Machine-learning algorithms for identifying climate-resilient corals in the Republic of Palau

Machine-learning algorithms for identifying climate-resilient corals in the Republic of Palau | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Machine-learning algorithms for identifying climate-resilient corals in the Republic of Palau Anderson Mayfield, Alexandra Dempsey This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6162052/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 9 You are reading this latest preprint version Abstract To restore degraded coral reef habitats, it is critical to ensure that the scleractinian broodstock utilized can withstand future heatwaves. However, reef coral resilience is normally assessed only after catastrophic stress events. By tapping into a rich, “molecules-to-satellites” dataset acquired during the Living Ocean Foundation’s research mission to the Republic of Palau, we trained an artificial intelligence to accurately predict pocilloporid coral thermotolerance from relatively cheap, easy-to-measure environmental and ecological survey parameters. Specifically, a neural network featuring 22 predictors, such as coral cover and colony size, could forecast the whereabouts and properties of climate-resilient colonies of Pocillopora acuta with ~ 90% accuracy. This machine-learning model enables practitioners to 1) estimate the climate resilience of local pocilloporid populations and 2) identify habitats characterized by high pocilloporid coral resilience. climate change corals ecological forecasting machine-learning resilience Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 1. Introduction Although climate change threatens coral reefs across the globe [ 1 ], there are well established gradients in thermotolerance both across and within scleractinian species [ 2 ]. In some instances, extreme oceanographic conditions have led to coral stress-hardening; corals from a) high-temperature pools in American Samoa [ 3 ] and b) upwelling reefs of Southern Taiwan [ 4 ] have shown superior ability to acclimatize to both natural and experimentally induced temperature spikes. Unfortunately, there is nothing that can be observed in situ that signifies which colonies are of high climate resilience prior to their manifesting late-stage signs of stress. A practitioner probing a new region for the most thermotolerant colonies of a particular species needs to instead either conduct a laboratory-based thermal stress challenge requiring expensive equipment (e.g., CBASS [ 5 ]) or wait for a natural stress event to transpire to determine the biological material that may make the most sense to use as broodstock for selective breeding or coral reef restoration programs. In remote regions where monitoring is not routinely undertaken, researchers may only have the chance to identify the most resilient corals by timing their expeditions to fall during hypothesized bleaching events (e.g., via consulting NOAA’s Coral Reef Watch forecasts [ 6 ]) or instead attempting to identify habitats featuring oceanographic conditions known to stress-harden corals [ 7 ]. Since even uncovering whether such conditions exist in a given under-studied (or unstudied) area might be challenging, we attempted to “work backyards” from a holistic, “molecules-to-satellites” dataset acquired along the leeward (western) coast of the Republic of Palau during the Living Oceans Foundation’s Global Reef Expedition ( GRE [ 8 ]) to generate predictions of coral climate resilience from easier-to-measure parameters. During this 2016 expedition, divers surveyed the coral reef benthos (0–35 m) using a standardized protocol [ 9 ], fish biomass and diversity assessments were conducted, reefs were mapped and integrated with satellite imagery [ 10 ], and pocilloporid corals ( Pocillopora acuta ) were sampled along the leeward reefscape; their climate resilience was assessed via calculation of a “coral health index” (CHI) as described previously [ 11 – 12 ], followed by fate-tracking a subset of 22 colonies. CHI ranges from 0 (dead) to 5 (immortal) and scales positively and linearly with stress tolerance + longevity. However, it involves measuring many expensive (~ $ 500USD/sample) physiological and molecular response variables [ 13 ] that require highly trained personnel; it is neither a rapid nor a practical solution for those working in developing nations. We therefore sought to take a modeling approach in an effort to predict the CHI in unmeasured samples from cheaper environmental (e.g., habitat type, seawater quality), ecological (namely benthic composition), and “ in situ physiological parameters” (i.e., those that can be measured underwater by SCUBA divers). Could high accuracy in predicting bleaching susceptibility, or even long-term survival, be obtained without the need for laboratory analyses, restoration practitioners and others seeking knowledge on the whereabouts and properties of climate-tolerant corals could 1) expedite their search more strategically and 2) make superior projections about the future state of local reefs. 2. Materials and Methods 2.1. Surveys & sample collection. Standardized protocols were used across all legs of the GRE , the largest coral reef survey to date, with key facets mentioned in Carlton et al. [9]. The benthic data can be found in a companion article [8], where we took a similar modeling approach to instead predict the whereabouts of unexplored reefs with high coral cover. The environmental and ecological parameters assessed are detailed in Table 1; readers interested in Palau’s windward reefs, which were not visited due to high seas, should consult Colin [14]. Although 86 sites were surveyed (see Fig. 1 of Mayfield & Dempsey [8].), P. acuta -like corals were sampled from only 37. Briefly, ~ 50-mg branches were removed by bone-cutting pliers from each of 150 colonies spanning a number of habitat characteristics (Table 1). Care was taken to ensure at least 10-m separation between colonies to reduce risk of sampling clonemates. Each biopsy was broken in two: one was placed in a cryo-tube and rapidly frozen in a liquid nitrogen (-150ºC) dry-shipper (Chart MVE, USA), and the second was coated with 1.5-ml of RNALater™ (Life Technologies, USA) and frozen at -80ºC for 1–2 weeks before later transferring to the dry-shipper. The samples were exported under a CITES permit issued by Palau’s Bureau of Marine Resources and then imported to Taiwan several days later. 2.2. Coral biopsy extractions & molecular response variables. In addition to four physiological response variables that were documented in situ (or via assessment of images on a computer)-maximum (max.) colony length, planar surface area, color [15], and polyp extension (expanded or retracted)- several biomarkers were measured to calculate the CHI after extracting RNA, DNA, and protein from a subset of 118 of the 150 biopsies as described previously [16]. The co-extracted DNAs were used for four purposes. First, a multiplex Taqman® assay from Thomas et al. [17] was employed to species-type the coral hosts. From this same DNA aliquot, the Symbiodiniaceae assemblage (to genus-level) was elucidated as in Correa et al. [18]. Next, the Symbiodiniaceae genome copy proportion (GCP), a molecular proxy for endosymbiont density, was calculated via real-time PCR as in prior works [19]. Finally, the microbial communities associated with the sampled corals are now being analyzed via high-throughput metagenomic sequencing at the National University of Singapore. The co-isolated proteins were precipitated in acetone and stored at -80ºC for a future comparison of light vs. dark proteomes (a subset of colonies was sampled both during the day & again at night.). Coral host and dinoflagellate endosymbiont biomarkers determined to be stress-responsive from prior transcriptomic analyses [20] are discussed in detail with respect to their use in calculation of the CHI in prior works [11–12] and are shown in Table 2. Table 1 Factors predicted to influence the coral health index (CHI) and coral lifespan. With the exception of the Symbiodiniaceae (Sym) assemblage, all other parameters can be determined by eye, via cheap (<$100USD) meters deployed in situ (e.g., temperature & photosynthetically active radiation), or in silico (e.g., ImageJ [NIH, USA] for determining colony size). NA = not applicable. Name Abbreviation Units Options PHYSIOLOGICAL FACTORS (PHYS) maximum colony length max. length cm continuous (range = 4–41 cm) planar surface area a planar SA cm 2 continuous (range = 9-508 cm 2 ) colony color color 1–5; % as either a mean score for the colony (1–5) or a percentage of the colony with anomalous pigmentation (see Appendix A for details.) endosymbiont assemblage Sym assemblage Durusdinium , Cladocopium , Symbiodinium , or combinations thereof polyps extended? -- NA yes or no ENVIRONMENTAL FACTORS (ENV) latitude Lat ºN continuous (range = 6.92–8.17ºN) longitude Lon ºE continuous (range = 134–135ºE) island a -- NA see Carlton et al. [9] (n = 10). survey site a site see Fig. 1 of Mayfield & Dempsey [8]. lagoon -- NA inside or outside fore reef -- NA is or is not reef emergence emergence NA submerged or emergent reef exposure exposure NA protected, intermediate, or exposed reef location location NA lagoon, back reef, or fore reef reef type type NA fringing, barrier, patch, back, or channel time (binned) time 18:30 night? -- NA day vs. night temperature temp. ºC continuous (range = 26-29.4ºC) salinity -- none continuous (range = 31-34.4) photosynthetically active radiation PAR µmol m − 2 s − 1 continuous (range = 0-335 µmol m − 2 s − 1 ) colony depth depth m continuous (range = 5–31 m). As binned in 5-m increments in certain analyses (< 5 m, 5–10 m, etc.) ECOLOGICAL (BENTHIC) FACTORS (ECO) % barren substrate PB % continuous (range = 0.5–42%). % non-coral invertebrates PINVS % continuous (range = 1.2–20.6%) % algal cover by taxon 6 taxa % see online supplemental data file (Appendix C). % coral cover by genus 73 genera b % see online supplemental data file (Appendix C). total live coral cover LCC % continuous (range = 13–70%). As binned in certain analyses: 10–20%, 20–30%, 30–40%, etc. total live algal cover LAC % continuous (range = 12–71%) coral/algae ratio coral/algae unitless (range = 0.2–5.1) #coral genera present sum(#genera) n continuous (range = 17–45 genera) #coral genera present-binned sum(#genera)-binned n in 5-genera increments: <5 genera, 5–10 genera, 10–15 genera, etc. a Excluded from most analyses. b Only 63 were observed in the vicinity of the sampled coral colonies. Table 2 Diagnostic results for a representative sick coral: PaPd21.1-2. Although the same data were collected from pocilloporids of other Global Reef Expedition missions, the reference (Ref) range is local to Palau. The colony color score, % of colony bleached/necrotic, and Symbiodiniaceae (Sym) genome copy proportion (GCP) were used to calculate the “color metric” as described in Appendix A. Full gene names are as follows: copper-zinc superoxide dismutase ( cu-zn-sod ), green fluorescent protein-like chromoprotein ( gfp-cp ), ubiquitin ligase ( ubiq-ligase ), heat shock protein 90 ( hsp90 ), catalase-peroxidase ( katG ), & zinc-induced facilitator-like 1-like ( zifl1l ). NA = not applicable. NM = not measured. Response variable Value Ref range In spec? Diagnosis/notes colony color score 5 > 4 yes normal appearance % of colony bleached/necrotic 0 < 10 yes no evidence of bleaching or disease max. colony length 27 (4,41) yes planar surface area 220 (9,508) yes polyps extended? yes yes yes The following 10 response variables were used to calculate the heat map score (see Appendix A.): RNA/DNA ratio 1.93 > 0.5 yes Sym GCP 0.05 (0.05,5) borderline low endosymbiont densities despite normal appearance; evidence for non-dinoflagellate algae? host coral cu-zn-sod 0.91 (0.5,1.5) yes host coral gfp-cp 0.05 (1,6) too low could point to lack of evidence for self-shading host coral cytochrome P450 0.01 (0.02,0.4) too low could point to problems with host metabolism host coral catalase 1.5 (0.5,5) yes Sym ubiq-lig 10.2 (0.01,0.3) too high strong evidence for endosymbiont stress response Sym hsp90 15.9 (0.1,10) too high strong evidence for endosymbiont stress response Sym katG 1.1 (0.01,5) yes Sym zifl1l NM (1,100) NA extracted insufficient RNA to measure this mRNA The following four metrics were used to calculate the CHI : heat map score 3 (0,2) too high color metric 0.17 >-0.7 yes variability index 2.8 1 too low diagnosis : colony is abnormally stressed prognosis : colony will not survive future heat wave outcome : colony perished after subsequent heat wave 2.3. CHI. The CHI was calculated as described previously [21] to allow for comparability with other pocilloporids sampled during the GRE . Briefly, it is calculated from the “heat map score” (Table 1), the “color metric,” the variability index, and the Mahalanobis distance. Details can be found in the online supplemental methods (i.e., Appendix A). 2.4. Predictive modeling. An approach analogous to that of Mayfield et al. [13] was taken to attempt to predict the CHI from combinations of cheaper and easier-to-measure environmental and ecological parameters (i.e., “predictors”). These include those of Table 1 and subsets thereof. A two-step approach was taken in JMP® Pro (USA), which was used for all statistical and modeling analyses. First, JMP Pro’s “model screening” platform was used to simultaneously generate and compare a large number of models-decision tree, stepwise regression, generalized regression, bootstrap forest, boosted tree, Naïve Bayes, discriminant, XGBoost, support vector machines, partial least squares, k-nearest neighbors, neural network, and others- with CHI as the Y and second-order factorial combinations of subsets of predictors as the X’s (Tables A1-A2 in Appendix B). CHI was modeled three different ways: as a continuous value spanning 0 to 5, as binned into categorical scores of 0, 1, 2, 3, or 4–5 (values rounded to the nearest integer), or as a colony being either weak (CHI ≤ 1) or resilient (> 1). The data table was randomly partitioned into 89 (75%) and 29 (25%) training and validation samples, respectively, and all error terms below represent std. dev. Of the latter 29 colonies, a subset of 22 were fate-tracked for one year (2016–2017) to field-test the models. When the superior modeling type identified-defined as the one with the highest validation data R 2 for the continuous CHI analyses (acceptance threshold > 0.8) and the highest accuracy for the categorically binned health values (acceptance threshold > 90%)-was a neural network, a tuning GUI developed by Diedrich Schmidt was used to test hundreds to thousands of combinations of the core neural network hyperparameters (e.g., number of hidden layers, learning rate) as described previously [22]. Another analysis was undertaken in which images of the sampled corals at both mm- and cm-scales were analyzed alongside the other predictors to determine whether the powerful Torch AI could uncover image features that are diagnostic of particular health states. In general, Torch did not outperform neural networks (Table A3), though it is important to note that, for image analysis AIs to be properly leveraged, care must be taken to standardize image parameters in situ , such as distance from the colony, light level, etc. Many of these conditions might only lend themselves to laboratory analyses at present. 3. Results 3.1. Data access & coral species-typing. All numerical data presented can be found in the online supplemental data file (Appendix C), and images of the sampled corals, as well as their surrounding habitats, can be found here. Precipitated RNAs, DNAs, and proteins (all frozen at -80ºC), as well as decalcified fragments of a subset of 30 samples embedded in paraffin, are available for sharing; their whereabouts and other key information (e.g., sample size) are described in Mayfield [ 23 ]. All sampled colonies were typed as Pocillopora acuta , and not Pocillopora damicornis ; based on our benthic survey data, it is doubtful that P. damicornis is present in the west of Palau. To verify whether the probe-based assay of Thomas et al. [ 17 ] was accurate, the mitochondrial open reading frames (mORF) of three random DNA samples were PCR-amplified and sequenced in both directions, and all were 100% matches to published P. acuta mORF sequences on the NCBI database. 3.2. Predictor screening. Reef site was found in > 42 of the 100 random bootstrap forest models generated during the predictor screen and was the most influential predictor of CHI (Fig. 1 a); however, because our goal was to produce a generalizable model that could be used to find resilient corals in sites not-yet-surveyed , we removed this term from subsequent analyses. Of the remainders, Symbiodiniaceae assemblage requires molecular benchwork and is discussed in more detail below. Max. colony length and surface area co-vary since both are proxies of colony size; as such, their relative influences on the CHI are similar. 3.3. Response screening. A response screen is a false discovery rate (FDR)-controlled means of looking at each individual predictor’s influence on a Y, and only a single predictor, Pachyseris spp. cover (Fig. 1 b), was associated with a p -value < 0.01. However, this correlation was not statistically at the FDR-adjusted p -value of 0.1, and cover of this genus explained only ~ 8–9% of the variation in the CHI. Although it may be tempting to interpret this inverse relationship as evidence for competition, confounding factors are likely responsible; Pachyseris is a low-light coral found at depths > 20 m, whereas P. acuta is more common in the shallows. Those areas where the former is relatively more abundant are likely to be associated with lower CHI scores for P. acuta . 3.4. Variation in CHI . Because reef site was the top predictor in the predictor screen, we plotted CHI across an 0.1 x 0.1-º grid (Fig. 2 a). Although variability can be observed, there were no clear trends with respect to latitude, longitude, proximity to land, or other geographic variables that would be useful to managers. There was also no correlation between CHI and either total live coral (Fig. 2 b) or pocilloporid cover (Fig. 2 c); simply documenting the abundance of these corals, as is the extent of most survey efforts [ 24 ], is insufficient for making predictions of climate resilience. In fact, the few remaining corals in a highly marginalized habitat could be expected to out-perform conspecifics from relatively less impacted areas [ 25 ]. 3.5. A neural network for predicting CHI-a: 23-predictor model ( Fig. 3 ). Because the CHI (as continuous & binned as five integers) could not be confidently predicted (R 2 10%, respectively; Table A1), corals were scored as either resilient (CHI > 1) or weak (CHI ≤ 1; “sick” in some tables & figures), and models approached 97% accuracy when including virtually all environmental and ecological predictors (Table A2); 100% of the models surpassing the 90% accuracy cutoff were neural networks. A more parsimonious model is shown in Fig. 3 . In this single-hidden layer, boosted neural network featuring Gaussian activation (Fig. 3 a), 91% accuracy was achieved in predicting whether a coral would perish from or resist bleaching, and only 23 environmental factors were incorporated as X’s; this represents only a ~ 6% drop in accuracy whilst needing ~ four-fold less data than for the 97%-accuracy model. In the subset of field test samples (n = 22 colonies), 5 (25%) were predicted to perish by summer of 2017, and all did (100% accuracy); 2 of the remaining 17 were “false-positives:” predicted to be climate-resilient but actually died (89.5% accuracy or 91% overall accuracy across 22 colonies). The 23 environmental factors featured in this model have been listed with respect to their total effect in the independent uniform inputs variable importance analysis in Fig. 3 b. A desirability analysis was conducted in which the AI was prompted to generate the conditions resulting in the highest probability of a coral being climate-resilient (“ p (resilient”= maximized); T 2 extrapolation control was implemented to ensure that environmentally unrealistic conditions were not generated (e.g., an exposed lagoonal patch reef). By modifying the 23 predictors, a scenario was forecasted in which there is near-100% certainty that a coral would be climate-resilient; the results of the top-five predictors and their optimal levels are shown in Fig. 3 c. Coral cover had the greatest weight in the model, despite the fact that there was no clear relationship between CHI and coral cover in the simple, univariate approaches (Fig. 2 b-c). However, the CHI did not vary across coral cover bins less than 60%; in other words, P. acuta resilience only plummets on reefs whose total live coral cover is > 60% (15 out of the 86 surveyed reefs). As reefs with coral cover over 60% would be very space-limited environments, competition with other corals could account for this decline in CHI. Of the remaining four predictors in Fig. 3 c, the horizontal line for summed coral genera is deceiving because the variable importance analysis clearly points to an influence of this diversity metric. This discrepancy is in part due to the fact that generic diversity covaries with several other environmental factors, such as coral cover (R 2 = 0.05); changes in other parameters confound that of generic diversity to where, under the projected conditions of the desirability analysis, p (resilient) does not vary. When performing a simulation (10,000 rows) and allowing generic diversity to fluctuate randomly across the entire range documented in Palau, there is a ~ 6% decrease in p (resilient) upon increasing the generic diversity from the optimal level of 27 genera to 41 genera (1.5-fold increase); in other words, generic diversity is a relatively weak predictor of P. acuta climate resilience when assessed by itself. The optimal colony diameter of 14 cm (Fig. 3 c) is significantly different from the mean colony size in Palau of 17 cm (± 8 cm; p < 0.0001). Since a typical colony can extend its diameter by ~ 2 cm per year [ 26 ], this 3-cm difference between the optimal and mean coral diameter represents a ~ 1.5-year age difference. Why a slightly younger and smaller colony would have higher climate resilience is currently unclear, and upon looking at the various physiological response variables assessed (e.g., gene expression, Symbiodiniaceae density, etc.), few showed variation across size. Therefore, a simulation was run in which the other optimized parameters from the desirability analysis were fixed at their optimal levels while max. colony length was allowed to vary randomly across its Palauan distribution. The results (Fig. 4a) reveal that p (resilient) actually does not vary much once a colony reaches ~ 10 cm. There is a “danger zone” of 2–5 cm, possibly due to corallivory or lack of ability to effectively compete whilst at such a small size. Note that while Fig. 4a corroborates the findings of Fig. 3 c at the lower end of the size range, the latter analysis shows another increase in the probability of colony being sick beyond about 30 cm that is only partly reflected in Fig. 4a. This discrepancy is likely due to the low number of large colonies sampled (only one was > 30 cm in diameter.). Colony depth was also weighted highly in the superior predictive model of coral resilience (Fig. 3 b), and the optimal depth was 10.8 m (PAR = 91 µmol m -2 s -1 ), significantly shallower than the mean sampling depth of 15 m (± 7; PAR = 77 µmol m -2 s -1 ). However, this is close to the median depth of 10.9 m (due to the highly skewed nature of the dataset, in which most corals were collected in the 8-12-m window). PAR alone did not have a strong impact on CHI or p (resilient) (Fig. 3 b). Wave energy (as “exposure” in the tables & figures), which would also vary by depth, was also a relatively weak predictor. From the desirability analysis (Fig. 3 c), there is not a large decrease in p (resilient) until the depth increases to about 25 m. To attempt to corroborate this, another simulation was performed (10,000 samples) in which the optimal values were locked for all parameters except depth, which was allowed to fluctuate randomly across its Palauan dataset range (std. dev.=5 m). The results (Fig. 4b) are similar to those of Fig. 3 c, albeit with a more gradual tapering at greater depths in the simulation. Another key difference is that the optimal depth in the simulation is notably lower, with p (resilient) maxing out at around 8 m instead of 11 m. 3.6. A neural network for predicting CHI-b: 22-predictor model ( Fig. 5 ). Were one interested in maximizing the chance of selecting a climate-resilient P. acuta colony in Palau, they would look for colonies around 14 cm in diameter at around 10–11 m depth in an area of relatively low (for Palau) 1) coral cover (20–30%) and 2) generic diversity (25–30 genera). All of these predictors can be assessed by diving equipment (depth gauge), simple scientific instruments like rulers (colony size), or trained surveyors (coral cover & generic diversity). However, there is a drawback to the model of Fig. 3 ; despite the fact that Symbiodiniaceae assemblage has a relatively low total effect size (~ 4-fold lower than coral cover, for instance), if removed from the model, the accuracy could decrease. Because Symbiodiniaceae assemblage requires laboratory benchwork that normally occurs days to weeks after a field trip is completed, it is not a desirable predictor for one looking for a fast, cheap means of identifying resilient corals. For this reason, the term was removed, and a similarly structured neural network was rebuilt (Fig. 5 a); upon removing this expensive, difficult-to-measure response variable, the accuracy decreased to 89%, just lower than the a priori accuracy cutoff of 90%. This 2% decrease in accuracy may be worth sacrificing given that the resulting model features only metrics that can either be assessed in situ or shortly after the dive upon analysis of image and benthic survey data. In fact, the top-five predictors are the same as in the model of Fig. 3 , albeit with a different relative ordering. As with the 23-predictor model (Fig. 3 c), the desirability analysis (Fig. 5 b) recommends a relatively small colony (11 vs. 14 cm in Fig. 3 ) on a low coral cover reef (20–30% coral cover in both models) with relatively low coral diversity (20–25 genera vs. 27 genera in Fig. 3 ). The main difference between the two models is the recommended depth: 18.5 vs. <11 m in the models of Figs. 5 and 3 , respectively. 4. Discussion In even the top two neural networks, post-survey analyses must be performed to calculate coral cover and coral generic diversity; whether or not a particular reef habitat is colonized by resilient P. acuta specimens would, in other words, not be known until after the dive, and a repeat visit would be necessary to collect samples (were the conditions indeed even suggestive of resilient corals). What is more likely to be the case is that a scientist or manager will have some pre-existing data on certain reefs, such as a long-term monitoring site. In this case, the relevant values can be put into the model of Fig. 5 (as a GUI) to calculate the probability that a particular habitat might be characterized by resilient pocilloporids. As a hypothetical example, say that a surveyor determined that a particular reef area is characterized by a coral cover of 40–50% with 25–30 coral genera present and pocilloporid corals found around 15 m. By inputting these values into the model of Fig. 5 and setting them as “fixed,” we can then re-calculate p (resilient) across differing colony sizes. Under these conditions, a colony of 17-cm diameter would have a 91% chance of being climate-resilient (Fig. 6 a); were coral cover and colony size instead allowed to vary, while the other 20 environmental parameters are fixed at their optimal levels, a contour plot (Fig. 6 b) instead provides a means of searching for resilient corals. However, the model of Fig. 5 is contingent upon measuring all 22 environmental factors; what if there is insufficient time to do so? Some can be known from consulting maps (e.g., reef type, reef exposure), yet others require divers or at least drone or satellite imagery. If one knew all mapping-related parameters-latitude, longitude, lagoon (inside vs. outside), reef type, reef emergence, reef exposure, forereef (is or is not a forereef)- in addition to coral cover and pocilloporid colony depth, but did not know the abundance of any other organism or anything about the seawater quality, the probability of correctly finding a climate-resilient pocilloporid would be ~ 71%. If temperature and coral polyp extension were also known, the probability increases to 75%. If one did not know the sizes of the pocilloporid colonies, the probability decreases only slightly to 72%. Whether or not a proper survey of the benthos is worth the effort of increasing the probability of correctly sourcing resilient corals from 75 to ~ 90% will ultimately be up to the practitioner. The value of an AI trained with data from 2016 may be suspect, especially considering that coral reefs have become increasingly marginalized since then; are predictions that were accurate in 2016–2017 of any use to us nearly 10 years later? Under the most pessimistic view, this dataset might only have value in highlighting what these reefs looked like when they were still dominated by stony corals; coral cover averaged ~ 45% on Palau’s leeward reefs at that time, which is relatively high by global standards. Of the 118 pocilloporids analyzed in detail, 36 (31%) were predicted to perish from climate change-related temperature increases and/or local stressors; this includes 5 fate-tracked colonies, all of which died (100% accuracy). Of the 22 fate-tracked colonies, 17 were predicted to be climate-resilient, though 2 of these were found dead in follow-up surveys in 2017 (“false-positives”). This means that the overall field accuracy of 20/22 (91%) is similar to the validation sample accuracy (i.e., samples simply held back from the AI). Follow-up surveys must now be conducted to determine whether the additional 65 colonies predicted to be climate-tolerant remain so to this day, though as was the case with most GRE survey sites, the majority of the reefs were difficult to reach and had never before been surveyed or sampled at the time. Hopefully, funds from international initiatives like CORDAP can be used by local scientists to periodically check on these corals such that the long-term predictive capacity of the models can be validated. If these models are only adept at identifying corals that survive only, for instance 1–2 years longer than conspecifics, they may ultimately be of little utility to those looking to stock their nurseries with the most robust corals. However, others have predicted that the thermotolerance of some Palauan corals might actually be increasing over time [ 27 ], though it is important to note that these models were not ground-truthed and do not consider key physiological benchmarks of resilience, only organismal abundance. Just as a city with more people is not associated with more resilient citizens, high-coral cover reefs are no more climate-resilient; local acroporids in particular have struggled to recover after recent disturbances [ 28 ]. As it is unlikely that adaptation is keeping pace with the rapid rate of seawater temperature rise in Palau, the degree to which accelerated evolution-based approaches may play a role in boosting coral resilience is currently under investigation by scientists from The Nature Conservancy (Palau; Y. Golbuu), Newcastle University (UK; J. Guest), and the University of Queensland (Australia; P. Mumby) through a CORDAP “Coral Accelerator Program” award. Perhaps the most resilient P. acuta colonies uncovered in our surveys could be cross-bred to yield more thermotolerant larvae for later reef reseeding efforts. Aside from our omission of samples from Palau’s windward reefs, another limitation of this study is that it features only a single species: P. acuta . Very different optimal conditions will be associated with other corals, and it is entirely possible that model accuracy for many of them will be too low to be of any value. We know more about P. acuta than any other coral [ 29 ], and molecules-to-satellites datasets do not exist for any other scleractinian except Seriatopora hystrix [ 30 ]; the accuracies presented herein likely represent amongst the highest that could be expected for predicting coral resilience to climate change. This is not to say that resilient corals have not been identified in Palau via other approaches. For instance, thermotolerant corals have been found in sheltered regions of the Rock Islands, such as Nikko Bay [ 31 ], from where corals were also sampled herein, and the Porites lobata genotypes living there are distinct from conspecifics on the outer reefs [ 32 ]. Our species-level molecular analysis was too crude to resolve individual P. acuta populations, so the DNAs have been sent to colleagues to analyze host coral, dinoflagellate, and microbial genetics at a higher resolution to determine the degree to which adaptation is driving the marked divergence in climate resilience documented [ 33 ]. 5. Conclusions By tapping into a dataset featuring molecular and physiological data, as well as simple, easy-to-measure survey parameters, we generated two machine-learning-based models for accurately predicting the climate resilience of the model coral P. acuta in Palau. Accuracies were ~ 90%, and colony size, depth, and local coral cover and diversity were all important predictors of pocilloporid resilience. Host coral genetics surely play a role (i.e., adaptation), as well, especially given the wide diversity of habitats and vast spatial extent of the survey [ 34 ]. Although a colony’s genotype could likely never be discerned in situ by a diver, knowing whether distinct populations tend to be those of highest resilience, regardless of their habitat preferences and physiological characteristics, will nevertheless be useful information for local practitioners, especially those looking to ensure that restored reefs are both “climate-proofed” and diverse [ 35 – 36 ]. Abbreviations The following abbreviations are used in this manuscript. See also Table 1 for non-standard abbreviations and Table 2 for full gene names of mRNA biomarkers. CHI Coral health index CORDAP Coral Research and Development Accelerator Platform EF Environmental factor (i.e., predictor) FDR False discovery rate GCP Genome copy proportion GRE Global Reef Expedition GUI Graphical user interface HL Hidden layer HMS Heat map score mORF Mitochondrial open reading frame NN Neural network PAR Photosynthetically active radiation Sym Symbiodiniaceae Declarations Supplementary Materials: The following supporting information can be downloaded at: www. xxx/, Appendix A: supplemental methods; Appendix B: supplemental tables (A1-A3); Appendix C: online supplemental data file. Appendices A-B are in the same Word file. Author Contributions: Conceptualization, A.B.M.; methodology, A.B.M., A.C.D.; software, A.B.M., A.C.D.; validation, A.B.M., A.C.D.; formal analysis, A.B.M., A.C.D; investigation, A.B.M., A.C.D.; resources, A.B.M., A.C.D.; data curation, A.B.M., A.C.D.; writing—A.B.M.; writing—review and editing, A.B.M.; visualization, A.B.M.; project administration, A.B.M.; funding acquisition, A.B.M., A.C.D. All authors have read and agreed to the published version of the manuscript. Funding: This research was funded by the Khaled bin Sultan Living Oceans Foundation, as well as the Coral Research and Development Accelerator Platform (CORDAP). Data Availability Statement: All data can be accessed via the online supplemental data file (Appendix C). All images of the coral reef habitats surveyed, as well as the sampled coral colonies, can be found on coralreefdiagnostics.com. The GUI necessary to run the neural networks (for those without coding or machine-learning knowledge) are also posted on this coral physiology database. Acknowledgments: We would like to thank the team at JMP for assistance with predictive modeling and machine-learning. ABM would like to thank the Fulbright and MacArthur Foundations for supporting his stay in Taiwan, where laboratory analyses were conducted. Conflicts of Interest: The authors declare no conflicts of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript; or in the decision to publish the results. Consent to Participate declaration: not applicable. Consent to Publish declaration: not applicable. Ethics declaration: not applicable References Reimer, J.D.; Peixoto, R.S.; Davies, S.W.; Traylor-Knowles, N.; Short, M.L.; Cabral-Tena, R.A.; Burt, J.A.; Pessoa, I.; Banaszak, A.T.; Winters, R.S.; Moore, T.; Schoepf, V.; Kaullysing, D.; Calderon-Aguilera, L.E.; Wörheide, G.; Harding, S.; Munbodhe, V.; Mayfield, A.B.; Ainsworth, T.; Vardi, T.; Eakin, C.M.; Voolstra, C.R. The Fourth Global Coral Bleaching Event–where do we go from here? Coral Reefs 2024 , 43 , 1121-1125. Loya, Y.; Sakai, K.; Yamazato, K.; Nakano, Y.; Sambali, H.; van Woesik, R. Coral bleaching: the winners and the losers. Ecol. Letters 2001 , 4 , 122-131. Barshis, D.J.; Ladner, J.T.; Oliver, T.A.; Seneca, F.O.; Traylor-Knowles, N.; Palumbi, S.R. Genomic basis for coral resilience to climate change. Proc. Natl. Acad. Sci. USA . 2013 , 110 , 1387-1392. https://doi.org/doi:10.1073/pnas.1210224110. Mayfield, A.B.; Chan, P.H.; Putnam, H.M.; Chen, C.S.; Fan, T.Y. The effects of a variable temperature regime on the physiology of the reef-building coral Seriatopora hystrix : results from a laboratory-based reciprocal transplant. J. Exp. Biol. 2012 , 215 , 4183-4195. Voolstra, C.R.; Buitrago-López, C.; Perna, G.; Cárdenas, A.; Hume, B.C.; Rädecker, N.; Barshis, D.J. Standardized short-term acute heat stress assays resolve historical differences in coral thermotolerance across microhabitat reef sites. Global Change Biol. 2020 , 26(8) , 4328-4343. Liu, G.; Heron, S.F.; Eakin, C.M.; Muller-Karger, F.E.; Vega-Rodriguez, M.; Guild, L.S.; De La Cour, J.L.; Geiger, E.F.; Skirving, W.J.; Burgess, T.F.R.; et al. Reef-scale thermal stress monitoring of coral ecosystems: new 5-km global products from NOAA Coral Reef Watch. Remote Sensing 2014 , 6 (11), 11579-11606. Bachman, S.D.; Kleypas, J.A.; Erdmann, M.; Setyawan, E. A global atlas of potential thermal refugia for coral reefs generated by internal gravity waves. Front. Mar. Sci. 2022 , 9 , 921879. https://doi.org/10.3389/fmars.2022.921879. Mayfield, A.B.; Dempsey, A.C. AI-driven coral reef bioprospecting in the Republic of Palau. Platax 2024 , 21 , 1-30. Carlton, R.; Dempsey, A.C.; Lubarsky, K.; Lindfield, S.; Faisal, M.; Purkis, S. Global Reef Expedition: The Republic of Palau. Final Report . Khaled bin Sultan Living Oceans Foundation, Annapolis, MD, USA, 2020 ; Vol. 12 Purkis, S.J.; Gleason, A.C.R.; Purkis, C.R.; Dempsey, A.C.; Renaud, P.G.; Faisal, M.; Saul, S.; Kerr, J.M. High-resolution habitat and bathymetry maps for 65,000 sq. km of Earth’s remotest coral reefs. Coral Reefs 2019 , 38 , 467-488. Mayfield, A.B.; Chen, C.S.; Dempsey, A.C. Biomarker profiling in reef corals of Tonga’s Ha’apai and Vava’u Archipelagos. PLoS ONE 2017 , e0185857 . Mayfield, A.B.; Chen, C.S.; Dempsey, A.C. Identifying corals displaying aberrant behavior in Fiji’s Lau Archipelago. PLoS ONE 2017 , e0177267 . Mayfield, A.B.;Dempsey, A.C.; Chen, C.S.; Lin, C. Expediting the search for climate-resilient reef corals in the Coral Triangle with artificial intelligence. Applied Sci. 2022 , 12 , 12955. Colin PL (2009) Marine Environments of Palau , 1st edition. Mutual Publishing, Honolulu, Hawaii, USA, 2009 . Siebeck, U.E.; Marshall, N.J.; Klüter, A.; Hoegh-Guldberg, O. Monitoring coral bleaching using a colour reference card. Coral Reefs 2006 , 25 , 453-460. https://doi.org/10.1007/s00338-006-0123-8 Mayfield, A.B.; Wang, L.H.; Tang, P.C.; Hsiao, Y.Y.; Fan, T.Y.; Tsai, C.L.; Chen, C.S. Assessing the impacts of experimentally elevated temperature on the biological composition and molecular chaperone gene expression of a reef coral. PLoS ONE 2011 , e26529 . Thomas, L.; Stat, M.; Evans, R.D.; Kennington, W.J. A fluorescence-based quantitative real-time PCR assay for accurate Pocillopora damicornis species identification. Coral Reefs 2016 , 35 , 895-899. Correa, A.M.S.; McDonald, M.D.; Baker, A.C. Development of clade-specific Symbiodinium primers for quantitative PCR (qPCR) and their application to detecting clade D symbionts in Caribbean corals. Mar. Biol. 2009 , 156 , 2403-2411. Putnam, H.M.;Mayfield, A.B.; Fan, T.Y.; Chen, C.S.; Gates, R.D. The physiological and molecular responses of larvae from the reef-building coral Pocillopora damicornis exposed to near-future increases in temperature and p CO 2 . Mar. Biol. 2013 , 160 , 2157-2173. Mayfield, A.B.; Wang, Y.B.; Chen, C.S.; Chen, S.H.; Lin, C.Y. Compartment-specific transcriptomics in a reef-building coral exposed to elevated temperatures. Mol. Ecol. 2014 , 23 , 5816-5830. Mayfield, A.B.; Chen, C.S.; Dempsey, A.C. The molecular ecophysiology of closely related pocilloporid corals of New Caledonia. Platax 2017 , 14 , 1-45. Mayfield, A.B.Machine-learning-based proteomic predictive modeling with thermally challenged Caribbean reef corals. Diversity 2022 , 14 , 33. Mayfield, A.B.Establishment of a long-term reef-building coral biopsy archive as a tool for the marine biology research community. Platax 2023 , 20 , 7-30. McClanahan, T.R.; Donner, S.D.; Maynard, J.A.; MacNeil, M.A.; Graham, N.A.J. Prioritizing key resilience indicators to support coral reef management in a changing climate. PLoS ONE 2012 , e42884 . Rubin, E.; Enochs, I.; Foord, C.; Kolodziej, G.; Basden, I.; Manzello, D.P.; Mayfield, A.B. Molecular mechanisms of coral persistence within highly urbanized locations in the Port of Miami, Florida. Front. Mar. Sci. 2021 , 8 , 695236. Anderson, K.D.; Cantin, N.E.; Heron, S.F.; Pisapia, C.; Pratchett, M. Variation in growth rates of branching corals along Australia’s Great Barrier Reef. Sci. Rep. 2017 , 7 , 2920. Lachs, L.; Donner, S.D.; Mumby, P.J.; Bythell, J.C.; Humanes, A.; East, H.K.; Guest, J.R. Emergent increase in coral thermal tolerance reduces mass bleaching under climate change. Nat. Commun . 2023 , 14 , 4939. https://doi.org/10.1038/s41467-023-40601-6 Lachs, L.; Biondi, P.; Gouezo, M.; Nestor, V.; Olsudong, D.; Guest, J.; Golbuu, Y. Demographic recovery of corals at a wave-exposed reef following catastrophic disturbance. Coral Reefs 2024 , 43 , 193-199. https://doi.org/10.1007/s00338-024-02464-1 Mayfield, A.B.; Fan, T.Y.; Chen, C.S. Physiological acclimation to elevated temperature in a reef-building coral from an upwelling environment. Coral Reefs 2013 , 32 , 909-921. Mayfield, A.B.;Chen, Y.J.; Lu, C.Y.; Chen, C.S. Exploring the environmental physiology of the Indo-Pacific reef coral Seriatopora hystrix using differential proteomics. Open J. Mar. Sci. 2018 , 8 , 223-252. Woesik, R.v.; Houk, P.; Isechal, A.L.; Idechong, J.W.; Victor, S.; Golbuu, Y. Climate-change refugia in the sheltered bays of Palau: analogs of future reefs. Ecol. Evol. 2012 , 2 (1), 2474-2484. https://doi.org/10.1002/ece3.363. Rivera, H.E.; Cohen, A.L.; Thompson, J.R.; Baums, I.B.; Fox, M.D .; Meyer-Kaiser, K.S. Palau’s warmest reefs harbor thermally tolerant corals that thrive across different habitats. Commun. Biol. 2022 , 5 , 1394. https://doi.org/10.1038/s42003-022-04315-7 Palumbi, S.R.; Walker, N.S.; Hanson, E.; Armstrong, K.; Lippert, M.; Cornwell, B.; Nestor, V.; Golbuu, Y. Small-scale genetic structure of coral populations in Palau based on whole mitochondrial genomes: Implications for future coral resilience. Evol. Appl. 2023 , 16 (2), 518-529. doi: 10.1111/eva.13509. Baums, I.B. A restoration genetics guide for coral reef conservation. Mol. Ecol . 2008 , 17 (12), 2796-2811. Quigley, K.M.; Hein, M.; Suggett, D.J. Translating the 10 golden rules of reforestation for coral reef restoration. Conserv. Biol. 2022 , 36(4), e13890. https://doi.org/10.1111/cobi.13890. Quigley, K.M.; Bay, L.K.; van Oppen, M.J.H. Genome-wide SNP analysis reveals an increase in adaptive genetic variation through selective breeding of coral. Mol. Ecol . 2020 , 29 (12), 2176-2188. Additional Declarations No competing interests reported. Supplementary Files MayfieldDempseysupplementalmaterialappendicesAC.zip Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 27 Jun, 2025 Reviews received at journal 22 Jun, 2025 Reviews received at journal 09 May, 2025 Reviewers agreed at journal 18 Apr, 2025 Reviewers agreed at journal 03 Apr, 2025 Reviewers invited by journal 01 Apr, 2025 Editor assigned by journal 28 Mar, 2025 Submission checks completed at journal 28 Mar, 2025 First submitted to journal 05 Mar, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6162052","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":445165447,"identity":"0731fbf0-6150-4fd0-b7fe-5d7db4a7850e","order_by":0,"name":"Anderson Mayfield","email":"data:image/png;base64,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","orcid":"","institution":"Coral Reef Diagnostics","correspondingAuthor":true,"prefix":"","firstName":"Anderson","middleName":"","lastName":"Mayfield","suffix":""},{"id":445165448,"identity":"0c822742-fa1a-4579-b66d-178143a3bc56","order_by":1,"name":"Alexandra Dempsey","email":"","orcid":"","institution":"Khaled bin Sultan Living Oceans Foundation","correspondingAuthor":false,"prefix":"","firstName":"Alexandra","middleName":"","lastName":"Dempsey","suffix":""}],"badges":[],"createdAt":"2025-03-05 11:23:10","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6162052/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6162052/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":81131900,"identity":"8a0873fe-05a4-41cc-9cd5-659d3f3162a5","added_by":"auto","created_at":"2025-04-22 14:47:36","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":409502,"visible":true,"origin":"","legend":"\u003cp\u003ePredicting the coral health index (CHI) from common physiological+ environmental (n=22) and ecological (n=77) parameters (all of which being relatively easier \u0026amp; cheaper-to-measure than the CHI itself). The physiological and environmental predictors included those 4 and 16 in Table 1 (n=20), respectively, as well as #21) depth binned into pre-defined categories and #22) site. The ecological parameters included those 76 of Table 1 plus #77) coral cover binned into 10-% increments (10-20%, 20-30%, etc.). Only the top 10 predictors have been shown (\u003cstrong\u003ea\u003c/strong\u003e). CHI was plotted against the top predictor identified in the response screen: \u003cem\u003ePachyseris\u003c/em\u003espp. (“PACH” in [\u003cstrong\u003ea\u003c/strong\u003e]) abundance (\u003cstrong\u003eb\u003c/strong\u003e). The shading about the line-of-best fit represents 95% confidence. A thinly branched \u003cem\u003ePocillopora acuta\u003c/em\u003e sample with a relatively high CHI score (\u0026gt;3.5) has been shown in \u003cem\u003e\u003cstrong\u003eb-1\u003c/strong\u003e\u003c/em\u003e. Scale bar=3 mm.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-6162052/v1/ceceda6bd812dec04976c8f5.png"},{"id":81131201,"identity":"439d89b6-e913-45cb-be72-a96581d0d229","added_by":"auto","created_at":"2025-04-22 14:39:36","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":1171299,"visible":true,"origin":"","legend":"\u003cp\u003eLarge-scale spatial variation in the \u003cem\u003ePocillopora acuta\u003c/em\u003e coral health index (CHI) in Palau.\u003cstrong\u003e \u003c/strong\u003eIn panel \u003cstrong\u003ea\u003c/strong\u003e, means were calculated within 0.1 x 0.1-º squares, and no samples were collected from the windward (eastern) side of the country. Panels \u003cstrong\u003eb\u003c/strong\u003e and \u003cstrong\u003ec\u003c/strong\u003e show the correlation between CHI and total live cover of 1) all observed hard coral genera and 2) pocilloporid corals only, respectively. Bands about the regression lines represent 95% confidence, and neither correlation was statistically significant (\u003cem\u003ep\u003c/em\u003e\u0026gt;0.01).\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-6162052/v1/7b6a7f539e3b76bbf8f79562.png"},{"id":81132672,"identity":"e6fbfdad-1230-42ed-bcf2-8d60e5dbc5f3","added_by":"auto","created_at":"2025-04-22 14:55:36","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":2513652,"visible":true,"origin":"","legend":"\u003cp\u003eA neural network for predicting coral resilience from 23 environmental factors (EF). The structure of the single-hidden layer (HL), Gaussian activation, boosted neural network is shown (\u003cstrong\u003ea\u003c/strong\u003e), with other hyperparameters presented in an inset; see Table A2 for additional model details. Results from a variable importance analysis (independent uniform inputs) are shown in \u003cstrong\u003eb\u003c/strong\u003e and reflect relative model weights of the various EF. A desirability analysis was performed with all 23 EF (\u003cstrong\u003ec\u003c/strong\u003e) to maximize the probability of a coral being climate-resilient (“\u003cem\u003ep\u003c/em\u003e(resilient)”). The optimal values for the top-five EF are written in red below the plots and shown in the plots themselves as hatched, vertical red lines. A colony with a markedly low coral health index (CHI) of only 1.2 (\u003cstrong\u003ed\u003c/strong\u003e), which perished on account of anomalously high temperatures in the year following its sampling (2017), has been shown (scale bar=5 mm).\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-6162052/v1/a48ae90cb688037a323e0a29.png"},{"id":81131903,"identity":"7be2c350-a21f-4f45-ba6a-1a51b88c8e9d","added_by":"auto","created_at":"2025-04-22 14:47:36","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":138444,"visible":true,"origin":"","legend":"\u003cp\u003eSimulated data (n=10,000) in which the optimal levels of the environmental factors (i.e., predictors) of the neural network in Fig. 3 were fixed with the exception of maximum (max.) colony length (a), which was allowed to vary randomly across a distribution with mean=17±8 cm, or colony depth (b), which was allowed to vary randomly across a distribution with mean=15±7 m.\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-6162052/v1/30c7957d362cb5d92948e095.png"},{"id":81132671,"identity":"072f615a-998f-4ae3-b39a-31e8da19d8e8","added_by":"auto","created_at":"2025-04-22 14:55:36","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":1970001,"visible":true,"origin":"","legend":"\u003cp\u003eNeural network (single-hidden layer [HL]) for predicting coral resilience (resilient vs. weak/sick) from 22 environmental factors (EF). The structure of the neural network has been shown (\u003cstrong\u003ea\u003c/strong\u003e) followed by a desirability analysis (\u003cstrong\u003eb\u003c/strong\u003e) in which the AI was programmed to maximize the probability of a coral being climate-resilient: \u003cem\u003ep\u003c/em\u003e(resilient). Only the top-five EF from a variable important analysis (independent uniform inputs) have been shown, and the values in red below the plots, as well as the hatched, red, vertical lines in the plots themselves, represent the optimal values (at which \u003cem\u003ep\u003c/em\u003e(resilient)=~99%). MR=misclassification rate.\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-6162052/v1/b2fab53dc19dba83f444ddd6.png"},{"id":81131905,"identity":"c1c68c7e-dd30-439d-8cc3-8756d9dbc8fc","added_by":"auto","created_at":"2025-04-22 14:47:36","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":936210,"visible":true,"origin":"","legend":"\u003cp\u003eA hypothetical analysis in which a practitioner can enter specific values for maximum (max.) colony diameter and/or depth into the model of Fig. 5 (\u003cstrong\u003ea\u003c/strong\u003e), while coral cover, coral generic diversity, and the remaining 18 environmental factors are fixed at the optimum levels from the desirability analysis. In this example, pocilloporids were common at 15 m. Upon entering “15” into the second red box, the probabilities then update accordingly, and it was found by manually dragging the vertical, hatched red line in the max. colony length panel that, under these conditions, the colony size with the highest probability of being climate-resilient (\u003cem\u003ep\u003c/em\u003e(resilient)=0.910) under the conditions shown in \u003cstrong\u003ea\u003c/strong\u003eis 17 cm. Assuming all 20 EF except max. colony length and coral cover are fixed at their optimal levels, a contour plot (\u003cstrong\u003eb\u003c/strong\u003e) can be used to visualize scenarios associated with pocilloporids of high climate resilience (as the continuous coral health index [CHI]).\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-6162052/v1/162ea3cd290d7075f4faecfa.png"},{"id":81132999,"identity":"ecf19f53-e839-44cd-b94f-0972ec6f3434","added_by":"auto","created_at":"2025-04-22 15:03:40","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":9071856,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6162052/v1/799f4cca-4f40-454b-bb44-866590ad6f2a.pdf"},{"id":81131202,"identity":"0083b31f-c747-4fc5-991b-285afa2a8b6d","added_by":"auto","created_at":"2025-04-22 14:39:36","extension":"zip","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":267484,"visible":true,"origin":"","legend":"","description":"","filename":"MayfieldDempseysupplementalmaterialappendicesAC.zip","url":"https://assets-eu.researchsquare.com/files/rs-6162052/v1/4b2ec879775e35530573c27c.zip"}],"financialInterests":"No competing interests reported.","formattedTitle":"\u003cp\u003eMachine-learning algorithms for identifying climate-resilient corals in the Republic of Palau\u003c/p\u003e","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eAlthough climate change threatens coral reefs across the globe [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e], there are well established gradients in thermotolerance both across and within scleractinian species [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. In some instances, extreme oceanographic conditions have led to coral stress-hardening; corals from a) high-temperature pools in American Samoa [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e] and b) upwelling reefs of Southern Taiwan [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e] have shown superior ability to acclimatize to both natural and experimentally induced temperature spikes. Unfortunately, there is nothing that can be observed \u003cem\u003ein situ\u003c/em\u003e that signifies which colonies are of high climate resilience prior to their manifesting late-stage signs of stress. A practitioner probing a new region for the most thermotolerant colonies of a particular species needs to instead either conduct a laboratory-based thermal stress challenge requiring expensive equipment (e.g., CBASS [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]) or wait for a natural stress event to transpire to determine the biological material that may make the most sense to use as broodstock for selective breeding or coral reef restoration programs.\u003c/p\u003e \u003cp\u003eIn remote regions where monitoring is not routinely undertaken, researchers may only have the chance to identify the most resilient corals by timing their expeditions to fall during hypothesized bleaching events (e.g., via consulting NOAA\u0026rsquo;s Coral Reef Watch forecasts [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]) or instead attempting to identify habitats featuring oceanographic conditions known to stress-harden corals [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. Since even uncovering whether such conditions exist in a given under-studied (or unstudied) area might be challenging, we attempted to \u0026ldquo;work backyards\u0026rdquo; from a holistic, \u0026ldquo;molecules-to-satellites\u0026rdquo; dataset acquired along the leeward (western) coast of the Republic of Palau during the Living Oceans Foundation\u0026rsquo;s \u003cem\u003eGlobal Reef Expedition\u003c/em\u003e (\u003cem\u003eGRE\u003c/em\u003e [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]) to generate predictions of coral climate resilience from easier-to-measure parameters.\u003c/p\u003e \u003cp\u003eDuring this 2016 expedition, divers surveyed the coral reef benthos (0\u0026ndash;35 m) using a standardized protocol [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e], fish biomass and diversity assessments were conducted, reefs were mapped and integrated with satellite imagery [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e], and pocilloporid corals (\u003cem\u003ePocillopora acuta\u003c/em\u003e) were sampled along the leeward reefscape; their climate resilience was assessed via calculation of a \u0026ldquo;coral health index\u0026rdquo; (CHI) as described previously [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e], followed by fate-tracking a subset of 22 colonies. CHI ranges from 0 (dead) to 5 (immortal) and scales positively and linearly with stress tolerance\u0026thinsp;+\u0026thinsp;longevity. However, it involves measuring many expensive (~\u003cspan\u003e$\u003c/span\u003e500USD/sample) physiological and molecular response variables [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e] that require highly trained personnel; it is neither a rapid nor a practical solution for those working in developing nations. We therefore sought to take a modeling approach in an effort to predict the CHI in unmeasured samples from cheaper environmental (e.g., habitat type, seawater quality), ecological (namely benthic composition), and \u0026ldquo;\u003cem\u003ein situ\u003c/em\u003e physiological parameters\u0026rdquo; (i.e., those that can be measured underwater by SCUBA divers). Could high accuracy in predicting bleaching susceptibility, or even long-term survival, be obtained without the need for laboratory analyses, restoration practitioners and others seeking knowledge on the whereabouts and properties of climate-tolerant corals could 1) expedite their search more strategically and 2) make superior projections about the future state of local reefs.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e"},{"header":"2. Materials and Methods","content":"\u003cp\u003e\u003cstrong\u003e2.1. Surveys \u0026amp; sample collection.\u003c/strong\u003e Standardized protocols were used across all legs of the \u003cem\u003eGRE\u003c/em\u003e, the largest coral reef survey to date, with key facets mentioned in Carlton et al. [9]. The benthic data can be found in a companion article [8], where we took a similar modeling approach to instead predict the whereabouts of unexplored reefs with high coral cover. The environmental and ecological parameters assessed are detailed in Table 1; readers interested in Palau’s windward reefs, which were not visited due to high seas, should consult Colin [14]. Although 86 sites were surveyed (see Fig. 1 of Mayfield \u0026amp; Dempsey [8].), \u003cem\u003eP. acuta\u003c/em\u003e-like corals were sampled from only 37. Briefly, ~ 50-mg branches were removed by bone-cutting pliers from each of 150 colonies spanning a number of habitat characteristics (Table\u0026nbsp;1). Care was taken to ensure at least 10-m separation between colonies to reduce risk of sampling clonemates. Each biopsy was broken in two: one was placed in a cryo-tube and rapidly frozen in a liquid nitrogen (-150ºC) dry-shipper (Chart MVE, USA), and the second was coated with 1.5-ml of RNALater™ (Life Technologies, USA) and frozen at -80ºC for 1–2 weeks before later transferring to the dry-shipper. The samples were exported under a CITES permit issued by Palau’s Bureau of Marine Resources and then imported to Taiwan several days later.\u003cbr\u003e\u003cstrong\u003e2.2. Coral biopsy extractions \u0026amp; molecular response variables.\u003c/strong\u003e In addition to four physiological response variables that were documented \u003cem\u003ein situ\u003c/em\u003e (or via assessment of images on a computer)-maximum (max.) colony length, planar surface area, color [15], and polyp extension (expanded or retracted)- several biomarkers were measured to calculate the CHI after extracting RNA, DNA, and protein from a subset of 118 of the 150 biopsies as described previously [16]. The co-extracted DNAs were used for four purposes. First, a multiplex Taqman® assay from Thomas et al. [17] was employed to species-type the coral hosts. From this same DNA aliquot, the Symbiodiniaceae assemblage (to genus-level) was elucidated as in Correa et al. [18]. Next, the Symbiodiniaceae genome copy proportion (GCP), a molecular proxy for endosymbiont density, was calculated via real-time PCR as in prior works [19]. Finally, the microbial communities associated with the sampled corals are now being analyzed via high-throughput metagenomic sequencing at the National University of Singapore. The co-isolated proteins were precipitated in acetone and stored at -80ºC for a future comparison of light vs. dark proteomes (a subset of colonies was sampled both during the day \u0026amp; again at night.). Coral host and dinoflagellate endosymbiont biomarkers determined to be stress-responsive from prior transcriptomic analyses [20] are discussed in detail with respect to their use in calculation of the CHI in prior works [11–12] and are shown in Table\u0026nbsp;2.\u003cbr\u003e\u003c/p\u003e\n\u003cdiv\u003e\n \u003ctable id=\"Tab1\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv\u003eTable 1\u003c/div\u003e\n \u003cdiv\u003e\n \u003cp\u003eFactors predicted to influence the coral health index (CHI) and coral lifespan. With the exception of the Symbiodiniaceae (Sym) assemblage, all other parameters can be determined by eye, via cheap (\u0026lt;$100USD) meters deployed \u003cem\u003ein situ\u003c/em\u003e (e.g., temperature \u0026amp; photosynthetically active radiation), or \u003cem\u003ein silico\u003c/em\u003e (e.g., ImageJ [NIH, USA] for determining colony size). NA = not applicable.\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003ccolgroup cols=\"4\"\u003e\u003c/colgroup\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003eName\u003cbr\u003e\u003c/th\u003e\n \u003cth align=\"left\"\u003eAbbreviation\u003cbr\u003e\u003c/th\u003e\n \u003cth align=\"left\"\u003eUnits\u003cbr\u003e\u003c/th\u003e\n \u003cth align=\"left\"\u003eOptions\u003cbr\u003e\u003c/th\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003cth align=\"left\" colspan=\"3\"\u003ePHYSIOLOGICAL FACTORS (PHYS)\u003cbr\u003e\u003c/th\u003e\n \u003cth align=\"left\"\u003e\u0026nbsp;\u003cbr\u003e\u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003emaximum colony length\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003emax. length\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003ecm\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003econtinuous (range = 4–41 cm)\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003eplanar surface area\u003csup\u003ea\u003c/sup\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eplanar SA\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003ecm\u003csup\u003e2\u003c/sup\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003econtinuous (range = 9-508 cm\u003csup\u003e2\u003c/sup\u003e)\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003ecolony color\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003ecolor\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e1–5; %\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eas either a mean score for the colony (1–5) or a percentage of the colony with anomalous pigmentation (see Appendix A for details.)\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003eendosymbiont assemblage\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eSym assemblage\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u003cem\u003eDurusdinium\u003c/em\u003e, \u003cem\u003eCladocopium\u003c/em\u003e, \u003cem\u003eSymbiodinium\u003c/em\u003e, or combinations thereof\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003epolyps extended?\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e--\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eNA\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eyes or no\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colspan=\"3\"\u003e\u003cstrong\u003eENVIRONMENTAL FACTORS (ENV)\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003elatitude\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eLat\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eºN\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003econtinuous (range = 6.92–8.17ºN)\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003elongitude\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eLon\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eºE\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003econtinuous (range = 134–135ºE)\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003eisland\u003csup\u003ea\u003c/sup\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e--\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eNA\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003esee Carlton et al. [9] (n = 10).\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003esurvey site\u003csup\u003ea\u003c/sup\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003esite\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003esee Fig.\u0026nbsp;1 of Mayfield \u0026amp; Dempsey [8].\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003elagoon\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e--\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eNA\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003einside or outside\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003efore reef\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e--\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eNA\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eis or is not\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003ereef emergence\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eemergence\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eNA\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003esubmerged or emergent\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003ereef exposure\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eexposure\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eNA\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eprotected, intermediate, or exposed\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003ereef location\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003elocation\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eNA\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003elagoon, back reef, or fore reef\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003ereef type\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003etype\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eNA\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003efringing, barrier, patch, back, or channel\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003etime (binned)\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003etime\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026lt; 10:00, 10:00–14:00, 14:00–18:30, or \u0026gt; 18:30\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003enight?\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e--\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eNA\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eday vs. night\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003etemperature\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003etemp.\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eºC\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003econtinuous (range = 26-29.4ºC)\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003esalinity\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e--\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003enone\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003econtinuous (range = 31-34.4)\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003ephotosynthetically active radiation\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003ePAR\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eµmol m\u003csup\u003e− 2\u003c/sup\u003e s\u003csup\u003e− 1\u003c/sup\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003econtinuous (range = 0-335 µmol m\u003csup\u003e− 2\u003c/sup\u003e s\u003csup\u003e− 1\u003c/sup\u003e)\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003ecolony depth\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003edepth\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003em\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003econtinuous (range = 5–31 m). As binned in 5-m increments in certain analyses (\u0026lt; 5 m, 5–10 m, etc.)\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colspan=\"4\"\u003e\u003cstrong\u003eECOLOGICAL (BENTHIC) FACTORS (ECO)\u003c/strong\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e% barren substrate\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003ePB\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e%\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003econtinuous (range = 0.5–42%).\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e% non-coral invertebrates\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003ePINVS\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e%\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003econtinuous (range = 1.2–20.6%)\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e% algal cover by taxon\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e6 taxa\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e%\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003esee online supplemental data file (Appendix C).\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e% coral cover by genus\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e73 genera\u003csup\u003eb\u003c/sup\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e%\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003esee online supplemental data file (Appendix C).\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003etotal live coral cover\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eLCC\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e%\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003econtinuous (range = 13–70%). As binned in certain analyses: 10–20%, 20–30%, 30–40%, etc.\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003etotal live algal cover\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eLAC\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e%\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003econtinuous (range = 12–71%)\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003ecoral/algae ratio\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003ecoral/algae\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eunitless (range = 0.2–5.1)\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e#coral genera present\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003esum(#genera)\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003en\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003econtinuous (range = 17–45 genera)\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e#coral genera present-binned\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003esum(#genera)-binned\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003en\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003ein 5-genera increments: \u0026lt;5 genera, 5–10 genera, 10–15 genera, etc.\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003e\u003csup\u003ea\u003c/sup\u003eExcluded from most analyses. \u003csup\u003eb\u003c/sup\u003eOnly 63 were observed in the vicinity of the sampled coral colonies.\u003c/p\u003e\n\u003cdiv\u003e\n \u003ctable id=\"Tab2\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv\u003eTable 2\u003c/div\u003e\n \u003cdiv\u003e\n \u003cp\u003eDiagnostic results for a representative sick coral: PaPd21.1-2. Although the same data were collected from pocilloporids of other \u003cem\u003eGlobal Reef Expedition\u003c/em\u003e missions, the reference (Ref) range is local to Palau. The colony color score, % of colony bleached/necrotic, and Symbiodiniaceae (Sym) genome copy proportion (GCP) were used to calculate the “color metric” as described in Appendix A. Full gene names are as follows: copper-zinc superoxide dismutase (\u003cem\u003ecu-zn-sod\u003c/em\u003e), green fluorescent protein-like chromoprotein (\u003cem\u003egfp-cp\u003c/em\u003e), ubiquitin ligase (\u003cem\u003eubiq-ligase\u003c/em\u003e), heat shock protein 90 (\u003cem\u003ehsp90\u003c/em\u003e), catalase-peroxidase (\u003cem\u003ekatG\u003c/em\u003e), \u0026amp; zinc-induced facilitator-like 1-like (\u003cem\u003ezifl1l\u003c/em\u003e). NA = not applicable. NM = not measured.\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003ccolgroup cols=\"5\"\u003e\u003c/colgroup\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003eResponse variable\u003cbr\u003e\u003c/th\u003e\n \u003cth align=\"left\"\u003eValue\u003cbr\u003e\u003c/th\u003e\n \u003cth align=\"left\"\u003eRef range\u003cbr\u003e\u003c/th\u003e\n \u003cth align=\"left\"\u003eIn spec?\u003cbr\u003e\u003c/th\u003e\n \u003cth align=\"left\"\u003eDiagnosis/notes\u003cbr\u003e\u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003ecolony color score\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e5\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026gt; 4\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eyes\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003enormal appearance\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e% of colony bleached/necrotic\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e0\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026lt; 10\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eyes\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eno evidence of bleaching or disease\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003emax. colony length\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e27\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e(4,41)\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eyes\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003eplanar surface area\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e220\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e(9,508)\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eyes\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003epolyps extended?\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eyes\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eyes\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eyes\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colspan=\"5\"\u003e\u003cstrong\u003eThe following 10 response variables were used to calculate the heat map score\u003c/strong\u003e (see Appendix A.):\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003eRNA/DNA ratio\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e1.93\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026gt; 0.5\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eyes\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003eSym GCP\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e0.05\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e(0.05,5)\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eborderline\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003elow endosymbiont densities despite normal appearance; evidence for non-dinoflagellate algae?\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003ehost coral \u003cem\u003ecu-zn-sod\u003c/em\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e0.91\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e(0.5,1.5)\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eyes\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003ehost coral \u003cem\u003egfp-cp\u003c/em\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e0.05\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e(1,6)\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003etoo low\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003ecould point to lack of evidence for self-shading\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003ehost coral cytochrome P450\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e0.01\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e(0.02,0.4)\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003etoo low\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003ecould point to problems with host metabolism\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003ehost coral catalase\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e1.5\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e(0.5,5)\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eyes\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003eSym \u003cem\u003eubiq-lig\u003c/em\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e10.2\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e(0.01,0.3)\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003etoo high\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003estrong evidence for endosymbiont stress response\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003eSym \u003cem\u003ehsp90\u003c/em\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e15.9\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e(0.1,10)\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003etoo high\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003estrong evidence for endosymbiont stress response\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003eSym \u003cem\u003ekatG\u003c/em\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e1.1\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e(0.01,5)\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eyes\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003eSym \u003cem\u003ezifl1l\u003c/em\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eNM\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e(1,100)\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eNA\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eextracted insufficient RNA to measure this mRNA\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colspan=\"5\"\u003e\u003cstrong\u003eThe following four metrics were used to calculate the CHI\u003c/strong\u003e:\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003eheat map score\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e3\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e(0,2)\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003etoo high\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003ecolor metric\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e0.17\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026gt;-0.7\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eyes\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003evariability index\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e2.8\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026lt; 1\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003etoo high\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eevidence for loss of control of homeostasis\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003eMahalanobis distance\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eNM\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e(1,4)\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003eNA\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003ecould not calculate due to missing data\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003eCHI\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e0.8\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026gt; 1\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003etoo low\u003cbr\u003e\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u003cstrong\u003ediagnosis\u003c/strong\u003e: colony is abnormally stressed\u003cbr\u003e\u003cstrong\u003eprognosis\u003c/strong\u003e: colony will not survive future heat wave\u003cbr\u003e\u003cstrong\u003eoutcome\u003c/strong\u003e: colony perished after subsequent heat wave\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003e\u003cstrong\u003e2.3. CHI.\u003c/strong\u003e The CHI was calculated as described previously [21] to allow for comparability with other pocilloporids sampled during the \u003cem\u003eGRE\u003c/em\u003e. Briefly, it is calculated from the “heat map score” (Table\u0026nbsp;1), the “color metric,” the variability index, and the Mahalanobis distance. Details can be found in the online supplemental methods (i.e., Appendix A).\u003cbr\u003e\u003cstrong\u003e2.4. Predictive modeling.\u003c/strong\u003e An approach analogous to that of Mayfield et al. [13] was taken to attempt to predict the CHI from combinations of cheaper and easier-to-measure environmental and ecological parameters (i.e., “predictors”). These include those of Table\u0026nbsp;1 and subsets thereof. A two-step approach was taken in JMP® Pro (USA), which was used for all statistical and modeling analyses. First, JMP Pro’s “model screening” platform was used to simultaneously generate and compare a large number of models-decision tree, stepwise regression, generalized regression, bootstrap forest, boosted tree, Naïve Bayes, discriminant, XGBoost, support vector machines, partial least squares, k-nearest neighbors, neural network, and others- with CHI as the Y and second-order factorial combinations of subsets of predictors as the X’s (Tables A1-A2 in Appendix B). CHI was modeled three different ways: as a continuous value spanning 0 to 5, as binned into categorical scores of 0, 1, 2, 3, or 4–5 (values rounded to the nearest integer), or as a colony being either weak (CHI ≤ 1) or resilient (\u0026gt; 1). The data table was randomly partitioned into 89 (75%) and 29 (25%) training and validation samples, respectively, and all error terms below represent std. dev. Of the latter 29 colonies, a subset of 22 were fate-tracked for one year (2016–2017) to field-test the models.\u003cbr\u003e\u003c/p\u003e\n\u003cdiv\u003e\n \u003cp\u003eWhen the superior modeling type identified-defined as the one with the highest validation data R\u003csup\u003e2\u003c/sup\u003e for the continuous CHI analyses (acceptance threshold \u0026gt; 0.8) and the highest accuracy for the categorically binned health values (acceptance threshold \u0026gt; 90%)-was a neural network, a tuning GUI developed by Diedrich Schmidt was used to test hundreds to thousands of combinations of the core neural network hyperparameters (e.g., number of hidden layers, learning rate) as described previously [22]. Another analysis was undertaken in which images of the sampled corals at both mm- and cm-scales were analyzed alongside the other predictors to determine whether the powerful Torch AI could uncover image features that are diagnostic of particular health states. In general, Torch did not outperform neural networks (Table A3), though it is important to note that, for image analysis AIs to be properly leveraged, care must be taken to standardize image parameters \u003cem\u003ein situ\u003c/em\u003e, such as distance from the colony, light level, etc. Many of these conditions might only lend themselves to laboratory analyses at present.\u003c/p\u003e\n\u003c/div\u003e"},{"header":"3. Results","content":"\u003cp\u003e\u003cstrong\u003e3.1. Data access \u0026amp; coral species-typing.\u003c/strong\u003e All numerical data presented can be found in the online supplemental data file (Appendix C), and images of the sampled corals, as well as their surrounding habitats, can be found here. Precipitated RNAs, DNAs, and proteins (all frozen at -80\u0026ordm;C), as well as decalcified fragments of a subset of 30 samples embedded in paraffin, are available for sharing; their whereabouts and other key information (e.g., sample size) are described in Mayfield [\u003cspan class=\"CitationRef\"\u003e23\u003c/span\u003e]. All sampled colonies were typed as \u003cem\u003ePocillopora acuta\u003c/em\u003e, and not \u003cem\u003ePocillopora damicornis\u003c/em\u003e; based on our benthic survey data, it is doubtful that \u003cem\u003eP. damicornis\u003c/em\u003e is present in the west of Palau. To verify whether the probe-based assay of Thomas et al. [\u003cspan class=\"CitationRef\"\u003e17\u003c/span\u003e] was accurate, the mitochondrial open reading frames (mORF) of three random DNA samples were PCR-amplified and sequenced in both directions, and all were 100% matches to published \u003cem\u003eP. acuta\u003c/em\u003e mORF sequences on the NCBI database.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.2. Predictor screening.\u003c/strong\u003e Reef site was found in \u0026gt;\u0026thinsp;42 of the 100 random bootstrap forest models generated during the predictor screen and was the most influential predictor of CHI (Fig. \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003ea); however, because our goal was to produce a generalizable model that could be used to find resilient corals in sites \u003cem\u003enot-yet-surveyed\u003c/em\u003e, we removed this term from subsequent analyses. Of the remainders, Symbiodiniaceae assemblage requires molecular benchwork and is discussed in more detail below. Max. colony length and surface area co-vary since both are proxies of colony size; as such, their relative influences on the CHI are similar. \u003cspan\u003e\u003cstrong\u003e3.3. Response screening.\u003c/strong\u003e A response screen is a false discovery rate (FDR)-controlled means of looking at each individual predictor\u0026rsquo;s influence on a Y, and only a single predictor, \u003cem\u003ePachyseris\u003c/em\u003e spp. cover (Fig. \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003eb), was associated with a \u003cem\u003ep\u003c/em\u003e-value\u0026thinsp;\u0026lt;\u0026thinsp;0.01. However, this correlation was not statistically at the FDR-adjusted \u003cem\u003ep\u003c/em\u003e-value of 0.1, and cover of this genus explained only\u0026thinsp;~\u0026thinsp;8\u0026ndash;9% of the variation in the CHI. Although it may be tempting to interpret this inverse relationship as evidence for competition, confounding factors are likely responsible; \u003cem\u003ePachyseris\u003c/em\u003e is a low-light coral found at depths\u0026thinsp;\u0026gt;\u0026thinsp;20 m, whereas \u003cem\u003eP. acuta\u003c/em\u003e is more common in the shallows. Those areas where the former is relatively more abundant are likely to be associated with lower CHI scores for\u0026nbsp;\u003cem\u003eP. acuta\u003c/em\u003e.\u003cbr\u003e\u003c/span\u003e\u003c/p\u003e\n\u003cdiv class=\"BlockQuote\"\u003e\n \u003cp\u003e\u003cstrong\u003e3.4. Variation in CHI\u003c/strong\u003e. Because reef site was the top predictor in the predictor screen, we plotted CHI across an 0.1 x 0.1-\u0026ordm; grid (Fig. \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003ea). Although variability can be observed, there were no clear trends with respect to latitude, longitude, proximity to land, or other geographic variables that would be useful to managers. There was also no correlation between CHI and either total live coral (Fig. \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003eb) or pocilloporid cover (Fig. \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003ec); simply documenting the abundance of these corals, as is the extent of most survey efforts [\u003cspan class=\"CitationRef\"\u003e24\u003c/span\u003e], is insufficient for making predictions of climate resilience. In fact, the few remaining corals in a highly marginalized habitat could be expected to \u003cem\u003eout-perform\u003c/em\u003e conspecifics from relatively less impacted areas [\u003cspan class=\"CitationRef\"\u003e25\u003c/span\u003e].\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv class=\"BlockQuote\"\u003e\n \u003cp\u003e\u003cstrong\u003e3.5. A neural network for predicting CHI-a: 23-predictor model (\u003c/strong\u003eFig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e\u003cstrong\u003e).\u003c/strong\u003e Because the CHI (as continuous \u0026amp; binned as five integers) could not be confidently predicted (R\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.8 \u0026amp; misclassification rate\u0026thinsp;\u0026gt;\u0026thinsp;10%, respectively; Table A1), corals were scored as either resilient (CHI\u0026thinsp;\u0026gt;\u0026thinsp;1) or weak (CHI\u0026thinsp;\u0026le;\u0026thinsp;1; \u0026ldquo;sick\u0026rdquo; in some tables \u0026amp; figures), and models approached 97% accuracy when including virtually all environmental and ecological predictors (Table A2); 100% of the models surpassing the 90% accuracy cutoff were neural networks. A more parsimonious model is shown in Fig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e. In this single-hidden layer, boosted neural network featuring Gaussian activation (Fig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003ea), 91% accuracy was achieved in predicting whether a coral would perish from or resist bleaching, and only 23 environmental factors were incorporated as X\u0026rsquo;s; this represents only a\u0026thinsp;~\u0026thinsp;6% drop in accuracy whilst needing\u0026thinsp;~\u0026thinsp;four-fold less data than for the 97%-accuracy model. In the subset of field test samples (n\u0026thinsp;=\u0026thinsp;22 colonies), 5 (25%) were predicted to perish by summer of 2017, and all did (100% accuracy); 2 of the remaining 17 were \u0026ldquo;false-positives:\u0026rdquo; predicted to be climate-resilient but actually died (89.5% accuracy or 91% overall accuracy across 22 colonies).\u003c/p\u003e\n \u003cp\u003eThe 23 environmental factors featured in this model have been listed with respect to their total effect in the independent uniform inputs variable importance analysis in Fig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003eb. A desirability analysis was conducted in which the AI was prompted to generate the conditions resulting in the highest probability of a coral being climate-resilient (\u0026ldquo;\u003cem\u003ep\u003c/em\u003e(resilient\u0026rdquo;= maximized); T\u003csup\u003e2\u003c/sup\u003e extrapolation control was implemented to ensure that environmentally unrealistic conditions were not generated (e.g., an exposed lagoonal patch reef). By modifying the 23 predictors, a scenario was forecasted in which there is near-100% certainty that a coral would be climate-resilient; the results of the top-five predictors and their optimal levels are shown in Fig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003ec. Coral cover had the greatest weight in the model, despite the fact that there was no clear relationship between CHI and coral cover in the simple, univariate approaches (Fig. \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003eb-c). However, the CHI did not vary across coral cover bins less than 60%; in other words, \u003cem\u003eP. acuta\u003c/em\u003e resilience only plummets on reefs whose total live coral cover is \u0026gt;\u0026thinsp;60% (15 out of the 86 surveyed reefs). As reefs with coral cover over 60% would be very space-limited environments, competition with other corals could account for this decline in CHI.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv class=\"BlockQuote\"\u003e\n \u003cp\u003eOf the remaining four predictors in Fig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003ec, the horizontal line for summed coral genera is deceiving because the variable importance analysis clearly points to an influence of this diversity metric. This discrepancy is in part due to the fact that generic diversity covaries with several other environmental factors, such as coral cover (R\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.05); changes in other parameters confound that of generic diversity to where, under the projected conditions of the desirability analysis, \u003cem\u003ep\u003c/em\u003e(resilient) does not vary. When performing a simulation (10,000 rows) and allowing generic diversity to fluctuate randomly across the entire range documented in Palau, there is a\u0026thinsp;~\u0026thinsp;6% decrease in \u003cem\u003ep\u003c/em\u003e(resilient) upon increasing the generic diversity from the optimal level of 27 genera to 41 genera (1.5-fold increase); in other words, generic diversity is a relatively weak predictor of \u003cem\u003eP. acuta\u003c/em\u003e climate resilience when assessed by itself.\u003c/p\u003e\n \u003cp\u003eThe optimal colony diameter of 14 cm (Fig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003ec) is significantly different from the mean colony size in Palau of 17 cm (\u0026plusmn;\u0026thinsp;8 cm; \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.0001). Since a typical colony can extend its diameter by ~\u0026thinsp;2 cm per year [\u003cspan class=\"CitationRef\"\u003e26\u003c/span\u003e], this 3-cm difference between the optimal and mean coral diameter represents a\u0026thinsp;~\u0026thinsp;1.5-year age difference. Why a slightly younger and smaller colony would have higher climate resilience is currently unclear, and upon looking at the various physiological response variables assessed (e.g., gene expression, Symbiodiniaceae density, etc.), few showed variation across size. Therefore, a simulation was run in which the other optimized parameters from the desirability analysis were fixed at their optimal levels while max. colony length was allowed to vary randomly across its Palauan distribution. The results (Fig. 4a) reveal that \u003cem\u003ep\u003c/em\u003e(resilient) actually does not vary much once a colony reaches\u0026thinsp;~\u0026thinsp;10 cm. There is a \u0026ldquo;danger zone\u0026rdquo; of 2\u0026ndash;5 cm, possibly due to corallivory or lack of ability to effectively compete whilst at such a small size. Note that while Fig. 4a corroborates the findings of Fig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003ec at the lower end of the size range, the latter analysis shows another increase in the probability of colony being sick beyond about 30 cm that is only partly reflected in Fig. 4a. This discrepancy is likely due to the low number of large colonies sampled (only one was \u0026gt;\u0026thinsp;30 cm in diameter.).\u003c/p\u003e\n \u003cp\u003eColony depth was also weighted highly in the superior predictive model of coral resilience (Fig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003eb), and the optimal depth was 10.8 m (PAR\u0026thinsp;=\u0026thinsp;91 \u0026micro;mol m\u003csup\u003e-2\u003c/sup\u003e s\u003csup\u003e-1\u003c/sup\u003e), significantly shallower than the mean sampling depth of 15 m (\u0026plusmn;\u0026thinsp;7; PAR\u0026thinsp;=\u0026thinsp;77 \u0026micro;mol m\u003csup\u003e-2\u003c/sup\u003e s\u003csup\u003e-1\u003c/sup\u003e). However, this is close to the median depth of 10.9 m (due to the highly skewed nature of the dataset, in which most corals were collected in the 8-12-m window). PAR alone did not have a strong impact on CHI or \u003cem\u003ep\u003c/em\u003e(resilient) (Fig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003eb). Wave energy (as \u0026ldquo;exposure\u0026rdquo; in the tables \u0026amp; figures), which would also vary by depth, was also a relatively weak predictor. From the desirability analysis (Fig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003ec), there is not a large decrease in \u003cem\u003ep\u003c/em\u003e(resilient) until the depth increases to about 25 m. To attempt to corroborate this, another simulation was performed (10,000 samples) in which the optimal values were locked for all parameters except depth, which was allowed to fluctuate randomly across its Palauan dataset range (std. dev.=5 m). The results (Fig. 4b) are similar to those of Fig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003ec, albeit with a more gradual tapering at greater depths in the simulation. Another key difference is that the optimal depth in the simulation is notably lower, with \u003cem\u003ep\u003c/em\u003e(resilient) maxing out at around 8 m instead of 11 m.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv class=\"BlockQuote\"\u003e\n \u003cp\u003e\u003cstrong\u003e3.6. A neural network for predicting CHI-b: 22-predictor model (\u003c/strong\u003eFig. \u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003e\u003cstrong\u003e).\u003c/strong\u003e Were one interested in maximizing the chance of selecting a climate-resilient \u003cem\u003eP. acuta\u003c/em\u003e colony in Palau, they would look for colonies around 14 cm in diameter at around 10\u0026ndash;11 m depth in an area of relatively low (for Palau) 1) coral cover (20\u0026ndash;30%) and 2) generic diversity (25\u0026ndash;30 genera). All of these predictors can be assessed by diving equipment (depth gauge), simple scientific instruments like rulers (colony size), or trained surveyors (coral cover \u0026amp; generic diversity). However, there is a drawback to the model of Fig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e; despite the fact that Symbiodiniaceae assemblage has a relatively low total effect size (~\u0026thinsp;4-fold lower than coral cover, for instance), if removed from the model, the accuracy could decrease. Because Symbiodiniaceae assemblage requires laboratory benchwork that normally occurs days to weeks after a field trip is completed, it is not a desirable predictor for one looking for a fast, cheap means of identifying resilient corals. For this reason, the term was removed, and a similarly structured neural network was rebuilt (Fig. \u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003ea); upon removing this expensive, difficult-to-measure response variable, the accuracy decreased to 89%, just lower than the \u003cem\u003ea priori\u003c/em\u003e accuracy cutoff of 90%. This 2% decrease in accuracy may be worth sacrificing given that the resulting model features only metrics that can either be assessed \u003cem\u003ein situ\u003c/em\u003e or shortly after the dive upon analysis of image and benthic survey data. In fact, the top-five predictors are the same as in the model of Fig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e, albeit with a different relative ordering. As with the 23-predictor model (Fig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003ec), the desirability analysis (Fig. \u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003eb) recommends a relatively small colony (11 vs. 14 cm in Fig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e) on a low coral cover reef (20\u0026ndash;30% coral cover in both models) with relatively low coral diversity (20\u0026ndash;25 genera vs. 27 genera in Fig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e). The main difference between the two models is the recommended depth: 18.5 vs. \u0026lt;11 m in the models of Figs. \u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003e and \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e, respectively.\u003c/p\u003e\n\u003c/div\u003e"},{"header":"4. Discussion","content":"\u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eIn even the top two neural networks, post-survey analyses must be performed to calculate coral cover and coral generic diversity; whether or not a particular reef habitat is colonized by resilient \u003cem\u003eP. acuta\u003c/em\u003e specimens would, in other words, not be known until after the dive, and a repeat visit would be necessary to collect samples (were the conditions indeed even suggestive of resilient corals). What is more likely to be the case is that a scientist or manager will have some pre-existing data on certain reefs, such as a long-term monitoring site. In this case, the relevant values can be put into the model of Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e5\u003c/span\u003e (as a GUI) to calculate the probability that a particular habitat might be characterized by resilient pocilloporids. As a hypothetical example, say that a surveyor determined that a particular reef area is characterized by a coral cover of 40\u0026ndash;50% with 25\u0026ndash;30 coral genera present and pocilloporid corals found around 15 m. By inputting these values into the model of Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e5\u003c/span\u003e and setting them as \u0026ldquo;fixed,\u0026rdquo; we can then re-calculate \u003cem\u003ep\u003c/em\u003e(resilient) across differing colony sizes. Under these conditions, a colony of 17-cm diameter would have a 91% chance of being climate-resilient (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e6\u003c/span\u003ea); were coral cover and colony size instead allowed to vary, while the other 20 environmental parameters are fixed at their optimal levels, a contour plot (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e6\u003c/span\u003eb) instead provides a means of searching for resilient corals.\u003c/p\u003e \u003cp\u003eHowever, the model of Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e5\u003c/span\u003e is contingent upon measuring all 22 environmental factors; what if there is insufficient time to do so? Some can be known from consulting maps (e.g., reef type, reef exposure), yet others require divers or at least drone or satellite imagery. If one knew all mapping-related parameters-latitude, longitude, lagoon (inside vs. outside), reef type, reef emergence, reef exposure, forereef (is or is not a forereef)- in addition to coral cover and pocilloporid colony depth, but did \u003cem\u003enot\u003c/em\u003e know the abundance of any other organism or anything about the seawater quality, the probability of correctly finding a climate-resilient pocilloporid would be ~\u0026thinsp;71%. If temperature and coral polyp extension were also known, the probability increases to 75%. If one did not know the sizes of the pocilloporid colonies, the probability decreases only slightly to 72%. Whether or not a proper survey of the benthos is worth the effort of increasing the probability of correctly sourcing resilient corals from 75 to ~\u0026thinsp;90% will ultimately be up to the practitioner.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eThe value of an AI trained with data from 2016 may be suspect, especially considering that coral reefs have become increasingly marginalized since then; are predictions that were accurate in 2016\u0026ndash;2017 of any use to us nearly 10 years later? Under the most pessimistic view, this dataset might only have value in highlighting what these reefs looked like when they were still dominated by stony corals; coral cover averaged\u0026thinsp;~\u0026thinsp;45% on Palau\u0026rsquo;s leeward reefs at that time, which is relatively high by global standards. Of the 118 pocilloporids analyzed in detail, 36 (31%) were predicted to perish from climate change-related temperature increases and/or local stressors; this includes 5 fate-tracked colonies, all of which died (100% accuracy). Of the 22 fate-tracked colonies, 17 were predicted to be climate-resilient, though 2 of these were found dead in follow-up surveys in 2017 (\u0026ldquo;false-positives\u0026rdquo;). This means that the overall field accuracy of 20/22 (91%) is similar to the validation sample accuracy (i.e., samples simply held back from the AI). Follow-up surveys must now be conducted to determine whether the additional 65 colonies predicted to be climate-tolerant remain so to this day, though as was the case with most \u003cem\u003eGRE\u003c/em\u003e survey sites, the majority of the reefs were difficult to reach and had never before been surveyed or sampled at the time. Hopefully, funds from international initiatives like CORDAP can be used by local scientists to periodically check on these corals such that the long-term predictive capacity of the models can be validated.\u003c/p\u003e \u003cp\u003eIf these models are only adept at identifying corals that survive only, for instance 1\u0026ndash;2 years longer than conspecifics, they may ultimately be of little utility to those looking to stock their nurseries with the most robust corals. However, others have predicted that the thermotolerance of some Palauan corals might actually be increasing over time [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e], though it is important to note that these models were not ground-truthed and do not consider key physiological benchmarks of resilience, only organismal abundance. Just as a city with more people is not associated with more resilient citizens, high-coral cover reefs are no more climate-resilient; local acroporids in particular have struggled to recover after recent disturbances [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]. As it is unlikely that adaptation is keeping pace with the rapid rate of seawater temperature rise in Palau, the degree to which accelerated evolution-based approaches may play a role in boosting coral resilience is currently under investigation by scientists from The Nature Conservancy (Palau; Y. Golbuu), Newcastle University (UK; J. Guest), and the University of Queensland (Australia; P. Mumby) through a CORDAP \u0026ldquo;Coral Accelerator Program\u0026rdquo; award. Perhaps the most resilient \u003cem\u003eP. acuta\u003c/em\u003e colonies uncovered in our surveys could be cross-bred to yield more thermotolerant larvae for later reef reseeding efforts.\u003c/p\u003e \u003cp\u003eAside from our omission of samples from Palau\u0026rsquo;s windward reefs, another limitation of this study is that it features only a single species: \u003cem\u003eP. acuta\u003c/em\u003e. Very different optimal conditions will be associated with other corals, and it is entirely possible that model accuracy for many of them will be too low to be of any value. We know more about \u003cem\u003eP. acuta\u003c/em\u003e than any other coral [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e], and molecules-to-satellites datasets do not exist for any other scleractinian except \u003cem\u003eSeriatopora hystrix\u003c/em\u003e [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]; the accuracies presented herein likely represent amongst the highest that could be expected for predicting coral resilience to climate change. This is not to say that resilient corals have not been identified in Palau via other approaches. For instance, thermotolerant corals have been found in sheltered regions of the Rock Islands, such as Nikko Bay [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e], from where corals were also sampled herein, and the \u003cem\u003ePorites lobata\u003c/em\u003e genotypes living there are distinct from conspecifics on the outer reefs [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]. Our species-level molecular analysis was too crude to resolve individual \u003cem\u003eP. acuta\u003c/em\u003e populations, so the DNAs have been sent to colleagues to analyze host coral, dinoflagellate, and microbial genetics at a higher resolution to determine the degree to which adaptation is driving the marked divergence in climate resilience documented [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e"},{"header":"5. Conclusions","content":"\u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eBy tapping into a dataset featuring molecular and physiological data, as well as simple, easy-to-measure survey parameters, we generated two machine-learning-based models for accurately predicting the climate resilience of the model coral \u003cem\u003eP. acuta\u003c/em\u003e in Palau. Accuracies were ~\u0026thinsp;90%, and colony size, depth, and local coral cover and diversity were all important predictors of pocilloporid resilience. Host coral genetics surely play a role (i.e., adaptation), as well, especially given the wide diversity of habitats and vast spatial extent of the survey [\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]. Although a colony\u0026rsquo;s genotype could likely never be discerned \u003cem\u003ein situ\u003c/em\u003e by a diver, knowing whether distinct populations tend to be those of highest resilience, regardless of their habitat preferences and physiological characteristics, will nevertheless be useful information for local practitioners, especially those looking to ensure that restored reefs are both \u0026ldquo;climate-proofed\u0026rdquo; and diverse [\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003eThe following abbreviations are used in this manuscript. See also Table 1 for non-standard abbreviations and Table 2 for full gene names of mRNA biomarkers.\u0026nbsp;\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"524\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 80px;\"\u003e\n \u003cp\u003eCHI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 443px;\"\u003e\n \u003cp\u003eCoral health index\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 80px;\"\u003e\n \u003cp\u003eCORDAP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 443px;\"\u003e\n \u003cp\u003eCoral Research and Development Accelerator Platform\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 80px;\"\u003e\n \u003cp\u003eEF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 443px;\"\u003e\n \u003cp\u003eEnvironmental factor (i.e., predictor)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 80px;\"\u003e\n \u003cp\u003eFDR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 443px;\"\u003e\n \u003cp\u003eFalse discovery rate\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 80px;\"\u003e\n \u003cp\u003eGCP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 443px;\"\u003e\n \u003cp\u003eGenome copy proportion\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 80px;\"\u003e\n \u003cp\u003e\u003cem\u003eGRE\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 443px;\"\u003e\n \u003cp\u003e\u003cem\u003eGlobal Reef Expedition\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 80px;\"\u003e\n \u003cp\u003eGUI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 443px;\"\u003e\n \u003cp\u003eGraphical user interface\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 80px;\"\u003e\n \u003cp\u003eHL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 443px;\"\u003e\n \u003cp\u003eHidden layer\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 80px;\"\u003e\n \u003cp\u003eHMS\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 443px;\"\u003e\n \u003cp\u003eHeat map score\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 80px;\"\u003e\n \u003cp\u003emORF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 443px;\"\u003e\n \u003cp\u003eMitochondrial open reading frame\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 80px;\"\u003e\n \u003cp\u003eNN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 443px;\"\u003e\n \u003cp\u003eNeural network\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 80px;\"\u003e\n \u003cp\u003ePAR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 443px;\"\u003e\n \u003cp\u003ePhotosynthetically active radiation\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 80px;\"\u003e\n \u003cp\u003eSym\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 443px;\"\u003e\n \u003cp\u003eSymbiodiniaceae\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eSupplementary Materials:\u0026nbsp;\u003c/strong\u003eThe following supporting information can be downloaded at: www. xxx/, Appendix A: supplemental methods; Appendix B: supplemental tables (A1-A3); Appendix C: online supplemental data file. Appendices A-B are in the same Word file.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Contributions:\u003c/strong\u003e Conceptualization, A.B.M.; methodology, A.B.M., A.C.D.; software, A.B.M., A.C.D.; validation, A.B.M., A.C.D.; formal analysis, A.B.M., A.C.D; investigation, A.B.M., A.C.D.; resources, A.B.M., A.C.D.; data curation, A.B.M., A.C.D.; writing\u0026mdash;A.B.M.; writing\u0026mdash;review and editing, A.B.M.; visualization, A.B.M.; project administration, A.B.M.; funding acquisition, A.B.M., A.C.D. All authors have read and agreed to the published version of the manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding:\u003c/strong\u003e This research was funded by the Khaled bin Sultan Living Oceans Foundation, as well as the Coral Research and Development Accelerator Platform (CORDAP).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Availability Statement:\u003c/strong\u003e All data can be accessed via the online supplemental data file (Appendix C). All images of the coral reef habitats surveyed, as well as the sampled coral colonies, can be found on coralreefdiagnostics.com. The GUI necessary to run the neural networks (for those without coding or machine-learning knowledge) are also posted on this coral physiology database. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgments:\u003c/strong\u003e We would like to thank the team at JMP for assistance with predictive modeling and machine-learning. ABM would like to thank the Fulbright and MacArthur Foundations for supporting his stay in Taiwan, where laboratory analyses were conducted.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConflicts of Interest:\u003c/strong\u003e The authors declare no conflicts of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript; or in the decision to publish the results.\u003c/p\u003e\u003cp\u003eConsent to Participate declaration: not applicable.\u003c/p\u003e\n\u003cp\u003eConsent to Publish declaration: not applicable.\u003c/p\u003e\n\u003cp\u003eEthics declaration: not applicable\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eReimer, J.D.; Peixoto, R.S.; Davies, S.W.; Traylor-Knowles, N.; Short, M.L.; Cabral-Tena, R.A.; Burt, J.A.; Pessoa, I.; Banaszak, A.T.; Winters, R.S.; Moore, T.; Schoepf, V.; Kaullysing, D.; Calderon-Aguilera, L.E.; W\u0026ouml;rheide, G.; Harding, S.; Munbodhe, V.; Mayfield, A.B.; Ainsworth, T.; Vardi, T.; Eakin, C.M.; Voolstra, C.R. The Fourth Global Coral Bleaching Event\u0026ndash;where do we go from here? \u003cem\u003eCoral Reefs\u003c/em\u003e \u003cstrong\u003e2024\u003c/strong\u003e, \u003cem\u003e43\u003c/em\u003e, 1121-1125. \u003c/li\u003e\n\u003cli\u003eLoya, Y.; Sakai, K.; Yamazato, K.; Nakano, Y.; Sambali, H.; van Woesik, R. Coral bleaching: the winners and the losers. \u003cem\u003eEcol. Letters\u003c/em\u003e \u003cstrong\u003e2001\u003c/strong\u003e, \u003cem\u003e4\u003c/em\u003e, 122-131.\u003c/li\u003e\n\u003cli\u003eBarshis, D.J.; Ladner, J.T.; Oliver, T.A.; Seneca, F.O.; Traylor-Knowles, N.; Palumbi, S.R. Genomic basis for coral resilience to climate change. \u003cem\u003eProc. Natl. Acad. Sci. USA\u003c/em\u003e. \u003cstrong\u003e2013\u003c/strong\u003e, \u003cem\u003e110\u003c/em\u003e, 1387-1392. https://doi.org/doi:10.1073/pnas.1210224110.\u003c/li\u003e\n\u003cli\u003eMayfield, A.B.; Chan, P.H.; Putnam, H.M.; Chen, C.S.; Fan, T.Y. The effects of a variable temperature regime on the physiology of the reef-building coral \u003cem\u003eSeriatopora hystrix\u003c/em\u003e: results from a laboratory-based reciprocal transplant. \u003cem\u003eJ. Exp. Biol. \u003c/em\u003e\u003cstrong\u003e2012\u003c/strong\u003e, \u003cem\u003e215\u003c/em\u003e, 4183-4195.\u003c/li\u003e\n\u003cli\u003eVoolstra, C.R.; Buitrago-L\u0026oacute;pez, C.; Perna, G.; C\u0026aacute;rdenas, A.; Hume, B.C.; R\u0026auml;decker, N.; Barshis, D.J. Standardized short-term acute heat stress assays resolve historical differences in coral thermotolerance across microhabitat reef sites. \u003cem\u003eGlobal Change\u003c/em\u003e \u003cem\u003eBiol.\u003c/em\u003e \u003cstrong\u003e2020\u003c/strong\u003e,\u003cem\u003e26(8)\u003c/em\u003e, 4328-4343.\u003c/li\u003e\n\u003cli\u003eLiu, G.; Heron, S.F.; Eakin, C.M.; Muller-Karger, F.E.; Vega-Rodriguez, M.; Guild, L.S.; De La Cour, J.L.; Geiger, E.F.; Skirving, W.J.; Burgess, T.F.R.; et al. Reef-scale thermal stress monitoring of coral ecosystems: new 5-km global products from NOAA Coral Reef Watch. \u003cem\u003eRemote Sensing\u003c/em\u003e \u003cstrong\u003e2014\u003c/strong\u003e, \u003cem\u003e6\u003c/em\u003e(11), 11579-11606.\u003c/li\u003e\n\u003cli\u003eBachman, S.D.; Kleypas, J.A.; Erdmann, M.; Setyawan, E. A global atlas of potential thermal refugia for coral reefs generated by internal gravity waves. \u003cem\u003eFront. Mar. Sci. \u003c/em\u003e\u003cstrong\u003e2022\u003c/strong\u003e, \u003cem\u003e9\u003c/em\u003e, 921879. https://doi.org/10.3389/fmars.2022.921879.\u003c/li\u003e\n\u003cli\u003eMayfield, A.B.; Dempsey, A.C. AI-driven coral reef bioprospecting in the Republic of Palau. \u003cem\u003ePlatax \u003c/em\u003e\u003cstrong\u003e2024\u003c/strong\u003e, \u003cem\u003e21\u003c/em\u003e, 1-30.\u003c/li\u003e\n\u003cli\u003eCarlton, R.; Dempsey, A.C.; Lubarsky, K.; Lindfield, S.; Faisal, M.; Purkis, S. \u003cem\u003eGlobal Reef Expedition: The Republic of Palau. Final Report\u003c/em\u003e. Khaled bin Sultan Living Oceans Foundation, Annapolis, MD, USA, \u003cstrong\u003e2020\u003c/strong\u003e; \u003cem\u003eVol. 12\u003c/em\u003e\u003c/li\u003e\n\u003cli\u003ePurkis, S.J.; Gleason, A.C.R.; Purkis, C.R.; Dempsey, A.C.; Renaud, P.G.; Faisal, M.; Saul, S.; Kerr, J.M. High-resolution habitat and bathymetry maps for 65,000 sq. km of Earth\u0026rsquo;s remotest coral reefs. \u003cem\u003eCoral Reefs \u003c/em\u003e\u003cstrong\u003e2019\u003c/strong\u003e, \u003cem\u003e38\u003c/em\u003e, 467-488.\u003c/li\u003e\n\u003c/ol\u003e\n\u003col start=\"11\"\u003e\n\u003cli\u003eMayfield, A.B.; Chen, C.S.; Dempsey, A.C. Biomarker profiling in reef corals of Tonga\u0026rsquo;s Ha\u0026rsquo;apai and Vava\u0026rsquo;u Archipelagos. \u003cem\u003ePLoS ONE \u003c/em\u003e\u003cstrong\u003e2017\u003c/strong\u003e,\u003cem\u003e e0185857\u003c/em\u003e.\u003c/li\u003e\n\u003cli\u003eMayfield, A.B.; Chen, C.S.; Dempsey, A.C. Identifying corals displaying aberrant behavior in Fiji\u0026rsquo;s Lau Archipelago. \u003cem\u003ePLoS ONE \u003c/em\u003e\u003cstrong\u003e2017\u003c/strong\u003e,\u003cem\u003e e0177267\u003c/em\u003e.\u003c/li\u003e\n\u003cli\u003eMayfield, A.B.;Dempsey, A.C.; Chen, C.S.; Lin, C. Expediting the search for climate-resilient reef corals in the Coral Triangle with artificial intelligence. \u003cem\u003eApplied Sci. \u003c/em\u003e\u003cstrong\u003e2022\u003c/strong\u003e, \u003cem\u003e12\u003c/em\u003e, 12955.\u003c/li\u003e\n\u003cli\u003eColin PL (2009) \u003cem\u003eMarine Environments of Palau\u003c/em\u003e, 1st edition. Mutual Publishing, Honolulu, Hawaii, USA, \u003cstrong\u003e2009\u003c/strong\u003e.\u003c/li\u003e\n\u003cli\u003eSiebeck, U.E.; Marshall, N.J.; Kl\u0026uuml;ter, A.; Hoegh-Guldberg, O. Monitoring coral bleaching using a colour reference card. \u003cem\u003eCoral Reefs\u003c/em\u003e \u003cstrong\u003e2006\u003c/strong\u003e, \u003cem\u003e25\u003c/em\u003e, 453-460. https://doi.org/10.1007/s00338-006-0123-8 \u003c/li\u003e\n\u003cli\u003eMayfield, A.B.; Wang, L.H.; Tang, P.C.; Hsiao, Y.Y.; Fan, T.Y.; Tsai, C.L.; Chen, C.S. Assessing the impacts of experimentally elevated temperature on the biological composition and molecular chaperone gene expression of a reef coral. \u003cem\u003ePLoS ONE\u003c/em\u003e \u003cstrong\u003e2011\u003c/strong\u003e, \u003cem\u003ee26529\u003c/em\u003e.\u003c/li\u003e\n\u003cli\u003eThomas, L.; Stat, M.; Evans, R.D.; Kennington, W.J. A fluorescence-based quantitative real-time PCR assay for accurate \u003cem\u003ePocillopora damicornis\u003c/em\u003e species identification. \u003cem\u003eCoral Reefs\u003c/em\u003e \u003cstrong\u003e2016\u003c/strong\u003e,\u003cem\u003e35\u003c/em\u003e, 895-899. \u003c/li\u003e\n\u003cli\u003eCorrea, A.M.S.; McDonald, M.D.; Baker, A.C. Development of clade-specific \u003cem\u003eSymbiodinium \u003c/em\u003eprimers for quantitative PCR (qPCR) and their application to detecting clade D symbionts in Caribbean corals. \u003cem\u003eMar. Biol.\u003c/em\u003e \u003cstrong\u003e2009\u003c/strong\u003e,\u003cem\u003e156\u003c/em\u003e, 2403-2411.\u003c/li\u003e\n\u003cli\u003ePutnam, H.M.;Mayfield, A.B.; Fan, T.Y.; Chen, C.S.; Gates, R.D. The physiological and molecular responses of larvae from the reef-building coral \u003cem\u003ePocillopora damicornis\u003c/em\u003e exposed to near-future increases in temperature and \u003cem\u003ep\u003c/em\u003eCO\u003csub\u003e2\u003c/sub\u003e. \u003cem\u003eMar. Biol. \u003c/em\u003e\u003cstrong\u003e2013\u003c/strong\u003e,\u003cem\u003e160\u003c/em\u003e, 2157-2173.\u003c/li\u003e\n\u003cli\u003eMayfield, A.B.; Wang, Y.B.; Chen, C.S.; Chen, S.H.; Lin, C.Y. Compartment-specific transcriptomics in a reef-building coral exposed to elevated temperatures. \u003cem\u003eMol. Ecol. \u003c/em\u003e\u003cstrong\u003e2014\u003c/strong\u003e,\u003cem\u003e23\u003c/em\u003e, 5816-5830.\u003c/li\u003e\n\u003cli\u003eMayfield, A.B.; Chen, C.S.; Dempsey, A.C. The molecular ecophysiology of closely related pocilloporid corals of New Caledonia. \u003cem\u003ePlatax\u003c/em\u003e \u003cstrong\u003e2017\u003c/strong\u003e,\u003cem\u003e14\u003c/em\u003e, 1-45.\u003c/li\u003e\n\u003cli\u003eMayfield, A.B.Machine-learning-based proteomic predictive modeling with thermally challenged Caribbean reef corals. \u003cem\u003eDiversity \u003c/em\u003e\u003cstrong\u003e2022\u003c/strong\u003e,\u003cem\u003e14\u003c/em\u003e, 33.\u003c/li\u003e\n\u003cli\u003eMayfield, A.B.Establishment of a long-term reef-building coral biopsy archive as a tool for the marine biology research community. \u003cem\u003ePlatax\u003c/em\u003e \u003cstrong\u003e2023\u003c/strong\u003e, \u003cem\u003e20\u003c/em\u003e, 7-30.\u003c/li\u003e\n\u003cli\u003eMcClanahan, T.R.; Donner, S.D.; Maynard, J.A.; MacNeil, M.A.; Graham, N.A.J. Prioritizing key resilience indicators to support coral reef management in a changing climate. \u003cem\u003ePLoS ONE\u003c/em\u003e \u003cstrong\u003e2012\u003c/strong\u003e,\u003cem\u003ee42884\u003c/em\u003e.\u003c/li\u003e\n\u003cli\u003eRubin, E.; Enochs, I.; Foord, C.; Kolodziej, G.; Basden, I.; Manzello, D.P.; Mayfield, A.B. Molecular mechanisms of coral persistence within highly urbanized locations in the Port of Miami, Florida. \u003cem\u003eFront. Mar. Sci. \u003c/em\u003e\u003cstrong\u003e2021\u003c/strong\u003e,\u003cem\u003e8\u003c/em\u003e, 695236. \u003c/li\u003e\n\u003cli\u003eAnderson, K.D.; Cantin, N.E.; Heron, S.F.; Pisapia, C.; Pratchett, M. Variation in growth rates of branching corals along Australia\u0026rsquo;s Great Barrier Reef. \u003cem\u003eSci. Rep. \u003c/em\u003e\u003cstrong\u003e2017\u003c/strong\u003e\u003cem\u003e,\u003c/em\u003e \u003cem\u003e7\u003c/em\u003e, 2920. \u003c/li\u003e\n\u003cli\u003eLachs, L.; Donner, S.D.; Mumby, P.J.; Bythell, J.C.; Humanes, A.; East, H.K.; Guest, J.R. Emergent increase in coral thermal tolerance reduces mass bleaching under climate change. \u003cem\u003eNat. Commun\u003c/em\u003e. \u003cstrong\u003e2023\u003c/strong\u003e,\u003cem\u003e14\u003c/em\u003e, 4939. https://doi.org/10.1038/s41467-023-40601-6\u003c/li\u003e\n\u003cli\u003eLachs, L.; Biondi, P.; Gouezo, M.; Nestor, V.; Olsudong, D.; Guest, J.; Golbuu, Y. Demographic recovery of corals at a wave-exposed reef following catastrophic disturbance. \u003cem\u003eCoral Reefs\u003c/em\u003e \u003cstrong\u003e2024\u003c/strong\u003e,\u003cem\u003e43\u003c/em\u003e, 193-199. https://doi.org/10.1007/s00338-024-02464-1\u003c/li\u003e\n\u003cli\u003eMayfield, A.B.; Fan, T.Y.; Chen, C.S. Physiological acclimation to elevated temperature in a reef-building coral from an upwelling environment. \u003cem\u003eCoral Reefs \u003c/em\u003e\u003cstrong\u003e2013\u003c/strong\u003e,\u003cem\u003e32\u003c/em\u003e, 909-921.\u003c/li\u003e\n\u003cli\u003eMayfield, A.B.;Chen, Y.J.; Lu, C.Y.; Chen, C.S. Exploring the environmental physiology of the Indo-Pacific reef coral \u003cem\u003eSeriatopora hystrix\u003c/em\u003e using differential proteomics. \u003cem\u003eOpen J. Mar. Sci. \u003c/em\u003e\u003cstrong\u003e2018\u003c/strong\u003e,\u003cem\u003e 8\u003c/em\u003e, 223-252.\u003c/li\u003e\n\u003cli\u003eWoesik, R.v.; Houk, P.; Isechal, A.L.; Idechong, J.W.; Victor, S.; Golbuu, Y. Climate-change refugia in the sheltered bays of Palau: analogs of future reefs. \u003cem\u003eEcol. Evol. \u003c/em\u003e\u003cstrong\u003e2012\u003c/strong\u003e, \u003cem\u003e2\u003c/em\u003e(1), 2474-2484. https://doi.org/10.1002/ece3.363.\u003c/li\u003e\n\u003cli\u003eRivera, H.E.; Cohen, A.L.; Thompson, J.R.; Baums, I.B.; Fox, M.D\u003cem\u003e.; \u003c/em\u003eMeyer-Kaiser, K.S. Palau\u0026rsquo;s warmest reefs harbor thermally tolerant corals that thrive across different habitats. \u003cem\u003eCommun. Biol.\u003c/em\u003e \u003cstrong\u003e2022\u003c/strong\u003e,\u003cem\u003e5\u003c/em\u003e, 1394. https://doi.org/10.1038/s42003-022-04315-7\u003c/li\u003e\n\u003cli\u003ePalumbi, S.R.; Walker, N.S.; Hanson, E.; Armstrong, K.; Lippert, M.; Cornwell, B.; Nestor, V.; Golbuu, Y. Small-scale genetic structure of coral populations in Palau based on whole mitochondrial genomes: Implications for future coral resilience. \u003cem\u003eEvol. Appl.\u003c/em\u003e \u003cstrong\u003e2023\u003c/strong\u003e, \u003cem\u003e16\u003c/em\u003e(2), 518-529. doi: 10.1111/eva.13509. \u003c/li\u003e\n\u003cli\u003eBaums, I.B. A restoration genetics guide for coral reef conservation. \u003cem\u003eMol. Ecol\u003c/em\u003e. \u003cstrong\u003e2008\u003c/strong\u003e, \u003cem\u003e17\u003c/em\u003e(12), 2796-2811.\u003c/li\u003e\n\u003cli\u003eQuigley, K.M.; Hein, M.; Suggett, D.J. Translating the 10 golden rules of reforestation for coral reef restoration. \u003cem\u003eConserv. Biol. \u003c/em\u003e\u003cstrong\u003e2022\u003c/strong\u003e, 36(4), e13890. https://doi.org/10.1111/cobi.13890.\u003c/li\u003e\n\u003cli\u003eQuigley, K.M.; Bay, L.K.; van Oppen, M.J.H. Genome-wide SNP analysis reveals an increase in adaptive genetic variation through selective breeding of coral. \u003cem\u003eMol. Ecol\u003c/em\u003e. \u003cstrong\u003e2020\u003c/strong\u003e,\u003cem\u003e29\u003c/em\u003e(12), 2176-2188.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"discover-oceans","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"Learn more about [Discover Oceans](https://www.springer.com/journal/44289)","snPcode":"44289","submissionUrl":"https://submission.nature.com/new-submission/44289","title":"Discover Oceans","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Discover Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"climate change, corals, ecological forecasting, machine-learning, resilience","lastPublishedDoi":"10.21203/rs.3.rs-6162052/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6162052/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eTo restore degraded coral reef habitats, it is critical to ensure that the scleractinian broodstock utilized can withstand future heatwaves. However, reef coral resilience is normally assessed only after catastrophic stress events. By tapping into a rich, \u0026ldquo;molecules-to-satellites\u0026rdquo; dataset acquired during the Living Ocean Foundation\u0026rsquo;s research mission to the Republic of Palau, we trained an artificial intelligence to accurately predict pocilloporid coral thermotolerance from relatively cheap, easy-to-measure environmental and ecological survey parameters. Specifically, a neural network featuring 22 predictors, such as coral cover and colony size, could forecast the whereabouts and properties of climate-resilient colonies of \u003cem\u003ePocillopora acuta\u003c/em\u003e with ~\u0026thinsp;90% accuracy. This machine-learning model enables practitioners to 1) estimate the climate resilience of local pocilloporid populations and 2) identify habitats characterized by high pocilloporid coral resilience.\u003c/p\u003e","manuscriptTitle":"Machine-learning algorithms for identifying climate-resilient corals in the Republic of Palau","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-04-22 14:39:31","doi":"10.21203/rs.3.rs-6162052/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-06-27T13:55:55+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-06-22T06:13:52+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-05-10T02:37:50+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"318727152299718997745852077510069785875","date":"2025-04-18T09:05:32+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"280274686034238023315033303744775785223","date":"2025-04-03T20:09:06+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-04-01T15:51:27+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-03-29T01:55:47+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-03-29T01:54:29+00:00","index":"","fulltext":""},{"type":"submitted","content":"Discover Oceans","date":"2025-03-05T11:08:08+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"discover-oceans","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"Learn more about [Discover Oceans](https://www.springer.com/journal/44289)","snPcode":"44289","submissionUrl":"https://submission.nature.com/new-submission/44289","title":"Discover Oceans","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Discover Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"81cc4777-5504-49e0-83f7-1938536a479b","owner":[],"postedDate":"April 22nd, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2025-09-16T13:08:44+00:00","versionOfRecord":[],"versionCreatedAt":"2025-04-22 14:39:31","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6162052","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6162052","identity":"rs-6162052","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}