Defining spatial units for analysis

Temperature (°C), oxygen (kPa, expressed as PO₂), and fish distribution (presence-absence) data were spatially organized into strata to provide a common scale of measurement. Strata selection was first based on those designated in the Northeast Fisheries Science Center (NEFSC) fishery trawl survey and were obtained from the NOAA github server (https://github.com/NOAA-EDAB/FisheryConditionLinks/tree/master/NES_BOTTOM_TRAWL_STRATA). Some smaller strata were combined to create larger sampling units to increase sample sizes of temperature and dissolved oxygen data within a stratum based on modifications from Walsh et al. [35] and Chen et al. [36]. Specifically, smaller offshore and inshore strata were combined within the same latitudinal bands. To designate strata, letter codes were provided based on across-shelf location (inshore (I), midshelf (M) and offshore (O) locations), and number codes were provided to designate along-shelf positions with 01–19 from south to north (Fig 1). Two NEFSC strata units spanned across (south to north) the Hudson Canyon. Due to observed differences in physical dynamics between the northern and southern flanks of the Hudson Canyon [37], these strata were divided accordingly, whereby the split north and south flanks were combined with their respective northern and southern counterparts (M05, O05, M04, and O04 in this paper; in Fig 1). In total, 43 distinct strata were defined.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Spatial units (strata) defined for this study.

Letter codes indicate across-shelf position (I = Inshore, M = Midshelf, O = Offshore) and number codes indicate along-shelf position from lower numbers in the south and higher in the north The colors indicate offshore (green), midshelf (magenta), and inshore (purple). The map inset shows the designation of regions across the shelf where SS = Southern Shelf, NYB = New York Bight, SNE = Southern New England, GB = George’s Bank, and GOM = Gulf of Maine. United States state outline data was obtained from the R package ‘maps’ (cran.r-project.org/package = maps).

https://doi.org/10.1371/journal.pclm.0000357.g001

The shelf was also divided into regions with similar hydrographic characteristics based on Walsh et al. [35] and Mountain [38]. Regions included the Southern Shelf, New York Bight, Southern New England, Georges Bank, and Gulf of Maine (Fig 1). Strata were classified within these regions based on their locations.

Temperature and oxygen data

Temperature and dissolved oxygen data, the two environmental parameter inputs for MI calculations, were compiled across the USNES. Bottom values (at depth) for temperature and oxygen were chosen because they represented the spatial habitat where fish were collected in bottom trawl surveys, and can also be a better predictor of species distributions from bottom trawl surveys when compared to sea surface temperature (i.e., [39]). Bottom temperature data were obtained from du Pontavice et al. [40]. This long term, high-resolution bottom temperature time-series is derived from bias-corrected simulations of the Northwest Atlantic Ocean Regional Ocean Modelling System (ROMS-NWA; [41]) for the period 1972–1992 and from raw output from the Global Ocean Physics Reanalysis (Glorys12v1 reanalysis; [42]) for the period 1993–2019. Bottom temperature data were available in a regular 1/12° horizontal grid over the USNES. Temperature data were assigned to specific strata and classified by season and year. Seasons were defined as winter (Dec-Feb), spring (Mar-May), summer (Jun-Aug) and fall (Sept-Nov). For each specific grouping of individual years from 1972–2019 and season, a mean temperature was calculated. The highest resolution calculation was at the stratum, year, and season level to make direct comparisons with the fishery data that were collected during spring and fall seasons.

Bottom dissolved oxygen data from 1972 to 2019 across the spatial domain of the USNES were retrieved from CTD casts fitted with oxygen optodes and accessed through the NCEI database, MARMAP survey, and BCO-DMO database (accessed through https://www.ncei.noaa.gov). There are currently no consistent surveys that measure oxygen at a regular basis within the USNES. Dissolved oxygen units from these data sources were converted into PO₂ if they were not already reported in units of partial pressure. Bottom PO₂ was defined as the deepest measurement from a CTD cast if it was within the bottom 20% of the total water column.

The final number of PO₂ data measurements was 7,752 and highlighted the sparsity of PO₂ data across time and space on the USNES, and provided insufficient coverage needed to calculate mean PO₂ at the resolution of stratum/season/year. With a goal of at least 5 measurements per strata/season/year and a total of 43 strata x 4 seasons x 48 years, ~41,000 measurements would be needed. As such, we used shelf-wide seasonal mean PO₂ values calculated from the available data for the inputs to calculate MI to provide seasonal temporal variation (Table 1). These seasonally averaged oxygen values were used across all years and strata. PO₂ averages over region or interannual time scales was not considered due to uneven distribution of raw data within regions and decades (Fig A in S1 Text).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Levels of oxygen (kPa) used in MI calculations for present and future climatologies.

https://doi.org/10.1371/journal.pclm.0000357.t001

Incorporation of fishery data

Fishery-independent data were used to provide information on species distribution over space and time to define the species-specific MI_crit (see below). Five fish species were selected for this analysis: Atlantic cod, black sea bass, cunner, spiny dogfish, and summer flounder. These species were chosen because they have the required physiological parameters determined from laboratory studies and are collected in the spring and fall NEFSC trawl surveys. Additionally, these species represent an array of cold and warm water species. Data from the NEFSC bottom trawl survey was accessed to include data from 1972–2019 and across the strata domain. After 2008, the NEFSC bottom trawl survey no longer sampled the furthest inshore strata within our “inshore” designations, and complementary trawl surveys were employed including the trawl surveys from the NorthEast Monitoring and Assessment Program (NEAMAP) and Massachusetts Divisions of Marine Fisheries (MADMF). For each fish species and season (spring, fall), data were aggregated into the strata. For each stratum, season and year, a presence or absence was assigned for each fish. Presence/absence data were chosen instead of biomass because the derived species-specific MI_crit represents a threshold of where fish are or are not found. This contrasts to other uses of species distribution data that defines optimal habitat based on conditions that lead to higher biomass or abundance.

Identification of thermal habitat through species distribution models has been assessed before using data from all of these trawl surveys (i.e., [39, 43]), and presence/absence data tends to be more robust to potential error in survey differences. These fishery-independent data sources were chosen because they provide standardized fishery distribution data, have been used in prior studies that develop species distribution models for USNES fishes (i.e., [29]), and are used regularly in stock assessments. Additionally, while some of the focal species in our study can inhabit different portions of the water column, the NEFSC bottom trawl survey has been shown to resolve sufficient data coverage for the stock assessments of pelagic species that are caught in the trawl [44]. Also, for species that are associated with structure (i.e., black sea bass), recent stock assessments showed no bias in estimation of biomass from bottom trawl surveys vs. incorporating fishing gear better suited for rocky, structured bottom [45].

Calculation of species-specific MI parameters

MI (ϕ, Eq 1) was calculated for the five species by first obtaining the species-specific metabolic parameters, A_o and E_o. The natural log of the critical PO₂, or P_crit, from laboratory studies (S1 Table) was regressed against the inverse measurement of temperature (Fig 2). A_o and E_o are species-specific terms derived from the y-intercept and the slope, respectively, of that regression [13]. A_o is a measure of the physiological oxygen supply (i.e., capacity of organism to supply tissues with oxygen, not environmental oxygen supply) and E_o is the temperature sensitivity parameter where a steeper slope (higher E_o) indicates increased temperature sensitivity. MI was then calculated according to Eq 1, (1) where K_B is the Boltzmann constant (K_B = 8.6173324⁻⁰⁵ eV; relates kinetic energy with temperature), T is temperature (K), PO₂ is environmental oxygen pressure (kPa), and B is body mass raised to a scaling coefficient, n. The P_crit is largely insensitive to body mass [46, 47] meaning that the scaling coefficient, n, is typically near 0 (i.e., no effect of body mass on the MI), because physiological oxygen supply evolves to meet demand regardless of size [16]. In this study, comparisons of MI within a single season (e.g., fall) were made with constant averaged seasonal environmental PO₂ (oxygen availability) and varying MI was primarily due to temperature.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. The temperature sensitivity of five USNES species.

The y-intercept (A_o) represents the hypoxia tolerance of the species, while the slope of the line (E_o) indicates the thermal sensitivity of the fish, whereby a steeper (more negative due to inverse temperature) slope indicates higher temperature sensitivity. The cold water species (Atlantic cod, cunner, spiny dogfish) are found to the right of the warm water species (summer flounder, black sea bass), which reflects the range of experimental temperatures tested based on species thermal tolerances.

https://doi.org/10.1371/journal.pclm.0000357.g002

After species-specific parameters were obtained, MI for each species was calculated across all strata, years, and seasons using Eq 1. The stratum-based mean was used for temperature for each year, and for PO_2, a seasonal-based mean across the entire spatial domain and years was used. Additionally, we identified the presumptive limiting MI value (MI_crit) [14] for each species using presence-only subset of fishery data in the fall season (the warmest season) throughout the entire timeseries. Data from fall was used because 1) fall in the USNES is typically the warmest season with concomitantly slightly lower PO₂; and 2) spring tends to be one of the coldest seasons, and without summer fishery data, the combination of the two seasons (i.e., the warmest and coldest seasons) leads to a bimodal distribution of MI that slightly skews the data. Preliminary analyses assessed the potential bias of eliminated spring fishery presence data and found 1–6% change in MI_crit estimate. MI_crit was defined as the 2.5^th percentile of the cumulative distribution function of the subset data (S1 Fig). The MI_crit can also be visualized in an oxygen-temperature state space to show combinations of temperature and oxygen values that lead to MI_crit, where minimal species presences are expected on the right-hand side of the MI_crit line. A full example from the derivation of species-specific parameters to visualizing MI_crit in the oxygen-temperature state space is shown in Fig 3.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Derivation of Metabolic Index parameters for Atlantic cod.

The regression of the ln(P_crit) to inverse temperature (same as Fig 1) (a) is used to define the species-specific parameters, which can be used to calculate MI from temperature and oxygen values for each year of the time series, for example 1972 (b) and 2019 (c). Across the time series, MI values from strata where Atlantic cod were present during fall are used to build the cumulative distribution function of MI for presence-only scenarios (d). The MI_crit (vertical dashed line) is defined as the 2.5^th percentile (horizontal solid line) of the cumulative distribution function. The presumed MI_crit can also be visualized in an oxygen-temperature state-space (e) with the same presence-only fishery data and the corresponding MI values for a given temperature and seasonally averaged oxygen value; species presences are expected to be on the left-hand side of the curve where the oxygen level is sufficient at the given temperature and the corresponding MI values are higher than MI_crit. Because the absolute lowest MI value was not used to define MI_crit, there are some species presences on the right-hand side of the curve. United States state outline data was obtained from the R package ‘maps’ (cran.r-project.org/package = maps).

https://doi.org/10.1371/journal.pclm.0000357.g003

The performance of the presumed MI_crit, which was identified using mean temperature and PO₂ in each stratum, can be assessed based on whether or not the derived line delineates oxygen and temperature conditions where species are likely present (left hand side; see Fig 3e). MI_crit was tested against the raw PO₂ and temperature values from the CTD data that was used to generate the averaged seasonal oxygen calculations. These data were binned to the nearest 0.5kPa and 0.5°C, and merged with the stratum-based fishery data based on the season and CTD cast latitude and longitude. Across the entire time series (1972–2019), fishery data were used to provide the proportion of times fish were present within the 0.5kPa and 0.5°C bins. These proportion of presences were compared to the individual species MI_crit and whether or not presences were more common on the left-hand side of the curve. While using the CTD data provided coarse data coverage, it allowed us to assess the performance of MI_crit on a different environmental data set than what was used to derive MI_crit.

Trends in MI

The change in metabolically suitable habitat across space and time was assessed for both the historical (1972–2019) and contemporary (2010–2019) climatologies (S2 Fig). A historical climatology was used to represent the average of a long-term trend in oceanic conditions within the USNES and provide the largest range of MI values (also to calculate MI_crit from). The contemporary climatology was used to explore the recent effect of warming on suitable habitat and compare to the longer time series; for this reason, we chose overlapping year ranges. For each season, strata were given a “risk” level based on the number of years the stratum MI reached MI_crit or lower across the entire time series. On this scale, a stratum was given a value of 0 if it never reached MI_crit and a 1 if it reached MI_crit in every year of the time series. These “risk” levels were calculated across all years of the time series for both climatologies, historical and contemporary, and for each species. While the contemporary climatology had fewer years to assess (10 vs 48), calculating the “risk” for this time period allowed us to assess the frequency of a stratum reaching MI_crit over the last 10 years in concert with recent warming.

Projection of future MI and habitat loss

Future MI was projected under a doubling of atmospheric CO₂ in the CM2.6 high-resolution global climate model (1/10° ocean component) developed by the NOAA Geophysical Fluid Dynamics Lab (GFDL) [24]. The CM2.6 model simulation is an idealized climate scenario where atmospheric CO₂ increases 1% per year until it doubles by model year 70 and remains stable until model year 80. We used the projected change in ocean bottom temperature and dissolved oxygen from model years 60 to 80, averaging these two decades of projections, to provide a delta value under a doubling of atmospheric CO₂. The temperature projections were available on a monthly time-scale allowing for the calculation of seasonally averaged deltas for each stratum (S3 Fig). The stratum- and season-based deltas were then added to the respective strata and seasonal means from the historical climatology (1972–2019). The CM2.6 model provides projected annual dissolved oxygen (moles/kg) on the same 1/10° spatial scale. To remain consistent with the spatial resolution of PO₂ in the historical climatology, we calculated an average oxygen delta for the entire shelf, which provided one oxygen projection value that was converted to kPa and then added to the oxygen averages for each season (Table 1). Converting oxygen units (moles/kg to kPa) is temperature-dependent, so for each season an average seasonal projected temperature was used, which led to minor seasonal deviations in the PO₂ projection (PO₂ projection delta range 0.50–0.57). Altogether, similar to the analyses on past MI trends, the future projections provided higher resolution temperature estimates and less spatially explicit oxygen estimates.

The projected change in MI was calculated for each species using model projections under a doubling of atmospheric CO₂. The MI values for both the historical and future climatologies were compared with the MI_crit values for each species to determine if a certain stratum was historically metabolically unsuitable or if a stratum will become metabolically unsuitable into the future, respectively. We identified the increase in metabolically unsuitable habitat as synonymous to habitat loss. To assess the potential sensitivity of the average projected oxygen values used, an oxygen level was calculated for each species and season where 50% and 100% habitat loss would occur. These oxygen levels were back-calculated using the species-specific MI_crit and projected stratum-based temperature to provide the limiting PO₂ level for each stratum, and the PO₂ level that resulted in an accumulation of 50 and 100% habitat loss was identified. These projected limiting oxygen levels were used to compared with the oxygen projections used in our analyses to assess the degree of additional decrements in oxygen that would lead to further metabolically suitable habitat loss.

Species-specific MI_crit

For the five USNES species, the three colder water species (Atlantic cod, cunner and spiny dogfish) had slightly lower MI_crit values than the warmer water species, black sea bass and summer flounder (~2.8–2.9 vs ~3.2; Table 2), which was still relatively similar compared to MI_crit values derived across many marine species [14]. MI_crit performance was assessed based on the ability of MI_crit to delineate where species occur in an oxygen and temperature state space based on raw CTD data that was not used to define MI_crit. Under this metric, MI_crit performed well for Atlantic cod and cunner (Fig 4). For spiny dogfish, MI_crit was less able to delineate species presence/absences at colder temperatures, which suggests oxygen may not be a limiting mechanism below ~17°C. For the warm water species, black sea bass and summer flounder, MI_crit did not perform as well. First, there were a number of presences on the right-hand side of MI_crit, which suggests fish were caught in ocean conditions that should be metabolically limiting using this framework. Second, the proportion of presences increased in temperature and oxygen conditions near MI_crit and species absences were more prevalent further away from MI_crit. This suggests that for the warm water species MI_crit either does not define conditions that lead to limiting habitat or the warm water species are living near their habitat limits within the USNES.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Performance of species-specific MI_crit.

The species-specific MI_crit (black line) indicate the oxygen and temperature conditions that lead to MI_crit. The heat map of proportion present (yellow = present to purple = absent) for each 0.5°C and 0.5kPa grid is also plotted in the temperature and oxygen state space for each species. If MI_crit successfully delineates species presence/absence based on environmental conditions, then species presences would be expected on the left-hand side of the line and absences on the right-hand side of the line.

https://doi.org/10.1371/journal.pclm.0000357.g004

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. The species-specific parameters defined for MI calculations.

https://doi.org/10.1371/journal.pclm.0000357.t002

MI trends: 1972–2019

Within the framework of assessing the number of times MI reached or fell below MI_crit, winter and spring never reached MI_crit for any of the species, which suggests no limitations on metabolic habitat suitability in these seasons. In the summer and fall, portions of the shelf reached MI_crit or below throughout the time series, but the degree to which MI_crit was repeatedly reached differed by location and species. During summer, metabolically limiting habitat was primarily constricted to the inshore and southern portions of the shelf (Fig 5), and the proportion of times a stratum was at or below MI_crit increased in those locations during the recent warming period (2010–2019; Fig 5). During fall, the proportion of observations of sub-critical MI increased in strata northward and eastward across the shelf compared to summer (Fig 6), and this pattern intensified during the recent warming period (Fig 6). Across the different USNES species, the cold water fishes (i.e., Atlantic cod, cunner and spiny dogfish) experienced a higher instance and amount of habitat at or below their MI_crit, while the warm water species rarely experienced metabolically limiting habitat except for the most southern and inshore portions of the shelf.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Strata “at-risk” during summer.

The strata “at-risk” based on proportion of years a stratum was at or below a species-specific MI_crit across the historical climatology (top panel) and contemporary climatology (bottom panel) during the summer. If a stratum is tan (0% risk), then it never reached MI_crit during any year across the respective timeseries while if a stratum is dark purple (100% risk), then that stratum reached or surpassed MI_crit for 100% of the years. United States state outline data was obtained from the R package ‘maps’ (cran.r-project.org/package = maps).

https://doi.org/10.1371/journal.pclm.0000357.g005

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. Strata “at-risk” during fall.

The strata “at-risk” based on proportion of years a stratum was at or below a species-specific MI_crit across the historical climatology (top panel) and contemporary climatology (bottom panel) during the fall. If a stratum is tan (0% risk), then it never reached MI_crit during any year across the respective timeseries while if a stratum is dark purple (100% risk), then that stratum reached or surpassed MI_crit for 100% of the years. United States state outline data was obtained from the R package ‘maps’ (cran.r-project.org/package = maps).

https://doi.org/10.1371/journal.pclm.0000357.g006

The amount of warming during the recent climatology from 2010 to 2019 led to an increase in the amount of habitat at or below critical limits for Atlantic cod, cunner and spiny dogfish when compared to the historical climatology. For example, during the fall for Atlantic cod, the number of strata that reached MI_crit at least 75% of the time increased from 7 to 10 strata out of 43 total (16.28 to 23.26%). Cunner and spiny dogfish also experienced an increase in number of strata reaching MI_crit at least 75% of the time but to a lesser degree than Atlantic cod. For black sea bass and summer flounder, there were no strata that reached MI_crit at least 75% of the time during 1972–2019, yet during the 2010–2019 climatology, 1 stratum (I01; summer and fall) reached MI_crit 80% of the time for summer flounder. This indicates that recent warming is increasing the frequency that certain portions of the shelf become metabolically unsuitable.

Projection of future MI and habitat loss under a doubling of atmospheric CO₂

Presumed habitat loss, defined by locations where MI is less than MI_crit, is substantial under a doubling of atmospheric CO₂ (Fig 7) for Atlantic cod, cunner, and spiny dogfish, particularly in summer and fall. The warm water species, black sea bass and summer flounder, are projected to experience minimal habitat loss. Winter and spring show minimal to no habitat loss (S4 Fig), respectively, for all species. Although warming is occurring in winter and spring, the absolute temperatures are not warm enough to reach MI_crit for any examined species. During summer, the cold water species are projected to experience some metabolically suitable habitat loss in the southern inshore and midshelf portions of the USNES as well as inshore portions of Southern New England; fall is projected to see the largest amount of metabolically suitable habitat loss with some species (i.e., Atlantic cod) habitat restricted to the Gulf of Maine.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 7. Future projections of metabolically suitable habitat loss.

Projections are shown under a doubling of atmospheric CO₂, 60–80 years into the future for each species for both summer (top) and fall (bottom). Individual species are shown in columns. Green strata indicate MI_crit was reached in the current climatology (1972–2019), pink indicates MI_crit will be reached under a doubling of CO₂ and blue shows that the stratum does not reach limiting MI_crit values into the future. United States state outline data was obtained from the R package ‘maps’ (cran.r-project.org/package = maps).

https://doi.org/10.1371/journal.pclm.0000357.g007

There were substantial differences in loss of metabolically suitable habitat across the five species. For Atlantic cod and cunner, most of the USNES south of the Gulf of Maine is projected to reach MI_crit in the fall, with slight refugia in the midshelf portion of the New York Bight. During summer, these two species are projected to experience less metabolically suitable habitat loss than in fall with inshore southern strata as well as inshore strata in Southern New England reaching MI_crit. Spiny dogfish are projected to experience a loss of metabolically habitat in the Southern Shelf and inshore portions of the New York Bight and Southern New England. Black sea bass and summer flounder may experience minimal loss in habitat under a doubling of CO₂, reaching MI_crit only in the southern inshore and midshelf portions of the Southern Shelf. This result may reflect the fact that the presumed MI_crit for each of the warm water species did not perform well within the study domain of the USNES.

The limiting oxygen levels to reach 50 and 100% habitat loss under projected warming differed by species and season (Table 3), and follows the progression of losing ~5% of MI per kPa loss below air-saturation, regardless of temperature [16]. During fall, the oxygen value used in the above projections (Fig 7) led to 50% habitat loss or more for Atlantic cod, and would have to drop an additional ~1kPa to reach 50% habitat loss for cunner and spiny dogfish. For black sea bass and summer flounder, oxygen would have to decrease to ~13kPa for a 50% loss of habitat, a greater than 3kPa decrease from the value used in the projections, and far greater than any drop expected from large-scale ocean deoxygenation [48]. Across all seasons, the cold and warm water species each had similar oxygen levels that would lead to 100% habitat loss at ~13kPa vs ~10kPa, respectively. These oxygen levels may represent an oxygen threshold for those species where regardless of a wide range of temperatures across the seasons (~5-25°C), habitat would become metabolically unsuitable throughout the USNES.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. The oxygen level (kPa) for 50 and 100% habitat loss for each species and season under a doubling of atmospheric CO₂.

https://doi.org/10.1371/journal.pclm.0000357.t003

Species-specific differences in MI_crit

For the five USNES fish species, MI_crit values fell within the 2–5 MI range proposed by Deutsch et al. [13, 14], but differed between species based on native thermal ranges. MI_crit values were lowest in the colder water species while the warmer water species had higher MI_crit values. For the cold water species, Atlantic cod and cunner, the MI_crit (defined by mean conditions for a stratum) agreed well with temperature and oxygen conditions delineating the boundary of individual fish occurrences across strata and seasons (Fig 4). However, for the warm water species, black sea bass and summer flounder, MI_crit did not appear to discriminate habitat where fish are absent and fish presences were higher closer to and below the MI_crit (Fig 4). These fish also have southern conspecifics inhabiting the US Southeast shelf, but the degree of genetic similarity between the northern and southern stocks differs for the two species. For black sea bass, the southern and northern stocks are identified as genetically distinct [52], yet for summer flounder there is no major genetic dissimilarity between the two stocks [53]. Moreover, Deutsch et al. [14] identified MI_crit values for 72 marine species, including some in this study. Our MI_crit values for black sea bass and summer flounder were almost identical to Deutsch et al. [14] (S2 Table), which is intriguing because both studies used different data sources both for the physical parameters and for species distributions. Because Deutsch et al. [14] also used species distribution data from the southern stocks of black sea bass and summer flounder, we suggest that MI_crit may be an evolutionary trait that can delineate metabolically suitable habitat at the lower latitudinal limits of an entire population, but may not represent physiological limits of higher latitude conspecifics with localized adaptations.

For Atlantic cod and cunner, our MI_crit values differed from Deutsch et al. [14]. In addition, species-specific E_o values between the two studies were similar for cunner but differed for Atlantic cod (S2 Table). For Atlantic cod, physiological values used to define E_o and A_o were from a different study than in Deutsch et al. [14] because we purposely only used data from studies that measured P_crit via the onset of anaerobic metabolism and not mortality (i.e., L50). MI_crit disagreement for the cold water species may also be due to our study using a more spatially explicit domain (i.e., USNES) than in Deutsch et al. [14]. The spatial domain and resolution of our environmental and fishery data were at the strata level covering the extent of the USNES while for Deutsch et al. [14] data were analyzed beyond the USNES and in a 1° x 1° grid. Our temperature and oxygen measurements also differed in that Deutsch et al. [14] used modeled data from the World Ocean Atlas in the same 1° x 1° grid. This resolution is insufficient to use for analyses assessing local dynamics along the USNES due to the size, complex bathymetry, and oceanography of the shelf, which require finer-resolution models to resolve the biogeochemistry [24]. While we simplified oxygen, temperature dynamics were more representative of the USNES dynamics than would have been resolved at a 1° x 1° grid.

Historical and contemporary MI trends

Across all seasons, MI decreased between the historical (1972–2019) and contemporary (2010–2019) climatologies, but MI fell below MI_crit for the cold water species (and rarely for the warm water species) only in the summer and fall seasons. The USNES experiences cold bottom temperature during winter and spring seasons, and starts to warm during late summer and fall when storms pass through mixing the warm surface layer to depth [54]. As such, late summer and fall are the warmest seasons for bottom temperature [55]. Therefore, most demersal species on the USNES are more likely to experience limiting MI values in the late summer and fall when the temperature is highest. Summer and fall coincide with fish spawning seasons and juvenile settlement, both life stages that can have narrower thermal windows [56]. This suggests that, at the warm limits of a species distribution, habitat may be constrained by the changes in summer and fall bottom temperature. However, within the domain of the USNES, this trend was only evident for the cold water species and there was minimal metabolically suitable habitat loss for the warm water species, which only experienced loss in the most inshore and southern strata.

While warming across the 1972–2019 time period has been occurring in winter and spring, no fish in this study experienced a sub-critical MI value during winter and spring throughout the study period. Warming winters may be problematic for some species, such as Atlantic cod in Southern New England [51], yet this was not apparent in our data. However, for some USNES species, winter warming may be beneficial, especially to warmer water species living at their northern range edges (i.e., black sea bass, [52]). Warmer winters can facilitate increased young-of-the-year survival [57] and can benefit species by increasing recruitment [58]. MI values well above MI_crit are of unknown significance, and as such, the consequences of warming in winter and spring may be better explored using another physiological metric that can distinguish suitable habitat based on cold-intolerances.

Our MI trend results provide a broad account of how the extent of metabolically suitable habitat has changed over time, but may mask finer-scale and/or new emerging oceanographic phenomena that lead to abrupt, intense and abnormal periods of warming within a season. For example, marine heatwaves are becoming more persistent in the USNES [30, 31], increasing water temperature for periods of time throughout a given year. Specific to the USNES, the summer season may also experience rapid warming driven by an increase in warm core ring activity. These warm core rings are eddies shed off of the Gulf Stream that can distribute warm and salty water onto the mid- to outer-shelf during summer months [59, 60]. Altogether, the increase in frequency and magnitude of abrupt and intense warming events could lead to rapid decreases in MI on shorter temporal scales not demonstrated in our data. The cumulative effect of these events is unknown, but how species respond to these changes may differ based on their propensity and/or time-course to return to previously thermally inhabitable water [61].

Future metabolically-suitable habitat loss

Under a doubling of atmospheric CO₂, future metabolically suitable habitat loss, as defined as strata where MI is at or below MI_crit, will be substantial for the cold water species, yet the warm water species will experience minimal habitat loss. Fall may see the highest extent of habitat loss, with many species restricted to the Gulf of Maine and small mid-shelf portions in the New York Bight. Habitat loss may also occur in summer, mainly for inshore/mid-shelf and southern portions of the USNES, but increased model resolution including higher frequency of warm core ring intrusions would project warmer temperatures on the mid- and outer-shelf than used in our model [60]. We project no spring- and only minimal winter loss of metabolically suitable habitat loss in the Southern Shelf. Despite a decrease in MI, winter warming could also lead to the opening of new habitat in mid-shelf and inshore regions that were historically too cold. Finally, while our projections provide individual assessments on season-specific habitat loss, an unexplored but important component to consider will also be the connectivity between populations who undergo seasonal migrations, and how shifts in distribution during one season may alter habitat use in other seasons.

Atlantic cod may experience the most amount of habitat loss under a doubling of atmospheric CO₂, leading to metabolically suitable habitat restricted to the Gulf of Maine, as has been shown in other studies focused on Atlantic cod (i.e., [39]). Importantly, MI_crit was defined for Atlantic cod across all fish presences within the NOAA NEFSC bottom trawl survey, yet within the USNES, Atlantic cod is managed as two stocks (Gulf of Maine and Georges Bank), and Southern New England fish are managed with the Georges Bank stock, although they may be genetically different [62]. Therefore, within this dataset, the different cod stocks were not differentiated when defining MI_crit (and were also not distinguished in the laboratory physiology studies). The general shift in Atlantic cod distribution has been poleward and deeper, especially as Atlantic cod are at their southern population limits in Southern New England and Georges Bank [26]. Our assessment between the historical and contemporary time periods would also support this observation.

Black sea bass may experience minimal habitat loss into the future, and only in a few areas in the most southern inshore and midshelf portions of the USNES. Black sea bass are a warm water species, and the Northern stock (the focus of this study) are the most northern within the full range of black sea bass [52], which may influence MI_crit (see above). Winter and spring warming have been linked to black sea bass distribution shifts in the spring [27] and increased recruitment from larval overwintering survival [58], which suggests warming in these seasons may be beneficial for black sea bass. Walsh et al. [35] showed that adult black sea bass center of biomass rate of movement was nonsignificant during fall, and perhaps extirpation of black sea bass out of warming regions has not occurred yet. As such, black sea bass will likely experience lower amounts of habitat loss in the summer and fall seasons compared to the other USNES species. Black sea bass maximum tolerable temperature was around 24°C [63]. The only stratum in this data set whose mean temperature reached 24°C or higher was stratum I01 (inshore Southern Shelf; Fig 1), which started to reach this temperature in the early 2000s.

Cunner were similar to Atlantic cod in that they may experience a substantial amount of habitat loss into the future. Out of the five species studied, cunner are the only non-targeted species by recreational and commercial fisheries, and previous attempts at discerning thermal habitat have been unsuccessful [64]. Cunner abundance has decreased in inshore portions of Southern New England [65] and has increased in parts of the Gulf of Maine [66]. We project that cunner will experience suitable habitat loss within the Southern Shelf, New York Bight, Southern New England, and Georges Bank. This result corroborates with Morley et al. [29] who showed a shift in distribution poleward with accompanying southern habitat loss.

Spiny dogfish projected summer and fall habitat loss was less extensive than Atlantic cod or cunner, but showed significant loss in the Southern Shelf and inshore regions throughout the USNES. This result contrasts with other studies that demonstrate a benefit to spiny dogfish with recent warming [28, 39, 49, 67], and a potential southward distribution shift [26]. Importantly, some of these studies (e.g., [39]) analyzed change in thermal habitat during the spring (March, April, May), where we also show no habitat below MI_crit (i.e., no habitat loss in spring; S4 Fig). Physiology-based studies on spiny dogfish have shown 21°C to be a physiological thermal limit under normoxia [68, 69], and under a doubling of atmospheric CO₂, spiny dogfish thermal limit would be around 16°C (at PO₂ = 16.64kPa). A challenge for identifying current and future changes to spiny dogfish is also driven by their highly migratory behavior and division into two distinct subpopulations, which can differ in their thermal tolerances [70]. The animals used in the laboratory study that provided species-specific parameters likely captured individuals from the northern subpopulation of spiny dogfish [69], which are shown to have lower thermal tolerances than their southern counterparts [70].

Summer flounder minimal projected habitat loss is similar to black sea bass for both summer and fall. Interestingly, summer flounder distribution has been shifting northward and eastward over time in both spring and fall [27]. Suggested drivers of this shift include bottom temperature trends [7], climate variability with the Gulf Stream Index [71], and fishing pressure [27]. However, other studies suggest the drivers remain unresolved as temperature and fishing pressure explain some but not most of the variation in the data [72]. Our MI_crit analyses suggest that temperature is a driver in limiting habitat at the most inshore and southern portions of the shelf, and other temperature dynamics along the USNES may affect the population through other mechanisms (i.e., [73]).

Considerations when building the MI framework

Species-specific physiological parameters are valuable tools that can be used to investigate how changing environments may affect species distributions. Most papers focused on species distribution and temperature use statistical relationships between where fish are found and the temperatures they reside in to develop species distribution models (e.g., [28]). These studies are instrumental for investigating how ocean warming will affect the USNES ecosystem as they can provide estimates of optimal habitat across a range of temperatures as well as incorporate more physical features such as salinity and rugosity into their models. Only focusing on temperature can sometimes mask true vulnerability to environmental change [50], especially for some benthic organisms with a high affinity towards structure [74]. The use of MI differs in that oxygen is also incorporated and the assessment of suitable habitat is based on current physiological understanding, using data that are defined independently of the collection of fishery data. Also, the framework of assessing MI with respect to MI_crit primarily focuses on defining habitat limits instead of optimal habitat. For species distribution models that do focus on habitat limits, developing an MI-based habitat model can be a complementary method that, along with oxygen data, provides an additional step of interpretation between statistical correlations of temperature to presence/absence and physiological limits that were derived in the laboratory. When the two methods are in alignment (i.e., Atlantic cod), it can be theorized that the trends seen in the species distribution models are likely driven by oxygen-limited thermal tolerance. However, when the two methods do not agree (i.e., spiny dogfish), additional data gaps are exposed that can guide subsequent research areas to improve our projections.

By defining limiting habitat, MI is more useful in identifying and tracking the shift in species ranges at the warm edge, which necessitates a clear shift in the edge that is synchronized with changes in temperature [75] and occurs within the spatial domain of the study. Furthermore, the MI approach is predicated on the assumption that temperature-dependent hypoxia tolerance determines the warm-water range edges, which may not be true for some species. If the range edges are set by some other factor, then MI_crit may be misleading or irrelevant [18]. Understanding the context of MI_crit is also relevant for black sea bass and summer flounder where MI_crit did not seem to distinguish limiting habitat within the USNES, which may have been a consequence of defining habitat limits in the northern portion of an entire range. However, our study domain was purposefully constrained to the USNES which is defined by specific spatial management zones used by NOAA’s Northeast Fisheries Science Center as well as regional management councils that include the Mid-Atlantic and New England Fishery Management Councils. Our results demonstrate the use of one physiological metric, MI, may be better suited for species whose range edge occurs within the USNES, which is an important distinction when identifying appropriate physiological metrics that could be integrated into management frameworks.

How the species-specific metrics are defined is important to consider when making comparisons across species and applying them towards describing and projecting species responses to environmental change. The calculation of the laboratory-based species-specific parameters used in the MI equation come from studies that measure P_crit at multiple temperatures and the methodologies used to calculate P_crit can differ between studies (e.g., [76, 77]). Many laboratory experiments subject animals to acute exposures of different temperatures and oxygen, obfuscating the potential to understand acclimation potential, but the duration of exposure can affect performance of particular physiological measurement end points [63]. As such, while we may have shown differing sensitivities between the cold and warm water species, it is still unknown whether the species would respond similarly to a slow, climate-induced temperature change. Finally, for species with subpopulations or spatially explicit stocks, the collection locations of experimental animals may also skew the derived species-specific parameters should P_crit (i.e., hypoxia tolerance) differ between these subpopulations or stocks. Altogether, these caveats highlight the need for continued assessment of species physiological performance in the laboratory so that potential influence in study design can be disentangled from actual responses in the natural environment.

In this study, we did not use modeled bottom ocean PO₂ because the sparsity of PO₂ observations in the USNES precluded model validation, which led to a coarser investigation on MI in the USNES. High-resolution ocean models are required for the USNES due to its high complexity in regional ocean circulation and bathymetry [24], and biogeochemical models that can resolve PO₂ at these resolutions are not common or have yet to be validated. We did test a modeled oxygen product for potential use, but found poor agreement with the raw data (methods described in S1 Text). However, the modeled oxygen product provided a range of spatial anomalies in oxygen data that could potentially represent variation in oxygen from the seasonal averages we used. To investigate the potential source in error when calculating MI, the oxygen spatial variability defined in the modeled oxygen was tested using a Monte Carlo simulation. Overall, we found our seasonal oxygen means did not significantly affect our interpretation of changes in metabolically suitable habitat (full explanation in S1 Text). In addition, the relationship between environmental PO₂ and MI is documented (i.e., [16]), and can be used to identify threshold PO₂ values that lead to substantial habitat loss (Table 3). While it is very unlikely these oxygen levels would represent future oceanic conditions in the USNES, they are useful when gauging the potential effect of hypoxic events. As methods and technology for ocean observing advance, future incorporation of PO₂ values could be used with critical threshold PO₂ values to identify regions of concern along the USNES.