What this research found
Some reefs bleach far less than their heat exposure would predict, and others far more. Four decades of global bleaching surveys were joined to satellite sea-surface temperatures, a fractional logistic model was fitted to estimate how much bleaching each survey's pre-visit heat exposure should have produced, and 1,223 repeatedly surveyed reef sites were ranked by how far they departed from that expectation. Thirty resilient and thirty sensitive candidates emerged, 28 of which survive false-discovery correction, with the resilient ones concentrated in the central-equatorial Pacific.
- Heat exposure predicts bleaching but leaves plenty unexplained. The model reached a McFadden pseudo-R-squared of 0.102, and ten-fold cross-validation holding out whole reef sites gave a mean absolute error of 15.02% bleaching and a root-mean-square error of 21.69%.
- Cumulative heat is the dominant driver. Fitted one at a time, all three exposure metrics were positively and significantly associated with bleaching: odds ratio 1.91 for cumulative positive anomaly (p = 9e-104), 2.00 for event duration (p = 5e-61), and 1.39 for maximum anomaly (p = 2e-31).
- Of 1,223 repeat-survey sites, the 30 most negative and 30 most positive mean residuals were nominated as resilient and sensitive candidates, and 28 pass Benjamini-Hochberg correction at q below 0.05 (11 resilient, 17 sensitive), meaning the anomaly recurs across repeat visits rather than resting on one survey.
- Resilient reefs cluster in the central-equatorial Pacific around the Phoenix and Line Islands, consistent with upwelling and advective cooling. One site near 2.8 degrees south, 171.7 degrees west bleached 11.6% where the model expected 47.1%.
- The most extreme sensitive sites form a tight cluster in the Ryukyu and Yaeyama lagoons of Japan near 24.2 degrees north, where observed bleaching reached roughly 100% against about 4% predicted, pointing to sub-grid heat stress that a quarter-degree satellite product cannot see.
How it was done
Bleaching survey records were paired with NOAA's daily quarter-degree sea-surface temperature product. Reef identity was resolved by DBSCAN clustering at 1 km on the sphere rather than by rounding coordinates, which mitigates GPS drift and yielded 1,358 repeat-survey sites across 15,600 records. For each survey, temperature anomalies against a 1982-2011 daily climatology with 15-day circular smoothing were accumulated over the 84 days before the visit into three metrics: cumulative positive anomaly in degree-days, maximum anomaly, and event duration. A fractional logistic regression with region and year fixed effects, fitted on the 12,296 records carrying an observed bleaching percentage, supplied expected values, and site-level mean residuals were tested against zero. The write-up is an 18-page manuscript with 9 figures and 37 references that closes with a validation plan using in-situ temperature loggers, symbiont profiling, and standardised heat-tolerance assays.
Data sources
- Global Coral Bleaching Database, NCEI accession 0228498 version 1.1 (DOI 10.25921/9xfa-0v97), 33,244 survey records from 1963 to 2021
- NOAA OISST version 2.1 daily sea-surface temperature, 0.25 degree resolution, 1982-2021, retrieved as 1,840 box-year subsets
Limitations
Roughly 43% of survey records lack a full calendar date and were excluded, because a daily 84-day exposure window cannot be defined without day resolution, and only 12,296 of the 15,600 retained records carry an observed bleaching percentage. Because the satellite product averages over quarter-degree cells, an apparent hypersensitivity may reflect unresolved local heat rather than coral biology, which is what the proposed field validation is designed to separate.
Figures from this analysis
How this research was produced
K-Dense Web planned and ran this environment investigation end to end — gathering the sources, carrying out the analysis, producing the figures, and drafting the report. The full session transcript, including every intermediate step, is available to view.
