Study setting

The CCE is an eastern boundary upwelling system extending from Baja California, Mexico (~27° N) to the North Pacific Transition Zone (~50° N) [62]. Ecosystem conditions vary between three oceanographic seasons (i.e., Davidson Current, Upwelling & Oceanic), with distinct effects on the distribution of marine biota [61,65]. There are a number of fixed management areas in the CCE, including a network of NMS (Fig 1). These sanctuaries were established to safeguard both natural and human heritage resources in the CCE [66]. A number of highly mobile pelagic species of ecological and conservation importance (e.g., protected species) occur seasonally within these NMS for diverse habitat uses including feeding and reproduction [67]. Seabirds have played a particularly important role in the NMS system since their inception. Colony locations were important information used in the designation of some NMS in light of, e.g., emergent oil spill risks [68–70]. Seabird abundance and diversity have also been proposed as quantitative ecological indicators of sanctuary condition [71]. NMS contain seabird assemblages with elevated levels of both metrics compared to surrounding areas [35, Santora et al. 2021; 63], and also feature in some sanctuary management plans (e.g., [72,73]). We included four of the six CCE NMS in our analyses: Chumash Heritage, Monterey Bay, Greater Farallones, and Cordell Bank (Fig 1). Olympic Coast NMS - the northernmost NMS and a hotspot for seabird prevalence located in Washington State off the Olympic Peninsula (Table 1) - was excluded because it falls predominantly outside of the model domain. Channel Islands NMS was also excluded due to the relatively low prevalence or complete absence of our focal species (Table 1), and the small coastal domain that is challenging to compare with other NMS that extend continuously from the coast to the continental slope.

Download:

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

Table 1. Summary of geographic properties and prevalence of seabird species from visual surveys within areas examined in the present study including the broader California Current reference areas, U.S. National Marine Sanctuaries, and offshore Wind Energy Areas (see Fig 1). BFAL = black-footed albatross; CAAU = Cassin’s auklet; COMU = common murre; RHAU = rhinoceros auklet; SOSH = sooty shearwater.

https://doi.org/10.1371/journal.pclm.0000687.t001

Offshore wind energy is a burgeoning new ocean-use sector within the CCE. The proposed development of offshore wind energy in the U.S. West Coast utilizes floating-platform technology, anchored to the seabed with transmission cables suspended in the water column in areas with depths exceeding 60 m along the continental shelf [74]. The U.S. Department of Interior’s Bureau of Ocean Energy Management (BOEM) is responsible for designating and leasing areas of the outer continental shelf for potential development in federal waters (> 3 nm offshore). The designation of areas for offshore wind energy development (i.e., WEAs) is guided by both national and state goals for renewable energy production and by marine spatial planning efforts that seek to avoid or minimize conflicts with important ecological components (e.g., critical habitat and species-of-interest) and ocean users (e.g., fishing and shipping industries). As of December 2024, there have been four WEAs proposed off the coasts of California and Oregon (Table 1). The two proposed WEAs off California are located respectively off Morro Bay and Humboldt in the north and are ~ 30 and 50 km offshore in water depths > 500m (Fig 1). The two proposed WEAs off the coast of Oregon are respectively located off Brookings and Coos Bay ~ 35 and 60 km offshore in water depths > 700m (Fig 1). There are a host of questions related to the potential impacts of this new industry on various components of the CCE, including seabirds [75]. A recent review suggests that the known impacts of offshore wind farms on seabirds have been consistently negative in other geographic areas [76]. In the CCE, collision risk at wind farms has previously been estimated as a function of seabird density and wind speeds [65]. However, displacement risk caused by climate-induced changes in suitable seabird habitat in the vicinity of WEAs remains unquantified. Given the cost and long lifespan of a typical offshore wind farm [~25 years; 77], it is important to consider potential species redistributions caused by this type of deleterious interaction in spatial planning efforts that seek to minimize potential ecological risks [27].

Seabird data

Our five focal species numerically dominate the CCE avifauna [61,63], and include two seasonal migrants: black-footed albatross (Phoebastria nigripes) and sooty shearwater (Ardenna grisea), that typically inhabit the region from late-winter through early fall (~March - October; Fig 1), as well as three species that are year-round residents: Cassin’s auklet (Ptychoramphus aleuticus), rhinoceros auklet (Cerorhinca monocerata), and common murre (Uria aalge). The conservation statuses of these species range from Near-Threatened (black-footed albatross, Cassin’s auklet, common murre) to Least Concern (rhinoceros auklet, sooty shearwater) on the International Union for Conservation of Nature Red List. We used data from the at-sea survey compilation presented in [78], which collated sightings from many scientific programs operating between 1980–2017 with strip-transect data divided into segments that were typically 4 linear km but included some shorter segments. We excluded segments shorter than 500 m in accordance with [78]. This compilation (Fig A in S1 Text) includes both ship-based (n = 91,011 segments) and aerial transect surveys (n = 41,215 segments), and has been used previously to assess spatiotemporal variation in seabird distributions throughout the CCE (e.g., [63,78]).

Environmental data

Environmental data used to train SDMs included a suite of dynamic variables extracted at the center of transect segments (Fig B in S1 Text). Environmental fields were obtained from two data-assimilative ocean model products for the CCE (https://oceanmodeling.ucsc.edu/) - a historical reanalysis spanning 1980–2010 and a near-real-time product for 2011–2017. Both model products are based on the Regional Ocean Modeling System (ROMS) and span the CCE from 30 to 48 °N and from the coast to 134 °W at 0.1° (~10 km) horizontal resolution [79]. The environmental fields used included sea surface temperature (sst), total kinetic energy (TKE), wind stress curl (wind_stress), upper ocean (0–200 m averaged) buoyancy frequency (bv), and isothermal layer depth (ild, defined by a 0.5°C departure from sst). The standard deviation of sea surface temperature (sst_sd) was calculated over a 3x3 neighborhood of pixels (0.3° x 0.3°) and included to characterize the spatial variability in SST. Together, these variables represent important oceanographic predictors of seabird habitat and/or prey distribution and abundance [80–83]. To account for seasonality in distributions, photoperiod (daylength) was also calculated using the day of year and mean latitude for each survey location using the ‘geosphere’ R package [84]. We also included an oceanographic season (season) variable, as seabird distributions vary across the year according to breeding phenology and in response to different upwelling-related regimes within the CCE – defined as Davidson Current (Nov. 15 - March 14), Upwelling (March 15 - Aug. 14), and Oceanic (Aug. 15 - Nov. 14) seasons (see also [61,65]). These seasons also roughly track breeding seasons, when spatial distribution patterns may be influenced by the constraints of central place from a limited number of suitable colony sites. Cassin’s auklet, common murre, and rhinoceros auklet breed during the Upwelling season; black-footed albatross breed during the combined Davidson/Upwelling seasons while sooty shearwater breed in the Southern Hemisphere, primarily in the Upwelling season [35,85–87]. We also included a region (province) variable to allow responses to variables to differ among biogeographic provinces [62–64,88,89]: 1) South of Point Conception (~34.45° N), 2) Central, from Point Conception to Cape Mendocino (~40.44° N), and 3) North of Cape Mendocino (Fig 1A). Bathymetric depth (z) and the standard deviation of depth (z_sd) at a resolution of 0.3° were included from the ETOPO1 1 arc-minute product [90]. The standard deviation of depth broadly represents the rugosity of the seafloor, and serves as a proxy for topographic features (e.g., the continental shelf breaks, the slopes of seamounts) that act as key foraging areas for marine predators [91]. We also present outputs from SDMs fit and projected without the fixed effects of province and season given uncertainties of assuming stationarity in species-environment responses under climate change (S2 Text).

Species distribution modeling

The goal of our modelling effort was to assess the spatiotemporal distribution of suitable habitat for each species from 1980-2100. We used presence-absence data for model fitting, and refer to our model outputs as a habitat suitability index (HSI) based on the assumption that seabird species are preferentially located in physiologically favorable habitats. For each species there were many more absences than presences (Fig A in S1 Text), which can bias SDM results [92,93]. Therefore, we selected absences for each species prior to model fitting to achieve a 1:1 ratio of presences to absences [94]. An equal number of absences were randomly selected from each decile of SST values for each species to minimize potential biases to a specific region or time of year due to prominent data-rich surveys that typically occur during spring and summer (e.g., California Cooperative Oceanic Fisheries Investigations [CalCOFI]). SST was selected for absence thinning because it varies spatially and seasonally and thus would sample absences across the range of conditions seabirds experience in the CCE. In addition, we expected SST to be relatively important in the models given the importance of temperature seabird SDMs in the region and elsewhere (e.g., [55,78]), and thus we sought to have the full range of SST conditions represented in the absences.

Boosted regression trees (BRTs) were used to fit SDMs. BRTs are efficient machine-learning models chosen for their flexibility in modelling non-linear responses, robustness to collinear predictor variables, and in accordance with established workflows that have been successfully employed for seabirds elsewhere [95] and for a range of taxa in the CCE (e.g., [83,96–98]). BRTs were fit using a tree complexity of 3, which allows each tree to model interactions among up to three variables; a bag fraction of 0.6, meaning 60% of the data was randomly sampled to train each tree, introducing stochasticity and reducing overfitting; and a learning rate that ensured that at least 1,000 trees were included in the final model for each species to balance model accuracy with generalization. These parameters have been found to lead to high performance without overfitting or sacrificing ecological realism that may occur when allowing complex variable interactions at higher tree complexities [99]. All BRTs were fit to presence-absence data with binomial response types using ‘dismo’ R package [100], which automatically selects the optimal number of trees during model fitting using 10-fold cross-validation. The relative importance of each predictor is determined based on the number of times a variable is selected for splitting, weighted by the squared improvement to the model as a result of each split, and averaged over all trees [99]. This metric is standardized to a relative contribution (%).

Model evaluation

For each species, we undertook multiple cross-validation analyses to assess spatiotemporal uncertainty in performance within our SDMs. First, we employed 10-fold cross-validation whereby, iteratively, data were randomly divided into 75–25% training-testing splits. Next, we used a leave-one-year-out (LOYO) cross-validation, where we withheld each year iteratively as testing data. Finally, we performed a spatial cross-validation where each biogeographic province (South, Central, & North) was iteratively withheld as testing data. For each cross-validation type (random, LOYO, & spatial), we calculated the mean (±1 SD) area under the receiver operating characteristic curve (AUC) and true skill statistic (TSS) across withheld iterations [101].

Model projections under climate change conditions

After training on the historical reanalysis product to ensure robust model calibration using high-quality, observationally-constrained data, SDMs for each species were projected using daily 10 km resolution outputs of the same variables as in model fitting for the 1980–2100 period from three dynamically downscaled earth-system models (ESM) under an Representative Concentration Pathway (RCP) 8.5 emissions scenario [60]. The three earth-system models were the Geophysical Fluid Dynamics Laboratory (GFDL) ESM2M, Institut Pierre Simon Laplace (IPSL) CM5A-MR and the Hadley Center HadGEM2-ES (HAD) from phase 5 of the coupled model intercomparison project (CMIP5). While RCP8.5 represents a severe and potentially unlikely future scenario (8.5 W/m² radiative forcing by 2100) that assumes continued growth in fossil fuel emissions ([102]; but see [103]), it allows greatest pattern recognition by maximizing signal (climate change) to noise (climate variability and model stochasticity), and is thus useful in spatial planning applications [28]. Additionally, the three ESMs represent a wide spread of CMIP5 ensemble members, and differences between ensemble members are often greater than scenario differences [104]. GFDL under RCP8.5, for example, has a relatively low warming rate that is typical of other models under the less extreme RCP4.5 (4.5 W/m2 radiative forcing by 2100) scenario (see [60] for more details).

Changes in suitable habitat overlap with boundaries of interest

Daily projections of HSI were used to assess historical (1985–2015), mid-century (2035–2065) and end-of-century (2070–2100) change across ecologically-important habitats and spatial management boundaries (Table 1). We calculated climatologies for 30-year periods of HSI scores; averaging across 30-year periods is common to ensure interannual to decadal climate variability is not inadvertently interpreted as long-term change [28]. For each area of interest, we calculated climatologies of the mean and standard deviation of HSI scores, by oceanographic season, for historical, mid-century, and end-of-century periods. We then calculated the standard deviation of HSI scores in each grid cell across the three downscaled ESMs to highlight uncertainty across the ESMs. Areas-of-interest included the NMSs and offshore WEAs on the U.S. west coast (see Fig 1) as well as two ecologically-important CCE-wide reference areas: the shelf (<200m bottom depth) and a combined shelf-outer slope area (Shelf/Slope) within 273 km of the coastline that represented the offshore extent covered by both ship-based and aerial surveys (Fig 1).

Next, we extracted time-series of species’ mean daily HSI scores for each model and subarea combination to assess intra-annual changes in HSI. Climatologies for each day of the year were calculated in each period, and the species-specific 75th percentile of HSI scores from the historical period were used to assess the proportion of the year in which highly suitable or ‘core’ habitat (i.e., > 75th percentile of climatological HSI scores from the historical period; [10,98]) is present. We defined areas containing highly suitable habitat in both the historical and future period(s) as potential refugia. Finally, basic additive time-series decomposition was used to identify long-term trends of HSI by removing the seasonal (within each year) and residual (random) components. The trend component highlights longer-term patterns in HSI, potentially reflecting sustained environmental changes. For each time-series, we calculated the Time of Emergence (ToE), which identifies the point at which a trend becomes distinguishable from historical variability [105], as the date at which the HSI’s trend component falls below or above one standard deviation from the mean HSI for the historical period.

Species-specific responses to environmental variables

Overall, SST, bottom depth, daylength, and province were the most important variables with contributions to each model being species-specific (Fig 2). SST was the most important variable (> 30%) for the auklet species, with both favoring temperatures less than ~15° C. While SST was less important for the other three species, they also favored cooler temperatures. Compared to the other species, SST was relatively unimportant (< 5%) in models for sooty shearwater and common murre, and partial response curves for SST were flatter (Fig C-G in S1 Text). Bottom depth was the most important variable (39.4%) for sooty shearwater and common murre (68.4%) models, with the latter strongly favoring shelf depths < ~400m, with habitat suitability elevated at the shelf break and shallower. Bottom depth was also relatively important in the black-footed albatross model, with depths >200 m favored by this species. Photoperiod had greater than 5% importance for all species. As expected, black-footed albatross and sooty shearwater were associated with longer photoperiods due to their seasonal occurrence within the CCE during spring-summer (Fig 1C-1D). Ocean season was marginally important in the sooty shearwater model (7.3%) and relatively unimportant (<5%) for all other species (Fig C-G in S1 Text). All species favored habitat in the Central and North provinces (Fig 2), with province being the most important variable for black-footed albatross (23.1%). Province was also relatively important for common murre (22.7%) and rhinoceros auklet (18.3%), and less so (<10%) for sooty shearwater and Cassin’s auklet. The standard deviation of depth was moderately influential for black-footed albatross (14.3%) and Cassin’s auklet (7.5%), with both favoring areas with elevated seafloor rugosity (i.e., higher values; Fig C-D in S1 Text). Upper ocean buoyancy frequency was moderately important (7.4%) only for the rhinoceros auklet model, with higher values favored, which indicate a more stable water column with higher stratification (Fig F in S1 Text). Isothermal layer depth, total kinetic energy, and wind stress were relatively unimportant (< 5%) in all models. A number of interactions between variables were important (Table B in S1 Text) with the strongest relationships between depth and either province (common murre) or daylength (Cassin’s auklet, sooty shearwater), indicating different physiographic associations (e.g., distance to breeding colonies, fronts, or other features) throughout the CCE and intra-annually.

Download:

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

Fig 2. Variable importance in each species SDM shown in radar plots for migratory (top left) and resident species (top right).

Response curves for the four most relatively important variables on average across species are shown below for each species along with their relative importance in parentheses. bv = upper ocean buoyancy frequency; TKE = total kinetic energy; ild = isothermal layer depth; sst = sea surface temperature; sst_sd = standard deviation of sea surface temperature; z = bathymetric depth; z_sd = standard deviation of depth.

https://doi.org/10.1371/journal.pclm.0000687.g002

Spatial extent and suitability of historical and future seabird habitat

Climatologies of ensemble mean HSI highlighted the general habitat preferences of each species in the historical period (Fig 3). Black-footed albatross HSI was low on the continental shelf, with higher HSI areas extending offshore throughout the extent of the shelf-outer slope study area (273 km coast-distance contour). Sooty shearwater, Cassin’s auklet, and rhinoceros auklet displayed similar habitat preferences to one another, with high HSI on the shelf and upper slope and HSI gradually declining further offshore. During the historical period, HSI was generally higher for these three species within the four central CCE NMS (0.5 ± 0.07) versus for black-footed albatross and common murre (0.41 ± 0.08). This same pattern was true of the WEAs, where HSI scores for sooty shearwater and the two auklet species were higher (0.52 ± 0.07) versus for black-footed albatross and common murre (0.39 ± 0.15). Highly suitable areas (i.e.,greater than the 75th percentile of historical HSI scores) were found within all four WEAs, although HSI for each species was slightly lower at the southernmost Morro Bay WEA for all species except sooty shearwater (Fig H-I in S1 Text). Both auklet species also exhibited lower HSI scores in the southern CCE, especially south of the Channel Islands. HSI scores were highest for the common murre on the continental shelf, sharply declined in areas with water depths > ~400 m and were relatively low in the southern CCE.

Download:

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

Fig 3. Mean habitat suitability index (HSI) scores in the Historical (1985-2015) period (top row) and the change in HSI (∆HSI; bottom row) in the end-of-century (2070-2100) for each species.

HSI are averaged across three downscaled earth-system models.

https://doi.org/10.1371/journal.pclm.0000687.g003

The strong relationship between suitability and SST in SDMs for the two auklet species led to the greatest overall HSI declines over the 21st century that extended throughout most of the study domain. These declines were slightly less pronounced in the northern CCE. Declines in HSI were also projected for black-footed albatross, common murre and sooty shearwater resulting in relatively few areas with no change or increases in HSI that overlap with areas of high historical suitability (i.e., refugia; Fig 3). Although HSI scores for much of the coastal and continental shelf common murre habitat were stable throughout the study period, notable declines in HSI were projected in areas with depths > ~400 m. Sooty shearwater was the only species with projected increases in ensemble mean HSI, primarily along the northern CCE shelf and upper slope. Some refugia areas (i.e., little-to-no HSI changes) were also present in the northern CCE in the end-of-century and overlapped with the three WEAs in this region (Fig I in S1 Text). The central CCE also contained end-of-century refugia for sooty shearwater, including in the four NMS (Fig I in S1 Text).

Overall habitat across the entire year was projected to decline across the CCE for all species by 2100 (Fig 3). Time-series revealed the different rates of HSI changes and their ToEs between species, ESMs, and CCE subareas (Fig 4). Generally, emergence was later for models projected with the GFDL ESM than for HAD or IPSL, due to slower rates of physical change in GFDL, although fluctuations caused some relatively early ToEs to be detected in some of the GFDL SDM projections (e.g., most locations for CAAU). ToEs were not found for any ESM or area for sooty shearwater (i.e., long-term change never emerged from natural variability). ToEs for black-footed albatross occurred relatively later (i.e.,after 2075) for black-footed albatross, and only for models projected with the HAD and IPSL ESMs, except for on the entire CCE shelf-slope, where HSI changes from GFDL emerged near the end of the century. Notably, ToEs were not found for black-footed albatross in NMSs despite an apparent declining HSI trend. Common murre ToEs for the shelf-slope study area, NMSs, and WEAs all occurred before 2050 despite a relatively lower rate of HSI decline in this species due to lower interannual variability. Relatively early (prior to 2050) ToEs were found for Cassin’s auklet, which also had the greatest rate of HSI declines in the study area, although later ToEs were found in the Coos Bay WEA and the shelf-slope where emergence was not found for the GFDL projections. For rhinoceros auklet, HSI changes emerged under all three ESM projections in NMSs, while emergent changes were not found consistently across the three ESMs in the Coos Bay (HAD only) and Brookings WEAs (IPSL and HAD).

Download:

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

Fig 4. Habitat suitability index (HSI) trend (seasonal and random components removed) time-series and Time of Emergence (ToE; vertical lines) for changes in HSI within CCE shelf-outer slope study area (within 273 km of the coast), four National Marine Sanctuaries (NMS), and four Wind Energy Areas (WEAs) across the the Earth-System Models (colors).

Both NMS & WEAs are arranged by decreasing latitude from top to bottom. ToE represents the first date where the HSI trend value is below one standard deviation of the mean HSI from the historical period (1985-2015).

https://doi.org/10.1371/journal.pclm.0000687.g004

Intra-annual changes in habitat suitability in future scenarios

Within oceanographic seasons there were several refugial areas with little-to-no change or even increases in HSI (Fig 5). Across the entire CCE, the strongest HSI declines occurred in the Upwelling and Oceanic seasons, while the Davidson Current season contained the widest range of changes, including habitat losses, refugia, and gains that represent potential habitat expansions. Changes in rhinoceros auklet and Cassin’s auklet HSI exemplified the variability in habitat changes across ocean seasons and provinces. In the Davidson Current season, highly suitable coastal and shelf areas in the central CCE remained suitable to the end of the century, while offshore areas were projected to decrease in HSI. HSI largely increased in the northern CCE (Fig 5). Sooty shearwater and black-footed albatross HSI was also projected to increase inshore and offshore, respectively, in the Davidson Current. In the Upwelling and Oceanic seasons, however, black-footed albatross HSI largely decreased, with some areas of little-to-no changes offshore in the central CCE (Fig 5). Meanwhile, common murre displayed the least seasonal variability. Shelf areas were projected to remain suitable for common murre near the end of the century, while potential foraging habitat seaward of the shelf break was projected to decrease in HSI. HSI increases were also apparent in the coastal southern CCE in the Oceanic and Upwelling seasons, although common murre were rarely sighted south of the Channel Islands NMS in the historical period (Table 1; Fig 1). We tested the effect of excluding the fixed-effect variables of season or province by fitting additional SDMs without them. Although the results from these models were largely similar (but slightly less skillful; Table A in S2 Text), the removal of these variables did lead to small increases (~5%) in the relative importance of SST for all species except sooty shearwater. Thus, projections of these models had lower HSI scores in the southern CCE, particularly in the future, and higher variability between ESMs (Fig A-E in S2 Text).

Download:

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

Fig 5. Mean percentage change in climatological habitat suitability index (HSI) scores [Future (2070-2100) - Historical (1985-2015)/ Historical * 100] of each species averaged across the three earth-system downscaled models for each oceanographic season.

Historical HSI scores are represented by the transparency of grid cells to highlight changes in areas with relatively high historical HSI scores versus those with low historical suitability (white).

https://doi.org/10.1371/journal.pclm.0000687.g005

Changes in the intra-annual timing of suitable habitat occurrence in the mid-century and end-of-century periods differed between migratory and resident species (Fig 6). HSI scores were highest for sooty shearwater and black-footed albatross in the Upwelling season, with the latter experiencing greater declines in the number of days with highly suitable habitat (i.e., > 75th percentile of historical HSI scores) in the examined areas in the mid-century and end-of-century periods (average declines of 43 and 68.14 out of 155 days, respectively). Of the areas examined, only the Humboldt, Brookings, and Coos Bay WEAs and the Cordell Bank NMS retained days where average conditions were highly suitable at the end-of-century for black-footed albatross (Fig 6). These declines in HSI were concentrated at the beginning and end of the Upwelling season, while HSI in the Davidson Current season was projected to remain relatively constant or even slightly increase. Common murre experienced declines in HSI throughout the year that were most pronounced in the Greater Farallones NMS and the Humboldt WEA, which respectively contained 138 and 155 days of highly suitable habitat in the historical period but had 14 and zero days at the end of century. Meanwhile, common murre HSI did not decline on the continental shelf, where average conditions were highly suitable year-round in each time period (Fig 6). Both auklet species were projected to experience HSI declines throughout the year that are concentrated in the Upwelling and Oceanic seasons, with only the Davidson Current season continuing to harbor high-HSI habitat at the end of the century for most areas except the Humboldt, Brookings, and Coos Bay WEAs, which retained some days of highly suitable habitat in the Upwelling season. The southernmost Morro Bay WEA and Chumash Heritage NMS contained highly suitable habitat for only common murre and sooty shearwater at the end of the century (Fig L-M in S1 Text).

Download:

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

Fig 6. Daily habitat suitability index [HSI] climatologies for depicting changes in the intra-annual timing and duration of suitable habitat occurrence within the CCE in the historical (1985-2015), mid-century (2035-2065), and end-of-century (2070-2100) periods.

Days with mean habitat suitability scores above the 75th percentile of each species’ scores during the historical period are shown with white points. Vertical dotted lines represent CCE oceanographic seasons and are indicated in white in the top left panel; DC = Davidson Current; U = Upwelling, O = Oceanic.

https://doi.org/10.1371/journal.pclm.0000687.g006

Model uncertainty and sensitivity evaluation

Overall, performance metrics (AUC, TSS) from our cross-validation analyses revealed good-to-excellent discrimination capacity by our SDMs with AUC values ranging from 0.82 for Cassin’s auklet to 0.95 for common murre (Table 2). Leave-one-year-out (LOYO) cross-validation scores were generally similar to those from random cross-validation. Meanwhile, the spatial cross-validation, in which each biogeographic province (North, Central, South; Fig 1A) was successively withheld as testing data, had slightly lower scores. Rhinoceros auklet had the most dramatic differences in performance among biogeographic provinces, with the lowest performance occurring when the North province was withheld (Fig J in S1 Text). LOYO cross-validation scores were fairly consistent across years for most species, and were lowest in years with relatively few testing data, and did not appear to notably lessen in environmentally anomalous years (e.g.,1998 El Nino; Fig J in S1 Text).

Download:

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

Table 2. Overview and evaluation of top models for each species. Mean (± SD) skill metrics [True Skill Statistic (TSS); area under the receiver operating characteristic curve (AUC)] from SDM evaluation via random cross validation with 75-25 training-testing splits (10-fold CV), leave-one-year-out (LOYO), and spatial (by biogeographic province) cross-validation.

https://doi.org/10.1371/journal.pclm.0000687.t002

There was little divergence in projected HSI scores between the three ESMs in the historical period (Fig N in S1 Text). In the end-of-century period, the standard deviation of HSI varied by species, season, and area of the CCE (Fig O in S1 Text). Generally, uncertainty was lower in the Upwelling season, which was the most data-rich season (Table A in S1 Text). Cassin’s auklet and rhinoceros auklet displayed similar patterns, with increased inter-model uncertainty north of the Greater Farallones NMS in the Oceanic season, and concentrated in the Central region in the Davidson season. Overall, inter-model differences were lower for black-footed albatross, sooty shearwater, and common murre. For common murre, uncertainty between ESMs was concentrated in the North around the shelf break and upper slope overlying the Oregon WEAs.

When the fixed effects of province and season were omitted, SDM performance slightly decreased (Table A in S2 Text). Dynamic variables such as temperature became slightly more important in the SDMs (Fig A in S2 Text), but climatologies of ensemble mean HSI and standard deviation between ESMs were similar to those from models with the fixed effect variables both in the historical and end-of-century periods (Fig B-E in S2 Text). However, some differences between these two model types emerged, with models without the fixed effect variables tending to produce outputs with lower HSI in the southern CCE, and with slightly higher temporal variability (Fig F-G in S2 Text).

Changing spatiotemporal distribution of seabird habitat

The species-specific environmental preferences we identified may provide insights into potential processes and threats that could affect seabird habitat distribution in the CCE under climate change. Important environmental variables matched previous efforts highlighting the influence of seasonality, temperature, and bathymetry as the primary physical drivers of these species’ occurrence [78,106–108]. Steeper HSI declines were projected for resident alcid species than for migratory species. This was largely driven by SST, as temperature had a more moderate association with HSI for black-footed albatross, sooty shearwater and common murre. These findings are consistent with expectations of broader environmental envelopes for large-ranged, migratory species versus resident species tied to narrower ranges of local conditions [109]. Common murre, a resident species, was a notable exception whose distribution was mostly associated with bottom depth and shelf habitat north of the Southern California Bight. As the southern CCE is currently the southern range limit for the three resident species, the relatively strong declines in HSI found in the southern portion of our study area match expectations for range contractions in the CCE based on northward movement of isotherms.

However, there are many moderating factors beyond the physical environment that may contribute to changes in seabird habitat distributions under future warming scenarios. Most important is food availability, which directly drives changes in seabird population dynamics, and their habitat distribution and abundance [42,80,110]. Seabirds do not occur in isolation; foraging assemblages and biotic interactions (both mutualistic and competitive) are critical to the ecology and marine habitat use of seabirds [111,112]. Furthermore, due to elements of conservation success, the recovery of whales and pinnipeds are altering food web interactions, (e.g., [113,114]), highlighting the potential utility of joint distribution models to better understand the overlap and associations of seabirds and other taxa [115–117]. In addition, we can expect that the degree to which shifting prey resources will affect seabirds varies according to their foraging ecology (e.g.,the ability to travel to offshore foraging areas), and that periods of anomalous prey availability will in turn affect the environmental preferences of species [111,118]. For example, anomalous ocean conditions caused by delayed upwelling in 2005 resulted in prey shortages (krill availability) and subsequent breeding failure of Cassin’s auklets in central California, resulting in colony abandonment and distributions towards the southern CCE [119]. Migrant sooty shearwaters and black-footed albatrosses are the least geographically constrained of our focal species, and thus the relatively small HSI declines we found for these species align with their broad habitat preferences. That is, migrant species can likely shift throughout the CCE based on food availability. Meanwhile, alcids have relatively high flight costs and small foraging ranges (<100s of km) that make them sensitive to horizontal shifts in prey [120], but may be able to switch between prey to partially ameliorate the negative effects of changing availability [121–123]. For example, the magnitude of projected seabird range contractions in Great Britain and Ireland have been associated with species’ foraging ecology, with smaller species that have high foraging costs and short foraging ranges the most negatively affected, as they are presumably most vulnerable to reductions in prey availability [55]. Overall, an integration of ecophysiological, behavioral, trophic, and historical factors (e.g., colony location) determine the magnitude of seabird distributional responses to changing climate and ecosystem conditions [110,124]. Therefore, the degree to which seabird habitat distributions are altered by climate change may be highly aligned with a suite of definable biological traits – particularly those related to specialization (e.g.,habitat breadth) and reproductive speed (e.g.,clutch size, generation time) – that may be used to quantify their climate vulnerability [125].

Timing of breeding and migration may also affect seabird species’ vulnerability to spatiotemporal predator-prey mismatches, which will likely increase if phenological changes in these traits are not commensurate with changes occurring at lower trophic levels [118,126]. For example, in the Benguela Current, no-fishing areas were established around seabird breeding colonies, but the available prey within these areas shifted elsewhere in response to changes in ocean temperature and salinity, altering the previously identified conservation benefits of these static management boundaries [127]. Our projections suggest that the timing of high-quality habitat conditions for seabirds is likely to change in the future, with a general narrowing of the annual window of high-HSI conditions for all of our focal species except sooty shearwater. This equates to a temporal habitat compression that may have wide-ranging potential effects on seabirds given the seasonal rhythms in many aspects of their ecology. For example, environmental conditions in winter that favor early upwelling has been shown to be an important indicator of breeding phenology (lay date) in common murres and Cassin’s auklets in the CCE [85]. For both of these species, we found that notably large areas with neutral or positive changes in HSI in the end-of-century period occurred north of Monterey Bay during the Davidson Current season, while areas to the south declined more strongly. This suggests winter warming and associated effects on seabirds will be concentrated in the southern CCE, as is true for overall annual warming [60], and that northward shifts into more favorable biogeographic regions for certain parts of the year could be an adaptive response for these taxa. Our projected intra-annual changes in HSI also suggest that earlier arrival times might benefit migrating shearwaters and albatrosses in the future. While a general trend of earlier summer arrivals has been observed for migrating seabirds [128], it is unknown if this response will keep pace with the velocity of climate and ecosystem change.

Implications of shifts in suitable habitat for marine spatial planning and management

The ambitious goal of placing 30% of the global marine environment in protected areas [129] has gained traction in recent years [130,131], and seabirds can play an important role in the delineation and monitoring of spatial management areas as indicators of ecosystem productivity [63,110,132]. By considering future changes in habitat suitability for seabird species within NMS and WEAs, managers can include considerations for habitat protections (e.g., special wildlife zones within NMS) in management plans, reducing other stressors (e.g., actions directed at reducing interactions with fisheries or offshore wind farms), or considering potential changes to boundaries in the long-term. The four NMS of the central CCE are located in important areas for the continued study and monitoring of climate change effects on seabirds [35,63], and we found they bound areas with both high HSI in the historical scenario and notable declines in projected future HSI, particularly in the Upwelling and Oceanic seasons. This includes the areas around the Farallon Islands and Cordell Bank where a major breeding colony for resident species (Southeast Farallon Island) and key foraging grounds for migrants occur [68,106,107]. However, some areas with high historical HSI remained suitable for at least part of the year until the end of the century, and thus may become important refugia. We projected these potential seasonal refugia to occur off the coast of Oregon and Washington where NMS are mostly absent. The Olympic Coast NMS, although largely outside our model domain and thus not included in SDM analyses, has a high prevalence of our focal species and extrapolation of our results to this area would likely identify this NMS as an important refuge.

Some seasonal habitat refugia also overlapped with offshore WEA boundaries (Fig H in S1 Text), suggesting that proposed WEAs are likely to displace habitat that would have otherwise remained relatively suitable in the long term (Fig K, M in S1 Text). For example, the four WEAs contained 5 and 7.1% of highly suitable habitat grid cells that remained in the study area in the end-of-century period for sooty shearwater and Cassin’s auklet, respectively. While the four proposed WEAs represent a small portion of the overall CCE seascape (1.3% of grid cells in the study area), additional research is needed to understand potential cumulative impacts. Avoidance of wind farms has been observed in many seabird species, and significant changes in abundance have been found at distances > 10 km from wind turbines [133,134], suggesting an even larger potential habitat displacement footprint than the boundaries of the WEAs used in our analyses.

While our study quantifies the availability of habitat in the vicinity of WEAs, there are additional factors that will mediate the impacts of wind farms on seabirds. [25] assessed the vulnerability across species in the CCE seabird community to collision and displacement impacts from offshore wind using a framework including population demographic and behavioral data. They found that alcids (e.g., auklets and murres) are among the most vulnerable to displacement from wind farms and relatively invulnerable to collisions, while larger-bodied, higher-flying species such as shearwaters are more vulnerable to collisions. Our analyses support the use of Cassin’s auklets habitat suitability for monitoring displacement impacts of wind farms, as their historical distributions overlap with all four WEAs the most (Fig I in S1 Text) and relatively strong impacts were projected at each (Fig M in S1 Text). Common murre have shown displacement effects elsewhere [135,136], and the CCE WEAs are within the seaward limits of their foraging range and may thus displace some foraging opportunities, particularly in the non-breeding season when foraging trips are less constrained to colony locations [137]. Similarly, the migrating, high-flying shearwaters could be used as indicators for testing the success of mitigation measures for collisions such as seasonal (or dynamic) operation scenarios (e.g., turning off or reducing the number or changing the spatial arrangement of turbines that are operating during migratory periods or when elevated wind conditions increase the likelihood of collisions [27,65]). Our SDMs could be coupled with wind models to generate short-term seasonal distribution forecasts that inform such measures. There is precedent for this type of strategy in the marine realm, such as moving fisheries closures that avoid bycatch [138], and dynamic management tools based on SDMs [139] or satellite data [140] that may show promise for meeting fishery and conservation goals.

Uncertainty, limitations, and directions for future research

Communicating uncertainty in SDMs is critical for interpreting potential climate change effects on species distributions [141]. The linkages between seabird habitat and static features such as the shelf break are well-established [78,80,106,142] and unlikely to be drastically altered in the future. However, there is considerable uncertainty in the future effect of dynamic variables, namely SST, on seabird habitat suitability. This uncertainty arises from the uncertainty inherent in ESM projections and in seabird distributions that respond to variability in ocean phenomena at multiple spatiotemporal scales [143,144] that cannot be entirely resolved by ocean models nor projected into the future. However, the declines in HSI that emerged from historical variability that we projected suggest that there may be fewer events with highly suitable conditions in the future, as habitat suitability often correlates with the upper limits of species’ local abundance [145]. We examined changes in seabird habitat suitability by oceanographic season to help account for some of the uncertainty driven by seabird responses to dynamic variables. The seasonal cycle is a critical element in understanding seabird vulnerability to stressors such as offshore wind [146], and will likely remain highly relevant to seabirds given the lack of overall changes in seabird phenological traits (e.g.,breeding times) observed since the mid-20th century [126]. However, there may be some alterations to seasonal patterns in the future, as projections suggest the CCE will experience intensification of upwelling in the spring and weaker summer upwelling [147] that may influence seabird phenology and the availability of prey.

Even with an extensive dataset of at-sea sightings spanning the CCE, there is also considerable uncertainty associated with the data underlying our SDMs (e.g.,the distribution of survey effort). Our cross-validation exercises demonstrated the importance of assessing SDM performance and uncertainty across both spatial and temporal dimensions. The spatial cross-validation analysis suggests that biogeographic province is an important variable to include so that responses may vary between subregions within large marine ecosystems, giving SDMs additional flexibility to represent the breadth of responses across species’ biogeography. Areas with relatively poor cross-validation performance (e.g., northern CCE for rhinoceros auklet) were associated with relative data scarcity, highlighting the potential use of model cross-validation outputs and other model uncertainty outputs (e.g.,inter-model disagreement) to identify areas to focus monitoring and additional data collection. This also demonstrates the importance of capturing as much of species’ ranges as possible when fitting SDMs, as certain areas (e.g.,those with important breeding colonies) may have outsized effects on model performance.

While our projections advance our understanding of potential climate change effects on seabird habitat in the CCE, there are a number of applications and paths for future modeling efforts that may expand upon our study. For example, habitat suitability declines from seabird SDMs in other regions are spatially associated with climate-driven population declines [52], suggesting our models can provide an avenue to explore the effects of habitat quality on population variability in future studies. Abundance/density is another important dimension of seabird distributions to understand their vulnerability to climate change and the risk of interactions with offshore wind turbines. A recent study modeled the probability of aggregation occurrence [108], a quantity which may be useful to project potential changes in interactions/risk from offshore wind development [148]. SDMs based on presence-absence data should be compared with density models, especially given uncertainties around future population changes that may affect seabird densities and the generally lower skill provided by abundance/density models, especially for taxa with patchy and high-density aggregations that may require relatively fine-scale data for realistic SDM predictions [149]. Biotic interactions (e.g., competition) between seabirds are also not included in our models, although are likely important in shaping seabird distributions given their co-occurrence and dietary overlap [111]. Joint SDMs that attempt to statistically account for species inter-dependencies may therefore be a useful tool to apply for future seabird SDMs [150].

Given the highly dynamic nature of seabird habitat use and foraging opportunities, similar projections of seabird prey habitat may also help identify additional requirements and refine projections for seabirds. Integration of forage and seabird projections would also allow for tests of whether changes in HSI between these groups are similar in magnitude or direction, and may identify associations between predator and prey habitats and refugia that would aid in future marine spatial planning efforts. Similarly, there is a growing amount of electronic tagging and behavioral data whose integration with SDMs could provide insights into foraging and migration patterns to better understand the connectivity and usage of areas we identified as highly suitable [151], or understand behavior-mediated risk of offshore wind interactions [65,136]. While tagging datasets have typically been used to evaluate SDMs or create presence-only SDMs [152,153], new methods for integrating different data types in SDMs (e.g., [154,155]) hold promise for bridging the divide between at-sea survey and tagging perspectives in seabird ecology with predictive models. Our SDMs can also be updated as new information is gathered on seabird distributions, or re-projected as ocean model products are iteratively refined or additional scenarios become available [156]. For example, the effect of large floating turbines planned for operation have been modeled in ROMS for the CCE and these models suggest that turbines will alter wind fields and oceanographic parameters in their vicinity [157,158]. These model runs may be used in future work to project these SDMs to quantify the potential ecological effects of built-out wind farms. There are many conceivable avenues for future research, the SDMs and derived outputs presented herein provide a first step in quantifying potential futures for seabird habitat across the CCE and the uncertainty associated with such projections. Our results illustrate how climate change can alter the benefits of conservation boundaries and trade-offs with proposed development, and these results could also be applied to marine spatial planning efforts for other ocean-use sectors, including offshore aquaculture, seabed mining, offshore oil and gas, wave or floating solar energy, marine carbon-dioxide removal, and commercial shipping activities.