Shifts in diatom and dinoflagellate biomass in the North Atlantic over 6 decades

Abstract The North Atlantic Ocean has large seasonal blooms rich in diatoms and dinoflagellates which can contribute disproportionately relative to other primary producers to export production and transfer of resources up the food web. Here we analyze data from the Continuous Plankton Recorder to reconstruct variation in the surface ocean diatom and dinoflagellate community biomass over 6 decades across the North Atlantic. We find: 1) diatom and dinoflagellate biomass has decreased up to 2% per year throughout the North Atlantic except in the eastern and western shelf regions, and 2) there has been a 1–2% per year increase in diatom biomass relative to total diatom and dinoflagellate biomass throughout the North Atlantic, except the Arctic province, from 1960–2017. Our results confirm the widely reported relationship where diatoms are displaced by dinoflagellates as waters warm on monthly to annual time scales. The common assumption that gradual ocean warming will result in a decadal-scale shift from diatoms to dinoflagellates was not supported by our analysis. Predicting the effects of climate change likely requires consideration of the consequences for the whole community, the simultaneous change of multiple environmental variables, and the evolutionary potential of plankton populations. Citation: Mutshinda CM, Finkel ZV, Irwin AJ (2025) Large, regionally variable shifts in diatom and dinoflagellate biomass in the North Atlantic over six decades. PLoS One 20(6): e0323675. https://doi.org/10.1371/journal.pone.0323675 Editor: Barathan Balaji Prasath, Gujarat Institute of Desert Ecology, INDIA Received: February 25, 2025; Accepted: April 12, 2025; Published: June 4, 2025 Copyright: © 2025 Mutshinda et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability: CPR data are available from Johns, D. 2019. Continuous Plankton Recorder Survey (CPR Survey), v 1.2. DOI: https://doi.org/10.17031/1629. HadISST data were obtained from: https://www.metoffice.gov.uk/hadobs/hadisst/. Funding: This work was supported by grants from the Simons Foundation (549935 to AJI, 549937 and 986772 to ZVF), the Ocean Frontier Institute (NWABCP to AJI and ZVF), and Discovery grant awards from the National Science and Engineering Research Council of Canada (AJI, ZVF). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Introduction North Atlantic phytoplankton communities are often dominated by blooms of diatoms in the spring and dinoflagellates in the late summer and early fall [1–3]. Diatoms are characterized by their need for silicon for their cell wall, their tendency to form spring blooms in the mid and high latitudes, and are estimated to contribute approximately 40% of total marine primary production and particulate carbon exported to depth [4–6]. Dinoflagellates often bloom later than diatoms, are amongst the most genetically, morphologically, and trophically diverse groups of plankton; and include photosynthetic, mixotrophic and heterotrophic members [7]. Dinoflagellates are typically bi-flagellated with an organic cell wall formed from specialized vesicles (alveolae) that can form carbon-rich plates of various thicknesses [8,9]. Their C-rich walls, and often large genomes [10], can make their biomass higher in C:N and lower in N:P compared to the siliceous-walled diatoms [11,12], influencing their quality as food to higher trophic levels [13]. In coastal waters, toxin production by some diatom and several dinoflagellate species can episodically disrupt fishing, aquaculture, and economic activities [14,15]. Changes in the relative success of diatoms relative to dinoflagellates are expected to have a large impact on pelagic food web dynamics, fisheries, and the biochemical cycling of carbon, nitrogen, and phosphorus. The Continuous Plankton Recorder (CPR) survey initiated by the Sir Alister Hardy Foundation for Ocean Science has been monitoring the plankton in the surface North Atlantic Ocean since the 1930s with millions of plankton abundance counts [16,17]. A silk mesh captures many larger phytoplankton species, predominantly diatoms and dinoflagellates. The methods used by the CPR survey have changed little since its initiation, providing an ideal dataset for quantifying how changes in physical and chemical conditions have altered plankton community structure and whether climate warming has left a detectable fingerprint on North Atlantic plankton community structure [17,18]. Analyses of the CPR dataset suggest there have been shifts in the abundance of diatoms relative to dinoflagellates over time and with environmental conditions in some regions of the North Atlantic over the past half century, but the direction of change may be regional. For example, Leterme et al. [19] report increases in the CPR phytoplankton colour index (PCI), the abundance of dinoflagellates, and decreases in the abundance of diatoms in the North Atlantic over the period 1958–2002, especially over the Southeast Atlantic since 1988. The annual development and decline of diatom and dinoflagellate blooms in the North Atlantic follow changes in irradiance, nutrient concentrations, temperature, and grazing, with diatoms proceeding dinoflagellates in cooler, more mixed, nutrient rich waters. Long-term trends in growth conditions may be expected to shift diatom and dinoflagellate communities in similar ways to the interannual cycle, so that warming and increased stratification may favour dinoflagellates at the expense of diatoms. In the central North Atlantic, Zhai et al. [20] reported a decadal increase in the ratio of dinoflagellates and a decrease in the ratio of diatoms, relative to all other phytoplankton, over 1985–2009. In contrast, Hinder et al. [21] reported an increase in the abundance of diatoms relative to dinoflagellates over 1960–2009, but with most of the increase occurring post 1995 in the northeast Atlantic (45–60°N, 15°W-10°E). Most recently, Holland et al. [22] reported increases in diatom abundance in the CPR record (1960–2019) and many additional shorter near-shore time-series across the Greater North Sea but decreases in the Northeast Atlantic. By contrast dinoflagellate abundance showed decreases in the North East Atlantic and a mixture of increases and decreases in the Greater North Sea. Changes in diatom and dinoflagellate community abundance over the last 60 years in the North Atlantic have been attributed to changes in temperature, water column stratification and nutrient supply, and wind-driven mixing [19,21,23]. Due to differences in scope and methods, it is difficult to determine if contrasting conclusions about diatom and dinoflagellate abundances across studies should be attributed to differences in the physical regimes across the regions, the time periods analyzed, or differences in how the data were analyzed [22]. Analyses across studies differ in the spatial and temporal subdivision of the North Atlantic examined and the taxa included in the analyses, potentially impacting conclusions. For example, Leterme et al. [19] included species with a frequency of occurrence >1%, Hinder et al. [21] examined the integrated abundance of diatoms and dinoflagellates present in > 4% of samples and with persistent abundance across their timeframe of analyses, while Zhai et al. [20] examined the total diatom and dinoflagellates abundance relative to all phytoplankton taxa identified. A major challenge in plankton time-series analysis is the large number of zero abundance data which is often overcome by focussing on dominant taxa and transforming data, both of which can induce biases [24]. Furthermore, interpretation of changes in abundance and the relative abundance of diatoms and dinoflagellates is complicated by large differences in the cell size and carbon content of the different diatoms and dinoflagellates making up the phytoplankton communities. Collections of species with similar ecological or biogeochemical roles are commonly aggregated in models or data analyses into functional groups. We develop our analysis at the functional group level, i.e., aggregate biomass of diatoms and dinoflagellates, rather than genera or species, to greatly reduce the number of zeros in our dataset, because many species within these groups have similar niches leading to approximately neutral fluctuations in abundance of species within functional groups [25] and because aggregations of species are inherently more predictable. Diatoms are generally more prevalent in colder, nutrient-rich and well-mixed waters, while dinoflagellates are found in warmer, lower nutrient concentration, and stratified waters. To more robustly assess if there have been temporal changes in diatom and dinoflagellates since 1960 across the North Atlantic, and if there are substantive differences in temporal trends across regions, we quantify the temporal trend in diatom and dinoflagellate biomass and a diatom index (= diatom/ (diatom + dinoflagellate) biomass) using a carbon content estimate of the species identified and a statistical model for the seasonal cycle, decadal-scale fluctuations, de-trended and de-seasoned temperature, latitude, and the long-term trend (linear change over years) in five biogeographic provinces in the North Atlantic. Materials and methods Data Phytoplankton abundance data were obtained from the CPR project [26]. The CPR programme initiated by the Sir Alister Hardy Foundation for Ocean Science is the largest multi-decadal plankton-monitoring programme in the world [17]. The CPR sampling device is towed by ships of opportunity at their conventional operating speeds at a standard depth of 7 m. Seawater passes over a continuously moving band of silk mesh, which is stored in preservative for later analysis. Cells from approximately 3 m3 of seawater are collected for each sample, but the exact volume is unknown. CPR practice is to treat the data as abundances proportional to cell number density, but not to convert abundances to cells L–1 [17,27]. The silk mesh captures many larger phytoplankton species, predominantly diatoms and dinoflagellates, as well as many smaller phytoplankton with diameters as small as 10 µm. Phytoplankton identified during zooplankton enumeration are sometimes recorded as present without quantification, but we discarded these presence-only data. For details, see the review by Richardson et al. [17]. Although phytoplankton data were first collected in the 1930s, broad spatial coverage of the North Atlantic, particularly in the Northwest North Atlantic, began in 1960. Each observation was associated with a particular Longhurst province [28] using geographic boundaries provided by marineregions.org (http://marineregions.org/mrgid/22538, version 4, March 2010). We used only data from six Longhurst provinces: Atlantic Arctic (ARCT) and Boreal Polar (BPLR), North Atlantic Drift (NADR), Northeast Atlantic Shelves (NECS), Northwest Atlantic Shelves (NWCS), and Atlantic sub-Arctic (SARC). Data from ARCT and BPLR were combined into one region identified as ARCT due to data scarcity in these provinces separately (Fig 1). PPT PowerPoint slide PNG larger image TIFF original image Download: Fig 1. Map of five biogeographic regions and location of CPR samples used in this study. Thin lines mark the boundaries of the regions. Provinces are Atlantic Arctic (ARCT), Atlantic sub-Arctic (SARC), Northwest Atlantic Shelves (NWCS), North Atlantic Drift (NADR), and Northeast Atlantic Shelves (NECS). See S2 Fig. for sample size of diatom and dinoflagellate biomass data. https://doi.org/10.1371/journal.pone.0323675.g001 To calculate the community biomass for diatoms and dinoflagellates we summed the product of the CPR abundance for each taxon and the biomass for each taxon using an average cell volume for the taxon [3] and a cell volume to cell carbon model [12]. Data for some taxa were not reported by Barton et al. [3] but were collected contemporaneously using the same methods and are included here (S1 File). Separate allometric estimates of cell carbon as a function of cell size for diatoms and dinoflagellates (using the general protist equation) from Menden-Deuer and Lessard [12] were used. Absent an estimate of the size of Ceratium pavillardii we approximated its size as the mean log biomass of all dinoflagellates in the table. Sea surface temperature (SST) is thought to be linked indirectly through changes in physical, chemical, and biological conditions to the absolute and relative prevalence of diatoms and dinoflagellates, but SST were not recorded simultaneously by the CPR project. We used the Hadley Centre sea surface temperature data product (HadISST, https://www.metoffice.gov.uk/hadobs/hadisst/) at 1° x 1° and monthly resolution [29]. This region contains 521,704 observations of diatom and dinoflagellate species abundance with relatively few gaps in months or years (S1 Table, S1 Fig, S1 Text). At the fine spatial resolution corresponding to a single square of sampling silk, it was common for a single sample (square of silk) to have 0 observations of diatoms or dinoflagellates, making it impossible to compute a biomass ratio at this scale. Data were aggregated to total functional group biomass within bins that were 2.5° wide in latitude, spanned the full longitudinal extent of the five provinces under consideration at monthly time resolution, yielding 12,162 observations of diatom biomass and 9,630 observations of dinoflagellate biomass (S1 and S2 Tables). During model development we subdivided provinces into 2.5° wide bins by longitude as well as latitude, but initial analysis showed that there was no longitudinal trend within provinces so we aggregated data across the full range of longitudes within a province to obtain a better estimate of mean biomasses of each group and the diatom index. Even after aggregation some locations had zero biomass of either diatoms (589 or 5%) or dinoflagellates (3121 or 24%). Almost half (45%) of the zero dinoflagellate biomass data were in the northernmost provinces (ARCT, SARC), consistent with our expectation of lower dinoflagellate biomass in colder waters (S3 Table). The frequency of temporal aggregation is somewhat arbitrary; we chose monthly resolution as a compromise between weekly and seasonal to capture a large fraction of the interannual variation and resolve the relative timing of diatom and dinoflagellate blooms without highlighting high-frequency stochastic variation or subdividing the data too much. Zero observations are a challenge for log-normally distributed data such as biomass since the log transformation drops out these zeros [24]. We computed the fraction of the combined diatom and dinoflagellate biomass due to diatoms, but zero biomass for either functional group was a problem here as we modelled the logit or log-odds of this biomass ratio. Zero biomass for both groups was replaced by 50% of the minimum community biomass over all samples in the aggregated dataset for each functional group (0.51 and 5.8 mg C, respectively for diatoms and dinoflagellates) and the distribution of the resulting data was examined to ensure it was approximately Normal. This imputation mirrors the tradition of increasing counts of 0 abundance to 1 and incorporates knowledge from the whole dataset about the minimum detectable biomass of each group. Our final aggregated data contained 12,751 observations of diatom and dinoflagellate biomass over five biogeographic provinces (observations per province varied from 1784 to 3837) and 6 decades (1960–2017, observations per decade ranged from 1491 to 2494). The diatom biomass, dinoflagellate biomass, and logit of the proportion of total (diatom + dinoflagellate) biomass due to diatoms, were roughly Normally distributed for all biogeographic provinces under consideration (S2 Fig). Imputation of missing (i.e., zero) dinoflagellate biomass resulted in a spike in the biomass histogram which would likely have been dispersed over a larger range of biomass values if the observations had the resolution to resolve them. The histogram of the biomass ratio did not exhibit this spike because of the amount of variation in diatom biomass across samples with imputed dinoflagellate biomass. The speed of ships towing CPR samplers has changed over the years, somewhat affecting the volume of water collected per sample. As average ship speed has increased the volume of water filtered per sample has decreased, and thus we might anticipate a decrease in the apparent abundance. No significant correlation has been found between ship speed and the phytoplankton color index, which should be proportional to total phytoplankton biomass [27]. Our diatom index should be less sensitive to changing ship speed over time since it is a ratio of two quantities expected to be affected equally by changes in speed. Model specification Our goal was to quantify the long-term linear rate of change in diatom biomass, dinoflagellate biomass, and the proportion of total (diatom + dinoflagellate) biomass due to diatoms, which we call the diatom biomass index. While the change in these quantities may be nonlinear in time, a linear approximation is a good first approximation to detect the difference between a directional trend in time and no trend. In addition to any long-term trend, we anticipated residual variation in diatom and dinoflagellate biomass and the diatom index within years, across ocean biogeographic provinces, and with latitude within each province. The provinces subdivide the complexity of the North Atlantic, but each province spans a large region (Fig 1, S3 Table), and locations of some sampling routes have shifted within provinces over time. We used latitude and longitude as predictors during model development to account for spatial variation in sampling and oceanographic conditions within each province. Initial results indicated no trend in biomasses or the diatom index with longitude when longitude was included in each model, so we omitted it from our models. We evaluated two options for deviations from a linear temporal trend: a time-space model that allowed for shifts in biomass and the biomass ratio with each decade and a time-space-temperature model that modelled changes in biomasses and the biomass ratio attributable to a sea-surface temperature anomaly. Sea-surface temperature varies across provinces, with latitude, from month to month, and following a long-term trend. To avoid including correlated predictor variables in our model, we removed the spatial and temporal variation in temperature that matched the predictors already in the model (month, year, province, latitude) by computing the residuals of a linear model predicting temperature from month, year, province, and latitude. We call this residual variation in sea-surface temperature the temperature anomaly. We denoted the mean diatom and dinoflagellate biomass (µg C) by and , respectively, in province at latitude during month (January, February, …, December) of year . The diatom index, equal to the proportion of total (diatom + dinoflagellate) biomass due to diatoms, is . The bivariate datum is compositional as its components and are bounded by 0 and 1 and sum up to 1 [30]. Bivariate compositional data are usually analyzed on the logit scale, where the logit transformation maps the interval (0,1) into the entire real line, allowing for the subsequent unconstrained data to be analyzed using conventional statistical techniques. Since compositional data only carry relative information, we also modeled spatio-temporal variation in each functional group’s log biomass to gain insight into the magnitude and direction of biomass changes. Time-space-temperature models Let denote the logit-transformed diatom index in province A at latitude during month of year . We modeled as Normally distributed with mean (1) and variance , where is the effect of month m, is the latitudinal gradient, with denoting the average latitude in province is the inter-annual trend, are sea-surface temperature anomalies, and is the linear effect of SST anomaly. We also modeled the ln-transformed biomass of each functional group and total biomass as in Eq. (1). The SST anomaly was defined using a model for observed SST. Let denote the observed SST in province A at latitude during month of year .We modeled this temperature as Normally distributed with mean (2) and variance , where , and were the effect of month (January through December), the latitudinal gradient, and inter-annual (year-to-year) trend, respectively. The SST anomaly is the residual error after removing the effect of province, seasonal cycle, inter-annual trend, and latitude and was modeled as . The SST anomaly in Eq. (1) is the posterior median of this quantity. We fit this model to SST data that coincided with the CPR sampling locations to develop the SST anomaly for the biomass and biomass-ratio models. To enable a comparison between the province as a whole and the province as represented at CPR sampling locations, we also fit this model to SST data sampled throughout the entire biogeographic province at 1° x 1° spatial resolution, except for the ARCT province which had large areas that were ice covered and extended far beyond the area surveyed by the CPR. Time-space models To quantify decadal-scale nonlinearities in the temporal trend, we modeled each of these quantities (ln q, r, and p) as Normally distributed with variance and expected value µ depending only on time and space by dropping the temperature anomaly from Eq. (1) and adding an effect for each decade: (3) where is the effect of month , is the latitudinal gradient, with denoting the average latitude in province is the inter-annual trend, and is the effect of decade (1960, 1970, …, 2010) for province A. We completed the specification of our models with explicit statements of priors on all unknowns. For the parameters common to the Time-Space and Time-Space-Temperature models, we assumed exchangeable , , and prior distributions on the month effects , the inter-annual trends , and the latitudinal gradients , with independent inverse gamma (InvGamma(0.1,0.1)) priors on the hyper-parameters , , and , and placed Uniform (0,10) priors independently on the province-specific standard deviations . On the temperature anomaly effects exclusive to the Time-Space-Temperature model, we placed exchangeable priors, with hyper-parameter drawn from the InvGamma(0.1,0.1) distribution. For the temperature anomaly model we assumed exchangeable , , and prior distributions on the month effects, the inter-annual trends, and the latitudinal gradients, with independent InvGamma(0.1,0.1) priors on the hyper-parameters , , and , and placed Uniform (0,10) priors independently on the province-specific standard deviations . On the decadal effects exclusive to the Time-Space models, we assigned hierarchical priors implementing the Bayesian LASSO regularization [31,32], following the Extended Bayesian LASSO prior design [33–35], to selectively shrink the irrelevant effects toward zero. More specifically, we assumed that , , where , and . Model fitting and validation Since the joint posterior of the model parameters is not available in closed-form, we used Markov chain Monte-Carlo (MCMC) methods [36], implemented in OpenBUGS [37], to simulate from the posterior. We ran 10,000 iterations of three parallel Markov chains and discarded the first 4,000 iterations of each chain as burn-in period, thinning the remainder by a factor of 10. We assessed the convergence of the Markov chains informally through visual inspection of traceplots and autocorrelation plots. Model fits were assessed using posterior predictive checks (see S1 Text). The importance of predictor variables was assessed by omitting one predictor at a time and observing the decrease in the proportion of variance in the response explained by the model (Bayesian R2). We used posterior predictive model checking to validate our models, which is the standard approach with Bayesian modelling [38]. Our posterior predicive model checking compared observed data with data simulated from the model using histograms and summarized differences using Bayesian p-values (see S1 Text). Discussion There is accumulating evidence that climate warming over the last century is altering ocean conditions, primary production, chlorophyll concentration, and the biogeography of the key phytoplankton taxa [40–43]. Diatoms and dinoflagellates are two of the major phytoplankton functional groups in the North Atlantic and changes in their relative biomass may impact export production [6]. It has been hypothesized that climate warming will shift plankton communities towards dinoflagellates and away from diatoms, altering food web composition with potentially significant consequences for fisheries [19,44]. The CPR data have been used to investigate trends in the biomass of phytoplankton, diatom and dinoflagellate functional groups, and the relative abundances of diatoms and dinoflagellates on many occasions, but a lack of standardization in how the plankton community has been assessed across regions (e.g., subset of species examined, geographical boundaries, statistical framework, using sums of abundances vs. biomass) and temporal range analyzed makes rigorous North Atlantic regional-temporal comparisons challenging. Here we synthesize nearly 60 years of data (1960–2017) from the Continuous Plankton Recorder project spanning much of the northern North Atlantic divided into 5 provincial regions (Fig 1) to identify if there has been a long-term trend in diatom and dinoflagellate biomass and the proportion of their biomass contributed by diatoms (diatom index) and whether these changes vary regionally. Since the CPR samples the ocean at 7 m depth with a coarse mesh, our analysis was restricted to changes in the biomass of relatively large diatoms and dinoflagellates in the upper-mixed layer of the ocean. In contrast to expectations, we find there has been temporal increase in the diatom index over much of the North Atlantic, except the Arctic region, and the sum of the total biomass of the diatoms and dinoflagellates increases in the shelf regions but decreases in the central and sub-Arctic regions. Our analysis confirms that there are stark regional differences in how diatom and dinoflagellate biomass has changed in the North Atlantic since the 1960s (Fig 2). There were large increases in diatom biomass in the eastern and western shelf regions (NWCS, NECS), diatom biomass was essentially unchanged in the central North Atlantic (NADR), and decreased in the polar regions (SARC, ARCT). Dinoflagellate biomass increased in the NWCS, was unchanged in the ARCT, and decreased in the other three provinces. The decline in diatoms in the NADR, sub-Arctic and Arctic are broadly consistent with a global model of ocean biogeochemistry coupled to a climate model [45]. There is relatively little comparable observational data from the Arctic and North Atlantic Drift province, but Zhai et al. [20] found a decrease in diatom relative to diatom + dinoflagellate abundance in the ARCT and NADR over 1991–2009, consistent with our longer-term trend. While Zhai et al. [20] link these changes in community structure to the NAO, sea surface temperature and stratification, Hátún et al. [23] hypothesize decreases in dissolved silicate may be responsible. Also consistent with our observations: Osman et al. [46] found primary productivity and diatom and dinoflagellate abundance has been in decline in the sub-Arctic, Head and Sameoto [47] reported increases in both diatom and dinoflagellate abundance between 1962–1971 versus 2001–2003 on the Newfoundland shelf, eastern Scotian shelf and central/western Scotian shelf, and Johns et al. [39] found that dinoflagellate abundance has been increasing in the Grand Banks region in winter. There is a wealth of studies from the Northeast Coastal Shelf, particularly in the North Sea, but reported results are not unambiguous. For example, Leterme et al. [19] found a decline in diatom abundance in the NE and SE Atlantic especially since the late 1960s/early 1970s, Hinder et al. [21] found an increase in diatom relative to dinoflagellate abundance using a subset of diatoms species related to sea surface temperature and windy conditions, while Edwards et al. [48] found a north-south contrast, with an increase in diatom abundance with warming in the more northerly versus more southerly part of the region. It is difficult to compare the results across these studies and establish the extent of regional differences across the North Atlantic without a consistent analysis of the entire dataset. Our analysis identified distinct regional differences in the long-term temporal trend in diatom and dinoflagellate biomass by accounting for intra-annual, interannual, and spatial variability in the diatom index and diatom and dinoflagellate biomass. Our analysis captured the strong interannual trend, including expected differences in timing of peak biomass of both functional groups (diatoms before dinoflagellates), and delay in blooms further north (Fig 3). We evaluated two mechanisms to account for a non-linear trend in biomass and the biomass index with time: decadal variation and alternatively a temperature anomaly, finding that the temperature anomaly approach produced the most explanatory power and reduced uncertainty in the parameter estimates (S3-S5 Figs). Spatial variation was incorporated using the standardized Longhurst provinces and latitude to allow for variation within provinces, while still permitting the spatial aggregation necessary to observe a strong signal. The regular CPR routes have sometimes shifted northward or southward over time, leading to additional variation within provinces that can be partially attributed to the latitude of the observation (see Edwards et al. [48] for a recent detailed description of the CPR dataset). Regional differences in the response of diatom and dinoflagellate abundance have been correlated to climate indices such as the NAO and associated differences in hydroclimatic conditions such as wind, mixed layer depth and upper water column stratification, that in turn affect nutrients and conditions such as sea surface temperature and light, e.g., [19,21,48]. Interactions between environmental conditions and trophic transfer to zooplankton and fish may also be altering the timing and abundance of diatoms and dinoflagellate populations, through differential effects of temperature on phytoplankton versus zooplankton or trophic cascades [49–51]. Given there has been warming throughout most of the North Atlantic over the past 6 decades [52–54](S9 Fig) and a diversity of responses in diatom and dinoflagellate biomass across regions, there is clearly no universal link between the NAO or increases in sea surface temperature and the temporal trajectory of diatom or dinoflagellate biomass. Although there is no coherent response of diatom or dinoflagellate biomass with increasing sea surface temperatures in the North Atlantic since the 1960s, anomalous temperature increases, after accounting for province, latitude, month, and year, result in a lower diatom index (shift towards dinoflagellates) in all provinces except the NWCS (Fig 4). We speculate that since the NWCS community is already relatively enriched in dinoflagellates with low interannual variation, there is limited potential for further shifts towards dinoflagellates (Fig 3). This almost universal response may be a direct effect of increasing temperature on growth rates, leading to increased biomass. Indirect effects of temperature can be positive or negative through correlated increases in stratification and windiness, affecting turbulence and the availability of light and nutrients. Differential effects of temperature on grazers and their prey may alter food web structure, diatom and dinoflagellate biomass and the diatom index. Increases in the temperature anomaly shifts the community towards dinoflagellates, while the long-term trend with time (and increasing temperature) shifts the community towards diatoms in most regions. We hypothesize that these short-term variations in temperature are too rapid for diatoms and dinoflagellates to adapt and therefore reflect eco-physiological responses to the associated hydroclimatic changes. Independent analyses in other parts of the ocean and experimental evolution studies in the lab show that phytoplankton can adapt to changes in temperature over several tens to hundreds of generations [55–58]. Our inference is that plankton adapt evolutionarily to temperature change over decadal scales and to predict or interpret the effects of decades of temperature change requires a consideration of evolutionary as well as eco-physiological responses to environmental variables. Here we document a long-term shift towards diatoms and away from dinoflagellates, on a biomass basis, throughout the temperate and sub-polar North Atlantic. This shift towards diatoms was the result of both increasing and decreasing trends in diatom and dinoflagellate biomass, depending on the biogeographic province. This analysis reveals a signature of a large-scale restructuring of the North Atlantic phytoplankton community not typically anticipated as the consequences of climate change. To document this trend required decades of investment in consistent sampling. Data from the CPR plankton survey are as valuable as they are rare. Continued investment in these surveys is essential for developing an understanding of the ocean-climate system. Although we reported results for the Arctic province, we recognize a severe under-sampling of the spatial extent of this province and caution our results are only representative of the southernmost extent of the Arctic and Boreal province in the North Atlantic. In the other four provinces, we see good agreement in the rate of warming at CPR sampling sites and the whole province in the NWCS and NECS, but note that the average temperature at the sites sampled by the CPR in the SARC and NADR decreased over 6 decades, despite the average temperature in these provinces increasing (S8, S9 Figs). As a result, we caution that CPR sampling in SARC and NADR may not be representative of the provinces as a whole, so there is some risk in extrapolating observed changes at sampling sites to the whole province in these provinces. The hallmark of our analysis is that the trends are highly variable across provinces, reflecting the consequences of regional changes in climate (e.g., increased windiness in NECS) and differences in the baseline community composition across provinces. Notably, dinoflagellate biomass was much lower in the Arctic resulting in a higher diatom index and dinoflagellate biomass was only weakly seasonal in the NWCS (Fig 3), so it seems reasonable to anticipate distinctive responses to climate change in these provinces. Assuming the underlying mechanisms do not change, the next few decades could bring further decreases in diatom and dinoflagellate biomass, with a shift towards diatoms in much of the North Atlantic and a shift towards dinoflagellates in the Arctic. These changes have likely had notable consequences for carbon export and the amount of biomass transferred up the food web. Future climate change might have similar effects, although non-linear or unanticipated changes are certainly possible. Extrapolating the long-term effects of climate change from this analysis is challenging because of the opposing factors documented by the long-term change (Fig 2) and the temperature anomaly (Fig 4). The effect of a 1°C temperature anomaly on diatom and dinoflagellate biomass and the diatom index is approximately equal, but opposite, to several decades of long-term climatic change (Figs 2, 4). Changes to future carbon export and trophic transfer will depend on the relative importance of long-term trends and short-term variation arising from temperature change, sea-ice melt, changes in freshwater inputs, wind-driven mixing and circulation. Acknowledgments We thank the CPR survey team past and present. The survey depends on the voluntary co-operation of owners, masters and crews of merchant vessels which tow the CPRs on regular routes. HadISST data were obtained from https://www.metoffice.gov.uk/hadobs/hadisst/ and are British Crown Copyright, Met Office provided under a Non-Commercial Government Licence http://www.nationalarchives.gov.uk/doc/non-commercial-government-licence/version/2/.