Phytoplankton play a key role in biogeochemical cycling and climate processes. Precise quantitative measurements of chlorophyll-a (Chl-a), a measure of phytoplankton biomass, have only been available globally since 1997 from the Sea-viewing Wide Field-of-view Sensor (SeaWiFS). In the North Atlantic, semi-quantitative measurements of chlorophyll (Phytoplankton Color Index, PCI) for >50 years have been collected by the Continuous Plankton Recorder. Here we demonstrate a significant correlation between PCI and SeaWiFS Chl-a from 1997–2002. Combining both time series allows quantification of the stepwise increase in biomass in the mid-1980s; this regime shift corresponded to a 60% increase in Chl-a. This was a result of an 80% increase in Chl-a during winter, alongside a smaller summer increase. This new high-resolution data set on the monthly variation of Chl-a in the North Atlantic since 1948 is now available for the development and validation of climate models, and for interpretation of ecological changes related to climate.
 Phytoplankton produce >45% of the primary production of plants on Earth [Falkowski et al., 2004], absorb the greenhouse gas carbon dioxide (CO2) from the atmosphere, and contribute to the biological pump, which ensures that the climate of the world is much cooler than would otherwise be the case [Reid and Edwards, 2001]. Changes in phytoplankton composition and abundance may influence the biodiversity of other organisms such as zooplankton, fish, seabirds and marine mammals [Nybakken, 1997]. Despite the significant role of algal production in the oceans, the short time-series of large-scale chlorophyll patterns limits our understanding of the impact of global change on primary productivity and vice versa. Acquiring this information is essential for the further development of global climate change models.
 Here we investigate the potential relationship between Sea-viewing Wide Field-of-view Sensor (SeaWiFS) chlorophyll-a (Chl-a) measurements in the Central Northeast Atlantic and North Sea (1997–2002) and simultaneous in situ measurements of the Phytoplankton Color Index (PCI). This index of chlorophyll was collected by the Continuous Plankton Recorder (CPR) survey, which is an upper-layer plankton monitoring programme that has operated in the North Sea and North Atlantic Ocean since 1931 [Reid et al., 2003]. By combining data from both instruments, it is now possible to extend the SeaWiFS Chl-a data set back more than 50 years.
2.1. CPR Plankton Data
 For this study, PCI data were extracted from the CPR database for the Central Northeast Atlantic and North Sea. For the period 1948–2002, the CPR survey collected more than 94000 samples (Figure 1a). The methodology of sampling and measurement of PCI has remained consistent since 1948 [Reid et al., 2003]. Samples were collected by a high-speed plankton recorder (∼15–20 knots) that is towed behind ‘ships of opportunity’ in the surface layer of the ocean (∼10 m depth); one sample represents 18 km of tow. Accumulation of phytoplankton cells on the silk gives it a greenish color [Batten et al., 2003]. Phytoplankton biomass or PCI is based on a relative scale of greenness and determined on the silk by reference to a standard color chart. There are four different ‘greenness’ values: 0 (no greenness), 1 (very pale green), 2 (pale green) or 6.5 (green). Categories of PCI are assigned numerical values based on acetone extracts [Colebrook and Robinson, 1965]. PCI is a unique estimate of phytoplankton biomass, as small phytoplankton cells that cannot be counted under the microscope contribute to the coloration of the filtering silk [Batten et al., 2003].
2.2. Satellite Data
 SeaWiFS data were acquired from the NASA GES DAAC and processed using SeaDAS version 4.4. Data used were Level 3 daily products (9 × 9 km2 resolution) of the near-surface Chl-a concentration (mg m−3), estimated using the Ocean Chlorophyll 4 - version 4 (OC4-v4) algorithm [O'Reilly et al., 1998]:
Processing these data requires a series of radiometric corrections (e.g. atmospheric) to eliminate the presence of clouds, haze and water vapour [Mueller and Austin, 1995].
2.3. Data Analysis
 Sixty-four months (September 1997–December 2002) of in situ measurements of PCI and satellite Chl-a values were compared for the area of the Central Northeast Atlantic and North Sea. Concurrent SeaWiFS and CPR measurements were compared for the same spatial and temporal (daily) coverage. In the area of study, the CPR survey collected 11149 different samples for the 5-year period. After screening the SeaWiFS data set for CPR match-ups, only 1585 samples could be used for comparison (86.7% of data did not have a SeaWiFS match-up, primarily due to cloud coverage) (Figure 1b). PCI data is on a ratio scale (i.e. not only can PCI categories be ranked but differences are quantified). Thus, Pearson correlation (or linear regression) is appropriate to assess the strength of the relationship between SeaWiFS and PCI data (StatSoft, Inc., Electronic Statistics Textbook, http://www.statsoft.com/textbook/stathome.html). SeaWiFS data were log-transformed to improve homogeneity of variance and normality [Zar, 1984].
2.4. Potential Biases
 Consistency and comparability of the methodology used in the CPR survey has been studied in some depth [Reid et al., 2003]. Although standard methods have been used for more than 50 years in the survey, the PCI has been measured by a number of different analysts during this time. However, measuring greenness is a simple task that is typically undertaken by 2 to 3 people in a year, many of whom have done this work for more than a decade. As well as referring to a standard color chart, apprentices are trained in assessing PCI for a year before they undertake the task on their own.
 The study area includes both Case I and Case II waters. In optically-complex Case II waters, Chl-a can not readily be distinguished from particulate matter and/or yellow substances (dissolved organic matter) and so global chlorophyll algorithms (such as OC4-v4) are less reliable [International Ocean-Colour Coordinating Group, 2000]. As the majority of the area included in this study comprises Case I water this bias influences only a small proportion of the data points (Figure 1b).
3. Results and Discussion
 We first examined the overall relationship between PCI and SeaWiFS Chl-a for the whole study area (Figure 2a). There is a significant positive relationship (r = 0.33, p < 0.001). This relationship is confirmed and strengthened when spatial and temporal autocorrelation are considered (r = 0.47, F = 15.38, Adjusted df = 53, p = 0.0003). (We removed spatial autocorrelation by calculating the monthly average for the entire area of interest for PCI and matched SeaWiFS data. The Pearson correlation between these monthly time series was then calculated, and the degrees of freedom and thus the significance level of this test procedure were adjusted [Pyper and Peterman, 1998]).
 As the relationship between SeaWiFS Chl-a and PCI is non-linear, we calculated the mean of SeaWiFS Chl-a for each PCI category (Figure 2a). There is a relatively small variation in the confidence intervals of Chl-a for the first three PCI categories (no green (NG) = 1.03 ± 0.21 mg m−3, very pale green (VPG) = 1.65 ± 0.16 mg m−3, pale green (PG) = 2.61 ± 0.29 mg m−3 with maximum variation in the fourth category (green (G) = 4.25 ± 0.98 mg m−3). There is clear differentiation in mean Chl-a among PCI categories (95% confidence intervals do not overlap), so these values can be used retrospectively to estimate Chl-a from PCI values.
 To explore the seasonal patterns of PCI and SeaWiFS, we plotted the monthly means of both data sets (Figure 2b). Seasonal cycles for both show similar patterns, with a peak during late spring (spring bloom) and a decline during autumn and winter. The correlation coefficient implies a significant positive relationship (r = 0.79, p < 0.01). For all months except July, the 95% confidence intervals overlap, indicating good agreement between the two Chl-a measures. It is possible that an increase of dinoflagellates in CPR samples in summer may have contributed to the difference in July, as they are the dominant phytoplankton at this time and give a brownish color to the CPR silks and so could potentially lead to overestimates of Chl-a from PCI.
 Using the significant relationship between the PCI and Chl-a (Figure 2a) and the results of the >94000 CPR samples analysed in the period 1948–2002 (Figure 1a), a retrospective calculation of Chl-a averaged for the Central Northeast Atlantic and North Sea has been produced (Figure 3). While the changes shown have been demonstrated previously in a semi-quantitative manner for the PCI [Reid et al., 1998], the current results confirm and quantify the observations. An increasing trend is apparent in mean Chl-a for the area of study over the period 1948–2002 (Figure 3). There is clear evidence for a stepwise increase after the mid-1980s, with a minimum of 1.3 mg m−3 in 1950 and a peak annual mean of 2.1 mg m−3 in 1989 (62% increase). Post 1986 levels of Chl-a have increased systematically during winter (80% increase), with generally higher values in summer as well (Figure 3). The marked increase in chlorophyll seen in the mid 1980s is part of what has been termed a regime shift, a stepwise alteration in the composition and productivity of the whole ecosystem at a regional scale that reflects major hydrographic change [Beaugrand, 2004; Reid et al., 2001]. Changes through time in the PCI are significantly correlated with both sea surface temperature and Northern Hemisphere Temperature [Beaugrand and Reid, 2003]. A climate signal is strongly evident in all trophic levels of the marine system in the North Atlantic and North Sea, although the mechanisms underlying such relationships are not fully understood [Richardson and Schoeman, 2004].
 We can develop new insights into decadal changes in phytoplankton standing stock by combining data from SeaWiFS Chl-a and PCI, although each have strengths and acknowledged weaknesses. The strength of satellite remote sensing (e.g. SeaWiFS) is its ability to obtain information on phytoplankton distribution and abundance over large spatial scales. For instance, SeaWiFS has been used extensively to assess the role that global oceanic photosynthesis plays in climate and fisheries [McClain et al., 1998]. The PCI is an alternative way of assessing the major temporal and spatial patterns of phytoplankton biomass over almost 60 years in the North Atlantic [Colebrook and Robinson, 1965; Reid et al., 1998]. However, both the PCI and SeaWiFS have limitations. The PCI provides a visual (semi-quantitative) estimate of phytoplankton biomass, which has only previously been coarsely calibrated with chlorophyll acetone extracts [Colebrook and Robinson, 1965], and its coverage is restricted to shipping routes [Reid et al., 2003]. By contrast, SeaWiFS has limitations due to its limited lifespan, making it impossible to investigate decadal changes in phytoplankton. Problems with SeaWiFS data associated with restricted coverage due to clouds [McClain et al., 1998] are also highlighted in the present study where only 13% of the in situ PCI data could be used for comparison with SeaWiFS, and in a recent comparative study where only about 2–3% could be used [Hooker and McClain, 2000].
 The oceans are increasingly recognised as a key component of the climate system [Bigg et al., 2003] and have recently been shown to be the only true sink for anthropogenic CO2 over the last 200 years [Sabine et al., 2004]. The same study showed that this oceanic sink of the key greenhouse gas CO2 may well be declining, at the same time as the strength of the terrestrial biosphere sink remains constant. If true, this result implies that concentrations of atmospheric CO2 are likely to increase at a more rapid rate over the next 100 years than currently predicted. Primary production and phytoplankton composition play a key role in the modulation of radiatively-important gases such as CO2 and also produces reactive gases that contribute to the formation of clouds and effect albedo. Increasing levels of atmospheric CO2, which consequently causes significant changes in surface ocean pH and carbonate chemistry, impact phytoplankton with calcareous body parts such as coccolithophores [Riebesell et al., 2000]. Given this background and the fact that primary production is at present not included in global climate models emphasises the importance of obtaining appropriate temporal-spatial data on phytoplankton.
 Our results make available for the first time data on the monthly variation of plant biomass (Chl-a) in the NE Atlantic and North Sea since 1948. This now allows quantification of the stepwise increase in plant biomass in the mid-1980s; this regime shift corresponded to a 60% increase in Chl-a. This increase is mainly due to the 80% increase in Chl-a during winter since the mid-1980s, alongside a smaller increase during summer. This new chlorophyll data set (based on >94000 stations), along with physical, biological and chemical parameters, can now be assimilated into the next generation of climate models. This will not only open up new possibilities for modelling marine ecosystems on a regional and oceanic scale, but should also advance our understanding of biogeochemical cycling and improve our predictive capability of the impacts of climate change.
 We are grateful to present and past staff of SAHFOS who have contributed to the maintenance of the CPR time series. We also acknowledge the important co-operation received from agents, owners, masters and crew of the towing vessels. The survey is supported by a consortium comprising IOC, the European Commission, and agencies from Canada, France, Faroes, Iceland, Ireland, the Netherlands, Portugal, Spain, the United Kingdom and the USA. Finally we would like to acknowledge the assistance of Gregory Beaugrand, Christos Maravelias, Yaswant Pradhan, Anthony Walne and Helena Oikonomou. D. E. Raitsos is supported by a scholarship from the University of Plymouth. This study was also supported by the UK Natural Environment Research Council through the Atlantic Meridional Transect consortium (NER/O/S/2001/00680) and Centre for Observation of Air-Sea Interactions and Fluxes (CASIX). This is contribution number 89 of the AMT programme and number 34 for CASIX. The authors would like to thank the SeaWiFS Project (Code 970.2) and the Goddard Earth Sciences Data and Information Services Center/Distributed Active Archive Center (Code 902) at the Goddard Space Flight Center, Greenbelt, MD 20771 for the production and distribution of the SeaWiFS Level 3 data. These activities are sponsored by NASA's Earth Science Enterprise.