A three-dimensional, multinutrient, and size-structured ecosystem model for the North Atlantic

Authors


Abstract

[1] We incorporate multinutrient and size-structured ecosystem dynamics into a three-dimensional ocean general circulation model for the North Atlantic. The model reproduces the magnitude and general spatial and temporal patterns in nutrients, chlorophyll and primary production seen in in situ (BATS, NABE, and OWSI) and satellite (SeaWiFS) data, showing substantial improvements over prior basin-scale simulations. Model skill is evaluated quantitatively against SeaWiFS data using a Taylor diagram approach. Model-data correlation R for the overall surface chlorophyll time-space distribution is ∼0.6, with comparable model and observed total variability. The agreement relative to satellite-based primary production is somewhat weaker (0.2 < R < 0.5). The simulations capture observed ecological characteristics, e.g., the dominance of picoplankton and episodic diatom blooms in the subtropics, nutrient-controlled plankton succession at higher latitudes, and associated seasonal/depth changes in new and regenerated production and particle export. In a sensitivity experiment that mimics behavior of simpler single-species models, removal of diatom silica limitation leads to major shifts in community structure and export and larger model-data errors similar to previous model studies. Model results also suggest that episodic diatom blooms at BATS may be related to interannual variations in the southward transport of nutrients, mainly SiO3, and plankton cells.

1. Introduction

[2] The need to quantify and project the ocean response to and feedbacks on anthropogenic perturbations has sparked a renewed interest in marine biogeochemical models the last decade [e.g., Fasham et al., 1990; Sarmiento et al., 1993; Doney et al., 1996; Doney, 1999; Dutkiewicz et al., 2001]. Most model implementations are based on comparatively simple trophic interactions with a few basic compartments: nutrients, phytoplankton, zooplankton and detritus (so called NPZD class of models). Nitrogen [Fasham et al., 1990] or phosphorus [Doveri et al., 1993] are used as natural “currencies”, under the assumption that they are the limiting nutrients in marine systems. This pragmatic approach to the functional categorization of organisms largely ignores the taxonomic and biogeochemical diversity and food web complexities characteristic to such systems [Pomeroy, 1974]. Nevertheless, this class of bulk models provides a useful general framework to describe and analyze ecosystem functioning and facilitates the evaluation of such models against in situ chlorophyll and primary productivity measurements as well as satellite data.

[3] Recently, this traditional NPZD model framework has been expanded to include multielement nutrient limitation, distinct phytoplankton functional groups, and more realistic biogeochemistry. One-dimensional simulations have been used to investigate nutrient and carbon fluxes at individual sites [Pondaven et al., 2000; Hood et al., 2001] regionally [Loukos et al., 1997; Denman and Peña, 1999; Leonard et al., 1999] and globally [Moore et al., 2002a, 2002b]. Multinutrient, size-structured ecosystem models based on fully three-dimensional time-dependent ocean general circulation models (OGCM) also have begun to appear in the literature [Gregg, 2002; Christian et al., 2002a, 2002b; Aumont et al., 2003]. In this study, we report results from a relatively complex ecosystem simulation in an OGCM for the North Atlantic.

[4] The ecosystem component includes distinct phytoplankton functional groups, as well as size structure, and a mechanistic phytoplankton growth and photoadaptation model [Geider et al., 1998] that accounts for multielement nutrient limitation (nitrogen and silica). Iron is not included in the model as it is typically not considered an important limiting nutrient in the North Atlantic [Martin et al., 1993; Fung et al., 2000] due to relatively strong mineral dust inputs from continents [Tegen and Fung, 1995; Mahowald et al., 1999]. The potentially important effects of nitrogen fixation, calcification and phosphate limitation are also not considered in the present model and are the subject of forthcoming work.

[5] Model results are compared with in situ data (BATS, Ocean Weather Station India and NABE) and SeaWiFS imagery. After quantitatively evaluating model skill against observations, we use our model as a heuristic tool to explore the dynamics of the pelagic ecosystem in the North Atlantic. In particular, we investigate seasonal and geographical patterns in nutrient limitation and phytoplankton community composition and the resulting impacts on new, regenerated and total primary production and the export of organic carbon. We also compare our model results with those from two previous basin-scale model studies of the North Atlantic, a seven-compartment biogeochemical model [Fasham et al., 1990] that includes ammonium, nitrate and suspended and sinking detritus, into a coarse-resolution (2°) OGCM [Sarmiento et al., 1993; Fasham et al., 1993] and a very simple four-compartment (NPZD) model embedded in a eddy-permitting (1/3°) OGCM [Oschlies et al., 2000; Oschlies, 2001, 2002]. The present work differs in that it combines a fairly complex and realistic ecosystem model with an intermediate resolution (0.8°) OGCM. Model-model comparison is complicated by the fact that each model uses a different OGCM and biogeochemical formulation. Nevertheless, by comparing results from such diverse models one can identify common strengths and deficiencies and gain insight into the basic features that are necessary to simulate ocean biology realistically.

2. Methods

2.1. Physical Model

[6] The physical component is the Los Alamos Parallel Ocean Program (POP) [Smith et al., 1992], a three-dimensional, primitive equation numerical ocean model that uses a level-coordinate system and has an implicit free surface treatment of the barotropic equations [Dukowicz and Smith, 1994]. The model domain comprises the Atlantic Ocean from 19.6°S to 72.5°N and 98°W to 16.8°E, including the Gulf of Mexico and the western part of the Mediterranean Sea. The horizontal mesh is a Mercator grid with resolution of Δλ = 0.8° and Δϕ = 0.8°cosϕ, where λ and ϕ are longitude and latitude, respectively. Thus horizontal resolution varies from 88.9 km at the equator to 26.7 km at the northern boundary. The vertical grid has 40 nonuniform vertical levels that vary in thickness from approximately 10 m at the surface to 250 m at depth. The topography is derived from the ETOPO5 database and interpolated to the 0.8° grid. The depth at each horizontal grid point is set equal to that of the nearest vertical level in the model.

[7] A Laplacian operator is used for horizontal mixing of momentum. The horizontal mixing of tracers is done using the [Gent and McWilliams, 1990] parameterization, which forces the mixing to take place along isopycnal surfaces, thus avoiding spurious horizontal diapycnal mixing across sloping isopycnals. The viscosity and diffusivity coefficients vary spatially with the cube of the horizontal grid spacing with values of 5 × 107 and 6 × 106 cm2 s−1 at the equator, respectively. The vertical viscosities and diffusivities are computed using the KPP mixing parameterization [Large et al., 1994] with background values of 1.0 and 0.1 cm2 s−1, respectively. The KPP model includes parameterizations for mixing due to internal wave breaking, shear instability and double diffusion [Large et al., 1994].

[8] The physical model is forced with daily wind stresses, derived from ECMWF TOGA Global Surface Analysis for the years 1985–1992, linearly interpolated to each model time step. The surface heat flux is computed using the ECMWF heat flux analysis of [Barnier et al., 1995], arranged in restoring form, using monthly mean climatologies of sea surface temperature, an ice mask and net downward short-wave radiation. An effective net freshwater flux is simulated by restoring the surface salinity to the monthly [Levitus et al., 1994] climatology using a 1-month restoring timescale. At the northern, southern, and eastern boundaries there is a 3° wide buffer zone where temperature and salinity are restored full depth to seasonal NODC climatology [Levitus et al., 1994; Levitus and Boyer, 1994] with a restoring constant 1/τ that varies linearly from 1/15 d−1 at the boundary to 0 d−1 at the interior end of the buffer zone.

2.2. Ecosystem Model

[9] The ecosystem component has eight basic compartments: two phytoplankton species (picophytoplankton P1 and diatoms P2), two classes of detritus (small/suspended D1 and large/sinking D2), zooplankton (Z), and the nutrients nitrate (NO3), ammonium (NH4+) and silicate (SiO3). The biotic and detrital compartments contain multiple elemental pools to track the flow of nitrogen, carbon and silica through the ecosystem (total of 17 ecological state variables). The structural relationship among the different compartments in the model is outlined in Figure 1. The general form of the time rate of change equations in terms of nitrogen are:

equation image
equation image
equation image
equation image
equation image
equation image
equation image
equation image

The phytoplankton, zooplankton and detritus source and sink terms are converted to carbon fluxes by multiplication with the respective carbon to nitrogen ratio ([C:N]) computed in the model. Parameter values and definitions are given in Table 1.

Figure 1.

General structure of ecosystem model. The model's 17 state variables comprise the three nutrient compartments and the elemental pools in each main biological compartment.

Table 1. Parameter Values and Definitions for the Ecosystem Model
ParameterValueUnitsDefinitionSourcea
ϕ0.45 PAR fraction of total irradiance 
α0.25mMol C (mg Chl)−1 d−1 m2 W−1initial slope of the PI curve1
κw0.04m−1light attenuation coefficient 
κchl0.035m2 (mg Chl)−1phytoplankton self-shading coefficient2
Tref30°Creference temperature 
PrefC3.0d−1phytoplankton maximum C-specific growth rate1
θN2.5mg Chl (mMol N)−1phytoplankton maximum Chl:N ratio1
QminN0.034mol N (mol C)−1phytoplankton minimum N:C ratio1
QmaxN0.170mol N (mol C)−1phytoplankton maximum N:C ratio1
QminSi0.041mol Si (mol C)−1diatom minimum Si:C ratio3
QmaxSi0.204mol Si (mol C)−1diatom maximum Si:C ratio3
VC refN0.51mMol N (mMol C)−1 d−1C-specific N uptake rate at Tref1
VC refSiPrefC × QmaxSimMol Si (mMol C)−1 d−1C-specific Si uptake rate at Tref 
Rref0.0d−1respiration/degradation at Tref1
ζ2.33mMolC mMolN−1cost of of biosynthesis1
image0.60mMol N m−3NO3 half saturation constant for picophytoplankton 
image0.007mMol N m−3NH4+ half saturation constant for picophytoplankton 
image1.5mMol N m−3NO3 half saturation constant for diatoms 
image0.07mMol N m−3NH4+ half saturation constant for diatoms 
image1.10mMol N m−3silicate half saturation constant for diatoms4
e10.1d−1picophytoplankton respiration rate5
e20.1d−1diatom respiration rate5
s10.1[mMol N m−3 d]−1picophytoplankton aggregation rate5
s20.1[mMol N m−3 d]−1diatom aggregation rate5
g13.25d−1zooplankton maximum growth rate when grazing picophytoplankton6
g22.75d−1zooplankton maximum growth rate when grazing diatoms6
KZ0.75mMol N m−3half saturation constant for zooplankton grazing 
c0.1d−1zooplankton respiration rate7
d0.25[mMol N m−3 d]−1zooplankton mortality rate 
a0.7 zooplankton assimilation efficiency8
λ0.5 zooplankton egestion allocation factor 
r10.2d−1remineralization rate for small detritus9
r20.2d−1remineralization rate for large detritus9
n0.1d−1nitrification rate6
w25m d−1large detritus sinking rate10

[10] Phytoplankton growth and photoadaptation Ui(Ipar, T, NH4, NO3) are computed following a modified version of the growth model of Geider et al. [1998] (hereinafter referred to as GD98). Environmental factors irradiance (Ipar), nutrients and temperature determine the instantaneous rates of light utilization, carbon and nutrient assimilation and chlorophyll synthesis. These instantaneous rates are then modulated by the effects of past environmental conditions through time-evolving intracellular quotas [Chl:C] and [N:C]. The GD98 model was modified for the present study to allow for two nitrogen sources (NO3 and NH4+) and for growth limitation by the [Si:C] quota for diatoms, in a similar manner as for nitrogen. Maximum and minimum cell quotas for each nutrient are input parameters (Table 1). Phytoplankton [N:C] ratios are taken from GD98, and the [Si:C] diatom cell quota is found by multiplying the GD98 [N:C] values by an average [Si:N] ratio for larger diatoms (1.2) [Brzezinski, 1985]. We use a maximum [Chl:N] ratio of 2.5 mg Chl (mMol N)−3 that is well within the range reported by Geider et al. [1997]. The generic form of the photosynthetic, nutrient uptake and photoadaptation rates computed in Ui for one limiting nutrient (N) are:

equation image

where the maximum photosynthetic rate PmaxC and carbon-specific nutrient uptake rate VCN are given by:

equation image

Tf is the temperature response function, Q is the phytoplankton [N:C] ratio, ζ represents the cost of biosynthesis, θC is the phytoplankton [Chl:C] ratio and ρChl is a dimensionless chlorophyll synthesis regulation term given by:

equation image

[11] The light intensity available for photosynthesis (Ipar) at depth (z) is given by:

equation image

where I0 is the total irradiance at the surface, ϕ is the fraction of total irradiance that is available for photosynthesis (PAR), κw and κchl are the light attenuation coefficients due to water and phytoplankton self-shading, respectively, and Chl1 and Chl2 are the small phytoplankton and diatom chlorophyll concentrations. The diurnal cycle of light intensity is not considered for consistency with the forcing in the physical model, and I0 is set to the temporally interpolated value of incoming short-wave radiation (W m−2).

[12] Total nitrogen (NO3 + NH4+) uptake by the phytoplankton and the inhibiting effect of NH4+ concentration on NO3 uptake is modeled using the substitutive equation of O'Neil et al. [1989]. The half saturation constants for NH4+ and NO3 are set to relatively low and high values, respectively (Table 1), to give the phytoplankton a reasonably strong preference for NH4+. In a series of test runs, these values provided the best fit of the model to observations. Silica uptake by the diatoms is modeled using standard Michaelis-Menten kinetics. Half saturation for SiO3 varies considerably among regions (0.5–4.6 mMol m−3) [Brzezinski and Nelson, 1989; Nelson and Tréguer, 1992; Nelson et al., 2001] and a medium value of 1.2 mMol m−3 is chosen.

[13] The single model zooplankton compartment is assumed to represent a diverse community of grazers, using a Holling type III predation functional response with different maximum growth/grazing rates for the zooplankton depending on the food source:

equation image

The grazing parameterization increases the overall stability of the system and simulates reasonably well the effects of size-dependent grazing on the phytoplankton community [Armstrong, 1999; Lima et al., 2002]. A relatively high zooplankton quadratic mortality rate is used to increase model stability [Steele and Henderson, 1992; Lima et al., 2002].

[14] The partitioning of zooplankton losses between the suspended and sinking detrital compartments depends on the type of food source and is parameterized in terms of an excretion allocation factor (δ):

equation image

where G(Pi) represents the Holling type III predation functional response for the phytoplankton species i (equation (13)). δ approaches 0.7 for P2P1. Thus, as the relative contribution of diatoms (P2) to total grazing increases, a proportionally larger fraction of zooplankton excretion and mortality is directed to the large sinking detritus compartment.

[15] The NO3 compartment receives input from nitrification at light levels corresponding to those at the bottom of the euphotic zone (1% of surface Ipar) and below. The temperature dependency of zooplankton growth rates and detrital remineralization rates is computed by multiplying the respective maximum instantaneous rates (Table 1) by the same temperature function (Arrhenius relation) used in the phytoplankton growth model (GD98).

2.3. Coupled Model Initialization and Integration Procedure

[16] All biogeochemical tracers are coupled into the model through the advection-diffusion equation:

equation image

where X represents each biogeochemical tracer and equation image the respective source-sink term in the biogeochemical model (equations (1)()()()()()()(8)). A third-order upwind advection scheme is used for all tracers to avoid negative tracer values near sharp gradients with minimal implicit diffusion [Hecht et al., 1998].

[17] The coupled model is initialized with the ocean at rest, temperature and salinity set equal to the June NODC climatology [Levitus et al., 1994; Levitus and Boyer, 1994], and nitrate and dissolved silica from the annual mean NODC climatology [Conkright et al., 1994]. All remaining biogeochemical compartments are initialized with a uniformly low value (0.1 mMol N m−3 or equivalent). The physical model is spun up for 2 years (1 June 1985 through 1 June 1987), driven by the ECMWF winds and the boundary forcing described above. The ecosystem model is then “switched on”, and the integration continues for another 10.5 years (through 1 December 1997). Because of the relatively short integration time and our focus on the upper water column, the deep nitrate and silica fields are restored to their respective NODC annual means, with a restoring constant 1/τ that varies linearly with depth from 0 d−1 at 400 m to 1/30 d−1 at 1100 m and remains constant for depths below that. The ecosystem model spins up rapidly and reaches an approximate repeating annual cycle after 2 years (Figure 2). The annual cycle does not repeat itself exactly because the model is forced with nonclimatological daily wind stresses.

Figure 2.

Time series of monthly means for the euphotic zone (107 m) average of (a) nitrate, (b) vertical particle flux at 107 m, (c) phytoplankton biomass, (d) total primary production, (e) new production, and (f) regenerated production for the picophytoplankton (P1), diatoms (P2), and total phytoplankton (P1 + P2).

2.4. Field and Satellite Data Sets

[18] Monthly climatological means of chlorophyll and primary productivity from the last 3 years of model integration are compared with estimates derived from SeaWiFS imagery. Surface chlorophyll data from monthly level 3 standard images covering the period from September 1997 to August 2003 were extracted for the model domain area and averaged and remapped to the model grid. Monthly primary productivity is estimated from SeaWiFS chlorophyll data using the VGPM model of Behrenfeld and Falkowski [1997]. Model skill is evaluated quantitatively against satellite estimates of chlorophyll and primary productivity using a Taylor diagram [Taylor, 2001], a two-dimensional plot comparing the total variance and space/time correlation between a test (model) and a reference (satellite) field. Four types of comparisons are made using different subsets of model and and satellite fields to explore how well the model does in replicating different aspects of the observed variability. The model and observed standard deviation and correlation are computed, respectively: for the sum of all 12 months and all model grid points (total space-time); for each month individually (monthly-spatial); for monthly anomalies from the local annual mean (monthly-time); and for the annual mean (spatial-annual). For a more detailed description of the Taylor diagram and the statistics and computations involved see the Appendix section. Oceanic bio-optical variability follows approximately a lognormal distribution [Campbell, 1995] so the natural log transformation is applied to the chlorophyll and primary productivity fields prior to the analysis.

[19] Observations from two time series stations, the Bermuda Atlantic Time series Study (BATS, 31°49′N, 64°09′W) [Michaels and Knap, 1996; Steinberg et al., 2001] and Ocean Weather Station India (OWSI, 59°N, 18°W) [Williams and Robinson, 1973] and from one process oriented study, the North Atlantic Bloom Experiment (NABE, 47°N, 20°W) [Weeks et al., 1993] are compared with model output from the corresponding grid points in the model domain. Because of the relative scarcity of data at this site, the NABE data include observations from the US JGOFS cruises, the Plankton Biogeochemical Ocean Flux Study (BOFS) [Lowry et al., 1994] and the Plankton Reactivity in the Marine Environment (PRIME) [Hadziabdic and Cramer, 1999]. For each site, all available observations on nutrients, chlorophyll and primary production were grouped by month and depth intervals to produce monthly average profiles.

[20] Model results on the relative abundance of the different phytoplankton groups are presented in chlorophyll units to facilitate comparison with available observations on community structure, which are usually reported in terms of pigment concentrations. Because of the scarcity of depth-resolving records of zooplankton abundance and the highly parameterized nature of the model's zooplankton compartment, we do not attempt to make direct comparisons of modeled zooplankton abundance and distribution with observations.

3. Basin-Scale Surface Chlorophyll and Primary Production

[21] The model reproduces the observed large-scale geographical and seasonal patterns of surface chlorophyll distribution and primary productivity derived from SeaWiFS imagery (Figures 3 and 4). Agreement in the subtropics is particularly good in the fall and winter. In the spring and summer, however, simulated subtropical chlorophyll levels are noticeably lower than observed, and the oligotrophic region is expanded over a larger area. Still, minimum modeled chlorophyll concentrations in the oligotrophic gyre (0.019 mg Chl m−3) during spring and summer are relatively close to the lowest satellite values (0.037 mg Chl m−3). The lowest simulated primary production values in the subtropical gyre also occur during spring and summer and are ∼50% lower than satellite derived estimates.

Figure 3.

Seasonal climatologies of surface chlorophyll concentrations (mg Chl m−3) from the last 3 years of model integration and SeaWiFS imagery (October 1998 to September 1999). White symbols represent the location of the BATS (squares), NABE (triangles), and OWSI (diamonds) sites.

Figure 4.

Seasonal climatologies of primary productivity estimates (g C m−2 season−1) from the last 3 years of model integration and from SeaWiFS chlorophyll data using the VGPM model of Behrenfeld and Falkowski [1997]. White symbols represent the location of the BATS (squares), NABE (triangles), and OWSI (diamonds) sites.

[22] The circulation model overestimates the winter mixed layer depth along the northwestern edge of the subtropical gyre by more than 150 m. The deep convection entrains an excess of nutrients that leads to an overestimation of surface chlorophyll and primary production along the northern flank of the subtropical gyre during winter and spring by ∼60% (Figures 3 and 4). High chlorophyll values are also advected northward and eastward by the Gulf Stream and North Atlantic Current, causing positive anomalies at high latitudes in winter and fall (Figure 3). In their eddy-permitting simulations, Oschlies et al. [2000] also report a similar problem that persisted despite changes in the parameterization of vertical turbulent mixing [Oschlies and Garçon, 1999].

[23] Because of its spatial resolution, the model does not capture coastal processes very well. Most of the high chlorophyll and primary production values observed on the northeast coast of South and North America and the upwelling regions off the coast of Africa have counterparts in the model, although weaker ones. The model also overestimates chlorophyll concentrations and primary production along the equatorial upwelling region, a result of high the equatorial upwelling due to the closed southern boundary [Smith et al., 2000] and the relatively high temperature-dependent phytoplankton growth rates in this region.

[24] Despite deficiencies, model-data correlation for the overall time-space distribution of satellite chlorophyll fields is ∼0.6 and the magnitude of the model total time-space variability is comparable to the observed value (point 3 in Figure 5a; see also Table 2). The correlation for the annual mean spatial pattern (spatial-annual) is somewhat better while the correlation for the seasonal cycle (monthly-time) is somewhat worse (Figure 5a). The largest errors occur in late winter and early spring (February and March) and are directly related to the excess convections along the northern edge of the subtropical gyre described above (Figure 3). This difference is also reflected in the monthly-time statistics which show a higher amplitude of variation in the simulated seasonal cycle (point 1 in Figure 5a).

Figure 5.

Taylor diagram comparing monthly climatological fields of log transformed surface chlorophyll estimates (mg Chl m−3) from the model and satellite and monthly primary productivity estimates (g C m−2 month−1) from the model and satellite data using the VGPM model of Behrenfeld and Falkowski [1997]. The radial distances from the origin are proportional to the ratio of the model and satellite standard deviations and the azimuthal positions correspond to the correlation between the model and satellite fields. The distances from the model points to the satellite point are proportional to the normalized RMS difference between the fields (Table 2). Black points represent the main run, and red points represent the “no silica limitation” run (NSLIM). (a) Chlorophyll monthly-time (1), spatial-annual (2), and total time-space (3) statistics. (b) Chlorophyll monthly-spatial statistics. (c) Primary productivity monthly-time (1), spatial-annual (2), and total time-space (3). (d) Primary productivity monthly-spatial statistics. Numbers in Figure 5b and 5d correspond to the months of the year.

Table 2. Normalized RMS Difference of Log Transformed Surface Chlorophyll and Monthly Primary Productivity Estimates From the Model and SeaWiFS Imagery for the Three Different Cases Presented in Figures 5a and 5c
 Main RunNSLIM Run
Chlorophyll, mg Chl m−3
  Monthly-time case1.3541.425
  Spatial-annual case0.8150.861
  Total time-space case0.9120.967
Primary productivity, g C m−2 month−1
  Monthly-time case1.2231.258
  Spatial-annual case1.1991.223
  Total time-space case1.2001.232

[25] Model performance in simulating monthly productivity (Figures 5c and 5d) is not as good as chlorophyll (Figures 5a and 5b). Modeled primary productivity is lower than the VGPM satellite estimated values over the most stratified and oligotrophic portion of the subtropical gyre (Figure 4) perhaps due to missing mesoscale nutrient inputs [McGillicuddy et al., 2003]. The model also overestimates primary production along the northern edge of the subtropical gyre during winter and early spring and in the equatorial upwelling region (Figure 4). Thus the model generally overestimates the spatial and temporal amplitude of variation of the satellite estimates by 20% and 70%, respectively, with weaker correlations (0.2 < R < 0.5) (Figures 5c and 5d). Primary production algorithms vary widely in performance, and the best performing algorithms agree with in situ estimates within a factor of two [Campbell et al., 2002]. Thus metrics of model skill will depend on the choice of primary production algorithm for the satellite reference field and may be lower as a result of deficiencies in the primary production algorithm used.

4. Annual Cycle at BATS

[26] There is generally good agreement between the observed and modeled annual cycles of surface chlorophyll (Figure 6a) and full depth NO3, chlorophyll and primary production fields at BATS (Figure 7). However, the onset and peak of the spring bloom in the model occurs a month later than observed, and the model overestimates the magnitude of chlorophyll concentrations and primary production during the spring bloom by ∼60% and ∼25%, respectively (Figures 6a and 7c–7f). These discrepancies result because the modeled winter mixing is 80% deeper, has a shorter duration and is delayed relative to the observed climatological mean (Figure 7b). The intensity and timing of winter mixing at BATS show a fair amount of interannual variability [Steinberg et al., 2001]. As a result, the climatological mean is is significantly shallower than the maximum in individual years, typically ∼200–300 m [Michaels and Knap, 1996] similar to the model average. For the remainder of the seasonal cycle, however, modeled surface chlorophyll concentrations are higher (∼15%) but closer to the range of observed values. Comparison with results from sensitivity runs show that the overestimation of chlorophyll concentrations in the DCML is also in part a result of excessive photoadaptation by the phytoplankton in the model. The suspended detritus → ammonium → picophytoplankton → micrograzer pathway with tight recycling under oligotrophic conditions allows the model to maintain primary production rates of ∼0.4 mMol C m−3 d−1 at very low ambient NO3 concentrations (0.01 mMol N m−3), which has proven to be very difficult with simpler ecosystem models [e.g., Fasham et al., 1993; Oschlies et al., 2000].

Figure 6.

Time series of monthly climatological means of surface chlorophyll concentrations (mg Chl m−3) from model, in situ observations, and SeaWiFS imagery (1998–1999) at the (a) BATS, (b) NABE, and (c) OWSI sites.

Figure 7.

Time series of monthly climatological means of observed and modeled NO3 (mMol m−3), chlorophyll concentrations (mg Chl m−3), and primary production (mMol C m−3 d−1) at the BATS site. White line represents mixed layer depth.

4.1. Community Dynamics

[27] Although both picophytoplankton and diatoms grow in response to the increased nitrate supply in late winter and early spring, the picophytoplankton dominate the bloom reaching maximum concentrations in March (Figure 8). Diatoms are significantly less abundant and reach maximum concentrations at depth (60–70 m) after the main bloom, in late spring and early summer. After the spring bloom, production in the upper part of the euphotic zone is mostly regenerated, while at depth, near the nutricline, nutrient supply (NO3, SiO3) is sufficient to maintain small but significant levels of new production in the DCML. These patterns are consistent with available observations at BATS [Brzezinski and Nelson, 1995; Michaels and Knap, 1996; Nelson and Brzezinski, 1997; Steinberg et al., 2001; DuRand et al., 2001].

Figure 8.

Time series of monthly climatological means of picophytoplankton (P1) and diatoms (P2) chlorophyll concentrations (mg m−3) and f ratios as function of depth from the model at the locations corresponding to the (a–c) BATS, (d–f) NABE, and (g–i) OWSI sites. Note different color scale for BATS.

4.2. Nutrient Limitation Patterns

[28] Simulated phytoplankton growth is limited by the nutrient with the lowest cell quota relative to its maximum. At BATS, nutrient entrainment in late winter allows both phytoplankton size classes to increase their nitrogen cell quotas (Figures 9a and 9b), with diatoms achieving near maximum silica cell quota. As spring progresses, both phytoplankton species become nitrogen limited, with diatoms more strongly limited because of their higher half saturation constant. By late spring and early summer, diatoms become severely silica limited and remain so throughout the rest of the seasonal cycle (Figure 9c), consistent with observations in the Bermuda region [Brzezinski and Nelson, 1995, 1996; Nelson and Brzezinski, 1997]. As found in field data [Lessard and Murrell, 1998], picophytoplankton abundance is controlled mainly by grazing.

Figure 9.

Time series of monthly climatological means of the picophytoplankton (P1) and diatom (P2) [N:C] and [Si:C] cell quotas relative to the respective maximum cell quotas as function of depth from the model at the locations corresponding to the (a–c) BATS, (d) NABE, and (e) OWSI sites. [N:C] and [Si:C] values below 0.2 near the surface in Figures 9b and 9c indicate P2 = 0. White contour lines indicate picophytoplankton and diatom chlorophyll concentrations (mg m−3) and correspond to the same contours in Figure 8. (f) Geographical nutrient limitation patterns for the diatoms averaged over the the euphotic zone (107 m). Rectangles indicate silica limitation, inclined lines indicate nitrogen limitation, and the cross-hatch pattern shows where the limiting nutrient varies seasonally. Symbols represent the location of the BATS (square), NABE (triangle), and OWSI (diamond) sites.

4.3. Production and Export

[29] Model new and regenerated production peak at the same time (March) at BATS. New production drops to very low levels after April while regenerated production remains relatively high throughout the rest of the year (Figure 10b). The peak in the vertical particle flux corresponds to the maximum in the combination of diatom mortality and zooplankton excretion (not shown), which occur 1–2 months after the peak in primary production and chlorophyll (Figures 10a and 10b) due to the time lag associated with the zooplankton response and the timing of the diatom bloom (late spring, early summer). Steinberg et al. [2001] report a weak correlation between vertical particle flux and primary production at BATS with a time lag of one week. The significant contribution of diatoms to the formation of sinking organic matter at BATS (50–67%) is consistent with observations [Brzezinski and Nelson, 1995; Nelson and Brzezinski, 1997]. The magnitude of carbon particle flux (POC) in the summer (Figure 10b) is approximately 150% higher than those estimated from sediment traps, which have typical values of ∼20 mg C m−2 d−1 at 150 m [Michaels and Knap, 1996; Steinberg et al., 2001]. This discrepancy is certainly related to the overestimation of phytoplankton abundance in the model and biases as well as uncertainties associated with sediment trap measurements [Steinberg et al., 2001].

Figure 10.

Time series of monthly climatological means of vertically integrated chlorophyll concentration (mg m−2), from the picophytoplankton (P1), diatoms (P2), and all phytoplankton (P1 + P2), vertically integrated new and regenerated production (mMol N m−2 d−1) and export of particulate organic carbon (mg C m−2 d−1) at 107 m from the model at the locations corresponding to the (a and b) BATS, (c and d) NABE, and (e and f) OWSI sites.

5. Annual Cycle at NABE and OWSI

[30] At NABE, model results show good general agreement (Figure 11) with available in situ observations, mostly restricted to the spring bloom and early summer periods [Weeks et al., 1993; Lochte et al., 1993]. Modeled maximum surface chlorophyll concentrations during the bloom agree well with satellite estimates but are ∼40% lower than in situ observations (Figure 6b). The onset and end of the spring bloom also occur a month earlier than in the in situ and satellite data, the difference possibly the result of deficiencies in the model's physical forcing. Numerical experiments with realistic forcing show that the onset of the North Atlantic spring bloom can vary by several weeks depending on surface heat and momentum fluxes [Oschlies et al., 2000]. In our modeled time series, the fall bloom is stronger and occurs a month later than in the SeaWiFS derived climatology (Figure 6b). The modeled seasonal cycle is also consistent with current available knowledge for this region of the ocean [Mann and Lazier, 1991; Longhurst, 1998]. Estimated winter NO3 concentrations [Glover and Brewer, 1988; Garside and Garside, 1993] are of the order of 10 mMol m−3, comparable to model results (Figure 11b).

Figure 11.

Time series of monthly climatological means of observed and modeled (a and b) NO3 (mMol m−3), (c and d) chlorophyll concentrations (mg Chl m−3), and (e and f) primary production (mMol C m−3 d−1) at the NABE site. White line represents mixed layer depth.

[31] OWSI data lack winter observations and vary in frequency but are commonly more frequent than monthly when available (Figures 6c and 12). The model fits the average annual cycle of the chlorophyll data remarkably well, with a seasonal cycle similar to that at NABE but shorter (Figure 12). However, maximum surface chlorophyll concentrations in the in situ observations are ∼55% lower than satellite and model estimates and the in situ data do not show the sharp decline in chlorophyll over the summer seen in the satellite and model data (Figure 6c). The model reproduces the seasonal NO3 trend and the relatively high surface NO3 values observed during summer (2–3 mMol m−3). However, nutrient and primary production measurements were relatively few and problematic (sometimes 2–3 observations per month), particularly at depth [Williams and Robinson, 1973; Fasham et al., 1993].

Figure 12.

Time series of monthly climatological means of observed and modeled (a and b) NO3 (mMol m−3), (c and d) chlorophyll concentrations (mg Chl m−3), and (e and f) primary production (mg C m−3 d−1) at the OWSI site. White line represents mixed layer depth.

5.1. Community Dynamics

[32] At NABE and OWSI, both phytoplankton groups grow rapidly in the spring reaching maximum concentrations near the surface (Figure 8) with diatoms comprising ∼55% of the total phytoplankton biomass. After the spring bloom, phytoplankton levels decline, and the community near the surface transitions into a picophytoplankton dominated regime fueled by regenerated production, with significant levels of new production and diatom concentrations only in the DCML. This community shift is consistent with observations at NABE [Sieracki et al., 1993; Lochte et al., 1993]. The weaker fall bloom at NABE and OWSI is, in the most part, the result of an increase in diatom concentrations.

5.2. Nutrient Limitation Patterns

[33] At NABE and OWSI, both phytoplankton size classes are near their maximum cell quotas during the winter (Figure 9) as their growth (carbon fixation) is strongly light limited. The phytoplankton are also nitrogen-replete during the spring and summer months, as NO3 and NH4+ concentrations remain relatively high during this period (Figures 11 and 12). Diatoms are silica limited near the surface during the summer months (Figure 9), leading to the demise of the diatoms in the upper layers (Figure 8) and the phytoplankton community shift noted above. The simulations are in excellent agreement with observations in the NABE region, where dissolved silicate depletion coincides with a shift in dominant phytoplankton from diatoms to small phytoplankton [Sieracki et al., 1993; Lochte et al., 1993].

5.3. Production and Export

[34] At NABE and OWSI, the peak in new production coincides with the spring bloom, while regenerated production reaches a maximum in early summer (Figures 10d and 10f). At both sites, the peak in the vertical particle flux occurs a month after the peak in new production and diatom abundance, and as at BATS is associated with the maximum in diatom mortality and zooplankton excretion (Figures 10d and 10f). The vertical particle flux minimum in late summer/early fall coincides with the minimum in diatom abundance (Figures 10c–10f) and a shift in community composition [Buesseler et al., 1992; Sieracki et al., 1993; Lochte et al., 1993]. The modeled particulate organic carbon flux at 150 m in the spring at NABE (153.2 mg C m−2 d−1) is approximately 30% higher than observed from sediment traps (117.6 mg C m−2 d−1) [Lochte et al., 1993], a result of uncertainties associated with both sediment trap measurements and model errors.

6. Effects of Silica Limitation on Ecosystem Dynamics

[35] In the model, euphotic zone (0–107 m) diatom growth is generally nitrogen limited in the subtropical gyre and silica limited at higher latitudes and in upwelling regions along the equator and the west coast of Africa (Figure 9f). There is a fair degree of seasonal variation between nitrogen limitation (late winter and early spring) and silica limitation (remainder of the year), mostly along the northwest side of the subtropical gyre near the BATS site. The spatial pattern of silica limitation is consistent with those from global ecosystem models with multinutrient limitation [Moore et al., 2002a, 2004; Aumont et al., 2003]. For the reference case, diatoms are a relatively small component of the phytoplankton community (Figure 2c). Primary production is mostly regenerated, associated with the picophytoplankton (Figure 2f), and diatoms are responsible for a disproportionately large amount of the new/export production (Figure 2e).

[36] In a sensitivity experiment where silica limitation is turned off (Table 3), diatoms consume more nitrogenous nutrients driving down NO3 and NH4+ concentrations in the euphotic zone, diatom relative abundance and total phytoplankton biomass increase, and the contribution of diatoms to new, regenerated and therefore total production increases. However, the intensification of the vertical particle flux associated the the higher diatom abundance reduces the residence time of sinking detritus and results in a significant drop in NH4+ concentrations and regenerated production in the euphotic zone, which depends mostly on the picophytoplankton. The combined increase in new and regenerated production by the diatoms is not large enough to offset the reduction in production by the picophytoplankton, resulting in a decrease in total primary production in the euphotic zone (Table 3). These results are consistent with those from a similar sensitivity experiment by Aumont et al. [2003], in which silica limitation for the diatoms is relaxed in a global ecosystem model.

Table 3. Euphotic Zone Averages for the Last 3 Years of the Main Run and the “No Silica Limitation” (NSLIM) Run and the Percentage Difference Between the Two Runs (NSLIM - Main)
 Main RunNSLIM RunPercent Difference
Nutrients, mMol N m−3
  NO34.01643.5891−10.64
  NH4+0.05370.0291−45.81
Phytoplankton biomass, mMol C m−3
  Picophytoplankton1.14891.16721.59
  Diatoms0.27620.464067.99
  Total1.42511.631214.46
Primary production, mMol C m−3 d−1
  Picophytoplankton0.28920.2098−27.46
  Diatoms0.05630.097973.89
  Total0.34540.3077−10.91
New production, mMol N m−3 d−1
  Picophytoplankton0.00430.0033−23.26
  Diatoms0.00510.006527.45
  Total0.00940.00973.19
Regenerated production, mMol N m−3 d−1
  Picophytoplankton0.03490.0191−45.27
  Diatoms0.00450.007260.00
  Total0.03930.0263−33.08
Vertical particle flux, mMol C m−2 d−13.57514.828735.06

[37] In all three JGOFS sites, the absence of silica limitation results in a increase in the diatom's relative abundance and a decline in picophytoplankton biomass (Figures 13a, 13c, and 13e). At BATS, where the picophytoplankton is the dominant group, total phytoplankton biomass decreases. At BATS and NABE, the higher export flux associated with the higher diatom abundance causes a reduction in regenerated (and total) production (Figures 13b and 13d). At OWSI, however, diatoms are more abundant during the summer than at NABE (Figure 8) and the increase in regenerated production by the diatoms compensates for the drop in regenerated production by the picophytoplankton, resulting in a small increase in total primary production (Figure 13f). At BATS, the effect of lack of silica limitation is smallest during the spring bloom season, when diatoms are mostly nitrogen limited (Figures 13a and 13b).

Figure 13.

Time series of the percentage difference in monthly climatological means of vertically integrated chlorophyll concentration (mg m−2), from the picophytoplankton (P1), diatoms (P2) and all phytoplankton (P1 + P2), vertically integrated new and regenerated production (mMol N m−2 d−1), and export of particulate organic carbon (mg C m−2 d−1) at 107 m between the “no silica limitation” run and the main run at the locations corresponding to the (a and b) BATS, (c and d) NABE, and (e and f) OWSI sites.

[38] Model skill for chlorophyll is adversely affected by the removal of silica limitation, particularly the monthly-time case (Figure 5a). In the absence of silica limitation, the decrease in regenerated production caused by the intensification of the particle flux causes a drop in picophytoplankton and total phytoplankton chlorophyll along the upwelling regions off the coast of Africa and inside the oligotrophic gyre (not shown). This effect is most pronounced in the summer. In addition, diatom and total chlorophyll concentrations in temperate and subpolar regions are further overestimated during spring and early summer. These effects increase the amplitude of variation of the model's chlorophyll fields with respect to the satellite's and reduce the correlation between simulated and satellite fields (Figure 5a). The removal of silica limitation has a similar but less noticeable effect on model skill for primary productivity (Figure 5c).

[39] In our model, the effects of silica limitation and species/size structure in the model are directly linked, as silica limitation controls over much of the domain the spatial and temporal shifts between a picophytoplankton-suspended detritus and a diatom-sinking detritus dominated community. In the absence of silica limitation, model behavior is similar to that of a simple NPZD model with one phytoplankton species and one large sinking detritus compartment. The similarity between the error patterns under the no silica limitation sensitivity simulation and previous model studies suggests that an intermediate level of complexity is required to capture the observed basin-scale variability.

7. Role of Horizontal Advection in the Subtropics

[40] In the model, horizontal advection is an important source of inorganic nutrients and dissolved organic matter for the plankton community in the subtropical gyre (Figure 14). Except for the Gulf Stream region and the eastern boundary, where transport is northward, the meridional advective transport is mostly southward, from higher latitudes into the central region of the subtropical gyre. Years with strong southward transport of nutrients (NO3 and SiO3) and chlorophyll correspond to larger spring blooms and an increase in diatom relative abundance at BATS (Figure 14f). Diatoms are normally a comparatively small component of the phytoplankton community at BATS but with significant interannual variability in their relative abundance and strong episodic blooms [Hulburt, 1990; Siegel et al., 1990; Michaels and Knap, 1996; Steinberg et al., 2001]. Diatoms blooms at BATS occur in late spring and early summer, indicating that they are not directly related to the increased nutrient input during winter mixing. Steinberg et al. [2001] attributes the diatom blooms to local nutrient input/injection by episodic events, such as mesoscale eddies. The model results suggest that the occurrence and strength of diatom blooms at BATS could be related to interannual variations in the advective/meridional transport of nutrients (mainly SiO3) and cells from higher latitudes. The severe silica limitation for diatoms after the spring bloom at BATS (Figure 9) is consistent with this hypothesis.

Figure 14.

Time series of the meridional transport (Mol s−1) of (a) NO3, (b) SiO3, and (c) dissolved organic carbon (DON) in the top 107 m across 31°32′N (around the BATS site). The dashed lines in Figures 14a and 14b correspond to the longitude of the BATS site and (d) the nutrient transport time series plot. (e) Time series of the meridional transport of picophytoplankton and diatom chlorophyll at the same location. Positive and negative values indicate northward and southward transport, respectively. (f) Time series of monthly means of total vertically integrated chlorophyll (P1 + P2) concentrations (mg m−2) and the percentage change in picophytoplankton (P1) and diatom (P2) chlorophyll concentrations in relation to their respective means.

[41] The lateral advection of DON from areas of higher biological production in the north (Figure 14c) is another important source of nitrogen for the phytoplankton (mostly picophytoplankton) in the oligotrophic gyre, where primary production is based on intense recycling of nutrients (Figures 8a and 10b). Sensitivity experiments using only one single large sinking detritus compartment produced significantly lower primary production rates and chlorophyll concentrations over a considerably larger area of the subtropical gyre (not shown). The importance of the suspended detritus → ammonium → picophytoplankton pathway for maintaining background chlorophyll concentrations is evident when we compare the relative magnitude of the annual mean distributions of detritus remineralization and new and regenerated production by the picophytoplankton and diatoms (Figure 15). Our results are consistent with the climatological analysis of Williams and Follows [1998], which demonstrates that southward surface Ekman flow transfers significant amounts of inorganic and organic nutrients from the subpolar into the subtropical gyre.

Figure 15.

Annual mean distribution of vertically integrated suspended (D1) and sinking (D2) detritus remineralization (mMol N m−2 d−1) and new and regenerated production by the picophytoplankton (P1) and diatoms (P2) (mMol N m−2 d−1).

8. Discussion

[42] It is evidently difficult to model an entire ocean basin with its wide range of biogeographical provinces and diverse plankton communities with one single set of parameters. However, our multinutrient, multispecies formulation seems to provide enough flexibility to begin to represent biological processes in both tropical/subtropical and high latitude areas of the North Atlantic. The model also reproduces observed characteristics of the ecosystem dynamics, e.g., the dominance of the picophytoplankton and episodic diatom blooms in the subtropics, the nutrient-controlled seasonal succession in the phytoplankton community at higher latitudes, and the associated seasonal/depth changes in new and regenerated production and export of particulate carbon.

[43] While not a full measure of model performance, we introduce formal, quantitative metrics of model skill using the Taylor diagram approach (time-space variance, model pattern correlation, and RMS error) with satellite ocean color and primary productivity estimates. The main discrepancy between model open ocean results and observations is the overestimation of chlorophyll concentrations and primary production along the northern edge of the subtropical gyre, due to excessive convective mixing in winter from the physical model. This systematic error is remarkably insensitive to changes in the OGCM's vertical mixing parameterization, indicating probable deficiencies in the surface forcing fields. The fact that Oschlies and Garçon [1999] and Oschlies et al. [2000] report similar errors in their eddy-permitting simulations suggests that the problem is not related to our model's lower resolution. We are currently investigating ways to eliminate or minimize or this problem. Another significant difference between model and observations is the underestimation of chlorophyll concentrations and primary production along the coastal areas. This is a result of the model's spatial resolution, which does not capture coastal processes very well.

[44] Our model shows better skill than many previous attempts in simulating the general spatial and temporal patterns in nutrients, chlorophyll and primary production seen in in situ and satellite data. A persistent feature in earlier ecosystem models, including both coarse-resolution [Sarmiento et al., 1993; Fasham et al., 1993] and eddy-permitting [Oschlies and Garçon, 1998; Oschlies et al., 2000; Oschlies, 2001, 2002] simulations, is the severe underestimation of chlorophyll concentrations and primary production rates in the oligotrophic subtropical gyre. The coarse-resolution (3.5°) global ecosystem model of Aumont et al. [2003] also significantly underestimates chlorophyll concentrations over large portions of the subtropical gyre in the North Atlantic. In the present study, chlorophyll and primary production values in the subtropical gyre, while still lower than satellite estimates, are considerably higher and more realistic. Chlorophyll concentrations below 0.02 mg m−3 occur only over a relatively small area in the center of the gyre, and minimum chlorophyll and primary production values in the summer are only 50% lower that satellite derived measurements. Lower than observed chlorophyll and primary production values in the subtropical gyre are expected, as the model does not include episodic upwelling of nutrients by mesoscale eddies [McGillicuddy et al., 1998; Siegel et al., 1999; Garçon et al., 2001; McGillicuddy et al., 2003] and the fixation of atmospheric nitrogen (N2) by cyanobacteria [Michaels and Knap, 1996; Gruber and Sarmiento, 1997; Capone et al., 1997; Lipschultz et al., 2002].

[45] The simulations of the seasonal cycle of nutrients, chlorophyll and primary production at the JGOFS sites presented here also show significant improvements with respect to those from previous similar studies [Fasham et al., 1993; Oschlies et al., 2000]. Despite the overestimation in winter mixing, maximum nitrate and chlorophyll concentrations in winter and early spring at the actual BATS site are significantly lower and more realistic in the present model (0.12 mMol N m−3 and 0.5 mg Chl m−3 compared to 1.4 mMol N m−3 and 1.2 mg Chl m−3 of Oschlies et al. [2000], and 3.5 mMol N m−3 and 5.5 mg Chl m−3 of Fasham et al. [1993]). Primary production levels during the summer at BATS are also considerably higher and closer to observations than those reported by Fasham et al. [1993] and Oschlies et al. [2000]. Simulated annual new production at the BATS location (0.21 Mol N m−2 yr−1) is lower than most observational estimates [see McGillicuddy et al., 2003, Table 5] which is to be expected as noted above. At NABE, Oschlies et al. [2000] overestimates phytoplankton biomass during the spring bloom by 30%, and their single-phytoplankton, single-detritus model fails to capture the positive correlation between phytoplankton biomass and particle export during and after the spring bloom. In addition, the termination of the bloom is dependent on an unrealistically large peak in zooplankton. In the present study, phytoplankton biomass during the spring bloom is 40% lower than in situ measurements but agrees well with satellite estimates, and the observed nutrient-controlled changes in phytoplankton abundance and vertical particle flux are nicely reproduced by the model. At OWSI, Fasham et al. [1993] overestimates chlorophyll concentrations and primary production rates during the spring bloom by more than 200%. In contrast, our model reproduces the magnitude and amplitude of variation of primary production rates at OWSI correctly, and fits the observed seasonal cycle of chlorophyll data remarkably well. Model estimates of annual new production at NABE (1.25 Mol N m−2 yr−1) and OWSI (0.80 Mol N m−2 yr−1) are also within the range of observed values [see McGillicuddy et al., 2003, Table 5].

[46] The fact that the eddy-permitting experiments of Oschlies et al. [2000], which include important nutrient inputs by mesoscale variability, provide only a relatively modest improvement over the results from the coarse-resolution models of Sarmiento et al. [1993] and Fasham et al. [1993] in the subtropical gyre suggests that a more realistic representation of the physical environment alone is not sufficient to simulate the observed biological fields and processes and that ecosystem dynamics are an important factor controlling biological production. The results presented here support this hypothesis. In the present study, the explicit inclusion of the suspended detritus → ammonium → picophytoplankton pathway in an intermediate resolution OGCM results in a significant improvement in the simulations in the oligotrophic subtropical gyre. In high-resolution experiments, chlorophyll and primary production levels in the subtropical gyre remain abnormally low despite the more realistic simulation of nutrient inputs by mesoscale processes, due to considerable underestimation of regenerated production by the simple ecosystem model [e.g., Oschlies et al., 2000; Oschlies, 2002]. Conversely, coarse-resolution models with increased ecological complexity (multinutrient limitation and suspended and sinking detritus) [e.g., Aumont et al., 2003] also tend to significantly underestimate chlorophyll concentrations in the subtropical gyre because of inadequate representation of the physical environment.

[47] The sensitivity experiments show that the inclusion of silica limitation for the diatoms results in an improvement in model skill. The sensitivity experiments also highlight the importance of diatoms in the ocean's carbon cycle and indicate that a reduction in silica limitation leads to a overall decrease in total primary production in the euphotic zone, through community shifts and counteracting effects of higher new/export production and lower regenerated production. This has important implications for projecting future climate change scenarios as floristic shifts, induced by climate variability, may have a significant impact on oceanic uptake of anthropogenic CO2 [Boyd and Doney, 2002].

[48] Marine ecosystems are composed of a diverse mixture of taxonomically and biogeochemically distinct groups. However, computational costs and the difficulty in parameterizing complex food webs impose severe limits on the construction of realistic ecosystem models. In fact, determining the required level of complexity to accurately simulate the marine ecosystem's response to climate change is one of the major issues confronting biological oceanographers today [Doney, 1999; Denman, 2003]. The present study suggests that an intermediate level of complexity above simple NPZD models is required for capturing basin-scale patterns of phytoplankton abundance and primary production, particularly in the oligotrophic gyre. We are currently working on expanding the ecosystem model to include additional phytoplankton functional groups (diazotrophs and coccolithophores) and limiting nutrients (phosphate and iron) and improving our physical forcing (e.g., including the diurnal cycle in light intensity) to address global-scale issues [Moore et al., 2004].

Appendix A:: Taylor Diagram

[49] A Taylor diagram provides a measure of the degree of pattern correspondence between a “reference” field, usually representing observations, and a “test” or model simulated field [Taylor, 2001]. This diagram combines the correlation coefficient (R) and the RMS difference (E) between the two fields along with the ratio of the standard deviations of the two patterns (σtestref) into one point in a two-dimensional plot (Figure 5). The ratio of the standard deviations indicates the relative amplitude of the simulated and observed variations, while the correlation coefficient indicates whether the fields have similar patterns of variation, regardless of amplitude. The normalized RMS difference reflects differences in the overall pattern of variations. In the diagram, the radial distances from the origin are proportional to the ratio of the standard deviations and the azimuthal positions give the correlation between the two fields (Figure 5). The point representing the reference field is plotted along the abscissa and has coordinate σrefref = 1 and R = 1. The distance between the test and reference point is proportional to the normalized RMS difference between the two fields.

[50] Consider a test field p and a reference field q defined in a M × N spatial grid and in T points in time, where each time (t) represents a monthly mean (t = {1, 2,…,12}). The area of each grid cell is given by ai,j (i = {1, 2,…,M}; j = {1, 2, …,N}).

A1. Total Time-Space Case

[51] In the total time-space case the standard deviations for p and q are computed as:

equation image

where equation image and equation image are total time-space means.The correlation coefficient between p and q is given by:

equation image

And the RMS difference is calculated as:

equation image

A2. Monthly-Time Case

[52] In the monthly-time case the standard deviations, correlation coefficients and RMS difference for each field are computed in the same way as in the total time-space case, except that deviations are computed with respect to the temporal (annual) means at each grid point equation image and equation image.

A3. Spatial-Annual Case

[53] In the spatial-annual case the standard deviation for fields p and q are computed as

equation image

The correlation coefficient between p and q is calculated as:

equation image

And the RMS difference is given by:

equation image

A4. Monthly-Spatial Case

[54] In the monthly-spatial case the standard deviations, correlation coefficients and RMS difference for p and q at each month (t) are computed in the same way as in the spatial-annual case, except that the deviations are given by pi,j,tequation image and qi,j,tequation image, where equation image and equation image are spatial means for each month (t).

Acknowledgments

[55] We thank Joanie Kleypas, Keith Lindsay and Keith Moore for their help in the development of the ecosystem model and for stimulating discussions pertaining to this work. Support for this work was provided by NASA SeaWiFS grant W-19,223 and NSF JGOFS SMP grant 0222033. The authors gratefully acknowledge the NCAR Climate Simulation Laboratory for the computer time provided. This is WHOI contribution 11204 and U.S. JGOFS contribution 1050.

Ancillary