Notice: Wiley Online Library will be unavailable on Saturday 30th July 2016 from 08:00-11:00 BST / 03:00-06:00 EST / 15:00-18:00 SGT for essential maintenance. Apologies for the inconvenience.
 Nitrogen fixing organisms, such as the cyanobacterium Trichodesmium, directly affect the oceanic nutrient inventory through the addition of new nitrogen to the ocean ecosystem and therefore have an important role in the strength and functioning of the biological carbon pump. Nonetheless, little is known about the distribution of Trichodesmium beyond limited shipboard observations. Even less is known about the occurrence and characteristics of very intense, transient blooms of this organism that have been observed historically throughout the world oceans. A new method for discriminating the occurrence of blooms from satellite ocean color data is used here to make the first global maps of Trichodesmium bloom occurrence and to examine their spatial and temporal distribution. As expected, Trichodesmium blooms are rare, occurring <5–10% of the time over most of the tropical and subtropical oceans for the time period examined (1998–2003). Areas of greatest persistence are found in the eastern tropical Pacific and the Arabian Sea, and reach recurrence levels of >30%. Many of the retrieved patterns are consistent with previously reported blooms, though differences exist. A strong seasonal cycle is observed in the Indian Ocean, probably related to monsoonal forcing, with weaker seasonal changes elsewhere. Estimated global nitrogen fixation rates by Trichodesmium blooms is ∼42 Tg N yr−1 which is biogeochemically significant on regional and global scales. Further, an estimate of the rate of Trichodesmium nitrogen fixation under nonbloom conditions is an additional ∼20 Tg N yr−1 suggesting that Trichodesmium is likely the dominant organism in the global ocean new nitrogen budget.
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.
 Regional and global estimates of N2 fixation rates have a large degree of uncertainty which is likely caused by methodological reasons. Areal estimates of N2 fixation are often scaled up on the basis of point measurements of N2 fixation [i.e., Capone et al., 2005] and spatial integration scales are not well defined and often differ from study to study (as pointed out by Hansell et al. ). Further, these rates generally do not include contributions due to infrequent, yet large blooms of Trichodesmium, owing to the lack of knowledge about their distribution in space and time. Observations reported in the literature document extremely extensive and intense blooms in many parts of the ocean [Carpenter, 1983; Capone and Carpenter, 1992, and references therein], but no comprehensive descriptive or quantitative assessment of the distribution of these blooms has been made to date. Geochemical estimates of net N2 fixation provide one way of accounting for this shortcoming, but impart their own biases [Michaels et al., 1996; Montoya et al., 1996; Carpenter et al., 1997; Gruber and Sarmiento, 1997; Karl et al., 1997]. Recent numerical modeling efforts explicitly including Trichodesmium N2 fixation provide another approach for estimating areal rates and biomass, but are sensitive to details of model parameterization [Doney, 1999; Hood et al., 2001, 2004; Moore et al., 2002]. Divergence in each of these approach's results is not unexpected owing to the extreme differences in the methods applied.
 Satellite remote sensing of Trichodesmium provides another approach that falls somewhere between the methods described above. Indirect approaches have yielded encouraging results consistent with observations and emphasize the importance of oceanic N fixation [Coles et al., 2004]. Early efforts at directly parameterizing Trichodesmium bio-optics worked for regional studies [e.g., Dupouy, 1992; Subramaniam et al., 1999, 2002], but were not appropriate for global scales. Westberry et al.  recently described a robust method for mapping the occurrence of Trichodesmium blooms that was independently validated with a globally representative data set (see section 2.2). Here this method is applied to describe the spatiotemporal distribution of Trichodesmium blooms in the tropical and subtropical oceans. Their occurrence and persistence is compared to previous observations of Trichodesmium blooms. The relationship between bulk chlorophyll (Chl a) and Trichodesmium blooms is investigated and their effect on the bio-optical assumption is discussed. Lastly, the contribution that Trichodesmium blooms make to global N2 fixation rates is estimated using physiological values taken from the literature.
2.1. Data Sources
 Ocean color imagery was provided by the Sea-viewing Wide Field of view Sensor (SeaWiFS). For the global analyses, 8-day Level-3 binned Global Area Coverage (GAC) composites for the time period January 1998 through December 2003 were used. The native spatial resolution of these files is ∼9 km at the equator, resulting in a grid size of 2096 × 4320. The data have been regridded to 1/4° (∼27 km) spatial resolution and only data between 45°N and 45°S are used. Normalized water leaving radiances were extracted at 5 wave bands in the visible (412, 443, 490, 510, and 555 nm) and converted to remote sensing reflectance, Rrs (0−, λ), just below the air-water interface. Derived atmospheric correction parameters were also extracted for later use in determining atmospheric contamination; ɛ78, the ratio of aerosol radiances in the short and long wavelengths (SeaWiFS band 7 and band 8), and Å510, the Angstrom coefficient at 510 nm. In addition, global fields of sea surface temperature (SST) were obtained from the Advanced Very High Resolution Radiometer (AVHRR) Pathfinder project (http://podaac-www.jpl.nasa.gov/) for the same time periods. These data were also binned and interpolated to the same 8-day, 1/4° domain as the SeaWiFS data.
2.2. Model Implementation
 The SeaWiFS Rrs(0−, λ) were subsequently used as input to the model of Westberry et al.  to create corresponding maps of Trichodesmium bloom occurrence. Briefly, the model provides an index for the presence of Trichodesmium biomass above a threshold value of 3200 trichomes L−1 (equivalent to 0.8 mg/m3Trichodesmium-specific Chl a found using average values of 200 trichomes per colony and 50 ng Chl a per colony [Carpenter, 1983; Subramaniam et al., 2002]), allowing for a “presence” or “absence” bloom designation. The robustness of this “bloom” estimate has been independently validated using both in situ and satellite observations of Rrs(0−, λ) and Trichodesmium abundance [Westberry et al., 2005]. In the in situ model development data set, 92% of observed bloom values (>3200 trichomes L−1) were correctly identified and 16% of nonbloom observations were falsely identified as blooms. Model performance in the independent satellite–in situ validation data set correctly described 76% of bloom observations and 29% false positive bloom retrievals. Given the small size of the model development and validation data sets, no consistent pattern in the location or timing of these “false blooms” is found. Further details describing model performance and validation is given by Westberry et al. .
 The resulting Trichodesmium bloom occurrence maps have undergone a series of image processing steps. A morphological erosion filter was employed to eliminate single bloom pixels surrounded by nonbloom retrievals [e.g., Brown and Yoder, 1994]. The reasoning is that single isolated bloom pixels amid nonbloom pixels are likely to be spurious or false positive retrievals. Further, a morphological closing operator was applied to “smooth” the features by filling in small holes using a disk of radius equal to one pixel as the structuring element [Soille, 2003]. A series of masks were also applied to the blooms maps to exclude regions of uncertain quality. Pixels with SST < 23.5C were neglected by evaluation of AVHRR SST at each location. Additionally, pixels with suspect atmospheric correction were also discarded. This was done by eliminating pixels where values of ɛ78 > 1.1 and/or A510 > 0.5. These values are upper bounds for a common maritime aerosol atmosphere (Bryan Franz, personal communication, 2005). As a final processing step, a depth mask was applied to neglect retrievals in shallow water (<100 meters) which are more difficult to interpret owing to complex optical properties of the coastal ocean and near-coastal atmospheric contamination.
 The overall effect of the masking procedures is small. While the combined masks eliminate positive Trichodesmium bloom retrievals, the overall pattern remains unchanged. Table 1 shows the extent of the masks and of changing those masks for the tropical latitudes where most blooms are found (20°S to 20°N). The SST mask is the most significant, eliminating ∼6–13% of retrieved bloom pixels in this region. If we consider higher latitudes, the SST mask excludes a much higher percentage of bloom identified pixels (not shown), particularly in the high latitudes in the Southern Hemisphere, where it is certain Trichodesmium cannot grow owing to physiological constraints (not shown) [Capone et al., 1997; Karl et al., 2002]. Both the depth mask and the atmospheric correction mask each eliminate ∼4–10% of additional bloom pixels. There is very little seasonality in the total fraction of pixels masked (Table 1). On average, approximately 23 ± 3% of classified bloom pixels were masked owing to the SST, atmospheric, and depth masks. In practice, most of these outliers were overlapping such that they met more than one of the criteria for masking.
Table 1. Percentage of Positively Identified Bloom Pixels (Within 20°S–20°N) Masked by Various Procedures by Seasona
SST = 21°
SST = 23.5°
SST = 25°
Atm. Corr. Mask
Also shown are results for changing SST thresholds (23.5°C is criterion used in this paper). Depth mask corresponds to water depths <100 m; atmospheric contamination mask is described in text.
3.2 ± 1.6
6.1 ± 2.0
11.5 ± 2.5
4.1 ± 0.6
9.6 ± 2.6
5.3 ± 1.2
12.2 ± 3.6
25.7 ± 6.5
4.1 ± 0.6
5.0 ± 2.2
6.8 ± 1.8
13.9 ± 3.2
24.5 ± 5.1
4.0 ± 0.5
9.3 ± 3.9
4.0 ± 1.1
10.4 ± 4.1
21.6 ± 6.1
3.6 ± 0.8
7.7 ± 2.6
Trichodesmium bloom occurrence maps were created for January 1998 to December 2003. The average percent time when a Trichodesmium bloom occurs can be calculated by sequentially summing the occurrence maps and dividing by the total number of clear sky observations. However, to account for biases due to clouds, this ratio is further scaled to the frequency of clear skies at each particular location (Figure 1). In general, blooms are rare, occurring less than 5% of the time for most locations. Maximum values approach 35% and are found in the northern Arabian Sea and in the eastern Pacific centered on 10°S and 120°W indicating that bloom quantities of Trichodesmium populate these regions approximately one third of the time. Other areas of persistent bloom occurrence (∼10% of the time) are the Caribbean, the southern Indian Ocean, the eastern tropical Atlantic and the eastern tropical north Pacific. The rest of the ocean sees Trichodesmium blooms infrequently with ∼70% of the tropical and subtropical surface ocean (45°S–45°N) experiencing blooms less than 5% of the time (Figure 2). The cumulative probability distribution function also indicates 90% of the ocean between 45°S and 45°N experiences blooms less than 10% of the time while 30% of the ocean examined never sees a Trichodesmium bloom as defined here.
 Over broad regions, seasonal patterns are evident in bloom occurrence that appear out of phase from one ocean basin to another (Figures 3a–3d). The Indian Ocean, whose atmospheric and oceanic responses are dominated by monsoonal flow, shows dramatic changes through the year. Maxima in persistence are observed during the Fall (September–November) in the Arabian Sea and along the western boundary near Oman and Somalia. These are the highest values found anywhere and show that blooms are observed ∼45% of the time during these months. A strong bloom persistence signal is also found in the winter, and in fact, the most widespread and persistent blooms in the Arabian Sea straddle the seasons (October, November, December (not shown)). This time period corresponds to the fall intermonsoon period and onset of the northeast winter monsoon (not upwelling favorable). It is possible that the decrease in upwelling during this period creates a N-starved environment. Satellite coverage during the summer is problematic owing to dust and clouds and it is difficult to diagnose bloom occurrence during this period [e.g., Banzon et al., 2004].
 In the Pacific (Figures 3a–3d), blooms are present throughout the year south of the equator growing to ∼5.4 × 106 km2 during the boreal winter (December–February) and they are about half this size during the fall (September–November). Maximum persistence is >40% in very small areas and are ∼20–30% over large areas of this region during the winter. Also in the eastern tropical Pacific, the bloom area off of Central America exhibits a strong seasonality, virtually disappearing during the summer. During the winter, Trichodesmium blooms occur up to ∼25% of the time. Bloom features in this area do not correspond exactly with the strong Chl a signal in the Gulf of Tehuauntepec or the Costa Rica Dome [Fiedler, 2002; Xie et al., 2005]. Trichodesmium bloom occurrence is ∼10% of the time around the Costa Rica Dome and appears on the flanks of the Chl a signal off of southern Mexico, except during the winter when much of the Chl a signal is coincident with retrieved Trichodesmium blooms (not shown).
 The Atlantic Ocean exhibits comparatively fewer blooms (persistence <20%) and a correspondingly smaller degree of seasonality. In the summer and fall, Trichodesmium blooms are found in the equatorial Atlantic as well as the Gulf Stream and features in the Caribbean and western equatorial Atlantic, are strongest in the winter and spring, but still only exhibit blooms <20% of the time (Figure 3). Features along the African coast are more transient and ill-defined possibly owing to dust contamination issues in the satellite imagery.
 Interannual variability in Trichodesmium bloom persistence exists in each ocean basin as well (Figure 4). By summing the Trichodesmium bloom retrievals zonally in 2° bands, the areal percent coverage of Trichodesmium blooms within each 2° band can be expressed. Figures 4a–4c shows this quantity throughout the first 6 years of the SeaWiFS mission in each ocean basin. The Pacific Ocean exhibits a large degree of interannual variability with maxima in intensity and zonal extent of blooms in 2000. The years 2001–2002 also have periods of widespread Trichodesmium blooms, while blooms in 1998–1999 and 2003 occur less frequently over smaller zonal scales. Most of the blooms in the Pacific occur south of the equator. Also plotted in Figure 4a is the monthly averaged Southern Oscillation Index (SOI) which gives a measure of interannual climate variability in the Pacific. There is a rough correspondence between relative changes in the amount of Trichodesmium blooms and the strength of the SOI, where positive excursions in the SOI correspond to greater areal extent of Trichodesmium blooms. This is especially evident in the beginning of the record, which is a strong El Niño–La Niña transition period. Other large-scale changes in the surface ocean biota have been well documented during this period [e.g., Behrenfeld et al., 2001]. However, the ENSO cycle has been more stable since this time and the levels of corresponding variability in bloom occurrence are also more stable. If the areal coverage is integrated meridionally, the correlation coefficient with the SOI is 0.25 (r value; significant at the 95% c.l.) where high SOI corresponds to higher Trichodesmium bloom occurrence.
 The Atlantic Ocean shows the least extensive or persistent Trichodesmium bloom occurrences (Figure 4b). Maxima in areal bloom coverage are also south of the equator but cover only ∼10% of a given zonal band. Variability in both latitude and time is more uniform than other ocean basins. Overlaid on the Atlantic pattern (Figure 4b) is the monthly averaged North Atlantic Oscillation (NAO) index which shows no obvious correspondence between the NAO and bloom coverage. Integrating bloom coverage meridionally as before yields a weak correlation with the NAO (r = 0.11, not significant at 95% c.l.). It seems likely that bloom formation is responding to more local forcing (i.e., PO4 input from the Amazon or aeolian dust deposition from North Africa).
 Blooms in the Indian Ocean appear regularly during the fall and winter, but with a large degree of interannual variability embedded in the seasonal pattern (Figure 4c). As in the global persistence map (Figure 1), most of this ocean sees very little coverage by Trichodesmium blooms, on the order of <5% of each 2° band in this case. Latitudes where blooms frequently occur (>30% of the time) correspond to the northern Arabian Sea and to a lesser extent around 18°S, on the western margin of the Indian ocean near Madagascar (Figure 1). These signals are out of phase with one another with maxima north of the equator closer to winter, while the maxima in the southern Indian Ocean appear in the Northern Hemisphere summer (Figure 4c). The Arabian Sea pattern is also out of phase with the evolution of the Chl a pattern as Chl a concentrations reach their peak in July–August for this region [Banzon et al., 2004]. Also plotted in Figure 4c is another index of interannual variability, the Dipole Mode Index (DMI) [Saji et al., 1999; Saji and Yamagata, 2002]. The DMI is a measure of the east-west SST gradient across the Indian Ocean and has been related to interannual variability in monsoon strength and other aperiodic phenomena such as ENSO. This index is not significantly correlated with the zonally integrated bloom abundance (r < 0.1).
4.1. Comparison to Previous Assessments of Trichodesmium Bloom Occurrence
 The global, satellite observation-derived maps of Trichodesmium bloom occurrence presented here are the first of their kind. It is therefore important to compare the present patterns of Trichodesmium bloom occurrence with previously published accounts. This is not a straight forward task as field measurements of Trichodesmium bloom occurrence are single point observations and are not areal estimates of biomass. Hence small-scale variations of abundance in horizontal distance or with depth can make assays of Trichodesmium abundance highly uncertain [Subramaniam et al., 2002]. Further, observations are often made using different techniques making their comparisons difficult. Owing to its rarity, relatively few observations of Trichodesmium blooms have been published making the elucidation of general patterns and trends difficult.
 The bio-optical approach used here to map Trichodesmium bloom occurrence has its limitations as well [Westberry et al., 2005]. Algorithm validation using an independent data set resulted in false positive bloom retrievals ∼29% of the time. This can be significant and so the image processing undertaken here was meant to minimize these retrievals (i.e., SST mask, atmospheric correction mask, removal of single bloom pixels, etc). In addition, the scale of the satellite measurements (both spatial and temporal) can affect the retrieved patterns by assuming uniformity over a pixel and the duration of the measurement (e.g., 8 days). For example, a bloom which covers just over half of a pixel and lasts for ∼5 days, may be enough to be identified as a Trichodesmium bloom. Thus the persistence values presented here can be viewed as an upper bound. Previous efforts to synthesize Trichodesmium distributions provide coarse descriptions of its distribution [e.g., Carpenter, 1983; Carpenter and Capone, 1992; Karl et al., 2002; Capone et al., 2005]. The emphasis of these reviews has been primarily on subbloom densities of Trichodesmium simply because they are observed more often. However, it is valid to compare against these representations and point out similarities and differences under the assumption that blooms will arise in regions where Trichodesmium abundances are found in general. Carpenter  pooled available literature reports of Trichodesmium occurrence in the world ocean and synthesized them into seasonal maps of abundance (see his Figures 4– 7) . Broad-sweeping assumptions needed to be made as each global seasonal composite was constructed from 15–20 individual observations. One notable feature in that analysis is the presence of Trichodesmium in the western boundary currents (Gulf Stream and Kuroshio Current system) during the summer and fall and its marked absence during the winter and spring. Figure 3 shows a similar pattern in the SeaWiFS retrievals of Trichodesmium bloom occurrence, especially in the eastern reaching extensions across the North Atlantic and western North Pacific. Carpenter  also showed elevated abundances of Trichodesmium in the central North Pacific during the same time period, summer and fall (June–November). Again, this pattern is observed in the SeaWiFS bloom estimates, although bloom occurrence still amounts to a small percentage of the total time (Figure 3). Carpenter  also showed fairly persistent levels of Trichodesmium (∼>105 trichomes m−3) throughout the northern Indian Ocean in the fall and winter. Still higher values (>106 trichomes m−3) were noted along the eastern margin of the African continent during the winter, as well. Both of these observations are broadly consistent with the patterns seen in Figure 3, which show strong seasonal appearance in this region during the fall and winter months.
Carpenter and Capone  compiled several dozens of reported Trichodesmium blooms and identified areas likely to experience persistent blooms. Those areas were the Arabian Sea, the west coast of Africa, the northwest coast of Australia, the Caribbean and Gulf of Mexico, and the southwest Pacific near New Caledonia and Vanuatu. The maps in Figure 1 and Figure 3 share many of these characteristics. The shallow seas north of Australia do not show up in this analysis owing to the depth mask used in satellite data processing which is unfortunate as there are repeated observations of intense blooms in this area [Carpenter and Capone, 1992]. Another area which does not exhibit persistent blooms in the analyses presented here but noted by Carpenter and Capone  is the southwest Pacific Ocean (20°S, 180°E). Extremely large blooms have been reported in this area [Dupouy, 1992], however, their frequency is not well documented in situ and are observed infrequently using the satellite-based maps presented here.
 In addition to direct shipboard observations, the satellite-based maps of Trichodesmium bloom occurrence can be compared with geochemical inferences of net N2 fixation [e.g., Michaels et al., 1996; Gruber and Sarmiento, 1997; Deutsch et al., 2001; Hansell et al., 2004]. Geochemical estimates are integrative measurements that reflect the net balance between N2 fixation and denitrification such that, if both processes are operating equally, a geochemical signature may not be apparent. Further, the geochemical approach includes contributions from all diazotrophs, not just Trichodesmium [e.g., Zehr et al., 2001; Capone et al., 2005].
 As a first-order proxy for geochemical N2 fixation we calculate excess nitrate concentrations from nitrate and phosphate data taken from the World Ocean Data Atlas 2001 [Conkright et al., 2002]. This is a simplified version of the N* parameter [Michaels et al., 1996; Gruber and Sarmiento, 1997] and is defined here as the difference between the nitrate and 16× the phosphate concentration objectively analyzed at 200 m depth (Figure 5). Positive N* values indicate excess nitrate relative to the Redfield ratio, presumably due to net nitrogen fixation; while negative values indicate a nitrate deficit relative to Redfield. Figure 5 shows that the North Atlantic Ocean has large regions of excess nitrogen as indicated by widespread positive N* anomalies, suggesting widespread areas of net N2 fixation. In contrast, the Pacific Ocean and Indian Ocean on the whole have significant nitrate deficits, indicating an excess of denitrification over N2 fixation.
 Interestingly, the present satellite maps of Trichodesmium bloom occurrence (Figures 1 and 3) show nearly the opposite pattern as the excess nitrate distribution (Figure 5). Regions of semipersistent Trichodesmium blooms, especially the eastern tropical Pacific and the northern Indian Ocean, more often than not overlay areas with neutral or large nitrate deficits, which suggest water column denitrification [Deuser, 1975; Ganeshram et al., 2000; Altabet et al., 1999]. This is opposite from that which might be expected from areas with abundant nitrogen fixers, such as blooms of Trichodesmium. It is possible that these regions may be good habitats for diazotrophs because the N:P ratio is driven down owing to inorganic nitrogen losses, possibly creating a nitrogen limited environment, and blooms of Trichodesmium. Indeed, nitrogen isotopic evidence shows that ∼40% of the nitrate at 80 m depth was locally derived from N2 fixation in the Arabian Sea, an otherwise active region of denitrification [Brandes et al., 1998]. This may also be happening in the eastern Tropical Pacific as well [Sigman et al., 2005]. Clearly, the issue of an apparent coupling between water column denitrification and N2 fixation cannot be resolved here and detailed studies are required. The present observations of Trichodesmium bloom occurrence near regions of net denitrification are consistent with the right conditions (high P, low N, high light, etc.) for Trichodesmium populations to thrive.
 Last, numerical simulations of pelagic ecosystems that explicitly include Trichodesmium N2 fixation or give quantitative estimates of Trichodesmium biomass [e.g., Hood et al., 2004; Moore et al., 2002] may be used as another measure for comparison with the satellite-retrieved patterns. Hood et al.  modeled N2 fixation for the domain spanning the tropical and subtropical Atlantic, Caribbean, and Gulf of Mexico. The authors found a springtime maximum in Trichodesmium biomass and N2 fixation rates for the Gulf of Guinea and an autumn maximum in the western tropical Atlantic off Cuba and the Dominican Republic and into the Gulf of Mexico. Many of these features are consistent with the remote sensing estimates of bloom occurrence while some are not (Figure 3). The remote sensing estimates find an autumn maximum in the Gulf of Mexico although it is displaced farther north than Hood et al.  and there is no large region of persistent blooms in the greater western Atlantic Ocean. There is some correspondence during the summer in the south equatorial Atlantic which shows elevated levels of Trichodesmium in the work of Hood et al.  and higher frequency of occurrence in the remote sensing estimates also. However, the patterns shown here do not exhibit much of the activity close to the west coast of Africa. Comparison with the model output of Moore et al.  also shows mixed results. The satellite retrieved bloom patterns (Figure 1) show similar features in the eastern tropical Pacific and off the African coast near Madagascar. Additionally, the blooms diagnosed in the Indian Ocean (Figure 1) are mostly in the Arabian Sea and do not extend to the Bay of Bengal as is shown by Moore et al. [2002, their Figure 11]. Again, owing to our screening process, we are unable to diagnose blooms in the shallow seas in the northwest of Australia and in the often dust-influenced west coast of Africa.
 In summary, the patterns of Trichodesmium blooms presented here share many of the features of previous descriptions, but also have some significant differences. The most prominent difference is in the central tropical South Pacific, which is grossly undersampled by traditional shipboard measurements. To the best of our knowledge, there are no reports of Trichodesmium blooms for this region. The other prominent feature is the pattern in the Arabian Sea. The suggestion that Trichodesmium blooms could be occurring in parts of the ocean traditionally thought of as dominated by denitrification processes is novel, but is not without basis [e.g., Brandes et al., 1998; Sigman et al., 2005]. In addition, Capone et al.  observed a large bloom (∼2 × 106 km2) in the central Arabian Sea and speculated that they may be much more frequent and widespread than previously thought. The authors also suggested that the blooms might be an important episodic source of combined N and organic matter to fuel denitrification.
4.2. Relationship to Bulk Chlorophyll a Retrievals
 It might be expected that Trichodesmium blooms and blooms of other phytoplankton, as assessed using SeaWiFS Chl a, would be mutually exclusive because Trichodesmium generally grow well where other phytoplankton do not (i.e., under nitrogen limitation). However, part of the Chl a signal must include the contribution from Trichodesmium. Therefore it is not surprising that the spatial nature of the Trichodesmium blooms corresponds to some of the features of the Chl a distribution in SeaWiFS images. To contrast with seasonal patterns in Trichodesmium blooms (Figures 3a–3d), Figures 6a–6d show similar views of bulk phytoplankton Chl a retrieved from SeaWiFS. The figure shows a similar measure of persistence in bulk Chl a blooms; specifically the percent of time during the period examined that Chl a > 0.8 mg m−3 which is roughly equivalent to the Trichodesmium bloom threshold of 3200 trichomes L−1. Clearly, the Chl a bloom occurrence patterns shown in all four seasons (Figures 6a–6d) are markedly different than those seen in Figures 3a–3d. Features are almost entirely confined to the coast and are found primarily in the productive upwelling regions. Many of the areas highlighted also show no corresponding signal in the Trichodesmium bloom maps (e.g., South China Sea, Benguela Current system). Maximum values of persistence in these regions are much higher than any instances found in the Trichodesmium bloom distributions and are easily 100% over broad coastal regions in all four seasons. Co-occurrences of Trichodesmium and Chl a bloom retrievals occur only rarely over most of the ocean. Only in the Arabian Sea do Trichodesmium and Chl a bloom retrievals overlap often (up to ∼30% of the time on a seasonal basis). In general though, overlap is minimal (<10% persistence).
 Mean values of the SeaWiFS Chl a distribution (±1σ) for all pixels classified as Trichodesmium blooms and those which are not classified as Trichodesmium blooms are shown in Figure 7. The distribution of Chl a values during non-Trichodesmium bloom conditions has a lower mode (∼0.1 mg m−3) compared with bloom conditions. Retrieved Chl a values during Trichodesmium blooms are greater (median = 0.18 mg m−3) and the distribution has higher skewness (1.88) and excess kurtosis (2.13). In addition, the coefficient of variation is generally much larger across the entire range of Chl a in the bloom classified pixels, indicating that the Trichodesmium fraction of total Chl a is much more variable.
 Only a small fraction of the Trichodesmium bloom classified pixels exhibit bulk Chl a retrievals greater than 0.8 mg m−3 (the Trichodesmium bloom threshold). However, this fraction relative to the total amount of occurrences is much greater in the bloom pixels than the nonbloom pixels. Nonetheless, there are numerous reasons that the bulk Chl a might be underestimated in the presence of Trichodesmium. Pigment packaging effects are large [Subramaniam et al., 1999] and “secondary” packaging effects due to colony formation are also very strong [Borstad et al., 1992]. Borstad et al.  demonstrated that the absorption cross section of an average colony would be one hundred times less than that of the individual trichomes which are contained, and that a factor of 3–10 should apply for the effect of shadowing. In addition, Subramaniam et al.  suggested this effect would underestimate the total Chl a by a factor of four. Therefore it is not surprising that very few of the bloom identified pixels have relatively low retrieved Chl a concentrations.
4.3. Contribution of Trichodesmium to Ocean Color Properties
 The effects of Trichodesmium on ocean color spectra (and the resulting chl a retrievals) can be investigated using the model of Westberry et al. . Run in forward mode, the Westberry et al.  model can estimate Rrs(0−, λ) for a given background Chl a, colored dissolved organic matter (CDOM) absorption, and Trichodesmium biomass. Figure 8 shows Rrs(0−, λ) for varying amounts of Trichodesmium biomass (0–4000 trichomes L−1) and a constant background Chl a = 0.1 mg m−3 and CDOM absorption = 0.02 m−1 (thin dotted lines in Figure 8). Also shown is a reference spectrum representing bulk phytoplankton from the model of Maritorena et al.  with the same Chl a concentration, CDOM absorption, and particulate backscatter equivalent to Chl a = 0.1 mg m−3 (solid line in Figure 8). For the extreme range of Trichodesmium biomass shown there is a subtle increase in the blue to green reflectance as Trichodesmium biomass increases but with no real changes above 500 nm. In contrast, the Rrs(0−, λ) estimated from Subramaniam et al.  overemphasizes Trichodesmium backscatter and phycoerythrin fluorescence as well as absorption by CDOM (thick dashed line in Figure 8). Both of these effects have been tempered by the optimization procedure described by Westberry et al.  and allow the subtle distinction between Trichodesmium blooms and nonbloom conditions. Further, if the Westberry et al.  modeled reflectances shown in Figure 8 are run through the operational SeaWiFS chl a algorithm (OC4v4 [O'Reilly et al., 1998]) the Chl a values retrieved are underestimated by a factor of ∼2x (not shown). This implies that the optical signature from Trichodesmium blooms is a much more subtle signal than once was believed, except, perhaps, for extremely dense surface blooms.
4.4. N2 Fixation by Trichodesmium Blooms
 The biogeochemical impact of the bloom patterns presented here can be directly assessed by estimating the global N2 fixation due to Trichodesmium blooms. The simplest approach is to apply a constant areal rate of nitrogen fixation taken from literature and extrapolate to the total surface area covered by blooms and their observed persistence. Observations of N2 fixation rates by Trichodesmium were recently summarized by Capone et al. . From that study, a value of 1500 μmol N m−2 day−1 is used here to represent typical values under bloom conditions. If applied to the seasonal persistence fields (Figure 3) and then summed over the seasons, the annual mean N2 fixation rate is ∼42 Tg N yr−1 (Table 2). Recent geochemical estimates of total pelagic N2 fixation are ∼110 ± 40 Tg N yr−1 [Gruber and Sarmiento, 1997; Codispoti et al., 2001; Gruber and Sarmiento, 2002]. Thus Trichodesmium blooms are a significant source of fixed N to the ocean.
Table 2. Areally Integrated Trichodesmium Bloom N2 Fixation Rates in the Different Ocean Basins During Different Seasonsa
Bloom N2 Fixation Rate, Tg N yr−1
Rates are reported as mean values for period 1998–2003 and are calculated using seasonal persistence fields shown in Figure 3. Values are expressed in Tg N yr−1 (1 Tg = 1015 g).
 Both the temporal and spatial pattern of the bloom N2 fixation follow the patterns in bloom abundance. On an annual basis the largest fraction comes from the Pacific Ocean (∼60%), while the least amount of Trichodesmium bloom N2 fixation arises in the Atlantic (∼11%). However, it is clear from recent syntheses that some of the times, much of the diazotrophic biomass and nitrogen fixation in the Atlantic can be due to picoplanktonic diazotrophs and diatom-cyanobacterial symbioses, i.e., Richelia and Hemialus, rather than Trichodesmium [Carpenter et al., 2004; Capone et al., 2005]. The period of peak bloom N2 fixation is different from basin to basin and never goes to zero. Pacific blooms fix the most N2 in the winter (i.e., blooms are most widespread and persistent). The annual cycle in the Atlantic Ocean is very small with only a slight decrease in the winter months. In the Indian Ocean, the maxima correspond to the fall and winter, although there is a slow increase throughout the late summer and fall (not shown). The timing of this suggests that the rise in N2 fixation (and bloom abundance) is not associated with the summer monsoon which brings vigorous upwelling and chl a biomass accumulation.
 Contributions from nonbloom Trichodesmium populations to global N2 fixation rates can also be estimated. Using observations of Trichodesmium abundance compiled in Westberry et al.  we can say something about the proportion of bloom observations to nonbloom observations. Roughly 50% of these observations can be considered negligible (<100 trichomes L−1), and Trichodesmium abundances ranging from 100–3200 trichomes L−1 represented 39% of the distribution, while bloom observations (>3200 trichomes L−1) comprise the uppermost 11th percentile of this distribution (Figure 9). Assuming that the relative fraction of bloom and subbloom observations given here is representative of Trichodesmium at any given time, we can estimate the area occupied by subbloom populations and their resulting N2 fixation rates, given the satellite-derived bloom estimates. For this case, we apply a more moderate nonbloom areal rate of N2 fixation equal to 200 μmol N m−2 d−1 which gives annual nonbloom Trichodesmium N2 fixation in the tropical and subtropical oceans ∼20 Tg N yr−1. Obviously, this result is sensitive to the areal rate applied, but this value is reasonable on the basis of a compilation of measurements made by Capone et al. . In a very rough sense, this is in line with previous investigations. Estimates of global N2 fixation based on biological rate measurements are ∼80 Tg N yr−1 and it is thought that ∼40–60% of this amount is directly attributable to nonbloom Trichodesmium [Capone et al., 1997; Capone and Carpenter, 1999].
 In early work, Carpenter and Capone  showed that inclusion of N2 fixation by Trichodesmium blooms doubled (2×) then present-day global N2 fixation estimates. For lack of improved estimates, this same relationship has been applied in more recent estimates as well [see Gruber and Sarmiento, 1997]. Using a much more accurate method of accounting, the bloom results presented here somewhat support this logic, but go even further. Owing to the intensity of N2 fixation within blooms, we find that although somewhat limited in space and time, Trichodesmium blooms account for a majority of the annual biological N2 fixation. Our estimates of nonbloom background N2 fixation are ∼20 Tg N yr−1, while the bloom results add an additional 42 Tg N yr−1.
 Blooms of Trichodesmium spp. are shown to be a widespread, yet infrequent phenomena of the tropical ocean. This much was suggested from decades worth of shipboard observation. However, vast areas of the ocean remain undersampled in space and/or time, and the remote sensing approach used here is an excellent compliment able to fill some of the spatial/temporal gaps. The region of semipersistent blooms in the equatorial Pacific is a good example, as very few expeditions have traversed this portion of ocean, much less sought to observe Trichodesmium ecology. The overall impact of these blooms on N cycling in the ocean is significant, and on regional scales their role may be much more important. The magnitude of N2 fixation estimated here shows that an accounting for Trichodesmium blooms must be included in determinations of the global N budget. In our estimation it is difficult to say whether this inclusion leads to a more or less balanced budget due to the large degrees of uncertainty in the various sources and sinks [e.g., Codispoti et al., 2001; Gruber, 2004]. Estimates of total oceanic N2 fixation differ greatly from investigation to investigation depending on approach and details of methodology (see Karl et al.  for review). Denitrification estimates suffer from similar constraints. Further, the idea that these blooms, and nitrogen fixation in general, could be occurring in regions with strong denitrification is an enticing future research topic and warrants further investigation. Unraveling this potential coupling may have significant effects on future estimates of the oceanic N budget.
 The authors would like to thank Ajit Subramaniam, Norm Nelson, Stephane Maritorena, Natalie Mahowald, Tony Michaels and many others for helpful discussions. This work would not have been possible without the many years worth of data collection by Doug Capone, Ed Carpenter, and others. Support was provided by NSF and NASA for both T. K. W. and D. A. S.