Climatic and tectonic drivers shaped the tropical distribution of coral reefs

Despite covering less than 0.1% of the oceans, warm-water coral reefs support exceptionally high marine biodiversity and are presently confined to approximately 34°N–32°S, primarily limited by minimum sea surface temperatures around 18 °C. This latitudinal restriction also structures the diversity of reef-associated organisms, producing a steep latitudinal biodiversity gradient concentrated in tropical and subtropical regions, with a major hotspot in the Indo-Australian Archipelago. Fossil evidence indicates that coral reefs previously extended to higher latitudes, suggesting less tropical skew in past intervals, but the drivers of these shifts—abiotic (climate) versus biotic (competition, adaptation)—remain debated. Given modern thermal tolerances, past poleward extensions are often attributed to expanded tropical/subtropical climates; however, plate tectonics also modulate shallow marine substrate and continental configuration, potentially constraining reef development regardless of climate if suitable benthic substrates are lacking. The study asks to what extent climate and the distribution of shallow marine area (palaeogeography) explain the observed fossil distribution of coral reefs, and how these drivers shaped the spatiotemporal distribution of climatically suitable habitat and, by extension, the latitudinal biodiversity gradient of reef-associated taxa. Modern correlations between reef habitat area and biodiversity, and the role of ancient reefs as biodiversity cradles, imply that historical changes in reef habitat could have structured deep-time biodiversity patterns. Understanding these dynamics is urgent as reefs face rapid warming, bleaching, and ocean acidification, with projected CO2 and temperature increases posing severe risks. The fossil record can provide empirical insight, but pre-Quaternary correlations between reefs and abiotic controls have been weak, likely due to patchy sampling and strong Northern Hemisphere bias, as well as socioeconomic sampling effects. Proxy-based palaeotemperature reconstructions share similar biases, complicating inference about climatic control. To address these challenges, this study integrates habitat suitability modelling with Earth system simulations and the ~247 Myr fossil record of scleractinian reefs to test whether modern climatic tolerances predict past reef distributions, evaluate model performance against fossil localities, and infer latitudinal shifts in suitable habitat and their implications for reef-associated biodiversity.

Prior work establishes that modern warm-water reefs are thermally constrained (~18 °C minimum SST) and concentrated in the tropics/subtropics, with peak richness in the Indo-Australian Archipelago. Fossil records show that reefs extended to higher paleolatitudes in certain intervals (e.g., an Eocene reef at ~46°N), suggesting past climates allowed broader distributions. Studies have proposed both abiotic factors (global climate states, sea level, temperature) and biotic factors (competition, adaptation) as drivers of these shifts. Plate tectonics are widely recognized as a fundamental driver of biodiversity over geological timescales through changes in continental configuration and shallow marine substrate, influencing reef and associated biota distributions. However, pre-Quaternary analyses often failed to find strong correlations between reef distribution and abiotic controls, likely due to sampling biases (Northern Hemisphere concentration, wealth-related sampling intensity) and uneven proxy data coverage. The literature also highlights that late Paleogene onset of ‘icehouse’ conditions may have steepened the modern latitudinal biodiversity gradient. Tectonic reconfiguration, notably the formation of the Indo-Australian Archipelago via Australian plate convergence with Eurasian and Philippine plates, is implicated in creating extensive shallow tropical substrates and fostering a biodiversity hotspot. Supercontinent phases (e.g., Pangaea) reduced shelf area and may have limited habitat for climatically sensitive taxa. These insights motivate an approach that combines climatic niche modelling with palaeogeographic constraints to mitigate fossil record biases and test climatic versus tectonic controls.

Data and study design:

• Present-day reef occurrences: Downloaded from ReefBase (>10,000 points), filtered to include only ‘true reefs’ (excluding non-reef coral communities), removed records on land or >200 m depth, and spatially subsampled to 1° × 1° grid to match climatic layers, yielding 790 calibration presences.
• Climate variables: Chosen as four predictors known to constrain modern reef distribution and reconstructable in deep time—mean maximum and mean minimum sea surface temperature (SST), and mean maximum and mean minimum insolation. Simulations used the HadCM3BL-M2.1aE (HadCM3L) coupled AOGCM with 2.5° × 3.75° horizontal resolution (19 atmospheric and 20 ocean vertical levels). Stage-level simulations span Anisian (Middle Triassic) to Piacenzian (late Neogene). Idealised atmospheric CO2: 1120 ppm (Triassic–Eocene), 560 ppm (Oligocene), 400 ppm (Miocene–early Pliocene), 280 ppm (late Pliocene and pre-industrial), each run 1422 years to near-surface equilibrium. Climate outputs downscaled to 1° × 1° via bilinear interpolation.
• Shallow-water mask: All climate layers clipped to <200 m substrate depth (approximating photic zone) to avoid inflating validation metrics. Bathymetry from Getech stratigraphic stage-level DEMs (0.5° × 0.5°), aggregated to 1° × 1° preserving minimum depth to retain oceanic islands. Missing climate values within masks imputed from 3 × 3 neighbourhood means.

Habitat suitability modelling (HSM):

• Algorithm: MaxEnt v3.4.4, presence-background modelling. Features restricted to linear and quadratic to obtain realistic response curves; clamping disabled; extrapolation allowed (reefs at upper thermal limit today). Background: 10,000 points sampled from shallow-water mask. Model trained with 100 bootstrap replicates using 85% of occurrences for training and 15% for testing; max iterations 5000; logistic output.
• Validation: Assessed with AUC (ROC-based discrimination) and continuous Boyce index (consistency against random expectations along prediction gradient).
• Hindcasting and thresholds: Modern-trained model projected to each geological stage’s climate and shallow-water mask. Median suitability grids converted to binary presence/absence with two thresholds: Least Training Presence (LTP) and MaxSSS (maximising sum of sensitivity and specificity). Multivariate Environmental Similarity Surface (MESS) analyses identified novel environmental conditions relative to calibration.

Fossil data and evaluation:

• Fossil reef occurrences compiled from PaleoReef Database (PARED) and Paleobiology Database (PBDB). Filters removed non-coral reefs, cold-water reefs, non-‘true reefs’, subsurface seismic-only occurrences, and PBDB records inconsistent with PARED filtering (e.g., perireef/subreef contexts, unsuitable lithologies). Occurrences were binned to stratigraphic stages, palaeorotated using Getech plate rotations, subsampled to 1° × 1°, and clipped to study bounds, yielding 535 unique fossil reef localities. Temporally unconstrained data were assigned to a stage if >50% of the occurrence’s age range overlapped that stage.
• Predictive performance testing: For each stage, intersected fossil localities with binary hindcasts (with and without a one-cell queen-move buffer). To control for the effect of total suitable area, generated 1000 random point sets per stage equal in size to fossil counts, computed the mean percentage intersecting suitable habitat, and compared with fossil predictive success using one-sample (one-sided) Wilcoxon rank-sum tests (α = 0.05). Also compared continuous suitability values at fossil localities vs. random points via two-sample (one-sided) Wilcoxon rank-sum tests per stage.

Spatiotemporal analyses:

• Computed area-weighted palaeolatitudinal centroids of suitable habitat per hemisphere from binary maps, correcting for variable cell area. Defined ‘reef zone’ as the most poleward suitable cell in each hemisphere per stage. Calculated global and 20° palaeolatitudinal bin areas of suitable habitat through time. Tested relationship between global suitable habitat area and number of fossil reef sites via OLS regression.

Reproducibility:

• Data, code, and simulations available via Zenodo and GitHub; climate model simulations accessible from the University of Bristol repository.

• Model performance: Modern-trained HSM achieved mean AUC = 0.856 (SD 0.003) and Boyce index = 0.988 (SD 0.006), indicating strong discrimination. Stage-level hindcasts correctly predicted ~60–87% (SD ~19–29%) of fossil reef localities across stages (threshold-dependent); with a one-cell buffer, ~62–90% (SD ~17–28%). Random expectation was ~3–8% intersection across stages, significantly lower than fossil predictive success (Wilcoxon one-sided P < 0.001 for nearly all stages; exceptions noted for Bartonian under both thresholds and Hettangian, Sinemurian, Bathonian, Selandian under MaxSSS P > 0.05). In 37/45 stages, continuous suitability at fossil sites exceeded random (P < 0.05).
• Error analysis and novelty: 71 of 73 false negatives were associated with minimum SST < 18 °C. MESS indicated 80 fossil localities in novel environments; 62 intersected mean maximum SST above present-day ocean upper thermal limits (33.8–35.6 °C), implying broader realized climatic space in some past intervals.
• Latitudinal shifts: Northern Hemisphere centroid shifted equatorward from ~24–29° (Sinemurian) to ~13–17° (Piacenzian). Both hemispheres show ~5–6° equatorward shift since the Priabonian (~late Eocene). Northern shift driven by ~54–72% decline in suitable habitat at 30–50°N due to cooling SSTs; Southern shift driven by increased suitable area in southern tropics/subtropics from expanded shallow marine substrate linked to plate reconfiguration.
• Habitat area through time: Suitable habitat was less tropically skewed through much of the Mesozoic–early Paleogene, with more available high-latitude (30–50°N) habitat than today. Notable reduction across the Jurassic/Cretaceous (J/K) boundary: 37% reduction in shallow marine area and decline of suitable habitat at 30–50°N; likely tied to global sea-level fall and interpreted brief ‘cold snaps’. Global suitable habitat peaked in the Late Jurassic and Late Cretaceous–Eocene; major declines occurred across the J/K boundary and from the Priabonian onward, primarily due to loss at higher northern latitudes.
• Reef zone extent: Maximum palaeolatitudinal extent during Bartonian (LTP threshold): 46.5°N to 49.5°S; under MaxSSS, Albian maximum: 36.5°N to 43.5°S. Minimum extent during Piacenzian: 29.5–37.5°N to 36.5–45.5°S. Overall reef zone contracted by ~11–12° from Norian to Piacenzian. Most fossil localities (~73–92%) fall within the estimated reef zone, with some high-latitude exceptions.
• Biodiversity proxy: OLS shows no significant relationship between number of fossil reef sites and global suitable habitat area (R² = 0.000–0.001, P = 0.878–0.979), likely reflecting sampling biases and non-climatic controls.
• Implications: Late Paleogene onward saw increasing tropical skew due to combined cooling and increased tropical shallow substrate from Indo-Australian Archipelago tectonics, likely steepening modern latitudinal biodiversity gradients of reef-associated taxa.

The findings demonstrate that warm-water coral reefs largely tracked large-scale latitudinal shifts in tropical/subtropical climates through the Mesozoic and Cenozoic, with past distributions less tropical-skewed than today. An exception is the earliest Cretaceous, where a pronounced reduction in shallow marine area across the J/K boundary and inferred brief cooling episodes led to diminished high-latitude suitable habitats, likely affecting marine ecosystem turnover. From the late Paleogene (~37 Ma), suitable habitat became increasingly concentrated in the tropics due to two intertwined processes: (1) global cooling reduced temperate (30–50°N) SSTs below reef tolerances, contracting northern suitable areas; (2) tectonic evolution, particularly the equatorward motion and convergence of the Australian Plate with Eurasian and Philippine plates, expanded shallow tropical substrates, contributing to the Indo-Australian Archipelago’s development as a biodiversity hotspot. These drivers collectively likely steepened the latitudinal biodiversity gradient of reef-associated taxa. The model’s high predictive success against fossil occurrences indicates that modern climatic tolerances, coupled with palaeogeographic constraints, can explain much of the fossil distribution despite sampling biases. Instances where fossil reefs occur at minimum SST <18 °C and maximum SST >33.8–35.6 °C suggest that reefs occupied a broader climatic space in some past greenhouse intervals. However, adaptation to such conditions developed over millions of years, cautioning against direct analogs for the rapid anthropogenic warming underway. Although warming could theoretically permit future poleward range expansions, the current rate and additional stressors (e.g., acidification, nutrient dynamics) likely preclude reefs from keeping pace. Overall, the results underscore the joint roles of climate state transitions and plate tectonics in shaping long-term reef habitat distributions, providing a framework to interpret deep-time biodiversity patterns and to anticipate spatial shifts under ongoing climate change.

This study integrates habitat suitability modelling, Earth system palaeoclimate simulations, and a global fossil reef dataset to show that climate and palaeogeography jointly governed the distribution of warm-water coral reefs over the past ~247 Myr. Reefs were less tropically skewed through much of the Mesozoic–early Paleogene, with significant contractions in high-latitude suitable areas across the J/K boundary and from the late Paleogene as Earth entered an ‘icehouse’ climate and tropical shallow substrates expanded via Indo-Australian Archipelago tectonics. The modern-trained HSM predicts fossil reef occurrences with high fidelity, validating the approach and suggesting potential for targeted fossil prospecting. While future warming might open higher-latitude habitats, the rapid rate of anthropogenic change and other stressors make it unlikely that reef ecosystems can adapt and expand quickly enough. Future work should incorporate ensembles of palaeoclimate models, refine CO2 histories and within-stage variability, increase spatial resolution where feasible, improve fossil sampling and age constraints, and include additional ecological variables (e.g., nutrients, aragonite saturation, upwelling intensity) to further constrain deep-time reef distributions and better link habitat to biodiversity dynamics.

• Idealised atmospheric CO2: Constant or stepwise values (e.g., 1120 ppm Triassic–Eocene) do not capture within-stage variability, potentially smoothing climatic transitions.
• Single climate model: Reliance on HadCM3L; inter-model geographic differences could affect hindcasts. An ensemble would better capture uncertainty but was unavailable for the full interval.
• Spatial resolution: 1° × 1° analyses (climate downscaled from 2.5° × 3.75°) may miss fine-scale bathymetry, local SST variations, and microhabitats; within-cell bathymetric heterogeneity may bias suitable area estimates.
• Sampling limitations: Some stages have few fossil reef localities, limiting robust stage-specific validation; fossil record is spatially biased (notably towards the Northern Hemisphere and wealthier regions).
• Model assumptions: Assumes accessibility and occupancy of all climatically suitable areas; does not model dispersal barriers, ocean circulation constraints, or seaway connectivity explicitly.
• Omitted/limited variables: Key ecological drivers like nutrient concentrations, aragonite saturation state, siliciclastic input, and detailed upwelling nutrient effects are not included (thermal upwelling resolved but nutrient consequences not), potentially overestimating suitability in regions like the West African coast.
• Potential climate biases: Possible mid/high-latitude cold bias in palaeoclimate simulations may underestimate temperate SSTs in some intervals (e.g., last 37 Myr).
• Threshold sensitivity: Binary presence/absence results depend on threshold choice (LTP vs. MaxSSS), introducing uncertainty in estimated extents.