Introduction
Recent years have seen an intense effort to better understand
the global biogeochemical cycle. Scientific cruises organized by programs
such as CLIVAR, WOCE, and GEOTRACES have generated a wealth of information
about physical and chemical tracers in the global oceans, most of which have
been aggregated to climatologies e.g.,. These field
data, together with satellite observations of phytoplankton biomass, have
helped in assessing the mean state and variability of the global marine
ecosystems. Concurrently, a number of global ocean general circulation models
(OGCMs) with added functionality to simulate biogeochemical processes have
been developed, mainly to study trends and variabilities in earth's climate
and the global carbon cycle.
This type of global climate model is predominately constructed and tuned to
simulate decadal large-scale processes such as the global wind-driven and
thermohaline circulation or the distribution of major water masses. However,
an increased interest in biophysical interactions on smaller temporal and
spatial scales, together with the recent ability to run global eddy-resolving
OGCMs, has raised the question of how well coarser global climate models
perform in daily to seasonal time domains. A model which has a high skill in
estimating the basin-wide annual mean phytoplankton biomass might not
correctly simulate process-level details in the seasonal cycle of biological
production, such as the onset of spring blooms.
With this study, we aim to do such an evaluation by comparing two
state-of-the-art OGCMs with observations of Chl and biological production.
Since our main focus is the seasonal cycle and regional variability, we use
two properties with high temporal and spatial resolution – satellite-derived
Chl and in situ estimations of net community production (NCP) based on the
ΔO2/Ar method. Chl and NCP are particularly well suited
because they represent the “end product” of bottom-up-driven biogeochemical
models and are both of general interest in climate change studies. Satellite-derived net primary production (NPP) is not included in this study since it
is closely correlated to Chl on the timescales of interest. We focus on the
Southern Ocean south of 40∘ S since a large number of
ΔO2/Ar measurements have been collected from this area
. While satellite-derived
Chl is a well-known and widely used property, ΔO2/Ar
based NCP will require a more detailed introduction.
The ΔO2/Ar method was developed to estimate oceanic NCP
by measuring the saturation of O2 and Ar in the mixed layer
e.g.,. O2 supersaturation occurs from both
biological O2 production and three physical processes: warming,
changes in air pressure, and bubble entrainment. It is possible to decouple
the biological component from the physical component by using the saturation
of Ar, since Ar has similar physical properties as O2
but is biologically inert. Following , we define
O2 supersaturation relative to Ar as
ΔO2/Ar=O2/ArsampleO2/Areq-1.
This term is equivalent to the biological O2 supersaturation. Knowing
ΔO2/Ar, one can approximate the loss of biological
O2 via gas transfer across the air–sea interface (hereafter denoted
O2 bioflux) via the relationship
O2bioflux≈κ⋅ΔO2/Ar⋅O2eq,
where κ is gas transfer velocity based on
and O2eq the concentration of O2 at equilibrium. It is possible to
use either ΔO2/Ar or O2 bioflux when comparing
observations and models. The relative advantages of the two properties are
further discussed in Appendix A; we will use O2 bioflux in this study.
One key challenge for OGCMs is to simulate the vertical exchange of water and
tracers. Most physical processes that generate vertical mixing or advection
act on length scales several orders of magnitude smaller than the model
resolution. Instead, the models use different types of parameterizations to
mimic the transport between surface waters and the deep ocean
. The results are evaluated by comparing
model simulations with observed global distributions of transient tracers
(e.g., radiocarbon or chlorofluorocarbons), especially in the deep sea. The
boundary layer parameterizations embedded in OGCMs are also often tested in
1-D against high-frequency observations e.g.,. It is,
however, less clear how well vertical processes in the mixed layer and
thermocline are represented by these parameterizations. One reason is that,
until now, there have been few good methods for evaluating vertical
transports on these spatial and temporal scales.
In this study we explore the feasibility of using O2 bioflux to evaluate
how well vertical processes are resolved on shorter timescales. This is
possible due to the fact that the actual O2 balance of the mixed
layer is
dO2/dt=NCP+-FO2+Din/hml,
where dO2/dt is the time rate of change of
dissolved oxygen in units of mmolm-3day-1, FO2 is
the sea to air gas exchange flux, Din is net input (or loss) of
O2 to the mixed layer from ocean physics (i.e., mixing and
advection), and hml is the mixed layer depth
. Assuming steady state, this equation can be rewritten
as a negative relationship between O2 bioflux and the net downward vertical
transport of O2:
NCP=O2bioflux+O2vertical flux.
O2 bioflux can hence be used to give a combined evaluation of how well
models simulate upper ocean biogeochemical rate processes, sea–air
O2 fluxes, and vertical mixing. Such a combined assessment is
valuable since the interaction between physics and biology is a key source
for variability on short timescales (days to months).
We will first compare regional and seasonal patterns in Chl and O2 bioflux
between observations and models, continue with exploring how O2 bioflux
can show discrepancies in how the models simulate vertical transports, and
finally discuss how the results could help to identify mechanisms that
contribute to mismatches between observations and models.
Methods
The main focus of the study is to compare model output with in situ
observations and satellite-derived properties. The data from each source have
different temporal and spatial resolution and span different ranges in
time. To compensate for these discrepancies, we regrid in situ observations
and satellite fields to a model grid with roughly 1∘ resolution at
equal intervals in time. We also combine observations from different years
but the same day of the year (e.g., 1 January 2003, 1 January 2004, 1 January 2005)
to a climatological year with a daily time resolution.
Satellite data
We use remotely observed chlorophyll concentrations from MODIS (Moderate
Resolution Imaging Spectroradiometer)/Aqua on the Level-3 9km×9km grid. Daily satellite images from 2003 to 2010 were aggregated
to the model grids and averaged to a day-of-the-year climatology. Satellite
data, particularly in the Southern Ocean, suffers from a high frequency of
days where clouds, light conditions, sea ice, or other problems disqualify
the observations. The relative frequency of days with valid information in
our data set is between 5 and 15 % on the original grid and between 20
and 50 % on the aggregated model grid.
In situ observations
We use ΔO2/Ar observations from 19 Southern Ocean
cruises between 1999 and 2009, all occurring during the austral summer. The
geographical locations of the respective ship tracks are presented in
Fig. . The measurements were conducted by two different
methods: water was collected in bottles and analyzed in the lab (sampling locations shown in blue in
Fig. ) on 16 cruises, and ΔO2/Ar was
measured directly using a ship-borne flow-through system on 3 cruises (shown
as red in Fig. ). The measurements are clustered in space
and time reflecting tracks of the ships of opportunity used in this study. We
use these sampling clusters in our analysis as natural areas to compare and
contrast different parts of the Southern Ocean. A more detailed description
of the sampling strategies, measurement methods, and data sources can be
found in and .
Bioflux is calculated as the product of the biological O2
supersaturation and the gas transfer velocity. The latter is determined using
the parameterization expressing gas transfer
velocity in terms of a quadratic function of wind speed and the Schmidt
number. We do the calculation using daily averages of the NESDIS wind product
with a 0.5∘ resolution based on data from the QuikSCAT (Quick
Scatterometer) satellite . The gas transfer velocity for
each ΔO2/Ar measurement is calculated from the daily
mean local wind speeds during the 60 days preceding collection. A
time-weighted value for the gas transfer velocity is calculated based on the
fraction of the mixed layer flushed in each subsequent interval until
sampling . The resulting gas transfer
velocity is then used in Eq. (3) to calculate O2 bioflux from
ΔO2/Ar supersaturation. A detailed analysis of possible
uncertainties associated with the method can be found in
.
Finally we use mixed layer depth (MLD) as a baseline diagnostic of how well
vertical physical processes in the surface ocean are resolved. A total of
about 75 000 vertical profiles from Argo floats between 2001 and 2012 are
used to estimate in situ MLDs for the different regions. The observed MLD is
defined as the depth where density is 0.03 kgm-3 higher than at
the most shallow observation.
Map of O2 observations used in the study. Grey shadings signifies
the four regions we focus on in this study. Dots indicate discreet sampling
locations, while red lines indicate continuous sampling.
Models
The observations are compared with output from the ocean components of two
Intergovernmental Panel on Climate Change (IPCC)-class ocean biogeochemical
models. The TOPAZ (Tracers in the Ocean with Allometric Zooplankton) ocean
model is based on version 4 of the modular ocean model (MOM4;
) with a vertical z coordinate and a horizontal
B-grid with a tripolar coordinate system (North America, Siberia, and
Antarctica) to resolve the Arctic. The model has a nominally 1∘
horizontal resolution globally, with higher meridional resolution near the
equator (to 1/3∘). There are 50 vertical layers; resolution is
10 m in the upper 200 m, and coarser below. MOM4 includes
a representation of the K-profile parameterization (KPP) planetary boundary
layer scheme , Bryan–Lewis deeper vertical mixing,
Gent–McWilliams isopycnal thickness diffusion , bottom
topography represented with partial cells, isotropic and anisotropic
friction, and a multiple-dimensional flux-limiting tracer advection scheme
using the third-order Sweby flux limiter. For these studies, the ocean model
is forced by prescribed boundary conditions from the reanalysis effort of the
ECMWF (European Centre for Medium-Range Weather Forecasts) and NCAR (National
Center for Atmospheric Research) Common Ocean-ice Reference Experiments
(CORE).
BGCCSM (BioGeochemistry Community Climate System Model) is based on the Los
Alamos National Laboratory Parallel Ocean Program (POP) .
In our application, the grid is symmetric in the Southern Hemisphere
with a zonal resolution of 3.6∘. Meridional resolution decreases from
1.8∘ at midlatitudes to about 0.8∘ at high and low latitudes.
The surface layer is 12 m thick; in total, there are 5 layers to
111 m and 25 layers to the bottom. This model invokes
Gent–McWilliams' isopycnal mixing and the KPP upper ocean model and is also
forced by prescribed boundary conditions from CORE.
The air–sea fluxes of O2 and CO2 in both models are computed
using prescribed atmospheric conditions (surface pressure, mole fraction),
model-predicted surface water concentrations, NCEP (National Centers for
Environmental Prediction) surface winds, and the quadratic dependence of the
gas exchange coefficient on wind speed . Argon was
added as a prognostic tracer to the simulations in both models in an
analogous fashion to O2; i.e., O2 and Ar solubility are
similarly determined using model temperature and salinity, and Ar uses
the same gas exchange parameterization as O2.
Both models include complex biogeochemistry/ecosystem components with macro-
and micronutrients, organic matter, and three phytoplankton functional
groups: small phytoplankton, large phytoplankton, and diazotrophs. Both
invoke co-limitation by iron, light, and nitrogen, mediated in part by their
influence on the Chl : C ratio, and slower rates of photosynthesis at lower
temperatures. Diazotrophs have high N : P ratios and low photosynthetic
efficiencies (TOPAZ) or high iron requirements (TOPAZ, BGCCSM). NCP is
calculated as the difference between production and consumption of carbon by
the different functional groups.
It is possible to use either ΔO2/Ar or O2
bioflux for comparing observations with models. One could argue that
ΔO2/Ar is a more robust property since it is an
observed quantity, whereas O2 bioflux is a derived product that
depends on the wind field. On the other hand, model
ΔO2/Ar depends on the choice of wind forcing, whereas
O2 bioflux is assumed to be directly linked to NCP. Our tests show
that both methods give similar accuracy, and we choose to use O2
bioflux since its units are more appropriate for the current study (see Appendix A for further discussion).
Results
First we test the feasibility of aggregating data from different years into
a single climatological year. This approach is useful only if the difference
between seasons is significantly larger than the interannual variability. We
test the assumption by comparing how model O2 bioflux changes from
one date to another in a climatological year against the standard deviation
at the same dates between a number of years. As a test case, fields from
BGCCSM were used to generate climatologies for the days 15 November and
15 December by averaging data for the entire Southern Ocean from four
consecutive model years. Our results show that the mean difference is
19.8 mmol O2 m2d-1 between the two different
dates, whereas the standard deviation over 4 years on the respective dates is
only 8.7 mmol O2 m-2d-1. It is hardly an
unexpected result that spring values of O2 bioflux differ from summer
since the Southern Ocean is a high-latitude region. This result encourages us
to aggregate observations and model data from different years into
a climatological year.
Next, we compare the model simulations with observations. Chlorophyll
simulated by BGCCSM and TOPAZ is related to satellite observations retrieved
by the MODIS/Aqua mission from 1 January 2003 to 31 December 2010. Each daily
satellite image is reprojected from a native 9km×9km resolution to the TOPAZ 1∘×1∘ grid, and the
fields are aggregated to a climatological year, as mentioned earlier. The
resulting data sets are analyzed in four geographical sectors, shown as shaded
areas in Fig. . We zonally average the data in each sector
to a Hofmøller diagram with latitude on the y axis, time on the
x axis, and daily zonal Chl averages as colors
(Figs. –). It should be noted
that one problem with using satellite-retrieved chlorophyll is a systematic
lack of data during winter due to low-light conditions and sea-ice cover in
combination with the satellite's track. These periods are shown as gray areas
in the figures. Figure shows a more detailed representation of
the data at 60∘ south.
Figures – present model NCP
and bioflux vs. our observations. Panel a in each figure shows the temporal
evolution of NCP vs. latitude in the two models, and panel b the
corresponding model O2 bioflux. The sampling locations are indicated
on the model plots by gray circles. Panel c, finally, presents O2
bioflux from the observations shown in Fig. . As mentioned
earlier, all measurements that fall on the same day of the year and grid cell are
combined to one mean value, which is indicated by the color of the dot. Note
that the aggregated values of observed O2 bioflux presented together
with TOPAZ are somewhat different from the ones connected to BGCCSM, since
the two model grids are different.
Hofmøller plots of satellite Chl (top panel), Chl simulated by
TOPAZ (middle panel), and Chl simulated by BGCCSM (bottom panel) in Drake
Passage. All panels are zonal medians within the box. All data that fall on a
specific grid cell on a specific day of the year are averaged to one
value.
Hofmøller plots of satellite Chl (top panel), Chl simulated by
TOPAZ (middle panel), and Chl simulated by BGCCSM (bottom panel) south of
South Africa. All panels are zonal medians within the box. All data that fall
on a specific grid cell on a specific day of the year are averaged to one
value.
Hofmøller plots of satellite Chl (top panel), Chl simulated by
TOPAZ (middle panel), and Chl simulated by BGCCSM (bottom panel) south of
Australia. All panels are zonal medians within the box. All data that fall on
a specific grid cell on a specific day of the year are averaged to one
value.
Hofmøller plots of satellite Chl (top panel), Chl simulated by
TOPAZ (middle panel), and Chl simulated by BGCCSM (bottom panel) south of New
Zealand. All panels are zonal medians within the box. All data that fall on a
specific grid cell on a specific day of the year are averaged to one
value.
Detail of the Chl concentrations in mgm-3 from
Figs. –. The lines show a
slice of each original panel at 60∘ S. All data that fall on a
specific grid cell on a specific day of the year are averaged to one
value.
Hofmøller plots of model NCP (mmolm-2d-1, top
panel), model O2 bioflux (mmolm-2d-1, middle panel) , and
observed O2 bioflux (mmolm-2d-1, bottom panel) in Drake
Passage (locations of observations are presented as red points on the map).
All model values are zonal medians within the box. All observations that fall
on a specific grid cell on a specific day of the year are averaged to one
value.
Hofmøller plots of model NCP (mmolm-2d-1, top
panel), model O2 bioflux (mmolm-2d-1, middle panel), and
observed O2 bioflux (mmolm-2d-1, bottom panel) in the
Southern Ocean south of South Africa (locations of observations are presented
as red points on the map). All model values are zonal medians within the box.
All observations that fall on a specific grid cell on a specific day of the year are
averaged to one value.
Hofmøller plots of model NCP (mmolm-2d-1, top
panel), model O2 bioflux (mmolm-2d-1, middle panel), and
observed O2 bioflux (mmolm-2d-1, bottom panel) in the
Southern Ocean south of Australia (locations of observations are presented as
red points on the map). All model values are zonal medians within the box.
All observations that fall on a specific grid cell on a specific day of the year are
averaged to one value.
Hofmøller plots of model NCP (mmolm-2d-1, top
panel), model O2 bioflux (mmolm-2d-1, middle panel), and
observed O2 bioflux (mmolm-2d-1, bottom panel) in the
Southern Ocean south of New Zealand (locations of observations are presented
as red points on the map). All model values are zonal medians within the box.
All observations that fall on a specific grid cell on a specific day of the year are
averaged to one value.
A general pattern arises where the models simulate upper ranges of Chl well
but underestimate spring levels significantly. Each model also shows
differences in skill between early and late parts of the growing season,
performing better during the spring and early summer. While there is a great
deal of scatter in the bioflux observations, the four sectors each show
distinct patterns, whereas the models exhibit little distinction between the
Australian and New Zealand sectors and between the Drake Passage and African
sectors. Next, we compare the simulated fields and data for each of the study
regions in more detail.
Drake passage
The seasonal change in satellite-derived chlorophyll for the Drake Passage
sector is shown in the top panel of Fig. . It is
possible to discern a seasonal cycle, even though we lack winter and early
spring data in higher latitudes. Towards the south, the earliest retrieved
concentrations are between three and five times or more lower than maximum
values at the late spring/summer peak. Both the magnitude and timing of
maximum summer Chl concentrations vary with latitude, with the strongest and
latest blooms occurring in high latitudes. [Chl] reaches 1 mgm-3
south of 70∘ S, decreases to 0.5 mgm-3 between 60 and
70∘ S, and increases again north of 50∘ S. The growing
season, as inferred from the period of elevated summer Chl, is about 3 months
south of 70∘ S and generally lengthens going to the north. The bands
of very high Chl north of 45∘ S and south of 70∘ S are likely
due to the transport of sedimentary iron from coastal waters generating
elevated biological production.
Panels b and c present Chl climatologies from TOPAZ and BGCCSM. The grey
shadings indicate where satellite coverage is missing. It is clear that the
seasonal chlorophyll progression in both models diverges from observations,
whereas the magnitude of peak Chl concentrations is simulated quite well.
TOPAZ underestimates Chl concentrations somewhat early in the season and
generates an intense spring bloom extending over the entire sector from 75 to
40∘ S. (“Bloom” as used here indicates a transient period of high
[Chl] lasting no more than 1 month.) In the observations, high [Chl], once
established, continues throughout most of the summer, and concentrations >1 mgm-3 are limited to the region south of 70∘ S. In
these southern regions, the onset of the TOPAZ bloom comes 1–2 months
earlier than in the observations. The spring bloom in TOPAZ collapses after
about 1 month with concentrations of Chl that are too low over the austral summer as
a result. Finally, the season ends with TOPAZ generating a weaker fall bloom
during part of the season without satellite coverage. South of 50∘ S,
BGCCSM begins the season with much lower Chl concentrations than both TOPAZ
and satellite observations. The model generates a much more intense bloom
between 60 and 70∘ S than observations suggest but has good skill in
predicting the timing and magnitudes south of 70∘ S. Both models
suggest that during the bloom, biomass is significantly higher in the frontal
regions compared to observations.
O2 bioflux observations in the Drake Passage region originates both
from transects crossing the Drake Passage and nearby cruises in the Palmer
Long-Term Ecological Research program annual survey west of the Antarctic
Peninsula. Our results are thus partly influenced by coastal processes
outside the models' domain. In the Drake Passage sector, TOPAZ and BGCCSM
show similar patterns in NCP (panel a) north of 65∘ S. South of
65∘ S, TOPAZ simulates intensive NCP over the summer, whereas BGCCSM
has close to zero net production. Both models have predominantly negative
O2 biofluxes south of 50∘ S in October (panel b) and
a progressive change to positive fluxes from north to south with time. The
negative values of O2 bioflux in TOPAZ are about twice as large as
they are in BGCCSM. In TOPAZ, O2 bioflux is positive throughout the
domain in January, whereas in BGCCSM there is a southerly region of negative
flux throughout the season.
Panel c of Fig. shows observations of O2
bioflux in the region. The pattern of negative fluxes south of 60∘ S
in November and December corresponds qualitatively with the models. There is
a wealth of observations south of 60∘ S in January and February, but
they show significant variability. Some observations suggest biological
production rates of up to 50 mmolm-2d-1, which is similar
to the levels predicted by TOPAZ. Others correspond to the results from
BGCCSM with O2 bioflux estimates near or below 0. The mean observed
O2 bioflux values in January and February south of 65∘ S are
8±20 mmol O2 m-2d-1. The corresponding
values for BGCCSM are 9±13, and for TOPAZ, 16±27. The
observations in April suggest undersaturated conditions that correspond with
BGCCSM but not with TOPAZ.
South Africa
The observations in the South Africa sector (Fig. , top
panel) show similar patterns to the Drake Passage sector for
satellite-derived Chl, with the main exception of a weaker bloom north of
50∘ S. Note the lack of observations south of 70∘ S where of
Chl concentrations may be very high. TOPAZ (panel b) begins the season with
somewhat higher concentrations of Chl than observed and, with time, generates
a significantly stronger bloom. As in the Drake Passage, the TOPAZ bloom
crashes midsummer, after which model Chl concentrations are significantly
lower than observations. BGCCSM starts the season with much lower Chl
concentrations than the satellite data and generates an exaggerated bloom as
well, especially south of 55∘ S.
NCP and O2 bioflux climatologies (Fig. ) show
that TOPAZ has an earlier and more intense band of positive NCP than BGCCSM
(panel a). BGCSSM has a slow southward progression of the positive NCP
band with NCP reaching 20 mmolm-2d-1 at 40∘ S in
October and at 65∘ S in February. TOPAZ, on the other hand, has
a more uniform pattern, in which most of the region reaches these levels of
NCP by mid-November. The two models have O2 bioflux patterns (shown
in panel b) that vary in a fashion similar to that of the Drake Passage
region. Both models start with a negative O2 bioflux:
< -30 m2d-1 in TOPAZ and
∼ -10 mmolm-2d-1 in much of the domain in BGCCSM.
This negative bioflux switches to positive values faster in TOPAZ than in
BGCCSM, most likely due to the earlier onset of biological production.
Finally, BGCCSM has a slightly earlier and stronger switch to negative
O2 bioflux at the end of the growing season than does TOPAZ.
We have O2 bioflux observations from four crossings in the South
Africa sector (Fig. ) distributed in time from December
to late March. The November–December transect has a pattern of weakly
positive O2 bioflux in the north that turns negative below
50∘ S. Both models show a somewhat different pattern with positive
O2 bioflux south of 50∘ S as well. While both models show
positive values, the observed January transect has predominantly negative
biofluxes. Finally, the two transects in March show a pattern with high
positive O2 bioflux in the north and significant negative bioflux at
55∘ S. BGCCSM has a better skill in recreating these patterns than
does TOPAZ.
Australia and New Zealand regions
Chl in the Australian and New Zealand sectors (Figs.
and ) follow the general pattern of Drake Passage and
South Africa. The main exception is BGCCSM generating spring blooms further
north and in a spottier pattern than in the sectors discussed earlier. In
this model, the New Zealand sector has four distinct areas/periods of high
Chl concentrations: in January–February south of 70∘ S,
November–December between 60 and 70∘ S, January between 45 and
55∘ S, and October–November north of 45∘ S. This peculiar
pattern could be due to interactions with hydrology in the frontal regions.
The Australian sector shows similar patterns for the latter three areas, whereas the region south of 70∘ S lacks data. TOPAZ shows indications
of being out of phase with both MODIS and BGCCSM after the beginning of
December, whereas BGCCSM captures the seasonal cycle in MODIS better.
NCP climatologies from the Australian sector (Fig. ,
top panel) show that TOPAZ generates short intense periods of positive NCP
early in the growing season. In the southern reach of the domain, NCP stays
high into the fall, whereas elsewhere summer and fall NCP values are low
(10 mmolm-2d-1). BGCCSM, on the other hand, has a much
longer and more intense period of positive NCP than in other parts of the
Southern Ocean. In this model, spring conditions south of 45–50∘ S
are dominated by weaker negative O2 biofluxes when compared to other
regions in both models (panel b). BGCCSM switches to negative bioflux at the
end of the summer, whereas O2 bioflux in TOPAZ stays positive
throughout March.
The observations in this part of the ocean differ from those simulated by
TOPAZ. For example, almost all measurements north of 55∘ S report
higher O2 biofluxes, >25 mmolm2d-1, in February.
These field data also exhibit low or even negative biofluxes at
65∘ S where TOPAZ predicts strong biological production and high
O2 bioflux from mid-November to mid-February. BGCCSM is more in line
with observed patterns, with the main exception of an early onset of negative
O2 bioflux in the fall at 40–50∘ S in the model where the
observations still suggest strong positive biofluxes. In general, between 50
and 60∘ S, TOPAZ and BGCCSM both simulate a strong spring bloom,
whereas observations show a simpler pattern of sustained high production
throughout the latter part of spring and summer. Model NCP and O2
bioflux in the New Zealand region are similar to values in the Australian
sector (Fig. ). The O2 bioflux observations in
panel c show considerable scatter but in general are marked by high values
north of the polar front at 60∘ S, with low or negative values of
bioflux to the south at most times. Both models capture the highly negative
values of bioflux early in the growing season south of 60–65∘ S and
occasionally negative values of bioflux later in the growing season. BGCCSM
simulates our observations of sustained high NCP north of the polar front
from November through February.
Cross-regional, Southern Ocean analysis
The data–model comparisons of O2 bioflux for the different regions
have certain patterns in common. In both models, the spring period of strong
positive flux starts later in high latitudes. In general, TOPAZ tends to have
an earlier, shorter, and more intense period of high Chl concentrations than
does BGCCSM. Observed values of O2 bioflux are much more variable
than simulated values. This is expected, given the relatively smooth and
coarse fields of the models, and likely reflects the absence of mesoscale
processes in the models.
Our next step is to compare the seasonal range in O2 bioflux values
between observations and models as a function of latitude
(Fig. ). We aggregate the observations to days of the year and
model grid cells as described above but compare the resulting values with
corresponding individual data points in the models matched by both location
and collection time. Such comparison using individual data points suffers
more from small-scale spatial mismatches than the zonal model averages used
earlier, but it allows us to better compare the seasonal range of O2
bioflux values between models and observations. Figure
shows O2 bioflux vs. latitude for the two models and observations in
each of the previously defined regions. We find that BGCCSM generally
predicts the meridional variability of ranges in O2 bioflux,
suggesting that processes constraining NCP are simulated well. The seasonal
maximum of model O2 bioflux might not occur at the same time in the
models as in the real world, but the magnitude of the maximum values seems to
be fairly well predicted. Equatorward of about 60∘ S, the models also
tend to capture range and meridional structure of negative O2
bioflux, which reflects both low wintertime NCP and physical transport. In
contrast, the models do not capture well the observed strong negative
O2 bioflux at high latitudes. TOPAZ also tends to exaggerate high
positive O2 bioflux in some areas, such as 60–70∘ S in the
Drake Passage and New Zealand regions, whereas O2 bioflux is
underestimated in TOPAZ between 40 and 50∘ S in the Australian and
New Zealand regions.
Scatterplots of observed (blue) and model (red) O2 bioflux
(mmolm-2d-1) versus latitude in four regions of the Southern
Ocean. All observations that fall on a specific grid cell at a specific year
day are averaged to one value. Models are subsampled at the location and year
day of the observations.
Scatterplots of model versus observed O2 bioflux
(mmolm-2d-1) for the observational sampling sites shown in
Fig. 1. All observations that fall on a specific grid cell at a specific day
of the year are averaged to one value. Models are subsampled at the location
and day of the year of the observations.
We compare observed and simulated O2 bioflux values for the same
locations and times. The scatterplot in Fig. compares
O2 bioflux simulated by TOPAZ (blue) and BGCSSM (red) with
observations binned to the same days of the year on the respective model's grid. It
is clear from this figure that both models show a low correlation with the
observations (r2=0.024 for TOPAZ and r2=0.23 for BGCCSM). This low
correlation is expected: lags in time or displacements in space can generate
large differences between the models and observations even if the fundamental
processes are simulated with high skill . Further, the
field data contain mesoscale variability that cannot be captured in the
models (even if the models were eddy-resolving, the details of the simulated
turbulent fields would differ from observed). It is also clear that the
distributions of model-data residuals (difference from a diagonal 1:1 line)
are asymmetrical. The upper-right quadrant, where both observations (obs) and
model data (mod) are positive, has considerable scatter about the 1:1 line
but no apparent bias. The lower-right quadrant (negative model, positive
observation) is mainly empty, showing that the model rarely predicts
undersaturation when the observations report supersaturation. An exception is
the clustering of TOPAZ NCP values around 0, a consequence of low
production simulated by that model after an intense bloom. The upper-left
quadrant is heavily populated, showing that the models frequently simulate
positive values of bioflux when the ocean is in fact undersaturated.
Consistent with this pattern, the results in the lower-left quadrant show
that, when data and models agree that bioflux is negative, the observations
are more negative than the models. The overall picture is that both models
have a positive bias in predicting bioflux, mainly due to fewer negative
values compared to the observations (Fig. ). For all data
points the model bias is statistically significant for both a paired t test
(BGCCSM: n=271, t=-4.803, p=0.000; TOPAZ: n=273, t=-2.870,
p=0.004) and a one-tailed binomial test (BGCCSM: pos =167, tot =271,
p=0.00005; TOPAZ: pos =151, tot =273, p=0.035). Neither model shows
a statistically significant bias when only positive values are considered.
Mixed layer depths
Finally, we create MLD climatologies for the different regions by integrating
Argo and model MLDs using methods mentioned earlier (Fig. ).
Both models are able to simulate the general trends rather well, with deep
mixed layers in winter, shoaling in spring and shallow mixed layers during
the summer. Regional structures are also similar in general, with the main
exception of deep winter mixing extending too far south in TOPAZ, even if
this is somewhat inconclusive due to lack of observations. The two main
differences between models and observations are that the spring shoaling tend
to be later and more gradual in the observations. There is also much
smaller-scale variability in the Argo climatology, which could be explained
by sparse data but also by the models having a mixed layer dynamics that are
too smooth.
Climatological Hofmøller plots of MLD (m) in different
regions of the Southern Ocean. All model values are zonal medians within the
box. All observations that fall on a specific grid cell on a specific year
day are averaged to one value.
Discussion
When comparing model and satellite climatologies of Chl concentrations and
O2 bioflux, we find both similarities and significant differences.
The models are able to predict spring and summertime maximum levels of Chl
and O2 bioflux well, but levels are much too low during the winter
and early spring. Such underestimations are particularly important in the
case of Chl, which by nature has a lognormal distribution
and is hence skewed towards low values. One
explanation for this behavior is a combination of model grazing and/or
phytoplankton mortality being too strong in the winter, as reported for the
BGCCSM in the subpolar North Atlantic by . These
patterns could also be explained by the two models simulating vertical export of phytoplankton that is too weak during summer and export that is too strong during
winter. have shown that MODIS/Aqua generally
underestimate the dynamical range of Chl in the Southern Ocean, which would
suggest that these differences might be even larger.
Another general pattern is that the models tend to simulate the increases in
Chl during spring and early summer rather well but show highly diverging
behavior later in the season. In TOPAZ, the ecosystem tends to crash in early
January with much too low biomass as an effect, whereas BGCCSM has patches
where far too much Chl is produced. Work by suggest
that such differences in skill over the season could be explained by changes
in processes that control the ecosystem. The spring bloom onset is thought to
be controlled mainly by physical factors such as light, temperature, and
vertical stratification, whereas the summer peak magnitude is thought to be controlled
by nutrient availability, grazing, mortality, and other ecological and
biogeochemical factors. Finally, the decrease in Chl concentrations after
a summer peak mainly depends on ecosystem dynamics such as grazing,
succession, and other interactions between organisms, as well as vertical
export of particulate organic matter. Phytoplankton production is in general
better resolved by the models, whereas heterotrophic processes and vertical
transports are challenging to implement well. These processes are often
stochastic in their behavior, and there is a lack of observation to
parameterize them accurately.
Panel (a) shows mixed-layer integrated NCP from the TOPAZ
model at 61.5∘ S, 159.5∘ W over a 400-day period.
Panel (b) shows two wind time series for the same time. The blue
line represents the NCEP/CORE reanalysis winds from the same location as the
model NCP data, and the red line represents Quickscat-satellite-derived winds
from the same location. The two lines in panel (c) represent the
resulting ΔO2/Ar supersaturation from a box model
simulation based on the time series in panels (a) and (b).
Panel (d) shows O2 bioflux calculated from the time series in
panels (b) and (c). The red line is based on reanalysis
winds and ΔO2/Ar, the blue line is based on Quickscat
winds and ΔO2/Ar, and the green line is based on
Quickscat winds and ΔO2/Ar calculated from reanalysis
winds.
With respect to models and observations, the misfit in spring Chl biomass is
likely explained by the differences in mixed layer dynamics between models
and observations. Consistent winter mixed layers that are too deep followed
by a rapid shoaling without much small-scale variability, as observed in the
models, would lead winter biomass that is too low and exaggerated spring
blooms, again as observed. Different factors, such as the timing and
progression of springtime mixed-layer shoaling, diapycnal mixing, and
mesoscale or submesoscale processes, can strongly affect the onset of the
spring bloom since the availability of light and nutrients in the mixed layer
will be impacted. Problems with MLD dynamics have been shown by earlier
studies in climate models similar to TOPAZ and BGCCSM e.g.,and
references therein.
Both models show significant skill in simulating how the ranges of O2
bioflux vary meridionally in the Southern Ocean north of 60∘ S. The
fact that the models provide such reasonable results suggests that the models
constrain ecosystem processes in the region rather well. The models show,
however, large regional variations that tend to follow the patterns discussed
earlier in Chl concentrations, both when compared with each other and with
observations. We also find different and varying patterns in the seasonal
cycle of biological net community production, with TOPAZ often having a
shorter, more intense growing season than BGCCSM. The fact that ecosystem
processes seem to be well simulated but the models still have regional and
timing issues could suggest problems with the physical model, such as how
well lateral currents are represented or how mixed-layer and thermocline
dynamics are parameterized.
One important difference between models and observations is that models fail
to predict observed events of negative O2 bioflux in waters south of
60∘ S. We suggest two possible explanations for why the models lack
these events: problems with the ecosystem dynamics and problems with the
vertical transport of oxygen. It is possible that observed summertime
undersaturation is generated by net heterotrophy (negative NCP) temporally
decoupled from earlier biological production. O2 supersaturation from
periods of positive NCP would then be lost via air–sea exchange before the
start of net respiration. It is also necessary for particulate organic carbon
to remain in the mixed layer for long enough to be respired, and hence particle
export has to be limited. Such events have been observed during the GasEx III
experiment but are not generated by either model. Both
TOPAZ and BGCCSM simulate NCP to evolve more smoothly than observations
across time and space, even if significant variability is seen
.
Our second explanation as to why models fail to capture summertime mixed
layer O2 undersaturation is that they underestimate rates and
characteristics of vertical mixing. Several studies have shown the potential
for entrainment and submesoscale processes to transport waters between the
thermocline and the mixed layer on short timescales. Such events would
introduce O2 undersaturated waters into the mixed lay and generate
the type of conditions we find in our observations. OGCMs such as TOPAZ and
BGCCSM lack the spatial resolution to resolve these kinds of processes and
hence the ability to generate the undersaturated conditions we observe.
We cannot conclusively disprove either explanation for the models' failure to
produce negative O2 bioflux in summer, but both the mismatches in MLD
dynamics and the specific patterns in O2 undersaturation strongly
suggest that vertical transport is the dominating factor. This explanation is
also supported by other studies, such as and
. Both find much stronger subduction rates south of
60∘ S when comparing high-resolution models to ones with lower
resolution. The areas of increased subduction described by Sallee and Rintoul
correspond very well with the latitude bands where we observe undersaturated
conditions not generated by the models. (Below 60∘ S for the New
Zealand, Australia, and Drake Passage sections; between 60 and 40∘ S
for the South African section.) Other studies furthermore suggest that such
physical processes captured by higher spatial resolutions tend to be
episodic in nature, making them prime candidates to generate the
undersaturated O2 conditions seen in our observations. One might
argue that the lower frequency of negative O2 bioflux in the models
is simply due to model overestimation of NCP. Such a conclusion is not
consistent with the models' accurate simulations of the seasonal range of
positive O2 bioflux (cf. Fig. ). The models appear
to selectively overestimate O2 bioflux when the observations of
O2 bioflux are negative.