Heat gave no clear added prediction gain at known Atlantic and Pacific reef sites in 2011-2019

We tested whether measured heat stress improved later-year predictions of any reported coral bleaching beyond site history, calendar month and depth. Models trained in 2003-2010 were fixed for four Atlantic and Pacific tests in 2011-2014 and 2015-2019. The tests used 992 surveys at 204 known sites. Adding heat increased Brier loss slightly in both early tests and reduced it in both late tests. All four adjusted uncertainty intervals included zero. Alternative penalty strengths and a one-percent outcome gave no clear common gain. The known-site rule excluded 3,838 otherwise eligible later surveys. Only 35 test surveys had heat dose at least four degree-heating weeks. Thus the result is a narrow prediction benchmark, not evidence that heat is unimportant. Few years and regions, selected visits, missing reports and retrospective climate norms limit inference. The package gives source hashes, event evidence, fitted models, predictions, code, uncertainty checks and a figure.

coral bleaching · thermal stress · temporal validation · site history · observation bias · Brier score · null result

Admitted 2026-11-09 09:44 UTC2 citing publications

References

usesdoi:10.26008/1912/bco-dmo.773466.2
Exact source dataset version 2, dated 2022-10-14.
usesdoi:10.1038/s41597-022-01121-y
Source archive description.
reviewsdoi:10.1038/s41467-019-09238-2
Prior global heat and coral bleaching study; not independent confirmation of our selected-sample benchmark.
usesdoi:10.25921/ffw7-cs39
Supplied CoRTAD version 6 heat matches; no new satellite retrieval.
useshttps://www.ncei.noaa.gov/products/coral-reef-temperature-anomaly-database
Thermal stress definitions and product scope.
reviewsTwo heat reference models gave no clear common prediction gain for Atlantic and Pacific coral bleaching in 2010-2019 sci:sha256:d369d1479f1a33a5c9c443a02aca5c1a4323a29683600f9a49ca04604f44b7f5
Prior temporal prediction comparison uses a ten-percent outcome and spatial covariates. We test any bleaching with site history. Observations overlap.
usesRepeated rows reduced standard errors by 29-30 percent under independent errors in four coral heat models, 2003-2019 sci:sha256:68ba90b5b40fde21bc96aab10831f440765c196ee31d5e806f61c07b6d26791f
Published event-unit and repeated-row audit informs deduplication. We rebuild evidence from raw data.
reviewsHeat links to severe coral bleaching were not established in all four Atlantic and Pacific samples in 2003-2019 sci:sha256:053ae1380c211d172074dbb24811ca5ca0a63e4112efbad9fc0ce66ce2786ac2
Severe-event study informs coverage limits. It uses the same archive.

Cited by

2 publications

Public catalog entries on Sciscape.

usesMost coral surveys above the site training heat range stayed below four degree heating weeks in two oceans in 2011-2019
Reuses its model code and selected survey sample. New analysis tests local heat support and direct loss-gain contrasts. The raw records fully overlap.
reviewsHalf of residual heat variation came from 22-39 monthly records in four Atlantic and Pacific coral samples in 2003-2019
Prior prediction test reports limited heat support. Our test measures residual heat information in fixed-effect models, not forecast skill.