The phenology of the spring phytoplankton bloom in the North Atlantic does not trend with temperature

Vernal phenology of processes within temperate to polar marine ecosystems has shifted in recent years, and climate change is anticipated to exacerbate these changes in decades to come (Poloczanska et al., 2016; Chmura et al., 2019). The spring transition period is pivotal in the progression of lower trophic level production in these marine ecosystems owing to the importance of seasonal mixing dynamics and nutrient redistribution (Rebstock and Shil Kang, 2003; Dessier et al., 2018). Generally, nutrient availability coupled with increasing light penetration during the spring transition period results in spring bloom formation (Townsend et al., 1994). The spring bloom, which serves as the engine of annual system productivity, is not homogeneous with respect to the composition (Spilling et al., 2018) and size of species (Devred et al., 2011) of primary producers, mixotrophs, and heterotrophic predators (Heneghan et al., 2019). The spring bloom initiates trophic responses in the food chain that influence organisms at different trophic levels. For example, vernal migrations of coastal migratory fishes to a fixed geographic terminus has been shown to vary with changes in thermal regimes (Langan et al., 2021; Dalton et al., 2022). Likewise, marine birds adjust their migratory arrival times in response to similar changes (Staudinger et al., 2019). Considering the importance of the spring bloom to North Atlantic marine ecosystems, the foci of this paper are the recent thermal changes that have occurred in the North Atlantic as actuating factors of the spring bloom and the dynamics of spring bloom initiation, duration, and magnitude as climate change responses.

A myriad of factors control the initiation and duration of the spring bloom. However, the main factors are believed to revolve around the preconditions of the availability of light and the structure of the water column. From experimental work and field observations, three main hypotheses have emerged to explain the formation of a phytoplankton bloom, which provides the basis of continuing research today (Chiswell et al., 2015). The seminal work on the topic is the critical depth hypothesis (Sverdrup, 1953), which relates the depth of the mixed layer to the depth of effective light penetration (Smetacek and Passow, 1990). The second main hypothesis, the critical turbulence hypothesis, suggests a bloom may develop if the rate of mixing or turbulence in the mixed layer does not exceed a level that transports phytoplankton cells out of the photic zone (Huisman et al., 1999). Finally, the third main hypothesis is the dilution-recoupling hypothesis that suggests bloom formation is mainly related to the top down control of grazing by planktivores (Behrenfeld, 2010). These concepts provide a framework to consider retrospective change in the spring bloom and anticipated adaptations of the bloom to climate change. Moreover, all the hypotheses are intimately connected to the effects of thermal regime on bloom development.

Regional studies within the North Atlantic basin suggest that the application of broadly applied hypotheses of bloom formation are reasonable, but at the same time, other factors affecting bloom development have been identified. The classical application of critical depth (Rumyantseva et al., 2019; Yang et al., 2020) and the formation of mixed layer dynamics (Mignot et al., 2018; Morison et al., 2019) support the contention that some form of these more universal hypotheses capture the range of factors that initiate the spring bloom. However, evidence indicates that localized effects, like the availability of nutrients (Browning et al., 2020; Browning et al., 2021) and change in freshwater runoff (Torres et al., 2020), are other important factors (Sommer and Lengfellner, 2008) in shaping bloom dynamics. In the application of these concepts to the phenology of blooms, except for the constant annual cycle of solar radiation, all other factors may be affected by thermal regime, including the species composition of the phytoplankton communities in the North Atlantic (Bolaños et al., 2020). In shallow coastal waters, mixing dynamics will probably remain unaltered with climate change, so bloom initiation is more likely to be controlled by light availability, whereas in deeper systems, thermal stratification and wind mixing will likely play a more dominant role in the timing of blooms (Winder and Sommer, 2012). Experimental evidence supports the contention that because light availability contributes strongly to spring bloom timing, other transitional factors affected by warming conditions will play a smaller role (Sommer and Lengfellner, 2008). However, these additional factors still impact the timing of peak phytoplankton production (Hallegraeff, 2010). Across the North Atlantic basin, there are regions of varying depth and mixing dynamics with different seasonal thermal regimes. These settings provide a test of climate change effects already underway, which can inform what may be anticipated with future change in thermal conditions. Hence, any investigation into the effect of climate change on bloom phenology should logically consider the change in ecosystem temperature that has already occurred.

Generally, sea surface temperatures (SSTs) are increasing globally, including in the North Atlantic (Barbosa and Andersen, 2009; Garcia-Soto et al., 2021; Karnauskas et al., 2021). The distribution of thermal conditions is determined to a large degree by the Atlantic meridional overturning circulation (AMOC), which is responsible for the northward transport of warm water into the North Atlantic. The AMOC is sensitive to many factors, including freshwater input (Sévellec et al., 2017), variations in wind forcing (Yang et al., 2016), rising global temperatures (Rahmstorf et al., 2015), and anthropogenic climate change (Liu et al., 2020). A direct relationship between rising North Atlantic SST and weakening of the AMOC is a matter of ongoing discussion (Ditlevsen and Ditlevsen, 2023). While the direction of change with respect to temperature is well established in the North Atlantic, its associated impact on the timing of spring bloom formation is not well known. Temperature change can have many non-linear effects on water column structure, phytoplankton metabolism, and grazers, to name only a few factors that interact to produce observed spring bloom formation and senescence.

The goals of this study were two-fold: first, to analyze SST dynamics in the North Atlantic; and second, in a complementary fashion, to examine spring bloom phenology, noting that the spring bloom is the dominant feature of the production cycle throughout most of the North Atlantic. Though widely reported on both global and regional scales, ocean warming associated with climate change varies substantially with the region. Therefore, we examined changes in temperature and looked for evidence of seasonal warming regimes. The dynamics of the spring bloom were examined to evaluate the distribution of change in bloom timing and size. Patterns in bloom phenology were then compared to changes in seasonal thermal regimes.

The study was restricted to the North Atlantic Ocean basin and to latitudes with the availability of remotely sensed temperature and ocean color data for the detection of thermal regimes and spring blooms (Figure 1). The study grid was sized to 2-degree latitude and longitude cells to improve spatio-temporal consistency of ocean color data and reduce the number of cells with missing values in the phytoplankton bloom analysis.

The change in thermal conditions of the North Atlantic were examined using three metrics: (1) SST trends across the North Atlantic, (2) the relative change in seasonal warming, and (3) the change in spring transition phenology. These analyses used the SST data sourced from the NOAA Optimum Interpolation (OISST) 0.25 Degree Daily SST Analysis dataset (Reynolds et al., 2007). Change in North Atlantic SST was visualized over the study grid as the change in mean annual temperature over the study period 1998–2022. Time series change in the annual means, and the other climate metrics listed below, was tested using a two-tailed Mann-Kendall non-parametric test of trend (Mann, 1945). We guarded against the issue of inflated significance caused by temporally auto-correlated data by using the Yue and Pilon correction method (Yue et al., 2002), with a Theil-Sen slope estimated for each time series. All trend estimates were made using the R package “zyp” (version 0.10-1.1). Trend and other statistical tests were performed for the entire sample grid that numbered 755 locations. When trend and correlations (further described below) were mapped, outliers in the data were identified using the “boxplot” command in R and were replaced with the highest or lowest value after outlier removal for positive and negative values, respectively.

In addition to analyzing the mean annual temperature trends, seasonal warming and cooling metrics were developed following the approach in Li et al. (2021). For each grid cell, mean SST was calculated by quartile or 25% quantiles. Therefore, the first quantile mean represents the lowest 25% of the daily SST values in a given year. In the same way, the next three quantile means were calculated as the means of the successive 25% increments of the data, respectively. The fourth quantile mean contained the highest 25% of the daily SSTs in a given year.

The phenology of spring thermal conditions was characterized using the date of arrival of a spring transition temperature (Friedland et al., 2015a). The spring transition temperature is the average annual temperature for each grid cell; the same mean temperature serves as the autumn transition temperature, though not used in this analysis. For each year, the daily SST data were smoothed with a 5-day moving average filter; the first day of the year that SST exceeded the transition temperature was scored as the spring transition day of the year for that year (see Figure 2a for an illustrated example).

Phytoplankton spring bloom parameters were derived from a time series of chlorophyll-a (CHL) concentration from the Hermes GlobColour website (see data availability statement) at a spatial resolution of 4 km and temporal resolution of 8 days from 1998 to 2022. The CHL data were a merged product using the Garver, Siegel, Maritorena model that combined the data using a bio-optical model inversion algorithm (Maritorena et al., 2010). The procedure combines data from the Sea-viewing Wide Field of view Sensor (SeaWiFS), Moderate Resolution Imaging Spectroradiometer (MODIS, on the Aqua satellite), Medium Resolution Imaging Spectrometer (MERIS), Visible and Infrared Imaging/Radiometer Suite (VIIRS), and Ocean and Land Colour Instrument (OLCI) sensors. These satellite remote sensing data provided a weighted average of the CHL in the top two optical depths of the sea, covering typically the surface to 5–15 m depth depending on water clarity (Gordon and McCluney, 1975).

The dynamics of the spring bloom, including initiation, senescence, and duration, were determined using the change point algorithm used previously in both regional and global analyses of bloom dynamics (Friedland et al., 2015a; Friedland et al., 2018). Bloom periods were identified with the change point algorithm sequential t-test analysis of regime shifts (STARS; Rodionov, 2006) also over the period 1998–2022. The STARS algorithm was parameterized following an extension of a method designed to increase bloom detection sensitivity (Friedland, 2021). The STARS α parameter was tested at the values of 0.05, 0.1, and 0.15; the length criteria, at values of 4, 5, and 6 time steps. The Huber weight was kept constant at 3. In previous assessments, blooms that exceeded a duration of nine 8-day periods (equivalent to 72 days) were considered ecologically different from discrete blooms and were excluded from the analysis (Friedland et al., 2015b). For each bloom detection, bloom start day of the year was determined, bloom duration was expressed as the number of consecutive 8-day periods associated with the event, bloom magnitude was calculated as the sum of CHL concentrations during the bloom period (mg m⁻³ 8-day), and bloom intensity was the average CHL concentration (mg m⁻³) during the bloom period (see Figure 2b for an illustrated example). The detections were attempted on a half-year, or a time series of 23 8-day CHL periods, incremented by one 8-day period throughout the entire time series. In total, over 7.7 million bloom searches were conducted. The summation for all bloom detection start dates over the study grid is shown in Figure 3. Over 2.01 million blooms were detected, keeping in mind that many of the detections were of the same bloom event. From this effort, we determined that most spring blooms start on the day of year 113 (April 22). Therefore, we restricted the analysis to using 15 bloom search windows that began on day of the year 1 through 97 and days 353 and 361 from the previous year covering the period 1998–2022; these search windows account for over half of the total number of bloom detections. Bloom parameters were based on the results of the 9 search windows that yielded the most detections per grid location.

Spring bloom dynamics were tested for potential relationships with the variables for thermal conditions. For each grid location, time series of bloom start, duration, magnitude, and intensity were correlated with mean annual temperature, the four quantile SST means, and spring transition day. We used Pearson product-moment correlation and evaluated correlation at a p-value of 0.05. Correlation coefficients were plotted over the North Atlantic study grid noting the distinction between significant and non-significant correlations.

The trends and correlations among SST and spring bloom dynamics are subject to spatial autocorrelation, which should be accounted for properly when drawing conclusions about their trajectories. Among uncorrelated locations, at least 5% of statistical tests are expected to identify significant trends (p < 0.05), by random chance alone. That is not the case, however, when locations are subject to spatial autocorrelation. To protect against spurious conclusions, we only considered a trend to be present at the map level when more than 10% of the individual locations yielded significant results (p < 0.05). This relatively conservative approach should reduce the likelihood of overstating effects. Simulations of conditions similar to those found in our dataset corroborated this threshold, as the most extreme cases identified significant trends in 10% of the locations.

Spatial autocorrelation was accounted for by formally assessing each SST and spring bloom metric, as well as their correlations. Moran’s I (Gittleman and Kot, 1990) was used to identify the degree of spatial clustering among locations; the statistic was significant at p ≤ 0.01 for all variables, so it is not mentioned in the results and can be taken to suggest that all the trend and correlation statistics were spatially correlated. Spatial generalized least squares (GLS) models were then fit to each trend and correlation statistic to test for map-scale effects independent of that autocorrelation. Two models were fit for each statistic: one to establish the overall effect, and another to test for latitudinal trends. The GLS models utilized distance-based covariances estimated from an exponential spatial covariance function, with range and nugget parameters estimated via maximum likelihood, following methods by Ives et al. (2021, 2022).

SST increased over most of the North Atlantic during the study period, but cool and warm temperatures did not trend equally throughout the basin. With the exception of the central portion of the basin, mean annual SST trended significantly positive and averaged a warming rate of 0.16°C per decade over the entire basin (Figure 4a). Of all grid locations, 32% were positive trends compared to just 1% having negative trends (Table 1). The overall trend throughout the basin, after correcting for spatial autocorrelation, was 0.22°C per decade (Table 2). The first quantile SST, representing cold SSTs, also trended mostly positive with a rate of 0.06°C per decade over the entire basin (Figure 4b). This lower overall rate was also reflected in a lower number of significantly positive trends (20% of the grid) versus a higher proportion of negative or cooling trends (8%), compared to the annual SST patterns. This pattern continued with the data for the second through fourth quantile SST means (Figures 4c–e). The proportion of significant and positive location-level trends shifted from 26% to 28% to 32% of the grid cells, where the proportion of significant negative trends shifted from 5% to 2% to 0% of the grid cells, for the respective quantiles. Though SST trends across the North Atlantic tended to be mostly positive, we saw a gradation of more to less trends in cooling to warming trends across the quantile data. Significant map-level effects were exhibited by the third and fourth quantile (0.29°C and 0.50°C per decade, respectively; Table 2), but not the first or second. In contrast to the significant warming trends across the quantiles, there were few significant trends in spring transition day (Figure 4f). There were 92 grid locations or 12% of the grids with significant negative trends, suggesting earlier spring transitions with warming, and only 21 or 3% of the grid with positive trends. The overall effect was negative (−2.5 days per decade; Table 2). Negative trends in transition day were concentrated in the southern part of the study area and coincident with areas of significant warming temperatures. This finding was reflected by a significant latitude effect in the corresponding GLS model (Table 2).

We detected no systematic change in spring bloom timing or duration across the North Atlantic. The mean trend in bloom start day was −0.69 days per decade indicating an insignificant advancement in blooms; however, over the study grid, only 8% of the trends were negative and significant (Table 3). The few significant trends were seemingly distributed randomly over the basin (Figure 5a). Spring bloom duration showed no signs of systematic change either with the mean trend over the study grid being −0.11 days per decade or a relatively modest shortening in bloom duration (Figure 5b). Only 7% of the grid cell trends were negative and significant (Figure 5b), and like bloom start, significant trends were randomly dispersed across the basin. Bloom magnitude and intensity, on the other hand, declined over time, with the mean trend of −0.71 mg m⁻³ 8-day period per decade and −0.09 mg m⁻³ per decade, with 16% and 17% of the grid field having significant and negative trends, respectively (Figure 5c, d). Trends of magnitude and intensity corrected for spatial correlation averaged −1.03 and −0.12 mg m⁻³ per decade (Table 2), respectively, for the basin. Most of the negative individual trends in bloom magnitude were concentrated in shelf sea areas in both the Northeast and Northwest Atlantic; these areas represent on average a 42% decline in magnitude and a 36% decline in intensity.

At the location level, there were no correlations between thermal variables and bloom start day or duration; however, there were significant negative correlations between cool temperature variability and the two bloom size metrics of magnitude and intensity. Overall, we found that none of the bloom parameters showed any pattern of positive correlation with the thermal parameters; hence, in the following we focus on the negative correlations only (Table 4). Bloom start day was significantly and negatively correlated with less than 10% of the study grid for all the thermal parameters except annual mean temperature (Table 4); the number of negative correlations between start day and annual mean SST represented 10% of the grid. However, these correlations were distributed randomly over a number of areas in the study grid (Figure 6). The correlation between bloom duration and the thermal parameters only showed 5% significant correlations over the study grid; like bloom start, there was little spatial cohesion to the distribution of significant correlations (Figure 7). Bloom magnitude, on the other hand, correlated significantly and negatively with mean annual SST in 24% of the grid field, while the second quantile SSTs correlated in 16% of the grid. These negative correlations with thermal variable were both mainly distributed in the southern part of the study grid (Figure 8a–c). Small regional clusters of correlated grid locations were found with the other thermal parameters (Figure 8d–f). Negative correlations between the thermal parameters and bloom intensity showed the most pronounced relationships. The highest proportion of grid cells was observed between first quantile SST and intensity in 31% of the grid; these correlations, like the magnitude correlations, were distributed in the southern portion of the study grid (Figure 9b). Strong negative correlation fields were also observed with annual mean SST and the second and third quantile SSTs, with similar spatial distribution to the negative correlations (Figure 9a, c, d). The remaining correlation fields were characterized by sparsely distributed significant correlations (Figures 9e, f). In summary, patterns of bloom size variability were associated most closely with winter temperatures, suggesting that cool winter conditions stimulate higher CHL magnitude and intensity while warm winter temperatures inhibit CHL levels.

At the basin scale, bloom start and duration were significantly, but weakly, associated with thermal variables (Table 2). All significant associations were negative, except with spring transition, which was slightly positively associated with both bloom start and duration. By contrast, thermal associations with bloom magnitude and intensity were more pronounced. The annual mean and first quantile of SST (which can be related to winter) were moderately associated with magnitude (−0.17 and −0.21, respectively) while the third quantile association was moderate (−0.09). Intensity was also associated with the mean and first through third quantiles of SST (−0.21, −0.25, −0.18, and −0.11, respectively). Thermal correlations with bloom magnitude and intensity were also more pronounced as latitude increased, except among the upper quantile.

Despite significant warming throughout the North Atlantic, we found little evidence of any concomitant pattern of change in spring bloom phenology. This analysis includes both key phenological indices of bloom start and duration, which taken together determine when in the production cycle a bloom occurs and contributes to the size of the bloom in terms of biomass accumulation that exceeds grazing, and thus can be transported beyond pelagic food webs (Stock et al., 2017). Our findings are consistent with analyses of seasonal blooms in the Norwegian Sea (Silva et al., 2021), where spring bloom timing and duration were without long-term trends. However, Silva et al. (2021) also examined later season blooms, characterized as a summer bloom, that did display a significant time series trend of delayed bloom timing. A similar pattern was observed in the Northwest Atlantic, specifically, on the US Northeast Continental Shelf where the spring bloom was found to be without trend. However, the autumn bloom within some sub-regions showed trends toward delayed bloom start and shortened bloom period (Friedland et al., 2023). Bloom timing has changed in higher latitudes of the North Atlantic, namely the Barents Sea, but these changes are related to earlier sea ice retreat and may be a special case (Song et al., 2021). Sea ice dynamics most likely do not influence bloom activity in our study region, where instead solar radiation and mixing dynamics are regulating bloom dynamics (Chivers et al., 2020; Dong et al., 2020; Song et al., 2021).

While we conclude that spring bloom phenology in the North Atlantic is essentially unchanged in recent decades, we acknowledge that may be a function of the length of our temporal lens. On multi-year timescales there is evidence that the North Atlantic Oscillation influences bloom timing (Zhai et al., 2013). During the period of our CHL data, the North Atlantic Oscillation, and for that matter the Atlantic Multidecadal Oscillation, were in one phase, so these basin scale drivers were unlikely to produce the observable trends. However, both advanced and delayed trends in spring bloom patterns have been observed over shorter periods of time (1998–2012) in the North Atlantic (Taboada and Anadon, 2014). This mixed pattern of change in bloom timing is similar to what Friedland et al. (2018) found in what was termed the dominant seasonal bloom, which for the North Atlantic was primarily characterized by a spring bloom. Long-term in-situ data (1957–2017) also show that the abundance of diatoms, dominant components of the spring bloom, is associated with multidecadal climate trends, but diatom phenology does not appear to have shifted in time (Edwards et al., 2022). There is evidence on the US Continental Shelf that 30–40 years of observations are necessary to detect trends in bloom phenology associated with basin-scale climate indices (Record et al., 2019). While our primary assertion is not in conflict with other CHL or taxon-specific observations, an explicit examination of future climate shifts as they become available in the North Atlantic time series is highly recommended.

For large extents of the North Atlantic, our findings suggest that bloom magnitude and intensity appear to be governed by cooler winter temperature variation, independently of any trend in these parameters. A general phenomenon of an association between temperature and size spectra of phytoplankton, within both marine and freshwater systems, reflects a range of dominant phytoplankton forms from picoplankton to microplankton (Venkataramana et al., 2019; Zohary et al., 2021). More specifically, thermal conditions at the onset of a bloom have been associated with the dominant taxa comprising the bloom, most generally an oscillation between diatoms and smaller forms such as picocyanobacteria. The dynamic between water temperature and phytoplankton size spectra has been observed in coastal waters where CHL tended to be lower and cyanobacteria, picoeukaryotes, and nanoeukaryotes dominated over diatoms under warm conditions (Trombetta et al., 2019). These thermal modifications to spring bloom phytoplankton community structure may also involve a replacement of diatoms with other diatom species and dinoflagellates (Hjerne et al., 2019), a species replacement that could modify the movement of fixed carbon through the ecosystem owing to the differences in sedimentation rates between the species. The generalization of any relationship between temperature and phytoplankton community composition should be tempered, as there is evidence that nutrient regime is also a significant modifier of community structure (Anderson et al., 2022). Because large portions of the southern part of our study area showed a negative correlation between CHL and temperature, a potential driver of this contrast in biomass may be related to shifting species composition of spring bloom in these areas. Moreover, and by analogy, we suspect it may be differences in the contribution of picoplankton to microplankton species that drives this effect.

Despite significant increases in SSTs throughout the North Atlantic, only limited portions of the basin showed any change in the timing of the spring thermal transition. The degree of stability in the spring transition day provides an indicator that suggests the timing of the onset of water column stratification may be relatively stable. However, the more important finding is that for most of the continental shelf seas of the North Atlantic, there was no trend in the day of spring transition. These highly productive shelf seas also showed no systematic change in bloom phenology. Whether these findings are consistent at depth is not possible to evaluate with our remote sensing observations. There is the hypothesis that SST influences the depth and gradient of the mixed layer and hence access to nutrients (Lewandowska et al., 2014); to the extent that this hypothesis might hold, it appears that temperature-induced changes in nutrient availability and physiology have not significantly affected what we can observe remotely in sea surface chlorophyll.

The absence of any coherent trends in spring bloom phenology supports the notion that photosynthetically active radiation, as controlled by sun zenith angle and day length, is among the primary factors controlling bloom timing. Factors that control timing of light availability vary with latitude but do not vary inter-annually. Similarly to our results, Silva et al. (2021) did not find any trend in spring bloom phenology, but they did observe coherent patterns of phenology change in later mid-summer blooms. Despite inter-annual variability on the order of months, Friedland et al. (2023) found no trend in the same spring bloom phenological metrics for the US Northeast Shelf ecosystem but did detect change in phenology of an autumn bloom. In both cases, the seasonal blooms that occurred later in the calendar year would be initiated when light is not limiting and would likely be more associated with change in water column structure, wind mixing, and nutrient distribution.

A decrease in spring bloom magnitude displayed some regional coherence and was mostly associated with blooms in continental shelf seas. These areas are also the locations where major oceanographic shifts have been documented, such as changes in the Gulf Stream, the Labrador Current, and the interactions between these currents (Gonçalves Neto et al., 2021). Bloom magnitude is driven by two factors, the CHL in the water column (bloom intensity) and the duration of the bloom period. Because bloom duration, like bloom start, displayed no coherent pattern of change in our study, we attribute the decline in bloom magnitude to a decline in CHL concentration, as seen in the trend in bloom intensity, and perhaps related to the latitudinal pattern of change in diatom contribution to spring blooms (Edwards et al., 2022). This view is consistent with the analysis of trends in global CHL concentration reported in Friedland et al. (2021), which is expected because those results are derived from the same datasets as the current study. Though spring bloom phenology may not have played a role in shaping the contribution of marine primary production to the ecological services of the North Atlantic such as those related to food production (Costanza et al., 2014), these data are further evidence of changes in total basin-scale spring productivity. While bloom phenology does not appear to be shifting in time, warming temperatures will increase metabolic rates of phytoplankton grazers (Franzè et al., 2023), hence causing a shift in resource-consumer dynamics. In addition, upper trophic level consumers often will have reproductive cycles that are driven by temperature, such as zooplankton diapause (Pierson et al., 2013) and development rates (Campbell et al., 2001), potentially creating phenology mismatches if the bloom-timing windows are relatively static (Ljungström et al., 2019).

Concern is mounting over the potential impacts of climate change on the phenology of relationships between fish reproduction and recruitment, and the production cycles of lower trophic level resources (Asch et al., 2019; Rogers and Dougherty, 2019; Ferreira et al., 2023). In attempts to be forward-looking, researchers have already developed analyses that show a range of climate change responses in the timing and duration of phytoplankton blooms that we might expect to see in future decades. Most of these nascent efforts suggest that the timing of the spring blooms will advance as the result of physical changes to the ecosystem like heat-mediated stratification and mixing (Hashioka et al., 2009; Henson et al., 2018; Asch et al., 2019; Mészáros et al., 2021). Expectations are that climate change will contribute to earlier phytoplankton bloom formation (Thackeray et al., 2016). However, these predictions are tempered by the fact that different phytoplankton taxa show varied phenological responses to warming. For example, diatoms have been observed to shift to an earlier component of the bloom in some ecosystems (Hjerne et al., 2019) but not in others (Chivers et al., 2020). Moreover, functional group responses within a single ecosystem are not usually coherent, as the bloom timing of other species, such as dinoflagellates and autotrophic ciliates, may change or shift later in the season (Hjerne et al., 2019), forming a zero sum game and buffering any measurable shift in bloom timing. The combined effect would generate a low signal-to-noise ratio that would be difficult to observe using only sea surface chlorophyll from remote sensing sources. Some model results suggest that zooplankton grazing may be the prime agent in future changes to bloom timing (Yamaguchi et al., 2022). Other analyses suggest bloom timing may be delayed by other physical forcing effects like wind stress at critical times of the year (Vikebø et al., 2019). Furthermore, predictions are complicated by anticipated change in the seasonal bloom dynamics of harmful algae (Clark et al., 2022). The predictive landscape of the anticipated change to bloom timing is varied, with perhaps a common thread forming around the idea that bloom start may be advanced. Ultimately, bloom dynamics are an emergent property that integrates multiple complex and interrelated processes, with the complexity of the interactions possibly providing more resilience than has been hypothesized (Caron and Countway, 2009).

Climate changes have already occurred on a scale that is similar to the anticipated changes suggested by climate projection models, so does the backward-looking exercise of recent events form a standard of expected future change? From the data presented here and elsewhere, we can assert that SST change in the North Atlantic has been on the order of 0.2°C–0.3°C per decade, which is what we can expect to see over the next half century (see figure 8 in Ruela et al., 2020). Under these conditions, we have seen advanced bloom timing in higher latitude systems where climate change has caused decreasing seasonal sea ice coverage (Kahru et al., 2011; Marchese et al., 2017). We have seen change in long-term data from the Continuous Plankton Recorder, suggesting that there have been changes in seasonal peaks of specific but not all phytoplankton taxa that would match a shift to earlier bloom timing in general (Edwards and Richardson, 2004). Results from a comparative study suggested that some parts of the world ocean have seen modest advancement in bloom timing; our analysis is in agreement with that work in suggesting that any change in bloom timing in the North Atlantic is not significant (Friedland et al., 2018). Combined with previous research, our findings suggest that for most of the North Atlantic, expected future changes in thermal conditions will likely have little effect on bloom timing observable by remote sensing. However, we note that the ocean climate is entering a state without empirical analog, and in this context, past trends are not always indicative of future responses. Analyses of global SSTs have shown a higher frequency of surprising conditions than would be expected based on trends alone (Pershing et al., 2019), a possibility that also exists for bloom dynamics.

Sea surface temperature data is available from the Optimum Interpolation Website: https://www.ncei.noaa.gov/products/optimum-interpolation-sst. Chlorophyll concentration data is available from the Hermes GlobColour website: https://hermes.acri.fr/.

The authors thank anonymous reviewers of this paper for their productive comments. They also thank A. Ives for guidance on study statistics. EcoFOCI contribution number is EcoFOCI-1038 and Cooperative Institute for Climate, Ocean, & Ecosystem Studies (CICOES) contribution number is 2023-1255.

The authors declare that they have no conflict of interest.

NRR was supported by NSF OCE 2049308.

Contributed to conception of the article: KDF.

Contributed to the design of the article: KDF, DCB.

Contributed to processing and analysis of the data: KDF, CJM.

Contributed to the interpretation of the data: All authors.

Drafted the article: KDF, JMN, DCB, CJM.

Contributed to revision of the article: All authors.

Approved submission of the manuscript: All authors.

How to cite this article: Friedland, KD, Nielsen, JM, Record, NR, Brady, DC, Morrow, CJ. 2024. The phenology of the spring phytoplankton bloom in the North Atlantic does not trend with temperature. Elementa: Science of the Anthropocene 12(1). DOI: https://doi.org/10.1525/elementa.2023.00111

Domain Editor-in-Chief: Jody W. Deming, University of Washington, Seattle, WA, USA

Associate Editor: Kevin R. Arrigo, Environmental Earth System Science, Stanford University, Stanford, CA, USA

Knowledge Domain: Ocean Science