A global assessment of the drivers of threatened terrestrial species richness

Safeguarding species against threatening processes remains a global priority, yet biodiversity continues to decline despite international agreements. Setting effective global conservation priorities requires understanding where threatened species concentrate and why threatened species richness diverges from total species richness. Conservation planning often targets overall species diversity as a surrogate for other biodiversity dimensions, but indices are spatially incongruent and total richness only partially explains threatened richness. Many regions (e.g., South America, sub-Saharan Africa, India, Southeast Asia) host more threatened species than expected from their species pools. Extinction risk is driven by human activities (e.g., population growth, land cover change, invasive species), but environmental conditions can predispose species to threat by promoting small-ranged, endemic, and extinction-prone taxa through energy availability and habitat heterogeneity. The study addresses the gap in globally disentangling predisposing environmental conditions from threatening human processes to explain spatial patterns in threatened terrestrial vertebrate species richness across taxa and regions.

Prior work shows limited spatial congruence among biodiversity indices and between hotspots of species richness, endemism, and threat. Determinants of total species richness have been widely studied (energy, water, environmental heterogeneity, speciation), but the drivers of threatened richness remain less explored. Human pressures (population density, land use change, invasive species, climate change) elevate extinction risk, yet environmental conditions may generate extinction-prone assemblages (e.g., small-ranged endemics). Islands are known centers of endemism and higher extinction proneness. Previous reports suggested correlations between human influence and productivity; however, links between speciation processes and extinction risk indicate environments fostering diversification can also raise contemporary imperilment. The study builds on this literature by quantifying, at global and regional scales and across vertebrate classes, the relative importance of environmental versus human impact covariates after accounting for total species richness.

Data: Global dataset of 26,746 terrestrial vertebrates (amphibians, reptiles, birds, mammals), with 20% listed as threatened (IUCN Red List: CR, EN, VU). Distributions were aggregated to a 0.5° equal-area grid. Predictors: Total species richness; Environmental covariates including insularity (landmass area per grid cell as a measure of island vs continental context), temperature seasonality, annual precipitation, long-term climate stability since the last interglacial (125 ka to present), mean temperature, elevation metrics (standard deviation, minimum), precipitation seasonality, and habitat (land cover) diversity (Shannon index). Human impact covariates included area of anthropogenic land use (crop/urban from GlobCover 2009), Human Influence Index, protected area coverage (IUCN categories I–VI), short-term land cover change (1992–2015), long-term land cover change (cropland change 1700–1992), and national-level counts of invasive alien species. All predictors were aggregated to the 0.5° grid. Regions: Analyses were repeated for 19 zoogeographic regions (per Holt et al. 2013; Polynesian region excluded due to insufficient cells). Modeling: Random Forest regression models (R, randomForest package) were used. Three model sets: (1) threatened species richness as a function of total species richness alone (global and regional); (2) threatened richness as a function of total richness plus environmental and human impact covariates (global, regional, and by taxonomic class); (3) threatened richness residuals from model (1) regressed on environmental and human covariates (to verify independence from species pool size). Collinearity checks indicated only temperature seasonality vs mean temperature had |r|>0.7. Cross-validation and spatial blocking: To address spatial autocorrelation, data were partitioned into 10 spatial blocks based on ecoregions, ensuring each block covered similar environmental ranges. Ten-fold cross-validation was used to tune number of trees and mtry; performance measured by R² on held-out blocks. Variable importance: Computed via randomization (permutation) importance: VI = (MSE_rand − MSE_obs)/MSE_obs, repeated 1000 times per variable; summarized by means (and distributions in figures). Functional relationships: Separate RF models fitted to total and threatened richness (without total richness as a predictor) to visualize partial response curves by predicting across focal variables while holding others at means/modes. Data-deficient species sensitivity: Models re-run assuming 0%, 50%, and 100% of DD species per cell are threatened; results qualitatively similar to main analysis. Model exclusion: Regional models with low explanatory power (R² < 0.25) were excluded from importance summaries.

- Environmental covariates generally explained more variation in threatened species richness than human impact covariates at the global scale, after controlling for total species richness. Repeated-measures ANOVAs confirmed higher mean importance of environmental vs human impact variables (total vertebrates: estimate 0.58, z=11.64, p<0.01), with total species richness being most influential overall. - Insularity (landmass area) was the single most important global predictor of threatened vertebrate richness, with island regions supporting more endemic, extinction-prone species. Temperature seasonality and annual precipitation ranked 3rd and 4th in importance globally. - Among human variables, long-term land cover change and invasive alien species ranked 6th and 7th globally; results align with habitat loss and invasive species as leading extinction drivers. - Model performance (R², mean ± SD): Using only total species richness, global models had moderate explanatory power (total vertebrates 0.49±0.03; amphibians 0.04±0.03; reptiles 0.45±0.04; birds 0.37±0.08; mammals 0.36±0.05). Including environmental and human covariates improved global R² substantially (total vertebrates 0.81±0.14; amphibians 0.63±0.20; reptiles 0.70±0.19; birds 0.77±0.16; mammals 0.79±0.15). Regional models (species richness only) were very strong (e.g., 0.94±0.01 for total vertebrates). - Taxon-specific drivers varied: • Amphibians: Elevation diversity was most influential after total richness; threatened amphibians concentrated in topographically diverse areas; patterns particularly strong in South American, Amazonian, and Australian regions, potentially linked to chytridiomycosis prevalence in montane habitats. • Birds: Temperature seasonality was most influential after total richness, with threatened birds concentrated where temperature variation is particularly high or low (e.g., migratory species vulnerability in seasonal climates; persistence of naturally rare species in benign climates). • Differences across taxa likely reflect variation in dispersal ability and sensitivity to fine-scale habitat heterogeneity. - Regional variation in drivers was pronounced: • In regions with higher-than-expected threatened richness (given total richness), human pressures often dominated. Examples: Amazonian region—long-term land cover change most influential; Madagascar—area of human-dominated land uses most important. • In Southeast Asia, environmental variables prevailed: insularity (Indo-Malayan), temperature seasonality (Chinese), and annual precipitation (Oriental) were key. - Contrasting responses between threatened vs total richness: • Amphibians: Elevation variation positively related to threatened richness but negatively to total richness; threatened richness increased with long-term land use change while total richness remained relatively constant. • Reptiles: Threatened richness showed a positive asymptotic relationship with long-term land use change; total richness relatively constant. • Birds and mammals: Threatened richness increased with area of human-dominated land uses, whereas total richness showed a hump-shaped relationship (initial increase then decline) with human-dominated land uses, consistent with productivity effects followed by habitat loss/fragmentation impacts. • Mammals: Insularity associated with higher threatened richness in insular areas, but total richness highest on larger landmasses. - Results robust to alternative assumptions about data-deficient species’ threat status; inclusion of DD species (0%, 50%, 100% threatened) did not materially change qualitative outcomes.

By separating the influence of total species pool size from environmental and human covariates, the study shows that environmental characteristics that predispose assemblages to threat—particularly insularity, climate regimes, and topographic complexity—often outweigh direct human impact variables in explaining where threatened species concentrate. Nevertheless, human pressures remain crucial, especially at regional scales where land cover change, anthropogenic land use, and invasions align with elevated imperilment in areas rich in narrow-range endemics. These findings clarify why hotspots of threatened richness diverge from hotspots of total richness and vary across taxa and regions. Recognizing inherent vulnerability provides a basis for prioritizing conservation actions: regions predisposed to generating extinction-prone assemblages may require stricter protection and proactive mitigation against human pressures, while strategies can be tuned to taxa-specific sensitivities (e.g., addressing disease and topographic refugia for amphibians, migration-linked vulnerabilities for birds). The divergent functional relationships between covariates and total versus threatened richness underscore that conservation planning cannot rely on total richness as a surrogate for imperilment patterns.

The study provides a global, taxon- and region-resolved assessment of the drivers of threatened terrestrial vertebrate species richness after accounting for total species richness. Environmental factors—especially insularity, climate, and topographic heterogeneity—are typically more influential than direct human impacts in shaping global patterns of imperilment, though human pressures are pivotal at regional scales. The work offers a framework to identify inherently vulnerable regions and taxonomic groups, informing global conservation prioritization and policy. Future research should integrate these models with projections of environmental and land use change to forecast shifts in threatened species richness and guide proactive, scalable conservation interventions.

- Spatial resolution is coarse (0.5° grid), which can favor climatic over land cover predictors and limits detection of fine-scale drivers of threat. - Some human impact measures (e.g., invasive alien species counts) are at national scales, potentially reducing within-region variation and apparent importance. - Potential circularity in including insularity given that range size informs IUCN threat status; included to assess island effects but may partially overlap conceptually with threat criteria. - Heterogeneity in global data quality and sampling effort across regions and taxa (notably reptiles) may bias estimates; sensitivity analyses with data-deficient species suggest robustness, but gaps remain. - Some regional models had low explanatory power (R² < 0.25) and were excluded from importance summaries, indicating uneven model reliability across regions.