We employ a global three-dimensional model to simulate diverse phytoplanktonic diazotrophs (nitrogen fixers) in the oceans. In the model, the structure of the marine phytoplankton community self-assembles from a large number of potentially viable physiologies. Amongst them, analogs of Trichodesmium, unicellular diazotrophs and diatom-diazotroph associations (DDA) are successful and abundant. The simulated biogeography and nitrogen fixation rates of the modeled diazotrophs compare favorably with a compilation of published observations, which includes both traditional and molecular measurements of abundance and activity of marine diazotrophs. In the model, the diazotroph analogs occupy warm subtropical and tropical waters, with higher concentrations and nitrogen fixation rates in the tropical Atlantic Ocean and the Arabian Sea/Northern Indian Ocean, and lower values in the tropical and subtropical South Pacific Ocean. The three main diazotroph types typically co-exist in the model, although Trichodesmium analogs dominate the diazotroph population in much of the North and tropical Atlantic Ocean and the Arabian Sea, while unicellular-diazotroph analogs dominate in the South Atlantic, Pacific and Indian oceans. This pattern reflects the relative degree of nutrient limitation by iron or phosphorus. The model suggests in addition that unicellular diazotrophs could add as much new nitrogen to the global ocean as Trichodesmium.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 Nitrogen fixers or diazotrophs, organisms that fix nitrogen gas (N2) into organic form, play an important role in the climate system as the availability of inorganic nitrogen can represent a significant limiting factor for marine primary production [Falkowski et al., 1998]. On the global scale, diazotrophy is the largest source of fixed nitrogen in the modern ocean [Galloway et al., 2004; Gruber, 2004], counterbalancing the sinks due to denitrification [Gruber and Sarmiento, 1997; Codispoti et al., 2001] and anammox [Devol, 2003]. It can also have important local consequences: In the oligotrophic North Pacific Ocean, nitrogen fixation is estimated to fuel up to 50% of the new production [Karl et al., 1997].
 Much of the current understanding of marine nitrogen fixation comes from the study of Trichodesmium; large, filamentous and non-heterocystous cyanobacteria that often aggregate at the surface of the ocean to form colonies. Trichodesmium has been observed to grow in most warm oligotrophic subtropical and tropical regions of the ocean [Carpenter and Romans, 1991; Capone et al., 1997], apart from the western equatorial and Southeastern subtropical Pacific Ocean [Mague et al., 1974; Bonnet et al., 2008, 2009].
 Both Trichodesmium and unicellular diazotrophs have been observed to have a relatively low maximum growth rate, high N:P and Fe:P ratios, and to be adapted to high light conditions and warm temperatures [Falcon et al., 2005; LaRoche and Breitbarth, 2005; Goebel et al., 2008]. These characteristics are probably related to the high energetic cost of breaking down the N2 triple bond using the enzyme nitrogenase.
 Ultimately, continued observations are necessary to define global habitats and nitrogen fixing contributions of each type of diazotroph. Models can help to synthesize and extrapolate the existing observations, as well as to explore hypothesized mechanisms of the regulators of community structure and habitat and to evaluate the integrated rate of nitrogen fixation in the global ocean. However, representing the diverse populations of marine microorganisms in models remains a challenge. Several published models include specific representations of a diazotrophic functional group [e.g., Fennel et al., 2002; Hood et al., 2004; Moore and Doney, 2007], typically based on physiological understanding of Trichodesmium. For the first time, Goebel et al.  resolved multiple diazotrophic types in a model, evaluating nitrogen fixation rates based on nifH observations of three diazotroph types (Trichodesmium and unicellular Groups A and B), in a one-dimensional representation of station ALOHA in the subtropical North Pacific Ocean.
 Here we introduce diazotrophy as a possible physiological function, associated with a variety of other characteristics, into a three-dimensional ocean circulation, biogeochemistry and ecosystem model in which the photo-autotroph communities “self-assemble” from a broad range of initialized phytoplankton types according to their relative fitness in the modeled environment [Follows et al., 2007]. In previous studies, this approach successfully reproduced the distribution of the diverse community of the cyanobacterium Prochlorococcus in the Atlantic Ocean [Follows et al., 2007], a plausible biogeography of the dominant functional groups of phytoplankton in the global ocean [Dutkiewicz et al., 2009], as well as the latitudinal gradient in marine phytoplankton diversity [Barton et al., 2010], though diazotrophy was not previously represented.
 We first describe the physical, biogeochemical, and ecosystem model (section 2). Then, in order to evaluate the veracity of the modeled plankton communities and nitrogen fixation rates, we compile and compare published direct observations of biomass and nitrogen fixation rates for Trichodesmium, unicellular diazotrophs and diatom-diazotroph associations (section 3). We subsequently show that the model system exhibits plausible patterns of abundance and nitrogen fixation rates for the main phytoplanktonic diazotrophs (section 4). This work provides a first cut at a global synthesis of the likely distributions and contributions in fixing nitrogen of three of the main types of marine diazotrophs.
 This study focuses on describing the model diazotroph biogeography in the light of a compilation of published observations of diazotroph biomass and nitrogen fixation rates. A deeper investigation into the ecological controls on the emergent model ecosystem is too extensive to include in this manuscript but is the subject of a companion paper (F. M. Monteiro et al., Biogeographical controls on the marine nitrogen fixers, submitted to Global Biogeochemical Cycles, 2010).
2. Model Description
 Our modeling framework is based on that of Follows et al.  and Dutkiewicz et al. , where a three-dimensional global ocean model underpins biogeochemical and ecosystem model components with “self-assembling” phytoplankton community structure.
2.1. Ocean Circulation and Biogeochemistry Models
 The ocean model (MITgcm; Marshall et al. ) is configured in a relatively coarse resolution (1° × 1° horizontally; 23 vertical levels), with sub-mesoscale eddy fluxes parameterized following the work of Gent and McWilliams  and boundary layer turbulent mixing following that of Large et al. . We use the ECCO-GODAE state estimates [Wunsch and Heimbach, 2007] in which the model circulation has been constrained to closely match in situ and remote observations of the physical environment.
 The model ocean transports inorganic and organic forms of phosphorus (P), nitrogen (N), iron (Fe) and silica (Si). None of the external sources of these elements (atmospheric, river or sediments) are represented in the model, except for nitrogen fixation (see later) and the atmospheric deposition of iron, using the model estimates made by Mahowald et al. . Here we assume that 0.8% of deposited iron is bioavailable [Mahowald et al., 2009]. The dynamic representation of the marine iron cycle, following Parekh et al. , includes complexation with an organic ligand and particulate scavenging. The biogeochemical and biological tracers interact through the formation, transportation and remineralization of organic matter. Photo-autotrophs are assumed to have fixed elemental ratios and may be consumed by two simple grazers while excretion and mortality, from both phytoplankton and zooplankton, transfer organic matter into sinking particulate and dissolved organic detritus pools. Heterotrophic bacteria are not explicitly represented. Organic detritus is simply respired back to inorganic forms on a fixed timescale.
 Here we also include a more detailed representation of the nitrogen cycling, including explicit representation of nitrogen fixation by diazotrophic phytoplankton (see next paragraph). The combined effect of deep denitrification, anammox and (possibly) heterotrophic diazotrophy is represented in the model very simply, by restoring excess nitrate towards the observed climatology from the World Ocean Atlas 2005 [Garcia et al., 2006] below 200 m depth (details in Appendix A).
2.2. Ecosystem Model
 As in the study of Follows et al. , we initialize many tens (here 78) of physiologically unique phytoplankton types with characteristics randomly generated from plausible ranges of possible physiologies. The randomization places the organism in one of two nominal size classes (“large” or “small”) and assigns parameters regulating the sensitivity of growth to light, nutrients and temperature (Appendix A). Trade-offs are imposed for several characteristics as a function of size: small phytoplankton are presumed to be more efficient at taking up nutrients in oligotrophic conditions, to be able to cope with lower light (due to a lack of package effect), but to be more likely to become photoinhibited and to have slower maximum growth rates. Though, in a general sense growth rate might be expected to decrease with cell size, taxonomic differences can dominate, as seen when comparing laboratory cultures of small, pico-cyanobacteria Prochlorococcus and fast-growing, populations of larger diatom cells [Sarthou et al., 2005]. The high maximum growth rates observed for diatoms may in part reflect an enhanced internal store, which can be larger for bigger cell types [Droop, 1983] though not accounted for explicitly in our model. The model also randomly selects for the nutrient strategy of the phytoplankton types. For example, a “coin-flip” determines whether a cell type requires silicon; if so it is assumed to be a diatom analog. The rate of change of the biomass of each phytoplankton is determined as a combination of growth (which is a function of light, temperature and nutrient resources), sinking, grazing, mortality and transport by the fluid environment (see Dutkiewicz et al.  for full prognostic equations).
 Here, we introduce to this framework the possibility for phytoplankton to fix nitrogen gas (N2). Diazotrophy is also assigned at random and may be associated with any other set of physiological characteristics. The diazotrophs are then assumed to be nitrogen non-limited, to have a smaller specific maximum growth rate (high energetic cost of nitrogen fixing) and higher elemental N:P and Fe:P ratios (high iron requirements of nitrogenase enzyme) than non-diazotrophic phytoplankton (Appendix A). We nominally assume generated diazotrophic cells in the large size class which are not diatoms, to be analogs of Trichodesmium; model diatoms which can fix N2 to be analogs of diatom-diazotroph associations (DDA); and small size-class cells which can fix nitrogen to be analogs of unicellular, cyanobacterial nitrogen fixers. In all simulations presented here, each of these three main diazotroph types is composed of several (typically 3 to 4) phytoplankton types, which differ slightly in physiology. Here we focus on the total population of each main diazotroph type (e.g. Trichodesmium), that is the sum of any generated phytoplankton that fall into that classification.
 As in our previous studies [Follows et al., 2007; Dutkiewicz et al., 2009], two size classes of grazers are represented in the model, where small (or big) size class preferentially eats small (or big) phytoplankton. For simplicity, here we choose not to have different palatability for the diazotrophs. Though it is probable that Trichodesmium has lower grazing pressure than other phytoplankton [Capone et al., 1997], there is evidence that at least one copepod group feeds on Trichodesmium [O'Neil and Roman, 1994]. More research on grazing pressure would certainly help us to better shape the trade-offs of diazotroph grazers.
 Most simulations were integrated for ten model years, after which time a robust seasonal cycle of productivity and phytoplankton community structure has developed and persists for several decades. Over longer timescales the background nutrient distributions drift slightly, driving minor adjustments in biogeography. One simulation was integrated for 100 years and the biogeography, in particular the diazotroph communities, remained relatively constant for that period. Here, all shown results are annual average of the tenth year of simulation. An ensemble of ten simulations was performed, each identical except for the randomization of physiological traits in the initialized organisms. The collection of the ten simulations is the “ensemble” and the individual simulations are the “ensemble members”. The biogeographical patterns that are established in the ensemble members were robust and broadly consistent and here we present the “ensemble mean”; the average of the results from the ten simulations.
 The resulting biogeography of the main phytoplankton functional groups is very similar to those found in the studies of Follows et al.  and Dutkiewicz et al. . Here the model additionally resolves multiple diazotroph types. Though typically at relatively low abundance relative to the main contributors to biomass, the model diazotrophs have a robust distribution which we compare to a compilation of published observations in the following sections.
3. Observations of Diazotroph Biogeography and Nitrogen Fixation
 Diazotroph abundance and nitrogen fixation rates have been observed in the ocean for more than 35 years, with early studies related to Trichodesmium [Dugdale et al., 1964] and DDAs [Mague et al., 1974]. Published data on the distribution of Trichodesmium have been compiled for the global ocean [Carpenter, 1983] and the North Atlantic Ocean [Hood et al., 2004]. Here we extend these two compilations to include most currently available observations of occurrence, biomass and nitrogen fixation rates for the three main phytoplanktonic diazotroph types and for the global ocean.
 For diazotroph abundance, we compile traditional observations of Trichodesmium and DDA (net tow combined with microscopy), video plankton recorder observations of Trichodesmium and evaluations of the abundance of the nifH gene for Trichodesmium, unicellular diazotrophs and DDAs. Measurement of nitrogen fixation rates comes from acetylene reduction assay or isotopic 15N2 methods which have been used to evaluate areal rates (molN m−2 d−1) for Trichodesmium, unicellular diazotrophs and DDAs. (Some additional published measurements of nitrogen fixation rates are not included here as they integrate the contribution of diazotrophic heterotrophic bacteria which are not represented in the model.)
 To compare model results with these observations, we convert the published values into common units (μmolP l−1 for diazotroph biomass and μmolN m−2 d−1 for nitrogen fixation rates) and include both the original measurement and estimated conversion in tabular form in the auxiliary material (Tables S1–S8). The tabulated values include both the uncertainty of the conversion factor and that of the original measurement (e.g. for nitrogen fixation rates, the depth distribution of the diazotroph and the time period of fixation used in the calculation can lead to uncertainties [Lipschultz and Owens, 1996]). Here we outline the method of conversion used for the abundance of each diazotroph type.
 By far the majority of observations on marine nitrogen fixation have been related to Trichodesmium. We convert observations of Trichodesmium abundances from filaments, cells, colonies or nifH copies into units of phosphorus biomass (μmolP l−1). To do so, we use average values of field and laboratory measurements of the amount of phosphorus present in the observed entity (see Text S1 and Tables S1 and S2). The resulting compilation shows that Trichodesmium observations (Tables 1 and 2) have relatively good coverage in the North Atlantic Ocean, North East subtropical Pacific Ocean and China Sea, but are more sparse in central South Atlantic Ocean, tropical East Pacific Ocean, middle West subtropical Pacific Ocean and East Indian Ocean.
Table 1. Comparison Between Model and Observations of Abundance of Trichodesmium (μmolP l−1) in the Global Ocean: Biomass of Trichodesmiuma
 Measurements of the abundance of unicellular cyanobacteria diazotrophs in the ocean are made using nifH gene sequencing methods. Here we convert the number of nifH copies into biomass units (μmolP l−1) for unicellular GroupsA and B by assuming 1 nifH copy per cell [Goebel et al., 2007]. Following Goebel et al. , since there are no direct measurements of the phosphorus quota in unicellular diazotrophs, we scale Trichodesmium phosphorus conversions to the unicellular diazotroph volume, using the allometric relationship for marine phytoplankton developed by Verity et al.  (see auxiliary material, Text S1). This scaling introduces some additional uncertainty.
Goebel et al.  estimated cell volumes of Trichodesmium, unicellular Group B and Group A to be between 200–1000 μm3, 4–270 μm3, and 0.2–4.0 μm3 respectively. Unicellular Group B cell volume is therefore on average two orders of magnitude larger than for unicellular Group A. We sum Group A and Group B biomass to estimate the total populations of unicellulars (Table 3). In most observations of unicellular diazotrophs, nifH concentrations are higher for unicellular Group A than for unicellular Group B (see Tables S4 and S5). However the biomass of Group B is generally greater than for unicellular Group A due to the difference in cell volume, thus suggesting a greater contribution to global nitrogen fixation from Group B. Current observations of unicellular diazotrophs in the ocean are mostly restricted to the North Atlantic region, the East subtropical Pacific Ocean and the Arabian Sea (Tables 3 and 4).
Table 3. Comparison Between Model and Observations of Abundance of Unicellular Diazotrophs (μmolP l−1) in the Global Ocean: Biomass of Unicellular Diazotrophsa
Model (μmolP l−1)
Observations (μmolP l−1)
The areas are labeled with letters and are reported in Figure 1c. We assume a ratio for nifH copies:cell = 1 [Goebel et al., 2007] and use the allometric relationship by Verity et al.  to convert phosphorus cell content from Trichodesmium estimate to unicellular Groups A and B. Data are detailed in the auxiliary material (Text S1 and Tables S4, S5, and S6) and values have been rounded.
 Due to the complexity of symbiotic organisms, the understanding of diatom-diazotroph associations (DDA) remains limited. Detection of DDAs in the ocean relies on specific microscopy (transmitted light and epi-fluorescent) [Carpenter et al., 1999] and nifH gene sequencing [Foster and Zehr, 2006]. Since no calibration of DDA nifH copies to biomass is available, we do not attempt in this study to compare quantities of abundance between the model and the data, but simply make the distinction between presence or absence of DDAs (Table 5, completed by the sparse observations of DDA nitrogen fixation rate in Table 6). We note though that the model itself provides biomass values for the DDA analogs, presented in Figure 1d.
Table 5. Comparison Between Model and Observations of Occurrence of Diatom-Diazotroph Associations (DDA) in the Global Ocean: Occurrence of Diatom-Diazotroph Associationsa
The areas are labeled with letters and are reported in Figure 1d.
Table 6. Comparison Between Model and Observations of Nitrogen Fixation Rates of Diatom-Diazotroph Associations in the Global Ocean (μmolN m−2 d−1): Nitrogen Fixation Rate of Diatom-Diazotroph Associationsa
Model (μmolN m−2 d−1)
Observations (μmolN m−2 d−1)
The areas are labeled with letters and are reported in Figure 2d. Data are detailed in the auxiliary material (Text S1 and Table S8) and values have been rounded.
 In the following section we present the model results of the abundance and nitrogen fixing activity of the three major marine photo-autotrophic diazotroph types and compare with the compiled observations for occurrence, biomass and nitrogen fixation rates.
4. Model Results: Diazotroph Biogeography and Nitrogen Fixation
Figure 1 presents the ensemble-average, annual mean biomass (μmolP l−1) of the total diazotrophs, Trichodesmium, unicellulars, and DDAs in the surface waters (0–50 m) for the tenth year of the ensemble simulation. Here we concentrate on the 0–50 m layer as most observations of diazotroph activity are taken in the upper water column. Figure 2 presents the corresponding model result for nitrogen fixation rates (μmolN m−2d−1). Concentrations and rates are presented on a log scale, as values span large ranges. The emergent diazotroph biogeography is relatively robust between the ten-ensemble-member simulations, illustrated in Figure 3 which displays the standard deviation, relative to the mean, of the simulation ensemble. (The standard deviation of nitrogen fixation rates is shown in Figure S1). Over the ten simulations, biomass and rates are consistent for total, Trichodesmium and unicellular-diazotroph analogs which have a low standard deviation relative to the mean in most regions. However the standard deviation, relative to the mean is high for the DDAs indicating a greater role of the randomization of physiological characteristics in regulating the habitat of this diazotroph type. This reflects, in part, our inability to define clear trade-offs and the likelihood that we have over-simplified this complex symbiotic relationship in the model.
 We discuss now the modeled biogeography of diazotrophs both collectively and as separate types, in comparison with their observed distributions.
4.1. Sum of Diazotrophs
 Although there is no explicit demand that modeled diazotroph analogs grow only in warm waters, the total autotrophic diazotroph population occupies warm subtropical and tropical regions of the ocean (Figure 1a), as is generally observed. While some previous models parameterize their diazotrophic functional types to only grow in warm waters [i.e., Moore et al., 2004, 2006; Breitbarth et al., 2007; Moore and Doney, 2007] or stratification conditions [Hood et al., 2004], here this feature is an emergent result of the “self-assembling” model communities. Cold-adapted diazotrophs were initialized in the model but were selected against due to other factors. Collectively, the modeled diazotrophs occupy warm subtropical and tropical waters (Figure 1a for biomass and Figure 2a for nitrogen fixation rate), with higher concentrations and nitrogen fixation rates in the tropical Atlantic Ocean, the Arabian Sea, and the Northern Indian Ocean (≥10−2μmolP l−1 for biomass, and ≥102μmolN m−2 d−1 for nitrogen fixation rates). Much of the central (sub)tropical South Pacific Ocean has very low diazotroph activity. The low iron concentrations there (Figure 5d) and the high requirement for iron imposed on the model diazotrophs limit diazotroph growth in this region of the Pacific Ocean.
4.2. Trichodesmium Analogs
 The modeled habitat and nitrogen fixation rate of Trichodesmium are compared with observations on a regional basis (Tables 1 and 2). Each region is designated by a letter (see Tables 1–6) indicated on the map of Trichodesmium surface biomass (Figure 1b) and nitrogen fixation rate (Figure 2b). Both observed and modeled Trichodesmium biomass concentration and nitrogen fixation rates vary by an order of magnitude or more even within regions (Tables 1 and 2). The large observed variability within regions may, in part, be due to seasonality, eddies, and sparse sampling.
 The large scale patterns in modeled Trichodesmium biomass and nitrogen fixation rates is generally consistent with those observed: High Trichodesmium concentrations (on the order of 10−3μmolP l−1, equivalent to 104 fil m−3) and nitrogen fixation rates (from 101 to 102μmolN m−2 d−1) occur in the tropical Atlantic Ocean, the Northern Indian Ocean and Arabian Sea. Somewhat lower values are found in the subtropical North Atlantic Ocean (concentrations around 10−5 to 10−4μmolP l−1 and rates of 101μmolN m−2 d−1). In the subtropical South Atlantic Ocean, model and observations show a sharp gradient in concentrations from about 10−4μmolP l−1 in the West, to almost nothing in the East of the basin. Low values characterize most of the North Pacific Ocean in both the model and the observations with about 10−5μmolP l−1 of biomass (equivalent to 102 fil m−3) and 1 μmolN m−2 d−1 of nitrogen fixation rate. Finally the model does not produce any Trichodesmium in the East South Pacific: Cruises in this region (Q and T) also did not find any Trichodesmium [Mague et al., 1974; Bonnet et al., 2008, 2009].
 A few regions in the model have lower biomass of Trichodesmium than what is observed. These include the Gulf of Mexico (C), some areas in the Western Pacific Ocean (L, P, R, S) and the Western part of the Indian Ocean (Y). Some of these (C, R, Y) are coastal regimes where tidal mixing, riverine inputs, and other influences unresolved in the model may be important. The modeled Western tropical and subtropical Pacific (L, P, R, S) have very low iron concentrations, potentially due to the lack of parameterization of sediment supply, which might also lead to unrealistically low diazotrophic activity. The Western tropical Indian Ocean (X) and the Mediterranean Sea (Z) have biomass of Trichodesmium which are too high in the model. Finally, the modeled nitrogen fixation rates are much lower than observed in the East subtropical North Pacific (M), though the modeled biomass concentrations in the region is similar to the observations. Similarly in the Central tropical Pacific Ocean (region Ø), the model reproduces observations of nitrogen fixation rates but underestimates Trichodesmium biomass. These discrepancies between biomass and rate of nitrogen fixers are not understood but might reveal an interesting dynamics present in the real ocean, where Trichodesmium fix more (or less) nitrogen than they actually need in the East subtropical North (or Central tropical) Pacific Ocean.
 Estimates of the occurrence of Trichodesmium blooms based on remote sensing [Westberry and Siegel, 2006] found high bloom concentrations in the tropical Atlantic Ocean, the Arabian Sea and the Western Indian Ocean; supporting the modeled distribution presented here. Westberry and Siegel  also inferred high Trichodesmium concentrations in the Eastern tropical Pacific Ocean consistent with inferences based on geochemical diagnostics [Deutsch et al., 2007]. Here, however, our model solutions show almost no presence of Trichodesmium. It is difficult to reconcile the results of both studies by Deutsch et al.  and Westberry and Siegel  to the direct in situ observations, as so far very little autotrophic diazotrophy has been observed in these regions [Mague et al., 1974; Bonnet et al., 2008, 2009], and seems to support our model solutions.
 The model captures the seasonal cycle of Trichodesmium at the Bermuda Atlantic Time-series Station, with a notable late summer bloom (Figure 4, shown here for a single ensemble member). Trichodesmium biomass varies between about 2 × 10−5μmolP l−1 in the late Spring to about 13 × 10−5μmolP l−1 in the late summer, consistent with time-series observations [Orcutt et al., 2001]. The model predicts a similar, late summer bloom of Trichodesmium in the subtropical North Pacific Ocean, as observed at the Hawaii Ocean Time-Series [Goebel et al., 2007; Church et al., 2009]. Here though the seasonality is captured, the modeled abundances are gradually drifting due to a long term trend in the modeled iron concentrations in this region.
4.3. Unicellular Diazotroph Analogs
 Both model and observations suggest high concentrations of unicellulars (about 10−3μmolP l−1), with corresponding high nitrogen fixation rates, in most of the East North Atlantic Ocean and Arabian Sea/Indian Ocean (Figure 1c and Table 3 for concentrations, and Figure 2c and Table 4 for rates). Both model and observations suggest lower concentrations and fixation rates in the East subtropical North Pacific (about 10−4μmolP l−1 and 101μmolN m−2 d−1 respectively) as well as no unicellular diazotroph activity in the East South Pacific, except for a small area in the middle of the subtropical gyre.
 The model and observations only differ significantly in the Western subtropical North Atlantic Ocean (A). Here however, the observations provide a mixed perspective with a measured rate of nitrogen fixation of the same order of magnitude as in the North Pacific [Falcon et al., 2004; Foster et al., 2007], consistent with the model solution. In contrast, another cruise found no evidence of unicellular diazotroph abundance from nifH analysis [Langlois et al., 2005], whereas the model supports a population of unicellular diazotrophs in agreement with the modeled fixation rate.
 Here we treat Group A and Group B both as strict autotrophs. However recent discoveries about the physiology of unicellular Group A indicate that they must rely on organic carbon as they lack the PSII oxygenic photosynthetic apparatus [Zehr et al., 2008] and may participate in a symbiotic relationship [Tripp et al., 2010]. In the current model, we do not clearly represent Group A and their rather special physiology and ecology. As understanding increases, future models may need to treat these organisms differently, with potentially significant implications for overall nitrogen fixation rates.
4.4. Diatom-Diazotroph Association Analogs
 Since we are unable to convert the biomass observations of DDAs to model units, we compare patterns of presence or absence of the organisms (Figure 1d and Table 5). DDAs have been observed in the Western side of the (sub)tropical North Atlantic Ocean, the Eastern subtropical and Western tropical North Pacific, the Kuroshio current and the Eastern Mediterranean Sea. The model solutions are consistent except in the Eastern subtropical North Pacific and the Mediterranean Sea where there are no model DDAs. In the model, DDAs are most abundant in the Arabian Sea and the Indian Ocean, as well as the west tropical Atlantic Ocean. The nitrogen fixation rates of modeled DDAs are typically too low at any locations for which there are measurements (Figure 2d and Table 6), perhaps indicating that the parameterization of the symbiotic relationship of a DDA is too simplistic.
 Overall, the model captures the main features of the observed habitats for the three diazotroph types and is broadly consistent with observed abundances and rates for Trichodesmium and unicellular diazotrophs. The model might underestimate nitrogen fixation by DDAs globally and by Trichodesmium in areas of the Western Pacific Ocean.
5.1. Biogeography of Marine Diazotrophs
 The three main diazotrophic types co-exist over much of the modeled subtropical ocean (Figure 5a). Unicellular-diazotroph analogs tend to dominate in the model in most of the regions inhabited by diazotrophs (Figure 5b), especially where iron is limiting (Figures 5c and 5d). The lower nutrient half-saturation constant, a consequence of their smaller size, confers in these regions an advantage on the unicellular diazotrophs. Unicellular-diazotroph analogs also dominate the diazotroph population in a small region of the equatorial Atlantic Ocean, which is characterized by low phosphate concentration rather than low iron concentration (Figures 5c and 5e). When the nutrient limitation is lifted, Trichodesmium analogs are no longer disadvantaged. Trichodesmium dominate particularly in regions where iron is more plentiful (subtropical Atlantic Ocean and Arabian Sea). DDAs are found in the model to dominate only in small regions, though we note that the simulated distribution of DDA analogs is quite variable between ensemble members and we are less confident of these predictions.
Trichodesmium analogs in our model are responsible for about 43% of the global total nitrogen fixation rate (25 TgN yr−1; section 5.2), while unicellular-diazotroph analogs are responsible for about 49% (30 TgN yr−1) of the global rate. Consistent with previous suggestions [Carpenter et al., 1999; Zehr et al., 2001], our global ocean model explicitly indicates that unicellular diazotrophs could be as important in the global nitrogen budget as Trichodesmium. DDA analogs contribute to about 8% of the global nitrogen fixation rate (5 TgN yr−1), though we are not sufficiently comfortable with the parameterization of this symbiotic relationship and the robustness of their distribution in the ensemble to be confident of this result.
5.2. Nitrogen Fixation Rates
 The global nitrogen fixation rate of our ensemble average is 60 ± 15 TgN yr−1 (where the uncertainty is the standard deviation of the ten ensemble members), within the range of published direct and indirect estimates, which vary from 5 to 135 TgN yr−1.
 Direct in situ biological estimates increased from early estimates of 5–10 TgN yr−1 [Capone and Carpenter, 1982; Carpenter et al., 1992] to more recent estimates of 80–85 TgN yr−1 [Capone et al., 1997; Galloway et al., 2004]. However these have accounted only for the contribution of Trichodesmium, extrapolating sparse measurements to a basin scale or global value for warm oligotrophic regions [Capone et al., 1997; Galloway et al., 2004]. Our model results (and we believe the data compilation) suggest that there are large regional variations in nitrogen fixation even in these warm oligotrophic waters. Thus we suggest that direct estimations of Trichodesmium activity [Capone et al., 1997; Galloway et al., 2004] extrapolated from a few measurements have large uncertainties. Our model suggests that Trichodesmium only contributes 25 TgN yr−1, though we underestimate Trichodesmium nitrogen fixation rates in the Western Pacific Ocean (where iron limitation is too strong) and in coastal zones.
 Geochemically based estimates of nitrogen fixation, relying on interpretations of N* and DINxs, reflect contributions from the entire diazotrophic community (including heterotrophic nitrogen fixers). Geochemical estimates range between 110 and 135 TgN yr−1 globally, subject to significant uncertainties [Gruber and Sarmiento, 1997; Gruber, 2004; Hansell et al., 2004; Landolfi et al., 2008; Deutsch et al., 2007]. The difference with our model result might be in part due to the activity of diazotrophic heterotrophic bacteria which is not resolved here.
 Published biogeochemical models have estimated global nitrogen fixation rates ranging from 55 to 100 TgN yr−1 [Moore et al., 2002, 2004, 2006] with a clear relationship between rate of iron supply to surface ocean and global nitrogen fixation. Increasing dust-borne iron solubility to 4% in our model induces a global nitrogen fixation rate of 80 ± 30 TgN yr−1 but at the expense of fidelity in comparison to the observed data of diazotroph activity. The distribution of the combined diazotroph types in this model is similar to that in previous model studies [Moore et al., 2004; Moore and Doney, 2007], which represented a single, generic marine diazotroph based on Trichodesmium. Therefore it is possible that the first order global biogeochemical impact of autotrophic nitrogen fixers in the ocean can be captured without the level of diversity modeled in this study. The degree of nitrogen fixation may be rather dictated by the relative shortfall of nutrients (e.g. the available amount of iron and phosphate to the diazotrophs). However the spatial extent of the diazotroph habitat, the stability of the biogeochemical nitrogen system and response to variability and climatic changes are probably all affected by the level of the captured ecological resolution.
 We coupled a three-dimensional ocean model with a biogeochemical and ecosystem parameterization to simulate the diverse population of diazotrophic phytoplankton in the global ocean. In this “self-assembling” ecosystem model, analogs of Trichodesmium, unicellular diazotrophs and diatom-diazotroph associations (DDA) are enabled and are successful and abundant in each of the ten members of the ensemble of simulations. In addition, we compiled currently available observations of biomass and nitrogen fixation rates for Trichodesmium, unicellular diazotrophs and DDAs in order to evaluate the veracity of the model solutions.
 The simulations capture the broad character of the observed distribution of Trichodesmium, unicellular diazotrophs and DDAs, as well as the abundances and nitrogen fixation rates of Trichodesmium and unicellular diazotrophs. They suggest that the three main types of photo-autotrophic diazotrophs co-exist over most of the warm subtropical and tropical waters of the ocean with higher abundances and rates in the tropical North Atlantic Ocean and the Northern Indian Ocean, consistent with the data compilation, diagnostic studies and other models. In this model, the confinement of diazotrophs to warm oligotrophic waters is not imposed but is a result of selection from a broad variety of initialized diazotroph analogs including some adapted to cold temperatures, suggesting that other factors (resource availability) are more significant in defining the habitat. In the Eastern South Pacific Ocean, the model suggests an absence of diazotrophs attributable to iron limitation. This finding is at odds with some other models and diagnostic studies [Westberry and Siegel, 2006; Deutsch et al., 2007] but appears to be supported by observations [Mague et al., 1974; Bonnet et al., 2008, 2009]. Diazotroph activity is underestimated in our model in the Western Pacific Ocean because iron limitation is too strong (perhaps due to a lack of sedimentary input) and in the unresolved coastal zones. Our globally integrated nitrogen fixation rate (60 ± 15 TgN y−1) is about half that estimated by geochemical tracers. This may reflect, in part, the fact that we do not resolve heterotrophic diazotrophs and that some regions are too strongly iron limited. An inclusion of sedimentary iron sources and variable iron solubility may help for the latter issue.
 In the model, Trichodesmium analogs dominate the total diazotroph population in most of the North and tropical Atlantic Ocean and in the Arabian Sea, whereas unicellular-diazotroph analogs dominate in the South Atlantic, Pacific and Indian oceans where iron or phosphorus are less plentiful. The model predicts in particular a high level of nitrogen fixation by unicellular diazotrophs in the Indian Ocean. Observations are still too sparse to support or refute this result. Finally, the model suggests that unicellular diazotrophs globally contribute to add as much new nitrogen to the ocean as Trichodesmium, supporting Zehr et al.  hypothesis on the importance of small diazotrophs in the marine nitrogen cycle. Thus, while Trichodesmium may still be the most observed diazotrophic phytoplankton, any attempt to estimate total global nitrogen fixation rates should include the contribution from other diverse diazotrophic communities. Much remains to be explored as new observations improve our understanding of these different nitrogen fixers and suggest better parameterizations in our model.
Appendix A:: Model Ecosystem Parametrization
 The ecosystem model equations are similar to that used in the studies of Follows et al.  and Dutkiewicz et al. . We use the exact same equations and parameter values as the multiple resource case experiment of Dutkiewicz et al. , which are provided in their Appendix. Here we added changes for the nitrogen cycle to include the source of nitrogen fixation and the sink of denitrification (detailed below). We also used a smaller Fe solubility constant (αFe = 0.008), ideal to represent the distribution of Trichodesmium in the North Atlantic Ocean model, and in the range of the observations that vary between 0.01% and 80%, with most values in the lower range [Mahowald et al., 2009].
A1. Nitrogen Fixation
 Nitrogen fixation in our model is represented by several autotrophic diazotroph types. When a phytoplankton is generated to be a diazotroph (Diaz), we assume the growth to be nitrogen non-limited and to have a smaller maximum growth rate (μmaxDiaz = 1/2 μmax) as well as higher N:P and Fe:P elemental ratios (N:PDiaz = 40 mol:mol; Fe:PDiaz = 3 × 10−2 mol:mol; see Text S1 in auxiliary material). Sensitivity studies were carried out in the model testing different values of αFe and Fe:PDiaz: We used those values that provided the best match to the observations of diazotrophs. Diazotroph abundances are observed on average to be ten times smaller that other phytoplankton [LaRoche and Breitbarth, 2005]. In our model we initialize diazotroph concentrations to be ten times less than for other phytoplankton; this avoids a large, unrealistic spike of new nitrogen at the beginning of the simulation and therefore reduces the computational time to reach steady state. In addition, we assume a half-saturation constant for unicellular-diazotroph analogs to vary between 0.012 and 0.017 μmolP l−1, which is between the range of Prochlorococcus analogs and other small organisms (such as Synechococcus). This is because we assume the half-saturation constant to be size-dependent and that unicellular diazotrophs are smaller than most small phytoplankton, and slightly bigger than Prochlorococcus.
A2. Nitrate Restoring Scheme
 We restore in the model the concentrations of nitrate to the observations (World Ocean Atlas 2005, Garcia et al. ) everywhere below 200 m, but only if the concentration exceeds the observation. This provides a crude representation of the combined effect of denitrification, anammox and (possibly) heterotrophic nitrogen fixation in the deep ocean. The model expression for the nitrate restoring scheme is equivalent to:
τrest is the restoring time scale, set to 1 year for our presented results. Sensitivity studies reveal that varying τrest from one year to a decade shows very similar distributions of diazotrophs in the global ocean model.
 Thanks to Jon Zehr, Chris Edwards, Nicole Goebel, Rachel Foster, Ed Boyle, and Maureen Coleman for discussions and advice, as well as Raleigh Hood and an anonymous reviewer for their very insightful comments. We thank the many investigators who have contributed to the growing body of observations of marine diazotrophs and nitrogen fixation rates which have been compiled here. This work was supported by the Gordon and Betty Moore Foundation Marine Microbiology Initiative and NASA.