Future climate change vulnerability of endemic island mammals

Islands host unique biotas but are particularly susceptible to climate change due to limited area, isolation, and the insularity syndrome that constrains dispersal and adaptation. Prior assessments seldom focus on islands and often measure only exposure, overlooking biological traits that influence species’ capacity to cope with changing climates. This study asks where and why islands (and archipelagos) will be most vulnerable to climate change by 2050, explicitly integrating exposure, sensitivity, and adaptive capacity for endemic mammals. The purpose is to identify spatial hotspots of vulnerability, determine the main drivers among the three vulnerability components, and link species’ ecological traits to vulnerability to better inform conservation prioritization and potential climate refugia identification.

The paper reviews the evolution of climate-change vulnerability concepts since the IPCC’s 2007 definition of exposure, sensitivity, and adaptive capacity. It notes that most biodiversity assessments emphasize exposure using correlative species distribution models, with fewer integrating mechanistic or trait-based approaches due to data limitations. Trait-based assessments are increasingly feasible as trait data accumulate but are challenged by arbitrary thresholds for continuous traits; multicriteria decision analysis (e.g., TOPSIS) offers a way to avoid such thresholds. Previous island-related assessments often measured only exposure and rarely considered endemic species’ traits, limiting conservation prioritization. The authors build on prior work quantifying local climate change (SED metrics) and on frameworks identifying traits linked to climate responses in vertebrates to deliver an integrated, island-focused assessment.

Study scope: 873 endemic mammal species (single- and multi-island endemics) were compiled from IUCN distributions and overlapped with a global island dataset, yielding 340 islands across 14 archipelagos (each with >5 endemic mammal species). Analyses were run in R 3.6.0.

Vulnerability framework: Vulnerability was computed as a function of three standardized components—exposure, sensitivity, and adaptive capacity—combined using the TOPSIS multicriteria decision method to rank each island by similarity to positive (no vulnerability) and negative (maximum vulnerability) ideal solutions.

Exposure: A local climate change (LCC) metric was calculated following Williams et al. as standardized Euclidean distances (SED) between current (WorldClim v2, 1970–2000, ~1 km) and future climates (2041–2060; five CMIP5 GCMs: CCSM4, HadGEM2-ES, MIROC5, MRI-CGCM3, NorESM1-M) under RCP 6.0 and RCP 8.5. Six variables were used: annual mean temperature, maximum temperature of warmest month, minimum temperature of coldest month, annual precipitation, precipitation of wettest month, and precipitation of driest month, standardized by temperature and precipitation seasonality. For each island, mean SED was computed across on-island grid cells; additional SEDs were computed within 100 km and 500 km buffers to reflect potential nearby climatic analogues. Higher SED indicates higher exposure. All islands had ≥10 climate grid cells (mean ± s.d.: 14,517 ± 85,973). Mean SED (all islands): 11.33 ± 8.35 (0.86–69.78) for RCP 6.0; 13.89 ± 9.74 (0.75–67.74) for RCP 8.5.

Sensitivity: Calculated as the island-level mean of four species-level attributes for endemic mammals present: (1) habitat specialization (number of IUCN habitat types used; fewer = more specialized), (2) dietary specialization (number of diet categories from EltonTraits; fewer = more specialized), (3) generation length (years; Pacifici et al.), and (4) ecological redundancy (number of species sharing identical combinations of ecological traits: main diet, foraging niche, foraging period, habitat niche breadth, and body mass; lower redundancy implies higher risk of co-extinctions from disrupted interactions).

Adaptive capacity: Four island-level attributes: (1) geographic isolation (proportion of surrounding land mass within 100, 1,000, and 10,000 km; more isolation reduces potential for dispersal to refugia), (2) protected area coverage (percent of island area under protection from WDPA), (3) phylogenetic distinctiveness of the island’s endemic mammal pool (mean fair-proportion index from PHYLACINE trees; greater distinctiveness implies higher evolutionary potential), and (4) extinction rate since 1500 CE (number of vertebrate extinctions divided by total vertebrate richness; higher historical filtering implies remaining species may be more disturbance-resistant).

Standardization and robustness: Each variable was normalized to 0–1 using three alternative methods tested for robustness: (a) min–max linear rescaling, (b) ln(x+1) transform followed by min–max, and (c) percentile rank (CDF). Component scores (exposure, sensitivity, adaptive capacity) were sums of their variables and re-standardized to 0–1. Pairwise correlations among variables within components were checked (|Spearman r| < 0.7). Robustness analyses removed one variable at a time and compared spatial congruence (Spearman’s rho > 0.8 across alternatives), and tested both RCP 6.0 and 8.5.

Vulnerability quantification (TOPSIS): Criteria: exposure and sensitivity (benefit to vulnerability; higher increases vulnerability) and adaptive capacity (cost to vulnerability; higher decreases vulnerability). Positive ideal: minimal exposure and sensitivity, maximal adaptive capacity (no vulnerability). Negative ideal: maximal exposure and sensitivity, minimal adaptive capacity (maximum vulnerability). Each island’s vulnerability is its relative distance to these ideals, scaled 0–1 (low–high).

• Spatial patterns: Island vulnerability values ranged from 0.18 to 0.71. 63% of islands had high vulnerability (>0.5), with hotspots concentrated in the Pacific. Low vulnerability islands included parts of Japan, Tasmania, Sri Lanka, and the Caribbean.
• Archipelagos: Mean archipelago vulnerability ranged from 0.25 (Tasmania) to 0.67 (New Hebrides). Six archipelagos were highly vulnerable (>0.5): New Hebrides, Bismarck Archipelago, New Caledonia, Solomon Islands, Malay Archipelago, and Sulawesi.
• Species richness: No correlation between vulnerability and endemic species richness at island or archipelago scales.
• Component magnitudes (archipelago scale): exposure 0.25 ± 0.21; sensitivity 0.50 ± 0.20; adaptive capacity 0.40 ± 0.14. Only two highly vulnerable archipelagos were also highly exposed (>0.5): Bismarck Archipelago and New Hebrides. Eight archipelagos (including the six highly vulnerable) had high sensitivity (>0.5) and low adaptive capacity (<0.5).
• Component relationships and drivers: PCA axis 1 (33.5% variance) aligned with higher vulnerability (Pearson r = 0.98), higher exposure (r = 0.85), and lower adaptive capacity (r = −0.73). Axis 2 (18.7%) associated with sensitivity (r = 0.71). Vulnerability correlated positively with exposure (Spearman rho = 0.85) and negatively with adaptive capacity (rho = −0.71); sensitivity had a weaker positive association (rho = 0.41). Sensitivity and exposure were positively related; other pairwise component relations were not significant.
• Trait associations: Across islands, vulnerability increased with dietary specialization and with longer generation length (both significant positive correlations). No global relationship was found with ecological redundancy or habitat specialization, though patterns varied by archipelago (e.g., West Indies showed negative associations for some traits, Japan the opposite).
• Robustness: Spatial patterns of vulnerability and components were robust to alternative normalizations and RCPs (Spearman’s rho > 0.8). Removing individual variables modestly shifted component values but did not change the identification of highly vs. lowly vulnerable archipelagos.

The study demonstrates that future climate change will impose additional threats on all island mammal communities, with pronounced vulnerability in Pacific archipelagos. Crucially, exposure alone did not explain vulnerability patterns; many highly vulnerable archipelagos were not the most exposed. Instead, high sensitivity and low adaptive capacity frequently drove vulnerability, underscoring the importance of incorporating biological traits, historical context, protection status, and isolation into assessments. Differences among archipelagos with similar component scores reflected distinct underlying drivers (e.g., adaptive capacity in West Indies linked to extinction history and isolation versus phylogenetic distinctiveness in Japan), implying that conservation strategies should be tailored to local determinants. Species most at risk on vulnerable islands are dietary specialists with long generation lengths, whose loss could trigger trophic downgrading and functional impairments. Protected areas may bolster resilience and facilitate range shifts, but coverage differs widely (e.g., null in Bismarck vs. ~10% in Sulawesi). Overall, comprehensive, component-based assessments, rather than exposure-only approaches, are essential to prioritize actions and identify potential refugia.

By integrating exposure, sensitivity, and adaptive capacity via a trait-informed, quantitative framework, the study identifies islands and archipelagos where endemic mammals are most vulnerable to climate change by 2050. Priority hotspots include Bismarck Archipelago, Malay Archipelago, New Caledonia, New Hebrides, Solomon Islands, and Sulawesi; notably, only two are highly exposed, revealing the dominant roles of sensitivity and adaptive capacity. Conservation should prioritize strengthening adaptive capacity (e.g., protected area networks, connectivity) and addressing intrinsic sensitivities of diet-specialized, slow-reproducing mammals. Future research should: develop finer-resolution island climate data to capture microrefugia; broaden assessments to entire communities and additional climate drivers (sea-level rise, extreme events); and refine trait sets and thresholds for island contexts, including historical extinction filters and indirect human responses to climate change.

• Climate data resolution: Use of coarse global projections may underestimate microclimatic refugia, particularly on topographically complex oceanic islands.
• Trait and component selection: Vulnerability outcomes can shift with trait choice; some traits (e.g., dietary specialization, ecological redundancy; extinction rate, PAs) influence component magnitudes when included/excluded.
• Ecological redundancy estimation: Calculated using only endemic mammals, potentially underestimating true community redundancy by excluding non-endemic native species.
• Taxonomic and endemic focus: Results apply to endemic mammals and may not generalize to other taxa or to non-endemic assemblages.
• Climate drivers considered: Focused on temperature and precipitation; other important factors (e.g., sea-level rise, extreme events) were not integrated.
• Potential biases in historical extinction data and phylogenetic uncertainty, though multiple trees and robustness checks were used.