Systematic review of the uncertainty of coral reef futures under climate change

Anthropogenic climate change is anticipated to push Earth system components beyond critical climate tipping points (CTPs), with warm-water coral reefs listed as a regional tipping element likely to be exceeded if warming surpasses 1.5 °C. Reefs are highly biodiverse and crucial for food security, coastal protection, and livelihoods, yet are experiencing more frequent, intense, and widespread bleaching driven by ocean warming. Recent syntheses, using the IPCC confidence framework, identify a coral reef CTP at ~1.5 °C (1–2 °C) and rely heavily on 'excess heat' threshold models (degree heating weeks/months applied to SSTs) to estimate bleaching frequency and the proportion of reefs at risk. While early projections largely used threshold approaches, their predictive performance varies and later studies developed alternative methods (species distribution models, eco-evolutionary, and population dynamics). Influential assessments have largely overlooked these developments. This study conducts a systematic review of published projections of coral reef futures under climate change, evaluates the landscape of modeling approaches, identifies key uncertainties, and proposes pathways—drawing from climate science—for incorporating uncertainty and coordinating modeling efforts toward probabilistic projections.

The review identified 79 studies (1999–2023) modeling coral reef responses to future climate change, with 59% at regional scales and 41% at global scales. Five major methodologies encompassed 76 of 79 articles: (1) 'excess heat' threshold models (32% of studies; 68% of cumulative citations), forecasting bleaching risk based on degree heating weeks/months and event frequency but rarely assessing biological/ecological processes directly; (2) population dynamic models (23%), process-based simulations of recruitment, growth, mortality, herbivory, predation, algal–coral interactions, and connectivity, often regionally focused due to data demands and attracting 13% of citations; (3) species distribution models (SDMs; 23%), projecting habitat suitability changes using environmental correlates (commonly SST and aragonite saturation), cost-effective and scalable but limited by reliance on mean conditions, assumptions about environment-driven distributions, and limited accounting for dispersal and evolution (<7% of citations); (4) eco-evolutionary models (12%), simulating interactions, adaptation, and dispersal with increasing emphasis on heat-tolerant symbionts and larval migration, but constrained by scarce trait/genetic data; (5) projective meta-analyses (5%), aggregating experimental responses to warming and acidification to parameterize projections of biological processes, with limitations from predominantly short-term experiments (<3% of citations). Heat stress modeling techniques split into thermal threshold versus continuous-variable approaches: overall 53% of studies used threshold techniques; by method the proportions were approximately: 'excess heat' models 95% threshold, population dynamics 40% threshold/60% continuous, SDMs 20% threshold/80% continuous, eco-evolutionary 50%/50%, and meta-analyses 20%/80%. The literature shows inconsistent reporting of outputs and scenarios, frequent coarse spatial resolutions, geographic biases (e.g., more focus on eastern Australia and the Caribbean, paucity in the eastern Pacific, western Atlantic Brazil, Indian Ocean, Arabian Seas), and limited inclusion of multiple stressors beyond warming.

The review followed PRISMA guidelines. A Web of Science search (March 6, 2023) retrieved 2705 peer-reviewed articles; screening titles, abstracts, and displays identified 2073 potentially eligible articles. Inclusion criteria: (1) projections of tropical/sub-tropical coral reef responses to future warming alone or with other drivers; (2) stated future emissions pathways and/or warming scenarios; (3) projections across more than one reef site. The final database comprised 79 studies (1999–2023). Key study characteristics were extracted: focal variables, model inputs, spatial scale, geographic area, methodological category, heat-stress modeling technique (threshold vs continuous), purpose, assumptions, spatial resolution, and downscaling application. Models were classified into five categories: 'excess heat' threshold models, population dynamics, SDMs, eco-evolutionary, and projective meta-analyses; a few were categorized as other. For an exploratory quantitative synthesis, the authors targeted three common outputs: fraction of reef cells at risk of severe/recurrent bleaching (or long-term degradation), changes in coral cover, and fractional habitat suitability. Studies included in the meta-analysis had to provide extractable projection estimates and uncertainty, end-of-century projections (2090–2100), and a baseline between 2000–2015 (0.86–0.96 °C warming). When necessary, values were digitized from figures. Eight studies (39 scenarios) met criteria. Effect sizes (Hedges’ g) and variances were computed comparing end-of-century to baseline, with direction standardized (negative denotes adverse ecological outcomes), using pooled standard deviations and small-sample bias correction. Calculations used R (v4.3.0) and the metafor package (v4.2-0).

- Five main modeling approaches dominate projections; 'excess heat' threshold models make up 32% of studies but receive 68% of citations, indicating outsized influence relative to diversity of methods. - More than half (53%) of studies use thermal threshold techniques; DHW/DHM metrics are prevalent despite evidence that other marine heatwave characteristics (e.g., peak temperatures, temperature variability, cool-duration, bimodality) often better predict bleaching severity; in one analysis DHW explained only 9% of variance. - Only one study directly compared threshold vs continuous-variable techniques; DHW-based projections yielded more severe declines in coral cover than a multivariate approach integrating acute and chronic stressors in the Indian Ocean. - Threshold choice and local adaptation matter: varying DHW algorithms and intra-population heat tolerance differences can shift projected timing of annual bleaching by up to 17 years under SSP2-4.5. - Deterministic modeling dominates, limiting formal uncertainty quantification. The authors recommend uncertainty incorporation via Monte Carlo and sensitivity analyses, and advocate multi-model ensemble approaches (analogous to CMIP) to derive probabilistic projections. - Reporting gaps hinder synthesis: 89% of studies did not report basic output metrics and/or extractable variation in common units; outputs often expressed as fractions of reef cells at risk or habitat suitability, with coral cover projections in only 29% of studies. - Emissions scenarios: RCP8.5 (CMIP5) is most used, typically paired with a lower-forcing scenario; recent work suggests RCP8.5 is an unsuitable baseline high-emission reference, underscoring need for coordinated adoption of AR6 SSPs (including 1.9, 3.4, 7.0 W m−2). - Spatial scale: 49% of studies operate at resolutions coarser than 0.25° (~770 km² at equator); downscaling used in 19 studies (85% statistical). A Caribbean comparison showed dynamical downscaling detected earlier severe bleaching onset via current changes, while statistical methods missed local features (eddies). - Geographic bias: Eastern Australia and Caribbean have diverse, numerous regional models; eastern Pacific, western Atlantic (Brazil), Indian Ocean, and Arabian Seas lack regional models and often rely solely on SDMs. - Multi-stressor modeling is limited: 76% of studies considered warming alone or with only one additional stressor. Pollution featured in 16 studies; fishing in 4; disease in 2; pest species in 1. Local stressors (nutrients, fishing) are known to modulate heatwave susceptibility and are manageable. - Exploratory effect-size synthesis (8 studies, 39 scenarios) shows most projections are negative; thermal-threshold studies tend to produce more negative effect sizes and, under >4 °C, commonly project >93% of global reef cells at risk by century’s end. Continuous-variable approaches yield more variable, generally less severe effect sizes, reflecting regional heterogeneity. - Comparison to IPCC: AR6 cites threshold-based studies (Schleussner 2016; Frieler 2013) projecting 70–90% coral reef decline at 1.5 °C and >99% at 2 °C; these effect sizes are as severe as alternative-method projections under >4 °C in this review, suggesting syntheses based on a narrow subset may overstate severity relative to broader methodologies.

Findings indicate that a narrow set of 'excess heat' threshold models has disproportionately shaped prevailing assessments of coral reef futures, potentially biasing conclusions toward more severe outcomes compared to alternative methods that consider broader stress regimes and ecological processes. The heterogeneity of model outputs, emissions scenarios, and spatial scales, combined with limited reporting of uncertainty, hampers robust cross-study synthesis and probabilistic inference. Incorporating uncertainty into deterministic models via Monte Carlo and sensitivity analyses can quantify parameter/input-driven uncertainty, while a coordinated multi-model ensemble framework—common scenarios, shared output metrics, and standardized reporting—can address system and structural uncertainties and yield probabilistic projections. Improving ecological relevance requires aligning modeled outputs with observable metrics (e.g., coral cover alongside community composition and functional indicators), selecting emissions pathways consistent with AR6 SSPs, increasing spatial resolution (or applying tested downscaling, preferably dynamical where feasible), and expanding regional coverage to address geographic gaps. Including multiple, interacting stressors and management interventions is essential for decision-relevant projections, given the modulating influence of local stressors and the policy focus on restoration.

This systematic review maps the methodological diversity and uncertainty in coral reef climate projections, revealing the dominance and limitations of threshold-based approaches and the underutilization of models capturing ecological, evolutionary, and multi-stressor dynamics. The study contributes an initial cross-method synthesis and an exploratory effect-size comparison, highlighting discrepancies that suggest widely cited threshold studies may project more severe impacts than other approaches. The authors propose practical steps to improve projections: adopt uncertainty analyses in deterministic models; coordinate a multi-model ensemble with standardized emissions scenarios (AR6 SSPs), common and observable output metrics, and open reporting of outputs and uncertainties; enhance spatial resolution via improved SST products and validated downscaling; address geographic biases; and integrate multi-stressor and intervention scenarios (including restoration and assisted evolution). These steps can generate probabilistic, decision-relevant projections to better guide coral reef management and policy under climate change.

Synthesis was constrained by heterogeneity in model outputs, emissions scenarios, and reporting: 89% of studies did not provide extractable uncertainty metrics in common units, limiting quantitative comparison. The exploratory meta-analysis included only 8 studies (39 scenarios), so effect-size comparisons are not definitive. Many models are deterministic, impeding formal uncertainty estimation. Spatial resolutions are often coarse (>0.25°), and downscaling introduces additional uncertainty and potential biases. Geographic coverage is uneven, with underrepresentation in several major reef provinces. Multi-stressor interactions and management interventions are sparsely modeled, limiting relevance for local decision-making. Data and parameter limitations, particularly for eco-evolutionary processes, constrain generality beyond well-studied taxa and regions.