Reassessment of the risks of climate change for terrestrial ecosystems

Global circulation models forecast strong climatic changes this century, implying both continuous and abrupt shifts in ecosystem and species distributions. Managers face uncertainty in the magnitude and direction of change, compounded by the potential emergence of novel and disappearing climate states. Prior global risk assessments commonly computed climate exposure indices (for example, standardized Euclidean distances between present and future climates, or climate change velocity) to identify areas at risk. However, these indices do not incorporate nonlinear, colimiting physiological responses to multiple climatic drivers, nor differences among species and plant growth forms. Consequently, ecological responses are unlikely to map directly to raw climate exposure. This study addresses that gap by transforming climate into phytoclimate—quantifying how climates support 14 plant growth forms—using process-based, ecophysiological species models for 135,153 vascular plant species. The research asks where and how the climatic forcing of terrestrial ecosystems is projected to change by 2070, and where novel or disappearing phytoclimates may emerge, to better inform conservation priorities.

Previous work has assessed climate change exposure through metrics like standardized Euclidean distances in multidimensional climate space and climate velocity, and identified regions of novel and disappearing climates. While such indices are tractable, their ecological interpretation is limited because they ignore nonlinear and colimiting physiological responses and reorganizing hierarchies of limiting factors. Ecosystem simulation models offer ecological interpretability but are hampered by process and parameter uncertainty and model disagreement. Species-focused studies have mapped when animals will experience conditions beyond their realized niches. Past analyses often emphasized temperature-driven novelty in the tropics due to standardization by historical variability, which tends to upweight temperature over precipitation. Process-based species distribution models such as the TTR-SDM have shown improved transferability beyond the calibration climate domain compared with correlative models (for example, Maxent), providing a pathway to more ecologically meaningful projections. Prior literature has also documented tundra shrub/tree encroachment, treeline advance, and potential Amazon dieback, highlighting ecosystem sensitivity to climatic shifts.

Study design: The authors transform ambient (1979–2013) and future (2061–2080; referred to as 2070) climatologies into phytoclimates that quantify climatic suitability for 14 plant growth forms, then derive indices of local phytoclimatic change, novelty, and disappearance. Species modelling: They fit the process-based TTR-SDM, an ecophysiological species distribution model based on Thornley’s transport resistance framework with a Farquhar-type photosynthesis module (distinct for C3 vs C4 plants). The model simulates monthly biomass accumulation constrained by minimum/mean/maximum temperature, soil moisture, solar radiation, atmospheric CO2 and soil nitrogen availability (nitrogen uptake in this application driven by temperature and soil moisture; soil nitrogen content assumed constant across grid cells). Species occurrence data: Vascular plant occurrences were obtained from the BIEN v4.1 database (non-public version), cleaned with CoordinateCleaner, deduplicated to 1 km, and linked to environmental data. Growth forms: 135,153 species were assigned to 14 growth forms (evergreen trees, dry-deciduous trees, cold-deciduous trees, needleleaf trees, evergreen shrubs, dry-deciduous shrubs, cold-deciduous shrubs, forbs, geophytes, therophytes, C3 grasses, C4 grasses, succulents, climbers) using BIEN, GIFT, specialized checklists (succulents; C3/C4 in grasses), and taxonomic rules; problematic categories (e.g., epiphytes, aquatic species, variable phenology for trees/shrubs) were excluded. Model fitting: For each species with ≥7 presences, the TTR-SDM parameters (n = 18) were inferred by maximizing the likelihood of a logistic regression (complementary log-log link), where the linear predictor is the log of simulated equilibrium biomass under monthly forcing. Pseudo-absences were sampled using a stratified scheme across 20 environmental zones (derived via clara clustering and DAPC on ambient forcing) with inverse weighting to reduce sampling bias. Optimization used Differential Evolution (DEoptim) for up to 1,000 iterations. Model evaluation: Models with TSS ≤ 0.7 were discarded. Final counts by growth form included, e.g., 24,362 evergreen trees; 439 needleleaf trees; 24,853 evergreen shrubs; 21,903 therophytes; 6,888 C3 grasses; 2,814 C4 grasses; 3,609 succulents; 9,990 climbers (total 135,153). Environmental forcing: Ambient monthly temperature and precipitation from CHELSA v1.2 (1979–2013), solar radiation from Global Aridity and PET Database v1, soil moisture model built by the authors using Hargreaves-type PET and soil field capacity/wilting point (IGBP-DIS). Ambient CO2 set to 338 ppm. Data were resampled to 1 km and projected to Eckert IV equal-area; species models were later projected to a 25 km grid for global analyses. Future projections: Ten downscaled CMIP5 climatologies for 2070 (2061–2080) from five GCMs (CCSM4, CNRM-CM5, FGOALS-g2, MIROC-ESM, MPI-ESM-LR) under RCP 2.6 and RCP 8.5; future soil moisture computed with the same model; solar radiation held constant; CO2 assumed 438 ppm (RCP 2.6) and 677 ppm (RCP 8.5). Phytoclimate construction: For each 25 km cell, the growth-form climatic suitability is the proportion of species of that growth form with predicted suitability above threshold (binary after optimizing true positives + true negatives). The phytoclimate is the 14-dimensional vector of these proportions. Phytoclimatic zones: Ambient cells were clustered via Gaussian mixtures (mclust, VEV model), selecting 18 clusters balancing interpretability and BIC performance; these zones reflect similarity in growth-form suitability (akin to bioclimatic zones). Risk indices and zone shifts: Multivariate Euclidean distances (ED) in growth-form suitability space were computed. Local change is ED between ambient and future for the same cell; novelty is the minimum ED between a future cell and any ambient cell; disappearance is the minimum ED between an ambient cell and any future cell. A threshold for ecological significance was set to the 5th percentile of inter-centroid distances among the 18 ambient zones. Future zone assignment maps each future cell to the ambient zone of its closest ambient analogue in phytoclimate space; cells beyond the threshold are designated novel. Summaries are reported as medians across the five GCMs for each RCP. Sensitivity analyses examined thresholds and uncertainty across GCMs; species subsampling showed stability of growth-form suitability surfaces. Spatial resolutions: Models fitted at 1 km; global projections and analyses at 25 km.

• By 2070, 33% (RCP 2.6) to 68% (RCP 8.5) of terrestrial land (excluding current ice on Greenland/Antarctica) will experience an ecologically significant change in phytoclimate (growth-form suitability profile).
• Hotspots of large local phytoclimatic change (median across five GCMs) include: south China mountains, Himalayas, northwestern Russia, Baltic countries, Scandinavia, southeastern and northeastern USA, Alaska, central Mexico, tropical Andes, southeastern South America, southeastern Australia, and northern New Zealand.
• Phytoclimatic zone shifts are widespread, strongest in temperate, boreal, and polar regions. Under RCP 8.5:
  • Zone 16 (tundra-supporting climates) shrinks by 72% (cannot shift poleward further).
  • Zone 18 (very continental cool-temperate) loses 54% of current extent, principally to zone 17 (less cold-limited continental cool-temperate, generally more suitable for most growth forms).
  • Cool-temperate and hemiboreal zones (13 and 14) and boreal (15) shift poleward.
• Tropical zones are comparatively more stationary in extent, but notable structural changes are projected:
  • Southeastern/eastern Amazon shifts toward zone 4, which supports savanna and drier forest ecosystems; this advance occurs under all GCMs but with variable magnitude.
• Novel phytoclimates are relatively rare: 0.3% of land under RCP 2.6 and 2.2% under RCP 8.5. Highest novelty indices occur in southeastern South America and Australia, plus tropical/subtropical mountain ranges (Andes, Himalayas, Sierra Madre Occidental). Many novel areas are currently in ambient zone 7 (high suitability for many growth forms) and are projected to gain suitability for evergreen and dry-deciduous trees and climbers, and lose suitability for needleleaf and cold-deciduous woody forms, likely due to slightly wetter conditions and hotter summers/milder winters.
• Disappearing phytoclimates are also rare: 0.1% (RCP 2.6) to 1.3% (RCP 8.5). Disappearance is concentrated in mesic subtropical eastern continental margins. Spatial overlap between novelty and disappearance is limited: only about 25% of cells with disappearing phytoclimates also become novel by 2070; 75% transition to a phytoclimate found elsewhere today.
• Contrary to climate-variable-based novelty studies emphasizing tropical temperature-driven novelty, equatorial regions in this phytoclimate framework see generalized declines in suitability across growth forms, making them converge toward existing drought-adapted and seasonal phytoclimatic zones rather than becoming functionally novel.
• The spatial pattern of change, novelty, and disappearance in phytoclimates differs markedly from patterns derived from raw climate exposure indices, implying different conservation priority areas.
• Overall, results indicate a profound, ongoing transformation in climatic support for terrestrial ecosystem structure, with substantially milder changes under RCP 2.6 compared to RCP 8.5.

By transforming climates into physiologically grounded phytoclimates, the study directly links climate change to potential structural changes in terrestrial ecosystems, addressing the limitations of exposure-only indices. The findings suggest that large portions of the planet—especially high latitudes—will experience strong shifts in the climatic support for plant growth forms, consistent with observed shrub/tree encroachment and treeline advance. The analysis also reveals limited functional novelty in many regions previously flagged as climatically novel, indicating that future conditions may resemble existing phytoclimates elsewhere, reframing expectations for ecosystem change and management. Conservation and restoration strategies should pivot from preserving static baseline states toward managing dynamic trajectories aligned with climatic forcing (for example, adopting Resist–Accept–Direct frameworks, facilitating species movements, improving landscape connectivity, and considering assisted migration). Where novel phytoclimates do emerge, managers face deep uncertainties due to lack of analogues, implying the need for enhanced monitoring, experimentation, and adaptive management. Regions with disappearing phytoclimates are high-risk for biodiversity loss because their associated ecosystem states may not be restorable elsewhere, underscoring priority areas for conservation action. Differences from previous climate-exposure maps highlight that ecophysiologically informed assessments can redefine priority regions and reduce misinterpretation risks inherent in raw exposure metrics.

This study introduces a scalable, ecophysiologically informed framework that transforms ambient and future climates into phytoclimates describing the climatic support for 14 growth forms, thereby reducing the gap between climate projections and ecological interpretation. It forecasts substantial global reorganization of phytoclimates by 2070, with limited but nontrivial emergence of novel and disappearing phytoclimates and pronounced poleward shifts in high-latitude zones. The results redefine conservation priorities relative to traditional exposure-based approaches and advocate for management strategies that facilitate ecosystem transitions along climate-forced trajectories. Critically, projected changes are markedly less severe under RCP 2.6 than RCP 8.5, reinforcing the imperative to cut greenhouse gas emissions to mitigate risks to biodiversity, ecosystem functioning, and land-based production systems. Future research should refine high-resolution projections to capture micro-phytoclimates, systematically quantify uncertainty from species models and data (occurrence, trait-based growth-form assignments), and integrate biotic interactions, disturbance regimes, and dispersal dynamics to better anticipate realized ecosystem responses.

Key limitations include: (1) Spatial resolution: global analyses at 25 km obscure fine-scale topographic and microclimatic heterogeneity (for example, alpine tundra in the Alps); higher-resolution projections could reveal micro-phytoclimates and local refugia. (2) Model structural dependence: projections rely on the process-based TTR-SDM and its assumptions (for example, trapezoidal/saturating response shapes, monthly time step, universal C3/C4 photosynthesis parameterization), and alternative species models were not systematically compared for this task. (3) Forcing assumptions: solar radiation held constant for 2070; soil nitrogen content assumed spatially uniform; soil moisture modeled using a simplified PET approach; such assumptions may affect suitability estimates. (4) Species data and classification uncertainty: occurrence data biases, thresholds for binary suitability, and potential errors in growth-form assignments (species/genus/family-level imputation) introduce uncertainty. (5) Uncertainty exploration focused on RCPs and five GCMs; parameter and structural uncertainty at the species model level was not propagated to final maps. (6) Ecological processes beyond physiological suitability—biotic interactions, dispersal limitations, disturbances, legacy effects, and microclimate buffering—were not explicitly modeled and will modulate realized ecosystem change and lags.