Regional dynamics drive differences in future heat stress and reveal where Hawaiian corals are most likely to persist Authors: Kate Feloy, Brian Powell, Tobias Friedrich, Hannah Barkley Published: 2026 ABSTRACT AbstractThe Hawaiian Islands cause significant disruption to North Pacific oceanic and atmospheric flows creating highly localized spatial variability of ocean conditions on coral reefs. This variability affects climate impacts on these corals but is unresolved in global climate models. Using a dynamically downscaled model to resolve future conditions at high temporal (daily) and spatial (4 km) resolution for the Hawaiian Islands, we investigate projected heat stress and the adaptation rate required for corals to withstand projected end of century warming for three different emission scenarios. In the region, the Island of Hawai‘i is projected to experience the greatest heat stress by the end of the century; however, we find that coral adaptation is possible to allow Hawaiian coral to persist in the three emission scenarios. Robust projections of future conditions and required adaptation rates on a local scale are critical for successful management of these crucial ecosystems.Keywords: Coral, Climate, Downscaling, Hawai‘iSubject terms: Climate sciences, Ecology, Ecology, Ocean sciences Introduction Hawaiian coral reefs have spiritual, biocultural, social and ecological relevance to the Native Hawaiian people1, and provide an important source of food2 and coastal protection3 to the islands. Since the first major bleaching event in 1996, Hawaiian corals have been subject to increasing heat stress and more prevalent bleaching events4–6. Effective future management of coral reefs requires an understanding of the projected temperatures that corals may be exposed to. Projecting future ocean temperatures is challenging in the Main Hawaiian Islands (MHI). The presence of the islands interrupts the large scale atmospheric and oceanic circulation, generating complex dynamics across a range of temporal and spatial scales. Small-scale regional dynamics, such as mesoscale eddies and local wind stress, modulate coastal temperatures around the islands4,7. Regional dynamics have been shown to impact future temperature projections8–10 but are not captured in most global climate models (GCMs). Without resolving these dynamics, GCMs cannot project ocean temperatures on sub-island scales that are essential for management. Dynamical downscaling, where a regional model is embedded within a GCM, can produce high-resolution climate projections that resolve regional dynamics. Corals may be able to adapt to, and survive, increasing temperatures by increasing their thermal tolerance over time11,12. Evidence of adaptation or acclimatization to increasing temperatures has been found in Hawaiian corals13,14. To understand the potential of corals to survive increasing temperatures, we must account for adaptation of local coral species. The objective of this study is to determine the rate that corals around the Hawaiian Islands will need to adapt to survive projected 21st century warming. We utilize output from recently-completed dynamically downscaled climate projections for the MHI15. In this approach, the Regional Ocean Modeling System (ROMS) is embedded within the Community Earth System Model (CESM2). The model is forced with three different CMIP6 Shared Socio-economic Pathways (SSP) forcing protocols: SSP1-2.6, SSP2-4.5, and SSP3-7.0 representing low, moderate and high emissions scenarios respectively16. We refer to the three downscaled simulations as SSP1, SSP2 and SSP3. The prevailing trade winds are interrupted by the Hawaiian Islands. The island-modified wind field drives important ocean dynamics around the MHI, particularly the generation of mesoscale eddies17. To capture these dynamics, the wind field is downscaled using a neural network approach detailed in18. A full description of the dynamical downscaling method can be found in15. The model configuration is submesoscale-permitting19, resolves regional processes that contribute to temperature variability15 (Fig. 1), and captures in-situ temperature and temperature variability on coral reefs in the MHI20. We average projected temperatures over the upper 30 m of the water column, corresponding to the depth range of coral reef monitoring in Hawai‘i20. Snapshot of daily temperature averaged over 0-30 m from dynamically downscaled projections for August 8th 2005. Islands are labeled in white. The North Equatorial Current (NEC) and Hawai‘i Lee Counter Current (HLCC) are labeled in black. We show that regional dynamics modulate the climate warming signal and produce spatial patterns in projected heat stress around the MHI. We find significant spatial variability in the adaptation rate required for Hawaiian corals to survive 21st century temperatures. Heat stress projections Heat stress depends on long-term temperature trends and local temperature variability. The MHI are mountains rising out of the ocean lacking typical continental shelves and coral reefs are exposed to open ocean conditions. Coastal temperature variability and trends are influenced by both offshore and regional dynamics. The westward-flowing North Equatorial Current (NEC) brings cool water to the windward (northeast) coasts. To the west of the MHI, the Hawai‘i Lee Counter Current (HLCC) delivers warm water to the leeward (west) coasts of the islands (Fig. 1)21,22. Across the MHI, temperatures typically vary by 1-2C on interannual and decadal timescales (longer than 375 days, see Fig. S2). The annual cycle dominates surface temperature variability around the MHI with a range of between 2 and 3C, with the highest range in the north23, (Fig. S2). On sub-annual timescales, temperatures around the MHI are controlled by winds, eddies, tides, and waves. Northeasterly trade winds are funneled between the islands, generating a region of high eddy kinetic energy (EKE) in the lee of the islands17,24,25. Mesoscale eddies in the region typically persist for weeks to months26, and can perturb surface and subsurface temperatures by more than 3C7 (Fig. 1). Low winds on the leeward side of Hawai‘i Island can also result in increased local temperatures because of lower evaporative cooling, clearer skies, and higher solar radiation4. These regional dynamics impact coastal temperature trends and variability. We focus our analysis on coastal locations, defined as all model grid cells adjacent to land. Average temperature trends over all coastal locations are driven by emissions scenario. For each location, we use ordinary least squares to fit a quadratic function to the temperature time series from dynamically downscaled projections between 2010 and 2100. The total projected warming over the 21st century for each location is calculated as the temperature difference due to the quadratic trend (Fig. S1). Over the 21st century, average coastal temperatures are projected to increase by 1.3 to 2.8C, depending on emissions scenario (Fig. 2a). Warming of 3C would make future wintertime temperatures equivalent to present-day summertime temperatures. Regardless of forcing scenario, coastal temperatures follow the same trajectory until mid-century, when they begin to diverge. This pattern is also apparent in CESM2 and is partly due to the fact that CO levels across emissions scenarios do not diverge until 203015. Temperatures diverge later than CO because the temperature trend around the MHI is weak27,28 relative to the background temperature variability. For each scenario, we only use one realization. Each realization has independent interannual and decadal variability and long-term fluctuations in temperature will occur at different times in different scenarios. The three scenarios diverge only once the trend is sufficiently high (around 2060) to emerge beyond the envelope of background variability. Regardless of mitigating action to curb global emissions, the future of coastal temperature in the MHI is likely already determined until approximately 2060. Temperature trends over the century. (a) Mean 0- 30 m coastal temperatures over the 21st century under different emissions scenarios. Solid lines indicate a 10-year rolling mean and the shading represents the spread in coastal-mean temperature throughout the year i.e., the annual cycle. Lower row: Spatial anomalies of projected temperature change calculated between 2010 and 2100 averaged over 0- 30 m from dynamically downscaled projections under (b) SSP1, (c) SSP2, and (d) SSP3. The spatial mean (shown in the upper right of each panel) has been removed to highlight spatial patterns. The waters around the MHI are not warming at a uniform rate. To compare the total 21st century temperature change across locations, we subtract the domain-mean total temperature change from the total temperature change at each location. The spatial anomalies of projected temperature change highlight patterns in the temperature trends that may play an important role in driving spatial patterns in coastal heat stress (Fig. 2b–d). Across all emissions scenarios, spatial patterns in the trends are consistent (Fig. 2b–d), providing confidence that these are largely robust. Ocean temperatures are projected to warm slower to the west of the islands than to the east of the islands, particularly between 19 and 20N (Fig. 2b–d). An exception to this is the area directly west of Hawai‘i Island, where temperatures are warming faster than the rest of the region. The spatial pattern in warming is partly due to the pattern imposed by CESM2 (see Fig. S3). Analyzing our domain boundaries in CESM2, we find that water advected from the west by the HLCC is projected to warm slower than water brought in by the NEC at the eastern boundary in SSP1 and SSP2. In SSP3, the stronger temperature increase leads to more uniform warming across the western boundary of the domain (163W). Despite these differences, the patterns of warming trends are consistent across the three scenarios, suggesting a contribution from domain-specific dynamics. Specifically, cyclonic mesoscale eddies generated in the lee of the islands typically propagate away from the islands north of 19N7 and upwell cooler waters17 that dampen the long-term warming signal. Of all the islands, Hawai‘i Island is projected to warm the fastest (Table 1). The coastal waters around O‘ahu are projected to experience the least warming, between 5 and 9% less warming than Hawai‘i Island, depending on scenario (Table 1). These patterns can be partly explained by the exposure of Maui, Moloka‘i, and O‘ahu to the slowly-warming waters advected in from the western boundary. These islands are also influenced by cyclonic eddies that typically propagate north of 19N. Hawai‘i is the highest island and is an effective barrier to the northeasterly trade winds. The resulting clear skies, weak winds, and strong, positive surface heat flux in the lee of the island results in more overall warming. Table 1Summary of results by island, with mean results for leeward (LW) and windward (WW) coasts shown separately. Maximum monthly mean (MMM) calculated between 2002 and 2020. Annual temperature range (difference between the 95th and 5th percentiles) calculated from temperature filtered to 355–375 days. Temperature difference due to long-term quadratic temperature trend between 2010 and 2100. Required adaptation rate such that the recovery period between significant heat stress events is always greater than 5 years. indicates metrics that are scenario-specific, SSP3 is shown. All values are rounded to one decimal place.IslandHawai‘i IslandMaui/Moloka‘iO‘ahuKaua‘iLWWWLWWWLWWWLWWWMMM [C]27.326.726.926.326.826.626.926.7Annual temperature range [C]2.72.42.72.53.02.93.23.2Trend [C per century]2.92.92.82.82.72.72.82.8DHW [C weeks]16.616.516.016.615.516.215.716.1Adaptation rate [C per century]1.91.91.81.91.81.91.81.8 Summary of results by island, with mean results for leeward (LW) and windward (WW) coasts shown separately. Maximum monthly mean (MMM) calculated between 2002 and 2020. Annual temperature range (difference between the 95th and 5th percentiles) calculated from temperature filtered to 355–375 days. Temperature difference due to long-term quadratic temperature trend between 2010 and 2100. Required adaptation rate such that the recovery period between significant heat stress events is always greater than 5 years. indicates metrics that are scenario-specific, SSP3 is shown. All values are rounded to one decimal place. To investigate how the combined effects of temperature trends and variability may impact corals, we examine temperature in the context of degree heating weeks (DHWs). The DHW metric quantifies a cumulative temperature anomaly relative to a local, climatology-derived threshold11,29–31 and is used to indicate bleaching conditions. Corals have been observed to bleach under significant heat stress (DHWs greater than 4) and experience severe bleaching and mortality under severe heat stress (DHWs greater than 8)11. The projected number of annual DHWs that corals will be exposed to over the century is highly sensitive to emissions scenario (Fig. 3). Under SSP1, coastal locations are projected to experience an increase in significant heat stress and intermittent severe heat stress that is either localized or short-lived (Fig. 3a). Under SSP2, coastal locations are projected to experience regular widespread significant heat stress over the second half of the century and extended periods (more than 5 years) of severe heat stress in most locations by the 2080s (Fig. 3b). Under SSP3, coastal locations are projected to experience a marked increase in DHWs from the 2060s onwards and extended periods of severe heat stress by the 2070s depending on location. By the 2080s, all coastal locations are projected to experience severe heat stress every year and all locations are exposed to more than 20 DHWs at least once by 2090 (Fig. 3c). The magnitude of DHWs in any given year depends on interannual variability. The phase of interannual variability is independent for each scenario and warm and cool years do not occur at the same time across scenarios. Projected Degree Heating Weeks (DHWs, in C-weeks) over the 21st Century for the Main Hawaiian Islands. Solid lines show mean DHWs across all coastal locations and the shading represents the spread over all coastal locations. The shaded yellow and red backgrounds represent levels of DHWs indicative of significant heat stress and severe heat stress, respectively. Individual panels show results from different emissions scenarios. In panel c, the dashed horizontal line indicates DHW=20. Projected heat stress is not uniform across coastal locations. In some years, all coastal locations experience a similar number of DHWs. In other years, only some coastal locations are projected to experience severe heat stress (Fig. 3, blue shading). Projected DHWs follow a predominantly latitudinal pattern. DHWs are highest around Hawai‘i Island, and also on the windward coasts of Maui, Moloka‘i, O‘ahu, and Kaua‘i (SSP3 shown in Fig. 4, Table 1). The spatial patterns are consistent across all three scenarios. For the period between 2080 and 2100 under SSP3, locations with the highest DHWs experience on average 17.6 DHW per year (Fig. 4a). Locations with the lowest DHWs experience on average 14.7 DHW per year. Locations with the highest DHWs experience on average 2.9 more DHWs per year than locations with the lowest DHWs (Fig. 4b). Figure 4c shows the contribution to the DHW pattern that is attributable to the long-term trends. The contribution from the trends explains the primary latitudinal pattern in average DHWs. More rapid warming of coastal waters in the south than in the north contributes an average difference of 3 DHWs between the locations with the fastest and slowest warming (Fig. 4c). We isolate the influence of the trend on DHWs by replacing the temperature trend from the downscaled projections with the temperature trend from CESM2 and comparing the calculated DHWs. For each location in the downscaled projections, we first remove the local quadratic long-term warming trend, then replace it with the quadratic long-term warming trend from the nearest CESM2 grid cell. Using these temperature time series, we calculate DHWs for each location. Finally, we subtract the DHWs calculated with CESM2 trends from the original DHWs calculated from downscaled simulations. Projections with CESM2 trends over-predict DHW in the center of the island chain, and under-predict DHWs in the lee of Hawai‘i Island and Kaua‘i (Fig. 4d). These results demonstrate that the regional dynamics resolved in the downscaled projections alter projections of DHWs through the modulation of long-term warming patterns. For example, in the downscaled model, the leeward coast of Hawai‘i Island is sheltered from wind-driven mixing and is projected to experience the most rapid warming and high DHWs. The island-modified wind and ocean dynamics are not represented in CESM2. Spatial patterns in projected mean annual DHWs (in C weeks) between 2080 and 2100 under SSP3. (a) Projected annual mean DHWs between 2080 and 2100. (b) Same as (a) but with the spatial mean removed to highlight spatial variability. (c) Contribution to total DHWs from the long-term trend, determined by calculating DHWs with a uniform trend and subtracting this from (a). (d) Difference between annual DHWs calculated using ROMS trend (a) and calculated using CESM2 trend. This represents the spatial information gained by using the downscaled trends, relative to using the CESM trends (see text for details). Spatial differences in temperature variability also contribute to spatial patterns in projected DHWs. The windward coasts of Maui, Moloka‘i, O‘ahu, and Kaua‘i are projected to experience higher DHWs than their respective leeward coasts (Fig. 4b, Table 1). Model results show that the windward coasts of Maui, Moloka‘i, O‘ahu, and Kaua‘i experience annual cycles that are between 2 and 9% weaker than the corresponding leeward coasts (Table 1). The weaker annual cycle may explain the spatial pattern in DHW projections. For locations with weaker annual cycles, temperatures are close to the annual maximum for a longer period of time. For a given temperature anomaly, the annual maximum will be exceeded for a higher number of days in locations with a weaker annual cycle. Coral response Corals are capable of recovering after a heat stress event if temperatures return to background conditions32. However, severe or sustained elevated temperatures can drive significant coral bleaching and mortality. Recovery timescales can range from months to multiple years32–34. We have shown that Hawaiian corals may be exposed to severe heat stress every year by the end of the century, with little time to recover. If temperatures exceed a coral’s thermal limit or if there is insufficient time to recover between heat stress events, corals will need to adapt in order to survive. We investigate how rapidly corals will need to adapt to survive projected 21st century temperatures. We explore a range of hypothetical adaptation rates between 0 and 3 °C per century, at intervals of 0.1 °C per century. Each adaptation rate is represented by a temperature time series that increases linearly between 2000 and 2100. To account for the effect of adaptation, we subtract the linear adaptation time series from the temperature time series from downscaled projections then calculate DHWs for each coastal location. The required adaptation rate is the minimum adaptation rate such that the recovery period between significant heat stress events is always greater than 5 years35,36. The timing of particularly warm years depends on natural internal variability and differs between scenarios (Fig. 3) and between different realizations of the same scenario. This timing affects the recovery period and required adaptation rate. To account for the effect of internal variability, we calculate required adaptation rates using 100 realizations of interannual and decadal variability from the CESM2 large ensemble. The median of all 100 adaptation rates provides an estimate of required adaptation rate that is not sensitive to a particular realization of internal variability. Required adaptation rates vary by location (Fig. 5). To report a single required coastal adaptation rate for each emissions scenario, we calculate the 75th percentile of required adaptation rates across all coastal locations. Under SSP1, an adaptation rate of 1.2 °C per century is required to ensure that 75% of coastal locations have a recovery period always greater than 5 years (Fig. 5a). Higher emissions scenarios require a higher adaptation rate. Under SSP2 and SSP3, the equivalent adaptation rates are 1.4 °C and 1.9 °C per century, respectively (Fig. 5b,c). Adaptation rate required so that the recovery period between significant heat stress events is always greater than 5 years under (a) SSP1, (b) SSP2, and (c) SSP3. The 75th percentile calculated across all coastal locations is shown in the bottom left of each panel. A recent study using corals collected at depths of 2 m suggests that adaptation could increase the thermal tolerance of coral species around the MHI by between 1 and 1.7 °C over the next 50 years, depending on the species14. If these rates are sustained throughout the century, corals may adapt at a rate of between 2 and 3.4 °C per century. Our results show that adaptation lower than this range (1.9 °C per century) would avoid widespread and frequent significant heat stress under all three emissions scenarios considered. Projected heat stress is not uniform and required adaptation rates vary by up to 0.4 °C per century across the MHI (Fig. 5). Spatial patterns in required adaptation rate reflect the spatial patterns in DHWs. DHWs are highest on Hawai‘i Island, and also on the windward coasts of Maui, Moloka‘i, O‘ahu, and Kaua‘i (Fig. 4, Table 1). Generally, the fastest adaptation is required for Hawai‘i Island and the windward sides of the islands (Fig. 5a–c, Table 1). Discussion Dynamically downscaled projections are rarely used for quantifying future heat stress on coral reefs due to the computational cost. A recent review37 identified 19 studies that apply downscaling to project coral reef responses to climate change. Of these, only one study8 used dynamical downscaling. Comparing statistical and dynamical downscaling methods across the Caribbean, van Hooidonk et al.8 found that the two methods provided consistent bleaching projections for most locations. However results differed significantly where GCMs did not capture climate-driven changes to regional currents. Studies using machine learning downscaling10 and nested dynamical downscaling9 show that localized upwelling impacts temperature and temperature projections on the Great Barrier Reef. GCMs cannot resolve small-scale temperature trends and variability and cannot project heat stress on sub-island scales for the MHI. Our results reinforce the finding that regional dynamics can play an important role in determining future heat stress on reefs at management-relevant spatial scales. Over the 21st century, Hawai‘i is projected to see significant temperature increases that represent a major shift away from present-day conditions. Depending on the global emissions scenario followed, coastal temperatures in the MHI are projected to increase by between 1.3 and 2.8C over the century. Regardless of scenario, coastal temperatures around the MHI are projected to increase by at least 1C by 2060. The global emissions scenario and the rate of adaptation are the two primary factors determining projected heat stress on Hawaiian coral reefs. Our results suggest that the following cases would maintain heat stress below the level at which persistent, widespread bleaching would occur: a low emissions trajectory with adaptation of 1.2C per century; a moderate emissions trajectory with adaptation of 1.4C per century; or a high emissions trajectory with adaptation of 1.9C per century. Importantly, an adaptation rate of 1.9C per century is within the range identified for Hawaiian corals by previous studies14. Experimental estimates of adaptation rate typically rely on short-term (less than one year) studies and it is uncertain whether these adaptation rates could be maintained long-term. Without aggressive emissions reductions or rapid adaptation, Hawaiian corals will be permanently exposed to severe bleaching conditions (greater than 8 DHWs) throughout the later half of the century. Both the emissions scenario and the rate of adaptation are unknown. Projected coastal temperatures follow the same trajectory regardless of scenario until 2060. Beyond this time, temperatures are highly sensitive to global emissions levels that depend on complex and uncertain socioeconomic factors. The results here present projections of the heat stress that MHI corals may face under a range of possible but uncertain futures. Adaptation rates vary among coral species and depend on both intrinsic and extrinsic factors, including the magnitude and rate of climate change38. Adaptation rate may also depend on a coral’s recent history of exposure to high temperatures39 and temperature variability38. Given these factors, adaptation rates are likely to be non-linear throughout the century. Overall heat stress depends on emissions scenario and adaptation but local variability adds spatial details that may be important for identifying potential refugia on local scales. Around the MHI, spatial patterns in DHWs are primarily driven by the modulation of long-term temperature trends by regional dynamics. Background temperature variability also contributes to spatial patterns in DHWs. Towards the end of the century under SSP3, spatial differences in DHWs are likely negligible since all locations are projected to be permanently exposed to high (greater than 20) DHWs. Although spatial differences in DHWs are small in magnitude, they may play a significant role under SSP1 and SSP2 and earlier in the century under SSP3 when DHWs are close to levels expected to result in bleaching. During these times, our results suggest that the leeward coasts of Maui, Moloka‘i, O‘ahu, and Kaua‘i are less likely to experience bleaching conditions than their respective windward coasts. Of all the islands, Hawai‘i Island is projected to experience the highest number of DHWs. In addition to increasing heat stress, Hawaiian reefs face many other stressors40,41. In the face of multiple, compounding stressors, small spatial differences in warming trends may become crucial. Due to the computational cost of dynamic downscaling, we present results from a single GCM, CESM2. Different GCMs will produce different warming trajectories for a given emissions scenario. Relative to other climate models, CESM2 has high equilibrium climate sensitivity42,43. As a result, the 21st century warming presented in our results for the Main Hawaiian Islands is likely to be high, relative to other CMIP6 models. Regardless of climate sensitivity, we are able to investigate and present a broad range of warming conditions because we run downscaled simulations under multiple warming scenarios. In the calculation of DHW, a number of choices have to be made that may impact the results. Firstly, we use years 2002- 2020 to represent background temperature conditions and to calculate the MMM. In operational use, the period between 1985- 2012 is typically used to represent background conditions30. Different definitions of background conditions may result in different calculated DHW values and impact whether heat stress is determined to be significant or severe. Secondly, we choose to define the MMM as the maximum of a 30-day running mean of the daily climatology. This may provide different results to the maximum of calendar monthly means typically used30. We use a 30-day running mean to capture typical maximum temperatures on monthly timescales, regardless of arbitrary calendar dates. Previous studies have demonstrated that temperature variability can alter heat stress projections and required adaptation rates across geographic regions with distinct climatic characteristics (seasonal and interannual variability)29. Our results show that temperature variability also plays an important role at much smaller, sub-island scales that are important for management. If the adaptation rates of Jury and Toonen14 are achieved, Hawaiian corals have the possibility of adapting to 21st century temperatures. Regional dynamics impact coastal heat stress in the MHI, and the local adaptation rate required depends on regional dynamics. Our results show that in locations where regional dynamics impact coastal temperatures, there can be significant spatial variability in the adaptation rate required for corals to survive 21st century temperatures. Methods Dynamical downscaling Dynamically downscaled climate projections are generated by embedding a high resolution regional model within a lower resolution GCM. This method has the advantage of resolving small scale dynamics within the region of interest without the vast computational expense of running a global model at the equivalent resolution. To investigate ocean temperatures around Hawai‘i over the 21st century, we examine output from dynamically downscaled climate projections documented in Friedrich et al.15. In this approach, the Regional Ocean Modeling System (ROMS) with a domain spanning the MHI is embedded within the Community Earth System Model (CESM2). CESM2 is a fully-coupled global climate model with a horizontal resolution of one degree in the ocean44. The model is forced with three different CMIP6 Shared Socio-economic Pathways (SSP) forcing protocols: SSP1-2.6, SSP2-4.5, and SSP3-7.0 representing low, moderate and high emissions scenarios respectively16. The last two digits of each scenario indicate the global average radiative forcing level (W m) achieved by the end of the century. Output from CESM2 simulations forced with SSP1-2.6 and SSP2-4.5 were accessed directly from the CMIP6 archive. For SSP3-7.0, output from a single ensemble member of the CESM2 large ensemble (LE) was selected. The CESM2-LE explores forced changes in climate variability under SSP3-7.0 by initializing 100 ensemble members with slightly different initial states45. For this study, the ensemble member with the SST and wind speed variance that deviated least from historical observations in the MHI was selected (see15 for details). Prior to 2015, CESM2 simulations forced with SSP1-2.6 and SSP2-4.5 have identical forcing. The CESM2 simulation forced with SSP3-7.0 was initialized from different conditions in 1850 and has different forcing for the duration of the simulation. Output from these three CESM2 simulations provided the initial, boundary, and forcing conditions for the downscaled simulations using ROMS between 2000 and 2100. We refer to these three downscaled simulations as SSP1, SSP2, and SSP3. ROMS is a hydrostatic free-surface model that solves the primitive equations over terrain-following vertical levels46. The grid has a horizontal resolution of 4 km and 32 vertical layers. The downscaled projections are submesoscale-permitting19 and have been shown to retain the long-term warming and the large-scale climate modes such as ENSO that are present in the global model15. The projections also resolve regional processes, such as mesoscale eddies, that contribute to temperature variability15, (Fig. 1c). A ten-year reanalysis of the MHI based on the same model configuration has been found to capture in-situ temperature and temperature variability on coral reefs in the upper 30 m across the MHI20. Depth averaging We base our calculations of temperature trends, DHWs, and adaptation rates on projected temperatures averaged over the upper 30 m of the water column. This range was selected to cover the depth range of most coral reef monitoring in Hawai‘i20. With respect to temperature, the upper 30 m of the water column around the Hawaiian Islands is well-mixed on daily timescales47,48. Our primary interest is investigating temperature at the locations and depths of monitored coral reefs. Since corals are fixed in place, we limit the analysis to a fixed depth range. Temperature trends Long-term temperature trends were calculated for all locations in the domain. For each grid cell, the time series of 0-30 m depth-averaged temperature was fit to a quadratic function using ordinary least squares, and the total projected warming was calculated as the total temperature change due to the trend (see Fig. S1). To test whether trends are sensitive to the specific period over which they are calculated, we compared trends for SSP3 calculated with start years between 2005 and 2015, all ending in 2100. Different start years resulted in a maximum difference in domain-wide temperature change of 0.05C by 2100, showing that the trend is robust and insensitive to start date. Spatial patterns were the same regardless of time period used. The trends presented here were calculated using a start date of 2010. Temperature variability To quantify how regional dynamics contribute to temperature variability, we decompose 0-30 m depth-averaged temperature time series into contributions at interannual and annual timescales, defined as periods of >375 and 355-375 days, respectively. We decompose de-trended temperature time series by applying a boxcar filter in frequency space. For each timescale, we quantify the temperature range (difference between the 95th and 5th percentiles) for each location. We use the temperature range because the distribution of temperatures is not Gaussian for all timescales. Results are shown in Fig. 2 of the supplementary information. Degree heating weeks We examine total heat stress in the context of annual degree heating weeks (DHWs), a widely used metric that quantifies a cumulative temperature anomaly relative to a local, climatology-derived threshold (e.g.,11,29–31). For each coastal location, we begin by calculating a daily climatology. Specifically, we calculate a daily climatology using years 2002-2020 from all three simulations. Years 2000-2002 are removed from each simulation to account for model spinup. The three simulations are subject to different forcing between 2015-2020; however, the differences are sufficiently small and the impact on ocean temperatures is sufficiently delayed that it is more beneficial to include these years than to use a shorter climatology. SSP1 and SSP2 were initialized from the same state and have identical temperatures until 2015 when the different emissions scenarios are applied. SSP3 was initialized from a slightly different state starting in 1850 (see15 for details). Therefore, for the years spanning 2002-2015 we have only two independent realizations of background conditions (SSP1/SSP2 and SSP3). For the years 2015-2020, each simulation is subject to different forcing and we have three independent realizations of background conditions (SSP1, SSP2, and SSP3). Each independent realization provides an equally valid representation of background conditions. To get the best possible representation of local climatology, we combine only independent realizations of background conditions. Specifically, we combine temperatures from the two independent realizations (SSP2 and SSP3) for years 2002-2020. We combine temperatures from all three independent realizations (SSP1, SSP2 and SSP3) for years 2015–2020. We then calculate a running 30-day mean over the daily climatology. We use a 30-day running mean rather than calendar monthly means to eliminate the influence of arbitrary month start and end dates. The maximum monthly mean (MMM) for each grid cell is the maximum of the running 30-day means. For all coastal locations, the MMM occurs during calendar month September. The MMM ranges from 26 to 27.5C with the coolest MMM on the northeast coasts of the islands that are exposed to the relatively cool waters of the NEC. The leeward coasts of the islands have warmer MMM, particularly the leeward coast of Hawai‘i Island which is most sheltered from the trade winds and has the highest MMM. These patterns are consistent with MMM calculated from satellite-derived SST49 over the same time period that range from 26.6 to 27.4C and are highest on the leeward sides of all islands. For each coastal grid cell, a time series of daily averaged temperatures from 2020-2100 is used to calculate anomalies, defined as temperatures that exceed the MMM plus 1C11,30. Daily anomalies are summed using a 12-week running total. The highest 12-week sum is retained for each year and divided by seven to convert from degree heating days to DHWs. Although the DHW metric uses only a single stressor (temperature), it has been found to be a useful predictor of bleaching50. Coral bleaching has been observed to occur when DHWs exceed 4 and DHWs exceeding 8 have been found to be indicative of severe bleaching and mortality11. We use the terms ”significant heat stress” and ”severe heat stress” to refer to instances of DHW and DHW, respectively. Adaptation To explore the possible impact of coral adaptation, we remove linear trends from the 2000-2100 temperature time series equivalent to adaptation rates between 0 and 3C per century at intervals of 0.1C per century. To achieve this, we first generate multiple time series, each representing a linear temperature increase between 2000 and 2100. We generate individual time series for temperature increases between 0 and 3C per century at intervals of 0.1C. This results in a total of 30 linear time series. Each time series represents a fixed, linear coral adaptation rate. We then subtract these linear adaptation trends from the temperature time series from downscaled projections. This method resolves the time-evolving effects of adaptation (as opposed to simply increasing the threshold). Multiple studies suggest that coral adaptation depends on a range of factors, for example coral species14, and historical exposure to high temperatures39. There is also evidence that adaptation is likely to occur non-linearly51. Since adaptation rates are variable and depend on many factors, we apply average, linear adaptation rates. This method allows for the direct comparison of multiple adaptation rates. Regardless of adaptation rate, corals are capable of recovering after a heat stress event. Recovery timescales can range from months to multiple years32–34. We impose a minimum 5-year recovery period for corals after being exposed to DHW greater than 435,36. For each coastal location, we impose linear adaptation rates and identify years when DHW exceed 4. We then test whether DHWs remain below 4 for the next 5 years (i.e., recovery period always greater than 5 years). The required adaptation rate is the lowest adaptation rate that meets this criteria. Because of the imposed recovery period, our required adaptation rates are sensitive to the phasing of interannual and decadal temperature variability. To remove this effect, we generate a psuedo-ensemble for each scenario by replacing the interannual and decadal variability from the dynamically downscaled simulations at each location with 100 realizations of interannual and decadal variability from the CESM2 large ensemble52. For each scenario, we do this individually for the temperature time series at each location via the following steps: We first fit and remove a quadratic trend from the dynamically downscaled output. We then remove the interannual and decadal variability by high-pass filtering at 375 days. We then add the trend back in. This gives us a downscaled time series with no interannual or decadal variability. Next we isolate the interannual and decadal variability from each ensemble member of the CESM2 large ensemble by removing a quadratic trend then low-pass filtering at 375 days. This gives us 100 realizations of interannual and decadal variability. We add each of these 100 time series to the downscaled time series with no interannual and decadal variability. This gives us 100 temperature time series, each with the same trend, annual, and sub-annual variability but different realizations of interannual and decadal variability. We calculate the required adaptation rate from all 100 time series, for each location and each scenario. Lastly, we take the median over the 100 calculated adaptation rates for each location and each scenario. Supplementary Information Supplementary Information. Acknowledgements This is SOEST publication number 12097 and UCAO contribution number 38. Dr. Feloy was supported by a graduate fellowship and research funds from the Uehiro Center for the Advancement of Oceanography. Drs. Powell and Friedrich were supported by NOAA grant NA20OAR4310445. Dr. Barkley was supported by NOAA’s Coral Reef Conservation Program, National Coral Reef Monitoring Program-Pacific Project # 743 and NOAA’s Ocean Acidification Program. The authors thank the World Climate Research Programme for the coordinated CMIP6. We thank international modeling groups for making CMIP6 outputs available via the Earth System Grid Federation (ESGF). We also thank the CESM2 Large Ensemble Community for the SSP3.70 case used in this work. We acknowledge the IBS Center for Climate Physics (ICCP) that made the CESM2 Large Ensemble possible. Finally, we thank the two anonymous reviewers along with the editor, Alankar Sapkal, for their comments that improved this manuscript. Author contributions K.F., B.P., and T.F. conceived of and designed the study. K.F. analyzed the data and wrote the manuscript. K.F., B.P., T.F., and H.B. revised and finalized the manuscript. Data availability All data presented in this manuscript is available via the Open Science Framework at https://osf.io/gwkev/overview. Declarations Competing Interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supplementary Information The online version contains supplementary material available at 10.1038/s41598-026-47781-3. References Associated Data Supplementary Materials Supplementary Information. Data Availability Statement All data presented in this manuscript is available via the Open Science Framework at https://osf.io/gwkev/overview.