2.1 Study region

The CCLME spans the western North American coast between 22–48^°N, extending as far west as 132^°W (Fig 1). For this study, we define two cross-shore bands within the CCLME, referred to as “nearshore” and “offshore,” defined to be within 0–75 km and 75–300 km of land, respectively. We further divide each of these bands into four alongshore zones, with delineation from north to south at 40^°N, 35^°N, and 30^°N. We present analyses of the CCLME, its nearshore and offshore bands, as well as their eight zones, as area-weighted, regional averages. We also define boreal winter as January-February-March (JFM), spring as April-May-June (AMJ), summer as July-August-September (JAS), and fall as October-November-December (OND).

2.2 Global ocean hindcast with biogeochemistry

We evaluate output from a 1/4 degree global ocean biogeochemical hindcast simulation (FREEBIORYS2V4; https://doi.org/10.48670/moi-00019) produced by Mercator-Ocean and made available through the Copernicus Marine Environmental Monitoring Service (CMEMS). FREEBIORYS2V4 uses the PISCES-v2 (Pelagic Interactions Scheme for Carbon and Ecosystem Studies) biogeochemical ocean model [43], which is initialized with global distributions of nutrients and dissolved oxygen from the World Ocean Atlas 2013 climatology [44] and forced offline by daily mean fields of atmospheric conditions from the European Center for Medium-Range Weather Forecasts (ECMWF) ERA-Interim atmospheric reanalysis [45] and physical ocean and sea ice conditions obtained from the Mercator-Ocean 1/4 degree FREEGLORYS2V4 numerical simulation (see extensive descriptions in [46] and [47]). We use bicubic interpolation to map the FREEGLORYS2V4 output onto the FREEBIORYS2V4 grid, consistent with the scheme used by Mercator-Ocean to produce the FREEBIORYS2V4 simulation. Taken together, the FREEGLORYS2V4 and FREEBIORYS2V4 output, hereafter collectively referred to as GLORYS-BGC, consist of daily or monthly mean physical and biogeochemical ocean variables from 1993-2019 at 1/4 degree (∼25 km) horizontal resolution and 75 vertical levels [48]. We limit our study of extremes to years 1996–2019 for which model salinity values are realistic for the region and exhibit no apparent drift (not shown). To standardize the temporal resolution across years, all leap day entries (February 29) were excluded from the dataset.

2.3 Detection of extreme events

In this study, we define extremes of temperature, chlorophyll, and hypoxic layer depth (HLD) as a MHW, low chlorophyll extreme (LCX), and shallow hypoxia extreme (SHX), respectively (Table 1). Prior to event detection, we define a bulk chlorophyll concentration by vertically-integrating the upper 100 m chlorophyll concentration (Chl; mg m⁻²); where bottom depth is shallower than 100 m, we integrate over the entire water column. Dissolved oxygen concentration is linearly interpolated in the vertical prior to defining the HLD (m) as the shallowest 2 mg L⁻¹ oxygen isopleth (2 mg L⁻¹ 62.5 mmol m⁻³ using a molar weight of 31.998 g mol⁻¹). As such, both chlorophyll and oxygen-related extremes are defined using a single value while still representing a water column extreme with relevance to ecosystem dynamics. While we assess MHWs throughout the water column, we have chosen to present results largely focused on surface-defined MHWs (SMHWs) across the CCLME, and, given their implications for benthic and demersal species, bottom MHWs occurring along the seafloor of the nearshore region (see [34]). In the latter case, we isolate bottom temperatures only at locations where the ocean bottom is shallower than 1000 m. Trends in temperature, Chl, and HLD were assessed but not removed prior to event detection, given the relatively short study period and consistent with recent studies of extremes (e.g., [49]); a sensitivity analysis on detrended data shows no significant change in the results presented here (not shown).

Download:

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

Table 1. Percentile (P) thresholds and naming conventions used to define extreme events.

https://doi.org/10.1371/journal.pclm.0000638.t001

From the model grid cell level to the regional level, single and compound (double and triple) extremes are detected over a given time period using a seasonally-varying relative threshold approach with an 11-day window size centered on the day of the year and extended over the 24 year record (as in [50]). In the case of a MHW, we define an extreme temperature event as any ≥ 5 day period exceeding the 90th percentile (P90; Table 1; [51]). For LCX and SHX, any ≥ 5 day period below the 10th percentile (P10) is considered an extreme event (Table 1; [49]). In the latter case, a SHX indicates a persistent, extreme shoaling of the HLD (similar to the THREEs metric used by [52]). If any consecutive extreme events occur ≤ 3 days apart, the events are combined and counted as a single event. Any time periods where two or more events co-occur in the water column for at least 5 days are deemed compound extreme events (e.g., Fig 2 illustrates a single day of a triple extreme as defined in this study).

Download:

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

Fig 2. Example of a triple extreme in the water column.

Snapshot of the upper 370 m water column on May 4, 2005 during a triple extreme event identified off the southwest coast of Oregon (43.25^°N, 125^°W) that began on May 4 and ended on May 8. The solid contours represent the daily (May 4) climatological mean vertical profiles of temperature (orange) and chlorophyll (green) and dissolved oxygen (blue) concentration. The corresponding dashed contours indicate the instantaneous or extreme profile of these properties when a concurrent SMHW (P90 = 13.1^°C), LCX (P10 = 31.7 mg m⁻²), and SHX (P10 = 159.6 m) was identified on May 4, 2005. The climatological (solid green bracket) and instantaneous (dashed green bracket) vertically-integrated (0–100 m; light green shading) chlorophyll concentration values are annotated. Waters below the HLD (blue stars) are hypoxic (light blue shading); the caret along the x-axis of dissolved oxygen marks the hypoxia threshold of 62.5 mmol m⁻³.

https://doi.org/10.1371/journal.pclm.0000638.g002

It is common to define hypoxia using a threshold of 2 mg L⁻¹, a cutoff conventionally used in fisheries management decisions, despite some species having lower or higher tolerances (see [53] and references therein). Our choice to define oxygen-related events based on extreme vertical displacements of the HLD is robust for a variety of hypoxic thresholds (1.5 and 4 mg L⁻¹) and less sensitive to model biases in absolute oxygen concentration (not shown). In shallower water columns along the continental shelf, the absence of an HLD is not uncommon, as the presence of hypoxia can vary by location and season (e.g., see [24]), with approximately 30% of the GLORYS-BGC nearshore domain experiencing some form of this. For nearshore assessments, undefined HLDs are treated as NaNs prior to calculating the spatially-weighted median, limiting contributions to areas where hypoxia is present and helping minimize seasonal biases. For pointwise calculations used to describe SHXs–including both the climatology and percentile threshold–we instead replace an undefined HLD with the ocean bottom depth at that location, effectively representing a fully oxygenated water column (“no hypoxia”). Such instances are not considered valid extreme days and are excluded from SHX event identification. A sensitivity test in which undefined HLDs are retained as NaNs (not shown) yields generally shallower threshold values, effectively imposing a stricter event criterion and resulting in modest changes in pointwise SHX frequency, intensity, and duration along the shelf.

2.4 Characterization of extreme events

Extreme events are characterized by three metrics: duration, intensity, and frequency. Duration is defined as the number of days an event remains above or below its respective percentile threshold, and intensity as the average anomaly over that period. For compound events, duration is the length of time over which the univariate extremes occur, and intensity is the product of their standardized anomalies or z-scores (Z; as in [54] and [55]; see Sect [2.8]2.8 for the computation of the geometric z-score for chlorophyll). Our analysis assumes that both positive and negative anomalies reflect departures from typical conditions that contribute to ecological stress, with the multiplication of z-scores ensuring positive compound intensities when extremes align in stress-associated directions.

For example, for a given location, the intensity (Ψ) of compound MHW-SHXs over the study period is expressed as

which is then standardized by subtracting the temporal mean () and dividing by the temporal standard deviation ():

This final standardization ensures comparability across compound event types by transforming intensities into a common z-score scale, enabling direct comparison of event anomalies relative to typical behavior. In addition to the product of z-scores method, computing compound event intensities (Ψ) using the Euclidean norm approach (as in [56]) yields consistent results, with key takeaways remaining unchanged across both methods (not shown).

The frequency of a single or compound event is the number of times it occurs within a given temporal range. For example, the triple extreme event present on May 4, 2005 at 45^°N and 235^°E off Oregon (Fig 2) included a surface MHW (SMHW) with an SST value (13.4^°C) exceeding the 90th percentile (13.1^° C) of all SST values simulated between April 29 and May 9 (11-day window) across all 24 years (264 total seasonally-independent SST values). This particular SMHW is characterized by a mean intensity of 1.2^°C (1.35σ) and a duration of seven days (May 2–May 8), occurring in a location that has seen 38 SMHWs over the study period (i.e., a frequency of 1.58 events per year; not shown).

2.5 Observations

We use daily SST values from the NOAA Optimum Interpolation Sea Surface Temperature (OISSTv2) blended satellite product available on a 0.25 degree grid [57,58]. Satellite estimates of surface chlorophyll concentration are provided by the European Space Agency Climate Change Initiative (ESA-CCI) as a blended 4 km product (version 6; [59]). Prior to analysis, we regrid the satellite-derived SST and chlorophyll products onto the standard 0.25 degree GLORYS-BGC grid using inverse distance weighting based on the four nearest neighbors. We then mask the model output according to the temporal availability of satellite retrievals.

We employ in situ observations collected seasonally through the California Cooperative Oceanic Fisheries Investigations (CalCOFI; [40]) sampling program along lines 76.7, 80, 83.3, 86.7, 90, and 93.3 (see map in S2 Fig). In accordance with CalCOFI quality control procedures, data flagged as suspect or otherwise unreliable were excluded from analysis, though the removal of individual oxygen samples does not result in undefined HLDs at that location, as sufficient remaining samples in the vertical profile allow for the calculation of the HLD. We subsample GLORYS-BGC in time (day) and space (latitude, longitude, depth) using nearest neighbor interpolation to match CalCOFI observations between January 1996 and December 2019. We quantify the modeled and observed bulk chlorophyll concentration by vertically integrating over the upper 100 m of available samples (Chl); we also vertically interpolate oxygen concentration prior to defining the HLD. Then, we spatially average both modeled and observed CalCOFI data into nearshore and offshore cross-shore bands for direct comparison and compute seasonal averages.

We also utilize in situ observations collected along the repeat Trinidad Head (TH; [60]) and Newport Hydrographic (NH; [61]) lines located at 41.05^°N and 44.65^°N, respectively, and within 75 km of the coast. We construct a composite of stations TH05 and TH06 for temperature, dissolved oxygen, and chlorophyll concentration over sampling years 2007–2019. At Newport, we isolate temperature (1997–2019) and dissolved oxygen (1998–2019) sampled at stations NH10, NH20, and NH25. For both observing programs, we compute monthly averages and linearly interpolate observations in the vertical. We then use bicubic interpolation to map monthly GLORYS-BGC variables onto the observed locations.

2.6 Hindcast evaluation

GLORYS-BGC has been previously validated against observational products [46] and utilized in studies of MHWs [62], low oxygen extremes [63], MHW-LCX compound extremes [55], ENSO-driven variability in Eastern Boundary Upwelling Systems [64], biogeochemistry along Lagrangian pathways [47,65], and as environmental context to observational sampling expeditions [66]. To assess the ability of the GLORYS-BGC hindcast to represent realistic extreme relationships in the CCLME, we leverage a combination of satellite products and ship-based observations of surface and subsurface temperature and chlorophyll concentration, as well as ship-derived HLD in this region over the last two decades.

We use satellite estimates of surface ocean properties to assess the model’s ability to reproduce historical extreme events, defining a probability of detection (POD) as the percentage of observed extremes correctly simulated by the model:

where h is the number of observed extreme events that the model correctly simulates (“hits”) and m is the number of observed extreme events that the model fails to simulate (“misses”). In the case of CCLME SMHWs, GLORYS-BGC has a POD of 82% when compared with OISSTv2 over the 1996–2019 period (S1 Fig). Geographically, we find the highest POD values in the CCLME north of 30^°N as well as the nearshore region of the central Baja California Peninsula (hereafter referred to as Baja; S1 Fig). The lowest POD values are found in the Southern California Bight, where values drop below 40% (S1 Fig), suggesting that GLORYS-BGC has some difficulty simulating MHW dynamics in this highly variable region of the CCS.

While OISSTv2 is a blended observational product, including cloud-penetrable microwave SST retrievals, sparse availability of visible light-reliant ESA-CCI chlorophyll retrievals precludes a robust POD analysis for low surface chlorophyll extremes (not shown). In the portions of the CCLME where at least 50% of a given season is available, we can partially evaluate regional seasonal mean biases. Indeed, we find that the model tends to overestimate seasonal surface chlorophyll concentration in the nearshore north of ∼28^°N (< 0.7σ) and underestimate it south of ∼28^°N (> -0.7σ) throughout the year (not shown). However, we note that event detection is not influenced by seasonal mean biases when using a seasonally-varying relative threshold approach.

Using in situ observations from the CalCOFI program, we evaluate the GLORYS-BGC hindcast’s ability to reproduce observed variance and covariance among extreme variables. In the nearshore CalCOFI region, modeled relationships between environmental stressors align well with those observed, while offshore agreement is somewhat weaker, particularly for hypoxic layer depth (HLD) relationships (S2 Fig). Across both regions, warm surface temperature anomalies are generally associated with low Chl anomalies, and vice versa. This inverse SST–Chl relationship is the strongest among the three variables and is most accurately captured by the model, as indicated by higher correlation coefficients and narrow, elongated confidence ellipses. By contrast, relationships involving HLD are weaker and more variable. Deep HLD anomalies show only modest correlation with warm SST and low Chl anomalies, as evidenced by lower correlation coefficients and broader, more circular ellipses—likely reflecting the more complex biogeochemical and physical drivers of dissolved oxygen variability (see [67]).

We further evaluate the correlation between observed and modeled anomalies (S3 Fig), with attention to variability associated with ENSO. During strong ENSO events, the model reasonably captures the expected anomaly signatures for warm and cold phases. For instance, during the 1997–1998 El Ni no, GLORYS-BGC successfully reproduces the anomalously warm nearshore SSTs, low Chl, and deep HLDs observed in CalCOFI records, in line with our broader findings on compound extremes presented in subsequent sections (S3 Fig; [68]). These results provide confidence in the ability of GLORYS-BGC to capture the relevant physical–biogeochemical relationships that underpin compound extremes in this region.

In other regions of the CCS, long-term observations from the Newport Hydrographic and Trinidad Head lines provide sufficient coverage to evaluate whether the model can meaningfully represent extremes. Among the various published model evaluation metrics commonly employed for general assessment (e.g., [69]), the anomaly correlation coefficient (ACC) is most directly relevant for assessing the model’s ability to capture the timing and variability of anomalies, which are critical for identifying percentile-based extreme events. We compute the ACC from monthly anomaly time series of 0–200 m vertically integrated temperature and dissolved oxygen from both lines, as well as 0–100 m vertically integrated chlorophyll concentration from the Trinidad Head line. Simulated upper-ocean temperature, dissolved oxygen, and chlorophyll concentration at Trinidad Head show good agreement with observations, with ACC values of 0.97, 0.72, and 0.64, respectively. Along the Newport line, station NH10, NH20, and NH25 temperature correlations remain relatively high (ACC = 0.74, 0.75, 0.77, respectively), while oxygen exhibits more modest anomaly correlations (ACC = 0.32, 0.32, and 0.39, respectively). These results are consistent with expectations for model–observation comparisons in highly dynamic, sparsely sampled coastal systems and, when considered in light of inherent observational limitations, indicate that GLORYS-BGC demonstrates sufficient skill to serve as a reliable tool for investigating ocean extremes.

2.7 Climate and upwelling indices

Throughout this study, we discuss multi-stressor extremes in the context of local and remote forcing on seasonal-to-decadal timescales, employing various climate indices describing ENSO, PDO, NPGO, and coastal upwelling. ENSO is described by the Ni no 3.4 index, which is defined as the area-averaged sea surface temperature (SST) anomaly over 5^°S–5^°N, 170–120^°W (see https://www.cpc.ncep.noaa.gov/data/indices/). The PDO index [70] is defined as the leading Principal Component of North Pacific SST anomalies (north of 20^° N), obtained by removing both the climatological annual cycle and the global average SST. The NPGO index is defined as the second Principal Component of monthly sea surface height (SSH) anomalies [37] in the Northeast Pacific region (180–110^° W; 25–62^°N). The Coastal Upwelling Transport Index (CUTI; [71]) is an estimate of the total vertical transport through the base of the mixed layer and is derived from SSH, mixed layer depth (MLD), and surface wind stress (τ); here, MLD is computed using a density threshold criterion; specifically, the MLD is the depth at which the density of the water column differs from the density by a value of 0.125 kg m⁻³, suitable for analyses focused on monthly and interannual timescales (e.g., [72]). All indices are computed using the common climatological reference period (January 1996–December 2019) and calculated as detailed above using monthly SST, SSH, MLD, and τ from GLORYS-BGC; model Ni no 3.4, PDO, and NPGO indices compare well with observed indices, with correlation coefficients exceeding 95% (not shown).

2.8 Statistical approach

Mean values reported for a given region are computed as area-weighted averages. Anomalies are expressed as the difference between the period of interest and a given reference climatology. To facilitate comparison across variables, anomalies are standardized by dividing by the reference period standard deviation, yielding z-scores–anomalies expressed in units of standard deviation (σ). Given that chlorophyll and net primary production are log-normally distributed, we compute statistics in space and time using the geometric mean (), standard deviation (), and standardized anomaly (z-score; Z_g) as

respectively, where n corresponds to the total number of samples (i) of a property (x) in a given time or space bin.

As in Le Grix et al. [49], the likelihood multiplication factor (LMF) is computed as the ratio of the observed joint frequency of all N ≥ 2 individual extremes occurring simultaneously (i.e., the compound event) to the product of their marginal frequencies (f(E_i); [73]):

For example, we would theoretically expect independent MHW-SHX events to occur 1% of the time (10% × 10%), by definition. However, effective percentile thresholds can be reduced after applying additional criteria such as a 5-day minimum duration, resulting in lower marginal frequencies due to the exclusion of shorter (1–4 day) events. As such, all frequencies used in the LMF calculation reflect the filtered definition of extremes. An LMF greater than 1 indicates that compound events occur more frequently than would be expected by chance (i.e., positive statistical dependence), suggesting shared drivers or physically coupled processes. Conversely, an LMF less than 1 suggests negative dependence or mutual exclusion, where the events co-occur less often than expected under independence, potentially reflecting competing or uncoupled physical mechanisms.

3.1 Characteristics of single and compound extremes and their spatial variability

We find that single and compound extremes in SST, Chl, and HLD exhibit distinct spatial patterns in mean frequency, intensity, and duration across the CCLME. SMHWs are frequently detected throughout the LME, with much of the region experiencing a consistent number of events exhibiting similar mean intensity and duration (Fig 3). In contrast, the highest occurrence of LCXs and SHXs is generally concentrated near the coast. LCXs are particularly frequent, with the most events found north of San Francisco Bay. Nearshore LCXs tend to be relatively weak and short-lived, except in zone 1 off Baja where durations can exceed 200 days (Figs 3 and 4). This pattern largely reflects the background mean and temporal variability of chlorophyll concentration in this region: where mean Chl is high and variability is frequent, more short-duration extremes are identified, and vice versa (Fig 3). In the case of SHXs, we find the most intense (up to 5σ; Fig 4) but relatively short-lived (<50 days; Fig 4) events throughout the nearshore (Figs 3 and 4). For example, off the Pacific Northwest coast (PNW; zone 4 latitudes), we find the strongest but shortest-lived SHXs along the shelf (Figs 3 and 4), consistent with Damien et al. [74] as a ‘hypoxia hotspot.’ In contrast, the Southern California Bight (SCB; zone 2 latitudes) stands out as a nearshore exception, where on average, relatively long-lasting (>200 days) but weak (<1.5σ) and infrequent SHXs occur (Figs 3 and 4). The longest-lasting events–whether single or compound–tend to occur far offshore (>300 km from the coastline), in open ocean areas characterized by low productivity and species diversity relative to the continental shelf and slope (Figs 3 and 5).

Download:

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

Fig 3. Maps of single extreme characteristics over the study period (1996–2019).

Frequency (top row), mean intensity (middle row; mean z-score (Z), expressed in sigma units), and mean duration (bottom row; days) of SMHWs (left column), LCXs (middle column), and SHXs (right column). The black contour marks the boundary of the CCLME.

https://doi.org/10.1371/journal.pclm.0000638.g003

Download:

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

Fig 4. Intensity–duration–frequency relationships for single extreme events across zones.

Single extremes of SMHWs (browns; left), LCXs (greens; middle), and SHXs (purples; right) are binned by duration (x-axis; days), intensity (y-axis; z-score (Z), expressed in sigma units), and frequency (color; average annual number of grid-cell extreme days) from 1996 to 2019. Markers (open circles) are overlain to represent the median of the distribution. Each extreme displays the nearshore zones 1–4 on the right and offshore zones 5–8 on the left, just as they appear in Fig 1 (zone numbers increase from south to north). Note, colorbar ranges differ between extremes.

https://doi.org/10.1371/journal.pclm.0000638.g004

Download:

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

Fig 5. Maps of compound extreme characteristics over the study period (1996-2019).

Frequency (top row), mean intensity (middle row; , expressed in sigma units), and mean duration (bottom row) of compound SMHW-LCX (first column), LCX-SHX (second column), SMHW-SHX (third column), and SMHW-LCX-SHX (fourth column). The black contour marks the boundary of the CCLME.

https://doi.org/10.1371/journal.pclm.0000638.g005

The compound events with the largest spatial footprint in the CCLME from highest to lowest are SMHW-LCXs, SMHW-SHXs, LCX-SHXs, and SMHW-LCX-SHXs (Fig 5), with triple extremes having only occurred over less than 23% of the LME between 1996 and 2019. The most intense (in a standardized sense) and longest-lasting SMHW-LCXs tend to occur far offshore (>300 km from the coast), though at a lower mean frequency than nearshore events (Fig 5). In contrast, the most recurrent SMHW-LCXs are found along the coast, with highest frequency values (>1 event per year) stretching across the majority of the LME coastline, from ∼27–45^°N (Fig 5). Coastal SMHW-LCXs occurring south of Point Conception (∼35^°N; Fig 5; zones 1–2 and 5–6; Fig 6) generally last longer than those to its north (zones 3–4 and 7–8; Fig 6). Notably, a large portion of the Baja coastline (within ∼300 km) spent the highest number of days (up to 620 days or ∼7% of total days) over the 24 year study period in SMHW-LCX status (see Sect [3.2]3.2).

Download:

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

Fig 6. Intensity–duration–frequency relationships for compound extreme events across zones.

Compound extremes of SMHW-LCXs (oranges; left), LCX-SHXs (greens; second from left), SMHW-SHXs (pinks; second from right), and SMHW-LCX-SHXs (grays; right) are binned by duration (x-axis; days), intensity (y-axis; , expressed in sigma units), and frequency (color; average annual number of grid-cell extreme days) from 1996 to 2019. Markers (open circles) are overlain to represent the median of the distribution. Each extreme displays the nearshore zones 1–4 on the right and offshore zones 5–8 on the left, just as they appear in Fig 1 (zone numbers increase from south to north). Note, colorbar ranges differ between extremes.

https://doi.org/10.1371/journal.pclm.0000638.g006

Unlike SMHW-LCXs, LCX-SHXs do not occur everywhere in the CCLME, with the least number of events in zone 1 off Baja (Figs 5 and 6). Where LCX-SHXs do occur in the nearshore and offshore bands, they are relatively short-lived (Figs 5 and 6). For example, off the coast of northern California in zone 7, we find elevated occurrences of this particular compound biogeochemical extreme, where the median event exhibits a 2.3σ standardized intensity and 14-day duration (Figs 5 and 6). While SMHW-SHXs are generally more prevalent across the CCLME than LCX-SHXs, large portions of the nearshore band off California and Baja do not experience any SMHW-SHXs (Figs 5 and 6). In the coastal region, the rarity of SMHW-SHX events is consistent with upwelling dynamics that couple low oxygen to low temperatures. At these nearshore locations, SMHW-SHXs that do occur are relatively short-lived and likely a result of independent processes, which we discuss in the following section (Sect [3.2]3.2). Outside of the nearshore, SMHW-SHX intensity appears generally correlated with duration, where relatively strong events are also long-lived, and vice versa (Figs 5 and 6). The opposite relationship generally holds true in the case of triple extremes: while SMHW-LCX-SHXs across the CCLME are generally strong (>2σ), they are also infrequent (<0.3 per year or <7 events total, approximately) and short-lived (Fig 5). Off the coast of northern California in zone 7, median triple event (standardized) intensity and duration is 3.6σ and 19 days, respectively (Fig 6). Triple extremes are a relatively rare occurrence in general, as there are more triple extreme-free areas in the CCLME than not, particularly in the nearshore (Figs 5 and 6).

3.2 The most common compound extremes in the CCLME

Based on total days over the study period, we find that SMHW-LCXs are the most commonly occurring compound extreme conditions in the CCLME (Fig 7). The nearshore in particular is most often experiencing SMHW-LCX conditions (Fig 7), with the total number of extreme days (1996–2019) generally increasing southward along the coast, from ∼100 days off Washington and northern Oregon to more than 500 days off portions of Baja (S4 Fig). The modeled frequency of SMHW-LCXs (Fig 5) across the region reflects largely positive LMF values (see Sect [2.8]2.8), with the highest values occurring near the coast (S5 Fig). The positive LMF values indicate a positive correlation between SMHWs and LCXs, such that SMHW-LCXs occur more frequently than by chance. For example, SMHW-LCXs are more than five times more likely to occur along the California and Baja coasts than expected, while LCX-SHXs and SMHW-SHXs tend to occur at a reduced likelihood (LMF<1) in the nearshore (S5 Fig), suggesting a negative correlation between their component events (i.e., when an LCX is present in the water column, it is less likely that an SHX is also present). In the case of nearshore LCX-SHXs, the processes that drive a SHX (e.g., coastal upwelling) would also tend to stimulate phytoplankton growth (e.g., via nutrient input), exemplifying how the likelihood of one event is coupled to the likelihood of the other. As for SMHW-SHXs, we know that warm SST anomalies in the nearshore are generally associated with deeper HLDs (S2 Fig; Sect [2.6]2.6) and that the relatively shallow water columns along the continental shelf are generally well-oxygenated, through a combination of physical (e.g., air-sea interaction, vertical mixing) and biological processes (e.g., photosynthesis; [75]). Coastal hypoxia can result from a sufficiently strong upwelling event that transports cold, low-oxygen, and nutrient-rich waters onto the shelf, a process that is generally more likely to be associated with a marine cold spell (MCS) than a MHW, and a high chlorophyll extreme (HCX) than a LCX (Table 1).

Download:

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

Fig 7. The most common compound extremes over the study period (1996–2019).

Maps of the most common compound extreme, defined as the highest number of extreme event days, from a surface (left) and bottom (right) ocean perspective (see respective keys): MHW-LCX (brown), LCX-SHX (purple), MHW-SHX (teal). Light gray contours outline the CCLME, offshore, and nearshore boundaries (see Fig 1 for reference). The maps on the right are regions isolated from the black open squares indicated on the left, delineated by zone 1–4 latitudes; here, only locations found within the CCLME where the ocean bottom is 1000 m or shallower are shown.

https://doi.org/10.1371/journal.pclm.0000638.g007

From a bottom ocean perspective, the continental shelf and slope are most commonly subject to one of two compound extremes, depending on the underlying bathymetry (Fig 7). In general, BMHW-LCXs are most common in relatively shallow water columns, such as along the continental shelf or island chains, while BMHW-SHXs are most common at locations characterized by relatively deeper water columns along the continental slope. For most of the coastline, this translates to a general cross-shore transition across the shelf break from BMHW-LCXs to BMHW-SHXs as ‘most common’ from east to west, respectively (Fig 7). This pattern likely reflects differences in hypoxia persistence, where shallower shelf waters typically experience seasonal hypoxia, while deeper slope waters are more likely to be hypoxic for the majority—or entirety—of the year, enabling a more consistent and extensive HLD. The SCB stands out as one exception, characterized by a relatively wide continental shelf with complex topography including multiple islands, with localized and variable patterns of upwelling and stratification. For the majority of the SCB shelf, we find that BMHWs are more likely to co-occur with SHXs (LMF>1) and less likely to co-occur with LCXs (LMF<1; S5 Fig). Here, BMHWs can be intense and persistent, driven by upwelling dynamics and depth-dependent thermal stratification [34]. Stratified conditions inhibit vertical mixing and oxygen replenishment and can exacerbate hypoxia [40,76]. While local species experience both BMHW-SHX and BMHW-LCX conditions here, the difference in total days is approximately two months (S4 Fig).

3.3 Spatiotemporal variability in nearshore and offshore ocean properties and their extremes

Seasonal variability of nearshore temperature, Chl, and HLD (Fig 8e–8h) is largely a reflection of the known seasonal variability in physical and biogeochemical processes such as coastal upwelling (e.g., [71]) and organic matter production and respiration [77]. Starting in late spring, surface heating and plentiful light combine with upwelling of cold, nutrient-rich, but oxygen-poor water to drive a productive upper ocean, with shallow HLDs and cold bottom water temperatures (Fig 8e–8h). Beginning in late fall, light wanes and upwelling weakens, resulting in a minimally productive surface ocean, more oxygenated water column, and warmer bottom water temperatures (Fig 8e–8h). Relative to the nearshore, the offshore region exhibits similar seasonality in SST and Chl, but is characterized by less phytoplankton biomass and deeper HLDs overall, with seasonal HLD phenology lagging the nearshore by approximately three months (Fig 8 vs. S6 Fig; see discussion in Sect [4.2]4.2).

Download:

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

Fig 8. Temporal variability of nearshore ocean extremes and their relationship to ENSO phases.

Daily time series (1996–2019; solid contours) of (a) SST (red), (b) Chl (green), (c) HLD (blue; inverted y-axis), and (d) bottom temperature (BT; black) regionally-averaged over the nearshore (0–75 km) coastal band (zones 1–4; see Methods). Additional contours in (a)-(d) show the daily climatology (dotted-dashed) and seasonally-varying percentile threshold (dashed; 90th for temperature, 10th otherwise). Extreme events of each ocean variable are indicated with under-the-curve and vertical shading. The duration of simulated El Ni no (brown) and La Ni na (teal) phases are indicated along the time (x) axis in (a)-(d), defined when the GLORYS-BGC Ni no 3.4 index exceeds ±1 standard deviation. (e-h) The same climatology shown in (a)-(d) is presented and enlarged as a single annual cycle beginning January 1, respectively.

https://doi.org/10.1371/journal.pclm.0000638.g008

Interannual variability of temperature, Chl, and HLD and the timing of their extremes (Fig 8a–8d) is largely a reflection of the influence of large-scale modes of climate variability, particularly ENSO, on the CCS thermocline (Figs 8a–8d, S6 Fig–S11 Fig; see [68] and references therein). In general, we find that nearshore SMHWs, LCXs, and BMHWs tend to occur during the strong El Ni no events of 1997–1998 and 2015–2016, as well as the 2014–2015 Blob period (Figs 8a, 8b, 8d, S7 Fig, S8 Fig and S10 Fig; [78]). In contrast, nearshore SHX occurrence is generally associated with La Ni na conditions, specifically the strong 2007–2008 and 2010–2011 events (Fig 8c and S9 Fig). However, nearshore HLD appears to exhibit a greater degree of lower frequency variability compared to temperature and chlorophyll (Fig 8 and S7 Fig–S9 Fig). While decadal modes of variability can modulate ENSO and mediate its influence on the CCS (e.g., [37]), this decadal modulation is less evident in SST and Chl, instead reflecting higher frequency surface variability (i.e., the forcing that drives these changes is more pronounced on shorter timescales). We also find that the timing of HLD extremes is lagged both (1) in the cross-shore direction, generally occurring in the offshore several years later than in the nearshore (Fig 8 vs. S6 Fig), and (2) in the along-shore direction, where extremes emerge progressively later from south to north, generally over the course of one year (S9 Fig). These spatio-temporal lags are largely reflective of the lower-frequency variability imparted by the large-scale circulation of the region (see discussion in Sect [4.2]4.2).

During the strong El Ni no events of 1997–1998 and 2015–2016, the nearshore is characterized by SMHWs, LCXs, and BMHWs, as well as deep HLDs (Figs 8 and 9). Indeed, the strongest nearshore deep hypoxia extreme (DHX; Table 1; 3.1σ maximum intensity on March 26, 1998) occurred during the 1997–1998 El Ni no, at a time when the region was experiencing surface-to-bottom MHW and 0–100 m LCX conditions (intermediate depths not shown in the former case). In the offshore region, a DHX began approximately 3-4 months after the nearshore one in 1997–1998, while no DHXs were observed during the 2015–2016 event, instead exhibiting shallower-than-normal HLDs, likely explained by low-frequency HLD variability and highlighting event-to-event differences in the CCS response to ENSO (S6 Fig; see [78]). One of the main differences between these two events, both among the strongest on record, lies in their distinct atmospheric drivers and the subsequent impacts on coastal upwelling. In particular, the 1997–1998 event was associated with more severe reductions in upwelling strength than the 2015–2016 event [79].

Download:

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

Fig 9. Annual snapshots of nearshore extremes during five case study periods.

Daily time series (solid contours) of nearshore-averaged extreme variables (rows; same as in Fig 8) isolated during a single year (July 1–June 30) of five case study periods (columns): from left to right, 1997–1998 (El Ni no), 2007–2008 (La Ni na), 2010–2011 (La Ni na), 2014–2015 (“Blob”), and 2015–2016 (El Ni no), respectively. Additional contours indicate the daily climatology (dotted) and both the 90th and 10th percentiles (dashed). High and low extreme events of each ocean variable are indicated with gray under-the-curve and vertical shading (refer to Table 1). Note the inverted y-axis for HLD (third row).

https://doi.org/10.1371/journal.pclm.0000638.g009

Immediately preceding the strong 2015–2016 El Ni no, the onset and peak of the Blob in 2013–2015 occurred against the backdrop of weak or aborted El Ni no conditions (not shown), when SST anomalies in the central equatorial Pacific and warm preconditioning along the coast may have aided its impact on the CCS (see [78,80]). As such, impacts to the nearshore during the Blob period resembled those of a strong El Ni no event, including surface and bottom MHWs and LCXs (Figs 8 and 9), but with relatively weaker and shorter BMHWs and without the DHXs (Fig 9). Indeed, the nearshore experienced only slightly deeper-than-normal HLDs during this period (Figs 8 and 9), while the offshore saw the end of a long-lasting SHX that began in late 2013, following a period of recurrent La Ni na-like conditions since the late 2000s (S6 Fig).

The La Ni na events of 2007–2008 and 2010–2011 are characterized by nearshore SHXs, with a strong and long-lasting SHX during the former and a relatively short and weak SHX during the latter (Fig 9). These La Ni na periods are also characterized by HCXs as well as MCSs in both the nearshore (Fig 9) and offshore (S6 Fig), as we would expect from enhanced trade winds and coastal upwelling. On average, the nearshore experienced strong and persistent SHXs during the 2007–2008 La Ni na, with conditions lingering well into 2009 (Figs 8c and 9), likely associated with a weaker second-year La Ni na in 2008–2009. While the offshore region also saw shallow HLDs during this time, it was not until the end of the 2010–2011 La Ni na that HLDs shoaled to an extremely shallow depth, with recurring SHXs through 2014 (S6 Fig). The temporal and regional differences in SHX timing in and around La Ni na events likely reflects the diversity of individual events and therefore the diversity in the CCS response to ENSO. In this case, the former was generally classified as a classic Eastern Pacific (EP) La Ni na, with the potential for more direct and immediate effects on the west coast thermocline, whereas the latter event was classified as a Central Pacific (CP) La Ni na, and likely to have a more delayed response in the CCS. While both are associated with enhanced coastal upwelling along the west coast, the location of the SST anomalies in the equatorial Pacific–known as ENSO diversity–may play an important role in shaping the associated atmospheric and oceanic teleconnections (see [81]), affecting the likelihood of extreme ocean conditions along the U.S. west coast [82], albeit with a large spread due to internal atmospheric variability [83].

We compare composite anomaly profiles of temperature, chlorophyll, and dissolved oxygen during all identified extreme events across the study period to average anomaly profiles derived from five case study one-year periods (S11 Fig; the same case study periods used in Fig 9). Across zones 1–8, composites represent the typical vertical structure associated with each type of extreme event, while the case study profiles reflect conditions during individual one-year periods characterized by distinct ENSO phases. Composites of extreme events largely reflect the physical and biogeochemical response to the ENSO state, consistent with [68] (S11 Fig). In general, composite MHW and LCX profiles mirror those during the three warm periods that include the 1997–1998 and 2015–2016 El Ni no events and the 2014–2015 Blob. During MHWs, there is a subsurface maximum temperature in the upper 100 m and with decreasing magnitude moving northward. During LCXs, the upper 170 m is characterized by anomalously low chlorophyll in the near-surface and high chlorophyll in the subsurface, with the inflection point found within the upper 100 m. In contrast, composite SHX profiles mirror those during the two cold periods that include the 2007–2008 and 2010–2011 La Ni na events. During SHXs, oxygen concentration is slightly elevated at the surface and reduced in the subsurface, a pattern that diminishes northward and offshore, with decreasing resemblance to the cold-period profiles; during La Ni na periods, this vertical structure reflects the thermal (solubility) and non-thermal (physical, biological) drivers of oxygen, respectively [68].

3.4 The most widespread extremes in the CCLME

How widespread are any of these extremes at a given time? During the 1997–1998 and 2014–2016 warm periods, SMHWs covered up to 80% and ∼100% of the CCLME, respectively (Fig 10a). The same holds true in the offshore band (Fig 10b) while nearshore coverages reached 100% during both periods (Fig 10c), reflecting the regional differences in their climate impacts (see Sect [3.3]3.3). As such, the most widespread single extreme during this 24 year period is the SMHW, covering 99.2% of the CCLME on March 29, 2015 (Fig 10a).

Download:

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

Fig 10. Spatial extent of single and compound extremes across the CCLME.

Time series of the total area affected by single (top panels) and compound (bottom panels) extremes, expressed as a percentage of the (a) CCLME, (b) offshore (making up 34.8% of the CCLME), and (c) nearshore (making up 13.9% of the CCLME). Note, y-axes ranges for compound extremes (bottom panels) differ between regions.

https://doi.org/10.1371/journal.pclm.0000638.g010

While LCX and SMHW-LCX prevalence is also elevated during El Ni no-related warm periods, the first half of 1996 features the most widespread of these extremes, covering up to 55% and 35% of the CCLME, respectively (Fig 10a). This occurs during the tail end of a weak-to-moderate La Ni na event in 1995–1996 (not shown). Despite the persistence of cold tropical conditions typically associated with La Ni na into 1997, atmospheric circulation in late 1995 sets up anomalously warm SST conditions along the west coast of North America in early 1996 (see [84]), highlighting event-to-event differences in ENSO impacts. These localized warm conditions in early 1996 are not typical of a La Ni na pattern, and the most widespread compound extreme, the SMHW-LCX, covers 35.2% of the CCLME on May 16, 1996 (Fig 10a). This underscores the importance of understanding how atmospheric circulation and local dynamics, rather than canonical ENSO states alone, can influence extreme events (see discussion in Sect [4.2]4.2).

We find that SMHW-SHXs and SMHW-LCX-SHXs are most prevalent outside of the nearshore and during El Ni no-related warm periods (Fig 10), a result largely obscured in a regional average (e.g., S6 Fig). At times during the 2014–2016 warm period, SMHW-SHXs cover nearly 15% of both the CCLME and its offshore band, fractional areas on par with SMHW-LCX coverage during that time (Fig 10a, 10b); the nearshore region appears to be largely unaffected by this particular compound extreme, as well as LCX-SHXs (Fig 10c). As for triple extremes, albeit a relatively rare occurrence, maximum coverage peaks at 1.4% of the CCLME on October 17, 2015 (Fig 10a) and 3.1% for the offshore band on November 15, 2014 (Fig 10b). Indeed, during the Blob, triple extremes covered nearly 15% of zone 7 by the end of 2014 (not shown), the largest fractional area covered by triple extremes at any one time in any of the 8 coastal zones; in zone 7, triple extremes can reach standardized intensities () as high as 4.6σ and can last as long as 78 days (Fig 6).

In a regionally-averaged sense, we have so far shown that SHX occurrence is generally associated with strong La Ni na conditions, such as during the 2007–2008 and 2010–2011 events. However, retaining the spatial dimension and quantifying the fractional area subjected to SHXs over time provides additional insights (Fig 10). Most notably, the spatial extent of SHXs exhibits relatively greater low frequency variability compared to other extremes, consistently covering between 2% and 23% of the CCLME at any given time over the study period, with maximum coverage occurring in 2012 (22.1%; Fig 10a). While conditions associated with the recurring La Ni na period between 2005 and 2012 could have contributed to the 2008 and 2012 peaks in SHX fractional area (Fig 10a, 10b), the comparable area values that persist after 2012 occur during a generally El Ni no-dominated period. We speculate that the low frequency spatiotemporal variability in SHXs identified outside of the nearshore band (Fig 10) is additionally driven by large-scale decadal variability in the region (e.g., gyre-scale circulation; see discussion in Sect [4.2]4.2). Conversely, the SHX-La Ni na relationship is more evident and strongest in the nearshore and exhibits higher frequency variability, with fractional areas peaking during La Ni na events, between 20–40% during any of the 1999-2000, 2007–2008, 2010–2011, and 2011-2012 events (Fig 10c).

4.1 Multi-stressors

Oceanic extreme events can alter both bottom-up (resource-driven) and top-down (predator-driven) ecosystem controls in unique and complex ways. As the landscape of compound extremes continues to shift in a rapidly changing world, the impacts to local food web dynamics and overall community structure become even more complex. While we show that the most common compound extreme in the CCLME is the SMHW-LCX (Fig 7) and that a triple MHW-LCX-SHX compound extreme is a relatively rare occurrence (Figs 5, 6, and 10), scientists and managers would benefit from a future comprehensive analysis of CCLME extremes that further includes combinations of additional known stressors, such as acidity and nutrients.

Like this study, previous research on compound extremes has analyzed combinations of up to three distinct physical and biogeochemical stressors (e.g., [56,85]). Lack of archived daily output of GLORYS-BGC pH precludes its explicit inclusion in this analysis of extremes, but we can assess the magnitude of its standardized monthly anomalies and comment on the timing of presumed multi-stressors (Fig 11). For example, during the 2007–2008 La Ni na event, anomalously low pH values (<−1.5σ) would suggest that in addition to a nearshore SHX (Fig 9), the ecosystem was additionally subject to acidic conditions, potentially threatening the ability of marine organisms like oysters and crabs to calcify [86]. Indeed, corrosive waters between 2007 and 2008 led to substantial larval mortality in commercial oyster hatcheries in the PNW [14].

Download:

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

Fig 11. Nearshore physical and biogeochemical variability in relation to ENSO phase.

(a) From top to bottom, time series of monthly standardized anomalies of the Ni no 3.4 climate index (brown bars) and nearshore-averaged monthly standardized anomalies of SST (red bars), Chl (green bars), HLD (blue bars), thermocline depth (TD; 3-month running mean; green contour), mixed layer depth (MLD; 3-month running mean; red contour), sea surface height (SSH; black bars), CUTI (gray bars; black contour indicates the 3-month running mean), vertically-integrated (0–100 m) net primary production (NPP; green bars), 100 m pH (pink bars), and 100 m nitrate (N; teal bars), phosphate (P; gold bars), and silicate (Si; purple bars) concentrations, respectively. All y-axes values have been restricted to a common range and are expressed in sigma units. (b-c) Schematic representation of a simplified view of some of the locally- and remotely-forced responses expected in the nearshore CCLME during a (b) El Ni no and (c) La Ni na. Positive (+) and negative (–) anomalies in atmospheric winds (northerly, out of the page; southerly, into the page), sea level pressure (SLP), the type of coastally trapped waves (CTWs; upwelling (↑) or downwelling (↓)), SSH, abundance of primary producers (star and oval shapes), temperature, TD (dashed line), upwelling (solid arrow), pH, oxygen concentration, and nutrient concentration (vertical colored gradient, where darker indicates higher). For illustrative purposes, size, weight, or density is used to indicate the relative differences in scalar properties between ENSO phases.

https://doi.org/10.1371/journal.pclm.0000638.g011

During the 1997–1998 El Ni no event, anomalously low nutrient concentrations (<−2σ; Fig 11a; [87–89]) are likely drivers of LCX conditions (Fig 9), while high pH values (>2σ; [90]) indicate a reprieve from ocean acidification (Fig 11a). While low nutrient availability can lead to LCXs, it can also contribute to HCXs under specific conditions. Some HCXs can be classified as harmful algal blooms (HABs) when certain toxin-producing phytoplankton (e.g., Pseudo-nitzschia) dominate, though such complex biogeochemical dynamics are not simulated in GLORYS-BGC. A striking example occurred in 2015, when an extensive and toxic Pseudo-nitzschia-based HAB poisoned food webs and closed fisheries along the U.S. west coast [10]. Ryan et al. [91] showed that this diatom bloom followed a strong upwelling event that injected nutrients into coastal waters, though with an anomalously low silicate-to-nitrate (Si:N) ratio signature. These silicate-limited, nitrate-replete conditions supported dense diatom populations and may have promoted elevated toxin production (see [91] and references therein). Consistent with observed conditions, GLORYS-BGC captures persistently low Si:N ratios beginning during the Blob and extending through the 2015–2016 El Ni no (Fig 11a), reaching a 24-year nearshore minimum (–0.25) on March 16, 2016 (not shown). Laboratory and field studies have shown that Pseudo-nitzschia from the CCS can thrive under warm anomalies and low Si extremes (Si limitation) and produce higher amounts of toxin, particularly when also exposed to acidic conditions [92]. Improving our understanding of the drivers and interactions of nutrient extremes–not just their absolute concentrations but also their stoichiometric ratios–is critical for advancing multi-stressor research, predictive capabilities, and risk management.

Our results and their limitations highlight the need for creating and maintaining observational platforms capable of capturing concurrent extremes. Satellites provide the greatest spatiotemporal resolution in the study of oceanic extremes, but are currently limited to surface ocean color- and SST-based events. For example, satellite chlorophyll retrievals are proxies for the total amount of phytoplankton present, with no distinguishing information on community structure. While some studies have attempted to infer phytoplankton functional types using post-hoc algorithms applied to standard ocean color products, these methods have generally been considered experimental, with limited accuracy and reliability. In contrast, NASA’s recently launched Plankton, Aerosol, Cloud, ocean Ecosystem (PACE) mission integrates advanced algorithms for phytoplankton group discrimination directly into its standard data processing, representing a significant step forward for future observational studies of LCXs and HABs. In situ observations (e.g., moorings, floats, gliders, ships) suffer from lower spatiotemporal resolution and/or coverage, but often boast a suite of concurrent measurements, including temperature, chlorophyll, oxygen, nutrients, and/or pH from which to study many multi-stressors, and potentially their vertical structure (as in [93]), which is an important factor in understanding the ecological implications of compound extreme events. Global and regional ocean reanalyses marry model output with observations to produce gap-free products, and the relatively few that include biogeochemistry are critical to the continued study of multi-stressors. While GLORYS-BGC is coarser in resolution than some regional models (e.g., [69,94,95]), this ocean hindcast with biogeochemistry provides consistent, multi-decadal, full water-column coverage across the entire CCLME at daily resolution–crucial for the spatial and temporal scope of this study–while its global domain and standardized output support future comparative applications in other LMEs.

4.2 Characterizing compound extreme events to inform marine resource management

We show that MHWs, LCXs, SHXs, and their combinations tend to occur during strong ENSO periods (Figs 8–10, S6 Fig–S10 Fig). The most common compound extremes in the CCLME are SMHW-LCXs in the nearshore and BMHW-SHXs and BMHW-LCXs along the continental slope and shelf, respectively (Fig 7). The ability to skillfully predict such extremes on subannual-to-decadal timescales would support coastal management (e.g., see [96]), such as the near-term decision to proactively close a fishery ahead of a strong MHW in an attempt to mitigate impacts. Seasonal-to-interannual forecasts leveraging ENSO as a source of predictability have already demonstrated skill for MHWs, ocean acidification extremes, and even chlorophyll anomalies in the CCS (e.g., [7,97–99]).

Taking a generalized view of large-scale climate variability on interannual-to-decadal timescales, we interpret standardized monthly anomalies in ecosystem-relevant quantities (Fig 11a) and present a schematic of a simplified view of the ENSO-CCLME relationship from 1996–2019 (Fig 11b, 11c). During the strong El Ni no events of 1997–1998 and 2015–2016 (Fig 11a, 11b), the nearshore ocean was characterized by elevated sea surface height, deeper thermocline and mixed layer depths, weaker upwelling (quantified by the coastal upwelling transport index, or CUTI; [71]), warmer SST, deeper HLD, higher pH (i.e., more basic conditions), and lower concentration of the macronutrients nitrate, phosphate, and silicate at 100 m (below the thermocline), along with reduced bulk chlorophyll and net primary production (Fig 11a). In contrast, during the strong La Ni na events of 2007–2008 and 2010–2011, the opposite patterns generally occurred (Fig 11a, 11c); however, the response in monthly thermocline and mixed layer depths during La Ni na is less consistent and more dependent on coastal location (see [100]). We also note that the upwelling (CUTI) response differed between the two El Ni no events, with the 2015–2016 event exhibiting greater spatial and temporal variability in key modulation pathways, including weaker-than-expected tropical teleconnection and anomalous upwelling-favorable winds in some regions (see [79]). These relationships (Fig 11a) align with previous modeling studies that highlight event-to-event variability in ENSO impacts along the coast (e.g., [68,78]).

While ENSO may have predictive power, it is important to remember that there is still uncertainty associated with the CCLME response to ENSO and more research on quantifying this uncertainty is needed. Drawing on an example from Sect [3.4]3.4, in the seasons following the moderate 1995–1996 La Ni na (not shown), our results show SMHWs and SMHW-LCXs extending unusually far across the CCLME–at areal extents more typically seen during weak-to-moderate El Ni no conditions (e.g., up to ∼15% in 2018-2019; Fig 10b, 10c). This occurred despite continued tropical La Ni na conditions into 1996–1997, likely due to an atmospheric circulation setup over the Northeast Pacific that favored warm SST conditions in early 1996 (not shown; see [84]). Such examples from our analysis underscore the importance of exercising caution when attributing local extremes solely to canonical ENSO phases.

Decadal variability may also provide a potential source of predictability. We know that the combined effects of physical (e.g., circulation) and biological (e.g., respiration) variability set the low-frequency variability of oxygen in the thermocline [101]. For the CCS, gyre-scale circulation of the North Pacific plays an important role in modulating subsurface oxygen through the propagation of water mass anomalies on decadal timescales [102,103]. For example, subsurface advection injects relatively young water masses with a high oxygen signature from the North Pacific gyre into the CCS ([102] and references therein). The predictability of oxygen and other ocean tracers (e.g., [104]) derives from the multiyear memory associated with these gyre-scale drivers [103]. The California Undercurrent has additionally been cited as an important source of water mass anomalies in the nearshore CCS, transporting relatively old water masses with little-to-no oxygen from the tropics northward along its path [105]. One additional hypothesis suggests that the PDO and/or NPGO generate this low-frequency variability, yet, similar to the findings of Pozo Buil and Di Lorenzo [103], we find statistically non-significant correlations between HLD and the PDO or NPGO indices (not shown). Recognizing that the PDO and NPGO are indices of statistical modes of variability that represent the summarized effects of underlying physical processes, we underscore the need to focus on these component physical processes, which may play a more direct role in modulating oxygen variability and extremes in the CCS. For example, the PDO reflects shifts in wind patterns, ocean circulation, and the advection of water masses, all of which influence oxygen levels and may be critical drivers of extremes [36]. More research is needed to better reconcile the potential interaction between gyre-scale and undercurrent drivers in (1) modulating nearshore HLD variability and extremes on seasonal-to-decadal timescales, (2) lending predictive skill, and (3) playing a role, if any, in the apparent lag we observe in the seasonality of HLD and timing of extremes in the along-shore and off-shore directions. Despite these complexities, to first order, water column dissolved oxygen–and thus, HLD–variability across the CCS is not completely understood, attributable to a historical lack of long-term, sustained observations across broad spatial scales and longer timescales.