Reef Check data from 1998 to 2019 do not show a higher coral bleaching threshold after 2006: SST rose equally at surveys without bleaching, and estimates change sign with the split year

Question: Sully et al. (2019) reported that coral bleaching started at a 0.5 to 0.6 °C higher sea surface temperature (SST) in 2007-2017 than in 1998-2006. Their statistic was the mean SST of Reef Check surveys with bleaching. Does this statistic show a change in the bleaching threshold? Approach: We used the table that the authors released (8,801 surveys, 3,597 sites) and the Reef Check rows of the Global Coral-Bleaching Database (10,749 surveys, 1998-2019). Surveys without bleaching were a negative control. We used binomial models with SST, degree heating weeks, period and ecoregion, and conditional logit models within sites. We changed the split year from 2004 to 2014, the ocean region and the bleaching definition (>0%, >=1%, >=10%). Results: Mean SST increased by +0.64 °C (95% site-bootstrap interval 0.46 to 0.81) at bleached surveys and by +0.63 °C (0.50 to 0.75) at surveys without bleaching. The difference is +0.01 °C (-0.18 to 0.21). Site mix, survey season and weekly anomalies caused the increase. The released likelihood-ratio test compares likelihoods of different data. A correct version also rejects for surveys without bleaching (p about 1e-42), and it rejected in 99.9% of 1,000 simulations with no change in coral response. At equal SST, the across-site threshold shift for the published split is +0.04 °C (-0.65 to 0.74). For the 11 split years, no interval is above zero and 7 are below zero (more bleaching later). The within-site shift is +1.18 °C (0.29 to 2.08) for 2007, but -0.29 °C when the 2006 surveys are removed, and 5 of 11 split years give intervals below zero. Atlantic sites bleached less after 2006 (+1.99 °C) and Indo-Pacific sites bleached more (-1.83 °C). Severe bleaching (>=10%) gives +1.39 °C (0.11 to 2.67) only for the 2007 split. Conclusion: The published increase occurs equally at surveys without bleaching. Thus, it does not show a higher threshold. At equal stress, the sign of the change depends on which years are in each period. Limitations: Reef Check surveys are not a random sample, 4 km CoRTAD SST does not show local stress, and our models cannot separate changes in coral community (ecological memory) from changes in sampling. The data do not show that thresholds stayed constant. They show that these designs do not detect a robust change.

coral bleaching · thermal threshold · Reef Check · Global Coral-Bleaching Database · sea surface temperature · degree heating weeks · reanalysis · negative control · selection on outcome · CoRTAD · climate adaptation · coral reefs