Remote Sensing of Environment 112 (2008) 3153–3159 Contents lists available at ScienceDirect Remote Sensing of Environment j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / r s e An absorption model to determine phytoplankton size classes from satellite ocean colour T. Hirata a,b,⁎, J. Aiken a,b, N. Hardman-Mountford a,b, T.J. Smyth a,b, R.G. Barlow c a b c Plymouth Marine Laboratory, Plymouth, UK Centre for observation of Air–Sea Interactions and fluXes (CASIX), Plymouth, UK Marine and Coastal Management, Cape Town, South Africa A R T I C L E I N F O Article history: Received 5 November 2007 Received in revised form 7 March 2008 Accepted 9 March 2008 Keywords: Absorption Phytoplankton Size, Ocean colour Remote sensing Pigment A B S T R A C T We have developed a model linking phytoplankton absorption to phytoplankton size classes (PSCs) that uses a single variable, the optical absorption by phytoplankton at 443 nm, aph(443), which can be derived from the inversion of ocean colour data. The model is based on the observation that the absolute value of aph(443) co-varies with the spectral slope of phytoplankton absorption in the range of 443–510 nm, which is also a characteristic of phytoplankton size classes. The model when used for analysis of SeaWiFS global data, showed that picoplankton dominated ~79.1% of surface waters, nanoplankton ~ 18.5% and microplankton the remainder (2.3%). The N. and S. Atlantic and the N. and S. Pacific Oceans showed seasonal cycles with both micro and nanoplankton increasing in spring and summer in each hemisphere, while picoplankton, dominant in the oligotrophic gyres, decreased in the summer. The PSCs derived from SeaWiFS data were verified by comparing contemporary 8-day composites with PSCs derived from in situ pigment data from quasiconcurrent Atlantic Meridional Transect cruises. © 2008 Elsevier Inc. All rights reserved. 1. Introduction Satellite remotely sensed observations of ocean colour have spatial– temporal resolutions that are appropriate for modelling global marine ecosystem dynamics. Presently chlorophyll-a (Chla) is the main biological variable that is derived operationally. The derivation of other biological variables (e.g. pigment composition) or other interpretations (e.g. phytoplankton taxa or species) could provide data to improve ecosystem models. There have been several empirical approaches that have used ocean colour data to derive phytoplankton functional types (size classes or representative species), most of which were linked to taxa-specific phytoplankton pigments (Ciotti et al., 2002; Sathyendranath et al., 2002; Alvain et al., 2005). Aiken et al. (2007) partitioned MERIS data of the Benguela ecosystem into size classes, microplankton (diatoms and dinoflagellates), nano-flagellates and picoplankton, using ranges of Chla and other bio-optical variables associated with each size class. Currently there is no systematic definition for phytoplankton functional types, but there are established links between phytoplankton size classes (PSCs) and major taxa (see Table 1). Recent oceanographic observations have shown links between optical properties, phytoplankton pigment composition and photosynthetic parameters (Aiken et al., 2004; Barlow et al., 2002, 2004; Moore et al., 2005; Fishwick et al., 2006). Besides the links between ⁎ Corresponding author. Plymouth Marine Laboratory Prospect Place, The Hoe, Plymouth, PL1 3DH, UK. E-mail address: tahi@pml.ac.uk (T. Hirata). 0034-4257/$ – see front matter © 2008 Elsevier Inc. All rights reserved. doi:10.1016/j.rse.2008.03.011 specific pigments and specific taxa (Jeffrey et al., 1997), distinctive biooptical characteristics have been observed for different taxonomic groups and PSCs. Notably, the Chla to accessory pigment ratio (Chla/ AP) was greatest for phytoplankton with highest Chla and the largest taxa (diatoms and dinoflagellates). Smaller taxa (flagellates) had lower Chla and Chla/AP. The smallest taxa (cyanobacteria and picoeukaryotes) had very low Chla and low Chla/AP. The photosynthetic quantum efficiency was positively correlated with Chla/AP and increased with phytoplankton size. The inference is that phytoplankton synthesise Chla preferentially when conditions are conducive to growth. These size, pigment and photosynthetic relationships are consistent with data from the Atlantic Meridional Transect (AMT, Aiken et al., 2000; Robinson et al., 2006) on: phytoplankton distributions (Zubkov et al., 1998; 2000; Tarran et al., 2006; Heywood et al., 2006); pigments (Gibb et al., 2000; Barlow et al., 2002, 2004; Aiken et al., in press); productivity (Maranon and Holligan, 1999; Maranon et al., 2000; Maranon, 2005; Poulton et al., 2006); photosynthetic activity (Suggett et al., 2006). Scaling of photosynthesis and cell size (Maranon et al., 2007; Maranon, 2008) shows a linear relationship over 9 orders of cell volume, from picoplankton to the large microplankton, is particularly significant in the context of the present research and by inference there should be a scaling of Cbiomass, Chla-biomass, photosynthesis and cell size. Chla has a distinctive blue absorption spectrum (400–470 nm, peak 443 nm), whereas carotenoids (both photosynthetic and photoprotective) have similar chemical structures, identical chromophores (similar absorption spectra) and are blue–green absorbers (400–550 nm, peak 3154 T. Hirata et al. / Remote Sensing of Environment 112 (2008) 3153–3159 Table 1 Taxonomic groups, marker pigments and size classes Taxonomic group Marker pigments (abbreviation) Size class Diatoms Dinoflagellates Flagellates Prymnesiophytes Chrysophytes Cryptophytes Chlorophytes Pico-flagellates or Pico-eukaryotes Cyanobacteria (prokaryotes; prok) Synechococcus spp. Prochlorococcus spp. Fucoxanthin (Fuc) Peridinin (Per); also Fuc Micro Micro 19′-Hexanoyloxyfucoxanthin (Hex) 19′-butanoyloxyfucoxanthin (But) Alloxanthin (Allo) Chlorophyll b (Chlb) (Hex, But, Allo, Chlb) Nano Nano Nano Nano Pico Zeaxanthin (Zea) Zeaxanthin (Zea) + DVChlb Pico Pico Pico Diagnostic pigments (DP) = 0.6Allo + 0.35But + 1.01Chlb + 1.41Fuc + 1.27Hex + 1.41Per + 0.86Zea. Micro (M) = 1.41(Fuc + Per)/DP. Nano (N) = (0.6Allo + 0.35But + 1.27Hex + 1.01Chlb)/DP. Prokaryotes (Cyanobacteria) = 0.86Zea/DP. PE = pico-eukaryotes (size b 2 μm, Chla b 0.25 mg m− 3). Pico (P) = Prok + PE. 490 nm). Phycobilins are major light-harvesting pigments in cyanobacteria (Synechococcus spp.) absorbing 580–630 nm, but occur in very low concentrations in the surface ocean. The peak of Chla absorption at 443 nm is the dominant influence on the shape and absolute magnitude of phytoplankton absorption spectra in the blue–green region (400–550 nm). This distinctive optical signature should be detectable in ocean colour data, as the basis of a model to classify PSCs. Here we explore the relationships between direct optical properties, phytoplankton absorption, aph(λ) and PSCs, using exclusively the NASA bio-optical data set NOMAD (Werdell & Bailey, 2006) that combines pigment and optical data acquired concurrently (contemporary and co-located) from widespread geographically distributed sources (new release 2007). NOMAD is a unique internationally accepted reference data set that is continuously expanding, benefiting the global bio-optical community. 1.1. Rationale for absorption model for phytoplankton size classes Fig. 1 shows phytoplankton absorption spectra, aph(λ), m− 1, for samples acquired in the Benguela (Fishwick et al., 2006) and AMT (Aiken et al., in press) for a range of Chla values and phytoplankton of different size classes: micro (diatoms and dinoflagellates); nano (several flagellate taxa); pico (prokaryotes and pico-eukaryotes). Fig. 1 inset shows the lower range of aph(λ) at expanded scale for nano and pico samples. All spectra peak at aph(443) and the blue– green difference δ ( = aph(443) − aph(510)) or the slope S ( = δ/|443–510|) have greatest values at largest aph(443) and largest Chla. Both δ and S decrease with decreasing aph(443) and Chla. Larger microplankton have the largest aph(443) and slope S, nanoplankton have mid-range aph(443) and S and picoplankton have the smallest aph(443) and S. This implies that the absolute magnitude of aph(443) or S, has a taxonomic link that can be used to classify phytoplankton by size. This conceptual model is unique in that it uses functions solely of aph, i.e. aph(443) (or S or δ), and aph is a physical observable, directly derivable from remotely sensed water-leaving radiance (Smyth et al., 2006). The slope S represents the spectral shape in terms of absolute difference, while the commonly used ratio, aph(443)/aph(510) is a measure of spectral shape in relative terms. controlled (QC) independently, by regression analysis of Chla vs AP, (as per Aiken et al., in press) and comparably by regression analysis of aph (443) vs aph(490); only a few ‘outlier’ data (N3 standard deviation) were deleted. The merged pigments and aph dataset that had been sampled and analysed separately by different processes were highly correlated (r2 =0.77), but had obvious ‘outlier’ data, possibly arising form small differences or errors of water sampling (timing) or data labelling (from different depths or stations). The merged data were quality controlled (de-spiked) by successively regressing aph and Chla, three times and deleting outliers (2 standard deviation) at each stage (final r2 =0.91). This reduced the data set to 256 stations, i.e. 84% of original merged data. Some miss-aligned aph and Chla data may remain in the merged dataset after QC, which adds noise to the inter-relationships and reduces the precision of the prediction from the final regression equations. NOMAD data were classified using the Diagnostic Pigment Analysis (DPA, Vidussi et al. 2000; revised by Uitz et al., 2006) based on marker pigments (MP, see Table 1), modified as described below. The major taxa were merged into size classes. Diatoms and dinoflagellates can not be separated absolutely, as Fuc (the MP for diatoms) often replaces Per as the major pigment in dinoflagellates. DPA is not definitive, but it is a useful approximation; its accuracy and utility is assessed by analyses of the NOMAD pigment data. Fig. 2 shows the fractional occurrence (MP/DP) of the 5 marker pigments (Zea, Fuc, Hex, Chlb, Per) in NOMAD, as a function of Chla, smoothed by a 5 point runningmean to reduce the remnant ‘spikiness’; the distributions of the MPs relative to aph(443) are similar, though more ‘spiky’. The occurrences of Allo and But (not shown) are very low (~0.02). The dominance of Hex in NOMAD suggests that most nanoplankton are prymnesiophytes and that chrysophytes and cryptophytes are not abundant in the global ocean. Fig. 2a shows that there is a step drop in the occurrence of Zea at ~ 0.25 mg m− 3, signifying the upper limit of prokaryote dominance (major picoplankton component). Fuc, the MP for microplankton, is distributed across the full range of Chla, partly because of its co-occurrence in nano-flagellates as the pre-cursor pigment for Hex and But, giving a potential anomaly in the DPA. There is a step increase of Fuc to N0.4 at Chla ~0.95 mg m− 3 and to ~0.6 at Chla N 1.8 mg m− 3. The abundance of Per (Fig. 2b) is generally low (b0.1) and very low (b0.03) for Chla b 0.25 mg m− 3. Hex is abundant (0.2–0.3) for Chla b 0.25 mg m− 3, indicating that these data probably correspond to small flagellates (b2 μm, e.g. prymnesiophytes); i.e. pico-eukaryotes as reported by Not et al. (2004); Fuller et al. (2006); Tarran et al. (2006). Supporting evidence comes from pigment 2. Data and methods A sub-set of NOMAD comprising contemporary, co-located aph data at key wavelengths and the key pigments, reduced the data total from N1000 to 306 stations. Pigments and aph data were quality Fig. 1. Phytoplankton absorption spectra for a range of Chla (24.6, 18.9, 13.0, 1.91, 0.68, 0.21 mg m− 3) and taxonomic size classes (pico, nano and micro) with decreasing slope from high to low aph(λ) and Chla; inset spectra of pico and nanoplankton at expanded range. T. Hirata et al. / Remote Sensing of Environment 112 (2008) 3153–3159 3155 Fig. 2. Fractional occurrence of major marker pigments (MP) in NOMAD over aph(443) range: a) Zea (MP for prok) and Fuc (MP for diatoms/microplankton); b) Hex (MP for prymnesiophytes), Chlb (MP for chlorophytes) and Per (MP for dinoflagellates sometimes); c) Hex/Fuc ratio. Data were smoothed with 5 point moving average. analyses for 15 AMT cruises (Aiken et al., in press) showing that the transition out of the sub-tropical gyres into temperate provinces was marked consistently by a step increase of Chla to N0.25 mg m− 3 with a coincident change of the dominant PSC from picoplankton (in gyre) to nanoplankton (out), confirmed by phytoplankton species counts (unpublished data). Hex is most abundant (0.3–0.4) in mid-range (N0.25 mg m− 3), with a step down at 1.8 mg m− 3 coincident with a step increases of Fuc; there is other step down at 0.95 mg m− 3. The widespread distribution of Chlb shows that it is not exclusively the MP for pico-eukaryotes (Chla b 0.25 mg m− 3); the step increase (to N0.20) at Chla N0.25 mg m− 3 arises from the occurrence of Chlb in larger eukaryotes e.g. chlorophytes (nanoplankton). The Hex/Fuc ratio (Fig. 2c, near-equivalent to the nano/micro ratio) has a general decline across the full Chla range, with a step-decrease at Chla ~0.95 and ~1.8 mg m− 3 corresponding to changes in the fractions of nano and microplankton in this “mixed zone”. The Hex/Fuc ratio is ~ 3.0 ± 0.7, in the picoplankton domain (Chla b0.25 mg m− 3), ~1.5 ± 0.5, in the mid-range (0.25–0.95 mg m− 3) with a step down to b0.5 (mean 0.20 ± 0.17) for Chla N 0.95 mg m− 3. These step changes in the MP/DP ratio are indicative of step changes of PSCs and phytoplankton community structure, which could be used for sub-partitioning (estimating fractions of nano and micro in mixed zones) though account must be taken of the likely overestimation of microplankton (MP Fuc). Fuc is totally dominant at Chla N3.0 mg m− 3, indicative of ‘pure’ microplankton. Here DPA has been modified to reduce the anomalies identified above. Firstly, Chlb (MP for chlorophytes) is included in the nanoplankton class, as it is most abundant at Chla N0.25 mg m− 3 and is a minor pigment at lower Chla. Secondly, all samples with Chla b0.25 mg m− 3 are defined as picoplankton, comprising prokaryotes and pico-flagellates designated pico-eukaryotes. Thirdly, the PSC is “dominant” if the MP/DP ratio is N0.45 rather than N0.5; this minimises the number of samples diagnosed as “mixed” (i.e. no size class N0.5). SeaWiFS monthly composite data of Level 3 mapped normalised water-leaving radiance (412, 443, 490, 510, 555, 670 nm) for 2004 were obtained from NASA Godard Space Flight Centre. The ocean colour inversion model (Smyth et al., 2006), was used to compute aph (443), released through International Ocean Colour Coordinating Group http://www.ioccg.org/groups/software.html. The original data had the nominal spatial resolution of 9 × 9 km at the equator, but were re-sized to the reduced resolution of 18 × 18 km globally. For model validation, 9 × 9 km 8-day composite images of the normalised waterleaving radiance (and Chla) were matched-up with AMT station positions, where in situ pigment data were collected. Resultant matchup image data have the maximum gap of 4 days from the station observation date. 3. Results 3.1. Analyses of NOMAD The relationship between aph(443) and S (from NOMAD) is a linear function that increases monotonically, meaning that the value of aph (443) is representative of S and vice versa: S ¼ 0:0082 aph ð443Þ þ 0:00002 r 2 ¼ 0:984 Clearly aph(443) and S have some degree of auto-correlation, which applies to all aph at all wavelengths and also aph(443) with ∫ aph(λ) dλ. Fig. 3 shows the log-transformed data, marked for the dominant PSC 3156 T. Hirata et al. / Remote Sensing of Environment 112 (2008) 3153–3159 Fig. 4 shows the log-transformed data marked for the dominant PSC, with the numbers of each PSC in each domain listed in the boxes with values for lower and higher limits for the nano–micro boundary limits. For the higher limit, the number of micro in the micro-domain are reduced (from 49 to 28) and the number of nano reduced (from 29 to 3), clearly a higher fraction of microplankton. In the nano-domain, the lower limit makes nano the more dominant PSC (64 nano to 29 micro) and the higher limit makes the domain more mixed (83 nano to 50 micro). The Chla to S relationships for the three size domains are: Spico ¼ 0:00016 log10 ½Chla þ 0:00029 r 2 ¼ 0:46 ; ChlaV0:025mg m3 ; Snano ¼ 0:00049 log10 ½Chla þ 0:0044 r 2 ¼ 0:32 ; Chla 0:25 toV1:8mg m3 ; Smicro ¼ 0:0011 log10 ½Chla þ 0:00036 r2 ¼ 0:32 ; ChlaNto 1:8 mg m3 Fig. 3. Observed relationship (from NOMAD) between aph(443) and the spectral slope S, marked with the phytoplankton size classes and nominal boundaries for pico at 0.023 and nano–micro at 0.046/0.069 m− 1 (lower/higher limits); upper curve, nanoplankton data displaced by +0.0004, pico-eukaryotes displaced by +0.0003; lower curve, prokaryote data displaced by − 0.0004. Boxes indicate the number of each size class in each domain for lower (L) and higher (H) nano–micro boundary limits. using the modified DPA. Picoplankton are most abundant at low aph (443) ≤ 0.023 ± 0.002, nanoplankton most abundant in mid-range and microplankton most abundant at aph(443) N 0.069. The partition between the picoplankton domain and larger PSCs is relatively sharp, but the partition between nano and microplankton classes overlaps in the aph(443) range 0.046–0.069 m− 1. The lower curve shows the prokaryote data displaced for clarity. The upper curve has the nanoplankton dominated data displaced, showing the overlap with the pico and microplankton domains across the nominal boundaries at 0.023 and 0.069 m− 1. We designated all flagellates (nanoplankton) with Chla b0.25 mg m− 3 as pico-eukaryotes; these are marked as squares in Fig. 3. Microplankton are dominant at high aph but co-exist with nanoplankton across the nano–micro domains, albeit overestimated (10–25%), by the co-occurrence of Fuc (MP for micro) in nanoplankton. The numbers of each PSC in each domain are listed in the boxes in Fig. 3, with values for both lower and higher limits for the nano–micro boundary. In the micro-domain with the higher limit, the number of micro are reduced (from 54 to 28) and the number of nano reduced (from 39 to 12), overall a higher fraction of microplankton. In the nano-domain, the lower limit makes nano more dominant (40 nano and 18 micro) and the higher limit makes the domain more mixed (67 nano and 44 micro). The relationship can be partitioned into 3 domains for each of the 3 dominant PSCs, according to the value of S or aph(443): The partitioning on the basis of Chla is better in number terms (fewer micro in the nano-domain), possibly because the integral pigment dataset is less ‘spiky’ than the merged aph and Chla dataset and it is the pigment data that are used to determine the dominant PSC. The Chla regressions have lower r2 than the aph(443) data, possibly due to lower precision of the pigment data and no autocorrelation; the data for S from aph and the size classification from pigments are the same for both analyses, so the increased variance must arise from the Chla data. These Chla-relationships could be used to partition ocean colour data (SeaWiFS, MODIS, MERIS) into PSCs (as per Aiken et al. 2007), but the standard error for prediction would be larger than for aph. 3.2. Analyses of SeaWiFS data Fig. 5 shows the global maps of PSCs (micro, nano and picoplankton) derived from SeaWiFS data in 2004, using aph(443) thresholds at ≤0.023 and N0.069 m− 1. Microplankton were dominant in coastal upwelling zones (e.g. Benguela) and temperate seasonally stratified regions (e.g. North Atlantic spring and summer). Nanoplankton were dominant at mid-latitudes after the spring bloom and in equatorial Spico ¼ 0:00027 log10 aph ð443Þ 2 þ 0:00064 r ¼ 0:89 ; aph ð443ÞV0:023m1 Snano ¼ 0:00076 log10 aph ð443Þ 2 þ 0:0014 r ¼ 0:88 ; aph ð443ÞN0:023 toV0:069m1 Smicro ¼ 0:0023 log10 aph ð443Þ þ 0:0032 r 2 ¼ 0:93 ; aph ð443ÞN0:069m1 The relationship between Chla (mg m− 3) and S (from NOMAD) is also a linear function, though the fraction of variance explained is less than for aph(443) and S, with no auto-correlation: S ¼ 0:00021 Chla þ 0:00017 r 2 ¼ 0:667 Fig. 4. Observed relationship (from NOMAD) between Chla and the spectral slope S, marked with the dominant phytoplankton size classes and nominal boundaries for pico at 0.25 and nano–micro at 0.95/1.8 mg m− 3 (lower/higher limits); upper curve, nanoplankton data displaced by +0.0006; lower curve, prokaryote data displaced by −0.0005. Boxes indicate the number of each size class in each domain for lower (L) and higher (H) nano–micro boundary limits. T. Hirata et al. / Remote Sensing of Environment 112 (2008) 3153–3159 3157 Fig. 5. Global distribution of PSCs from SeaWiFS for 2004: a) Jan; b) Mar; c) May; d) July; e) Sep; f) Nov. Red, microplankton dominant; green, nanoplankton dominant; blue, picoplankton dominant. regions. Picoplankton were dominant in equatorial zones, tropical and sub-tropical ocean gyres. Table 2 shows the monthly occurrence of pico, nano and microplankton in the global ocean, Indian, North and South Atlantic, North and South Pacific oceans, from Jan to Dec 2004. The occurrence of microplankton in the global oceans was 1.9–2.8% (annual mean 2.3%), while that of picoplankton ranged from 77.7 to 80.5% (mean 79.1%). Nanoplankton occurrence was 17.3–19.6 % (mean 18.5%). The N. and S. Atlantic and the N. and S. Pacific oceans showed seasonal cycles with both micro and nanoplankton increasing in spring and summer in each hemisphere, and picoplankton decreasing. The S. Pacific had the largest fraction of picoplankton (annual mean 83.4%) and the least microplankton (0.6%). The Indian Ocean shows seasonal features associated with the SW monsoon in Sep and NE monsoon in Jan and Feb. The mean occurrence for each size class have been converted to mean fraction of Chla (lower panel of Table 2), using mean factors (mg m− 3) calculated from NOMAD: pico (b0.25) = 0.13; nano (0.25–1.8) = 0.72; micro (N1.8) = 3.31 mg m− 3. These factors are comparable to values derived from AMT data (Aiken et al., in press). Table 2 Dominance of PSCs (M, N, P) by sea surface area (%) from SeaWiFS data 2004; cloud-covered or other invalid pixels are ignored in the calculation Global Month Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Mean Chla M 2.3 2.1 2.0 2.0 2.9 2.8 2.4 2.6 2.9 2.4 1.9 1.9 2.3 22.2 Indian Ocean N 19.0 17.9 17.6 18.7 18.6 17.3 17.1 18.9 19.5 18.9 19.1 19.6 18.5 45.3 P 78.6 79.9 80.4 79.3 78.5 79.8 80.5 78.5 77.6 78.7 79.0 78.5 79.1 32.6 M 1.5 1.2 0.7 0.5 0.5 0.5 0.6 0.7 1.7 0.8 0.7 0.8 0.9 9.2 N 21.0 17.0 11.7 11.7 15.3 17.8 19.9 21.1 21.9 22.0 24.4 22.9 18.9 52.9 N. Atlantic P 77.5 81.9 87.6 87.8 84.1 81.6 79.5 78.2 76.3 77.2 74.9 76.3 80.2 37.9 M 2.8 3.2 3.8 5.0 8.1 8.1 5.7 5.2 4.8 4.1 3.3 2.3 4.7 36.6 N 13.3 14.8 20.9 27.5 27.0 22.6 19.9 18.7 19.2 15.5 12.4 10.8 18.5 37.3 S. Atlantic P 83.8 82.0 75.3 67.6 64.9 69.3 74.4 76.2 76.0 80.4 84.3 87.0 76.7 26.1 M 5.5 4.3 3.3 1.8 1.8 2.3 2.7 2.0 1.5 1.9 2.6 4.4 2.8 21.5 N 30.5 27.5 27.2 26.0 26.5 23.0 22.5 27.1 29.8 36.1 33.2 30.0 28.3 55.6 N. Pacific P 64.0 68.1 69.5 72.1 71.7 74.8 74.8 70.9 68.7 62.0 64.2 65.6 68.9 22.8 M 2.4 2.5 2.7 3.2 4.6 4.4 3.7 4.0 4.7 3.3 2.4 2.0 3.3 29.1 S. Pacific N 16.6 18.9 19.4 21.6 20.9 19.0 17.5 19.0 19.5 16.6 14.1 13.7 18.1 41.0 P 81.1 78.6 77.7 74.7 73.4 75.6 78.3 76.2 74.7 78.8 82.9 84.0 78.0 29.8 M 1.1 0.8 0.7 0.5 0.4 0.5 0.5 0.5 0.4 0.4 0.6 0.8 0.6 7.3 N 17.6 15.2 15.3 15.6 14.0 12.9 13.9 5.3 16.0 15.0 17.0 21.3 15.8 49.0 P 80.9 83.5 83.7 83.7 85.2 86.6 85.6 84.1 83.6 84.3 82.2 77.6 83.4 43.7 The bottom rows show the mean annual occurrences of each size class by area (%) and mean annual Chla-biomass (%) using conversion factors derived from NOMAD: see text. 3158 T. Hirata et al. / Remote Sensing of Environment 112 (2008) 3153–3159 3.3. Preliminary verification of PSC aph-model PSCs derived by the aph-model from SeaWiFS data are compared with PSCs derived by DPA from pigment data acquired on AMT cruises (Aiken et al., in press) as a preliminary verification, illustrating the difficulties and the need for a comprehensive global validation. This comparison is not a true independent validation, since DPA was used to develop the aph-model and PSCs derived from pigments are indicative and not definitive. Table 3 shows results from AMT-07 (26 stations), UK to Uruguay (19 Sept to 16 Oct 1998) and contemporary SeaWiFS data (analysed for aph(443), Smyth et al., 2006) along the cruise track (47° N to 37° S); 3 stations had no valid SeaWiFS Chla and are suspect. The diagnoses showed good agreement (verification score: match = 2; near match = 1; no match = 0; suspect = S). In the gyres out of 18 stations (16 valid) all picoplankton, 15 scored 2 and 1 scored 1 about 94% success; in mesotrophic and eutrophic zones out of 8 stations (7 valid), 4 scored 2, 1 scored 1 and 3 scored 0, about 56% success. Comparison of PSC diagnoses for other AMT cruises, with contemporary SeaWiFS data showed practically 100% agreement for picoplankton in the gyres and ~50% elsewhere. The good results for pico can be explained by the low heterogeneity in the gyres; the poorer results elsewhere arise from the increased mesoscale variability that renders 8-day mean values of aph or Chla unsuitable for validation in dynamic zones. SeaWiFS protocols require in situ data acquisition within ± 1 h of overpass but this is unacceptably long in dynamic zones; Aiken et al. (2007) suggested that ± 10 min was needed in the Benguela upwelling ecosystem. Regional validation of in situ and satellite-derived PSCs with contemporary imagery is needed also. A preliminary analysis of Benguela data (Fishwick et al., 2006) using aph(443) data, derived PSCs comparable to the Chla analysis (Aiken et al., 2007) despite the extreme heterogeneity of the area. 4. Discussion Recent basin-scale observations from the Atlantic Meridional Transect (AMT) provide validation of the satellite-derived PSC distributions a priori. Notably, Zubkov et al. (1998, 2000), Heywood et al. (2006), Tarran et al. (2006) showed that phytoplankton in the tropical and sub-tropical gyres in the Atlantic Ocean were predominantly picoplankton (prokaryotes and pico-eukaryotes), quantitatively determined by flow-cytometry for cell counts and C-biomass equivalents. As shown by Aiken et al. (in press), the transition out of the sub-tropical gyres into the temperate provinces was sharp in terms of Chla (N0.25 mg m− 3) with phytoplankton types switching from picoplankton to nano-flagellates, confirmed by phytoplankton species counts. Microscope phytoplankton species counts for AMT-01 to -08 cruises have been converted to C-biomass for larger phytoplankton (nano-flagellates, diatoms, dinoflagellates), but counts for picoplankton (prokaryotes, pico-eukaryotes) are qualitative, due to inherent sampling and storage problems. Merging Flow Cytometry (some but not all cruises) and microscopic counts over the whole size range requires careful inter-calibration and is not complete. The classification of the dominant PSC (N0.45 MP/DP ratio) does not mean that only a single PSC is present, and usually a mixture occurs. Ambiguity may arise from errors in DPA and uncertainty for the threshold values of aph(443) in the regions of overlap and coexistence of the nano and micro PSCs. Additionally, Fuc (MP for micro) occurs in most nanoplankton and results in an overestimation of the micro class, with most impact in the nano-domain (Chla 0.25–1.8). The changes of the Hex/Fuc ratio may offer a method (mixing model) for estimating the fraction of nano and micro PSCs in this very mixed region. For a crude model, the value of the Hex/Fuc ratio for picoeukaryotes (zero microplankton) is the lower end-point for ‘pure’ flagellates and the ratio for Chla N 3.0 mg m− 3 is the upper end-point for ‘pure’ microplankton. Refinement of this model is needed and true validation (against phytoplankton species counts) would be more demanding than for the present analyses. Similar approaches could be used to partition the pico class into pico-eukaryotes and prokaryotes and the latter into Synechococcus spp. and Prochlorococcus spp. DPA is an approximate method but there is scope for enhancing the analyses by combining with backscatter characteristics of different phytoplankton groups (e.g. Aiken et al., 2007). Table 3 Comparison of PSCs from pigments for AMT-07 (Sept–Oct 1998) and contemporary SeaWiFS 8-day composite data using aph-model St. Lat. Long. SDY Chla Micro Nano or [PE] Prok Pico PSC DayDif ChlaDif SPSC Score A702 A704 A706 A709 A711 A712 A715 A717 A719 A721 A724 A726 A728 A731 A733 A735 A737 A739 A742 A744 A746 A748 A750 A753 A756 A757 46.61 41.63 38.78 36.31 32.55 30.06 25.82 22.02 17.69 14.59 14.39 12.18 08.08 04.28 00.41 −03.99 −08.04 −11.92 − 15.94 − 19.60 − 23.49 − 26.48 − 29.60 − 32.74 −35.77 −37.73 −10.80 −10.80 −13.60 − 17.49 −16.95 −19.94 −19.98 −19.99 −20.00 −17.76 − 17.73 − 21.32 −22.20 −23.75 −25.32 −27.09 −28.75 −30.32 −32.01 −34.94 −36.90 −39.90 −43.23 −46.66 −49.65 −51.96 259 260 265 266 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 0.28 0.28 0.08 0.05 0.04 0.08 0.08 0.21 1.24 0.56 0.26 1.06 0.22 0.15 0.16 0.11 0.06 0.06 0.03 0.06 0.04 0.13 0.16 0.07 0.34 0.27 23.8 19.8 0.0 0.0 0.0 0.0 0.0 0.0 0.0 8.1 11.5 26.8 8.8 11.6 0.0 0.0 0.0 8.6 0.0 17.3 13.6 16.7 23.6 15.1 48.9 16.1 47.1 30.6 [53.2] [27.5] [44.0] [42.5] [36.8] [68.3] 86.5 59.1 40.6 63.7 [32.1] [30.8] [34.8] [23.4] [32.3] [31.9] [31.9] [29.8] [17.8] [28.3] [29.9] [30.5] 44.4 79.1 29.0 49.4 46.7 72.4 55.9 57.4 63.1 31.6 13.4 32.7 47.7 9.4 59.0 57.4 65.1 76.5 67.6 59.4 68.0 52.8 68.5 54.9 46.4 54.2 6.6 4.6 29.0 49.4 100 100 100 100 100 100 13.4 32.7 47.7 9.4 91.1 88.2 100 100 100 91.3 100 82.6 86.3 83.2 76.3 84.7 6.6 4.6 N P P P P P P P N N P N P P P P P P P P P P P P M N −2 −1 −4 −3 −1 0 +1 +2 +3 −4 −3 −2 −1 0 +1 +2 +3 −4 −3 −2 −1 0 +1 +2 +3 +4 −0.05 0.03 −0.02 −0.03 S 0.0 −0.02 S S 0.21 0.03 0.65 0.08 0.01 0.01 0.00 −0.04 0.02 0.00 −0.01 −0.07 −0.02 0.07 −0.04 −0.01 −0.09 N P P P N P P N P P P P P P P P P P P P P N P P N N 2 2 2 2 1 2 2 0 0 0 2 0 2 2 2 2 2 2 2 2 2 0 2 2 1 2 Station No.; position; serial day year (SDY); in situ Chla (mg m− 3); DPA (% micro, nano, prok, pico); nano = PE if Chla b0.25; dominant PSC(M = micro, N = nano, P = pico = prok + PE); DayDif = SDY — (day 5 of SeaWiFS 8 day composite); ChlaDif = Chla — SeaWiFS Chla (mg m− 3); S = SeaWiFS suspect; SPSC (= SeaWiFS PSC); Score: match = 2, near match = 1, no match = 0. T. Hirata et al. / Remote Sensing of Environment 112 (2008) 3153–3159 NOMAD comprises a global data set of phytoplankton bio-optical properties. Other data sets from the English Channel in 2001 (Aiken et al., 2004) and other years (2003–2007), the Atlantic Ocean (Barlow et al., 2004), N. Atlantic (Moore et al., 2005) and Benguela (Fishwick et al., 2006) overlie NOMAD the S vs. aph(443) function precisely, concurring with NOMAD analyses and providing assurance of the aphmodel for diverse regional ecosystems. We demonstrate that the absolute magnitude of aph(443) provides a practical method for the analysis of ocean colour data because: 1) satellite ocean colour provides aph(λ) at discrete wavelength bands in the blue–green spectral region (443, 490, 510 nm) that can be retrieved with precision; 2) the model uses only one variable aph(443), so the number of error sources are minimal; 3) implementation is simple so large data sets (e.g. complete SeaWiFS, MODIS, MERIS or merged data) can be processed quickly, which is a priority activity, following comprehensive validation of the model. This study has shown that phytoplankton size classes can be derived from remotely sensed ocean colour data by an absorption model or a Chla model. From the remote sensing perspective, the aph-model is preferred, as aph(443) is derivable analytically from normalised water-leaving radiance, a direct physical observable of natural waters, whereas Chla is a biogeochemical variable, usually approximated by an empirical band-ratio algorithm that minimises the ‘bias’ over the full range of observations, but does not minimise the standard error of prediction. PSCs, C-biomass, pigments, productivity or other variables can be derived from aph. Marra et al. (2007) model production from aph(λ), showing that production for different PSCs could be derived from aph (443). Acknowledgments The authors acknowledge NASA SeaWiFS project team for satellite data and all contributors to the NOMAD data set for in situ bio-optical data. We commend NASA for the excellence of NOMAD and encourage continued enlargement and improvement of the dataset. This work is funded by Natural Environment Research Council, UK, through the Centre for observations of Air–Sea Interactions and fluXes (CASIX: publication number 52) and Plymouth Marine laboratory and was supported by AMT and MarQUEST projects (AMT publication number 164). References Aiken, J., Rees, N., Hooker, S., Holligan, P., Bale, A., Robins, D., Moore, G., et al. (2000). The Atlantic Meridional Transect: Overview and synthesis of data. Progress in Oceanography, 45, 257−312. Aiken, J., Fishwick, J. R., Moore, G. F., & Pemberton, K. (2004). The annual cycle of phytoplankton photosynthetic quantum efficiency, pigment composition and optical properties in the western English Channel. Journal of the Marine Biological Association in UK, 84, 301−303. Aiken, J., Fishwick, J. R., Lavender, S., Barlow, R., Moore, G. F., Heather, S., et al. (2007). Validation of MERIS reflectance and chlorophyll during the BENCAL cruise October, 2002: Preliminary validation of new demonstration products for phytoplankton functional types and photosynthetic parameters. International Journal of Remote Sensing, 28, 497−516. Aiken, J., Pradhan, Y., Barlow, R. G., Lavender, S., Poulton, A., Holligan, P. M., et al., (in press). Phytoplankton pigments and functional types in the Atlantic Ocean: A decadal assessment, 1995–2005. Deep-Sea Reserch., AMT Special Issue. Alvain, S., Moulin, C., Dandonneau, Y., & Breon, F. M. (2005). Remote sensing of phytoplankton groups in case 1 waters from global SeaWiFS imagery. Deep-Sea Research I, 52, 1989−2004. Barlow, R. G., Aiken, J., Holligan, P. M., Cummings, D. G., Mariotena, S., & Hooker, S. (2002). Phytoplankton pigment and absorption characteristics along meridional transects in the Atlantic Ocean. Deep-Sea Research I, 49, 637−660. Barlow, R. G., Aiken, J., Moore, G. F., Holligan, P. M., & Lavender, S. (2004). Pigment adaptations in surface phytoplankton along the eastern boundary of the Atlantic Ocean. Marine Ecology. Progress Series, 281, 13−26. 3159 Ciotti, A. M., Marlon, M. R., & Cullen, J. J. (2002). Assessment of the relationship between dominant cell size in natural phytoplankton communities and the spectral shape of the absorption coefficient. Limnology and Oceanography, 47, 404−417. Fishwick, J. R., Aiken, J., Barlow, R. G., Sessions, H., Bernard, S., & Ras, J. (2006). Functional relationships and bio-optical properties derived from phytoplankton pigments, optical and photosynthetic parameters; a case study of the Benguela ecosystem. Journal of the Marine Biological Association in UK, 86, 1267−1280. Fuller, N. J., Tarran, G. A., Cummings, D. G., Woodward, E. M. S., Orcutt, K. M., Yallop, M., et al. (2006). Molecular analysis of photosynthetic picoeukaryote community structure along an Arabian Sea transect. Limnology and Oceanography, 51, 2502−2514. Gibb, S. W., Barlow, R. G., Cummings, D. G., Rees, N. W., Trees, C. C., Holligan, P. M., et al. (2000). Surface phytoplankton pigment distributions in the Atlantic Ocean: An assessment of basin scale variability between 50 N and 50 S. Progress in Oceanography, 45, 339−368. Heywood, J. L., Zubkov, M. V., Tarran, G. A., Fuchs, B. M., & Holligan, P. M. (2006). Prokaryoplankton standing stocks in oligotrophic gyre and equatorial provinces of the Atlantic Ocean: Evaluation of inter-annual variability. Deep-Sea Research II, 53, 1530−1547. Jeffrey, S. W., Mantoura, R. F. C., & Wright, S. W. (1997). Phytoplankton pigments in oceanography: Guidelines to modern methods. Paris: UNESCO Publishing. Marra, J., Trees, C. C., & O'Reilly, J. E. (2007). Phytoplankton pigment absorption: A strong predictor of primary productivity in the surface ocean. Deep Sea Research I, 54, 155−163. Maranon, E., & Holligan, P. M. (1999). Photosynthetic parameters of phytoplankton from 50 N to 50 S in the Atlantic Ocean. Marine Ecology. Progress Series, 176, 191−203. Maranon, E., Holligan, P. M., Varela, M., Mourino, B., & Bale, A. J. (2000). Basin-scale variability of phytoplankton biomass and growth in the Atlantic Ocean. Deep-Sea Research, 47, 825−857. Maranon, E. (2005). Phytoplankton growth rates in the Atlantic subtropical gyres. Limnology and Oceanography, 50, 299−310. Maranon, E., Cermeno, P., Rodriguez, J., Zubkov, M. V., & Harris, R. P. (2007). Scaling of phytoplankton photosynthesis and cell size in the ocean. Limnology and Oceanography, 52, 2190−2198. Maranon, E. (2008). Inter-specific scaling of phytoplankton production and cell size in the field. Journal of Plankton Research, 30, 157−163. Moore, C. M., Lucas, M. I., Sanders, R., & Davidson, R. (2005). Basin-scale variability of 471 phytoplankton bio-optical characteristics in relation to bloom state and community structure in the Northeast Atlantic. Deep-Sea Research I, 52, 401−419. Not, F., Latasa, M., Cariou, T., Valout, D., & Simon, N. (2004). A single species, Micromonas pusilla (Prasinophyceae), dominates eukaryote picoplankton in the Western English Channel. Applied Environmental Microbiology, 70, 4064−4072. Poulton, A. J., Holligan, P. M., Hickman, A., Kim, Y. -N., Adey, T. R., Stinchcombe, M. C., Holeton, C., Root, S., & Woodward, E. M. S. (2006). Phytoplankton carbon fixation, chlorophyll-biomass and diagnostic pigments in the Atlantic Ocean. Deep-Sea Research II, 53, 1593−1610. Robinson, C., Poulton, A. J., Holligan, P. M., Baker, A. R., Forster, G., Gist, N., et al. (2006). The Atlantic Meridional Transect (AMT) Programme: A contextual view 1995– 2005. Deep Sea Reserch II, 53, 1485−1515. Sathyendranath, S., Watts, L., Devred, E., Platt, T., Caverhill, C., & Maass, H. (2004). Discrimination of diatoms from other phytoplankton using ocean-colour data. Marine Ecology. Progress Series, 272, 59−68. Smyth, T., Moore, G., Hirata, T., & Aiken, J. (2006). Semianalytical model for the derivation of ocean colour inherent optical properties: Description, implementation, and performance assessment. Applied Optics, 45, 8116−8131. Suggett, D. J., Moore, C. M., Maranon, E., Omachi, C., Varela, R. A., Aiken, J., et al. (2006). Photosynthetic electron turnover in the tropical and subtropical Atlantic Ocean. Deep-Sea Research, 53, 1573−1592. Tarran, G. A., Heywood, J. L., & Zubkov, M. V. (2006). Latitudinal changes in the standing stocks of nano- and picoeukaryotic phytoplankton in the Atlantic Ocean. Deep-Sea Research II, 53, 1516−1529. Uitz, J., Claustre, H., Morel, A., & Hooker, S. B. (2006). Vertical distribution of phytoplankton communities in open ocean: An assessment based on surface chlorophyll. Journal of Geophysical Reserch, 111, C08005. doi:10.1029/2005JC003207. Vidussi, F., Claustre, H., Manca, B. B., Luchetta, A., & Marty, J. (2001). Phytoplankton pigment distribution in relation to upper thermocline circulation in the eastern Mediterranean Sea during winter. Journal of Geophysical Reserch, 106(C9), 19939−19956. Werdell, P. J., & Bailey, S. W. (2005). An improved in situ bio-optical data set for ocean colour algorithm development and satellite data product validation. Remote Sensing of Environment, 98, 122−140. Zubkov, M. V., Sleigh, M. A., Tarran, G. A., Burkill, P. H., & Leaky, R. J. G. (1998). Picoplankton community structure on an Atlantic transect from 50 N to 50S. DeepSea Research I, 45, 1339−1355. Zubkov, M. V., Sleigh, M. A., Burkill, P. H., & Leakey, R. J. G. (2000). Picoplankton community structure on the Atlantic Meridional Transect: A comparison between seasons. Progress in Oceanography, 45, 369−386.