Most coral surveys above the site training heat range stayed below four degree heating weeks in two oceans in 2011-2019

We tested whether heat-model prediction gain changed above the heat range recorded at each reef during training. Fixed models trained in 2003-2010 used site history, month and depth, with or without heat. Tests used 992 later Reef Check surveys in the Atlantic and Pacific. Of 105 surveys above the site training heat maximum, 80 had less than four degree heating weeks. Fifty-eight had a site training maximum of zero. Thus a fixed high-heat count does not measure local model input support. In both Pacific periods, the heat model had a larger mean loss reduction above range. Atlantic contrasts had opposite signs. All four adjusted intervals included zero. Stricter heat range rules and three model checks kept the signs but did not establish a clear contrast. Few above-range events, selected visits and shared source records limit inference. The method audits model support; it does not identify a causal heat effect. Code, aggregate evidence, source hashes and an independent calculation check are included.

coral bleaching · thermal stress · prediction · training range · observation bias · temporal validation · null result