy, dispersal, biotic interactions). - Dispersal and barriers: Biogeographic barriers (mountains, isthmus), anthropogenic land use, urbanization, and cropland expansion constrain migration and reduce available range-shift area. - Species representation: PFTs represented by four species each; while chosen to be representative, this may not capture full intra-type diversity and responses.