Daily MODIS-Aqua imagery (Level 1A, 1 km²) from 2003-2016 were obtained from NASA’s Oceancolor web portal (https://oceancolor.gsfc.nasa.gov/). Details of satellite data processing are described in Suchy et al. (2019). Briefly, data were atmospherically corrected using SeaDAS (developed by the Ocean Biology Processing Group, OBPG) following regional methods outlined in Carswell et al. (2017) and applying the OC3M chlorophyll a (Chl a) algorithm (McClain et al., 2000). In addition, a mask (indicated by an asterisk in Figure 1) was applied to eliminate all satellite data from the region near the mouth of the Fraser River, as recommended in Komick et al. (2009), in order to minimize any error associated with the high turbidity known to occur in this region. Pixels were excluded with standard quality-control flags (e.g., contamination caused by straylight, high solar and sensor zenith angles). As a result, no data were available from mid-November to mid-February due to the position of the sun and extensive cloud cover during the winter months.

All Chl a products were spatially binned and mapped to a common grid at 1.1 km² spatial resolution and transformed to a base 10 logarithm (Campbell, 1995). Missing pixels due to persistent cloud cover in the region were spatially interpolated using the Data Interpolating Empirical Orthogonal Functions (DINEOF) methodology (Beckers and Rixen, 2003) with considerations of the optimal temporal resolution of the input dataset as defined in Hilborn and Costa (2018). Implementation of the DINEOF method resulted in a daily spatially-continuous, gap-filled time series that was subsequently binned into 8-day “weeks”. Maps showing the seasonal and spatial variability in Chl a during the same years considered in this study are presented in Figures 8, 9 in Suchy et al. (2019). Although interannual variability in the spatial extent of Chl a was observed, Chl a concentrations during spring (mid-February to May), on average, were highest in the southern portion of the Central SoG near the mouth of the Fraser River. Warmer years exhibited widespread Chl a concentrations > 5 mg m^-3 throughout the entire region, whereas Chl a blooms during colder years were more localized (Suchy et al., 2019).

Commonly used metrics for defining the start of the spring bloom in remote sensing studies include the date when chlorophyll concentrations first become 5% greater than the local annual median (e.g. Siegel et al., 2002; Henson et al., 2009) or selecting the date when chlorophyll concentrations first reach a pre-defined threshold concentration, e.g. 2 mg m^-3 (Jackson et al., 2015). Given that our 8-day weekly Chl a data showed substantial variability from spring through to fall (Supplementary Table 1; and see Suchy et al., 2019), we chose a more conservative definition of spring bloom initiation. Spring bloom initiation was defined using a threshold value of the median + 5% Chl a concentration for a given year. Bloom initiation occurred when: i) the first week in our Chl a time series reached the threshold Chl a concentration, and ii) at least one of the two following weeks were >70% of the threshold value. This additional criterion was imposed to ensure that we did not prematurely assign bloom initiation to a given week in the Chl a time series if that week was followed by a long period of low Chl a concentrations. It should be noted that although using threshold values of 75% or 80% yielded the same results for all years except 2012 (Supplementary Table 2), we used the 70% value so as to not limit our ability to interpret bloom initiation in 2012. From these results, we calculated the mean spring bloom initiation week. Chl a bloom years were considered “early” or “late” if bloom initiation was greater than or equal to 1 standard deviation about the mean. In addition, we characterized spring bloom intensity and spring bloom magnitude using slightly modified definitions given in Friedland et al. (2015; Friedland et al., 2016) in order to account for the high variability in Chl a concentrations throughout the months considered in this study. Weeks with Chl a concentration above the median + 5% threshold concentration were considered to be under ‘bloom conditions’. Bloom intensity was then defined as the average Chl a concentrations across all weeks under bloom conditions over the course of the spring (mid-February to May), whereas spring bloom magnitude (a proxy for bloom duration) was defined as bloom intensity multiplied by the number of weeks under bloom conditions. Due to the dynamic nature of the region, it was not possible to clearly define the end of the bloom in a given year. Therefore, we focused on bloom dynamics during the spring only; the impacts of summer and autumn bloom conditions on crustacean zooplankton will be considered in future studies.

Zooplankton samples from 2003-2016 were collected as part of larger sampling programs that have been conducted in the SoG since 1996 by Fisheries and Oceans Canada (see Perry et al., 2021 for details). Sampling frequency has varied throughout the years; however, since 2015 sampling has been conducted approximately monthly from February to October (Table 1). Spatial coverage of the zooplankton sampling also varied between years, with a few core stations being consistently sampled throughout the time series (see Figure 1 in Perry et al., 2021). While the spatial coverage during the early part of our time series (i.e., prior to 2015) was mostly limited to these core stations, the spatial extent of zooplankton sampling has expanded throughout the Central SoG since 2015. For our study, only vertical net tow data from Scientific Committee on Oceanic Research (SCOR), Bongo, or ring nets with mouth diameters of approximately 50 cm and mesh sizes of approximately 250 µm were selected from the database. Typical vertical tow profiles were obtained from ~10 m above the bottom of the seafloor to the surface. Nets were retrieved at 1 m s^-1, and calibrated flow meters were used to record the volumes filtered by each tow. Samples were preserved immediately in 10% sodium borate-buffered formalin in seawater. We focused our analyses on the deep (bottom depths deeper than 50 m) Central SoG as this region had the best temporal resolution over the time period of our study. In addition, we selected net tows that covered at least 70% or more of the water column to ensure that our samples included the diel and seasonal vertical migrating zooplankton taxa.

In the laboratory, samples were examined using stereomicroscopes (Wild dissecting microscope up to 2013, and Zeiss SteREO Discovery 8 thereafter) and processed in two parts based on zooplankton size. Whole samples were first scanned for large (> 5 mm) or rare individuals, then split using a Folsom splitter to approximately 100 individuals > 5 mm. Plankton in these subsamples were identified to species, sex and stage where possible, and subsequently removed. The remaining subsample containing organisms < 5 mm was split to approximately 400 individuals which were identified to the lowest possible taxonomic classification and life history stage or size class. Biomass was calculated from the abundance data using the average biomass of individuals derived from measured or literature values. The average biomass values per individual taxa (species and life stage) are provided in the Open Data Portal listed above (see also Perry et al., 2021). Abundance and biomass data are presented as numbers or weight (dry mass; mg) per m³, respectively.

For taxonomic groupings, we followed those defined in Table S1 in Perry et al. (2021). Specifically, we focused on the following crustacean groups: hyperiid amphipods, small (prosome length < 1 mm) calanoid copepods, medium (prosome length 1-3 mm) calanoid copepods, large (prosome length > 3 mm) calanoid copepods, non-calanoid copepods, cladocerans, ostracods, decapods (larval crabs), and euphausiids. Daytime adult euphausiid abundance/biomass was multiplied by 3 to adjust for under sampling due to net avoidance (Mackas et al., 2013; Perry et al., 2021).

Due to inconsistent and patchy temporal sampling throughout our study period, the zooplankton data were insufficient to look at peak timing in individual years using common mathematical methods such as the central tendency (e.g. Edwards & Richardson, 2004) or the cumulative percentiles (e.g. Mackas et al., 2012) methods. Therefore, we calculated the climatological peak timing of crustacean abundance and biomass by taking the mean value for a given month over the course of our time series (2003-2016) and smoothing the data using simple spline curves. Additionally, we calculated similar climatologies using mean monthly crustacean abundance and biomass values for the “early”, “average”, and “late” years as defined by the Chl a bloom metrics in order to assess the match/mismatch between phytoplankton and crustacean zooplankton under different bloom conditions.

Spring bloom initiation throughout our study (2003-2016) occurred, on average, during the last week of March (DOY 89; middle of the 8-day week). These results differ slightly from Schweigert et al. (2013) who used SeaWiFS level 3 data from 1997-2010 to show that satellite-derived bloom initiation, defined as either the first annual date when two consecutive mean Chl a values were greater than the annual median + 5% or the first date when a selected threshold of 6 mg m^-3 occurred, for the entire SoG typically occurred between DOY 60 and 80. However, our results were well within the spring bloom timing (defined as the peak in chlorophyll biomass) of mid-March to mid-April in the Central SoG identified by Bornhold (2000). Furthermore, Jackson et al. (2015) showed that Chl a climatology typically begins to increase in March in the SoG, however a peak in Chl a was not observed in their study until May, possibly due to the fact that they considered the entire SoG region as a whole, over different climatological years (1997-2010), using different satellite data and processing methods. Our results were also consistent with the long-term mean spring bloom start date (defined as the maximum phytoplankton concentration when nitrate was depleted) in the Central SoG of March 25 (DOY 84) determined by a one-dimensional biophysical model (Allen and Wolfe, 2013) despite the fact that our study considered a larger spatial scale (i.e., the entire Central SoG instead of a single sampling station) and a different range of years (2003-2016 in our study compared to 1968-2010 presented in Allen and Wolfe, 2013). Similarly, Peña et al. (2016) used a three-dimensional coupled biophysical model to determine that the spring bloom typically occurred between April 2 and 15, only slightly later than what we report in our study. Although Peña et al. (2016) also defined the spring bloom as the peak in phytoplankton biomass, and not initiation per se, their model did reveal that phytoplankton biomass begins to increase in February and typically reaches concentrations > 5 mg m^-3 in March, consistent with our findings for bloom initiation. Whereas the previous examples are consistent with spring phytoplankton bloom timing in the Central SoG being around the end of March, they differ in their details, in part because they used different metrics (i.e., bloom initiation, peak in phytoplankton biomass, initial drawdown of nutrients, etc.), over different spatial and temporal scales, even within the same relatively small study region.

Despite the various methods used to characterize bloom timing in the SoG, the “early” (late February/early March) Chl a bloom years defined by this study (2004, 2005, 2015) were corroborated by results from other studies. In situ surface Chl a samples previously determined that the 2005 bloom began during the third week of February in the Central SoG (Halverson and Pawlowicz, 2013). In addition, 2005 and 2015 were designated as early bloom years in the Central SoG based on high resolution surface measurements of chlorophyll fluorescence from ferry box sampling (Sastri et al., 2016) and a coupled bio-physical model (Collins et al., 2009; Allen et al., 2016). In fact, 2015 was found to be the second earliest bloom (the earliest being late February 2005) as measured by the ferry systems in the Central SoG since 2001 (Sastri et al., 2016). The “late” bloom years defined in this study (2007, 2008) did not always agree well with previous studies. For example, observation data of in situ Chl a indicated that the bloom in 2007 occurred between April 2-8 (DOY 92-98; Sastri et al., 2016), whereas modeled bloom timing for 2007 revealed a slightly later bloom data of April 10 (DOY 100; Allen and Wolfe, 2013), which was closer to our observed value of the week beginning with DOY 113 (end of April). In contrast, coarser resolution (standard SeaWiFS and MODIS) satellite-derived chlorophyll data from 1997-2010 suggested an early bloom (February-March) in 2007 (Thomson et al., 2012); however, they noted that the 2008 bloom was delayed relative to their observed early bloom of 2007. Similarly, Gower et al. (2013) determined relatively early but more localized Chl a blooms in both 2007 (mid-February) and 2008 (mid-March) using the fluorescence line height approach for MODIS satellite imagery. Discrepancies in results between this study and other satellite-based studies may be due to the use of region-specific algorithms for atmospheric correction and the models used to retrieve chlorophyll data in the present study (see Carswell et al., 2017; Suchy et al., 2019).

A detailed discussion of environmental drivers of phytoplankton variability in the SoG based on the same satellite-derived Chl a data used in this study is presented in Suchy et al. (2019). Results from this previous work suggested that local drivers (i.e., Fraser River discharge), rather than large-scale climate indices, influence spatial and temporal variability of chlorophyll concentrations in the Central SoG. However, examination of the environmental drivers of Chl a bloom dynamics in this study, as opposed to chlorophyll concentrations per se, revealed that spring Chl a bloom initiation was strongly positively related to NPGO and SOI and negatively correlated with SST and PDO (Figure 5); warmer years tended to have earlier blooms whereas colder years exhibited average or late blooms. This general pattern of early initiation of the spring bloom associated with warmer SST has been found in numerous other studies (e.g., Sasaoka et al., 2011; Henson et al., 2013; Zhao et al., 2013; Hunter-Cevera et al., 2016). In contrast to our results, Allen & Wolfe (2013) found no significant relationship between modeled spring bloom date and temperature in the Central SoG; yet their results did show a weak relationship between bloom initiation and NPGO (r = 0.36, p = 0.05), which supports our findings. The warm phase conditions (-NPGO, -SOI, +PDO; Suchy et al., 2019) observed during early bloom years in the SoG were also associated with increased light levels (PAR) and weaker winds during late winter/early spring, both of which are known to impact bloom initiation in the SoG (Yin et al., 1997; Collins et al., 2009; Allen & Wolfe, 2013).

Warm-phase conditions may have direct and indirect effects on phytoplankton. For example, higher SST can result in earlier blooms if increased temperatures also act to increase the photosynthetic (Henson et al., 2006) or cell division (Hunter-Cevera et al., 2016) rates of phytoplankton cells. Alternatively, warmer atmospheric temperatures may indirectly impact Chl a concentrations during the spring by causing earlier snowmelt and increased freshwater runoff, resulting in increased water column stratification (Yin et al., 1997; Foreman et al., 2001; Ji et al., 2007; Suchy et al., 2019). Stratification in the SOG typically peaks in the summer (June or July) coincident with the peak in Fraser River discharge (Suchy et al., 2019). Chiba et al. (2008) found that warmer conditions during the winter in the Oyashio region resulted in strong stratification in the spring, which prompted earlier bloom timing. We found no significant relationships between spring bloom initiation and either SSS, Fraser River discharge, or stratification; however, spring bloom intensity, the average Chl a concentration over weeks under bloom conditions, was significantly related to stratification (Adj R² = 0.38, p < 0.05; Table 2). These results further support our previous findings that increased stratification in the SoG during late winter/early spring as a result of warm-phase conditions contributes to early bloom formation and high Chl a concentrations compared to colder years (Suchy et al., 2019). Furthermore, warmer years with earlier Chl a blooms subsequently resulted in a higher spring bloom magnitude, a finding that has previously been reported for both the Grand Banks of Newfoundland (Zhao et al., 2013) the US Northeast Shelf (Friedland et al., 2015), the North Atlantic (Friedland et al., 2016), as well as on a global scale (Racault et al., 2012; Friedland et al., 2018). While no direct relationship was found between SST and bloom magnitude in our study, variability in spring bloom magnitude (our proxy for bloom duration) was best explained by NPGO (Adj R² = 0.57, p < 0.01; Table 2). Specifically, years associated with -NPGO (warmer years) were characterized by weaker spring winds (Figure 5), likely allowing blooms to persist for a longer duration, and thus resulting in the observed higher bloom magnitudes.

The climatological seasonal zooplankton cycle in the SoG is known to have a late-spring to late-summer peak and a winter/early-spring minimum (Mackas et al., 2013). Our results for the climatology (and average Chl a bloom years) showed that the maximum peaks in both abundance and biomass of total crustaceans occurred in July (Figure 4). The initial peak in crustacean abundance, however, occurred just prior to April and at the same time as the first peak in Chl a. This initial peak in crustacean abundance is likely attributed to the high numbers of juvenile crustaceans (e.g., euphausiids, copepod nauplii, and the copepodite stages included in the medium calanoid copepod grouping), which feed predominantly in the upper water column in response to rapid increases in phytoplankton. Euphausiid development cues are known to be strongly coupled with phytoplankton biomass (Ross & Quetin 2000) and their reproductive cycles have previously been shown to be triggered by episodic phytoplankton blooms (Ji et al., 2010). Results from our study indeed revealed a tight coupling between euphausiid abundance and Chl a biomass, wherein the peak timing of euphausiids corresponded to the main peak in Chl a regardless of Chl a bloom timing (Figure 6). Furthermore, there was an initial smaller peak in total crustacean biomass in May (Figure 4E), which is consistent with the findings from previous studies in this region that determined zooplankton biomass typically peaks about one month after the maximum phytoplankton biomass is reached (e.g., Stockner et al., 1977; Peña et al., 2016).

The seasonal patterns we observed in monthly median Chl a concentrations varied between early, average, and late Chl a bloom years. While our results indicate that bloom concentrations during late Chl a bloom years were associated with colder temperatures, lower light levels and stronger winds (which can cause extensive vertical mixing), zooplankton grazing may also disrupt or delay phytoplankton blooms. As such, the timing and composition of the zooplankton community is an important factor contributing to Chl a bloom dynamics given that high abundances of zooplankton grazers prior to a bloom initiation may inhibit its development (Parsons et al., 1966; Siegel et al., 2002). Below we discuss the synchrony in phytoplankton and zooplankton phenology during early and late Chl a bloom years in order to better understand what constitutes a match vs. mismatch year for lower trophic levels in the Central SoG.

Median monthly Chl a concentrations were > 5 mg m^-3 from February through April during early bloom years with a notable decrease in Chl a concentrations during March (Figures 4B, F). A decline in Chl a concentrations in the spring bloom in the SoG was previously shown to coincide with increased zooplankton biomass during a time when surface nitrate was still available, thus indicating that grazing, as opposed to nutrient limitation, was the cause of the decline (Peña et al., 2016). We suspect that this may be the case during the early bloom years in this study as the decline in Chl a in March coincided with an increase in the abundance of non-calanoid copepods, medium calanoid copepods, and amphipods that resulted from an earlier shift in the initial peak abundances of these groups (Figure 6). Variability in cohort timing of N. plumchrus, the copepodite stages of which are included in our medium calanoid copepod group, is known to be positively correlated with spring surface temperature anomalies in the Northeast Pacific (Mackas et al., 2012). Mackas et al. (1998) previously observed that the peak zooplankton biomass during “warm”, “normal”, and “cold years” at Ocean Station Papa in the North Pacific was distinctly early (and even higher) during years when the surface layers were anomalously warm. Our results support these findings and suggest that the shift in timing of herbivorous species may have contributed to the observed decrease in Chl a biomass during the spring of early years. While early bloom years were associated with the highest crustacean abundance observed in our study (> 1000 ind m^-3 during August; Figure 4B), this high abundance was mainly attributed to smaller-bodied non-calanoid copepods (Figures 6, Figure 10). Overall, the low biomass values observed during these years suggest that early Chl a blooms cause a mismatch for phytoplankton and larger crustacean zooplankton in the Central SoG.

Chl a concentrations remained < 5 mg m^-3 from February through April during late bloom years and were subsequently higher (5 to 10 mg m^-3) during summer and autumn compared to early and average bloom years (Figures 4D, H). Somewhat surprisingly, the earliest peak in total crustacean biomass during our study was observed during May to June for late Chl a bloom years (Figure 4H) coinciding with the time period when Chl a concentrations reached > 5 mg m^-3. This main peak in crustacean biomass during late Chl a bloom years was lower, but of a longer duration, relative to average bloom years. Friedland et al. (2015) suggested that grazing pressure of zooplankton on phytoplankton may be higher during late bloom years when zooplankton populations have had sufficient time to develop prior to the bloom start, thus resulting in smaller Chl a blooms. Total crustacean abundance initially peaked in March during late bloom years in our study and was largely driven by a high abundance of medium-sized calanoid copepods (Figure 6). Therefore, it is possible that late Chl a blooms may have experienced heavier grazing pressure by zooplankton already present in the surface waters, thus preventing Chl a concentrations from increasing during the spring of those years (Siegel et al., 2002; Zhao et al., 2013). That said, even though crustacean abundance was lowest during late Chl a bloom years (Figure 4D), crustacean biomass was still relatively high when 2007 and 2008 were considered together (maximum biomass of 56.1 mg m^-3; Figure 4H). Our results are in contrast to a study from nearby Rivers Inlet, BC, which found that the phenology of five main copepod species was delayed when the spring bloom was later (Tommasi et al., 2021); however, the authors note that phenological responses to environmental variables may be region-specific. In addition, our study considers larger taxonomic groupings and a longer time period (i.e., 14 years compared to the 3 years of observations in Tommasi et al., 2021). It is also important to mention that while our satellite-derived metrics defined both 2007 and 2008 as late Chl a bloom years, these years showed marked differences in both spring bloom intensity and magnitude, which were higher in 2007 compared to 2008 (Figure 3B), as well as in crustacean abundance and biomass, which was lower in 2007 relative to 2008 (Figure 8). Furthermore, the lack of summer samples in 2007 and 2008 may have biased our zooplankton-related phenology results for late Chl a bloom years and likely resulted in lower annual crustacean abundance and biomass estimates than would be observed with higher temporal sampling resolution.

One of the most notable findings of this study came from the clustering analysis of crustacean community composition. Our hierarchical clustering results based on crustacean abundance data, alone, clearly separated zooplankton communities associated with early bloom years (Group 2; 2004, 2005, 2015; Figure 9) as defined by the satellite-derived Chl a bloom phenology. In contrast, the main group (Group 1) included average Chl a bloom years, whereas Group 3 separated out the late bloom year of 2007 (but not 2008). Analysis of the crustacean taxa comprising each of these groups revealed a higher proportion of non-calanoid copepods in the years associated with Group 2 (i.e., early Chl a bloom years). These results are somewhat expected given that higher temperatures, associated with the early bloom years in this study, tend to favour smaller zooplankton taxa in the SoG (Mackas et al., 2013). However, smaller copepods such as non-calanoid and small calanoid copepods are not an ideal food item for juvenile salmon (Preikshot et al., 2010), which suggests that early bloom years may negatively impact the quality of zooplankton prey available for higher trophic levels.

In contrast, average Chl a bloom years associated with Group 1 had a slightly higher relative abundance of medium calanoid copepods and larger crustaceans, e.g. euphausiids and amphipods, resulting in the highest overall biomass (40.3 mg m^-3) of all three groups (Figure 10). This higher proportion of large crustaceans, particularly large-bodied copepods with higher lipid content, is indicative of better food quality available for higher trophic levels (Preikshot et al., 2010; Mackas et al., 2013). As was observed by Tommasi et al. (2013), we observed a shift in community composition resulting in a substantially lower abundance of euphausiids during the late bloom year of 2007 (Figure 10). Coupled with the fact that overall crustacean abundance was substantially lower in 2007, this contributed to low overall biomass values for Group 3 that were comparable to the low biomass values observed during the early Chl a bloom years comprising Group 2 (26.1 and 27.7 mg m^-3 for Groups 2 and 3, respectively). However, given our poor sampling resolution in 2007, any potential impacts of crustacean zooplankton as food, in terms of both quantity and quality, on higher trophic levels during late Chl a bloom years, remains unclear.

Previous studies have extensively examined the link between variations in zooplankton anomalies in relation to both large-scale climate indices and local factors in the SoG (Li et al., 2013; Mackas et al., 2013; Perry et al., 2021). Mackas et al. (2013) determined that zooplankton biomass from 1990-2010 was strongly related to NPGO, particularly on decadal scales; however, the authors highlighted that relationships between environmental variables and individual zooplankton taxa are often weak and confounded by the presence of multicollinearity amongst the environmental variable themselves in this region. More recently, Perry et al. (2021) found that zooplankton biomass anomaly trends from 1996-2018 were related to sea surface salinity, PDO, and peak spring phytoplankton bloom date. Given the consideration of these relationships elsewhere, we focused instead on determining the environmental variables associated with changes in crustacean community composition by comparing large-scale climate indices, environmental drivers, and bloom dynamics between the years grouped in our zooplankton cluster analysis. Results from this analysis indicated that spring bloom initiation was significantly different between clustering years (p < 0.05; Figure 11). In addition, SOI and two key variables associated with phytoplankton bloom formation, PAR and wind, were significantly different between clustering years at alpha = 0.10. Negative SOI values (warm-phase El Niño conditions), weak winds, and high PAR values likely contributed to early Chl a bloom initiation and were associated with the crustacean communities clustered into Group 2 (years 2004, 2005, and 2015; Figure 11). Previous studies have linked SOI to changes in the phytoplankton community (an increase in heterotrophic and autotrophic dinoflagellates) (Pospelova et al., 2010), the zooplankton community (Haro-Garay and Soberanis, 2008; Li et al., 2013), as well as to salmon production in the SoG (Beamish et al., 1997). The main El Niño events (negative values of SOI) during this study occurred from 2003 to mid-2005, and then again from mid-2014 to 2016 (Suchy et al., 2019), thus encompassing the early Chl a bloom years as defined by both satellite-derived metrics and zooplankton cluster analysis. Further evidence of these warm-phase conditions is revealed by the observed differences in NPGO and SST (non-significant) between the three cluster groups (highest SST and lowest NPGO values during years associated with Group 2; Figure 11).

Warming conditions are known to result in a shift in phytoplankton community composition, ultimately resulting in smaller size classes of zooplankton (Richardson, 2008). For example, El Niño events have been shown to result in earlier spring blooms (Yin et al., 1997; Yoo et al., 2008; Sasaoka et al., 2011) and may result in a shift of the phytoplankton assemblage to smaller phytoplankton cells (nanoflagellates; Harris et al., 2009). In addition, shifts from diatom-dominated to dinoflagellate-dominated communities have been linked to negative NPGO periods in the California Current System (Fischer et al., 2020). Although phytoplankton community composition data were not available during the years of our time series, HPLC data previously analyzed for this region showed that there was a higher proportion of smaller phytoplankton cells (i.e., haptophytes) compared to diatoms in the Central SoG during spring of the early bloom year 2015 (Nemcek et al., 2020). In terms of zooplankton, Haro-Garay & Soberanis (2008) determined that the low abundances of N. plumchrus and E. pacifica during the 1997 El Niño year were attributed to an early spring bloom which supports our results. Given that we have previously shown that El Niño years are associated with high Fraser River discharge in the SoG (Suchy et al., 2019), it is plausible that warmer conditions, increased stratification, and potentially higher nutrient limitation, may result in a shift in both phytoplankton and zooplankton sizes towards smaller species (Kamykowski and Zentara, 2005; Richardson, 2008) during warm-phase El Niño conditions. Future data are needed in order to fully elucidate these shifts by comparing both the phytoplankton and zooplankton communities simultaneously, as well as the nutrient concentrations, during El Niño versus La Niña years in the SoG. To our knowledge, this study provides the first evidence linking earlier spring bloom timing to a shift in the crustacean community towards smaller taxa in response to multiple warm-phase events in the SoG. In contrast, strong winds and low PAR values during the cold-phase La Niña year of 2007 resulted in the late spring bloom start date associated with Group 3. Previous analyses of zooplankton community variability from 1990-2010 in the SoG (including a portion of the data used in the present study) also indicated that 2007 was the largest outlier in zooplankton ordinations, and that the low zooplankton biomass during this year was related to strong winter winds (Mackas et al., 2013). Furthermore, Perry et al. (2021) determined that there was a significant shift in zooplankton trends between 2006 and 2007 over the time period 1996 to 2018. Therefore, results from these longer-term studies support our findings that the late Chl a bloom year of 2007 had anomalously low crustacean biomass despite our lack of summer zooplankton sampling.