Iron in the Ross Sea: 2. Impact of discrete iron addition strategies

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 110,, doi: /2004jc002568, 2005 Iron in the Ross Sea: 2. Impact of discrete iron addition strategies Kevin R. Arrigo and Alessandro Tagliabue Department of Geophysics, Stanford University, Stanford, California, USA Received 29 June 2004; revised 3 December 2004; accepted 6 January 2005; published 8 March [1] Presented are results of a regional-scale numerical investigation into the effectiveness of Fe fertilization as a means to increase the efficiency of the biological pump in Fe-limited waters of the Ross Sea, Antarctica. This investigation was conducted using a modified version of the Coupled Ice And Ocean (CIAO) ecosystem model of the Ross Sea sector of the Southern Ocean. Four sets of experiments were performed, investigating the impacts of differences in (1) timing of fertilization, (2) duration of fertilization, (3) amount of Fe added, and (4) size of the fertilized patch. Results show that the stimulation of air-sea CO 2 exchange (FCO 2 ) depends primarily on the timing of fertilization, regardless of the amount of Fe added. When Fe was added at the optimal time of year, FCO 2 from the atmosphere into the Ross Sea was increased by 3 22%, depending on fertilization strategy. Increasing patch size produced the largest response, and increasing initial Fe concentration produced the smallest. In all cases, as the intensity of Fe fertilization increased, the fertilization efficiency (increase in CO 2 uptake per unit added Fe) dropped. Strategies that maximized the fertilization efficiency resulted in relatively little additional CO 2 being drawn out of the atmosphere. To markedly increase oceanic uptake of atmospheric CO 2 would require the addition of large amounts of Fe due to the low fertilization efficiencies associated with maximum air-sea CO 2 exchange. Our results also show that differences in the fertilization strategy should be kept in mind when comparing the results of different Fe fertilization experiments. Citation: Arrigo, K. R., and A. Tagliabue (2005), Iron in the Ross Sea: 2. Impact of discrete iron addition strategies, J. Geophys. Res., 110,, doi: /2004jc Copyright 2005 by the American Geophysical Union /05/2004JC Introduction [2] Despite any efforts to curb anthropogenic release of carbon dioxide (CO 2 ), atmospheric concentrations are likely to increase for the foreseeable future. As a result, interest has been mounting to either find alternative sinks for anthropogenic CO 2 or devise mechanisms to remove CO 2 from the atmosphere [De Baar, 1992; Priddle et al., 1992; Riemer, 1996; Buesseler et al., 2004]. Currently, approximately 30% of the 7 Gt of CO 2 that is produced anthropogenically is removed by the oceans via natural processes [Takahashi et al., 2002]. It has been hypothesized, however, that this number could be increased by artificially stimulating the biological pump, the process whereby phytoplankton remove CO 2 from the surface ocean during photosynthesis and produce organic matter which is subsequently exported to depth when the algae die or are eaten by zooplankton [Longhurst and Harrison, 1989]. [3] The biological pump operates at varying efficiencies, depending on physical and chemical conditions of the water column. In the North Atlantic, where wintertime convection brings nutrient-rich waters to the surface, large phytoplankton blooms develop [Martin et al., 1993]. These blooms are composed primarily of large diatoms that are consumed by mesozooplankton and macrozooplankton such as copepods. Carbon is exported via sinking of both phytodetritus and fecal pellets with a relatively high efficiency [Lampitt, 1985; Lochte et al., 1993]. In contrast, much of the Southern Ocean is characterized by high nutrients but low chlorophyll a (Chl a) concentrations (HNLC), due mainly to limitation of phytoplankton growth by the scarcity of trace metals such as iron (Fe) [Martin et al., 1990; De Baar et al., 1990; Sedwick and DiTullio, 1997; Sedwick et al., 1997, 2000]. These conditions favor the growth of small phytoplankton whose fixed organic carbon is mostly remineralized in surface waters, resulting in low export efficiency. [4] In eight field experiments to date, exogenous Fe was added to a number of HNLC regions, including surface waters of the equatorial Pacific [Coale et al., 1998; Landry et al., 2000], the Southern Ocean [Boyd and Law, 2001; Gervais et al., 2002; Coale et al., 2004], and most recently, the Sub-Arctic Pacific [Tsuda et al., 2003; Boyd et al., 2004]. Although the chemical transformations accompanying the Fe additions are not completely understood, in all cases the growth of phytoplankton was stimulated (primarily diatoms, but sometimes small flagellates as well), albeit to varying degrees. Unfortunately, most of these artificially produced blooms were monitored for only a 1of16

2 ARRIGO AND TAGLIABUE: IRON IN THE ROSS SEA, 2 short time, and little is known about either the efficiency with which fixed carbon was exported from surface waters (but see Boyd et al. [2004]) or the depth and timing of organic matter remineralization following its production in the euphotic zone. In a modeling study of ocean fertilization in the equatorial Pacific using a global ocean general circulation model, Gnanadesikan et al. [2003] show that fertilization events can alter air-sea exchange of CO 2 over surprisingly large temporal (century) and spatial (1000 km) scales and that carbon sequestration efficiency is highly dependent on the depth of remineralization. [5] Here we present results of a regional-scale numerical investigation into the potential effectiveness of Fe fertilization as a means to increase the efficiency of the biological pump in waters of the Ross Sea, Antarctica, the most productive region of the Southern Ocean [Arrigo et al., 1998]. There are two major natural sources of Fe for surface waters of the Ross Sea, aeolian input associated with dust deposition on sea ice and its subsequent release during melting, and more importantly, upwelling of Fe-rich deep water (due to remobilization of Fe from sediments) [Loscher et al., 1997]. Together, these Fe sources are generally insufficient to allow complete utilization of macronutrients within the upper mixed layer (UML), where approximately one third of the initial nutrient load remains at the end of the spring-summer phytoplankton blooms [Arrigo et al., 2000]. By introducing additional Fe into the system, it is possible that the phytoplankton blooms can be extended, resulting in more complete utilization of available macronutrients and additional drawdown of atmospheric CO 2. [6] The study of the impacts of Fe fertilization on phytoplankton net primary production (NPP) and carbon sequestration presented here augments recent modeling studies by Hannon et al. [2001] and Gnanadesikan et al. [2003]. Hannon et al. [2001] used a one-dimensional ecosystem model to simulate the response by phytoplankton to Fe additions in the subantarctic zone during the Southern Ocean IRon Enrichment Experiment (SOIREE). Their results showed that carbon export only became important days after initial Fe fertilization, long after in situ sampling had ceased. By using a coarse grid global-scale model with low temporal resolution and highly simplified biogeochemistry, Gnanadesikan et al. [2003] were able to quantify both near- and far-field changes in oceanic CO 2 uptake resulting from Fe additions, as well as its potential sequestration efficiency over centennial timescales. In contrast to these two studies, the biological component of our regional-scale Coupled Ice, Atmosphere, and Ocean (CIAO) model is of much higher temporal (1 hour) and spatial (25 km) resolution and incorporates more mechanistic biogeochemistry, including an Fe cycle, state variables for multiple phytoplankton taxa, each with unique photophysiology and nutrient biochemistry, and multiple factors capable of limiting phytoplankton growth (e.g., NO 3, PO 4, Fe, Si(OH 4 ), light, and temperature). These aspects of the model allow us to expand the findings of Hannon et al. [2001] and Gnanadesikan et al. [2003] by resolving the three-dimensional dynamics at much smaller scales (e.g., shorter blooms, smaller fertilized patches) and adding important mechanistic biological detail to model predictions. Here we show how the timing, extent, duration, and intensity of discrete anthropogenic Fe fertilization events can alter the dynamics of phytoplankton biomass, nutrients, pco 2, and the air-sea flux of CO 2 (FCO 2 ) in Felimited waters. In addition, we show how changes in fertilization strategy can markedly influence the degree of CO 2 influx from the atmosphere. 2. Methods 2.1. Model [7] This investigation was conducted using a modified version of the CIAO ecosystem model of the Ross Sea, the full description of which is given by Arrigo et al. [2003] and Worthen and Arrigo [2003]. Briefly, the physics of CIAO are based on the Princeton Ocean Model (POM), a primitive equation ocean circulation model [Blumberg and Mellor, 1987] with vertical mixing calculated using a turbulence closure scheme [Mellor and Yamada, 1982]. The original biogeochemical state variables in CIAO included two phytoplankton groups (diatoms and the Haptophyte Phaeocystis antarctica), Si(OH) 4,NO 3, Fe, detritus, and zooplankton. For the purposes of this work, we have added an additional nutrient (PO 4 ), the carbonate system, calculation of the partial pressure of CO 2 (pco 2 ), and FCO 2. [8] Sinking of phytoplankton and detritus is parameterized as a linear function of particle density [Arrigo et al., 2003]. This simple approach allows for the increased sinking speeds that would be expected as aggregation becomes important at higher particle densities, and yields results that are consistent with typical sinking fluxes observed in the Ross Sea [Dunbar et al., 1998]. This formulation cannot, however, simulate the extremely high sinking fluxes that have been occasionally observed in these waters [DiTullio et al., 2000; Asper and Smith, 2003]. Sinking particles are assumed to be remineralized at a biomass specific rate of 0.03 day 1 [Tagliabue and Arrigo, 2005]. [9] Additional details of these new CIAO formulations can be found in the first part of this paper, by Tagliabue and Arrigo [2005]. It is worth noting here that the simple Fe cycle used in the present model does not include photochemistry or Fe scavenging by particles, processes that are likely to be important in surface waters with high surface Fe concentrations, but that are not yet well understood. A more mechanistic Fe cycle is currently being added to CIAO, and the impact of such a model improvement should be clearer in the near future (A. Tagliabue and K. R. Arrigo, manuscript in preparation, 2005) Model Experiments [10] In order to assess how addition of Fe to surface waters may alter biogeochemical dynamics, we compared results of model experimental manipulations to results obtained during the standard run of CIAO, where no exogenous Fe was added (although natural Fe sources were still included). In all model runs (experimental and standard), CIAO was initialized and forced with climatological fields as described by Arrigo et al. [2003]. [11] We ran four sets of model experiments, with each set investigating the impact of a different aspect of a possible 2of16

3 ARRIGO AND TAGLIABUE: IRON IN THE ROSS SEA, 2 Figure 1. Map of the southwestern Ross Sea study area (73 S 78 S, 160 E 155 W) showing the size and location of Site A (76.5 S, 177 E) and Box A (76 S to77 S and E to E). Fe fertilization strategy for the Ross Sea. These included the impacts of differences in (1) the timing of a fertilization event, (2) the duration of a fertilization event, (3) the amount of Fe added (Fe initial, defined as the desired Fe concentration in the upper 25 m, the approximate depth of the seasonal thermocline, after Fe addition), and (4) the size of the fertilized patch. In all of these experiments, we analyzed model results with respect to the temporal dynamics of phytoplankton biomass (Chl a), nutrient drawdown (PO 4,NO 3, and Fe), surface pco 2, and FCO 2. Silicate was ignored due to its extremely high abundance in the Ross Sea (80 mm). Only one aspect of the fertilization strategy was altered in each simulation, holding all others constant. Unless otherwise noted, the conditions for each fertilization event were that (1) Fe initial was 2 nm, (2) Fe was added to a single model grid cell (646 km 2 ) at Site A (for location see Figure 1), (3) Fe was added on 31 December, and (4) Fe was added over a single day and then allowed to vary as determined by physical and biological dynamics. [12] Although CIAO was run for the entire Ross Sea domain [Arrigo et al., 2003], we focused our analyses on the more localized region being fertilized with Fe so that high temporal frequency dynamics could be more easily seen and understood. This is because over time, the patch will evolve and increase in area via strain and rotation [Abraham et al., 2000], resulting in the entrainment of surrounding low Fe waters and reducing the strength of the biogeochemical signal caused by the fertilization event. Therefore, in most cases, we concerned ourselves with model results in the vicinity of the fertilized patch. Model results were analyzed within Box A (Figure 1) with horizontal dimensions of five model grid cells by five model grid cells (16,160 km 2 ) surrounding the fertilization site (Site A). The site chosen for simulated fertilization (Site A) is located near the center of the large Phaeocystis antarctica bloom that typically forms during December in the southeastern Ross Sea, north of the Ross Ice Shelf (Figure 1). Whenever temporal averages over the course of the experiment were desired, these were calculated from 11 December, the earliest date of simulated Fe addition, through 10 April, when the effects of experimental manipulations were no longer measurably different from the standard run. Past experience tells us that CIAO does a particularly good job of simulating the physical processes, phytoplankton biomass, and nutrient dynamics associated with this bloom [Arrigo et al., 2003; Tagliabue and Arrigo, 2005]. 3. Results and Discussion 3.1. Standard Run [13] During the standard run, concentrations of Chl a in Box A (Figure 2a) associated with the phytoplankton bloom began to increase in late November, coinciding with the rapid sea ice retreat from the waters of the Ross Sea polynya (Figure 2a) and the increase in surface irradiance. The phytoplankton bloom attained its maximum biomass of 9 mg Chl a m 3 in early December, at which time concentrations began to drop due to the exhaustion of Fe (Figure 2d) within the UML [Arrigo et al., 2003]. By mid-january, algal biomass in the UML had fallen by >90%. A second, much smaller increase in Chl a took place in the austral autumn, and was associated with a 3of16

4 ARRIGO AND TAGLIABUE: IRON IN THE ROSS SEA, 2 Figure 2. Temporal changes in (a) chlorophyll a, (b) phosphate, (c) nitrate, (d) iron, (e) pco 2, and (f) FCO 2 (negative values denote a flux of CO 2 from the atmosphere into the Ross Sea) for surface waters of Box A as a function of the timing of the fertilization event. Results from the standard run (STD) are also shown, as is the sea ice concentration (0 = no ice, 1.0 = 100% ice cover) during the simulations. transient increase in Fe availability due to the entrainment of Fe-rich deeper waters as winds increased and waters in the UML cooled. This second phytoplankton bloom was shortlived, however, because the rapidly expanding sea ice cover (Figure 2a) and lower sun angle quickly reduced ambient irradiance to growth-limiting levels. [14] The rapid increase in phytoplankton biomass during the spring bloom was accompanied by a sharp drop in 4of16

5 ARRIGO AND TAGLIABUE: IRON IN THE ROSS SEA, 2 Table 1. Time-Averaged (11 December to 10 April) Biogeochemical Quantities Fe, nm Chlorophyll a, NO 3, mg m 3 mm Box A PO 4, mm pco 2, matm FCO 2, nmol m 2 s 1 Ross Sea FCO 2,TgC Standard Run a 4.86 Timing of Fertilization b 11 December December February February Duration of Fertilization b 1 day days days Initial Fe Concentration of Fertilized Patch (Fe initial ) b 1 nm nm nm Fertilization Patch Size b 646 km ,818 km ,160 km a Negative FCO 2 value denotes flux from the atmosphere to the ocean. b Values are expressed as percent difference from the standard run. concentrations of both PO 4 (Figure 2b) and NO 3 (Figure 2c) within the UML. However, because of the early exhaustion of Fe (Figure 2d), PO 4 and NO 3 were not entirely depleted within the UML during the spring bloom, reaching minimum concentrations of 1 mm and 10 mm, respectively, in agreement with numerous field observations [Tagliabue and Arrigo, 2005]. These residual nutrients amount to approximately one half and one third, respectively, of the original PO 4 and NO 3 pools in the UML prior to the spring phytoplankton bloom, in good agreement with field observations by Arrigo et al. [2000]. These reduced macronutrient levels persisted for a period of 3 months, at which time concentrations began to rise as a result of increased mixing between the UML and sub-pycnocline waters during autumn. [15] Concurrent with the increase in phytoplankton biomass was a 50% reduction in pco 2 within the UML, from 400 matm prior to the bloom to <200 matm at the time of maximum macronutrient drawdown (Figure 2e). Reduced levels of pco 2 also persisted for approximately 3 months, resulting in a net FCO 2 from the atmosphere to the surface ocean during this time period (negative FCO 2 values denote the flux of CO 2 from the atmosphere to the ocean). FCO 2 was most intense in late austral summer (Figure 2f), with the maximum influx of 308 nmol CO 2 m 2 s 1 (317 mg C m 2 d 1 ) coinciding with the seasonal increase in winds. This maximum value is consistent with an estimated seasonal peak in FCO 2 of approximately 290 mg C m 2 d 1 into the southwestern Ross Sea made by Sweeney et al. [2000] based on direct pco 2 measurements. The dramatic decline in FCO 2 in March during the standard run was the result of rapidly increasing ice cover (Figure 2a), which greatly restricted air-sea gas exchange [see Tagliabue and Arrigo, 2005]. During the period of persistent open water, FCO 2 averaged 132 nmol CO 2 m 2 s 1 into the Ross Sea (Table 1), with slightly more than 0.24 Tg C (15 g C m 2 ) taken up annually by the surface ocean within Box A and 4.86 Tg C taken up within the Ross Sea study area. The influx of atmospheric CO 2 is approximately equal to 10% of the CO 2 fixed annually via phytoplankton NPP both in Box A (2.3 Tg C) and over the entire Ross Sea study area (48.9 Tg C) Timing of Fe Fertilization [16] The impact of the timing of Fe fertilization was tested by varying the date of Fe addition with respect to the peak of the phytoplankton bloom. Preliminary simulations indicated that addition of Fe prior to the start of the phytoplankton bloom (or in its early stages) had virtually no effect on bloom or CO 2 dynamics, due to the already high Fe content within the UML at this time. Most of the Fe that was added quickly diffused and advected away from the area of the fertilized patch prior to uptake by the already Fereplete phytoplankton. Fe fertilization of the UML of Site A on 11 December, at the peak of the phytoplankton bloom, resulted in a transient increase in the mean Fe concentration to 0.1 nm within Box A (Figure 2d). This value is much lower than Fe initial (2 nm) because Fe was added to Site A only and results were averaged over Box A, which has 25 times the volume of Site A (when integrated over depth). Fe concentrations remained slightly elevated for 1 month after fertilization as it was being depleted by the phytoplankton community, whose peak biomass increased by 5% within Box A over that in the standard run (Figure 2a). This 1-day infusion of Fe increased the Fe concentration of Box A by an average of 54% during the course of the experiment (11 December to 10 April), resulting in mean Chl a concentrations for this same time frame that were 21% higher than in the standard run (Table 1). [17] As a result of the increased algal biomass, the uptake of PO 4 and NO 3 also increased when Fe fertilization took place on 11 December, drawing down these macronutrient concentrations within the UML of Box A by 30 40% 5of16

6 ARRIGO AND TAGLIABUE: IRON IN THE ROSS SEA, 2 Figure 3. Depth profiles of POC through time at Site A (76.5 S, 177 E) showing effect of Fe fertilization on 31 December (denoted by black arrow). Also shown is the change in the depth of the mixed layer over time. below those in the standard run (Figures 2b and 2c, respectively). Averaged over the entire experimental period, NO 3 and PO 4 concentrations in surface waters of Box A were 9.1% and 5.8%, respectively, below those of the standard run (Table 1). The relatively greater depletion of NO 3 (over PO 4 ) reflects the high N/P uptake ratio for P. antarctica which dominated the waters encompassed by Box A [Arrigo et al., 1999, 2002]. In addition, pco 2 was reduced when Site A was fertilized with Fe on 11 December, but only by 3%. This decline in pco 2 within the UML resulted in a similar increase in the FCO 2 from the atmosphere to the surface waters of Box A by an average of 2.7% during the experimental period (Table 1). [18] When fertilization was delayed until 31 December, approximately 3 weeks after the peak of the phytoplankton bloom, its biogeochemical impact was the greatest of all the dates tested. By this time of year, phytoplankton biomass had dropped to one third of its peak value (Figure 2a) and Fe concentrations were near zero (Figure 2d). Adding 2 nm Fe on 31 December resulted in a transient increase in Chl a which persisted for nearly 2 months (Figure 2a). The mean Chl a in Box A during the experimental period was 36% higher than that during the standard run (Table 1), well above the 21% increase in Chl a calculated for a fertilization event on 11 December. The Fe added on 31 December stimulated additional phytoplankton drawdown of PO 4 and NO 3 by approximately 0.2 mm and 4 mm, respectively, within Box A (9% and 14% reductions, respectively, relative to the standard run, Table 1). Concentrations of particulate organic carbon (POC) also increased markedly after fertilization, particularly in the UML of Site A (Figure 3). By mid-january, the POC generated by the fertilization event began to sink out of the water column, reaching a depth of >400 m by the end of February (Figure 3). As a result of the formation and subsequent export of Fe-stimulated POC, mean pco 2 levels within the UML of Box A during the experimental period decreased by 5% relative to the standard run (Table 1), driving an increase in FCO 2 from the atmosphere of >6%, almost triple the increase determined for an Fe fertilization event on 11 December. [19] By 2 February in the standard run, phytoplankton had declined to near background levels due to low growth rates over the previous 50 days, resulting from the lack of Fe in surface waters. When Fe was added to Site A on 2 February, algal productivity and nutrient uptake rebounded, but by somewhat less than when fertilized on 31 December, resulting in a 10-fold increase in Chl a from 0.2 to 2 mg m 3 (Figure 2a). When averaged over Box A during the experimental period, changes in Chl a, NO 3, PO 4, and pco 2 relative to the standard run were very similar to those obtained for a fertilization event on 11 December (Table 1). The reason for the diminished response for a 2 February Fe fertilization event compared to one on 31 December is that soon afterward, the winds had picked up, the mixed layer had deepened, and sea ice had begun to form. As a result, the newly stimulated phytoplankton bloom was unable to persist and declined soon after fertilization due to lack of light, rather than because it had used up its newly acquired Fe supply (Figure 2d). While, as noted earlier, changes in Chl a, NO 3,PO 4, and pco 2 were very similar to those obtained for the 11 December run, the same was not true for Fe and FCO 2, whose magnitudes were twice as large as that of the 11 December fertilization 6of16

7 ARRIGO AND TAGLIABUE: IRON IN THE ROSS SEA, 2 Figure 4. Temporal changes in (a) chlorophyll a, (b) phosphate, (c) nitrate, (d) iron, (e) pco 2, and (f) FCO 2 (negative values denote a flux of CO 2 from the atmosphere into the Ross Sea) as a function of the duration of the fertilization event. Results from the standard run (STD) are also shown, as is the sea ice concentration (0 = no ice, 1.0 = 100% ice cover) during the simulations. simulation. The increased rate of CO 2 influx from the atmosphere was due to the fact that when Fe was added on 2 February, the depletion in upper ocean pco 2 lasted until much later in the season (Figure 2e), when increased wind speeds enhanced the rate of air-sea CO 2 exchange [Tagliabue and Arrigo, 2005]. [20] As might be expected, adding Fe even later in the year (20 February) had almost no impact on phytoplankton dynamics, nutrient utilization, or FCO 2 (Figure 2, Table 1). Despite the fact that there was >130% more Fe available during the experimental period than during the standard run (Table 1), concentrations of sea ice were increasing rapidly and solar elevation and day length were diminishing. This decline in solar elevation, coupled with increased vertical mixing and a deepening UML (Figure 3) due to higher winds, resulted in dramatically reduced phytoplankton growth rates, such that they were not able to utilize the additional Fe Duration of Fe Fertilization [21] In the simulations referred to above, Fe was added such that the Fe concentration in the upper 25 m was set to 2 nm on a single day. Here we investigate the impact of longer sustained fertilization periods of 1 week and 7of16

8 ARRIGO AND TAGLIABUE: IRON IN THE ROSS SEA, 2 1 month, with Fe fertilization in each case beginning on 31 December. The simulation of a 1-day fertilization period here is the same as the 31 December fertilization event described previously in section 3.2 and will not be described again here. [22] By increasing the length of the fertilization period from 1 day to 7 days, the mean Fe concentration in surface waters of Box A during the experimental period increased to values that averaged 227% above those of the standard run (Table 1). These levels of increase likely represent upper limits because oxidation, precipitation, and scavenging of added Fe by particles, which should be particularly important at high Fe concentrations, is not presently being accounted for in CIAO. Nevertheless, as a result of the persistently high Fe (Figure 4d), Chl a concentrations after the initial bloom peak remained elevated for more than 2 months, such that the usual autumn phytoplankton bloom characteristic of normal conditions was difficult to discern (Figure 4a). Mean Chl a abundance during the experimental period was elevated by 47% relative to the standard run (compared to a 36% enhancement for the 1-day fertilization), with concentrations in Box A remaining elevated throughout January and February. [23] Reductions in surface water PO 4 (Figure 4b) and NO 3 (Figure 4c) resulting from the 6-day increase in the duration of Fe fertilization were not as great as might be expected. While the mean PO 4 and NO 3 concentrations in Box A during the experimental period were 12% and 18%, respectively, below those of the standard run, this degree of depletion was not much greater than the 9% and 14% depletion obtained for the 1-day fertilization event (Table 1). Despite the additional Fe availability, neither PO 4 nor NO 3 were exhausted within Box A after 7 days of Fe additions, or were even reduced to growth-limiting levels (although they were at Site A, the specific site of fertilization). Likewise, levels of pco 2 within surface waters of Box A (Figure 4e) were only reduced by an average of 7% relative to the standard run during the experimental period (Table 1), slightly more than the decrease in pco 2 that resulted from the 1-day Fe fertilization event (5%). However, the lowered pco 2 levels resulting from the 7-day fertilization event stimulated an additional influx of atmospheric CO 2 of 9% into Ross Sea surface waters relative to the standard run (Figure 4f, Table 1), approximately 40% greater than the difference between the 1-day fertilization event and the standard run. [24] In the simulation where fertilization continued for 31 days, Fe concentrations in surface waters of Box A increased to almost 0.4 nm by early February (Figure 4d), similar to wintertime Fe concentrations, and were not exhausted for the remainder of the simulation. Mean Fe concentrations within the UML of Box A during the experimental period were >800% higher than those of the standard run and almost fourfold greater than in the 7-day fertilization run. These elevated Fe concentrations resulted in a prolonged phytoplankton bloom within Box A, which lasted approximately 1 month longer than in the 7-day fertilization run and almost 3 months longer than in the standard run (Figure 4). Averaged over the experimental period, Chl a concentrations in Box A of the 31-day Fe fertilization simulation were 73% higher than in the standard run (Table 1). [25] The extended phytoplankton bloom of the 31-day fertilization run resulted in additional depletion of macronutrients from within Box A. Although seasonal minimum concentrations were only marginally lower than in the other experimental runs, low macronutrient concentrations persisted for a longer period of time. Mean NO 3 and PO 4 concentrations for the experimental period in the 31-day fertilization run were 27% and 18% lower, respectively, than in the standard run, but like the 7-day fertilization run, did not become limiting when averaged over Box A (Figures 4c and 4d), although they did at Site A. [26] The temporal trend in pco 2 (Figure 4e) for the 31-day fertilization run was similar to that of NO 3 and PO 4, with reduced pco 2 levels persisting later into the season. Despite the long fertilization period, pco 2 levels within Box A during the experimental period were <10% below those of the standard run, a surprisingly small decrease given the relatively large changes in mean Fe concentration and Chl a. This depletion of surface water pco 2 resulted in an additional influx of atmospheric CO 2 to Box A of 14% above that of the standard run (Figure 4f, Table 1). Interestingly, when all three of the simulations are considered together, our results suggest that a 30-fold increase in the length of the fertilization event (from 1 day to 31 days) yields a surprisingly small (only twofold) increase in algal biomass, macronutrient depletion, and FCO 2 from the atmosphere into surface waters (Table 1). This is because phytoplankton Fe demands remained near saturation, allowing Fe to mix below the thermocline before being removed by phytoplankton Initial Fe Concentration (Fe initial ) of Fertilized Patch [27] One of the default conditions in the experimental runs was that Fe was added on 1 day such that the bioavailable Fe concentration in the upper 25 m was 2 nm. Here we investigate how model results changed when Fe initial was varied from 1 nm to 5 nm. Setting Fe initial to 1 nm at Site A on 31 December resulted in an increase in the Fe concentration in Box A to 0.05 nm immediately after the Fe addition, which diminished rapidly soon thereafter (Figure 5d). As a result, the mean Fe concentration in surface waters of Box A for the experimental period was approximately 32% above that of the standard run (Table 1). [28] Phytoplankton biomass in Box A was elevated relative to that of the standard run after the 31 December addition of 1 nm Fe to Site A (Figure 5a). This biomass increase lasted for 6 weeks, similar to results obtained by Hannon et al. [2001] in simulations of the SOIREE Feenrichment experiment, until by mid-february, Chl a concentrations had fallen back to levels equivalent to those of the standard run. Over the experimental period, the addition of 1 nm Fe increased phytoplankton biomass relative to the standard run by an average of 19% within Box A (Table 1). This increase in algal biomass within Box A resulted in a0.2mm decline in PO 4 relative to the standard run, and a 3-mM drop in NO 3, with reduced levels of both nutrients lasting slightly less than 2 months (Figures 5b and 5c). Similarly, pco 2 was also reduced over the same time period, by approximately 20 matm (Figure 5e). Averaged over the experimental period, NO 3,PO 4, and pco 2 were 8of16

9 ARRIGO AND TAGLIABUE: IRON IN THE ROSS SEA, 2 Figure 5. Temporal changes in (a) chlorophyll a, (b) phosphate, (c) nitrate, (d) iron, (e) pco 2, and (f) FCO 2 (negative values denote a flux of CO 2 from the atmosphere into the Ross Sea) as a function of Fe initial for the fertilization event. Results from the standard run (STD) are also shown, as is the sea ice concentration (0 = no ice, 1.0 = 100% ice cover) during the simulations. reduced by 8%, 5%, and 3%, respectively, in Box A relative to the standard run. This depression of pco 2 resulted in a slight increase in FCO 2 into Box A (Figure 5f), which exceeded that of the standard run by 3 4% (Table 1). [29] Doubling Fe initial on 31 December from 1 nm to 2 nm had a roughly proportional effect on the seasonal increase in Chl a and the decrease in PO 4,NO 3, and pco 2, relative to the standard run (Table 1), due to a 10-day increase in the length of the bloom. Chl a averaged over the experimental period increased from 19% above the standard run, when 1 nm Fe was added, to 36% over the standard run when 2 nm Fe was added. Similarly, seasonal mean NO 3,PO 4, and pco 2 in the 2-nM fertilization run were all reduced by about twice as much relative to the standard run as they were in the 1-nM fertilization run (Table 1). Reduced levels persisted for approximately 2 3 months after Fe fertilization. The decline in pco 2 resulted in a 6% increase in FCO 2 from the atmosphere within Box A (Figure 5f) when averaged over the experimental period (again, double that of the 1-nM fertilization run). [30] Increasing Fe initial to 5 nm resulted in Fe concentrations within Box A that were more than fivefold higher than in the standard run (Table 1). Again, it must be remembered that these values likely represent upper limits, 9of16

10 ARRIGO AND TAGLIABUE: IRON IN THE ROSS SEA, 2 Figure 6. Temporal changes in (a) chlorophyll a, (b) phosphate, (c) nitrate, (d) iron, (e) pco 2, and (f) FCO 2 (negative values denote a flux of CO 2 from the atmosphere into the Ross Sea) as a function of the size of the fertilized area (1 grid cell = 646 km 2 ). Results from the standard run (STD) are also shown, as is the sea ice concentration (0 = no ice, 1.0 = 100% ice cover) during the simulations. because of the lack of redox chemistry, Fe precipitation, and particle scavenging of added Fe in the present model formulations. The Fe concentration in Box A reached a peak of nm during January, but declined rapidly during late January and February as it was consumed by phytoplankton (Figure 5d). Despite these much higher Fe concentrations in Box A in the 5-nM Fe run, phytoplankton biomass was only 60% higher than in the standard run when averaged over the experimental period. As a result of the higher initial concentration of Fe, and slow rate of phytoplankton decline, the uptake of NO 3,PO 4, and CO 2 continued to exceed that of the standard run for more than 3 months (Figures 5c, 5d, and 5e). However, seasonal minimum values of NO 3, PO 4, and CO 2 resulting from the 5-nM Fe fertilization differed little from those of the 1-nM and 2-nM Fe fertilization runs. The pco 2 within Box A during the experimental period dropped by 8 9%, resulting in an increase in FCO 2 from the atmosphere of 12% relative to the standard run (Figure 5f, Table 1) Size of Fertilized Patch [31] In each of the experimental Fe fertilization runs described above, only a single model grid cell was fertilized, amounting to a total area of 646 km 2. Here we show 10 of 16

11 ARRIGO AND TAGLIABUE: IRON IN THE ROSS SEA, 2 Figure 7. Amount of CO 2 entering the southwestern Ross Sea (73 S 78 S, 160 E 155 W) from the atmosphere, above that obtained in the standard run (DFCO 2 influx), as a function of variation in (a) the date of fertilization, (b) the duration of fertilization, (c) Fe initial, and (d) the size of the fertilized patch. how the effect of Fe fertilization changed when larger patches of ocean were fertilized, including a three grid cell by three grid cell patch (5,818 km 2 ) and a five grid cell by five grid cell patch (16,160 km 2 ). Fe concentrations within Box A increased dramatically as the fertilized patch size increased (Figure 6d), demonstrating that model results were more sensitive to this parameter than they were to any of the other experimental treatments tested. For example, when the fertilized patch size was increased to 5,818 km 2, the mean Fe concentration within Box A during the experimental period increased to more than 11-fold above that of the standard run (Table 1). [32] Algal biomass exhibited a large secondary peak following the fertilization of the 5818 km 2 patch within Box A. Chl a concentration increased from 3 mg m 3 at the time of fertilization to nearly 6 mg m 3 2 weeks later (Figure 6a). Algal biomass remained above that of the standard run for approximately 3 months, resulting in the accumulation of an average of 70% more Chl a within Box A (during the experimental period) for the simulation with a 5818 km 2 fertilized patch than the standard run (Table 1). As a result of this enhanced algal biomass, seasonal minimum PO 4 and NO 3 concentrations decreased by 0.4 mm and 7 mm, respectively, below the minimum values of the standard run, and remained lower for more than 3 months (Figures 6b and 6c). Averaged over Box A during the experimental period, concentrations of PO 4 and NO 3 for this run were 22% and 35% below those of the standard run (Table 1). Likewise, pco 2 was reduced by an additional 50 matm relative to the standard run, to a minimum of <160 matm (Figure 6e). The seasonally averaged drop in pco 2 within Box A for the 5818 km 2 fertilized patch exhibited a decrease of about 12% over that of the standard run. This drop in pco 2 resulted in a 17% increase in FCO 2 into Box A relative to the standard run (Figure 6f), when averaged over the experimental period (Table 1). [33] Increasing the fertilized patch size to 16,160 km 2 (five grid cells by five grid cells, or the entire area of Box A) also resulted in dramatic changes compared to standard run, but considering that this patch was nearly triple the area of the 5818 km 2 patch in the previous experiment (or 25 times the one grid cell simulation), the results were not as pronounced as might be expected. Concentrations of Fe in Box A increased to >1.2 nm upon fertilization, the highest of any experimental treatment, but declined rapidly as Fe was removed by algal uptake and advection/diffusion into adjacent grid cells and out of Box A (Figure 6d). Averaged over the experimental period, the Fe concentration within Box A was 16-fold higher for the run with the 16,160 km 2 fertilized patch than in the standard run (Table 1). [34] Phytoplankton dynamics in this experiment were similar to those seen in the run with the 5818 km 2 patch, but with a slightly more pronounced autumn Chl a peak (Figure 6a). Mean Chl a within Box A during the experimental period for this run were 80% above that of the standard run (they were 70% higher in the 5818 km 2 patch run, Table 1). Dynamics of NO 3 and PO 4 were similar to, and depletion was somewhat greater than, that of the smaller patch (5818 km 2 ), with a further 0.1 mm decline in PO 4 (Figure 6b) and 1 2 mm decrease in NO 3 (Figure 6c). Compared to the standard run, however, seasonally averaged NO 3 and PO 4 were reduced by only 44% and 28%, respectively, within Box A, despite the 16-fold greater Fe concentration (Table 1). [35] Declines in pco 2 were modestly larger than those for a fertilized patch size of 5818 km 2, with minimum values falling to 150 matm (Figure 6e) and a seasonally averaged value for Box A that was 16% below that of the standard run (Table 1). Prediction of the influx of atmospheric CO 2 for the five grid cells by five grid cells simulation was higher than that of any other simulation, with maximum FCO 2 values of >350 nmol C m 2 s 1 (Figure 6f) and seasonally averaged rates within Box A that were 22% higher than those of the standard run (Table 1) Air-Sea CO 2 Fluxes Over the Southwestern Ross Sea [36] Because of horizontal advection/diffusion, the effects of Fe fertilization will extend beyond the boundaries of Box A. Therefore, to calculate the impact of each of the various fertilization strategies on total FCO 2 exchange, we expanded our area of interest to include the entire southwestern Ross Sea (73 S 78 S, 160 E 155 W), rather than just Box A. In the standard run of CIAO, the surface water pco 2 deficit caused by the biological pump drove an annual CO 2 flux of 4.9 Tg C yr 1 from the atmosphere into the southwestern Ross Sea (Table 2). This air-sea CO 2 flux is equal to approximately 10% of the annual rate of NPP for this same geographic region (48.9 Tg C yr 1 ). When integrated over the southwestern Ross Sea, each of the experimental Fe-fertilization runs exhibited a higher FCO 2 from the atmosphere to the ocean than was obtained in the standard run (the difference in the space and time-integrated air-sea CO 2 flux between the standard run and the experimental 11 of 16

12 ARRIGO AND TAGLIABUE: IRON IN THE ROSS SEA, 2 Table 2. Annual Rates of NPP, FCO 2, and POC Export for the Ross Sea Study Area a Rates for Standard Run NPP 48.9 FCO Export > 100 m 22.2 Export > 200 m 15.0 Export > 400 m 5.04 a Units are Tg C yr 1. Negative FCO 2 values denote flux from the atmosphere to the ocean. The boundaries of the Ross Sea study area are 73 S 78 S, 160 E 155 W. runs will be denoted as the DFCO 2 influx, Figure 7). While none of the fertilization scenarios tested altered FCO 2 over the southwestern Ross Sea by more than 5% (Table 1), this value is not inconsequential when one considers that the southwestern Ross Sea, as defined here, has an area of nearly 700,000 km 2 and the largest fertilized patch was only 16,160 km 2 in size (2.3% of the size of the Ross Sea study area). [37] As the date of Fe fertilization was varied, the DFCO 2 influx from the atmosphere into the Ross Sea ranged from a maximum of 0.05 Tg C yr 1, when the Ross Sea was fertilized on 31 December (Figure 7a), to virtually zero, when fertilization took place either before the phytoplankton bloom had begun or after sea ice concentration had started to increase. Results from this set of simulations demonstrate that the amount of CO 2 removed from the atmosphere as a result of a particular Fe fertilization strategy can be maximized by fertilizing at certain times of the year. In the case of the southwestern Ross Sea, Fe fertilization is most efficient, and results in the greatest amount of CO 2 uptake from the atmosphere, when it takes place a few weeks after the peak of the phytoplankton bloom. [38] It should be noted that for the purpose of discussion, the fertilization efficiency is being defined here as the amount of CO 2 removed from the atmosphere (FCO 2 ) per unit of Fe added over the model domain (mol C:mol Fe). We used FCO 2 rather than POC export below the thermocline as an index of carbon sequestration and fertilization efficiency for two reasons. First, estimates of POC export will vary widely depending on the depth over which export is evaluated in both models (Table 2) and in the field due to remineralization of POC as it sinks below the surface, making it difficult to define an absolute POC flux. In the case of the southwestern Ross Sea, annual FCO 2 is approximately equal to the POC flux below 400 m (Table 2), but this will not be true everywhere. Calculations of the CO 2 exchange across the air-sea interface are not subject to these depth scaling problems and should provide a more reliable measure of atmospheric CO 2 removal. More importantly, because the southwestern Ross Sea, unlike most oceanographic regions, undergoes intense winter convective mixing, a large fraction of the organic matter exported below the thermocline is remineralized to CO 2 and brought back to the surface. As a result, using POC export below 100 m, as has been done in calculations of fertilization efficiency from field experiments, yields estimates of carbon sequestration and fertilization efficiency that are approximately fivefold greater than those calculated from annual FCO 2 (Table 2), and that are unrealistically high (>40% of NPP). [39] Fertilization efficiency varied with date of Fe fertilization (Figure 8a) in a manner identical to that exhibited by the DFCO 2 influx. This similarity is due to the fact that when only the date of fertilization is varied between simulations, differences in fertilization efficiency are driven solely by changes in the FCO 2 from the atmosphere (DFCO 2 influx) because the amount of Fe added is the same. Other fertilization strategies (see below) involve the addition of varying amounts of Fe, and as a result, changes in fertilization efficiency are more complex. The maximum fertilization efficiency was realized when fertilization took place 2 3 weeks after the peak of the bloom, reaching 161,000 mol C:mol Fe. [40] Varying the duration of fertilization from 1 day to 31 days resulted in a roughly linear increase in the DFCO 2 influx to Ross Sea surface waters (Figure 7b), from 0.05 to 0.12 Tg C yr 1. At first glance, it might appear from these results that the effectiveness of the Ross Sea as a sink of atmospheric CO 2 could be increased by simply fertilizing for longer periods of time. While this may be true in principle, in practice such a strategy would produce diminishing returns. As the time period of fertilization is extended, an ever increasing Fe supply is required to maintain Fe concentrations at high levels within the UML. CIAO results show that the additional Fe needed to increase the length of the phytoplankton bloom would not be balanced by a proportional uptake of atmospheric CO 2 by Ross Sea surface waters. While fertilization efficiency was relatively high for fertilization scenarios of short duration, fertilization efficiency dropped as the fertilization duration began to Figure 8. Fertilization efficiency, defined here as the annual amount of CO 2 taken up from the atmosphere per unit Fe added to the water column (mol CO 2 influx/mol Fe added), as a function of variation in (a) the date of fertilization, (b) duration of fertilization, (c) the Fe concentration after fertilization, and (d) the size of the fertilized patch. The thick line in Figure 8b shows the increased fertilization efficiency when Fe is added to the surface ocean every other day instead of every day. 12 of 16

13 ARRIGO AND TAGLIABUE: IRON IN THE ROSS SEA, 2 Table 3. Fertilization Parameters of In Situ Fe-Enrichment Experiments a Study Date Location Patch Size, km 2 Fe initial, nm Efficiency, mol C:mol Fe Source IronEx-I October 1993 Galapagos 64 4 ND Coale et al. [1998] IronEx-II May S, 104 W 72 2 ND Coale et al. [1996] SOIREE February S, 140 E 50 3 ND Boyd et al. [2000] EisenEx Nov S, 21 E 490 ND ND Gervais et al. [2002] SEEDS July N, 165 E ND Tsuda et al. [2003] SOFeX-N Jan. Feb S, 172 W , ,000 Bishop et al. [2004] SOFeX-S Jan. Feb S, 172 W Buesseler et al. [2004] SERIES July Aug N, 145 W b Boyd et al. [2004] Modeled January 76.5 S, 177 E ,000 this study (all runs) Modeled January 76.5 S, 177 E , ,000 c this study (SOFeX runs) a ND means no data were available for that parameter. b Range is based on export from the upper 125 m and 50 m, respectively. c Range is from simulations with conditions similar to those of SOFeX-N. increase (Figure 8b). This is easy to understand upon closer inspection of Figure 7b. These results show that a 1-day fertilization event resulted in a DFCO 2 influx into the Ross Sea of 0.05 Tg C above that of the standard run. A 31-day fertilization event resulted in a DFCO 2 influx of only 0.12 Tg C, an increase of less than threefold, despite a large increase in the amount of Fe added. Hence, although there are no nonlinear feedbacks of sufficient importance to reverse the sign of the DFCO 2 influx trend, and thus, longer fertilization events always remove more CO 2 from the atmosphere in an absolute sense, they do so at an increasingly high Fe cost and a much reduced fertilization efficiency. [41] Similar results were obtained when evaluating the effect of changing either Fe initial (Figure 7c) or the size of the fertilized patch (Figure 7d). In both cases, the DFCO 2 influx rose as fertilization intensity increased, although changing patch size exhibited a decidedly more nonlinear response in the DFCO 2 influx. However, as was the case for the duration of fertilization (described above), the increases in the DFCO 2 influx during both of these sets of simulations were not proportional to the increase in the amount of Fe added, resulting in a reduction in fertilization efficiency as Fe fertilization was intensified. In the case of an increase in Fe initial, fertilization efficiency decreased from 215,000 to 118,000 mol CO 2 :mol Fe as the desired Fe concentration was raised from 1 nm to 5 nm (Figure 8c). More dramatically, as the area of the fertilized patch was increased from 646 to 16,160 km 2, fertilization efficiency dropped from 161,000 to 31,000 mol CO 2 :mol Fe (Figure 8d) CIAO Simulations Versus In Situ Fe-Enrichment Experiments [42] Unfortunately, there are few measurements of fertilization efficiency available from the in situ Fe enrichment experiments with which to compare our results. This is because these in situ experiments were generally of too short a duration to measure enhanced FCO 2 or export of carbon (Table 3). Exceptions are the recent Southern Ocean Iron Experiment (SOFeX) in the northern Ross Sea sector of the Southern Ocean and the Subarctic Ecosystem Response to Iron Enrichment Study (SERIES) conducted in the North Pacific Ocean. In these studies, fertilization efficiency was calculated differently than has been done here. Rather than using FCO 2, which is difficult to measure over the appropriate timescales, they defined fertilization efficiency as the ratio of the amount of Fe added to the increase in POC export (beyond that measured in an unfertilized control patch). Fertilization efficiency during SERIES was calculated to be 1205 mol C:mol Fe based on collection of POC by sediment traps placed at a depth of 50 m [Boyd et al., 2004], very low compared to results of the present study. This value decreased even further, to 476 mol C:mol Fe, when POC export below the permanent thermocline ( m) was evaluated for the 125 m sediment traps (Table 3), due to the high rates of remineralization in surface waters during SERIES. [43] As was observed during SERIES, export efficiency within the southern fertilized patch of SOFeX (SOFeX-S, Table 3) was also relatively low (3300 mol C:mol Fe), as determined from 234 Th deficits [Buesseler et al., 2004]. This was not the case in the northern patch of SOFeX (SOFeX-N), however, where export efficiencies of 10, ,000 were reported, based on measurements made by the Carbon Explorer POC sensor [Bishop et al., 2004]. As argued by Boyd [2004], the most likely reason for the large discrepancy between the estimates of fertilization efficiency for SOFeX-N and SOFeX-S (beyond the obvious differences in measurement technique) was that the SOFeX-S was only monitored for 30 days while SOFeX-N was monitored for 50 days [Bishop et al., 2004]. Given the high Fv/Fm values (an index of photosynthetic competence) measured for phytoplankton at the end of the SOFeX-S [Coale et al., 2004], it is likely that the bloom had not yet terminated before the study ended [Buesseler et al., 2004]. Hence, while large export events were observed during the 50-day study of the SOFeX-N, rates of POC export had not yet peaked by the end of the 30-day SOFeX-S study. [44] Because of its more complete temporal coverage and similar geographic location (within the Southern Ocean, unlike SERIES), the most reasonable in situ estimate of fertilization efficiency with which to compare the results for the Ross Sea from CIAO is from SOFeX-N. Moreover, this particular experiment was conducted in low-silicate waters north of the Polar Front, and the Fe-stimulated bloom was dominated by nonsiliceous flagellates (although it is not clear if export at SOFeX-N was dominated by flagellates), as was our simulated Fe-stimulated bloom in the southwestern Ross Sea (dominated by P. antarctica). Estimates of export efficiency for SOFeX-N fall within the range of values obtained for the southwestern Ross Sea by CIAO, which varied from 3000 to 215,000 mol C:mol Fe (Table 3). This range from the southwestern Ross Sea can be narrowed, however, by including only those simulations whose 13 of 16