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.
 Comparison of eight iron experiments shows that maximum Chl a, the maximum DIC removal, and the overall DIC/Fe efficiency all scale inversely with depth of the wind mixed layer (WML) defining the light environment. Moreover, lateral patch dilution, sea surface irradiance, temperature, and grazing play additional roles. The Southern Ocean experiments were most influenced by very deep WMLs. In contrast, light conditions were most favorable during SEEDS and SERIES as well as during IronEx-2. The two extreme experiments, EisenEx and SEEDS, can be linked via EisenEx bottle incubations with shallower simulated WML depth. Large diatoms always benefit the most from Fe addition, where a remarkably small group of thriving diatom species is dominated by universal response of Pseudo-nitzschia spp. Significant response of these moderate (10–30 μm), medium (30–60 μm), and large (>60 μm) diatoms is consistent with growth physiology determined for single species in natural seawater. The minimum level of “dissolved” Fe (filtrate < 0.2 μm) maintained during an experiment determines the dominant diatom size class. However, this is further complicated by continuous transfer of original truly dissolved reduced Fe(II) into the colloidal pool, which may constitute some 75% of the “dissolved” pool. Depth integration of carbon inventory changes partly compensates the adverse effects of a deep WML due to its greater integration depths, decreasing the differences in responses between the eight experiments. About half of depth-integrated overall primary productivity is reflected in a decrease of DIC. The overall C/Fe efficiency of DIC uptake is DIC/Fe ∼ 5600 for all eight experiments. The increase of particulate organic carbon is about a quarter of the primary production, suggesting food web losses for the other three quarters. Replenishment of DIC by air/sea exchange tends to be a minor few percent of primary CO2 fixation but will continue well after observations have stopped. Export of carbon into deeper waters is difficult to assess and is until now firmly proven and quite modest in only two experiments.
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.
 In August 1987, Martin and Fitzwater  demonstrated that phytoplankton Chl a growth was strongly stimulated when iron (Fe) was added to bottle-incubated seawater samples from the subarctic North Pacific (Figure 1a). They further suggested that Fe limited phytoplankton growth in the Southern Ocean, another region characterized by chronically high nutrients and low chlorophyll (HNLC). This long-forgotten hypothesis of Fe limitation of the Southern Ocean [Gran, 1931] was tested successfully [de Baar et al., 1990] 1 year later (Figure 1b), with iron addition significantly stimulating larger diatoms such as Nitzschia fragilaria (subsequently renamed Fragilariopsis kerguelensis), Corethron sp. and Thalassiothrix sp. [Buma et al., 1991]. With these events, the iron age in oceanography had begun [Coale et al., 1999; de Baar and La Roche, 2003].
2. Iron Fertilization Experiments in the 1993–2004 Era
 In October 1993, some 450 kg Fe (7800 mol) together with SF6 tracer (0.35 mol) was introduced into about 64 km2 surface waters for the IronEx-1 experiment [Martin et al., 1994]. A significant increase of photosynthetic quantum efficiency, Fv/Fm, from 0.3 to 0.6 occurred within 24 hours [Kolber et al., 1994]. An increase in primary productivity from 10–15 mg C m−3 d−1 to 48 mg C m−3 d−1, an increase of Chl a from about 0.24 to 0.65 mg m−3, and a modest decrease of fugacity of carbon dioxide, fCO2, of about 7 10−6 atm [Watson et al., 1994] were also observed within the first 3 days. No systematic differences were observed for the major nutrients (nitrate, phosphate and silicate) of the fertilized patch compared to out-patch ambient waters. Unfortunately, the patch subducted to about 30–35 m, after 4 days, and not much happened during the remaining 5 days of observations [Watson et al., 1994], apart from some increasing trend of Chl a and a decline in Fv/Fm [Kolber et al., 1994]. Obviously, the lower light level at depths exceeding 35 m, was detrimental for further ecological and biogeochemical response, but the dissolved Fe within the patch had also decreased below the shipboard detection limit of about 0.3 nM within 5 days, consistent with the decrease in Fv/Fm. Afterward, by Fe analyses in the home laboratory, the final dissolved Fe within the patch was still above the very low (<0.05 nM) natural dissolved Fe in ambient waters [Gordon et al., 1998].
 The follow-up experiment IronEx-2 in May 1995 was a major success, and it still is in the context of the seven subsequent Fe addition experiments (Table 1). One initial Fe infusion (225 kg) together with SF6 tracer at day 0 (29 May), was followed by two more Fe infusions of 112 kg each at days 3 and 7 [Coale et al., 1996]. The wind mixed layer (WML) deepened in a suite of small mixing events from 25 m at day 0 to 50 m by day 11. The Fv/Fm increased rapidly from 0.25 to 0.5 [Behrenfeld et al., 1996]. The maximum Chl a increased 27-fold from 0.15–0.20 mg m−3 to values approaching 4 mg m−3 on day 9, and then decreased to 0.3 mg m−3 by day 17. This was accompanied by a strong nitrate drawdown of 4 mmol m−3 and a maximum decrease of fCO2 of more than 70 10−6 atm, also at day 9 [Cooper et al., 1996, Table 1]. These dramatic impacts are convincingly illustrated by a day 5 transect across the patch (Figure 3). A strong initial increase of both DMSP and its conversion product DMS in the first 6 days [Turner et al., 1996], followed by a decreasing trend, appears consistent with a similar increase and then decrease in the haptophytes, the nanoplankton-sized major producers of DMSP. All size classes of phytoplankton responded to the added Fe by increasing their cellular photopigment concentration, indicating that they had previously been Fe limited [Cavender-Bares et al., 1999]. However the smaller pico- and nanoplankton size classes were kept at relatively constant concentrations by heterotrophic grazers, allowing the initially rare and large (>20 μm) pennate diatom Pseudo-nitzschia (rather than Nitzschia sp. as reported) to strongly dominate carbon biomass by the end of the experiment [Cavender-Bares et al., 1999; Landry et al., 2000a, 2000b]. The recently discovered strong organic complexation of dissolved Fe [Gledhill and van den Berg, 1994; van den Berg, 1995] was also nicely confirmed with two organic ligand classes L1 and L2 in the ambient waters, and a striking 400% increase of Fe(III)-binding ligands, largely in the stronger L1 ligand class, was observed within 48 hours of Patch-1 Fe fertilization [Rue and Bruland, 1997]. The 234Th deficiency method provided a first assessment of particulate organic carbon (POC) export increasing from 7 mmol m−2 d−1 prior to enrichment to ∼15 mmol m−2 d−1 in the day 2–7 period, and values approaching 50 mmol m−2 d−1 during days 8–14 [Bidigare et al., 1999].
 The first Southern Ocean Iron Release Experiment (SOIREE) [Boyd et al., 2000] took place at the end of austral summer in February 1999 with four Fe infusions (days 0, 3, 5 and 7) in the Antarctic Ocean proper, i.e., south of the Polar Front where all 3 major nutrients silicate, nitrate and phosphate are present. The depth of the Wind Mixed Layer was on average about 65 m [Boyd and Law, 2001]. Dissolved silicate at the onset was about 10 mmol m−3 which is well below winter values. This and the significant abundance of larger diatoms Fragilariopsis kerguelensis (>35 μm per cell, 4500 cells L−1), and other diatoms (∼1900 cells L−1) indicated that diatom blooms during the preceding summer had provided an “inoculate” population [Gall et al., 2001a]. Two haptophyte groups, with pigment signatures typical of Phaeocystis sp. and coccolithophores, respectively, increased steadily during the first 8–10 days, and then decreased somewhat [Gall et al., 2001a]. This trend was closely correlated with increases of DMSP and its breakdown product DMS at about days 8–9, consistent with cell lysis or grazing, and indeed, the abundance of heterotrophic ciliates grazers increased as well on day 9 [Hall and Safi, 2001].
 On day 13, the in-patch was dominated by diatoms [Gall et al., 2001a, Table 2], notably the chain-forming pennate, Fragilariopsis kerguelensis (16,500 cells L−1) which has a heavily silicified architecture as grazing protection [Hamm et al., 2003], and also Rhizosolenia sp. (4800 cells L−1) and Pseudo-nitzschia sp. (1400 cells L−1). The increasing chain length of Fragilariopsis kerguelensis (up to 14 cells/chain) indicated favorable growth conditions (chains up to 40 cells/chain were found in sediment traps [Waite and Nodder, 2001]). On the final day 13, primary production was indeed dominated by the >22 μm size class [Gall et al., 2001b] consisting of diatoms. Among these, various very large diatom species were abundant, notably Thalassiothrixantarctica (∼0.2 cells L−1), Asteromphalus flabellatus (∼300 μm; ∼1 cell L−1), Trichotoxonreinboldii (“needles” > 1 mm length, and ∼2 cells L−1), Nitzschia cf. sicula varieties (∼30 cells L−1), Coscinodiscus spp. (∼12 cells L−1), Eucampia antarctica (∼0.8 cells L−1) and various Navicula spp. (∼3.4 cells L−1) as shown by Waite and Nodder [2001, Figure 1]. These very large diatoms contribute significantly to the overall diatom biomass.
 Dissolved nitrate and silicate both decreased by 3 mmol m−3 during 13 days [Frew et al., 2001]. Similarly the fCO2 and DIC in the patch center had decreased by about 35 10−6 atm (Figure 4) and 17 mmol m−3, respectively [Bakker et al., 2001]. The area of the fertilized patch increased from 50 to about 250 km2 by day 13, a fivefold patch dilution constituting a “chemostat effect” where major nutrients are replenished while the trace element Fe and plankton biomass are diluted [Abraham et al., 2000]. The observation of the chlorophyll patch by remote sensing even 40 days after the start of the experiment, also requires to invoke continuation of a chemostat effect [Abraham et al., 2000] and has implications also for self-shading, coagulation and other processes [Boyd and Law, 2001]; yet due to lack of data in the 13–40 day period these cannot be verified. For example somehow retention of Fe during 40 days would presumably be required, somewhat in contrast to some 70% of added Fe being unaccounted for, i.e., lost, within first 13 days of observations [Bowie et al., 2001].
 The 234Th:238U ratio was less than 1 before the experiment, hinting at significant export in the preceding late summer. During the experiment the 234Th:238U ratio increased similarly at both IN and OUT stations by natural ingrowth of 234Th during conditions of near zero net particle flux [Charette and Buesseler, 2000], consistent with lack of clear IN versus OUT differences in shallow sediment trap fluxes [Nodder and Waite, 2001; Nodder et al., 2001].
 Comparison of the significant responses between IronEx-2 in the tropical Pacific and SOIREE in the Southern Ocean (Figure 5) shows much faster rates of primary production and nitrate removal in IronEx-2 experiment, consistent with the far lower temperature, deeper Wind Mixed Layer (WML) and stronger lateral patch dilution for SOIREE. This lateral patch dilution may also have slowed down the aggregation of particles as a step toward export of larger aggregates in SOIREE relative to IronEx-2 [Boyd et al., 2002].
 The Carbondioxide Uptake Southern Ocean (CARUSO)/Eisen(=Iron) Experiment (EisenEx) took place in the austral spring, November 2000, in the core of a cold Antarctic eddy that had spun off the Polar Frontal jet into the sub-Antarctic region. The eddy being Antarctic water had all major nutrients present but was surrounded by sub-Antarctic waters with ample nitrate and phosphate as well, but low in silicate. Three Fe infusions of 780 kg each at days 0, 7–8, and 16 were diluted downward and laterally by two major storms (days 5 and 13) and overall strong winds throughout the 23 days of observations [Bakker et al., 2005]. The WML increased from an initial 40–50 m to almost 100 m during the final days 16–22 (Figure 6). Patch size increased concomitantly. Shear stress in the energetic ocean frontal region also played a role in increasing the patch size from 50 km2 during the first Fe/SF6 infusion to 950 km2 in the final survey map. The Fv/Fm responded with a slight but distinct increase within a day [Gervais et al., 2002]. Within 48 hours, Fv/Fm increased from 0.3 to 0.4 and steadily increased to 0.52 by final day 21. An increase in primary production on day 2 preceded increases in Chl a on day 4 for all pico-, nano-, and microphytoplankton size classes (pico < 2; 2 < nano < 20; micro > 20 μm). After day 2, the picoplankton Chl a remained constant, likely due to a balance between growth and grazing, but the nanoplankton Chl a steadily increased, and the initially smallest pool of microplankton Chl a (>20 μm) increased exponentially to dominate primary production after day 16 [Gervais et al., 2002]. The first study of diurnal carbohydrate dynamics in an enrichment experiment showed increases of both daytime production and nocturnal consumption of polysaccharides in Fe enriched waters, the polysaccharide production furthermore being light limited [van Oijen et al., 2005]. During the initial phase, the microplankton diatoms were dominated by Fragilariopsis kerguelensis with a mean depth-integrated abundance of 5737 cells L−1 at T = 0 (day 0) of the experiment. This species increased to 22,146 cells L−1 inside and 9,389 cells L−1 outside the patch by day 21. Chain lengths on the order of 4–30 cells/chain were indicative of favorable growth, with maximum observed lengths of up to 160 cells/chain. During the second phase of the experiment the smaller Pseudo-nitzschia lineola became by far the most abundant, ultimately accounting for 53% of total diatom cells and 25% of total diatom biomass [Assmy, 2004]. A similar marked response was observed for the small centric diatom Chaetoceros debilis. Among all diatom species P. lineola showed the highest accumulation rate of 0.20 d−1, and even outside the patch was growing at 0.09 d−1 [Assmy, 2004]. By the end of the experiment the depth-integrated cell numbers of Pseudo-nitzschia had increased 11-fold within the patch (18.7 109 cells m−2) versus the control station (1.7 109 cells m−2). Very large diatoms (e.g., Rhizosolenia sp., Thalassiothrix sp., Corethron pennatum) also responded strongly. Although cell numbers of this very large size class were much lower, in terms of biomass (both organic and opal) they may be quite significant. Twenty days after the first Fe infusion, the maximum changes in the surface waters of the patch relative to the outside patch measurements [Bozec et al., 2005] were −15 mmol m−3 for DIC, −23 10−6 atm for fCO2, +0.033 units for pH, −1.61 mmol m−3 for nitrate, and −0.16 mmol m−3 for phosphate (Figure 7). Despite the significant increase in larger diatoms within the patch, the dissolved silicate showed a similar decrease of about 4 mmol m−3 for the in-patch and out-patch stations.
 The Subarctic Pacific Iron Experiment for Ecosystem Dynamics Study (SEEDS) enjoyed favorable summer weather (July 2001) and a shallow 10 m wind mixed layer during the single infusion of 350 kg Fe and 0.48 M SF6 into its 8 × 10 km patch size [Tsuda et al., 2003]. The Fv/Fm of the whole community rose from a low initial value of 0.2 to 0.3 on day 4, significantly exceeding the out-patch value. This coincided with the onset of a strong growth phase of the >10 μm size class on day 4 due to a very rapid increase of the chain-forming centric diatom Chaetoceros debilis, which initially had occurred in low numbers (∼400 cells L−1 at day 0). This gave rise to a very strong increase in Chl a in the >10 μm size class which fully dominated the total Chl a pool reaching a record of 19 mg m−3 at the final day 13 of observations (Figure 8). The smaller diatom Pseudo-nitzschia turgidula [Hasle, 1993] did have higher initial abundance (10,000 cells L−1) but increased slowly and was surpassed by C. debilis on day 6 in terms of cell numbers. Similar to the previous IronEx-2 and SOIREE experiments, albeit more trivial at SEEDS, there was an increase and finally a decrease, of the nanoplankton (2–10 μm). On the final day 13 both nitrate and silicate were virtually depleted (Figure 8), consistent with a decrease again in Fv/Fm on days 11 and 13 indicative of nutrient stress (N, Fe or both), and accompanied by record decreases in fCO2 (94 10−6 atm) and DIC (61 mmol m−3) [Tsuda et al., 2003].
 The Southern Ocean Fe Experiment (SOFeX), in January–February 2002 increased the effort compared with the previous experiments by fertilizing two patches (North and South) at the same time, each patch being larger (225 km2) than before, with sophisticated multiship logistics of R/V Revelle, R/V Melville and icebreaker Polar Star [Coale et al., 2004].
 The SOFeX-North patch with low-silicate (<3 mmol m−3) high-nitrate (20 mmol m−3) sub-Antarctic waters, had two infusions of 631 kg Fe on days 0 and 5 (10–12 and 16 January) and a third 450 kg infusion on day 30 (10 February). The initial WML was 45 m deep increasing to 55 m 1 month later [Coale et al., 2004, Table S1]. Owing to many frontal systems in the region, SOFeX-North experienced more shear stress than CARUSO/EisenEx. It was streaky immediately after the first infusion, and by day 38 the patch had evolved into a 7-km-wide by at least 340-km-long filament. The Fv/Fm increased from the initial 0.2 to 0.5 at the end. The SOFeX-North patch was characterized by an approximately 20-fold increase in Chl a (up to 2.6 mg m−3) and a fourfold increase in phytoplankton carbon biomass. In the early stages (first 2 weeks), the increase was almost entirely due to >5 μm nonsiliceous taxa (prymnesiophytes, pelagophytes, dinoflagellates). After a month, however, >20 μm diatoms, strongly dominated by Pseudo-nitzschia, had risen from 5 to 38% of total phytoplankton biomass despite low available silicate [Coale et al., 2004; S. Brown and M. Landry, personal communication, 2005]. Physical mixing and dilution processes entrained dissolved silicate into the patch [Coale et al., 2004; Hiscock and Millero, 2005], akin to the chemostat effect in SOIREE [Abraham et al., 2000]. This prevented the complete removal of silicate, thus allowing for sustained, however Si-limited, diatom production [Brzezinski et al., 2005; Hiscock and Millero, 2005]. The maximum changes in carbonate and nutrient parameters for the North Patch were −14 ± 5 mmol m−3 for DIC, −26 ± 5 10−6 atm for fCO2, −0.09 ± 0.03 mmol m−3 for phosphate, −1.1 ± 0.4 mmol m−3 for silicate, and −1.4 ± 0.2 mmol m−3 for nitrate [Hiscock and Millero, 2005].
 The SOFeX-South patch in high silicate (60 mmol m−3) and high nitrate (28 mmol m−3) Antarctic waters had four infusions of 315 kg Fe on days 0, 4, 7 and 11 (24 January–5 February), with the depth of the WML remaining constant at 35 m [Coale et al., 2004, Table S2]. The study site was strongly dominated by diatoms (2/3 of phytoplankton C at initial and out-patch “control” stations), notably Pseudo-nitzschia, Chaetoceros, Thalassiothrix and Fragilariopsis species and various smaller pennates. The community-wide Fv/Fm increased from 0.25 to 0.65, and overall maximum photosynthetic rates increased from 0.29 to 4.6 mmol C m−3 d−1 (Figure 9). Total fluorometric Chl a increased by about eightfold, from ∼0.5 on day 0 to 4 mg m−3 on day 21. This response comprises a greater than threefold increase in the phytoplankton Chl a:C ratio, and slightly greater than a twofold increase in carbon biomass (M. Landry and S. Brown, personal communication, 2005). Cells in the size ranges of 10–20 and 20–100 μm showed the highest net rates of change, increasing to about 3 times initial levels, with net rates of change dropping appreciably for >100 μm cells (very large diatoms) and negligible change for <10 μm cells. Thus while the community response was not dramatic, there was a significant shift to intermediate and large-sized cells and more diatoms (increasing from 66 to 80% of phytoplankton carbon biomass) (M. Landry and S. Brown, personal communication, 2005). The maximum changes of carbonate parameters in the South Patch mixed layer were −21 ± 5 mmol m−3 for DIC and −36 ± 4 10−6 atm for fCO2 [Hiscock and Millero, 2005]. The maximum changes for nutrients were −0.22 ± 0.03 mmol m−3 for phosphate, −3.6 ± 0.2 mmol m−3 for silicate, and −4.1 ± 0.2 mmol m−3 for nitrate [Hiscock and Millero, 2005]. Changes in the carbonate and nutrient systems of the South Patch were larger than those in the North Patch. In addition, the export of carbon as particles settling into deeper layers was determined to be significant by the 234Th deficiency technique (Figure 9), but modest with respect to regular estimates in the region [Buesseler et al., 2004, 2005].
 During summer of the same year (July 2002), another multiship experiment, the Subarctic Ecosystem Response to Iron Enrichment Study (SERIES) was conducted at ocean station “Papa” (50°N, 145°W) where it all had started [Martin and Fitzwater, 1988]. Upon two Fe infusions (days 0 and 6) of the quite favorable fairly shallow 30 m WML depth, the 77 km2 patch enlarged to >200 km2 by day 13 and to a maximum of about 1000 km3 by days 17–18 [Boyd et al., 2004a]. The Fv/Fm increased from an initial value of 0.2 to at most 0.4 by day 11, dropping to about 0.2 by day 19 then decreasing further to well below 0.2 at day 21 [Boyd et al., 2004a, suppl. Figure 1]. The Chl a increased [Harrison et al., 2004] from an initial 0.6 mg m−3 to a maximum 5.5 mg m−3 on day 17, then dropped sharply to about 1.5 mg m−3 by days 23–25 (Figure 10). The parallel increase of diatom biovolume was stronger relative to the low initial values, and the strongest increase (days 10–15) was nicely mirrored by a sharp decrease in dissolved silicate (Figure 10). The most dramatic blooming was by the small pennate diatom, Pseudo-nitzschia sp., which increased from an initial 174 cells L−1 to 192,000 cells L−1 at peak days 15–18 [Marchettti and Harrison, 2004]. Similarly the almost 100-fold larger cells of the pennate Thalassiothrix increased from 17 to 12,100 cells L−1, thus dominating by some 33% the carbon inventory of all diatoms. Centric diatoms of medium (Chaetoceros sp.), large (Thalassiosira, Proboscia) and very large (Rhizosolenia) sizes, together comprised over 60% of the diatom carbon inventory. On about day 16, silicate was depleted, and the bloom declined, leading to increased export flux into sediment traps at 50, 75, 100 and 125 m (Figure 10). This, and the above-mentioned export flux at SOFeX-South, were the first convincing observations of Fe-stimulated export of POC and Si into deeper waters. Over the 25 day experiment, about 14 mmol m−3 silicate and 5 mmol m−3 nitrate were removed (Figure 10). The fCO2 showed its maximum decrease of over 70 10−6 atm in the patch centre at day 18 [Boyd et al., 2004a, supplement]. On the same day, DIC had decreased by about 40 mmol m−3, but then it increased again due to the combined effects of bloom decline (respiration) and patch dilution with ambient waters.
 Finally the very recent EIFEX experiment (2004) in the Southern Ocean is mentioned but results are not yet available.
3. Light Limitation and Other Physical Forcings
 Wind mixing strongly influences the amount of light that phytoplankton receive for growth. When comparing the eight Fe experiments there is a wide range of average depths of the wind mixed layer (Figure 11, black bars), which vary from a shallow 10–15 m for SEEDS to ∼100 m for CARUSO/EisenEx (Figure 6). The maximum yield of Chl a abundance shows an inverse relationship with WML (Figure 11, green bars).
 Excluding IronEx-1, which is anomalous due to early subduction, one finds a striking and significant (R2 = 0.90) inverse relationship between maximum Chl a yield and average WML depth (Figure 12 (top)). Hence having dumped a total of 8795 kg of Fe into HNLC waters and utilizing about 1 year of shiptime, we may conclude that light is the ultimate determinate of the phytoplankton biomass response. However, Chl a may not be the most suitable variable for biomass. Since iron is required for synthesis of the chlorophyll molecule, and Fe-depleted phytoplankton tend to be short in Chl a, one of the first responses to Fe enrichment is an increase in cellular Chl a (Figure 13). For example, the mean C:Chl ratio of phytoplankton decreased by factors of 4–5 in the IronEx-2 and CARUSO/EisenEx patches [Landry et al., 2000a; M. J. W. Veldhuis and K. R. Timmermans, Photoplankton dynamics during the EISENEX in situ iron fertilization experiment in the Southern Ocean: A comparative study of field and bottle incubation measurements, submitted to Limnology and Oceanography, 2005] and only slightly less in SOFeX-South (M. Landry, personal communication, 2005). Since Chl a yield exaggerates the phytoplankton biomass response to Fe fertilization, a more suitable indicator may be preferable.
 Both the maximum nitrate removal and the maximum fCO2 drawdown (Figure 11, blue and red bars, respectively) also show trends opposite to the WML depth, but the fits (R2 = 0.69 and R2 = 0.63, respectively) are less convincing (Table 2). Perhaps the maximum net removal (photosynthesis/respiration) of DIC, which also shows an inverse trend with WML depth (Figure 11) with a better fit (R2 = 0.72; see Figure 12 (bottom)), is the most suitable overall biomass indicator.
Table 2. Apparent Inverse Relationships of Wind Mixed Layer (WML) Depth Versus Observed Maximum Chl a and Versus Observed Maximum Removals of Nitrate, Fugacity of CO2, and Dissolved Inorganic Carbon (DIC) in Seven Experimentsa
Function of WML Depth
WML depth is in meters. Observed maximum Chl a is in mg m−3. Fugacity of CO2 is in 10−6 atm. Dissolved inorganic carbon is in mmol m−3. The probability that the fitted curve of Chl a versus WML with an exponent of −1.342 is due to chance alone is low enough (P = 0.003, t = −5.55, R2 = 0.90, n = 7) to reject the hypothesis that WML depth has no effect. Excluded is IronEx-1 due to premature end at day 4 by patch subduction. The better fits for Chl a and for DIC are illustrated in Figure 12. The goodness of fit R2 for n = 7 data points is remarkable but certainly not perfect due to several other factors (temperature, photosynthetically active radiation, patch dilution, grazing, and Km for Fe limitation of predominantly responding phytoplankton species) also playing roles; see text.
Maximum chlorophyl = f(WML)
y = 523.21 x−1.3598 (exponential)
Maximum delta nitrate = f(WML)
y = 82.33 x−0.884 (exponential)
Maximum delta fCO2 = f(WML)
y = −1.087 x + 97.62 (linear)
Maximum delta DIC = f(WML)
y = −0.772 x + 59.04 (linear)
 Of course other physical factors are also at play here. First, the amount of incident light at the sea surface, or photosynthetically active radiation (PAR), varies from day to day, by region and by season (Figure 14). Moreover, when a bloom develops, the self-shading by the more abundant phytoplankton diminishes the available light below, and thus the maximum extent of the euphotic zone of positive phytoplankton growth (not shown).
 The optimal or maximum rate of growth (μmax after Monod ) is well known to be a function of temperature, where for each 10°C rise in temperature the maximum growth rate doubles [Eppley, 1972; Goldman and Carpenter, 1974]. Given the range in temperatures (Figure 14) one would expect the primary production during IronEx-2 (25°C) to be almost fourfold faster than during SOIREE (2°C), and such trend is apparent indeed (Figure 5).
 Similarly, SEEDS (9.5°C) showed growth rates about 50% faster than CARUSO/EisenEx. In fact, during days 4–7 of SEEDS, the major diatom C. debilis grew at a net rate of 2.6 doublings/day (μ = 1.8 d−1), exceeding the expected maximum of 1.5 doublings/day (μ = 1.0 d−1) at 9.5°C [Tsuda et al., 2003].
 Finally, due to combined wind mixing and shear stress, the initial patch tends to dilute, somewhat by mixing with underlying waters, but apparently more by lateral mixing. With most initial patch size areas of about 50–80 km2, and 225 km2 for SOFeX patches, the final patch area after 13–38 days of observations may well exceed 2000 km2 (Figure 15). Obviously the resulting dilution factor will be least at the patch centre and increase toward its edges, and in principle can be assessed from the distribution of SF6 tracer in time and space, after a correction for SF6 loss by gas exchange. However, matters are complicated by the need for several Fe infusions, while only the first infusion included the SF6 tracer. Also, patches can evolve very chaotically (CARUSO/EisenEx, SOFeX-North) defying quantitative description. Here, we simply have taken the ratio of initial patch size and reported final patch size as indicative of lateral patch dilution. This dilution factor varies from 3 and to 19, where obviously the length in days of the experiment plays a role (Figure 15 and its caption).
4. Linking the Peaceful Pacific With the Roaring Forties
 Given the wide range in physical conditions (WML depth, PAR, temperature, patch dilution) and biogeochemical responses (Figures 11, 14, and 15), we will take up the challenge of determining coherence between only the two extreme cases: SEEDS with favorable conditions and record responses, and CARUSO/EisenEx.
 After completing the 15 hours of the first Fe infusion for CARUSO/EisenEx, seawater samples were immediately collected in both the in-patch and at an out-patch control station. This seawater was placed in a suite of PMMA bottles (transparent for UVB + UVA + visible light) in deck incubators with screening to simulate a 12 ± 4 m light depth [van Oijen and Rijkenberg, 2004], as well as in large 20 L polycarbonate bottles incubated under artificial light on a 12/12 hour day/night cycle at a light level (3.5 mol m−2 d−1) corresponding to the 25 m light depth (K. R. Timmermans et al., manuscript in preparation, 2005). The PAR reaching the sea surface as well as the screened deck incubators varied between 12 and 55 mol m−2 d−1, averaging 30 mol m−2 d−1.
 Four cases can now be compared. The CARUSO/EisenEx in situ in-patch stations reached a maximum nitrate removal of 1.6 mol m−3 after 21 days (Figure 7 (bottom)). The in-patch samples of the CARUSO/EisenEx deck incubators were completely depleted of both nitrate and silicate after 12–14 days (Figure 16 (left)), strikingly similar to the SEEDS in situ result after 13 days (Figure 8). In these deck incubators, the same diatoms as found in the field showed healthy net growth rates of 0.4–0.6 d−1, with significantly (10–15%) higher growth for the Fe-fertilized in-patch bottles, notably for Pseudo-nitzschia which eventually was dominant both in these bottles and in the in situ in-patch stations [van Oijen and Rijkenberg, 2004].
 Remarkably, only Fragilariopsis kerguelensis did not show higher growth rates for the Fe-enriched in-patch bottles versus control out-patch bottles. This is consistent with its failure to eventually dominate the in situ in-patch stations (as opposed to SOIREE). Finally, the CARUSO/EisenEx in-patch samples cultured under lower artificial light were somewhat slower, only exhausting nitrate and silicate after 18 days (Figure 16 (right)).
 When comparing the physical forcing parameters temperature, WML depth and the patch dilution factor for the two extreme in situ experiments CARUSO/EisenEx and SEEDS (Figure 17 (top)), one notices that simulated WML depths of the CARUSO/EisenEx bottle treatments are similar to, or slightly deeper, than for SEEDS. Moreover, bottles obviously have a dilution factor of 1 (no dilution), closer to the modest factor of 3 of SEEDS than the large dilution factor of 19 for CARUSO/EisenEx. The resulting daily removal rates of nitrate and silicate (Figure 17 (bottom)) in both bottle treatments nicely bridge the gap between the nutrient removal rates of the two in situ experiments. The fact that both nutrients are also fairly rapidly removed in the control treatments (Figure 16), further confirms the major impact of light relative to Fe deficiency. Even though the differences of PAR as well as temperature between CARUSO/EisenEx and SEEDS (Figure 14) have been ignored here, the strong influence of WML depth, defining the mean light environment, as well as the dilution factor, is very obvious. Admittedly, this is just a very simple approach, only providing a working hypotheses for a more refined validation of day-to-day changes in these four cases by plankton ecosystem modeling [Lancelot et al., 2000; Pasquer et al., 2005], with all physical forcings (WML depth, PAR, dilution factor, temperature) in the model varying daily and acting simultaneously.
5. Large Diatoms
 In all of the experiments, added Fe produced a striking shift-up response of cell numbers of larger size classes of diatoms. In general the community biomass shifted from mostly nanoplankton (<10 μm) to mostly microplankton (>10 μm). As noted previously, the intermediate-sized pennate (Pseudo-) Nitzschia sp. took over in IronEx-2 [Landry et al., 2000a]. Pseudo-nitzschia species also dominated in CARUSO/EisenEx, SOFeX-North and SERIES. The large chain-forming pennate Fragilariopsis kerguelensis was most successful in terms of cell numbers in SOIREE, consistent with its spring bloom development and dominance in the naturally Fe-rich Polar Frontal jet [de Baar et al., 1995, 1997]. Both Fragilariopsis kerguelensis and Pseudo-nitzschia were numerically dominant in SOFeX-South, also at the control stations. During SEEDS, the centric chain-forming Chaetoceros debilis responded very strongly, replacing the smaller pennate Pseudo-nitzschia turgidula in both cell abundance and biomass after 6 days. These in situ results are consistent with the major conclusion of a preceding synthesis of all preceding bottle incubation experiments and natural Fe-replete regions which documented a systematic Fe stimulation of the large size class of diatoms [de Baar and Boyd, 2000].
 Reports on very large diatoms (0.1–1 mm) have been relatively sporadic. Nevertheless, very large taxa (Rhizosolenia sp., Thalassiothrix, Thrichotoxon, Asteromphalus, Actinocyclus sp.) with elongate “needles” as well as perfectly discoid forms should not be overlooked. In terms of biomass (organic matter or opal), these “giants” may in some experiments have grown to higher levels than the numerically more abundant (cells m−2) large diatoms.
Timmermans et al. [2001a, 2001b, 2004] have succeeded in maintaining such very fragile (spines, weak chains) diatoms of moderate (10–30 μm), medium (30–60 μm) and large (>60 μm) size classes in cultures of natural (no EDTA disturbances [Gerringa et al., 2000]) Antarctic seawater. From their curves (not shown) of growth rate responses versus added dissolved Fe, the constants Km for half-saturated growth (= 50% of maximum growth rate) now show a convincing inverse relationship with surface/volume ratio (Figure 18). Since surface/volume ratio is itself inversely related to size, the required ambient Fe concentration to achieve 50% of maximum growth rate clearly increases as a (fairly linear) function of diatom size. Moreover, the required Fe concentrations (0.2–1.2 nM) are above the typical dissolved Fe concentrations of remote oceanic surface waters. In other words, these size classes of moderate (10–30 μm), medium (30–60 μm) and large (>60 μm) diatoms can only bloom upon an extra delivery of dissolved Fe, either by wet dust deposition from above [Jickells and Spokes, 2001], by an Fe-rich oceanic front [de Baar et al., 1995, 1997], by an upwelling/mixing supply event from below [Hoppema et al., 2003], by iron supply from shallow topography [Blain et al., 2001], or by an in situ Fe fertilization experiment.
 The observed linear relationship between diatom size and Fe requirement for growth is a major step forward and key to understanding the in situ experiments. When superimposing the above Km of the large F. kerguelensis and very large Actinocyclus on the time series of dissolved Fe (<0.2 μm filtrate) in SOIREE, the minimum observed dissolved Fe is more or less adequate to support half-maximum growth of F. kerguelensis, but inadequate for Actinocyclus (Figure 19). The double Fe concentrations (dotted lines) for optimal 100% growth are occasionally met for F. kerguelensis and rarely, if at all, for Actinocyclus. Obviously, if one could maintain a steady and carefully chosen dissolved Fe concentration throughout an experiment, the relationship (Figure 18) might reliably predict the size class of diatom that would dominate in the end (in the absence of loss processes to grazing, sinking, etc.).
 Even this, however, has proven to be a major stumbling block of the in situ experiments. The intrinsic instability of the added reduced Fe(II) in oxygenated seawater (discussed below) leads to rapid removal of the Fe enrichment. In an effort to remedy this problem, a lignopolysulfonate ligand had been added during the two experiments (GreenSea 1, 2) in the Gulf of Mexico (Table 1), but its effect cannot be discerned from the brief report [Markels and Barber, 2001]. In all other experiments, only dissolved [Fe(II)] was added.
 In SOIREE, four successive Fe additions (vertical black bars) were applied in an effort to compensate for the major Fe loss. Despite a rigorous budgeting approach [Bowie et al., 2001], only 30% of the added Fe could be accounted for, the remaining 70% being “lost.” Similar repeat Fe infusions have been done for most other experiments (except IronEx-1 and SEEDS), with Fe budgeting hardly, if at all, being attempted anymore. The overall result tends to be a sawtooth pattern of dissolved Fe concentration inside the patch.
 This sawtooth pattern creates a conceptual dilemma in which the Km values (Figure 18) are only valid at steady state, i.e., constant dissolved Fe, while patch in situ Fe concentrations oscillate markedly. Presumably, during events of high natural input of dissolved Fe, diatoms are capable of luxury accumulation of Fe within their cells, to be used later when Fe concentration fall below a critical threshold level, e.g., Km. Perhaps for any given diatom species, a bandwidth of suitable dissolved Fe between, for example, 50% and 100% of optimal growth, defines the range to be maintained in an in situ experiment. Future culture experiments exposing diatoms to oscillating Fe levels may demonstrate such Fe storage capacity and help bridge the conceptual gap between our present understanding of linear steady state relationships (Figure 18) and the more dynamic algorithms that will be needed for the next generation of ecosystem simulation models of in situ experiments [Lancelot et al., 2000; Hannon et al., 2001]. Obviously, this is also desirable for understanding the real ocean where natural dissolved Fe varies over daily timescales in surface waters.
6. Fe Chemistry in Seawater
 Until now, the role of Fe has been described using dissolved Fe in filtered seawater (<0.2 μm filtrate) as the common variable. In reality, however, the physical chemistry of Fe in seawater is far more complicated.
 During CARUSO/EisenEx, polyethylene hollow-fiber ultrafiltration was used to distinguish another size class, colloids (200 kDa < colloids < 0.2 μm), with the ultrafiltrate (<200 kDa) being called the soluble fraction [Nishioka et al., 2005]. Prior to the first infusion of iron, the dissolved (<0.2 μm) iron concentrations in the ambient surface seawater were extremely low (0.06 ± 0.015 nM), with colloidal iron being a minor fraction. For the iron addition, the eddy was fertilized with an acidified FeSO4 solution (i.e., reduced [Fe(II)]) 3 times over a 23 day period [de Baar, 2001]. High concentrations of dissolved iron (2.0 ± 1.1 nM) were measured in the surface water until 4 days after the first iron infusion (Figure 20). After every iron infusion, when high iron concentrations were observed before storm events, there was a significant correlation between colloidal and dissolved iron:
These results indicate that a roughly constant proportion of colloidal versus dissolved iron (∼76%) was observed after iron infusion. Thus it appears that most of the added [Fe(II)] was rapidly converted into fine colloids.
 The operationally defined division of Fe into three size classes, soluble <200 kDa, 200 kDa < colloids < 0.2 μm, and particles >0.2 μm, is also reflected in the organic complexation of iron. Organic binding of Fe was investigated only in the soluble and colloid fractions [Boyé et al., 2005]. In the natural seawater before Fe addition, some 91 ± 3% of the organic Fe-binding ligands was found in the soluble fraction. In contrast, the size distribution of ligands in the mixed layer after Fe release was balanced between soluble (∼55%) and colloidal (∼45%) ligands. Soluble and colloidal ligands were produced rapidly within the mixed layer after the first and second Fe infusions. Here a dramatic increase of the colloidal ligand concentrations was observed, increasing concentrations by two- to threefold for soluble, and up to 35-fold for colloidal ligands, relative to their respective out-patch levels.
 Simultaneous measurements of Fe(II) and H2O2 showed detectable Fe(II) concentrations for up to 8 days after iron infusion [Croot et al., 2005]. Vertical profiles of Fe(II) showed maxima consistent with the plume of the iron infusion. Parallel H2O2 profiles revealed corresponding minima, showing the effect of ongoing oxidation of Fe(II) by H2O2. The H2O2 concentrations increased at the depth of the chlorophyll maximum when iron concentrations returned to preinfusion concentrations (<80 pM), possibly due to biological production related to iron reductase activity. During a later surface survey of the iron enriched patch, elevated levels of Fe(II) were found in surface waters presumably from Fe(II) dissolved in the rainwater that was falling at this time. Model results suggest that the reaction between uncomplexed Fe(III) and O2− was a significant mechanism helping to maintain high levels of Fe(II) in the water column, and the low temperature of Antarctic seawater slows down the reverse oxidation reaction considerably [Croot et al., 2001]. Finally, photochemical reduction of finely dispersed colloids may act as a source of reduced [Fe(II)] in surface waters [Rijkenberg et al., 2005], thus making Fe available again for plankton uptake.
 We are only beginning to understand the chemistry of Fe in seawater. Nevertheless two main lines of thought can already be identified.
 First, the artificial ∼100-fold increase of overall Fe levels after the addition of dissolved inorganic Fe(II) ions is a major disruption of the natural physical-chemical abundances and reactivity of Fe in seawater. Hence the ensuing plankton responses, while significant, are not necessarily representative of natural enrichment by dry or wet deposition of aeolian dust [Boyé et al., 2005], or from adjacent continental margins or ocean island plateaus. This notion has led to another experiment (FeCYCLE, 2003) where only SF6 tracer was added just to follow the natural physical chemistry of Fe in seawater [Boyd et al., 2004b] (available at http://aslo.org/honolulu2004). Moreover, two parallel programs BICEP and KEOPS were undertaken in austral summer 2004–2005 using natural gradients of dissolved Fe near the ocean islands of Crozet and Kerguelen [Blain et al., 2001; Bucciarelli et al., 2001] as a natural laboratory for unraveling the complex interactions of Fe chemistry and phytoplankton growth.
 Second, algal cells can only directly assimilate “truly dissolved” Fe and thus the Fe residing within the soluble (<200 kDa) size fraction. This soluble fraction consists of reduced [Fe(II)] as well as organically bound [Fe(III)ligand], with evidence for more than one ligand type, and also some inorganic [Fe(III)]. It is not clear which is the preferred form for uptake by algae. If the kinetics of exchange between these three major forms within the soluble pool are faster than the overall uptake rate, algal uptake preferences do not really matter. On the other hand, if one or more of the transformation reactions are slow, it (they) could potentially control growth rates of the algae. Moreover, the colloid size class (>200 kDa) is not directly available for uptake, but may rapidly replenish the soluble pool, as colloids dissolve by photoreduction [Rijkenberg et al., 2005], releasing dissolved [Fe(II)] for algal uptake. Yet, in due course, the [Fe(II)] will also be oxidized again into colloids. The kinetics of exchange between these various Fe pools must be quantified in order to understand how Fe controls the rate of phytoplankton growth. Thus the above consistency between Km values (Figure 18) from diatom growth curves [Timmermans et al., 2004] and dissolved Fe (<0.2 μm fraction) during an in situ experiment (Figure 19) is an important step forward, but only one step on a long road to unraveling the complexities of iron-phytoplankton interactions.
7. Grazing Impacts
 In discussing the potential roles of grazers in Fe-fertilized ecosystems, it is useful to distinguish between two major size classes: the micro- and mesozooplankton. Operationally, the “microzooplankton” includes all consumers <200 μm in size, a size class typically dominated by a diverse assemblage of heterotrophic and mixotrophic protists. Such organisms are the major direct consumers of phytoplankton in the open oceans [Calbet and Landry, 2004], and can grow at rates comparable to or greater than similarly sized phytoplankton. Because of these qualities, meaningful studies of the grazing impacts and population responses of the microzooplankton can be conducted on the days-to-weeks scale of the typical fertilization experiment. The “mesozooplankton,” on the other hand, are larger (>200 μm) animals with longer and more complex life histories, some (in polar and subpolar systems) with generation times of a year or more. Mesozooplankton play central roles in ocean food webs as consumers of large phytoplankton, as trophic intermediates to higher-level consumers (e.g., fish), as fecal pellet producers (thereby accelerating sinking and export flux), and as predators and regulators of the microzooplankton, all of which make them relevant to understanding system level responses to Fe. However, the temporal and spatial scales of experimental studies conducted to date are inadequate to study population and community responses of mesozooplankton to Fe, or to predict their direct and indirect implications for large-scale and long-term Fe fertilization. Because our current understanding of Fe effects on ocean ecosystems does not extend past single-celled organisms, the comments below are meant to apply mainly to this “microbial” portion of plankton communities.
 Since Fe concentration strongly affects growth rate and, therefore, the intrinsic competitiveness of phytoplankton taxa and size classes (Figure 18) and since light-related variables, like WML depth, appear to control maximum levels of biomass accumulation (Figure 12), grazing interactions would seem, at first glance, to have little influence on phytoplankton responses to Fe addition. However, temporal changes in populations and communities do not simply reflect growth rates or potential, but rather the net realized differences between growth and mortality rates. Consequently, if the goal is to predict the timing and magnitudes of Fe responses at the community and system levels, the mortality “environment” is as important as the growth environment for determining net rates of change. The role of mortality becomes particularly apparent when one considers how initially rare taxa rise to dominance quickly in Fe fertilization experiments. In IronEx-2, for example, the estimated Fe stimulated growth rate for the diatom Pseudo-nitzschia was ∼2 cell doublings d−1 (μ = 1.4 d−1 [Landry et al., 2000b]) while Prochlorococcus growth rate was observed to shift up from 0.7 d−1 to 1.1 d−1when Fe was added [Mann and Chisholm, 2000]. If left entirely to the growth rate difference of 0.3 d−1, the initially rare Pseudo-nitzschia would need about 8 days of growth (16 generations) to catch up to Prochlorococcus for each order-of-magnitude difference between the two in their initial contributions to community biomass. In reality, Pseudo-nitzschia was rapidly installed as the bloom dominant within 5 days of the IronEx-2 patch fertilization because Prochlorococcus and other small cells were held entirely in check by grazing (Figure 21). For these initially dominant phytoplankton, already growing at relatively high rates under ambient conditions, the adjustments that lead to continued grazing balance at Fe-stimulated rates of growth can be relatively modest, e.g., a 50% increase in grazer standing stock, or perhaps just slight increases in prey vulnerability due to Fe-enhanced changes in cell size or physiochemical cell surface properties [Monger and Landry, 1990; Monger et al., 1999].
 Diatoms do not dominate Fe-stimulated blooms because they physically cannot be eaten by microzooplankton. Within this diverse group, there are clearly organisms with appropriate size, behaviors and apparent preferences for feeding on diatoms [e.g., Gaines and Taylor, 1984; Jacobson and Anderson, 1986]. In IronEx-2, large dinoflagellates and ciliates eventually exerted a heavy grazing toll on Pseudo-nitzschia [Landry et al., 2000b], and large protists were also the major grazers in the diatom-dominated waters of SOFeX-South [Coale et al., 2004; M. Landry, personal communication, 2005]. It does appear to be true, however, that initially rare and relatively large diatoms enjoy a release from heavy grazing pressure in at least the early stages of an Fe-stimulated bloom because their specialized grazers are not sufficiently abundant to control them when their growth rates are strongly stimulated by the added Fe. The larger the difference between the growth rates of large rare species in Fe-deficient ambient waters and their maximal growth potential with added Fe, the larger will be their net growth rate advantage in the early bloom. It is also likely the case that regulatory potential by microzooplankton grazers diminishes with increasing cell size of phytoplankton, such that “giant” diatoms, at the extreme, are minimally vulnerable to such losses. Thus grazing pressure would seem to act in a way that systematically reinforces the selective growth advantages of large cells under Fe-replete conditions, accelerating, in fact, the rate at which they can rise relative to other cells to dominate the community response.
8. Carbon Fluxes
 Photosynthetic fixation of CO2 in all experiments depletes the pool of dissolved inorganic carbon (DIC) in seawater by converting it into particulate organic carbon (POC). The DIC loss is paralleled by a decrease of the fugacity of CO2 in surface waters (Figures 3, 4, 6, and 11), where an undersaturation versus atmospheric CO2 may develop. For a sufficient rate of gas exchange, this undersaturation may drive an influx of CO2 from the air into the sea, thus somewhat replenishing the DIC pool. In addition, part of the POC inventory may settle out into deeper layers, thus exporting carbon into the deep sea (Figures 8 and 9).
 A detailed comparison of carbon budgets among the eight Fe experiments would be desirable, but the designs, implementations, weather conditions and actual evolutions of these experiments have been quite different. Patch dilution has notably varied greatly (Figure 15) and efforts to quantify this using the SF6 (and 3He) tracer(s) have proven quite challenging [Goldson, 2004; Bakker et al., 2005; C. S. Law et al., Patch evolution and the biogeochemical impact of entrainment during an iron fertilization experiment in the subarctic Pacific, submitted to Deep-Sea Research, Part II, 2005], if pursued at all. In general, the variability and patch dilution interfere with sampling, for example at any given day of any experiment nobody can guarantee the true core (SF6 maximum) of the patch was sampled. In other words, the response reported is intrinsically a function of sampling strategy and intensity. Also sampling grids differ between variables, the Primary Production assay typically being done at one in-patch station, while some state variables have been mapped extensively. Finally, the C budgeting efforts for various experiments (sources as in caption Figure 22) have used different approaches. Thus overall, the comparison presented here is somewhat incompatible. Further study may improve this, but full comparability may never be achieved.
 Depth-integrated rate estimates (mmol m−2 d−1) for the different Fe fertilization experiments tend to vary less (Figure 22) than previously shown maximum changes per volume seawater (Figure 11). A deep WML is unfavorable for phytoplankton growth, for example, but the effect is partially offset by integrating over a greater depth. Thus the “rankings” of experimental impact shift somewhat between Figures 11 and 22. On this basis, the primary production of IronEx-2 now exceeds that of SEEDS. This is consistent with the twofold higher optimal growth rates of IronEx-2 predicted from temperature differences (Figure 14).
 For the experimental data represented in Figure 22, measured DIC drawdown averages half (51 ± 26%) of primary production rates, from 14C uptake measurements which typically underestimate true gross production by a factor of two or more [e.g., Laws et al., 2000]. In the same data set, the POC increases average one quarter (26 ± 22%) of primary production. Excluding the unusually high ratios for SEEDS, these POC ratios reduce to 45% of DIC drawdown and 18% of primary production. It is clear from these relationships that only SEEDS demonstrated the kind of response, very high efficiency in moving carbon from DIC to POC pools via photosynthetic fixation, that one might associate with a relatively pure phytoplankton “culture.” More generally, the transfers are characterized by large inefficiencies (only 18–26% of production accumulates as POC), which speaks to the importance of loss processes due to grazing and related trophic processes. Such results are consistent with rate estimates of phytoplankton instantaneous growth and biomass accumulation in the fertilization experiments. For example, in IronEx-2, growth rates of 1.5 to 2 cell doublings d−1 (μ of 1.0–1.4 d−1) produced only a net fivefold increase in phytoplankton carbon biomass over the course of a week [Landry et al., 2000a]. Similarly, in the SOFeX-South patch, phytoplankton biomass only doubled despite mean estimated growth rates (μ of 0.2–0.3 d−1) that could have led to a biomass increase of 20-fold or more [Coale et al., 2004; M. Landry, personal communication, 2005]. Because production/biomass accumulation precedes grazer responses, there must also be a substantial time dependency to calculated estimates of the DIC ⇒ primary production ⇒ POC transfer ratios. For example, phytoplankton POC estimates plateaued for 4–5 days at peak levels in the IronEx-2 patch, during which phytoplankton continued to grow at a high rate while their biomass remained static. Clearly, during such times, in the mid to late stages of bloom response to added Fe, the ratios of DIC and POC changes to time-integrated primary production should decline significantly.
 Of the average 70–80% of primary production that does not accumulate as POC in Fe-fertilized patches, some appears as DOC or contributes to export. Reports on DOC changes have been sporadic. In CARUSO/EisenEx over the first 12 days, during which the DIC decreased 12 mmol m−3 (Figure 6), the DOC showed a small increasing trend from about 48 to 52 mmol m−3 suggesting a net increase on the order of 4 mmol m−3 ∼30% of the DIC decrease. However the DOC data were scattered, and over the complete 21 days there was no real trend. In addition, the 4 mmol m−3 change is hardly discernible at an about 1 mmol m−3 analytical reproducibility.
 The CO2 gas flux into the sea has been calculated for most experiments and is a small ∼8% portion of the DIC removal rate, ranging from 2.7% at SOIREE to as much as 13% for both SOFeX sites (Figure 22). The gas flux tends to be about 3% of primary production (Figure 22). The percentage variability between experiments is partly real, partly also due to different calculation approaches and applied gas exchange coefficients [e.g., Wanninkhof, 1992; Wanninkhof and McGillis, 1999]. Nevertheless, the slow replenishment of CO2 from the air is as expected due to the slow relaxation time of this gas as a result of its equilibration reactions with the seawater, as opposed to nonreactive gases, e.g., O2 or SF6. On the other hand, this also implies that CO2 supply from the air continues long after the observer ships have left the scene. Hence the overall drawdown of CO2 from the air likely is more then shown here only for the 12–24 day observation periods of individual experiments. In the end, the DIC deficiency (e.g., Figures 6 and 11) will be largely compensated for from the atmosphere, as well as from net heterotrophic respiration, once the bloom has collapsed. Such a collapse was observed in SERIES, but the CO2 gas flux results are not yet available. The relative contribution of both terms (gas flux versus respiration) will remain the subject of debate until longer observation strategies are implemented in this type of experiment.
 Export flux of organic carbon into deeper waters has been assessed by difference from budget calculations. For SEEDS an elegant effort has been reported [Tsuda et al., 2003, supplement]. Similarly for CARUSO/EisenEx (not shown) such an effort has been made (U. Riebesell et al., unpublished manuscript, 2003). However with export being only a small fraction of primary production and the changes in DIC and POC, and given the difficulty in assessing DOC and its changes, calculated export signals hardly exceed the noise of the overall budget, if at all. The export of organic matter remained constant in SOIREE, perhaps due to its short duration and patch dilution [Charette and Buesseler, 2000; Nodder and Waite, 2001]. The most direct evidence for export was observed in sediment traps of SERIES upon collapse of the bloom (Figure 9 (bottom)). However, derivation of an overall export number does require extra information on, e.g., trapping efficiencies [Boyd et al., 2004a]. Finally, for SOFeX-South, the observed 234Th deficiencies have led to an estimate of carbon export which is significant, yet still modest relative to similar estimates for the natural Southern Ocean [Buesseler et al., 2004, 2005].
9. Light Climate Due to Wind Mixed Layer Depth
 Nine experiments in the 1993–2004 era have each been successful in following an Fe-enriched water mass and observing a distinct impact, small or large. The combined nine experiments have demonstrated that the depth of the wind mixed layer (WML), in regulating light climate, is the major factor controlling photosynthesis in the high-nutrient low-chlorophyll regions of the ocean. This is not surprising. Light climate with its various parameterizations (e.g., PAR, WML depth, self-shading) has been recognized in every oceanography textbook [e.g., Lalli and Parsons, 1993] as a key control of photosynthesis. Moreover, for some of the Fe enrichment experiments in HNLC regions, light effects have been considered explicitly, such as WML depth and self-shading in SOIREE [Boyd et al., 2000; Gall et al., 2001b], and self-shading in SOFeX [Coale et al., 2004], and the unfavorable WML depth during CARUSO/EisenEx [Bozec et al., 2005]. Nevertheless it is by combination of these experiments (Figure 12; Tables 2 and 3) that the dominant role of WML depth for the worldwide HNLC regions has become more evident. In other words, when challenged to specify just one over-ruling limitation for each individual Fe enrichment experiment, one might conclude from Figure 12 that SOIREE and CARUSO/EisenEx in the Southern Ocean were predominantly light limited, while the SEEDS and SERIES regions were predominantly Fe limited, but IronEx-2 and SOFeX-South and SOFeX-North were intermediate between limitation by light and iron. However, the concept of one overruling limiting factor is fruitless [de Baar, 1994; de Baar and Boyd, 2000]. Instead colimitation by iron and light [Sunda and Huntsman, 1997; van Leeuwe, 1997; Maldonado et al., 1999; Lancelot et al., 2000; van Oijen, 2004] best describes the HNLC regions in 40% of the world oceans. Otherwise, the case for predominantly Fe limitation in SEEDS and SERIES is further underlined by the virtual complete exhaustion of notably silicate as well as nitrate (Figure 8 and Figure 10 (middle)) after 13 and 18 days respectively, where during SERIES indeed the bloom was observed to collapse in days 19–23 (Figure 10 (top)).
Table 3. Preliminary Overview of Derived DIC/Fe Efficiencies for Several Experimentsa
The added Fe (kg) is accurately known (where 1 mole Fe is 55.847 g). Patch dilution hampers the accuracy of the calculation of overall DIC loss. The DIC removal from several sources is processed with various approaches (not necessarily consistent among each other) listed below (A–F) to obtain DIC loss (moles) for each experiment. The total DIC loss (moles) by addition, where alternative estimates (values in italic font) were ignored. The DIC/Fe efficiency scales inversely with WML depth (excluding IronEx-1): DIC/Fe = −240.3WML + 18818 (linear, with R2 = 0.73 for n = 7 data points), with SERIES deviating above the line and two SOFeXs deviating below the line. The grand total overall DIC/Fe efficiency is 5620 and about threefold below the optimum 15,000 calculated for SEEDS. The Antarctic summation of SOIREE, EisenEx, and SOFeX-South in high-silicate, high-nitrate Antarctic waters yields an efficiency DIC/Fe = 4347 and hinges largely on the most robust estimates by Bakker et al. . For the time being, this may serve as the state-of-the-art gross DIC removal estimate for austral spring-summer-early autumn season (October–February), e.g., when assessing the impact of the approximately 10-fold extra atmospheric Fe dust input [Edwards et al., 1998] during the Last Glacial Maximum.
(A) Steinberg et al. ; integrated using WML and patch size of Figures 11 and 15 for IronEx-1 initial patch size; for IronEx-2, both initial and final patch size provide lower and upper limit of C/Fe efficiency, respectively. (B) Boyd et al. ; providing an upper limit of 800 t from algal carbon integration over 200 km2 patch size and 400 t when assuming algal carbon follows SF6 distribution. (C) Bakker et al.  estimates for both SOIREE and EisenEx based on covariance analysis with the SF6 distributions, apparently the most robust DIC removal estimates available now. (D) Tsuda et al. [2003, Figure S1]. (E) Hiscock et al. ; see also text description of SOFeX and Figure 11. Integration over initial patch size is deemed to provide a lower limit estimate of C/Fe. For example, the larger final patch size of 1000 km2 as suggested by Buesseler et al.  would increase the DIC/Fe efficiency accordingly to a quite high ratio DIC/Fe = 35,680. (F) Boyd et al. [2004a, Table 2]. The change in particulate organic carbon (ΔPOC) estimate for the entire patch is based on Chl a signal from Sea-viewing Wide Field-of-view Sensor (SeaWiFS) ocean color image [Boyd et al., 2004a, 2004b, suppl. Figure 3] multiplied by C/Chl a ratio = 80 in patch center at day 18 and WML depth of 25 m. This yields a C/Fe efficiency similar to SEEDS. The ΔDIC estimate and the ΔPP for the iron-mediated increase in net primary production are both for the patch center only and yield quite high C/Fe efficiencies.
6 10−3 mol m−3
27 10−3 mol m−3
27 10−3 mol m−3
4 10−6 mol m−3
61 10−3 mol m−3
1.3 mol m−2
14 10−3 mol m−3
23 10−3 mol m−3
23 10−3 mol m−3
41.3 10−6 mol m−2
ΔPP 1.61 mol m−2
41.3 10−6 mol m−2
ΔDIC 1.1 mol m−2
ΔPOC 1776 t
 The well-documented climatology of wind velocity and WML depths in the world oceans [Monterey and Levitus, 1997] (available at http://www.nodc.noaa.gov) is a major resource when one wishes to understand, predict or simulate what prevents HNLC regions from achieving full biomass growth potential. Notably in the Southern Ocean this is the case [Mitchell et al., 1991; Lancelot et al., 1993, 2000; Hannon et al., 2001; Pasquer et al., 2005]. Here the most favorable conditions are in austral summer months December–January. Indeed SOFeX-South in January 2002 showed a major impact of Fe enrichment. These 2 months being favorable also for many other lines of research and expeditions, it has proven to be very difficult to secure suitable shiptime for Fe experiments in the Southern Ocean. Thus SOIREE had been accommodated during late summer, and CARUSO/EisenEx during early austral spring, when wind and WML conditions were not optimal for light.
10. Efficiency of Moles Carbon Removed per Moles of Iron Added
 The key role of light climate also implies that simple stoichiometric arguments for the amount of Fe required for a given level of C, N, P or biogenic Si production by phytoplankton, as popular during the past 15 years, do not hold in general. This is even more the case because the added [Fe(II)] is rapidly lost via the formation of colloids, and then into larger (>0.2 μm) particles. In other words, for any amount of dissolved Fe(II) added, only a minor portion of perhaps about 20% remains soluble and available for direct uptake by phytoplankton. Part of the colloid portion may, or may not, become indirectly available via photoreduction [Rijkenberg et al., 2005], and part of the particulate Fe may, or may not, again become available due to dissolution and remineralization via grazing [Barbeau and Moffett, 2000].
 In SEEDS, the most optimal and perhaps also most straightforward experiment, one single Fe infusion, bringing the initial total Fe to about 4 10−6 mol m−3, led to a maximum DIC removal of 61 10−3 mol m−3 (Table 3). Accordingly, each Fe atom removed 15,000 carbon atoms as DIC or CO2. Where in general the DIC removal was found to be about half of the primary productivity (Figure 22), in SEEDS it was 88% such that one expects a ratio C/Fe = ∼17,250 for diatom growth in SEEDS. The cellular Fe content of plankton has long been a vexing question, but the first direct measurements of Fe quotas of individual plankton cells have recently been made by Twining et al. [2004b] detecting Fe in single cells by synchrotron X-ray radiation (Table 3). Within the Fe-enriched patch of SOFeX-South, the reported ratio C/F e = ∼25,000 [Twining et al., 2004b] is remarkably consistent with above expected ∼17,250 in SEEDS. The C/Fe = ∼25,000 is substantially less than initial suggestions of oceanic phytoplankton cellular Fe requirements as high as 500,000 on the basis of the Fe′ paradigm in EDTA-manipulated laboratory cultures [Sunda et al., 1991]. The ratio C/Fe = ∼33,000 previously calculated from excess (i.e., biogenic) particulate Fe in the deep North Pacific [Martin et al., 1989] would agree more closely, but may be fortuitous as deep biogenic particles are not necessarily comparable with the Fe/C of primary production. The ratio C/Fe derived from dissolved Fe versus nitrate (C/Fe = ∼483,000) or versus O2 (C/Fe = ∼384,000) in the nutricline of the North Pacific [Martin et al., 1989; Martin, 1992] (discussed by de Baar and de Jong [2001, p. 164]) is much higher, but again not necessarily comparable with newly synthesized phytoplankton cells.
 In fact, the DIC/Fe ratio of 15,000 of SEEDS is a maximum value, occurring during the early build up of the bloom with more to be returned to DIC by heterotrophic respiration during the decline phase. For the sake of argument, the general POC buildup is about half of DIC removal (the other half is recycled by respiration), but higher at 79% for SEEDS such that the maximum POC/Fe efficiency of in situ Fe enrichment would be on the order of C/Fe = 11,800, potentially an upper limit for drawdown of CO2 from the atmosphere. In fact, the thus far observed net export flux is only a few percent (Figure 22), this in keeping with a typical f ratio in the order of 1–10% of primary production for ocean plankton blooms. The primary production in general being about twice the DIC removal but 115% for SEEDS, this implies 1.15–11.5% of DIC removal will eventually be exported. Thus the ultimate efficiency of CO2 removal from surface waters would be more on the order of 150–1500 C atom for each Fe added. For all other conditions with a deeper WML, notably in the Southern Ocean, and in all other experiments (Figure 11), the DIC/Fe efficiency would be less than the upper limit C/Fe = 11,800, here derived from SEEDS.
 The major uncertainty in such overall DIC/Fe efficiency estimate is due to the patch dilution somehow to be taken into account when converting from maximum DIC loss (mol m−3) to overall DIC removal (moles) for the overall experiment. Here the above estimate for SEEDS is relatively robust as its patch dilution factor was only threefold. Table 3 provides an account of overall DIC/Fe efficiencies which indeed tend to be less than the above derived optimum DIC/Fe = 15,000 ratio, but all suffering greater or lesser uncertainty due to patch dilution. Nevertheless, the derived C/Fe efficiencies, being either better or worse, also tend to scale inversely with WML depth (Table 3, caption).
11. Iron Fertilization During the Last Glacial Maximum and Anthropocene
 The implications of above DIC/Fe ratio values for the last deglaciation (17,000–11,000 y BP) as well as for intentional Fe fertilization of the modern ocean have been reported elsewhere (H. J. W. De Baar et al., Iron makes big diatoms blooming, but cannot change carbon dioxide and climate, submitted to Science, 2005, hereinafter referred to as de Baar et al., submitted manuscript, 2005). Briefly, during the Last Glacial Maximum (LGM) the Fe dust input into the Antarctic region was 11-fold the modern dust flux [Edwards et al., 1998]. Sometime after this dust flux terminated, the atmospheric CO2 has risen initially with 80 10−6 atm and eventually with 90 10−6 atm to the preindustrial value of 280 10−6 atm [Petit et al., 1999; Watson et al., 2000]. Taking the Antarctic summation of SOIREE, EisenEx and SOFeX-South in high-silicate high-nitrate Antarctic waters yields an efficiency DIC/Fe = 4347 (Table 3), which by assuming 20% export, yields an export efficiency C/Fe = 870. This combined with a factor 10 in Fe dust flux, 30% wet deposition of which 14% dissolves, and a 3 month austral summer growth season, yields an Fe fertilization effect which can account for only 0.5% of the observed rise of atmospheric CO2. Taking a more favorable export ratio C/Fe = 3257 after Buesseler et al. , the Fe effect would be higher, but still only 2 % of the observed rate of atmospheric CO2 increase. When also taking a more favorable overall mean ∼32% dust dissolution (operational defined range is 9–89% [Edwards and Sedwick, 2001]) instead of above 30% of which 14% dissolves, the Fe effect might be as high as ∼15% of the observed CO2 rise.
 Similarly extrapolation to the current anthropogenic fossil fuel CO2 emission rate of about 6.6 Petagram C yr−1 (0.55 1015 mol C yr−1) would lead to a required Fe fertilization of 0.63 1012 mol Fe yr−1 or 35 109 kg Fe yr−1, i.e., 35 million tons Fe yr−1 (de Baar et al., submitted manuscript, 2005). This is 40-fold more Fe than originally hypothesized (430,000 tons Fe to remove 3 PgC yr−1 [Martin, 1990]). Extrapolation of the most favorable C export of C/Fe = 3257 for only SOFeX-South in austral summer [Buesseler et al., 2004] would yield a lower required Fe fertilization of 0.17 1012 mol yr−1 or 9.4 109 kg yr−1. This is 9.4 million tons Fe yr−1.
 One major reason for in situ enrichment experiments was to independently verify bottle incubations, notably with respect to perceived artefacts of the latter. One such artefact, the control bottles also outgrowing the field biomass (Figure 1) due to more favorable light climate in deck incubators, has now been confirmed. Indeed, while Fe addition is effective when light is optimal (Figures 11 and 17), added Fe has little effect when light conditions are unfavorable.
 The in situ experiments have their own artefacts as well [Boyd et al., 2002]. Most notably, the 100-fold higher Fe is a strong perturbation of Fe chemistry, the effects of which on phytoplankton growth have yet to be understood. Moreover, the rapid formation of Fe colloids is reason for some caution when interpreting coagulation of particles and export fluxes with approaches and concepts developed in the unperturbed natural ocean. Briefly, freshly formed Fe colloids are adhesive and reactive and may not necessarily be inert versus coagulation of biogenic debris and scavenging processes of trace elements and isotopes. In fact, Fe colloid formation is a standard method worldwide for purifying natural fresh waters in the production of drinking water. Obviously the applied levels are much higher, but some effectiveness at the typical 3–5 nM concentrations of in situ experiments is not necessarily ruled out.
 Another physical phenomenon to be considered is patch dilution (Figure 15), which gives rise to the chemostat effect [Abraham et al., 2000]. Patch dilution can be overcome, in principle, by fertilizing very large patches vis-à-vis the expected shear stress and wind forcing, such that the core of the patch will be unaltered and dilution only a boundary effect. On the other hand, the patch sizes used thus far, 80 or 225 km2, are perhaps of similar order as natural wet deposition mesoscale events (combined rain/dust storms) or natural upwelling events, bringing in extra Fe from either above or below. In other words, patch dilution is, on the one hand, an artefact for understanding details of Fe responses in a controlled reproducible manner, yet, on the other hand, perhaps quite realistic in the natural ocean.
 Experimental oceanography has now been proven a powerful and exciting approach for unraveling the drivers and inner mechanisms of plankton ecosystems. This preliminary synthesis next needs more rigorous verification by application of a generic plankton ecosystem simulation model to most or all of the 8–9 experiments. This is crucial also for the design of the next generation of experiments, some by tinkering with innovative techniques for delivering the extra Fe more naturally and effectively, others by relying on natural Fe supply and gradients instead. Moreover one would like to be able to follow an experimental water body over longer time periods of several months. Finally new techniques are desirable to quantify more reliably and routinely the exchanges of CO2, DMS and other biogenic gases with the atmosphere, as well the thus far difficult to quantify biogenic export to the deep ocean.
 The authors are most grateful to the organizers of The Ocean in a High CO2 World symposium for the invitation, with special thanks to Maria Hood for realizing an excellent and exciting meeting within sight of the iron tower of Eiffel. During preparation the team of coauthors kindly and diligently provided much data, graphics, and other precious findings and insights, where the temptation for special thanks to special coauthors is here avoided. Silvio Pantoja and two anonymous referees are acknowledged for their constructive editorial comments as well as patience. This and detailed comments by coauthor Paul Harrison have remedied many flaws in the first submitted version. Hendrik van Aken, Yvonnick Le Clainche, and Jaap van der Meer kindly assisted with various data and graphics. The vision and enlightenment of the late John Martin as founding father of this exciting research field cannot be overestimated. This paper is dedicated to all the heroes who went out to sea to dump in the iron and do the accurate measurements. Their original research articles are the basis of this preliminary review. We are most grateful to the officers and crew, and our shipboard fellow scientists, aboard the research vessels Columbus Iselin, Melville, Tangaroa, Kaiyo-Maru, Revelle, John P. Tully, and El Puma and icebreakers Polarstern and Polar Star. The commitment and support of all these people, and the dedication of ships by AWI, Fisheries Agency Japan, NIWA, the Canadian and U.S. Coast Guards, and U.S. National Science Foundation Ocean Sciences, have created and realized this new era of experimental oceanography. This research was supported by the European Union through programs CARUSO (1998–2001), IRONAGES (1999–2003), and COMET (2000–2003); the Netherlands-Bremen Oceanography program NEBROC-1; and the Netherlands Organization for Research NWO through the Netherlands Antarctic Program project FePath. Both the U.S. National Science Foundation and the U.S. Department of Energy provided significant support for the SOFeX program. M.R.L. acknowledges the U.S. National Science Foundation for support of IronEx and SOFeX projects and related studies (OCE-9912230, -9911765, and -0322074).