By continuing to browse this site you agree to us using cookies as described in About Cookies
Notice: Wiley Online Library will be unavailable on Saturday 7th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 08.00 EDT / 13.00 BST / 17:30 IST / 20.00 SGT and Sunday 8th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 06.00 EDT / 11.00 BST / 15:30 IST / 18.00 SGT for essential maintenance. Apologies for the inconvenience.
 In this paper we use a coupled, 3-dimensional, biological-physical model, which includes an explicit, dynamic representation of Trichodesmium, to predict the distribution of Trichodesmium and rates of N2-fixation in the tropical and subtropical Atlantic Ocean. It is shown that the model reproduces the approximate observed meridional distribution of Trichodesmium in the Atlantic and elevated concentrations in specific coastal and open ocean regions where this organism is known to occur. The model also appears to reproduce the observed seasonality of Trichodesmium populations at higher latitudes (highest concentrations in summer and fall), but this seasonal cycle may be too pronounced at low latitudes. High and persistent Trichodesmium concentrations and rates of N2-fixation are generated by the model in the Gulf of Guinea off of Africa. This unexpected finding appears to be confirmed by historical measurements. In general, increased Trichodesmium concentrations develop in regions where the mixed layer is relatively thin (resulting in high mean light levels) and dissolved inorganic nitrogen (DIN) concentrations and phytoplankton biomass are low for extended periods of time. The model-predicted Trichodesmium distributions are therefore very sensitive to the fidelity of the physical model's representation of mixed layer depth variability, and upwelling intensity, and the biological model's estimated DIN and phytoplankton concentrations. The model generates a three-step successional sequence where (1) high DIN concentrations due to upwelling and/or mixing stimulate phytoplankton growth, followed by (2) Trichodesmium growth after DIN depletion and phytoplankton decline, followed by (3) enhanced phytoplankton growth due to new nitrogen inputs from N2-fixation. This sequence develops in response to seasonal variations in mixing in the southwestern North Atlantic and in response to upwelling along the coast of Africa and the equator. We interpret this sequence as representing a diatom- Trichodesmium-flagellate succession, which is consistent with observed species successions off of northwest Africa and in the Gulf of Mexico. The results presented in this paper lead us to conclude that our model includes the primary factors that dictate when and where Trichodesmium and N2-fixation occurs in the Atlantic. Moreover, it appears that our model reproduces some of the major effects that diazotrophically-derived inputs of new nitrogen have on the pelagic ecosystem.
 Recent revised estimates suggest that open ocean N2-fixation is globally significant, i.e., on the order of 80–110 Tg N yr−1 [Gruber and Sarmiento, 1997; Capone et al., 1997], and comparable to inputs of NO3 from the deep ocean in subtropical waters [Karl et al., 1997; Capone et al., 1997, D. G. Capone et al., New nitrogen input to the tropical North Atlantic Ocean by nitrogen fixation by the cyanobacterium, Trichodesmium spp., submitted to Nature, 2004 (hereinafter referred to as Capone et al., submitted manuscript, 2004)]. A large fraction of this fixation (perhaps as much as 25%) occurs in the Atlantic Ocean [Gruber and Sarmiento, 1997; Capone et al., 1997]. We now know that many different diazotrophic organisms contribute to this new nitrogen source [Hood et al., 2000; Zehr et al., 2001]. However, the conspicuous marine cyanobacterium, Trichodesmium, is still believed to be the most significant N2-fixer in the open ocean [Capone et al., 1997, Capone et al., submitted manuscript, 2004]. Although shipboard data are still limited, Trichodesmium distributions and rates of N2-fixation have been better characterized in the Atlantic than in any other ocean basin.
Trichodesmium has been the subject of quantitative scientific investigation for nearly a century [see, e.g., Wille, 1904; Dugdale et al., 1964; Carpenter and Capone, 1992]. The factors that control its growth are thought to include temperature, vertical mixing/light availability, competition with other phytoplankton species, and the availability of iron and/or phosphorus. Temperature control has been inferred from the observation that Trichodesmium is not found in significant densities in waters that are colder than 20°C, and rarely blooms below 25°C [Carpenter and Capone, 1992; Capone et al., 1997; Subramaniam et al., 2002]. The importance of wind mixing has similarly been deduced from reports that Trichodesmium blooms, which develop under calm conditions, dissipate rapidly when winds begin to increase. Accumulations of Trichodesmium are rarely observed when it is windy, even when other conditions are favorable for growth [Capone et al., 1997; Subramaniam et al., 2002]. The exact mechanism of this mixing control is not entirely clear. There is undoubtedly some dilution effect that occurs when the mixed layer deepens in response to increased winds, but there is also an impact on the average light in the mixed layer, and therefore the growth rate and physiology of Trichodesmium [Hood et al., 2001; Sañudo-Wilhelmy et al., 2001].
Hood et al.  hypothesized that, because of it's relatively slow growth rate (3–5 days per doubling [Capone et al., 1997]) and adaptation to high light [Carpenter et al., 1993], Trichodesmium concentrations will remain low when dissolved inorganic nitrogen (DIN) is replete in the euphotic zone and/or the mixed layer is thick. Under these conditions, attenuation of light by deep mixing and/or shading by other phytoplankton (which can grow much more rapidly under these conditions) will further reduce the growth rate of Trichodesmium, putting it at an even greater disadvantage. According to this hypothesis, Trichodesmium can only become dominant when the mixed layer is thin and the growth rate and biomass of other species is restricted by lack of DIN, which provides optimal (high) light conditions for Trichodesmium that maximize its growth rate. Regardless of the exact mechanism, it is abundantly clear that mixing strongly influences when and where Trichodesmium grows in the open ocean, and that blooms tend to occur where the mixed layer is thin and DIN concentrations are low [Capone et al., 1997; Hood et al., 2001].
 In this paper we model the distribution of Trichodesmium and rates of N2-fixation in the Atlantic using a coupled physical-biological model. Following Hood et al. , we hypothesize that Trichodesmium's fundamental physical, chemical, and ecological niche is defined by high light intensity, relatively weak vertical mixing, and low DIN concentrations, where the latter prevents the growth of other, faster growing, phytoplankton species. Further, we implicitly assume that although Fe and P limitation may place constraints upon the total amount of Trichodesmium biomass that can develop in any one location, these elements do not dictate when or where Trichodesmium occurs. This paper is therefore a test of the Hood et al.  hypothesis: We examine the predictions of our model and we determine the degree to which these factors can reproduce observed Trichodesmium distributions in the Atlantic. We argue here that changes in mixed layer depth (light) and DIN availability are, in fact, sufficient to explain the large-scale patterns. In addition, the model predicts that Trichodesmium and high rates of N2-fixation should occur in the Gulf of Guinea (off of central West Africa) and that N fluxes due to N2-fixation should have a significant impact on the distribution of phytoplankton in the open ocean.
2. Coupled Model
 The combined model consists of a six-compartment ecosystem model with an explicit representation of Trichodesmium coupled to an Atlantic implementation of the Miami Isopycnal Coordinate Model (MICOM). In the following subsections we describe these two models and how they are linked, as well as the forcing, boundary conditions, and parameter choices that were used to generate the solution described and discussed in section 3.
2.1. Ecosystem Model
 The biological model (Figure 1) is nitrogen based, and it includes six state variables: dissolved inorganic nitrogen (DIN), dissolved organic nitrogen (DON), phytoplankton (P), detritus (D), heterotrophs (H), and Trichodesmium (T). In this model, which is described in detail by Hood et al. , Trichodesmium is distinguished from phytoplankton by its ability to fix molecular nitrogen (N2), its slow growth rate and its almost complete immunity to heterotrophic (zooplankton) grazing. The growth rate of Trichodesmium in this model is entirely determined by light, and therefore also by mixed layer depth and phytoplankton concentrations which both influence the mean light levels in the euphotic zone. There is no Fe or P limitation.
 The dual wavelength optical model used by Hood et al.  is used here to calculate the average irradiance in each MICOM layer, which is then used to calculate phytoplankton and Trichodesmium growth rate. The average light is determined by integrating the irradiance in both wave bands over the layer in question and dividing by layer thickness. In this model, light is attenuated by both phytoplankton and Trichodesmium [Hood et al., 2001]. Although the light model includes both shortwave and longwave components, the latter is absorbed rapidly in the surface layer. Thus, for all practical purposes, there is only one band in the photosynthetically active radiation (PAR) range. There is also no feedback between this optical model and the radiative penetration in the physical model, which assumes that all radiation is absorbed in the mixed layer. The detrital sinking flux is calculated using a simple upstream, flux form advection scheme assuming a constant sinking rate.
Trichodesmium concentration is reported here in units of colonies/L in order to facilitate intercomparisons between the model estimates and direct measurements. Colony concentrations are calculated from the model-generated Trichodesmium mass (mmoles N/m3) by assuming an average colony nitrogen content of 2 μg N/colony, which is based upon direct measurements from tropical and subtropical western Atlantic waters [Carpenter, 1983]. The N2-fixation rate reported from the model is the net uptake rate of molecular nitrogen by Trichodesmium, i.e., the gross nitrogen uptake less any nitrogen taken up from the DIN pool.
2.2. Ecosystem Model Parameter Settings
 Except for those discussed below, all of the ecosystem model parameters are set as described by Hood et al. [2001, Table 1]. The parameters that are changed include the following: the natural mortality rate of Trichodesmium (ST), the grazing preferences for heterotrophic feeding on T, P, D, H, and DON, (ΦT, ΦP, ΦD, ΦH, and ΦDON, respectively), and the sinking rate of detritus (w).
 The values of ST and w were re-adjusted following essentially the same procedure as described by Hood et al. , except here the model was tuned to simultaneously reproduce satellite-derived, near-surface chlorophyll concentrations throughout the basin, and Trichodesmium colony concentrations in the western tropical and subtropical Atlantic. Specifically, this was done by first setting the Trichodesmium mortality rate, ST, to give 1–3 colonies/L between 17°N and 25°N along approximately 57.5°W in October/November. These concentrations are derived from measurements reported by Carpenter and Romans , which are generally consistent with the majority of the reports from this region [e.g., Carpenter and Price, 1977; Carpenter and Romans, 1991; Sander and Steven, 1973; Steven and Glombitza, 1972; Dunstan and Hosford, 1977]. Tuning to the observed Trichodesmium concentrations in this manner has the effect of setting the overall concentrations and rates of N2-fixation throughout the Atlantic basin. However, where and when Trichodesmium occurs, and relative concentrations in different ocean regions, are entirely emergent properties of the model.
 Once the colony concentrations are set, then w is adjusted to yield seasonal, basin-wide phytoplankton chlorophyll concentrations similar to chlorophyll concentrations derived from composite satellite (SeaWiFS) data. This model fitting is done subjectively and iteratively. The comparison between the model-generated and observed chlorophyll concentrations is described and discussed in detail in section 3.4. Adjusting the sinking rate of detritus in this manner effectively sets the nitrogen export flux so that it balances, on average, the total input of new nitrogen into the upper ocean (i.e., N2-fixation plus inputs from deep water due to mixing, upwelling and diffusion). The resulting parameter values are ST = 0.025 d−1 and w = 6 m d−1.
 In addition, ΦT was changed from 0 as per Hood et al.  to 0.01 in order to add a small amount of density-dependent grazing pressure (negative feedback) on Trichodesmium. This change had the effect of generally reducing the high Trichodesmium concentrations in the eastern equatorial Atlantic (e.g., in the Gulf of Guinea) which we deemed to be too high in initial model runs relative to concentrations on the western side of the basin. In addition, ΦP was similarly increased, from 0.25 to 0.3475, to provide stronger grazing pressure and reduce high phytoplankton concentrations in the equatorial upwelling regions. The other three grazing preferences, ΦD, ΦH, and ΦDON, were lowered from 0.25 to 0.2175 so that all of the preferences sum to 1.
2.3. General Circulation Model
 The 3-D general circulation model is the Miami Isopycnal Coordinate Model (MICOM) [Bleck and Smith, 1990; Bleck et al., 1992], whose vertical coordinate is potential density. The model is formulated as a vertical stack of layers, each obeying the shallow water equations and thus having vertically uniform dynamic, thermodynamic, and ultimately ecosystem properties. The uppermost layer is the interface between the atmosphere and the ocean, and acts as a Kraus-Turner bulk mixed layer [Kraus and Turner, 1967] with laterally varying thermodynamic properties. Below the mixed layer, advection and diffusion take place along isopycnal surfaces, with diapycnal mixing leading to an explicit exchange of mass and tracers between layers. The diapycnal flux is determined by the difference in concentration between layers and a stratification-dependent diffusion coefficient. In addition, interface smoothing in the model gives rise to isopycnal flux.
 For this application, MICOM is configured with a relatively coarse horizontal grid, 2° zonal by 2×cos(lat) mesh. At this resolution, the physical model resolves only the gross features of the ocean gyre systems. Thus we cannot expect to fully resolve the western boundary currents and the equatorial current systems. In addition, this low spatial resolution does not resolve the full spectrum of open ocean eddy variability, which gives rise to low overall eddy kinetic energy [McGillicuddy and Robinson, 1997; McGillicuddy et al., 1998; Mahadevan and Archer, 2000; Chassignet and Garraffo, 2001]. In an exploratory model development effort such as this, the advantage of a low-resolution model is that it allows us to run many (more than 90) iterations of the model for troubleshooting and biological model tuning, which is not feasible with current high-resolution, eddy permitting, and eddy resolving models. The model has 19 layers in the vertical, which are concentrated primarily in the upper ocean and tropical thermocline in order to resolve vertical biological structure and relevant physical processes.
 The model domain extends from 45°N to 20°S, 97°W to 13°E. The northern and southern boundaries are outfitted with 5° sponge layers with a tapered relaxation to monthly climatological layer thickness, salinity, and annual mean DIN concentrations from NODC analyzed fields from the Ocean Database 1998 CD-ROM set [Conkright et al., 1998]. The bathymetry is derived from NGDC 5-min gridded bottom topography data (“ETOPO5”) by averaging to the model's horizontal grid resolution. This averaged bathymetry resolves only major topographic features and shelf areas within the model domain. Islands are not fully resolved, for example, the Antilles Archipelago is represented as a shoal region extending up to about 2000 m depth. Vertical diffusion is implemented with a Nyquist frequency dependent mass transfer. Isopycnal diffusion of tracer quantities is set at 0.5 cm s−1 × the grid size, or 1. × 107 cm2 s−1. In MICOM, horizontal advection of salt and biological tracers is carried out using an upstream advection scheme which corrects for numerical diffusion [Smolarkiewicz, 1983].
 The model is forced using surface wind stress and speed, and air temperature and humidity from the COADS climatology [da Silva et al., 1994]. Precipitation and surface radiation are derived from the Oberhuber atlas [da Silva et al., 1994]. The radiation data from the latter (which provides daily averaged values with both long and shortwave components) are also used to specify the surface irradiance for the dual wavelength optical model that calculates the subsurface irradiance for the ecosystem model. River runoff is prescribed seasonally as an augmentation of the precipitation field for four freshwater sources: the Amazon, the Congo, the Orinocco, and the Mississippi. The flows are taken from Carton . Currently, these rivers do not act as sources of nutrients to the model ocean.
3. Results and Discussion
 In this section we discuss our main run solution, which was derived using the tuning procedure described in section 2.2, and we validate it against physical, chemical, and biological observations. In section 3.1 we compare the model-estimated mixed layer depth variability to seasonal climatological fields, and we describe how this influences the average light in the mixed layer. In section 3.2 we validate the modeled Trichodesmium distributions against all available direct measurements of Trichodesmium biomass. In section 3.3 we compare model-generated, meridional DIN sections with seasonal climatological DIN data from the World Ocean Database 1998 CD-ROM set [Conkright et al., 1998], and in section 3.4 we compare modeled phytoplankton concentrations in the mixed layer with SeaWiFS-derived near surface chlorophyll estimates. Finally, we conclude the results and discussion with section 3.5, where we describe the model-generated seasonality in Trichodesmium, phytoplankton, and DON concentrations along two meridional sections, focusing on successional patterns generated by the model.
3.1. Mixed Layer Depth and Light
 Proper representation of mixed layer depth (MLD) and its variability are crucial for modeling phytoplankton and Trichodesmium concentrations in the Atlantic (and in the ocean in general [McCreary et al., 1996, 2001; Hood et al., 2001, 2003]), because the MLD has a large impact upon the light and nutrient levels that are experienced by the autotrophs in the system. Shallow mixed layers generated by surface heating and detrainment tend to result in low nutrient and high light conditions, whereas deep mixed layers generated by surface cooling and entrainment give rise to the opposite. As discussed above, in our ecosystem model the latter will tend to lead to phytoplankton dominance in the mixed layer, whereas prolonged periods of the former will eventually allow Trichodesmium populations to increase near the surface.
 However, MLD can also be strongly influenced by upwelling and downwelling, which can lead to very different light and nutrient conditions. If the mixed layer in MICOM is at its minimum thickness (20 m), upwelling can inject nutrients into the surface layer which will result in a high light (thin mixed layer) and high nutrient condition. These circumstances will tend to arise in regions where there is continuous upwelling and strong subsurface stratification, such as the equator. In contrast, if the mixed layer deepens due to convergence of nutrient-depleted surface waters, then it can result in a low light and low nutrient condition. These circumstances will tend to arise in regions where there is continuous convergence, such as the central gyres. Thin mixed layers associated with continuous upwelling provide optimal light and nutrient conditions for phytoplankton growth in our ecosystem model, and therefore tend to result in phytoplankton dominance and low Trichodesmium concentrations. Thicker mixed layers associated with convergence and downwelling can result in conditions more favorable for Trichodesmium growth in our model, but these favorable conditions are ephemeral; that is, as the mixed layer deepens, increased light limitation progressively limits Trichodesmium growth.
 In the MLD analyses that follow we compare MLD variability generated by the model (Figure 2) with MLD estimated from monthly climatological temperature and salinity from NODC analyzed fields from the World Ocean Database 1998 CD-ROM set [Conkright et al., 1998] (Figure 3). Recall that MICOM estimates MLD using a Kraus-Turner-like model. Thus the two criteria are very different, but the patterns should be similar. The modeled light fields, however, are synoptic (Figure 4) to facilitate comparisons with the synoptic model output described in later sections.
 The modeled and observed patterns in MLD variability are basically similar at lower latitudes in winter (February), i.e., shallow MLDs along the northern coast of South America, in the Cape Verde/Sierra Leone region, and all along the coast of Africa in the Gulf of Guinea and southward (Figures 2 and 3). The model also reproduces the shallow mixed layers that are observed along the equator. However, in winter the model mixes too deeply (>100 m) at higher latitudes in the North Atlantic. Note in particular that the modeled mixed layer is much too deep (>100 m) in winter in the northern Gulf of Mexico and along the coast of Florida, and 20–30 m too deep throughout the Caribbean Sea.
 The relatively deep mixing generated by the model at higher latitudes in the Northern Hemisphere winter results in low mean light conditions (<10 W/m2) over much of the central and eastern North Atlantic (Figure 4). In contrast, the shallow MLDs (20 m) off of the North coast of South America and over large areas along the equator and off of Africa south of 10°N result in relatively high light conditions (>30 W/m2).
 In spring the MLDs in the model between 20°N and 30°N are shallower than observed (Figures 2 and 3) over much of the North Atlantic (<60 m). Although the pattern of MLD variability generated by the model is still essentially similar to the observations at low latitudes, there are some significant discrepancies. For example, the data reveal much more extensive shallow mixed layers all along the northern coast of South America compared to the model. Note also that in the model the MLDs are considerably deeper and extend over a broader area between the northeastern coast of Brazil and northwest Africa.
 Because of the shallow MLDs, the mean light levels in the model increase (to >20 W/m2) over much of the subtropical North Atlantic (including the Gulf of Mexico and Caribbean waters) in spring (Figure 4). However, the light levels remain relatively low (<20 W/m2) where the MLDs are deep in the center of the basin between 5°N and 25°N, and along the coast in the Gulf of Mexico, off of Florida, and in places along the coast of Central America. The light levels in the mixed layer off of the north coast of South America and over large areas along the equator and off of Africa remain relatively high (>35 W/m2).
 The agreement between the modeled and observed MLDs is probably best in summer (June) and fall (September) (Figures 2 and 3). The model reproduces all of the observed shallow regions during these time periods, though not in precise detail. The model also generates deeper MLDs in June and September in the Southern Hemisphere, and a band of relatively deep MLDs extending zonally between 10°N and 25°N, as observed. Note, however, that the shallow MLDs along the equator, which are associated with equatorial upwelling, extend too far to the west in the model compared to the observations. This discrepancy arises due to the relatively low resolution of the model, which results in overestimation of the strength and extent of equatorial upwelling.
Figure 4 shows that the average light in the mixed layer is relatively high in all of the regions where the mixed layer is shallow. In summer (June) the light levels are actually higher (>45 W/m−2) north of 20°N (in the open ocean and in the Gulf of Mexico) than in the coastal and open ocean upwelling regions, even though the mixed layer is at its minimum thickness in both. This happens because phytoplankton concentrations are much higher in the upwelling areas which results in more rapid light attenuation (see Figure 12 in section 3.4).
3.2. Observed Trichodesmium and Rates of N2-Fixation
 In this section we compile and summarize the available direct measurements of Trichodesmium biomass from the Atlantic between 30°N and 15°S, and compare them to the model-generated Trichodesmium concentrations. The measurements are reported in Table 1 along with the specific conversion factors that were applied to each data source to convert them to common units of colonies/L (see footnotes in Table 1). We also use this information to generate a crude seasonal/spatial map of Trichodesmium biomass (Figure 5) that directly compares the observed and modeled concentrations. Direct N2-fixation rate estimates are also available (and discussed below) in some cases, but the focus here is on biomass rather than the rates because there are many more biomass measurements.
Table 1. Observed Trichodesmium Concentrations in Atlantic Waters
Calculated from filament concentrations assuming 200 filaments per colony.
Calculated from cell concentrations assuming 100 cells per filament and 200 filaments per colony.
Calculated assuming 20–50 ng Chla per colony.
Calculated from nitrogen concentrations assuming 2 μg N per colony.
Figure 5 reveals that the measurements are very patchy in both space and time, and Table 1 shows that only a few time series exist. There are also far more measurements in the western Atlantic (especially the Caribbean). Moreover, comparisons with the model are confounded to some degree by the fact that surface accumulations of Trichodesmium populations often develop under stratified conditions in the field, which can give rise to extremely high measured concentrations which cannot be replicated by the model because it averages the concentrations over the minimum MICOM MLD of 20 m. We point out where these kinds of effects may be important.
3.2.1. Meridional and Zonal Extent of Trichodesmium and N2-Fixation
 The meridional extent of the Trichodesmium populations and rates of N2-fixation predicted by the model are essentially correct (Figures 6 and 7); that is, the model generates Trichodesmium distributions that are largely confined to tropical and subtropical waters as observed [Carpenter, 1983; Capone et al., 1997]. Although it has often been assumed that the northern extent of Trichodesmium populations is dictated by temperature (kinetic) control of their growth rate [Carpenter, 1983; Carpenter and Capone, 1992; Moore et al., 2002], our model does not require this mechanism to reproduce the observed range. Rather, variations in Trichodesmium biomass and rates of N2-fixation in the model are determined primarily by the depth and duration of winter mixing. At higher latitudes, deeper winter mixing results in lower light levels and higher DIN concentrations which favor phytoplankton growth. At lower latitudes, persistent net surface heating results in thinner mixed layers, higher mean light levels, and DIN depletion, which favors Trichodesmium growth. This dynamic in the model is consistent with recent observations which show that there is a strong correlation between MLD and N2-fixation rate in situ [Sañudo-Wilhelmy et al., 2001].
3.2.2. Spatial Pattern and Seasonality in the Gulf of Mexico and the South Atlantic Bight
Trichodesmium is commonly observed and highly variable in the Gulf of Mexico, with most accounts from summer and fall. Concentrations of 10–20 colonies/L are often observed in northwestern Gulf waters in summer, whereas they appear to be absent in winter (T. Villareal, personal communication, 2002). Krauk , for example, observed typical concentrations of 2–5 colonies/L in the northwestern Gulf in late July (24–28, 2000), but at one station concentrations were an order of magnitude higher. In the eastern Gulf off of the west coast of Florida, Lenes et al.  measured concentrations ranging from 2 to 22 colonies/L in surveys in June through September, with the maximum occurring in early July (1999). Walsh and Steidinger  report a mean concentration of 125 colonies/L in a bloom in early July (1980) in the same general area. An historical time series of Trichodesmium concentration measurements constructed by Walsh and Steidinger  reveals that blooms typically develop off the west coast of Florida in summer and fall, with concentrations varying between 0.5 and 50 colonies/L. However, this time series also shows several instances of bloom development in winter and spring. Lenes et al. and Walsh and Steidinger link bloom development to Fe deposition events.
Trichodesmium blooms also develop in the South Atlantic Bight (SAB) along the southeastern coast of the United States, with high concentrations most often in summer and fall. For example, using satellite remote sensing techniques, Subramaniam et al.  estimated Trichodesmium specific chlorophyll concentrations up to 3 mg m−3 in a coastal bloom off of Florida in October, which is equivalent to 20 –150 colonies/L. However, Dunstan and Hosford  reported significant concentrations (>5 colonies/L) in spring, as well as in summer and fall.
Figure 6 shows that the model predicts Trichodesmium concentrations of 1–3.5 colonies/L, which develop in summer and fall along the coasts in the Gulf of Mexico and the SAB, and very low concentrations in winter and spring. Note the development of two distinct patches in the Gulf in September: one off of southwest Florida and another dominating the western half of the region. These locations are consistent with the observations, and the concentrations are comparable to the observed “background” Trichodesmium concentrations in these waters. We speculate that the higher observed concentrations are associated with surface accumulations that develop in warm, stratified waters. As we mentioned above, our model cannot reproduce these conditions. We can, however, estimate what the model concentrations would be if it could: If the maximum model-generated mixed layer concentrations of 1–3.5 colonies/L in a 20-m MLD are concentrated in a layer 1 m thick, it would give 20–70 colonies/L, which is consistent with most of the higher measured concentrations in the Gulf and the SAB.
 The seasonality generated by the model is also generally consistent with the observed seasonal patterns. However, it appears that Trichodesmium blooms sometimes develop in the winter and spring in the Gulf and in the SAB (Walsh and Steidinger  and Dunstan and Hosford , respectively), which is not consistent with the model. This discrepancy may be due to the fact that the model substantially overestimates MLD in these waters in winter and spring, which would prohibit the development of Trichodesmium blooms during these seasons, and delay the onset of population increases in the summer. We also suggest that while blooms may occasionally develop in the winter and spring in the Gulf and SAB, this is probably unusual.
 Although Lenes et al.  and Walsh and Steidinger  attribute the Trichodesmium blooms off of the west coast of Florida to Fe deposition events, no such mechanism exists in our model; that is, the high concentrations that develop in the model in summer and fall are linked to stratification, increased light in the mixed layer, and depleted DIN concentrations, not Fe deposition. We believe that thin MLD and depleted DIN conditions are the first-order controls on Trichodesmium growth and therefore must be satisfied first for Trichodesmium to grow. However, it is also possible that the converse is true; that is, even if these physical conditions are met, a bloom may not develop in the absence of significant atmospheric Fe deposition.
3.2.3. Spatial Pattern and Seasonality in the Sargasso Sea, Caribbean, and the Northern Coast of South America
 Although numerous observations have been made of Trichodesmium in the Sargasso Sea and Caribbean waters over the last 20 years, these measurements are still very patchy in both space and time, and the concentrations vary tremendously. Carpenter and Romans  reported surface concentrations along a transect from Barbados to Bermuda in October/November. These data show about 1 colony/L between 10°N and 18°N, increasing to as much as 3 colonies/L between 18°N and 23°N, and then decreasing again to 1 colony/L or less approaching Bermuda. Comparison with the model results for September reveals essentially the same meridional pattern, with similar concentrations (Figure 6). Although the model was tuned to reproduce these observations between 18°N and 23°N, it was not tuned to reproduce the observed meridional pattern, which it does remarkably well.
Hulburt  published a February transect extending northeastward from Barbados to approximately 30°N, 48°W. North of 16°N they encountered Trichodesmium at only one station location, but southward of this location they measured individual filament concentrations equivalent to 5–10 colonies/L. Comparison with the model reveals essentially the same pattern, but with much lower concentrations, increasing to only about 0.5 colonies/L in the vicinity of Barbados. In another transect in the southwestern Caribbean Sea (running along the coast from the Panama Canal up along the coast of Nicaragua into the northern Caribbean), Hulburt  measured highly variable concentrations ranging from 0.150 to 4.5 colonies/L in early November (1965). The model predicts somewhat lower concentrations in this region in September, ranging from 0.25 to 0.75 colonies/L. As discussed above, the model overestimates the MLD throughout much of the Caribbean in winter and spring, which may lead to underestimation of Trichodesmium concentrations all year round (see also section 3.3.4).
 More recently, Orcutt et al.  measured colony concentrations along a transect near Bermuda in September of 1994 (between 26°N and 33°N) and July of 1995 (between 29°N and 35°N). Both transects show low Trichodesmium concentrations near Bermuda (0–0.025 colonies/L) with values increasing southward in September of 1994 to 0.250 colonies/L at 26°N. These observations are also consistent with the model, which predicts low concentrations near Bermuda all year round, and dramatic increases in colony concentrations just south of Bermuda (between 25°N and 30°N) in September. Although the model-estimated concentrations at 26°N are higher than observed (1–2 colonies/L estimated versus 0.250 colonies/L observed), Orcutt et al.  reported that a large fraction (up to 90% of the population) existed as free trichomes. If this is taken into account, then the observed Trichodesmium biomass at 26°N is roughly consistent with the model.
 Finally, compiling data from three broad area cruises in the Caribbean and the Sargasso Sea, Carpenter and Price  reported average Trichodesmium concentrations of 1–4 colonies/L in the Caribbean compared to 0.25–0.5 colonies/L in the Sargasso. Thus the range of values predicted by the model over all seasons in these two regions (<0.25 to 4 colonies/L) agrees quite well with the observations. However, the model predicts the highest concentrations in the southern Sargasso Sea, whereas the observations indicate the opposite. This pattern of higher concentrations in the Caribbean reported by Carpenter and Price  is also apparent in the data compiled by Carpenter and Romans . This discrepancy may also be related to the fact that the model generates MLDs that are too deep in the Caribbean in winter and spring.
 From a compilation of direct rate estimates from numerous cruises in the Caribbean and Sargasso Sea, Capone et al.  reported mean N2-fixation rates ranging from 0.004 to 0.228 mmoles N m−2 d−1 for Caribbean waters and mean values between 0.001 and 0.006 mmoles N m−2 d−1 for the Sargasso Sea. However, using some of the same data Carpenter and Romans  estimated substantially higher rates (ranging from 0.7 to 3.57 mmoles N m−2 d−1) for the same general area. Orcutt et al.  estimated maximum summer/fall rates at BATS of 0.01–0.1 mmoles N m−2 d−1 (depending upon the year in question) for colonies only, with values approaching zero in winter and spring.
 For comparison, the model generates integrated rates ranging from about 0.0 to 0.6 mmoles m−2 d−1 in September in these waters, with the highest values associated with the biomass maximum situated just north of the Greater Antilles, and lower values in the southeastern Caribbean and the Sargasso Sea (Figure 7). During other seasons the model-estimated rates drop to low levels in these waters. In the vicinity of BATS the model predicts low rates throughout the year (<0.06 mmoles m−2 d−1). Thus the range of rates generated by the model for the Caribbean and the Sargasso Sea are roughly comparable to those summarized by Capone et al. , and they are consistent with the colony-based estimates of Orcutt et al.  at BATS. The model-estimated rates are considerably lower than the estimates of Carpenter and Romans , but these investigators used very liberal assumptions in their calculations which have been questioned in a subsequent publication [see Lipschultz and Owens, 1996].
 The seasonal biomass cycle generated by the model in the Sargasso Sea and Caribbean waters is essentially the same as the seasonal cycle in the Gulf of Mexico. Trichodesmium biomass begins to increase in April/May following the late winter/early spring phytoplankton bloom in these waters and continues to increase through summer and fall, and then declines in December as the mixed layer deepens and mean light levels drop. This seasonal cycle is similar to that described by Hood et al. , and it is driven by the same physical and biological dynamics; that is, Trichodesmium growth rates increase and their biomass begins to accumulate after the late winter/early spring phytoplankton bloom declines due to DIN depletion. The combination of relatively shallow mixed layers in the late spring/early summer, high mixed layer light levels, and low DIN concentrations provide optimal conditions for Trichodesmium growth, while simultaneously inhibiting phytoplankton growth. As the summer progresses, increasing surface light intensities and surface heating maintain these optimal conditions for Trichodesmium. However, due to its slow growth rate, Trichodesmium biomass accumulates slowly, with maximum concentrations developing in October and November. Lower mean light levels and surface cooling in December results in deepening of the mixed layer, thereby shutting down Trichodesmium growth and mixing the accumulated Trichodesmium biomass downward, while simultaneously entraining DIN from depth. The latter sets the stage for the winter/spring phytoplankton bloom and a repeat of the seasonal cycle.
Figure 6 shows that this seasonal cycle in Trichodesmium concentration is strongly manifested in the model solution throughout the southwestern North Atlantic. As discussed by Hood et al. , Orcutt et al.'s  3-year time series (1995–1997) of Trichodesmium biomass and N2-fixation rate measurements from BATS reveals a seasonal cycle that is very similar to the model-predicted cycle. As discussed above, there is some evidence of this seasonality in the Gulf of Mexico as well. Similarly, high concentrations of Trichodesmium are most commonly observed in the southern Sargasso Sea in summer and fall [Carpenter and Romans, 1991; Carpenter et al., 2004, and references therein]. Farther south, the populations are more sustained over the seasons, but the seasonal dynamic of the winter trade winds does affect the distribution and growth of Trichodesmium. Even in the tropics proper, we expect peak growth and abundance during calm periods in the summer and fall (D. G. Capone, personal observations, 2002). Thus the model-predicted seasonal cycle appears to be consistent with observed patterns in the Sargasso Sea, the Gulf of Mexico, and the Caribbean Sea.
 These general observations are supported by a 25-month time series collected by Navarro  just south of Puerto Rico, which reveals a clear Trichodesmium abundance maximum in July–October 1996 (0.014–0.5 colonies/L) and a weaker maximum in October–November of 1997 (0.003–0.03 colonies/L). However, these data revealed no seasonal maximum in 1995. Moreover, Carpenter et al.  report higher concentrations east of Puerto Rico in spring (5–25 colonies/L in May/June 1994) than in fall (0.5–3.0 colonies/L in October/November 1996). Thus there appears to be considerable interannual variability, and it is clear that Trichodesmium blooms sometimes develop in these waters in the spring, which is not consistent with the model.
 Farther south, time series published by Steven and Glombitza  and Borstad  for a location about 9 km east of Barbados (13°15′N, 59°43′W), reveals strong oscillations in Trichodesmium concentrations with a period of about 3 months, but no evidence of the model-predicted seasonality. Rather, the variability in these waters appears to be dominated by the complex movements of the Guiana Current and the passage of Brazil Current eddies which pass through this region with a period of about 3 months [Carton and Chao, 1999]. Since the relatively low resolution of our model does not permit accurate formation and propagation of these eddies, we do not expect to see 3-month fluctuations in Trichodesmium biomass in the vicinity of Barbados as observed.
3.2.4. Eastern and Northeastern Coast of Brazil and Offshore Waters in the Vicinity
Carpenter et al.  report high concentrations (up to 8 colonies/L) in March–April of 1996 in two offshore patches off of the northeastern coast of Brazil (one centered at 14°N, 55°W, and another between 30°W and 45°W and 0° and 8°N, Table 1, Figure 5). Although the model does not produce high concentrations in the northern region (14°N, 55°W), it does produce a well-defined Trichodesmium population maximum in the southern region (30°–45°W and 0°–8°N) (Figure 6) that is part of a distinct band of high Trichodesmium concentration that extends westward along the equator between 0° and 5°N. These model results suggest that the southern patch reported by Carpenter et al.  may have been derived from offshore blooms that develop along the flanks of the equatorial upwelling region, which are then advected westward. Note, however, that the predicted concentrations are much lower than observed (<1 colony/L).
Goering et al.  reported both Trichodesmium biomass and rate measurements from two cruises in the region 0°–24°N, 45°–66°W (off of the northeastern coast of Brazil with most stations located in or near the Amazon River plume). Their Trichodesmium biomass measurements suggest concentrations of 1–10 colonies/L with generally higher values in fall [see also Calef and Grice, 1966], and they report a mean N2-fixation rate of about 0.01 mmoles N m−3 d−1 in fall and values near zero in spring. For comparison, the model predicts low Trichodesmium biomass and rates all along the northeastern coast of Brazil in spring (March) and distinctly elevated biomass and rates in summer and fall, with colony concentrations approaching 3 colonies/L and N2-fixation up to 0.02 mmoles N m−3 d−1 in June and September, respectively (Figures 6 and 7). Thus the model predicts Trichodesmium concentrations and N2-fixation rates in this region in summer and fall that are largely consistent with the observations, but it may underestimate values in spring.
 We know of only one data set from the Brazil coastal region between 5°S and 15°S. On a transect extending through the equator and along the east coast of South America in September–October, Tyrrell et al.  observed significant Trichodesmium concentrations (0.1–10 colonies/L) on the equator and south along the South American coast between 25°S and 35°S, but they did not observe Trichodesmium between 5°S and 15°S. This pattern is consistent with the model, which generates low concentrations (<0.25 colonies/L) off of the east coast of Brazil and higher concentrations in equatorial waters to the north (>0.50 colonies/L) in fall (Figure 6). We have no data to either confirm or refute the relatively high model-predicted Trichodesmium concentrations off the east coast of Brazil during other seasons (Figure 6).
 Finally, it should be noted that many of the locations discussed in this section are strongly influenced by freshwater inputs from the Amazon and Orinoco Rivers. We know from direct observations that Trichodesmium does not occur in low-salinity waters and is therefore absent in many places along the northeast coast of South America where one might otherwise expect it to occur (D. G. Capone and A. Subramaniam, unpublished observations, 2004). Although the freshwater input from these rivers is included in the physical model, our biological model has no mechanism that would prevent Trichodesmium from growing in low-salinity waters. In fact, in the model, stratification associated with freshwater input should tend to enhance Trichodesmium growth in regions where DIN is depleted. This may result in overestimation of Trichodesmium concentrations during periods when the river flow is high and coastal salinities are low (i.e., summer). This fact may help to explain some of the discrepancies discussed above.
3.2.5. Cape Verde/Sierra Leone Region and Equatorial Waters
 There are several reports of Trichodesmium in tropical waters off of the coast of northwest Africa. For example, Vallespinos  reported Trichodesmium between 6°N and 24°N from a cruise in November 1975 off of northwest Africa and compared them to measurements in the same general region in August 1971 by Margalef . In both studies, Trichodesmium was found in the upwelling region within 2° of the coast and extended offshore to at least 22°W, but the populations were actually associated with warm, nutrient-poor surface waters, derived from the south. The Trichodesmium concentrations were an order of magnitude higher in August (1971) compared to November (1975) (Table 1). Vallespinos concluded that standing crop fluctuates seasonally, being most abundant from August to November and absent during spring.
 These observations and conclusions are basically consistent with the model results for this region which show the highest concentrations and areal extent in the summer and fall and lowest concentrations and areal extent in the winter and spring (Figure 6). In the model, high Trichodesmium concentrations are also associated with nutrient-poor surface waters which are located adjacent to (south of) the upwelling center. However, Hernańdez-León et al.  reported Trichodesmium much farther offshore (between 30°W and 45°W) and farther north on a zonal (east-west) transect along 22°N in August–September. Trichodesmium blooms do not develop this far north or offshore in this region of the North Atlantic in the model. Again, we speculate that this discrepancy may be related to the fact that the model predicts MLDs that are much too deep in the winter months in this region of the North Atlantic which would tend to reduce the northward and seaward extent of Trichodesmium populations all year round.
Tyrrell et al.  report highly variable (0–10 colonies/L) Trichodesmium concentrations in September–October near the Canary Islands (20°N–30°N, 20°W), and high concentrations (1–10 colonies/L) in essentially the same areas where the model produces high concentrations in the Guinea Dome/Sierra Leone region (Figure 6). Their data also reveal generally high concentrations from 0° to 15°N, and 10°W to 25°W, with a patch in excess of 10 colonies/L just north of the equator. In Tyrrell et al.'s data set, Trichodesmium concentrations drop dramatically on the equator itself. These recent observations are consistent with earlier reports by Aleem , which revealed Trichodesmium in Sierra Leone coastal waters, and Bauerfeind , who reported the presence of Trichodesmium (up to 8.5 colonies/L) between 3°N and 2°S at 22°W in April through June of 1979.
 For comparison, the model generates maximum concentrations of 2–5 colonies/L in the Guinea Dome/Sierra Leone region in the fall, up to 9 colonies/L in summer, and between 1 and 4 colonies/L all year round just north of the equator at 10°W (Figure 6). The model also predicts distinctly lower concentrations in a zonal band between the equator and 5°S all year round, which is consistent with Tyrrell et al.'s  fall measurements. Thus the patterns generated by the model in the Guinea Dome/Sierra Leone region and in equatorial waters to the south generally agree with the direct observations, albeit with somewhat lower concentration ranges.
3.2.6. Gulf of Guinea and the South African Coast
 The model predicts higher Trichodesmium concentrations and rates of N2-fixation in the Gulf of Guinea in spring than anywhere else in the domain (>9 colonies/L, ≫0.02 mmoles N m−3 d−1), and according to the model the populations persist there all year round at fairly high levels (>2 colonies/L) (Figures 6 and 7). Dandonneau  reported that Trichodesmium is present all year in shelf waters in the vicinity of Abidjan (the western reaches of the Gulf of Guinea, 2°W–8°W), and that it is associated with the presence of oligotrophic, oceanic water masses. In addition, the harpacticoid copepod Macrosetella gracilis, whose nutrition and life cycle are closely associated with Trichodesmium spp. [O'Neil et al., 1996; O'Neil, 1997], has also been found, along with Trichodesmium, in the same general area (M. Pagano via R. Foster, personal communication, 2002). According to the model the highest Trichodesmium concentrations and N2-fixation rates actually develop farther east (0°E–10°E, 5°N–5°S). As discussed above, the model generates relatively thin MLDs and high average light levels throughout the year in these waters, as observed. The thin mixed layers in this region are due, at least in part, to high freshwater fluxes from direct precipitation and river runoff. Thus it appears that the physical conditions which give rise to the high Trichodesmium concentrations in the model actually exist in the Gulf of Guinea. However, the model may underestimate the strength of coastal upwelling in this region, which might lead to overestimation of Trichodesmium biomass and N2-fixation.
 Farther south, An  reported that Trichodesmium was “relatively common” along two meridional transects (11°S and 14°S) extending offshore from southern Africa (Angola) in April of 1968. In contrast, Tyrrell et al.'s.  southeastern transect (which ran from off the coast of Sierra Leone diagonally down to the South African coast at 20°S, in September–October) shows Trichodesmium (0.1–1 colonies/L) at only one station just south of the equator. No Trichodesmium was encountered farther south and east in the vicinity of An's transects. This difference probably reflects seasonal variations in the populations; that is, Trichodesmium is present in Austral fall and absent in Austral spring. The model results are consistent with both of these reports (Figure 6), predicting lower concentrations south of the equator in Austral Spring (September) as reported by Tyrrell, and elevated concentrations south of the equator in Austral fall (March) as observed by An, with concentrations ranging from 0.25 to 1 colonies/L.
3.2.7. Relation Between MLD and N2-Fixation
Figure 8 shows that the highest Trichodesmium biomass in the model occurs when the MLD is at or near the minimum thickness (i.e., <30 m) and the biomass gets progressively lower as MLD increases. This plot, however, also reveals that colony concentrations are often low when the mixed layer is thin and sometimes fairly high when the mixed layer is relatively thick. This lack of a tight correlation between MLD and biomass arises because (1) time is an important factor in determining whether or not high biomass develops; that is, the mixed layer must remain relatively thin long enough to allow significant Trichodesmium concentrations to accumulate; and (2) thin mixed layers that develop in response to upwelling can have high phytoplankton concentrations that lower light levels and prevent Trichodesmium growth. Conversely, relatively thick mixed layers (i.e., approaching 60 m, Figure 8) can give rise to high colony concentrations if they persist and become DIN depleted so that phytoplankton concentrations remain low and there is time to allow a significant Trichodesmium biomass to develop. Deep mixed layers can also contain elevated colony concentrations when high surface accumulations are mixed downward by a mixing event. Thus the model is capable of generating elevated Trichodesmium concentrations under windy/deep MLD conditions, which has been recently observed in the field (A. Subramaniam, personal communication, 2003).
3.3. Modeled and Observed DIN Concentrations
 The temporal and spatial patterns in mixed layer DIN concentrations generated by the model primarily reflect patterns in the physical processes which bring nutrients into surface waters (Figure 9); for example, DIN concentrations are enhanced in upwelling regions off of northwest Africa and along the equator. In addition, concentrations are substantially elevated in winter (January) prior to the spring phytoplankton bloom north of 20°N and in the Gulf of Mexico due to deep winter mixing. DIN concentrations are similarly elevated in the austral winter (June and September) south of 10°S.
 However, not all of the regions of enhanced DIN concentration in the model can be attributed to physical processes. Elevated concentrations (up to 0.4 mmoles N/m3) are also apparent in the Gulf of Mexico, the southern Sargasso Sea and generally throughout much of the southwestern North Atlantic in late summer and fall (September). A comparison between Figures 6, 7, and 9 shows that these regions of enhanced DIN concentration are coincident with regions of high Trichodesmium concentration and high rates of N2-fixation. Although the effects are more subtle, DIN concentrations are also significantly enhanced by N2-fixation off the coast of Africa where the model-estimated N2-fixation rates are high (i.e., in the Gulf of Guinea, and coastal Africa up into the Cape Verde/Sierra Leone region).
Figures 10 and 11 show sections of DIN concentration from the model (bottom panels) and from the NODC analyzed [Conkright et al., 1998] seasonal climatologies (top panels) along two meridional sections, one extending north from the coast of South America up through the Caribbean and the Sargasso Sea along 70°W (Figure 10), and another extending north through the equator and up along the coast of northwest Africa along 21°W (Figure 11). The 70°W sections show that the model generates the same broad patterns in the nutrient distributions as observed, with the nutricline extending up nearer to the surface at the northern and southern extremes of the transect, and generally lower values at depth in the vicinity of 20°N.
 The climatology also shows two distinct near-surface DIN maxima along the 70°W transect. One of these is centered at about 15°N in the middle of the Caribbean Sea, and it persists throughout the year. This feature, which is not represented in the model, has DIN concentrations which vary from 0.4 to >1.0 mmoles N/m3, and concentrations are highest in June and September. The second near-surface DIN anomaly is located between 20°N and 25°N, and appears only in June and September. This latter feature, which has concentrations in excess of 0.4 mmoles/m3 in September, does appear to be crudely represented in the model as a broad region of elevated DIN concentrations between 19°N and 25°N. In the model this anomaly is generated by Trichodesmium and N2-fixation, as discussed above. We speculate that the persistent near-surface DIN anomaly that is observed in the Caribbean Sea at about 15°N may also be generated by N2-fixation, but does not show up in the model because the model underestimates Trichodesmium and N2-fixation in these waters.
 A comparison between the model and observed DIN concentrations along 21°W (Figure 11) shows that the model reproduces the large-scale meridional patterns on the eastern side of the basin as well, i.e., with the nutricline depth gradually shoaling from 10°S to 10°N and then deepening rapidly at the northern end of the transect. The effects of equatorial upwelling are much more obvious in the model solution than they are in the observations. This difference between the model and the observations is probably due, in part, to the overly strong equatorial upwelling in the model, as discussed above, and in part to the smearing of the equatorial upwelling signal in the observations due to averaging and interpolation of the data. Note also that the nutricline is not as sharp in the model as observed. The fact that this discrepancy exists, even though the pycnocline structure in the model is as sharp as observed (not shown), suggests that remineralization of particulate matter may be occurring too slowly (deeply) in the model. Effects of Trichodesmium and N2-fixation are not obvious along this section in the model, but they do influence the near-surface DIN concentrations [Coles et al., 2004a].
3.4. Modeled and Observed Phytoplankton Concentrations
 The model reproduces the gross large-scale patterns and temporal variations in phytoplankton chlorophyll concentrations that are observed in SeaWiFS imagery (Figures 12 and 13). However, the model has a general tendency to underestimate coastal chlorophyll concentrations. Some of these discrepancies are probably due to the effects of nutrient inputs from rivers that are not represented in the model (e.g., in the vicinities of the Amazon and Orinoco River outflow plumes), and/or these turbid and DOM-rich plumes may cause SeaWiFS to overestimate chlorophyll concentrations [O'Reilly et al., 1998]. Off the coast of northwest Africa, a more likely explanation is that the model cannot properly represent the intense coastal upwelling very near shore due to its relatively low horizontal resolution. In contrast, the model consistently overestimates chlorophyll concentrations along the equator because the upwelling in the physical model is too vigorous and extends too far to the west along the equator (see discussion in section 3.1).
 Of course, the overall chlorophyll concentrations (and levels of primary productivity) generated by the model are a matter of choice; that is, we tuned chlorophyll concentrations in the mixed layer to roughly approximate the SeaWiFS observed values by adjusting export levels in the model. The general tendency of the model to underestimate chlorophyll concentration in some regions arises because the model had to be “tuned down” somewhat to avoid generating extremely high chlorophyll concentrations at the equator and adjacent waters. These kinds of discrepancies between modeled and observed physical fields and surface chlorophyll concentrations are commonly observed with low-resolution 3-dimensional biogeochemical models, and they have been discussed extensively in the literature [see Oschlies and Garcon, 1999; Oschlies, 2000, and references therein].
 The effects of new nitrogen inputs from N2-fixation on the model-generated chlorophyll concentrations can be see quite clearly in the southwestern North Atlantic in the fall (September) (compare Figures 6, 7, 9, and 12). Note in particular the distinctly elevated phytoplankton concentrations in the model solution in four regions: (1) around Cuba and Haiti extending northeastward into the open ocean; (2) off the northern coast of South America and in the Caribbean Sea; (3) along the western side of the Gulf of Mexico; and (4) off of the southwestern tip of Florida. Although more subtle, N2-fixation also enhances the model-generated phytoplankton concentrations off the coast of northwest Africa in the Cape Verde/Sierra Leone region, in the Gulf of Guinea, and in two broad zonal bands situated to the north and south of the equator. These effects are discussed in more detail in section 3.5 below.
 It is difficult to discern from these SeaWiFS surface chlorophyll maps (Figure 13) whether or not this enhancement of surface phytoplankton concentrations by N2-fixation actually occurs in nature. Figure 13 does not reveal any obvious enhancement of chlorophyll concentrations in the southern Sargasso Sea in the fall, as predicted by the model, even though the observed DIN sections suggest that there may actually be some enhancement of nutrient concentrations due to N2-fixation (i.e., observed elevated surface DIN concentrations in June and September between 20°N and 25°N in Figure 10, top panels). However, a recent EOF analysis of the SeaWiFS chlorophyll data has revealed a summertime anomaly (enhancement) of chlorophyll concentrations around Cuba and Haiti as predicted by the model [Coles et al., 2004b]. On the eastern side of the Atlantic basin, all we can say for certain about the effects of N2-fixation is that the modeled chlorophyll concentrations are too low compared to SeaWiFS in the Gulf of Guinea in model runs without input of new nitrogen from N2-fixation [see Coles et al., 2004a].
 This general sequence of events, where new nitrogen inputs from Trichodesmium blooms give rise to subsequent increases in other phytoplankton species, has been observed and discussed previously. For example, Burford et al.  argue that Trichodesmium enhances the growth of other algae. Devassy et al.  provide a particularly good example of phytoplankton succession following a Trichodesmium bloom, suggesting that nutrient inputs and “conditioning” provided by Trichodesmium promotes the growth of diatoms, specifically that of Nitzschia colosterium [Devassy, 1987]. In contrast, Revelante et al.  discuss specific floristic groups associated with Trichodesmium and allude to marked increases in dinoflagellates following Trichodesmium blooms.
 Recent work in the Gulf of Mexico suggests that inputs of new nitrogen from Trichodesmium and N2-fixation may be responsible for initiating harmful algal blooms [Lenes et al., 2001; Walsh and Steidinger, 2001; J. J. Walsh, personal communication, 2000]. Specifically, Walsh and Steidinger hypothesize that N2-fixation supplies the excess nitrogen that is required to support Karenia brevis (formerly known as Gymnodinium breve) blooms. Although our model does not include specific phytoplankton groups, the enhancement of phytoplankton concentrations due to new nitrogen inputs from N2-fixation in the model is essentially the same kind of effect; that is, N2-fixation supplies new nitrogen which stimulates phytoplankton growth. In the model this stimulation of phytoplankton growth happens only under stratified conditions that are conducive to Trichodesmium growth. These are exactly the kind of conditions under which flagellate and dinoflagellate blooms occur. Perhaps it is only a coincidence, but Figure 12 shows that the model predicts a N2-fixation induced “bloom” off of the southwestern tip of Florida in the fall in the same general area where K. brevis blooms occur, and in the western Gulf as well.
3.5. Meridional Sections
 The seasonality in Trichodesmium concentrations discussed above can be seen very clearly along 70°W (Figure 14), i.e., low concentrations in winter/spring (January and March) and highest concentrations in fall (September). Note that the increased Trichodesmium concentrations in June (at about 20°N, just north of Haiti and the Dominican Republic) are located in waters with relatively low DIN (Figure 9) and phytoplankton concentrations. By September the Trichodesmium biomass has increased to more than 4 col/L, and the phytoplankton, DIN (Figure 9) and DON concentrations have increased in the mixed layer as well due to new nitrogen inputs from N2-fixation, as discussed above.
Figure 14 also shows that the vertical distribution of Trichodesmium and phytoplankton along this transect is consistent with direct observations from these same waters [Hood et al., 2001; Carpenter et al., 2004], and our general understanding of the relationship between Trichodesmium and phytoplankton distributions; that is, Trichodesmium populations tend to be restricted to the upper 50 m of the water column where light levels are high and DIN concentrations are low [Capone et al., 1997]. In some sections we also see the development of subsurface phytoplankton maxima at the base of the mixed layer and the top of the nutricline, just below Trichodesmium. As discussed by Hood et al. , this happens in the model because Trichodesmium growth is maximized near the surface where irradiance is highest, whereas phytoplankton growth tends to be constrained by the availability of DIN supplied from depth.
 The section along 21°W reveals a more complicated pattern (Figure 15). One can discern three distinct Trichodesmium biomass maxima in March and June: one on either side of the equatorial upwelling region and a third off of northwest Africa in the Cape Verde/Sierra Leone region. In January and September, only two of the three maxima are apparent, with the southernmost feature absent. In addition, the Trichodesmium maximum off of northwest Africa is much more pronounced in summer and fall (June and September) than it is in winter and spring (January and March).
 A comparison of the Trichodesmium, phytoplankton, and DIN sections (Figures 11 and 15) reveals that these regions of elevated Trichodesmium are located adjacent to regions where the phytoplankton biomass and DIN concentrations are highest. Thus the patterns differ somewhat from what we observe along 70°W where all of the maxima coincide. For example, in June the two Trichodesmium concentration maxima on either side of the equator are situated just north and south of the phytoplankton and DIN maxima generated by equatorial upwelling. A similar pattern can be seen off of northwest Africa in June where the highest Trichodesmium concentrations are located just south of the highest phytoplankton and DIN concentrations. It should be emphasized, however, that there is also considerable overlap in these distributions, with Trichodesmium generally increasing as phytoplankton and DIN concentrations decline. Also note that the model generates distinct subsurface phytoplankton biomass maxima in all seasons, and that the regions of high Trichodesmium are always located above these subsurface features.
 The DON sections in Figure 15 also reveal distinct maxima. Some of these appear to be associated with elevated phytoplankton biomass and are likely generated by recycling of organic nitrogen derived from new nitrogen (DIN) inputs from upwelling. However, others (e.g., the DON maxima in the mixed layer in June and September between 10°N and 15°N) are clearly associated with elevated Trichodesmium concentrations and are likely generated by direct DON exudation from Trichodesmium. The latter is consistent with the interpretation of Vidal et al. , who argue that elevated DON off of northwest Africa is derived from N2-fixation.
 These model results are consistent with our current understanding of phytoplankton species succession in these waters [Margalef, 1963a, 1963b]. Margalef proposed that Trichodesmium populations increase along the periphery of upwelling regions in a time/space successional sequence after bloom forming species, such as chain-forming diatoms, have depleted surface nutrient concentrations from recently upwelled waters. We interpret the patterns along 21°W as a manifestation of this successional sequence; that is, we see an upwelling-induced phytoplankton bloom at the equator and off of northwest Africa in all of the sections. This bloom happens because the upwelling brings DIN to the surface and injects it into a thin mixed layer (high mean light). As this water moves laterally away from the upwelling center, the phytoplankton concentrations drop because they deplete the surface DIN concentrations and their growth becomes nutrient-limited. This, in turn, provides an opportunity for Trichodesmium populations to increase along the flanks of the upwelling region where the mixed layer is still thin (high light), and DIN and phytoplankton concentrations are lower. However, as the mixed layer gets progressively thicker as water moves laterally away from the upwelling region, the growth rate and biomass of Trichodesmium ultimately declines due to light limitation.
4. Summary and Conclusions
 In this paper we have attempted to model the distribution of Trichodesmium and rates of N2-fixation in the Atlantic using a coupled physical-biological model, and validate these results using available observations. Following Hood et al. , we have hypothesized that Trichodesmium's fundamental physical, chemical, and ecological niche is defined by high light intensity, relatively weak vertical mixing, and low DIN concentrations, where the latter prevents the growth of other, faster growing, phytoplankton species. Further, we have assumed that there is no temperature control of Trichodesmium growth rate in our model and that Fe and P limitation do not dictate when or where Trichodesmium occurs.
 In spite of these simplifying assumptions, the model appears to reproduce the observed large-scale (meridional), Trichodesmium distribution pattern in the Atlantic; that is, Trichodesmium occurs only in subtropical and tropical waters and concentrations are highest in the latter [Capone et al., 1997]. The model does this without invoking any temperature dependence or mechanical influence on Trichodesmium growth rate. Rather, it reproduces the observed distribution in response to meridional gradients in MLD and MLD variability.
 We have compiled a fairly large number of measurements from the southwestern North Atlantic which show that the model is reproducing major aspects of the observed temporal and spatial variability in Trichodesmium populations, i.e., highest concentrations in summer and fall and distinct population maxima in the Gulf of Mexico and the southern Sargasso Sea/Northern Caribbean. However, the model does not generate the extreme high densities that are sometimes observed in these waters, which we attribute to the formation of near-surface accumulations that cannot be reproduced by the model.
 Although we have fewer measurements to compare with, it is clear that the model is generating distinctly elevated Trichodesmium concentrations in locations where it has been observed off of northwest Africa, Cape Verde/Sierra Leone, and in equatorial waters. There is also some evidence of Trichodesmium blooms occurring off of the coast of South Africa as modeled. Perhaps the biggest question raised by the model-predicted Trichodesmium distributions is whether or not the high population densities and high rates of N2-fixation generated by the model in the Gulf of Guinea are correct. If the concentrations and rates in this area are as high and seasonally persistent as the model suggests, then it is likely that this region is a globally significant center of N2-fixation that needs to be explicitly considered in global N2-fixation rate estimates.
 Our comparisons also reveal some clear discrepancies between the model and direct measurements of Trichodesmium concentration. In some cases, these can be linked to infidelities in the physical model's representation of the mixed layer depth or its temporal evolution. For example, although we expect and observe seasonality in Trichodesmium populations in Caribbean waters due to seasonal changes in the strength of the trade winds, it appears that the seasonal cycle generated by the model is too strong due to overly deep winter mixing. Another conclusion that can be drawn from these comparisons is that the model has a general tendency to underestimate the observed Trichodesmium populations and rates of N2-fixation. In some locations, for example, in the open ocean off of the northeastern coast of Brazil, the model produces almost no Trichodesmium biomass and very low rates in regions where direct measurements clearly show substantial populations and high rates of N2-fixation.
 The temporal and spatial patterns in mixed layer DIN concentrations generated by the model reveal regions of enhanced surface DIN concentrations that can be clearly attributed to inputs of new nitrogen from Trichodesmium and N2-fixation. In particular, elevated concentrations are apparent in the Gulf of Mexico, the southern Sargasso Sea, and throughout much of the southwestern North Atlantic in the fall. Although the effects are more subtle, DIN concentrations are also significantly enhanced by N2-fixation off the coast of Africa where the model-estimated rates are high. Interestingly, climatological meridional DIN sections in the western Atlantic appear to confirm some of these model predictions.
 The effects of new nitrogen inputs from N2-fixation on the model-generated phytoplankton fields are readily apparent in the Gulf of Mexico, the Caribbean Sea, and the southern Sargasso Sea in the fall. The model-predicted enhancement of chlorophyll concentrations in the southern Sargasso Sea due to N2-fixation is not readily apparent in the climatological SeaWiFS data presented in this paper (Figure 13). However, a recent EOF analysis of SeaWiFS chlorophyll data has revealed enhancement in this region as predicted by the model [Coles et al., 2004b]. Moreover, there are several publications and accounts from other areas which suggest that new nitrogen inputs from Trichodesmium influence phytoplankton concentrations [e.g., Burford et al., 1995; Devassy et al., 1979; Devassy, 1987; Revelante et al., 1982; Lenes et al., 2001; Walsh and Steidinger, 2001; J. J. Walsh, personal communication, 2000].
 Stepping back and looking at the full sequence generated by the model reveals a three-step succession where (1) elevated DIN concentrations stimulate phytoplankton growth, which is followed by (2) Trichodesmium growth after DIN depletion, which is then followed by (3) enhanced phytoplankton growth due to new nitrogen inputs from N2-fixation. Although we do not resolve different phytoplankton species or forms of DIN in our model, we interpret this sequence as representing something like a diatom-Trichodesmium-flagellate succession, where upwelling or mixing stimulates a strong diatom bloom in a nitrate-rich environment and Trichodesmium stimulates a weaker flagellate bloom in a stratified ammonium, urea, and DON rich environment.
 The results presented in this paper lead us to conclude that our model includes the fundamental factors that control spatial and temporal variations in Trichodesmium concentrations and rates of N2-fixation in the Atlantic Ocean. Moreover, it appears that our model also reproduces some of the major effects that these diazotrophically derived inputs of new nitrogen have on the pelagic ecosystem. Thus we do not reject the Hood et al.  hypothesis; that is, we believe that Trichodesmium's fundamental physical, chemical, and ecological niche is defined by high light intensity, relatively weak vertical mixing, and low DIN concentrations, where the latter prevents the growth of other, faster growing, phytoplankton species. Further, we conclude that although Fe and P limitation may place constraints upon the total amount of Trichodesmium biomass that can develop in any one location, these elements do not dictate when or where Trichodesmium occurs.
 We would like to thank Ajit Subramaniam and Edward Carpenter for providing many of the references that were used to validate the modeled-generated Trichodesmium distributions, and Mercedes Pascual for helpful discussions on various aspects of this work. We thank Toby Tyrrell for generously allowing us to use his Atlantic transect data to validate our model before his data were published, and Tracy Villareal for assisting with efforts to validate the model in the Gulf of Mexico. We also thank Cara Wilson for generating the SeaWiFS chlorophyll composites shown in Figure 12. This work was supported by an NSF Biocomplexity Initiative grant to R. Hood, V. Coles (OCE-9981218), and D. Capone (OCE-9981371). This paper represents UMCES contribution 3743.