An Integrated Oceanographic and Physiological Framework for Identifying Climate Refugia for Marine Invertebrates

ABSTRACT

Climate refugia—areas buffered from environmental change—are increasingly important for marine conservation and fisheries management, yet identifying their location and extent remains challenging. We developed a framework using established ecological and physiological concepts describing temperature‐induced stress to map potential thermal refugia—areas expected to be resilient to thermal stress. Our framework uses three complementary approaches to analyze daily temperature records and quantify relative thermal stress across sites. We applied this to economically important green abalone ( Haliotis fulgens ) along 1200 km of Baja California (Mexico) coastline, in the southern California large marine ecoregion. Using satellite sea surface temperature data from 1186 sites spaced at 1 km intervals along the coastline, converted to bottom water temperature via in situ validation, we calculated relative thermal stress based on: (1) short‐term temperature variability, because stress from rapid changes can exceed acclimatization capacity; (2) exposure to thermal extremes, because prolonged heat exposure induces thermal stress; and (3) long‐term temperature deviations, because extreme cold and hot conditions may not be optimal for organism performance. Each approach revealed distinct spatial patterns, with relative stress varying substantially from individual bays (< 10 km) to regional gradients (> 500 km). Despite differences, 18% of the coastline (213 of 1186 sites) exhibited consistently low relative stress across all three definitions, concentrated north of 28° N. An additional 28% met low‐stress criteria for two definitions, spanning 28° N–32.5° N. To evaluate the robustness, we used 18 years of catch‐per‐unit‐effort fisheries data from Isla Natividad. Exposure to thermal extremes (stress definition 2) best explained abalone catch patterns, accounting for 61% of variance. Results demonstrate that refugia identification depends critically on how thermal stress is defined, yet consensus areas across multiple definitions provide a generalizable framework for identifying robust targets for climate‐resilient marine management in an era of accelerating ocean change.

We mapped thermal refugia for green abalone ( Haliotis fulgens ) across 1186 sites along 1200 km of Baja California coastline using three complementary definitions of thermal stress: short‐term temperature variability, exposure to thermal extremes, and long‐term temperature deviations from site‐specific means. Despite distinct spatial patterns across definitions, 18% of sites exhibited consistently low thermal stress across all three metrics, concentrated north of 28° N latitude. An additional 28% met low‐stress criteria for two definitions, spanning 28° N–32.5° N. These consensus refugia—robust to how thermal stress is defined—provide spatially explicit targets for climate‐resilient marine protected area design in an era of accelerating ocean warming.

1. Introduction

Ocean warming is impacting the biology of marine organisms (Frölicher et al. 2018; Poloczanska et al. 2013; Schulte 2014) and causing population ranges to shrink or shift, and, in some cases, expand in response to changes in water temperature (Pinsky et al. 2013, 2020). In addition to temperature change, other phenomena, including ocean acidification (Cooley and Doney 2009), hypoxia (Zillén et al. 2008), and sea level rise (Bellard et al. 2014), have already had negative impacts on marine populations, ecosystems, and ecosystem services globally, including fisheries (Brander 2007). Historical ocean warming has led to an estimated 4% decline of global harvests from 1930 to 2010, with multiple ecoregions experiencing losses up to 35% (Free et al. 2019). Given the escalating pace of ecological and economic impact and the urgent need to support climate mitigation and adaptation, it is imperative that scientists attempt to explain anomalies in the expected patterns of marine population responses to climate change and leverage this understanding to inform climate adaptation strategies.

One approach to understanding these anomalies involves identifying climate refugia—areas that are buffered from contemporary climate change and extremes (Ashcroft 2010; Isaak et al. 2015; Lenoir et al. 2017; Maher et al. 2017; Sanz‐Martín et al. 2026). Climate refugia arise because the impact of climate change and extremes, such as marine heat waves (Hobday et al. 2016), on marine environments is likely not uniform across space or time (Morelli et al. 2016). Consequently, climate refugia are becoming especially relevant to marine spatial management aimed at prioritizing climate resilience (Arafeh‐Dalmau et al. 2023; Micheli et al. 2024; Sanz‐Martín et al. 2026). Refuge habitats have the potential to promote persistence in small populations and therefore, may be conservation buffers to long‐term change (Keppel et al. 2015). Although managing climate refugia is frequently identified as a key climate change adaptation strategy in marine spatial planning, identifying the location and spatial extent, and quantifying the attributes of such habitats is rare, especially in marine ecosystems (e.g., Kroeker et al. 2023; Woodson et al. 2019) compared to terrestrial ecosystems (e.g., Keppel et al. 2015). Nevertheless, a growing body of work has begun to evaluate marine thermal refugia explicitly, particularly in eastern boundary upwelling systems and coral reef contexts. Upwelling regions have been proposed and modeled as potential climate refugia under warming scenarios (Dixon et al. 2022; Frazer et al. 2026; Lourenço et al. 2016), including assessments across geological timescales (Rodriguez‐Ruano et al. 2023). Regional assessments in Mexican waters have also evaluated upwelling systems—including western Baja California, the Gulf of California, and northern Yucatan—as potential thermal refugia using physiological and ecological niche‐based approaches (Angeles‐Gonzalez et al. 2023, 2024). Our framework builds on and extends this emerging literature by developing an integrated, multi‐criteria approach to refugia identification grounded in established physiological and oceanographic concepts.

Marine scientists and practitioners have made substantial efforts to predict where ocean habitats have and will become increasingly stressful for organisms to survive and reproduce, and eventually become uninhabitable (Duncan et al. 2023; Galli et al. 2017; Oliver et al. 2019). Low‐quality or unsuitable habitat is often defined based on observations from field and experimental studies and may involve multiple variables (Kroeker et al. 2023) and using statistical and modeling approaches, such as species distribution models (Angeles‐Gonzalez et al. 2024; Cheung et al. 2011; Frazer et al. 2026). Studies of organismal physiology, ecology, and oceanography in marine science suggest there may be multiple criteria for defining and identifying poor habitat even when using a single environmental variable, such as temperature. This follows directly from the physiology of thermal performance: organismal performance is nonlinear across temperature gradients, with distinct optimum, suboptimal, and critical threshold temperatures (Frederich and Pörtner 2000; Pörtner and Farrell 2008; Pörtner and Peck 2010). As a result, different thermal stress metrics, such as short‐term variability, cumulative heat exposure, and deviation from optimum, each capture different aspects of how temperature impairs performance, and no single criterion fully characterizes the thermal environment experienced by an organism. Such observations highlight a critical need to integrate across criteria, data, and approaches to understand the spatial distribution of climate refugia.

The need to integrate assessments of thermal stress from multiple disciplines to identify climate refugia in marine ecosystems is especially pertinent for economically and culturally important marine species, particularly invertebrates with limited ability to move and escape extreme conditions. Green abalone Haliotis fulgens , in Baja California, Mexico, exemplifies this challenge. The abalone fishery (Haliotis spp) from Baja California has been the most important economic activity in shaping the development of coastal villages in western Baja California (McCay et al. 2014; Ponce‐Díaz et al. 1998). Over the last 30 years, abalone fisheries have been managed by a relatively successful model of multispecies spatial rights (territorial use rights for fisheries, or TURFs) granted to fishing cooperatives under a comanagement system with the Mexican Government (McCay et al. 2014). Recent management approaches have expanded to include fully‐ or partially protected areas (MPAs). Protection through MPAs, minimum landing sizes, seasonal closures in fishing zones, and setting limits on annual catch quotas has promoted the recovery of abalone populations and fisheries following mass mortalities and fisheries collapse at some locations (Micheli et al. 2012; Olguín‐Jacobson et al. 2025; Smith et al. 2022). However, with escalating regional impacts from extreme ENSO, hypoxic events, and marine heat waves, there is great concern about the future of these important fisheries, and a critical need to inform the siting and design of MPAs and other spatial management approaches to promoting climate resilience for abalone and other vulnerable marine species (Micheli et al. 2024). Increased understanding of abalone vulnerability to environmental change, in particular ocean temperature, and identification of potential refugia is urgently needed. The connection between relative thermal stress and demographic outcomes operates through well‐established physiological mechanisms: temperatures outside an organism's thermal performance window reduce growth, reproduction, and survival rates (Pörtner and Peck 2010). Areas with consistently low thermal stress are therefore expected to support higher fitness and population persistence over time, providing the mechanistic basis for using stress‐based metrics as proxies for habitat quality in the context of MPA siting and climate‐resilient spatial management.

Abalone sensitivity to temperature has been documented using a suite of temperature‐related stress criteria (Table 1). Laboratory and field studies have investigated a variety of thermal stress metrics in response to prolonged exposure to warm temperatures, rapid changes over short periods (Chen et al. 2016; Hooper et al. 2014), and rare but extreme temperature events (Chen et al. 2016; Dahlhoff and Somero 1993a, 1993b; Hooper et al. 2014; Rogers‐Bennett and Catton 2019). These experiments and field observations have shown that when temperature increases, abalone growth can also increase, up to a threshold, but survival rates decline, especially when temperatures increase in variability (Boch et al. 2018). Heatwaves and associated disease outbreaks are generally considered one of the major causes of abalone mass mortality despite positive effects of incremental warm temperatures on abalone growth rates (Leighton 1974; Moore et al. 2000). Extended and prolonged marine heat waves in the California Current have caused large die‐offs in kelp and led to mass mortality events in abalone populations as well as increases in the presence of withering syndrome, which leads to reduced meat quality for commercial sale (Beas‐Luna et al. 2020; Micheli et al. 2024; Rogers‐Bennett and Catton 2019). Efficiently synthesizing the impacts of these different types of environmental change on abalone is a necessary next step to generalize the lessons learned from lab studies and natural observations into actionable information for abalone harvesters and managers.

TABLE 1.

Examples of studies measuring temperature‐induced stress in abalone from ecology, oceanography, physiology disciplines.

Thermal stress definition	Outcome measured	Criteria for temperature stress and time period	Species	Marine science discipline	Citation
Stress 1: Short‐term variability	Organ damage and immunosuppression	10°C ramp 3 days	Hybrid abalone H. laevigata  ×  H. rubra	Aquaculture, Physiology	Hooper et al. (2014)
	Cardiac performance	10+°C ramp 24 h	H. discus hannai , H. gigantea	Aquaculture, Physiology	Chen et al. (2016)
	Respiration rates	5°C ramp, ranging 5°C–55°C 2 months	H. cracherodii H. corregata H. rufesens H. kamtschatkana kamtschatkana H. fulgens	Physiology	Dahlhoff and Somero (1993a)
	Cytosolic malate dehydrogenases (an enzyme) in muscle tissue	< 24 h	H. cracherodii H. corregata H. rufesens H. kamtschatkana kamtschatkana H. fulgens	Physiology	Dahlhoff and Somero (1993b)
Stress 2: Exposure to thermal extremes	Immunity	Prolonged exposure at constant temperature 1 month	H. discus hannai	Physiology	Ding et al. (2016)
	Mortality	Presence of marine heat wave 4 years	H. rufesens	Ecology	Rogers‐Bennett and Catton (2019)
	Growth rate, mortality	Cumulative exposure to > 20°C 1 year	Green abalone H. fulgens	Ecology, Oceanography	Boch et al. (2018)

Note: Citations are classified into two thermal stress definitions: Stress 1 short‐term variability and stress 2 exposure to thermal extremes. To date, no studies have used long‐term temperature deviations (stress 3 definition) to study thermal stress in abalone.

Here, we present a general framework for identifying refugia from relative thermal stress—hereafter “thermal refugia”—that integrates different aspects of how organisms experience temperature‐related stress. Our approach borrows well‐established concepts from physiology, and marine ecology, to determine “consensus” areas emerging as refugia regardless of criteria. We then apply this new framework to a widely distributed benthic marine invertebrate to illustrate and evaluate this novel approach to refugia identification. Our framework encompasses three distinct definitions of thermal stress (Figure 1): short‐term variability (definition 1), which quantifies relative stress from rapid temperature changes that exceed an organism's acclimatization capacity; exposure above a thermal threshold (definition 2), which measures relative stress as cumulative exposure to temperatures above site‐specific thermal thresholds; and (3) long‐term temperature deviations (definition 3), which assumes relative stress is highest when daily temperature deviates from a site‐specific annual mean temperature and uses the shape of a thermal performance curve to calculate relative stress daily, recognizing that both unusually warm and cold conditions may be suboptimal for organism performance. We acknowledge that this definition implicitly equates the long‐term local mean temperature with an organism's thermal optimum, an assumption that requires justification. Thermal performance curves are not always symmetric around environmental means, and optimal temperatures may differ from experienced means (Angilletta 2009; Pörtner and Farrell 2008). In the absence of species‐specific thermal performance data for a broadly distributed species with short dispersal distance resolved at the spatial scales of our analysis, we use the local annual mean as a spatially explicit baseline, on the assumption that it approximates the thermal conditions to which local populations are most regularly exposed. Local adaptation is expected under spatially heterogeneous selection (Kawecki and Ebert 2004), and observed in coastal marine populations (Kelly et al. 2013). For all three definitions, experimental and field observations and theoretical frameworks show support for the mechanisms of these different types of relative thermal stress impacting marine organisms and abalone (definition 1: Dowd and Denny 2020; Kroeker et al. 2020; Peck et al. 2014; Pörtner and Gutt 2016; definition 2: Kayanne 2017; King et al. 2025; Morley et al. 2019 and citations in Table 1; definition 3: Brown et al. 2004).

FIGURE 1.

Three definitions of thermal refugia based on different measures of thermal stress. (a) Short‐term variability: Daily thermal stress calculated as the difference between ambient temperature and acclimatized temperature (expected temperature from modified LOWESS regression with decreasing weights for earlier temperatures, shown by circle sizes). (b) Degree‐heating days: Thermal stress measured as cumulative temperatures above a threshold (purple shaded area below the curve). (c) Long‐term temperature deviations: Daily thermal stress is the deviation of daily temperature from the annual mean multiplied by the stress factor of that day. The stress factor is the inverted thermal performance curve, and site‐specific annual mean temperature is assumed to be associated with the lowest stress (T _opt). Purple arrows indicate an example of daily temperature deviations.

The three stress definitions are conceptually distinct but not fully independent—they capture partially overlapping dimensions of the thermal environment. Definition 1 (short‐term variability) reflects the rate and magnitude of temperature fluctuations over short timescales and is most sensitive to rapid, acute thermal events. Definition 2 (exposure above threshold) is driven primarily by the frequency and duration of warm temperature anomalies and is most sensitive to persistent heat stress. Definition 3 (deviation from annual mean) is influenced by both warm and cold extremes relative to the local baseline and integrates performance‐curve‐based penalties across the full temperature distribution. In practice, sites that are simultaneously affected by multiple types of thermal stress—for example, those subject to both high variability and frequent thermal extremes—are likely to score poorly across all three definitions. Conversely, sites that are thermally stressful under one definition but not others may represent partial refugia. Identifying areas of low stress across all three definitions thus provides a conservative, robust criterion for refugia designation that accounts for the possibility of partial overlap among metrics while retaining the distinct mechanistic information each captures.

The goal of this paper is two‐fold: first, to present a new general framework for identifying marine climate refugia for marine invertebrates that combines well‐established concepts from marine science disciplines and applies it to green abalone and rocky reef habitat as a case study. Second, to create fine‐resolution maps and identify areas of “consensus” among the three refugia definitions as a way to identify potential refugia that are robust across different criteria. The resulting maps provide useful information about where thermal refugia might occur for a low‐dispersal, low‐mobility benthic species. Importantly, we document our process as a framework for researchers working at the nexus of scientific research and marine resource management on how to combine multiple data sources and stress estimation methods to develop a relatively straightforward tool that defines and identifies potential locations of climate refugia. We note that the individual stress definitions employed here draw on well‐established concepts in physiology and marine ecology. The novelty of our contribution lies not in the individual metrics themselves, but in three aspects of their synthesis: (1) the explicit integration of three mechanistically distinct definitions of thermal stress into a unified spatial framework; (2) the application of this integrated framework at fine spatial resolution (1 km intervals) along a 1200 km coastline, enabling detection of refugia from the scale of individual bays to regional gradients; and (3) the empirical evaluation of the framework against independent fisheries data, providing initial evidence for the ecological relevance of the stress metrics. Together, these elements advance the operationalization of climate refugia identification for marine invertebrate management in ways that single‐metric or nonspatial approaches cannot.

2. Methods

We analyze three thermal‐induced stress definitions and apply them to green abalone. This study is focused on the Pacific coast of Baja California, Mexico. Abalone harvest is regulated by fishing cooperatives that hold abalone concessions (exclusive access rights; McCay et al. 2014) that all belong to the Federacion Regional de Sociedades Cooperativas de la Industria Pesquera Baja California. We use the shorthand FEDECOOP to refer to these management areas.

Below we describe how a time series of sea surface temperature (SST), available from satellite data for the entire region, is converted to bottom temperature in near‐shore environments (Section 2.1). Second, we define the three thermal stress definitions and how each is estimated using available temperature data (Section 2.2). Third, we show how relative thermal stress from the definitions is used to identify low stress sites for abalone along the Baja California coastline (Section 2.3). And last, we provide a case study of Isla Natividad, in Baja California Sur, how the three stress definitions may be validated with existing catch data (Section 2.4: Isla Natividad Case Study).

2.1. Method for Converting SST to Bottom Temperature

Abalone are benthic marine species and therefore seawater temperature near the seafloor is best for characterizing their habitat. However, continuous records of bottom temperature are scarce because they require in situ measurements by instruments set and retrieved by divers, which is challenging to do across large geographic regions. Long‐term records of SST, however, are available at regional scales from satellite remote sensing (e.g., NOAA Coastwatch Environmental Research Division's Data Access Program, ERDDAP). We developed a method for converting satellite‐derived SST time series into time series of bottom temperature using in situ records of ocean bottom temperature time series recorded at 17 sites dispersed throughout the FEDECOOP concessions (Figure 2). At all sites, miniDOT temperature loggers were deployed for 3–15 years and the corresponding satellite‐derived SST records were retrieved over the same period. For plotting, we show approximately 1 year (10/2017–10/2018) (Figure 2).

FIGURE 2.

Map showing the 17 sites with miniDOT temperature loggers in Baja California Sur; 12 strong upwelling sites (green) and 5 weak upwelling sites (blue). Cooperative management boundaries are shaded gray polygons.

Time series of daily SST for the west coast of North America (26 N to 38 N) were obtained from the MODIS Aqua SST data product available on the NOAA Coastwatch ERDDAP server (https://coastwatch.pfeg.noaa.gov/erddap/index.html) over the period from 2003 to 2019. SST data were interpolated to 1‐km bins along the entire coastline to create time series for each block. Seasonal climatologies were created using a smoothing function with a 90‐day local weighted smoothing function (loess). Time series of satellite‐derived SST can be patchy due to cloud cover. To fill in missing daily values of SST we computed the variance in daily SST values for the entire time series at each block, then used a normal random number generator with a mean of 0 and variance equal to the measured value from the available SST data and added this value to the climatology for each day with missing data. Once full records of SST were available, we compute power spectra over the entire time series at each site and integrated it over frequencies above 0.25 day⁻¹ (4 day period) to obtain an estimate of high frequency variance that would be unaffected by seasonality and synoptic weather patterns which are largely driven by high and low pressure systems that cause upwelling and relaxation events at approximately 10 day cycles (e.g., Woodson et al. 2019). See Figures S2 and S3 for complete power spectra plots at each site in SST and miniDOT time series data, respectively.

SST and bottom temperature exhibit similar seasonal patterns, but mean values and high‐frequency variability (temperature fluctuations at timescales shorter than 4 days) differ significantly. SST was consistently warmer and more variable than bottom temperature (Figures S1.1 and S1.2), with mean bottom temperature averaging 2.5°C cooler. We found that high‐frequency variance in SST correlates with high‐frequency variance in bottom temperature (Figure 3a), enabling us to convert between these measures. To establish this conversion, we fit both linear and power law functions to the satellite‐derived SST and bottom temperature data (Figure 3a, Figure S4). Because the relationship between surface and bottom temperature in nearshore environments is strongly influenced by coastal upwelling along western North America, we classified each site as either “strong” or “weak” upwelling based on its coastline orientation (Table S1). Weak upwelling sites are located in protected coves or bays with coastlines to the south or west, while strong upwelling sites have coastlines to the east or north. We evaluated five functions relating high‐frequency variability between SST and bottom temperature: (1) a linear regression fit to all sites, separate linear regressions for strong and weak upwelling sites, power law regressions for all sites and for strong upwelling sites specifically. Based on R ² values, a linear regression provided the best fit to weak upwelling sites and the simple power law regression performed best for strong upwelling sites (Figure 3a, Figure S4).

FIGURE 3.

Temperature relationship between the surface (SST) and the bottom at 17 sites. (a) Correlation between high frequency variance in SST time series and high frequency variance in bottom temperature time series. Linear regression (blue solid) is fit to weak upwelling sites and power law regression (green dashed) is fit to strong upwelling sites. (b) Mean differences in SST and bottom temperature across 17 sites. (c) Time series of SST and bottom temperature at one example site, Los Gavilanes, to show that seasonally in temperature follows the same pattern at the surface and at bottom.

To convert SST time series to bottom temperature time series, we implemented a four‐step process: (1) extracting the seasonal climatology from SST data, (2) adjusting this climatology by subtracting 2.5°C to better align with typical bottom temperature, (3) applying site‐specific conversions for high‐frequency variance—using linear regression for weak upwelling sites and power law regression for strong upwelling sites (Figure 3a), and (4) combining the converted high‐frequency variability with the adjusted seasonal climatology to produce the final time series. Because our primary goal was to use these temperature time series to calculate relative thermal stress, we validated our SST conversion by comparing relative thermal stress values derived from the adjusted SST against those calculated using direct bottom temperature measurements from miniDOT sensors.

2.2. Thermal Stress Definitions

Using the derived bottom temperature time series from SST, we calculate relative thermal stress at 1 km intervals along the Baja California coastline. We processed 10‐year satellite‐derived SST datasets through our conversion method to generate corresponding bottom temperature profiles for each of the 1186 locations. Using the three thermal stress mathematical definitions, we calculate daily thermal stress (detailed below). Because we are seeking locations that potentially serve as long‐term refugia habitat, we numerically integrate the 10‐year daily stress time series, producing a single cumulative value for each site under each definition. For comparative analysis, we normalize the integrated values to a 0–1 scale, enabling clear assessment of relative stress levels across all 1186 sites. We use cumulative integration rather than annual or event‐based summaries because our goal is to identify sites that consistently experience low thermal stress over ecologically relevant timescales, e.g., the decade‐scale persistence conditions that would support stable abalone populations and justify long‐term management investment. A site experiencing rare but intense stress events will accumulate high cumulative stress over 10 years just as a chronically stressed site would, which is appropriate given that both scenarios are unfavorable for refugia designation. We acknowledge that this approach collapses temporal structure; future work using event‐based metrics (e.g., frequency and duration of extreme events) could provide complementary information. The normalization of integrated values to a 0–1 scale is intended to facilitate visual comparison across definitions with different magnitudes, and all interpretations are based on relative rankings within the study region rather than absolute stress values.

2.2.1. Stress Definition 1 (Short‐Term Variability)

This definition assumes that rapid temperature change results in greater stress compared to stable thermal conditions that allow for physiological acclimatization. We implemented methods developed by Dowd and Denny (2020) to quantify relative thermal stress in green abalone. Laboratory experiments show that rapid temperature increases (+5°C) caused decreased feeding activity and movement in abalone by > 90% over 24–48 h time periods (Boch et al. 2018), and yet the negative effects of warmer water dissipate over longer periods of time (~1–2 weeks, authors' personal observation in the field and supported by studies in Table 1), suggesting that abalone can acclimatize to warmer conditions with time. We use an acclimatization window, which assumes that abalone physiology is adjusted to temperatures experienced over the previous 4 days. This timescale is consistent with laboratory observations of abalone behavioral responses to temperature changes, where effects of rapid warming on feeding and movement persisted for 24–48 h but dissipated over 1–2 week periods, suggesting intermediate‐term physiology adjustment occurs rapidly, within days (Boch et al. 2018). We acknowledge that behavioral recovery timescales do not necessarily correspond to full physiological acclimatization, which can involve metabolic adjustments, such as heat shock protein upregulation, operating on timescales of hours to days (see references in Table 1). The 4‐day window is therefore best interpreted as capturing an ecologically relevant intermediate timescale, longer than acute stress responses but shorter than seasonal variation.

Stress on day $i$ at site $j$ is the absolute difference between the acclimatized temperature $a_{\mathit{ij}}$ and the temperature recorded that day $T_{\mathit{ij}}$,

$$s_{\mathit{ij}}=\mid a_{\mathit{ij}}-T_{\mathit{ij}}\mid$$ (1)

The acclimatized temperature $a_i$ is calculated with a moving weighted linear regression over a 4‐day window. We assume temperature on the fourth day matters more than temperature experienced on day one in the window, so we weigh recent temperatures more heavily than past temperatures when calculating the acclimatized temperature on day four. Total stress for the year at site j is the integral of the absolute values of daily stress; units are °C.

2.2.2. Stress Definition 2 (Exposure to Thermal Extremes)

This method is similar to how thermal stress is measured on coral reefs in terms of degree‐heating weeks and assumes that thermal stress accumulates when organisms experience temperatures above an upper thermal limit (Boch et al. 2018). We define the site‐specific upper thermal limit ($L_j$) as one standard deviation above the long‐term mean of temperatures observed at site j. This threshold is a statistical approach rather than an experimentally derived physiological limit. It was chosen because species‐wide upper thermal tolerance data for green abalone spanning the full latitudinal range of this study are unavailable at the spatial resolution required for our analysis. The one standard deviation threshold approximates the temperature at which conditions become infrequently experienced and therefore likely physiologically challenging, analogous in principle to the bleaching threshold approach used in coral ecosystems, though derived statistically rather than experimentally. By defining site‐specific upper thermal limits, we assume there is some local adaptation among sites such that extreme temperatures experienced at one site may not be considered extreme, and therefore stressful, at another site. The use of site‐specific thresholds does not require demonstrating local adaptation experimentally; rather, it reflects the precautionary assumption that populations along a 1200 km coastline may differ in their thermal tolerance, consistent with theoretical expectations of divergent selection under spatially heterogeneous environments (Kawecki and Ebert 2004). Whether such differentiation exists in green abalone populations along Baja California remains an open empirical question.

First, a site's climatology was calculated using a loess function with a 90‐day smoothing window to give the seasonal average daily temperature $C_{\mathit{ij}}$ (Boch et al. 2018). Stress on day i at site $j$ is the difference between daily climatology temperature, $C_{\mathit{ij}}$, and the site‐specific thermal limit $L_j$ for when $C_{\mathit{ij}}>L_j$,

$$s_{\mathit{ij}}=C_{\mathit{ij}}-L_j,C_{\mathit{ij}}>L_j$$ (2)

If the seasonal average daily temperature is cooler than the thermal limit $L_j$, then no stress is experienced that day. Total stress for the year at site j is the integral of $s_{\mathit{ij}}$; units are °C.

2.2.3. Stress Definition 3 (Long‐Term Temperature Deviations)

This method recognizes that organisms may experience stress when ambient temperatures deviate substantially from conditions that maximize key physiology traits, whether these deviations are toward warmer or cooler temperatures. This concept draws from thermal performance curve theory, which suggests that organism performance typically shows a unimodal response to temperature and declining performance at both temperature extremes (Brown et al. 2004; Huey and Stevenson 1979). For this definition, we take advantage of the general shape of the thermal performance curve to describe thermal‐induced stress—we invert the curve so that temperature describes relative stress instead of metabolic performance. In this new formation, intermediate temperatures are associated with the lowest values of stress and temperature extremes (cool and warm) are associated with higher relative stress (Figure 1c). We make the assumption that key physiology traits, such as growth, are on average maximized at the annual site‐specific mean temperature.

We approximate a thermal performance curve for each site using the annual mean temperature at each site and equations from Barneche et al. (2014). We include those equations here (Equations 3 and 4 below) and the parameters Barneche et al. published (see Table S2 in the Supporting Information for parameter descriptions and Table 1 in Barneche et al. 2014).

$$B_0=b_0\left(T_c\right)e^{E_r\left(\frac{1}{{\mathit{kT}}_c}-\frac{1}{\mathit{kT}}\right)}I\left(T\right)$$ (3)

$$I\left(T\right)={\left(1+\left(\frac{E_r}{E_i-E_r}\right)e^{E_i\left(\frac{1}{{\mathit{kT}}_{\mathit{opt}}}-\frac{1}{\mathit{kT}}\right)}\right)}^{-1}$$ (4)

We then invert the thermal performance relationship because high performance is assumed to be low stress (the inverted values we call “stress factor” values, $F_{T_{\mathit{ij}}}$). Next, we calculate the deviation between daily temperature (T _ij ) and the mean temperature at the site (T _{mean, j} –T _ij ), and multiply by the stress factor on day $i$ at site $j$ (Equation 5).

$$s_{\mathit{ij}}=F_{T_{\mathit{ij}}}*\left(T_{\mathit{\text{mean}},j}-T_{\mathit{ij}}\right)$$ (5)

It is biologically unrealistic to let each curve fully adjust to different sites because physiological processes break down at very warm temperatures; we set an upper limit of 25°C based on field observations (Boch et al. 2018). The 25°C serves as an upper limit anchor, so that if a given site has a warm mean temperature, for example 22°C, then small increases in temperature results in a relatively large increase in thermal stress. In contrast, sites with relatively cool mean temperature, the same increment in temperature increase results in relatively less thermal stress. The estimate of total stress at site j is the integral of $s_{\mathit{ij}}$; units are °C.

2.3. Identifying Consistent Refugia

We identified which of the 1‐km sites along the Baja California coastline (n = 1186) were considered relatively low‐stress according to all three stress definitions. A site was considered relatively low‐stress if the site's stress value fell below the 50th percentile, and consistently low‐stress sites meant that stress values were below the 50th percentile for all three definitions. We also identified sites that met 2 of the 3 definitions.

2.4. Isla Natividad Case Study

Isla Natividad is an island in the FEDECOOP with records of annual abalone catch‐per‐unit‐effort (CPUE) from 1993 to 2010 in five fishing zones around the island (Figure 6a,b). While CPUE is not always an accurate index of population abundance because increases in fishing effort can compensate for decreasing population size to give the impression that population size remains constant, we use CPUE in the Isla Natividad case study because annual fishing effort remained stable during this time frame (Rossetto et al. 2015) and fishery‐independent surveys are unavailable.

FIGURE 6.

Isla Natividad case study showing (a) the fishing zones A‐E around Isla Natividad, (b) the annual catch per unit effort of abalone from 1993 to 2010 in each zone, the start and end of the 1997 El Nino event is marked with vertical dashed lines in 1997 and 2000, and (c) thermal stress around Isla Natividad for the three stress definitions: Short‐term variability in temperature, exposure to thermal extremes, and thermal performance curve.

We use mixed effects linear models to explain variability in abalone catch‐per‐unit‐effort. Because abalone abundance declined right after the 1997 El Nino event (Figure 6b) we omitted CPUE data in 1998–1999 in the regression fits. To account for differences in abalone abundance before and after the El Nino event we include a before/after categorical variable in our models, and we include the fishing zone as a random effect because some fishing zones have strong or weak upwelling which affects overall abalone productivity. We use corrected Akaike's information criterion (AICc) model selection to distinguish which of the three possible models (Equations (6), (7), (8)) best describe the relationship between fishing zone, before/after the El Nino event, and thermal stress definition with abalone CPUE. We checked model assumptions by examining QQ plots and residuals vs. fitted values plots for each model. Residuals were approximately normally distributed, indicating that normality and homoscedasticity assumptions were met. To assess temporal autocorrelation, we computed the lag‐1 autocorrelation function (ACF) of model residuals for each fishing zone and each stress definition. No lag‐1 ACF values exceeded the 95% confidence bounds for white noise (Figure S6), indicating that temporal autocorrelation was not significant in any model. Stresses 1–3 are described in Section 2.2. Analysis was done using the lme4 package in R (Bates et al. 2015).

$$\text{annual CPUE}~\text{before}/\text{after}\mathrm{El}\text{Nino}+\text{Fishing Zone}+R\mathit{\text{efugiaDefinition}}1$$ (6)

$$\text{annual CPUE}~\text{before}/\text{after}\mathrm{El}\text{Nino}+\text{Fishing Zone}+R\mathit{\text{efugiaDefinition}}2$$ (7)

$$\text{annual CPUE}~\text{before}/\text{after}\mathrm{El}\text{Nino}+\text{Fishing Zone}+R\mathit{\text{efugiaDefinition}}3$$ (8)

3. Results

3.1. SST to Bottom Temperature Conversion

Sea‐surface temperature was positively correlated with bottom temperature across all 17 validation sites (Figure 3c), and high‐frequency variability at the surface was positively correlated with high‐frequency temperature variability at the bottom (Figure 3a). For sites with strong upwelling exposure (n = 12), the correlation between high‐frequency variability at the surface and bottom followed a simple power regression (R ² = 0.73). For weak upwelling sites (n = 5), surface and bottom temperature variability followed a linear correlation (R ² = 0.98). The mean difference between surface and bottom temperature ranged between 1°C and 5°C, with most sites experiencing a mean difference of 2°C–3°C (Figure 3b).

When comparing thermal stress calculated using in situ temperature loggers versus converted SST‐bottom temperature time series, we found positive correlations across all three stress definitions (Figure 4). Short‐term variability stress values (definition 1) and long‐term temperature deviations relative stress values (definition 3) showed the strongest correlations between logger and converted SST data (R ² = 0.83 and 0.92, respectively) with slopes close to unity. Exposure to thermal extremes relative stress values (definition 2) calculated using converted SST time series tended to overpredict stress compared to temperature logger data (R ² = 0.42), but still showed a significant positive relationship. This suggests that our method produced time series similar enough as if we had deployed temperature loggers at the sea‐surface.

FIGURE 4.

Correlation between stress values calculated from bottom temperature miniDOT logger data (x‐axis) and stress values calculated using modified sea‐surface temperature (y‐axis) for each stress definition (panels). Points represent individual sites. Solid black line is the linear regression of points. The dashed black line is the 1:1 line.

3.2. Spatial Patterns of Thermal Stress

Thermal stress varied substantially along the Baja California coastline, with distinct spatial patterns depending on the stress definition used (Figure 5). When thermal stress was measured by short‐term variability in temperature (definition 1), areas with the lowest stress were primarily concentrated near 28° N latitude, with additional low‐stress pockets dispersed throughout the entire coastline. In contrast, when thermal stress was measured using exposure to thermal extremes (definition 2), low‐stress areas showed a more patchy distribution, with concentrations in both northern and central regions. The long‐term temperature deviations (definition 3) revealed that low‐stress habitat was predominantly located in northern regions between 30° N–32° N latitudes, with relatively high stress conditions in southern areas.

FIGURE 5.

Thermal refugia according to three definitions of thermal stress; (1) short‐term variability, (2) exposure to thermal extremes, and (3) long‐term temperature deviations at 1 km sites along the Baja California coastline. Yellow colors indicate lower stress and darker colors indicate higher thermal stress. Sites with stress values in the lower 50th percentile for all three stress definitions are labeled consistent refugia (indicated with yellow circles along coastline), small black circles are sites that do not meet the consistent refugia criteria.

Despite these differences in spatial patterns, approximately 18% of the Baja California coastline (213 of 1186 one‐kilometer sites) exhibited relatively low thermal stress across all three definitions, with stress values falling below the 50th percentile for each metric. These consistently low‐stress sites were located exclusively near and north of 28° N latitude, with none identified south of Punta Eugenia (Figure 5). An additional 28% of coastline sites (332 sites) met the low‐stress criteria for two of the three definitions.

Within the FEDECOOP management region, which encompasses much of the traditional abalone fishing grounds in Baja California Sur, only areas around Punta Baja and in Isla Cedros met all three thermal stress refugia criteria (Figure 5). However, different areas within the region showed varying susceptibility to different types of thermal stress, with some areas primarily affected by short‐term temperature variability while others were more impacted by prolonged exposure to extreme temperatures. Sites that met all three definitions were highly patchy throughout the region (Table 2).

TABLE 2.

Summary statistics (mean, standard deviation, minimum, and maximum) describing the spatial distribution of consistent thermal refugia patches identified along the Baja California coastline (Figure 5), including the distance separating adjacent refugia patches and the length of individual patches, both in kilometers.

	Mean	SD	Minimum	Maximum
Distance between thermal refugia patches (km)	15.7	61.7	1	478
Thermal refugia patch length (km)	3.4	3.0	1	16

3.3. Isla Natividad Case Study

Model selection analysis revealed that exposure to thermal extremes (definition 2) provides preliminary evidence that explains spatial patterns in abalone catch per unit effort (CPUE) around Isla Natividad. The model including exposure to thermal extremes (definition 2, m2) had the strongest support (wi = 0.56, AIC = 581.7), followed by the model with short‐term temperature variability (definition 1, m1; wi = 0.31, ΔAIC = 1.2). The model incorporating long‐term temperature deviations metrics (definition 3, m3) had substantially less support (wi = 0.12, ΔAIC = 3.0) (Table 3).

TABLE 3.

Model selection table comparing linear mixed‐effects models of catch per unit effort (CPUE) with before/after treatment effects and refugia thermal stress values.

Model	Formula	AIC	dAIC	wi	Refugia_coef	SE	t_value
Short‐term variability (m1)	Before/after + fishing zone + definition1	582.9	1.2	0.31	−0.032	0.010	−3.35*
Exposure to thermal extremes (m2)	Before/after + fishing zone + definition2	581.7	0.0	0.56	0.048	0.013	3.63*
Long‐term temperature deviations (m3)	Before/after + fishing zone + definition3	584.7	3.0	0.12	−0.047	0.054	−0.87

Note: Models ranked by Akaike Information Criterion (AIC), with AIC differences (dAIC), Akaike weights (wi), and parameter estimates for refugia variables (Refugia_coef). All models included fishing zone as a random intercept (n = 73, 5 zones). Asterisks indicate significance (|t| > 2).

All models consistently demonstrated a strong before/after El Niño effect, with CPUE approximately 32 units higher before the 1997 El Niño event compared to after (β ≈ 32.0, t ≈ 9.4 across all models). Among the thermal stress variables, exposure to thermal extremes (definition 2) showed a significant positive relationship with CPUE (β = 0.048, SE = 0.013, t = 3.63), indicating higher catch rates in areas with greater exposure to thermal extremes. Conversely, short‐term temperature variability (definition 1) had a significant negative effect on CPUE (β = −0.032, SE = 0.010, t = −3.35), suggesting lower catch rates in areas experiencing high short‐term thermal fluctuations. The long‐term temperature deviations metric (definition 3) showed no significant relationship with CPUE (β = −0.047, SE = 0.054, t = −0.87).

The best‐supported model (definition 2, m2) explained 61.5% of the total variance in CPUE (conditional R ² = 0.61), with 59.3% attributed to the fixed effects (marginal R ² = 0.59). The small difference between marginal and conditional R ² indicates that random site effects explained only about 2.2% additional variance beyond the fixed effects, suggesting that the before/after the El Nino and thermal extreme exposure refugia variables accounted for most of the explained variance in the model.

The spatial distribution of thermal stress around Isla Natividad varied considerably among the three definitions (Figure 6c), with zones experiencing different combinations of stress types. This fine‐scale heterogeneity in thermal stress patterns was consistent with the broader coastwide analysis, demonstrating that thermal refugia identification depends critically on how thermal stress is defined and measured.

4. Discussion

Our framework utilizes seawater temperature time series and well‐established concepts in physiology and marine ecology to provide three distinct definitions and criteria—increased short‐term variability, greater exposure to thermal extremes, and deviations from annual mean temperature—for quantifying relative thermal stress and identifying potential refugia that are broadly applicable to benthic marine species. Application to the green abalone case study demonstrates that thermal refugia vary substantially along the Baja California coastline, with the pattern and intensity of stress depending on how thermal stress is defined. Each approach to measuring thermal stress revealed distinct spatial patterns of potential thermal stress refugia. This finding suggests that the types of thermal stress most influential to benthic species persistence likely vary spatially at fine scales, emphasizing the importance of considering multiple stress definitions when identifying climate refugia. Importantly, integration of the three approaches also highlights refugia that are robust to different definitions and metrics. Despite differences among individual thermal stress metrics, approximately 18% of the coastline exhibited consistently low stress across all three definitions, with these areas concentrated at and north of 28° N latitude. At one location, Isla Natividad, where detailed catch data allowed for preliminary evaluation of our approach, exposure to thermal extremes (definition 2) best explained spatial patterns in abalone catches, providing initial evidence that this stress metric may relate most closely to observed population responses. This is an important result as this definition is based on oceanographic data rather than species‐specific biological data, thereby providing an approach that is broadly applicable across multiple species. Assessing the generalizability of this approach, as well as additional investigation into if and how refugia support local populations, are critical future next steps.

4.1. Multi‐Scale Patterns of Thermal Stress

The spatial patterns of thermal stress we observed operate across multiple scales, from kilometer‐scale variations within individual bays to broader regional patterns spanning hundreds of kilometers. At smaller spatial scales (< 50 km), we found considerable heterogeneity in thermal stress, particularly in areas with complex coastlines or varying exposure to upwelling. This fine‐scale variation creates a mosaic of thermal environments, consistent with other studies in Baja California (Woodson et al. 2019) and other upwelling‐dominated coastal systems (Salois et al. 2022; Wang et al. 2015). At larger scales (> 500 km), we identified broader latitudinal patterns, with northern regions generally experiencing lower thermal stress across all three thermal refugia definitions.

This multi‐scale variability in thermal stress highlights the importance of considering both local oceanographic processes and regional patterns when identifying potential thermal refugia and more broadly, climate refugia. The heterogeneity we observed around Isla Natividad appears to be representative of coastwide patterns, suggesting that efforts to use oceanographic microclimates for spatial management (Kemppinen et al. 2024; Woodson et al. 2019) could be relevant across large regions of the Baja California coast.

4.2. Evaluation and Ecological Interpretation

The finding that exposure to thermal extremes (definition 2) best explains abalone CPUE patterns around Isla Natividad is initially counterintuitive, but likely reflects the confounding relationship between thermal stress and oceanographic productivity in upwelling systems. Along western Baja California, sites with stronger upwelling experience more frequent thermal excursions—captured by Definition 2—but these same upwelling events deliver nutrient‐rich water that drives primary productivity and food availability for abalone (Ryther 1969; Zaytsev et al. 2003). Definition 2 may therefore partly function as a proxy for upwelling intensity and productivity‐driven habitat quality, rather than purely as a measure of physiological stress. We cannot disentangle these signals with temperature data alone, and caution against interpreting the positive CPUE relationship as evidence that thermal extreme exposure directly benefits abalone. Separating the thermal stress and productivity components of Definition 2 will require datasets pairing temperature records with direct measures of primary productivity or kelp biomass at fine spatial scales—an important priority for future validation work.

However, we caution that this relationship was established at a single location and may not hold true across the entire range of green abalone. While this case study provides initial evidence that thermal stress metrics correlate with observed patterns in abalone abundance, the degree to which this relationship generalizes across the full 1200 km coastline remains untested. We therefore interpret the validation analysis as proof of concept rather than broad empirical confirmation, and encourage additional validation efforts at other locations. The relative importance of different types of thermal stress likely varies geographically depending on local bathymetry and oceanographic conditions related to upwelling (Zaytsev et al. 2003), food availability, and local thermal adaptation and gene flow (Miller et al. 2020).

The strong before/after El Nino effect observed in all models and documented in earlier field studies (Arafeh‐Dalmau et al. 2019; Cavole et al. 2016) suggests that large‐scale climate events can override local thermal stress patterns, at least temporarily. This is not unexpected: a refugia is a location that experiences chronically lower stress under baseline conditions and may recover more rapidly following disturbance, not necessarily a location immune to extreme events (Morelli et al. 2016, 2020). The relevant ecological question is, therefore, not whether refugia avoid El Nino impacts entirely, but whether populations in low‐stress sites recover more quickly after such events than populations in chronically stressed sites. Our data do not have sufficient temporal resolution post‐1999 to test differentials directly, but [look up figure of zones].

4.3. Implications for Fisheries Management

Different areas within the region showed varying susceptibility to different types of thermal stress. Some areas were primarily characterized by high short‐term temperature variability (particularly sites near 28° N around Punta Eugenia and dispersed along the coast), while others were more affected by prolonged exposure to extreme temperatures (notably in central regions and areas south of Punta Eugenia, with relatively lower exposure in the northern regions between 30° N and 32° N). Understanding where different types of thermal stress occur could help harvesters and managers anticipate population responses as oceanographic conditions change. For example, if climate change alters upwelling patterns or increases the frequency of marine heatwaves, it could particularly impact areas already experiencing high exposure to thermal extremes (Arellano and Rivas 2019; Villaseñor‐Derbez et al. 2024). Similarly, if climate change leads to deeper thermoclines, sites characterized by high short‐term variability may see a decrease in short‐term stress (definition 1) (Durazo 2009). Seasonal forecasts of marine heatwaves could inform strategic fisheries management decisions (e.g., quotas, fishing seasons) for populations that are especially susceptible (Holbrook et al. 2020). The approach to defining refugia used in this study provides a foundation for adaptive management strategies that account for varying types and intensities of thermal stress.

Within the FEDECOOP management region, our analysis revealed that only areas near Punta Baja and in Isla Cedros met all three types of thermal stress refugia. Most fishing grounds within the fishing cooperative concessions were not identified as consensus thermal refugia, particularly south of Punta Eugenia. Conversely, most of the consensus thermal refugia areas fall outside of fishing grounds managed as TURFs, which suggests significant challenges for long‐term fisheries sustainability as climate change continues to impact ocean temperature and variability (Arafeh‐Dalmau et al. 2023; Villaseñor‐Derbez et al. 2024) and highlights an urgent need to strengthen the management and protection of potentially resilient areas that currently lack local management and monitoring by strengthening local governance and through an expanded network of protected areas (e.g., Arafeh‐Dalmau et al. 2023).

To strengthen the direct application to fisheries management, future work should investigate the link between thermal stress indices directly to demographic parameters (e.g., growth, survival, and recruitment) (Pörtner and Peck 2010). The Isla Natividad case study provides an initial empirical evidence, but CPUE is an imperfect demographic indicator and was available at only one location. The framework is therefore best interpreted as a relative ranking tool for prioritizing sites for monitoring and potential protection, rather than a predictive model of absolute population dynamics, pending more direct demographic validation.

4.4. Framework Applications and Broader Implications

The framework we developed can be adapted for other marine species and systems, particularly those where multiple types of environmental stress exist (Georges et al. 2024; Gunderson et al. 2016; Kroeker et al. 2023) but have not been systematically compared. Three components of the framework require species‐specific inputs and would need recalibration for marine invertebrate taxa with different thermal responses: (1) the acclimatization window in Definition 1 (ideally set from laboratory recovery experiments), (2) the upper thermal threshold in Definition 2 (which should be replaced by experimentally derived critical limits where available), and (3) the thermal performance curve parameterization in Definition 3 (which should reflect published TPCs or known thermal optima for the target species). The mathematical structure of the framework remains unchanged in each case; only the biological inputs require updating.

Our approach of combining satellite‐derived temperature data with multiple stress metrics provides a cost‐effective method for identifying potential thermal refugia across large spatial scales without requiring extensive field monitoring. The identification of areas that meet multiple types of thermal stress criteria provides a conservative approach that accounts for uncertainty in how organisms respond to different types of thermal stress. This framework integrates well with adaptive management approaches to marine resource management, which emphasize learning through iterative management decisions despite incomplete knowledge (Walters 1986). When it is difficult to determine which type of thermal stress most significantly impacts a population, identifying thermal refugia that satisfy multiple definitions can help reduce management uncertainty and provide more robust conservation targets.

However, our approach has important limitations that should be considered in future applications. The conversion of satellite‐derived sea surface temperature to bottom temperature introduces uncertainty into spatial stress estimates that warrants explicit consideration. For strong upwelling sites (12 of 17 calibration sites), the power law regression between surface and bottom high‐frequency variance explained 73% of observed variance (R ² = 0.73), leaving approximately 27% unexplained. This residual variance most likely reflects physical processes not captured by surface conditions alone, including internal wave propagation and tidal mixing, both of which can decouple surface and bottom temperatures in nearshore environments (Woodson et al. 2019). Our validation analysis (Figure 4) provides empirical quantification of how this uncertainty propagates into cumulative stress estimates. Definitions 1 and 3 showed strong agreement between logger‐derived and converted SST‐derived stress values (R ² = 0.83 and 0.92, respectively), suggesting these definitions are relatively robust to bottom temperature reconstruction error. Definition 2 showed weaker agreement (R ² = 0.42) and a tendency to overpredict stress from converted SST relative to logger‐based estimates (slope = 0.71). This overprediction is a conservative bias for refugia identification: sites we classify as low‐stress under Definition 2 are likely genuinely low‐stress because our conversion systematically assigns them more stress than the bottom logger data would. One possible way to reduce this uncertainty is to use ocean reanalysis products (e.g., E.U. Copernicus Marine Service Information; https://doi.org/10.48670/moi‐00021 or CALCOFI) to provide the time series of environmental variables for refugia definitions but this comes with the cost of coarser spatial resolution.

Our stress definitions also make simplifying assumptions about acclimation timescales (e.g., we use a 4‐day window for green abalone in definition 1, based on observed behavioral recovery patterns in this species) and local adaptation (in definition 3 we use mean annual temperature to define site‐specific “optimal” temperature, T _opt) that may not apply universally. The window of time an organism needs to acclimate to warmer temperatures likely varies by species (Morley et al. 2019; Vinagre et al. 2016) and will, therefore need to be adjusted for other species. Additionally, our assumption about the shape of thermal performance curves (in definition 3 we assume that key physiology traits, such as metabolic rate or organismal growth, shows a unimodel response to temperature change) may not always apply and curves could have different shapes depending on species' phenotypic plasticity and capacity for acclimatization (Schulte et al. 2011). We use one thermal performance curve at each site, but do not consider different curves for different life stages of green abalone, which may differ over ontogeny (Sinclair et al. 2016). The metrics in this study focus solely on temperature, and do not account for many other factors important to green abalone that contribute to climate refugia, such as kelp presence, predator interactions, hypoxia, or disease that can mediate thermal stress responses. Future studies to identify species‐specific climate refugia in marine environments will need to take a more holistic approach beyond temperature.

4.5. Research Priorities and Future Directions

The identification of thermal refugia in marine systems requires continued integration across oceanography, physiology, and ecology. Our results demonstrate that different disciplinary perspectives on thermal stress can lead to different conclusions about habitat quality yet also reveal areas of consensus where multiple approaches agree.

Future research should prioritize: (i) field studies designed to investigate how much different stress metrics across multiple sites and species best correlate with population density or key physiology traits such as growth and long‐term trends in the environment and extreme events, (ii) investigation of how different types of thermal stress interact with other environmental stressors such as ocean acidification and hypoxia, (iii) development of more sophisticated models that account for local adaptation, acclimation capacity, and life‐history variation, (iv) integration of thermal stress with other resilience metrics and frameworks for identifying climate refugia, such as kelp persistence (e.g., Arafeh‐Dalmau et al. 2021). Long‐term monitoring programs that combine oceanographic measurements with population assessments (e.g., Alabia et al. 2021; Ban et al. 2016; Micheli et al. 2024), and fisheries stock assessments that incorporate climate or oceanographic conditions in population assessments (Pepin et al. 2022) will be essential for testing and refining thermal and other climate stress metrics to identify refugia. Additionally, more validation for definition 2 is needed at more sites because this definition had the weakest correlation in the conversion of SST to bottom temperature. Lastly, while the validation case study at Isla Natividad is valuable, it is geographically limited and CPUE data come with limitations as being an accurate proxy of population abundance. Future studies should seek fisheries‐independent data sources to provide additional validation.

4.6. Conclusions

Different thermal stress metrics revealed different patterns of potential refugia, but areas of consensus emerged that suggest areas with consistently low thermal stress conditions along the Baja California coast. These areas are concentrated around 28° N latitude (at Punta Eugenia) and to the north, suggesting greater vulnerability at low latitudes and highlighting opportunities for increased protection and management of identified thermal refugia outside well‐managed fishing concessions. This study is a proof‐of‐concept demonstration of how to identify potential climate refugia in marine environments. The three relative thermal stress metrics were tailored to green abalone as a case study, but their overall design is not species‐specific or system‐specific, and therefore, the approach may be widely applicable across regions and species. These methods and findings provide initial guidance for identifying potentially resilient areas while highlighting the complexity of thermal stress patterns at management‐relevant spatial scales. More broadly, our framework provides a template for how researchers and managers can integrate multiple perspectives on environmental stress to identify and protect refugia.

Author Contributions

Giulio de Leo: conceptualization, funding acquisition, writing – review and editing, methodology, visualization, supervision, validation. Mikaela M. Provost: writing – original draft, methodology, visualization, formal analysis, conceptualization, writing – review and editing, validation. Fiorenza Micheli: conceptualization, funding acquisition, writing – review and editing, supervision, visualization, methodology, validation. Mario Ramade Villanueva: data curation, writing – review and editing. Brock Woodson: funding acquisition, conceptualization, writing – review and editing, visualization, methodology, data curation, validation.

Funding

This work was supported by the National Science Foundation.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Table S1: MiniDOT temperature loggers were deployed on the seafloor at 17 sites; see Figure 2 in main text for a map of site locations. Each site was classified as having “strong” or “weak” upwelling based on the coastline's exposure to Pacific Ocean waters and prevailing upwelling‐favorable winds. Sites located along coastlines with western exposure (west, west‐southwest, or west‐northwest orientations) were classified as strong upwelling sites due to their direct exposure to northwest winds and offshore Ekman transport. Sites with eastern exposure (east, east‐southeast, or east‐northeast orientations) or sites located within protected bays or behind islands that lack direct exposure to the Pacific Ocean were classified as weak upwelling sites due to reduced exposure to upwelling‐favorable conditions.

Table S2: Descriptions of parameters used in Equations (3) and (4) in the main. Original Equations (3) and (4), parameters, and descriptions are published in Barneche et al. (2014).

Figure S1.1: Time series of satellite‐derived sea‐surface temperature (blue line) and miniDOT temperature logger data of the seafloor at 10 of 17 sites.

Figure S1.2: Time series of satellite‐derived sea‐surface temperature (blue line) and miniDOT temperature logger data of the seafloor at 7 of 17 sites.

Figure S2: Spectra of sea‐surface temperature time series at the 17 sites. X‐axis is frequency (units are 1/day), a frequency of 0.5 corresponds to a period of 2 days. Y‐axis is log(power), which can be interpreted as variance in the time series at specific frequencies.

Figure S3: Spectra of seafloor temperature from miniDOT logger data at the 17 sites. X‐axis is frequency (units are 1/day), a frequency of 0.5 corresponds to a period of 2 days. Y‐axis is log(power), which can be interpreted as variance in the time series at specific frequencies.

Figure S4: Correlation between high frequency variance in the seafloor temperature (y‐axis) and high frequency variance in sea‐surface temperature (x‐axis). Green sites are strong upwelling sites and blue sites have weak upwelling. This plot is similar to Figure 3a, but here we show other regressions we tested but omitted from the main paper analysis and their respective R ² values.

Figure S5: Autocorrelation in observed bottom temperature (top) and autocorrelation in reconstructed bottom temperature using the conversion regressions (bottom) for one site, Los Gavilanes. We only include the autocorrelation plots one site because the other 16 sites were similar.

Figure S6: We computed the lag‐1 autocorrelation function (ACF) of model residuals for each fishing zone (Zones A–E) across all three stress definitions. No lag‐1 ACF values exceeded the 95% white noise confidence bounds (±1.96/√n), indicating that temporal autocorrelation in the residuals was not statistically significant in any zone or model.

Data Availability Statement

The satellite‐derived sea surface temperature data used in this study were obtained from the NOAA Coastwatch Environmental Research Division's Data Access Program (ERDDAP; https://coastwatch.pfeg.noaa.gov/erddap/index.html). MODIS Aqua SST data are publicly available and free to download. Bottom temperature data derived from in situ temperature loggers (miniDOT sensors) deployed at 17 validation sites were used for model calibration but are not being made publicly available at this time due to restrictions from collaborating partner communities in Baja California, Mexico. However, the validated SST‐to‐bottom‐temperature conversion methodology and resulting converted temperature time series for all 1186 coastal sites are uploaded to Dryad (DOI: https://doi.org/10.5061/dryad.w3r22815t). Abalone catch‐per‐unit‐effort (CPUE) data from Isla Natividad were provided by FEDECOOP (Federación Regional de Sociedades Cooperativas de la Industria Pesquera Baja California). We have rescaled the confidential raw catch‐per‐unit‐effort data used in our manuscript to preserve the confidentiality of data from the fishing cooperatives in Mexico. All scripts for thermal stress calculations (Definitions 1, 2, and 3), rescaled CPUE data from Isla Natividad, and for the statistical analyses are available in this repository: https://doi.org/10.5061/dryad.w3r22815t. Processed thermal stress data for all 1186 sites (relative thermal stress values for each of the three definitions, normalized to 0–1 scale, and classification as low‐stress refugia) are also archived here. Summary maps and GIS layers showing the spatial distribution of thermal refugia are also included in this archive.