Journal of Plankton Research Vol.19 no.12 pp.1829-1858, 1997
Sea surface temperature gradients, baroclinicity, and vegetation
gradients in the sea
William M.Balch, Bruce C.Bowler and Charles F.Byrne
Bigelow Laboratory for Ocean Sciences, PO Box 475, W. Boothbay Harbor,
ME 04575, USA
Abstract. We describe the results of an analysis of the 1984 CalCOFI data set, in which sea surface
temperature (SST) gradients were found to be well correlated to isopycnal slope (baroclinicity) at the
mesoscale. A link was established between the SST gradients and gradients of integral biomass [mg
chlorophyll (Chi) a m~2] and integral primary production (mg C m~2 day'). The water column photosynthetic efficiency, ¥ , was calculated by combining the CalCOFI data base with satellite-based estimates of the surface irradiance. It was found that V was rather stable, but not constant, for a wide
range of isopycnal slopes, in which integrated nitrate concentrations varied by up to 500 x. Variability in *¥ was most strongly dependent on calendar date, with the highest values in winter months and
the lowest values in the summer, which was probably due to an effect of day length or in-adiance. The
relationship between baroclinicity and SST gradients was highly scale dependent; from 0 to 200 km,
the data were randomly distributed, then as the length scales exceeded 400 km, the relationship
between baroclinicity and the SST gradient showed a significantly negative slope, with steadily
improving correlation coefficient. Such relationships for individual cruises generally showed
improved correlation over the generic annual relationship, which highlights the dynamic hydrography
of the region. Baroclinicity/SST gradient relationships also improved in waters of increasing depth.
A technique is described for calculating net integral biomass or production gradients from SST
imagery. The gradients in biomass and productivity showed increasingly strong ties to isopycnal slope
at larger size scales. Although this topic has long been described by oceanographers, this paper is the
first to demonstrate a quantitative coupling between baroclinicity and primary production gradients
as a function of length scale. Also significant is the observation that the relationships between baroclinicity versus gradients of integral chlorophyll or integral productivity are consistent over seasons
and over an area exceeding 1 million km2. We conclude that for satellite algorithms to account for
>50% of the variance in primary production, the nutrient term must be considered. Information on
baroclinicity is relevant to estimates of the along-isopycnal nutrientfluxto the euphotic zone. Merging
this with the more traditional light-based estimates of photosynthesis will improve algorithm performance since the methods utilize entirely independent sources of information.
Introduction
Geostrophy, eutrophication and primary production
It has long been understood that primary production is strongly impacted by
geostrophy (defined as the maintenance of pressure gradient fields by the combined effect of water flowing across a rotating geoid such that there is a balance
between Coriolis and gravitational forces). This is evidenced by the fact that the
global primary production maps of Steemann Nielsen (1954), Fleming (1957),
Ryther (1969), Koblents-Mishke et al. (1970), Leith and Whittaker (1975), Platt
and Subba Rao (1975), Berger et al. (1987) and Longhurst et al. (1995) depict the
centers of basin-scale cyclones as being eutrophic, while the centers of anticyclones are oligotrophic. The main reason for this relationship is that Ekman transport combined with geostrophy impacts the sloping of isopycnals from the geoid,
and this brings nutrient-rich water closer to the euphotic zone, allowing increased
rates of primary production. Moreover, the isopycnal slope affects the rate at
© Oxford University Press
1829
W.M.Btlch, B.CBowfer and CEByrne
which nutrient-rich seawater ascends to the euphotic zone. A large fraction of
marine productivity is considered to be nitrate limited; thus, the rate of alongisopycnal nutrient flow into the euphotic zone will affect new production, and the
subsequent sinking flux of particulate matter (Eppley and Peterson, 1979). Obviously, sloping of isopycnals alone is insufficient to move nutrient-laden water
upwards; there has to be some force to drive this motion (e.g. winds and Ekman
pumping). Since geostrophic balance is often maintained by winds blowing at
right angles to the sea surface slope, this forcing function is typically present.
When the wind dies, isopycnal slope will be maintained for a period of time by
inertial effects; under this inertial condition, nutrient flux (hence primary production) is probably more dependent on the depth of the mixed layer than on
along-isopycnal flow (Yentsch and Phinney, 1992).
The effects of geostrophy on phytoplankton standing stocks and productivity
have been described qualitatively before. As described in Yentsch (1993),
A.C.Redfield, after attending a seminar by C.G.Rossby at Woods Hole Oceanographic Institute, recognized the importance of 'wake stream theory' in the Gulf
Stream for enriching nutrients in slope waters and enhancing primary production.
In short, Redfield (1936; and references within) suggested that along the eastern
(right) edge of the Gulf Stream, water moves upwards and westward into a wake
stream, and is then ejected shoreward into the Gulf Stream counter-current. The
net effect is that nutrient-rich water from the deep Atlantic basin shoals at the
edge of the continental shelf, thereby enhancing productivity. Another connection between primary production and geostrophy was made by Sverdrup (1955),
who qualitatively predicted global patterns of primary production based on
knowledge of dynamic height. His production map had a remarkable resemblance to later global maps of productivity based on 14C measurements.
The importance of baroclinicity to algal standing stocks was demonstrated by
Yentsch (1974,1988), who focused on frontal zones in the Indian Ocean and the
equatorial Atlantic. He showed that integrated chlorophyll a biomass was
strongly related to the ocean density at 100 m (which is directly related to isopycnal slope). The relationships, though, varied between the SW monsoon Somali
Current and the Atlantic South Equatorial Current. This was probably due to
differences in the temperature-nitrate relationship and density-nitrate relationship.
Haury et al. (1993) showed that in the region of the Ensenada Front (California
Current), highest surface chlorophyll and surface primary production were
observed in regions of reduced geopotential anomalies (<5.4 m2 s~2), not regions
of low temperatures and positive divergence. Highest algal standing stock was
associated with high-velocity gradient areas between coastal jets, not the jet cores.
The fact that primary production was a linear function of the nitracline depth
(Haury et al, 1993; their Figure 14) implies that baroclinicity estimates should be
useful in the remote detection of gradients of primary production (the subject of
this paper). Note that the depth of the nitracline has previously been shown to
impact the percentage of chlorophyll in small particles (Herbland et al, 1985), as
well as the standing stock (g C m~3) and the production rate (g C m~3 day-1; Eppley
et al, 1979). Geostrophy combined with Ekman pumping probably creates the
1830
Vegetation gradients in the sea
tight coupling between nitracline depth and primary production, and also may
have been the reason that Steemann Nielsen and Jensen (1957) observed their
well-known relationship between euphotic zone depth and photosynthesis. More
recently, Sosik and Mitchell (1995) showed that the chlorophyll-specific absorption coefficient was low at the surface only if the nitracline and pycnocline sloped
towards the surface. Clearly, there is a profound dependence of primary production on baroclinicity, euphotic zone depth and nitracline depth.
Diapycnal mixing (mixing across isopycnals) and aeolian fertilization (dust
deposition) are important to mention in any discussion about mechanisms of
nutrient enrichment of euphotic phytoplankton populations. Diapycnal mixing
typically is forced by strong surface winds which homogenize the density structure in the surface layer, thus bringing up nutrient-rich water from below. An
example of this type of mixing is the vertical turbulent mixing of deeper, nutrient-rich water into the nutrient-depleted surface layers (Lewis era/., 1986). Note
that diapycnal mixing is secondarily affected by geostrophy, since the closer a
'nutrient-rich' isopycnal is to the surface, the less diapycnal mixing will be
required to move this water upwards.
In terms of aeolian fertilization, there are the so-called high-nitrate, lowchlorophyll regions (Cullen, 1991) in which a trace metal (iron?) is considered
limiting instead of nitrate. Fertilization of these waters may be occurring via
iron-rich dust deposition into surface waters (Martin and Fitzwater, 1988). Any
link between iron fertilization and phytoplankton and dynamic height variability would not be expected to be strong like the nitrogen/geostrophy link
where the nutrients are being supplied along isopycnals from below the
euphotic zone.
Primary production algorithms
In terms of algorithms for the remote sensing of primary production, the nutrient input term has been largely overlooked in favor of the photon flux terms. This
is because photon flux estimates are easier to quantify by satellite (e.g. estimating cloud cover and diffuse attenuation). Semi-analytical primary production estimates typically have involved chlorophyll biomass, an irradiance term, an
absorption term and a quantum efficiency term (Kiefer and Mitchell, 1983; Bidigare etal, 1992):
P = f\CM,Eita{\),a;*ma)
(1)
where C u l is the satellite-derived pigment, Ed or Eu is the down- or upwelling
irradiance, respectively, a* is the chlorophyll-specific absorption and <&„,„ is the
maximum quantum yield of photosynthesis. These models have been cast from
the most complex extreme (time, depth and wavelength varying) to moderately
complicated (time, depth and wavelength integrated) to the most simple (empirical relationships). Such models have accounted for 30-50% of the variance of
integrated primary production (Balch et ai, 1992; Behrenfeld and Falkowski,
1997; J.W.Campbell, personal communication). This is understandable given that
1831
W.M.Balch, B.CBowler and CF.Byme
the models employ no estimate of the nutrient input to the euphotic zone, so
important to phytoplankton growth.
Related to this question is whether isopycnal slope, due to its relationship to
nutrient upwelling, has any relationship to water column photosynthetic
efficiency, ^ (Falkowski, 1981), and whether ^ could be predicted and incorporated into a primary production algorithm. ^ is defined as:
¥ = X/7[XChl £d(0+)]
(2)
where JJ3 is integrated primary production (mg C m~2 day 1 ), XChl is integrated
chlorophyll a (mg Chi a m~2) and ^(O*) is the downwelling irradiance (E m~2
day 1 ) just above the surface. ^ has been attributed by some workers to be constant [0.43 g C m2 (g Chi E)"1] such that a plot of Z/VIChl versus £ d gives a
straight line. If this were true, the remote sensing ramifications would be profound, since knowledge of Ed and XChl (both of which can now be measured from
satellite) would then allow the calculation of integrated primary production.
Unfortunately, observations from some large data sets such as MARMAP (midto northern Atlantic Bight; Campbell and O'Reilly, 1987), SCBS (Southern California Bight Study; Balch, 1989), SE Atlantic Bight (Yoder et al., 1985) and
Antarctic LTER (Claustre et al., 1997) give highly variable ^ values, ranging from
0.27 to 2.2 g C m2 (g Chi E)"1.
The purpose of our research was to examine whether the slope of isopycnals
offers any new information on integral chlorophyll, photosynthesis and water
column photosynthetic efficiency, with the express purpose of enhancing the performance of primary production algorithms. Specifically, our approach has been
to use one of the most comprehensive hydrographic and biological data sets available, the CalCOFI data set, to (i) examine whether baroclinicity could be predicted based on sea-surface temperature gradients (SST), and then (ii) to see
whether this baroclinicity provided any biological information for predicting
gradients of integral biomass or primary production. This approach exploits satellite SST gradients for the determination of the potential for along-isopycnal
nitrate flux into the euphotic zone. Our previous work with the GEOSECS data
(Balch and Byrne, 1994) provided an indication of a strong relationship between
gradients in SST and the slope of isopycnals in the top 200 m. The relationship
was best at mesoscales, not at smaller scales. Defining this scale remains critical
for using SST data to augment our space-based primary production estimates. In
short, the CalCOFI data set offers an unprecedented opportunity to address these
issues.
Physical setting
There are four sources of water to the California Current (Reid et al, 1958). Subarctic water is supplied from the north, with low temperature and salinity, and
high oxygen. Central North Pacific water enters from the west, with warm temperatures and high salinities. Equatorial Pacific water enters the region from the
south, below the thermocline, with high temperatures and salinities, and low
1832
Vegetation gradients in the sea
oxygen. Finally, Subarctic and Equatorial Pacific water mix to produce the deeper
water. Northern subarctic water dominates the surface layer, while southern
waters dominate at depth (Reid et al, 1958; Jackson, 1988). It has long been
known that the generalflowof the California Current is southward (Sverdrup and
Fleming, 1941; Tibby, 1943; McEwen, 1948) with nearshore surface northwards
flow in the winter. The southward component is believed to be in geostrophic
balance, therefore giving rise to coastal upwelling starting in March or April, continuing through July or August. Nevertheless, it also is well known that the flow
is not simple; eddies have commonly been noted (Sverdrup and Fleming, 1941;
McEwen, 1948; Reid, 1988). For example, circulation in the Southern California
Bight is dominated by a counterclockwise eddy south of Point Conception, which
is fed by Southern California Bight water traveling northwards at -5 cm s'1
(Jackson, 1988). Haury et al. (1993), using drogues, noted the complex eddy-like
flow associated with the Ensenada Front. Obviously, the presence of eddies and
nearshore countercurrents complicates the analysis of baroclinicity, as calculated
from SST gradients, and will probably set the minimal spatial scales over which
strong relationships can be found between baroclinicity and SST gradients.
Method
The data used in these analyses were taken from the 1984 CalCOFI data set,
cruises 8401,8402,8404,8407 and 8410 (see Scripps Institution of Oceanography,
1984a,b,c,d,e for a complete description of cruises). The appealing aspects of
these data over the GEOSECS data were: (i) the full CalCOFI grid spanned from
mid-Baja, California, to San Francisco (-1100 km) and was sampled quarterly up
until 1984; (ii) the minimum station resolution was -25 km, which allowed elucidation down to frontal scales; (iii) there was a wealth of nutrient, pigment and
productivity data; (iv) there were concurrent AVHRR and CZCS data for comparison to the sea-truth data.
Each sampling of the CalCOFI grid was accomplished by two ships, each starting from opposing ends of the grid. The sampling grid typically contained 146
stations (of which 30 had primary production data) bounded by the following four
coordinates: 36°16.8'N X 126°29.1'W; 37°56.8'N X 122°52.9'W; 29°47.1'N X
116°1.5'W; 27°37.7'N X 120°16.9'W (Figure 1). Methods used for measuring
chlorophyll a, primary production, nutrients, temperature and salinity are
described elsewhere (Scripps Institution of Oceanography, 1984a,b,c,d,e;
Mantyla et al, 1995). All computations were performed on a DEC Alpha workstation. Data analyses were performed using Excel and Deltagraph software as
well as statistical techniques described in Zar (1974).
SST gradients (8557-) were calculated using either systematic or random
combinations of any two stations within the CalCOFI station grid, in which:
hsr = (.Trem-Tref)/D
(3)
where 7 ^ was the surface temperature at the remote site and Tnf was the surface
temperature at the reference site, and D was the distance between the stations. If
1833
WJH.Balch, B.CBowler and CF.Byrne
• ' • T V•
CALCOFI CRUISE 8401
4 - 2 7 JAHUAKY 1984
STATION
POSITIONS
DIRECTION OF THiVlL
O
•
MID—MUM
mnoa
Q cm ncOTOws
Q
»* MOM WTKI torn
DAVID STARR
JORDAN
-27 J«NUMY I M 4
NEW HORIZON
4 - J 3 JMUMY SO4
Fig. L CalCOFI sample grid for cruise 8401 (Scripps Institution of Oceanography, 1984a). Other
cruises during 1984 followed the same station plan.
1834
Vegetation gradients in the sea
the shallowest temperature at any station was from >10 m depth, then that
station's data were not considered to be representative of the surface and were not
included in the analysis. Systematic station comparisons were made by starting
with the station in the SE corner of the sample area as the reference station, and
comparing it to all other stations in the grid (remote stations), then using the next
station along the cruise track as the reference station, and comparing it to all other
stations, etc. At the end of this comparison, any duplicate comparisons were
removed. Non-systematic comparisons involved random geographical choice of
reference and remote stations, and deletion of any duplicate comparisons.
Baroclinicity was calculated based on the slope of a particular crT isopleth. Typically, the 25.5 <rT was used, provided it did not intersect the surface between two
stations. The baroclinicity of the 25.5 <rT isopycnal (BaT) was calculated as
follows:
e ( i T = Z ref -Z rem /Z)
(4)
where Zref was the depth of the 25.5 isopycnal at the reference station and Z rem
was the depth of the 25.5 isopycnal at the remote station.
The impact of spatial scales and water column depth on the baroclinicity-SST
gradient relationship was examined. It is important to note that at length scales
>400 km and depths of >800 m, the total number of comparisons fell precipitously,
and statistics became unreliable. Thus, for our analysis of the combined effects
of depth and inter-station distance, we present only correlations where the sample
size was £100.
To apply these concepts to imagery, it was necessary to calculate a net SST
gradient and net gradient direction at every pixel of an image. This was accomplished using standard circular histograms (Zar, 1974), and making each pixel the
center of a circle of diameter d (which represented the scale over which all gradients were calculated). Every 15° of the circle, a diameter was constructed (for a
total of 12 diameters per pixel) and the net gradient across each diameter calculated. Each diameter was decomposed into its north-south and east-west vectors,
which were then averaged to calculate the net magnitude and direction of the SST
gradient. If any of the diameters fell on clouds or land, they were eliminated from
the analysis; an analysis would be performed only if there were six or more diameters without cloud or land contamination.
Estimation of the water column photosynthetic efficiency, W
We have used the CalCOFI data set to calculate V and to examine whether its
variability was related to isopycnal slope. One measurement not made in the
CalCOFI program in 1984 was downwelling irradiance (Ed). To this end, we
acquired the International Satellite Cloud Climatology Project (ISCCP) data
base (Bishop and Rossow, 1991). Bishop and Rossow (1991) compared the predicted irradiance data to buoy data with good results. For 1984, we tabulated the
total solar, clear sky and predicted cloudy sky irradiances from the version 2
ISCCP data set from J.Bishop (University of British Columbia, Victoria) at daily
1835
WALBalcfa, B.CBowler and CF.Byme
intervals and 2.5° X 2.5° resolution. We converted these to visible photosynthetically active radiation (PAR) values and integrated them into the 1984
CalCOFI data base. Given access to £ d , U3 and XChl values from the CalCOFI
cruises, we then calculated ^ , and examined its variability in relation to temperature, baroclinicity (slope of isopycnals), time of year, water depth and integral
nitrate in the euphotic zone. It is also important to point out that we calculated
^ based on single-day values as well as 4 day means. This was to investigate the
impact of light history on the primary production values.
Results
Effect of varying length and depth scales, data randomization
The relationships between baroclinicity and SST gradients were statistically
highly significant, and the relationships improved with length scale. The resulting
baroclinicity versus SST gradient relationships showed increasing coefficients of
correlation at higher length scales. Systematic (non-randomized) station comparisons produced a bias in the results depending where in the station grid the
first reference station was chosen. For example, when station comparisons were
made in a non-random way, there was a higher probability that northerly stations
were colder than the southerly stations, and since more of the remote stations
were in the north, and more reference stations were in the south, then this produced a majority of negative 8557 values. Moreover, random station comparisons
for all the 1984 CalCOFI cruises produced more symmetrical Bay versus SJST
plots than were found in the systematic station comparison. Henceforth, only the
results from randomized station comparisons will be described.
Summary of pooled data
The 1984 cruise data were first pooled in order to describe the generic BaT versus
Sss-j- relationship for the California Current, i.e. a best estimate for all seasons
(Figure 2). At the smallest size scales (0-200 km), the data were randomly distributed with virtually no trends visible. At 200-400 km, there was a barely perceptive negative slope. At higher length scales (>400 km), a negative slope was
clearly apparent. Two distinct data populations emerged on each plot at size
scales >600 km, which were minor images (Figure 2D, E and F). Note, it is important to emphasize that these plots represent all cruises from 1984. This pooled
analysis was carried out to illustrate the generic relationship between baroclinicity and SST gradients, applicable over the entire year. Relationships are tighter
when one examines cruises individually. Distributions of these data showed that
there were non-zero X intercepts (stations with positive or negative SST gradients associated with BaT - 0), as well as non-zero Y intercepts (positive or negative baroclinicity associated with 8557 = 0). The two populations in Figures 2D, E
and F resulted from the random sampling strategy, and only such comparisons
will be described henceforth. Table I provides statistical summaries of the data in
Figure 2. Coefficients of determination for plots of baroclinicity versus SST gradients as a function of inter-station distance and station sounding depth are shown
1836
Vegetation gradients in the sea
200-400
400-600
800-1000
1000-10000
D
600-800
-3-2-1
-3-2-1
0
1 2
0
1
2
3-3-2-1
0
1
2
3
3
SST gradient (x 1 0 5 °C m"1)
Fig. 2. Plots of baroclinidty versus SST gradient taken from all CalCOFI data of 1984 (cruises 8401,
8402, 8404, 8407 and 8410). Station comparisons have been pooled according to distance: (A)
0-200 km; (B) 2(XM00 km; (C) 400-600 km; (D) 600-800 km; (E) 800-1000 km; (F) 1000-10 000 km.
Station comparisons were also chosen randomly within each cruise data set.
in Figure 3 (pooled cruise data for all of 1984). The data show that for 60 038
random station comparisons, the coefficient of determination increased from 0 to
1000 km, and accounted for -55% of the variance in isopycnal slope at the
1000 km size scale. For the summed data set, the r2 values in BaT versus 8557 plots
actually decreased at the greatest depths (Figure 3B). The histogram of the
Table L Statistical summary of the data in Figure 2. Data are for all CalCOFI cruises in 1984. Data
for the slope have units of [(vertical meter/horizontal meter) ("C"1)]. All data for intercept, intercept
standard error and predicted baroclinirity (Y) are in units of [(vertical meter/horizontal meter)
X 10"6]. The degrees of freedom (d.f.) and correlation coeffident (r) are also given
Distance
range (km)
Slope
(°C-i)
Intercept
x 10-6
SE intercept
X10- 6
SEY
X 10"3
d.f.
r
(°c-')
0-200
200^400
400-600
600-800
800-1000
1000-10 000
0-10 000
-3.18
-7.43
-6.93
-7.46
-9.18
-9.00
-5.91
0.223
0.165
0.171
0.162
0.167
0.237
0.089
-0.33
-4.19
-1.84
-2.85
-1.95
-0.257
-2.565
2.00
0.993
0.837
0.733
0.721
0.998
0.535
0.195
0.142
0.105
0.067
0.046
0.035
0.130
9503
20 609
15 695
8316
3999
1204
59 336
0.145
0.300
0308
0.449
0.656
0.738
0.264
SE
1837
WALBakti, B.CBow1er and CF.Bynie
A
0.5 H
{
0.4/
0 3no
1
0.1-
F= =1
/
0-
8
8 88 8
Interstation Distance (km)
0.6
c 0.5-
B
= 0.4
I
Q0.3
0J2
OillM|Mii|lln|iiM|,ni|,,,,|inl|iii?
Station Sounoing Depth (m)
10000-
c
8000600040002000-
0-
1
1 1
i
11 1i i i
•
•
rt ^
s
n
©
l
w
o5 o
I8§I§§I§i
^^
^~*
*"^
•""*
^ ^
fc-
*
t
Inter-statlon Distance (km)
1838
~*
jjj
Vegetation gradients in the sea
7
o
X 1-
to
c
CO
CD -2
-
2
-
1
0
1
5
2
1
SST gradient (x10" °C m" )
Fig. 4. Enlarged view of (E) in Figure 2 showing 'salinity offset'. The explanation for a measurable
baroclinicity with zero SST gradient, or zero baroclinicity with measurable SST gradient, is given in
the text.
number of comparisons shows how they declined at the smallest and largest size
scales (Figure 3C). Table I provides statistical summaries of the data in Figures 2
and 3. As in Figure 1, these plots represent the generic relationships for all cruises
from 1984. Relationships are often stronger when one examines cruises individually.
Interestingly, the distributions of the data in Figure 2 showed that there were
cases where there was measurable baroclinicity with zero SST gradient and vice
versa. Figure 4 shows this effect in detail. The explanation for this behavior will
be discussed later.
Individual cruise relationships
When analyzing each cruise individually, the relationship between the gradient
in SST and the slope of the 25.5 <TT isopycnal improved over the relationship
observed in the pooled data set. For example, in cruise 8401 at spatial scales of
200 km or larger, 40% of the variance in isopycnal slope could be accounted for
by SST gradient (Figures 5 and 6). Between 42 and 75% of the variance in the
slope of the 25.5 oT could be explained by the gradient in SST over 1000 km
length scales for all cruises. Unlike the pooled data set, the baroclinicity versus
SST gradient relationship also improved for specific cruises when comparisons
were confined to deep stations (Figure 6). It can be seen that during CalCOFI
cruise 8404, including stations with water depths >100 or >500 km allowed only
Fig. 3. Coefficient of determination for plots of baroclinicity versus SST gradients as a function of
(A) inter-station distance (km) and (B) station sounding depth (m). All estimates of baroclinicity
based on 25.5 a T slope. Stations were chosen from each CalCOFI cruise at random (arbitrarily called
'reference' station) and compared to a random 'remote' station within a pre-determined inter-station
distance or water depth. (C) Histogram of the number of station comparisons of baroclinicity versus
SST gradients as a function of inter-station distance.
1839
WJM.Balch, BX.BowJer and CF.Byme
200
400
600
800
1000
lnter-station distance (km)
Fig. 5. Values of r2 for baroclinicity (Baj = 25-5) versus SST gradient relationships for each calCOFI
cruise in 1984. Data are segregated by inter-station distance, with all water column depths pooled.
CalCOFI cruises are denoted as follows: • , 8401; • , 8402; A, 8404; • , 8407; D, 8410.
0
100
2 0 0 3 0 0 4 0 0 5 0 0 6 0 0 7 0 0 8 0 0
MINIMUM IrfTER-STATlON DISTANCE (Km)
Fig. 6. Values of fi for baroclinicity (fi,,T = 25S) versus SST gradient relationships for CalCOFI cruise
8404. Data are segregated by water cohimn depths with all inter-station distances pooled. In order to
have reliable statistics, the A values are only shown for each inter-station distance when there are 100
or more data points. Depth regimes are denoted as follows: O, >100 m; A, >500 m; O, >800 m; O,
>1000m.
1840
Vegetation gradients in the sea
-25-30% of the variance in baroclinicity to be explained by SST gradients. Confining the depth comparisons to stations having water depths >800 m allowed 40%
of the variance in baroclinicity to be explained. Finally, when only stations with
water depths >1000 m were used, as much as 75% of the variance in baroclinicity
could be explained at length scales of 400 km. It is important to note that when
all of the 1984 cruise data were pooled, the effect of water column depth was
equivocal (Table II).
Two-ship surveys
In all CalCOFI surveys during 1984, the station grid was sampled simultaneously
with two ships. This allowed comparison of BCTT versus 8SST plots for two-ship
comparisons only, where the data from each station comparison were binned
according to the differences in their sample times. As examples, we chose two
cruises, 8404 (which had a moderately good relationship between Bar versus SSST)
and 8407 (which had a poor relationship between BCTT versus 8SST). The binned
data showed that the correlation coefficient for survey 8404 increased as the
difference in sample times increased (Table III), while the data for survey 8407
showed equivocal results. A plot of the difference in sample time versus the distance between the stations shows why this occurred (Figure 7). Because of the
station pattern, when the difference in sample times was ~2 weeks, then the
length between the stations exceeded -300 km. Thus, while the analysis was
meant to include all interstation distances, it was biased to large distances when
the time difference between stations was large. The results of all two-ship comparisons show that for four of the five expeditions, confining station comparisons
to time periods of <1 week actually increased the coefficient of determination
(Figure 8). As noted above, the one exception was for survey 8404. In some cases,
Table n. Data summary for all 1984 CalCOFI survey cruises for the BCTJ versus &SST relationship.
The minimum and maximum distances and depths are given along with the least squares fit correlation
coefficients, slopes and intercepts (and associated SE terms)
Minimum Maximum Minimum Maximum n
dist
dist
depth
depth
(km)
(km)
(m)
(m)
r
Slope
Slope
error
Intercept Intercept
error
(X 10-6)
0
0
200
400
600
800
1000
1400
200
400
600
800
1000
1400
0
0
0
0
0
0
0
5000
5000
5000
5000
5000
5000
5000
59338
9505
20611
15 697
8318
4001
1206
-0.26
-0.14
-0.30
-0.31
-0.45
-0.66
-0.74
-5.91
-3.18
-7.43
-6.93
-7.46
-9.18
-9.00
0.089
0.222
0.165
0.171
0.162
0.167
0.237
-237
-0.33
-0.42
-1.84
-2.85
-1.95
-0.26
0.54
2.00
0.99
0.84
0.73
0.72
1.00
0
0
0
0
0
0
1400
1400
1400
1400
1400
1400
0
200
400
600
800
1000
200
400
600
800
1000
5000
7172
3150
16006
10843
2907
19260
-0.39
-0.40
-0.22
-0.29
-0.11
-0.23
-8.24
-8.39
^t.85
-6.90
-1.83
-5.57
0.229
0342
0.171
0.219
0301
0.169
-3.20
-28.09
-5.03
-2.97
13.11
1.191
1.62
2.14
1.03
1.17
2.24
0.97
1841
W.M.Balch, RCBowier and CF.Byrne
TaWe HL Change in the correlation coefficient of baroclinidty versus SST gradient relationships as a
function of time between randomly chosen stations. Data from all two-ship surveys, all depths and
distances are included. The time range used in binning the data is given as the minimum and maximum
times (days) used in each correlation. In all cases, the baroclinicity was calculated using the slope of
the 25J>(rT. For each cruise, times are binned in decreasing then increasing fashion
Minimum time
Maximum time
Cruise 8404
0
0
0
0
0
0
0
0
22
20
10
5
4
3
2
1
-0.326
-0.326
-0303
-0.286
-0.277
-0.268
-0.270
-0.165
8880
8845
6366
3433
2729
1964
1161
330
1
2
3
4
5
10
15
20
22
22
22
22
22
22
22
22
-0.333
-0.340
-0.352
-0.361
-0.369
-0.438
-0.530
-0.476
8550
7719
6916
6151
5447
2514
739
35
0
0
0
0
0
0
0
0
0
23
20
15
10
5
4
3
2
1
-0.003
-0.003
-0.001
+0.031
+0.078
+0.108
+0.084
+0.070
+0.103
13 068
12 854
11749
9115
4763
3694
2638
1550
471
1
2
3
4
5
10
15
20
23
23
23
23
23
23
23
23
-0.010
-0.020
-0.040
-0.077
-0.082
-0.172
-0.134
-0.356
12597
11518
10430
9374
8305
3953
1319
214
the coefficient of determination doubled by confining the binned times to <1 week
(e.g. surveys 8402 and 8410). The BaT versus ossr plots were examined for each
ship of each cruise individually (data not shown) to verify that there were no statistical biases associated with the data from each ship.
Fig. 7. Examples of the difference in station sample time versus the distance between the stations for
cruises (A) 8404 and (B) 8407. Data are shown for comparisons where two stations were sampled by
different ships. Results show that for the baroclinidty calculations, the largest range in spatial scales
sampled corresponded to the smallest time between stations. The largest amount of time between
stations was associated with distances of -600-800 km.
1842
Vegetation gradients in the sea
200
400
600
200
400
600
800
800
1000
1000
1200
1400
1200
Distance (km)
1843
W.M.Bakh, B.C.Bowier and CF.Byroe
C\J
8401
8402
8404
8407
8410
Cruise Number
Fig. 8. Coefficient of determinations of baroclinicity versus SST gradient for all two-ship comparisons. For each cruise period, the left-most bar of the histogram provides the r2 value for all times
sampled, the middle bar is for station comparisons with <1 week and the right bar is for station comparisons with >1 week between each two-ship comparison. Data show that for four of the five expeditions, confining station comparisons to time periods of <1 week increased the coefficient of
determination.
SST gradients and vegetation gradients
Gradients in integral chlorophyll were significantly correlated with SST gradients,
and, as with the baroclinicity versus SST gradient results above, the integral
chlorophyll gradient versus SST gradient correlations improved with increasing
inter-station distance (Figure 9A). This trend was not observed for every cruise;
during cruise 8407, the coefficient of determination between the integral chlorophyll gradient versus SST gradient never exceeded 0.15. The opposite extreme
was demonstrated during cruise 8410, in which almost 60% of the variance in the
gradient of integrated chlorophyll could be accounted for by the SST gradient at
a 600 km size scale.
Even more noteworthy was the fact that the integral productivity gradient was
better correlated to the SST gradient than the integral chlorophyll gradient
(Figure 9B). For example, while the integral chlorophyll gradient versus 8557 data
for cruise 8407 showed r2 values of <0.1 at the 600 km length scale, the integral
primary production gradient versus 8557- relationship had an r2 of 0.45 at the same
length scale. While r2 values for plots of integral chlorophyll gradient versus Sssr
improved at increasing size scale, r2 values between integral primary production
Fig. 9. (A) Values of r2 for plots of integral chlorophyll gradient versus SST gradient for each
CalCOFl cruise in 1984. One can see improvements in the coefficients of determination between SST
gradients and gradients in integral chlorophyll at larger spatial scales. (B) Values of r2 for plots of
integral productivity gradient versus SST gradient for each CalCOFl cruise in 1984. For both panels,
data were segregated by interstation distance, with all water column depths pooled. The key for each
cruise is shown on the figure.
1844
Vegetation gradients in the sea
(A)
0.8-
0.5-
0.4-
/'
i0.3-
02-
. . - •
0.1-
/•
~
100
(B)
a-
200
300
400
500
600
0.7
200
300
400
Intsr-SWloo Distance (km)
500
600
1845
W.M.Bakh, BX-Bowier and CF.Bynie
gradient versus 8SST showed even more improvement with increasing size scale,
and the trends were more consistent over all cruises (compare Figure 9A and B).
Variability of V in the CalCOFl data base
Given access to Ed, LP and ZChl values for the CalCOFl cruises, we then calculated ^ , and examined its variability in relation to temperature, baroclinicity
(slope of isopycnals), time of year, water depth and integral nitrate in the euphotic
zone. It is important to point out that we calculated ¥ based on both single-day
values as well as 4 day means. This was to investigate the impact of light history
on the primary production values. The results indicated that, as expected from
previously published results, the relationship between £/7XChl and Ed accounted
for only -9% of the variance in the California Current, reaffirming that *P is not
constant. Moreover, ¥ had no relationship to baroclinicity. Obviously, W will be
of little use for enhancing algorithm performance, given that so much of the variance is left unexplained. Nevertheless, the next question was to define the statistics of ¥ , to document its dynamic range, relative to the range of biomass and
production observed in the region. Interestingly, over the entire CalCOFl grid,
integral nitrate nitrogen varied by almost 500X, integral production varied 10X,
integral chlorophyll biomass varied 6x and ¥ only varied ~3x.
The envelope of ¥ variability showed the highest values at low SST (and largest
range, too). As temperature increased, the ¥ values fell, as did their range (data
not shown). This was probably due to the highly non-linear decrease in nitrate
concentrations at wanner temperatures. Within any single cruise, the relationship
between V and SST improved, accounting for up to 58% of the total variance.
While the SST data suggested some influence on "ty, baroclinicity had virtually no
impact on ^ , implying no strong relationship between ^ and along-isopycnal
mixing of nutrients. Variance of W was most dependent on the time of year, but
again, the error limits were large (Figure 10). The pooled cruise data showed that
the highest values were in the winter months and the lowest values were seen in
the summer.
Application of results to imagery
The results of the gradient analysis calculations applied to SST images can be
seen in Figure 11. This shows an original AVHRR image (84024) with net SST
gradient vectors drawn every 25 pixels. It can be seen that there is relatively good
coherence between the various vectors of the image in terms of their net direction within the California Current. Note, at 200 km length scale (Figure 11 A),
SST gradients explained -42% of the baroclinicity in the top 200 m (Figure 5),
provided the station depth was >1000 m (Figure 11). At the 200 km size scale,
only 23 and 35% of the variance in integral chlorophyll or integral primary production gradients could be predicted from SST gradients, respectively. Therefore,
we increased the size scale to 400 km (Figure 11B). While SST gradients did not
explain much more variance in baroclinicity at this scale (44%), the information
about integrated chlorophyll and production was significantly enhanced (SST
1846
Vegetation gradients in the sea
c
iu
*L3
O
O
50
100
150
200
250
300
350
Calendar day
Fig. 10. Water column photosynthetic efficiency (V) plotted against calendar day. Values are based
on CalCOFI data for integrated primary production and biomass, and the ISCCP irradiance data from
J.Bishop (University of British Columbia, Victoria), light data are tabulated at daily intervals and
2.5° x 25° resolution. Results show that the highest •*• values (with the highest range) are seen in the
winter months, and the lowest ¥ values are observed in April. Subsequent months show increasing
y values with increasing range into winter. Dashed lines indicate the upper and lower range of the
bulk of data.
gradients explained 42% of the variance in the gradient of integrated chlorophyll
and 55% of the variance in the gradient of integrated primary production). The
direction and magnitude of the net vectors in the 200 km image show good coherence with the 400 km image, although some differences are visible. As an indication of this coherence, the mean angle for 8SST at all the ocean pixels in the
200 km image was 135.95 ± 71.03° (n = 104 302) and the mean angle in the 400 km
image was 129.22 ± 42.25° (n = 63 652). At the 600 km scale (data not shown),
the angles are even more spatially coherent with a mean angle of 131.61 ± 16.38°
{n = 22 851). The scale bar on the image indicates the SST gradient, the resultant
baroclinicity, integral chlorophyll gradient or integral primary productivity
gradient (Figure 11). Note, these results apply to the top 200 m, not just the top
optical depth of the water column; thus, we are statistically recovering information about primary production to depths completely invisible to CZCS or
SeaWiFS.
Discussion
The baroclinicity versus SST gradient relationships described here for the California Current (slope = -3 to -9 "C"1 and intercept = -0.33 X 10"6 to -2.85 X 10"6;
Table II) were markedly different from the global GEOSECS relationships
described earlier by Balch and Byrne (1994; slope = -24.4 "C"1 and intercept of
17.5 X 10"6; n = 47; r2 = 0.82). Recall, there were far fewer stations in the
1847
WJM.Bakh, RCBowler and CF.Byrne
Fig. 11 (A) Results of gradient analysis applied to AVHRR SST image 84024 (24 January 1984).
Each line represents a net gradient vector calculated from a circular histogram consisting of 12 individual gradients taken sequentially at 30° increments around the pixel of interest. The angle and
magnitude of each gradient are shown every 25 pixels. All gradients are calculated at 200 km size
scale. Note coherence in the angle and magnitude. The scale bar represents 0.01°C km"1, and also
shows equivalent parameters, r2 values are given to show how much of their variance is accounted for
by the gradient in SST. (B) As (A), except that calculations were performed at 400 km length scales.
Note the similarity in gradient magnitudes and directions between this panel and (A) (based on
200 km length scales). Details of calculations are given in the text.
GEOSECS data, only larger scales were sampled (200-500 km between
stations), but much larger hydrographic features were crossed. The latter aspect
might have caused the differences that we observed here. Ironically, temperature gradients measured across the Ensenada Front in the California Current by
Haury et al. (1993; their Table I) were also at the upper range of what we
observed in the 1984 CalCOFI data set. The high values of Haury et al. (1993)
were probably due to the fact that their sample grid was more highly resolved
1848
Vegetation gradients in the sea
than the standard CalCOFI grid so that sharp frontal gradients were more adequately sampled horizontally.
Our new CalCOFI results are of interest for two reasons, (i) The smallest
spatial scales in the GEOSECS data were several hundred kilometers, whereas
the CalCOFI data covered a much larger range of spatial scales and demonstrated
how the baroclinicity versus SST gradient relationships improved with size scale.
There are few areas of the world where biological and physical gradients are
sampled as well as in the California Current, (ii) While the GEOSECS data were
global in coverage, biological data were minimally sampled. The new CalCOFI
results point to strong relationships between SST gradients and gradients in integrated production and chlorophyll at larger spatial scales.
1849
WJW.Baldi, B.CBowler and CF.Byrne
Factors affecting the baroclinicity/SST relationships
Plots of baroclinicity versus SST showed points where there was measurable
baroclinicity with zero SST gradient. This was due to the fact that
temperature-salinity plots from the top 200 m of the California Current showed
a statistical variance in salinity (0.55 PSU) at a given temperature (Churgin and
Halminski, 1974). Thus, it would be possible to have a salinity increase or
decrease between stations of 0.55 PSU with no change in SST. The effect of this
would be to cause a gradient in density of -0.4CTTunits (and concurrent change
in baroclinicity), with zero gradient in SST. The converse situation, stations with
zero baroclinicity yet a measurable SST gradient, was also observed. These could
also have been due to the counteracting effects of temperature and salinity
observed in this eastern boundary current region. As mentioned earlier, the California Current is cold and fresh in the north, and warm and salty in the south.
Therefore, the effect of increasing SST towards the south is countered by increasing salinity, which tends to counteract a change in density (and baroclinicity).
The time scales associated with these analyses were between 0 and 24 days,
given that the two ships involved in each CalCOFI cruise took that period to
complete the station grid. This means that the link between baroclinicity and SST
gradients described here is relevant to time scales of about a month. Moreover,
the fact that the California Current flows with a velocity of 0.1 m s"1 to the SE
(relative to the 1000 decibar surface; Reid, 1961), and that it is in geostrophic
balance, means that isopycnals will shoal towards the coast. This shoaling would
be expected mostly in the top 300 m because only very weak geostrophic currents
have been observed at deeper depths (Sverdrup and Fleming, 1941; Barcelona et
ai, 1982). The SE flow of the California Current would have been the primary
factor driving the onshore-offshore baroclinicity changes, but this would have
been affected by the seasonally varying counter-current. It is well known that the
California Current forms a coastal counter-current around 32°N (Sverdrup and
Fleming, 1941; Hickey, 1979; Lynn and Simpson, 1987), which becomes the Southern California Counter-current (Jackson, 1986). The region where the California
Current turns onshore is the principal cause of the Ensenada Front (Haury et ai,
1993); the northward-flowing surface counter-current is most pronounced in
winter and nearshore flow is predominantly southward in the summer (Reid,
1988). The worst baroclinicity/SST gradient relationships occurred during July,
and the best were during October and January. It remains speculative as to
whether the strength of the baroclinicity versus SST gradient relationships is
related to the strength of the counter-current.
Biological impact of baroclinicity
While this paper is the first to demonstrate a quantitative coupling between baroclinicity and gradients of integrated primary production, the topic has long been
discussed by oceanographers (see Introduction). What is unique in this study is
that the relationship between baroclinicity versus gradients in integral chlorophyll or productivity is reasonably consistent over seasons and over an area the
1850
Vegetation gradients in the sea
size of the CalCOFI station grid (>1 million km2). Clearly, nitrate flux was the
connecting factor between isopycnal slope and productivity; this became obvious
in a plot of integral production versus the depth of the 25.5CTT(Figure 12). During
all the 1984 CalCOFI cruises, as the pycnocline shoaled, there was an exponential increase in integral productivity. This is virtually identical to the patterns
observed by Yentsch (1974, 1988) for integral chlorophyll versus the density at
100 m, but here we extend this relationship to production. The most obvious
explanation for this was that the photon flux increased exponentially at shallower
depths. Balanced growth would have required both light and nutrients; as the
nitracline shoaled into an exponentially increasing light field, integral production
would also have increased exponentially (provided that nitrate did not become
growth limiting). Figure 12 also suggests that diapycnal and along-isopycnal
mixing of nitrate-rich water was sufficient to sustain higher productivity levels
when the pycnocline shoaled. Of course, factors like photoinhibition of shallow
20
• •
40
• .
60-
H
A
T
80-|
CO 100-1
"5
Q. 120-
140-
160-
180
500
1000
1600
2000
Int. Production (mg m
2600
2
3000
3600
1
tf )
Fig. 12. Plot of the depth of the 25.5 <TT versus the integral production rate for all CalCOFI cruises
of 1984. • , 8401; • , 8402; A, 8404; • , 8407; T, 8410.
1851
WJVi.Btlch, aCBowter and CF.Bymc
phytoplankton populations would also have affected the production estimates,
but they were probably of secondary importance in determining water column
production.
The nitrate-temperature relationship in the California Current shows no
nitrate above ~15-20°C (Kamykowski and Zentara, 1986); thus, one would expect
a priori that productivity gradients would be reduced in warm waters, where the
horizontal gradient of integral nitrate was also reduced. Obviously, the reason
that productivity gradients were still observed in warmer portions of the California Current (where surface nitrate was completely depleted) was that there
was still nitrate at the base of the euphotic zone to allow for overall balanced
growth. Moreover, the system probably self-regulated. That is, if baroclinicity was
reduced, then less integral nitrate would have been available in the euphotic zone,
the integral primary production would have decreased, and zooplankton grazing
and/or sinking would have diminished the remaining algal biomass. This, in turn,
would have allowed greater penetration of light so that deeper algal populations
(closer to the nitrate supply) could grow. The key to this interpretation is the
focus on integrated production, as opposed to production in the top optical depth
visible to satellites.
Most of the variance in Z/7XChl was unexplained by daily integrated PAR, and
water column photosynthetic efficiency (^) showed no relationship to baroclinicity. Nonetheless, the variance of water column photosynthetic efficiency was
within a factor of 3X, in the face of orders of magnitude of biological and chemical variability. The quasi-stability in ^ results from the acclimation of the phytoplankton assemblage to ambient light and nutrient conditions. Species succession
may also have occurred in the transition to oligotrophic conditions, such that lightharvesting efficiencies remained constant or increased (due to the package effect;
see Morel and Bricaud, 1981) while the algal biomass and production decreased.
Such variability might be expected in view of day length changes (the full
CalCOFI grid extends to ~42°N which would have annual changes in day length
of -85%), changes in C:Chl ratios (from -20-200 or 1000%; Steele and Baird,
1962; Eppley et al., 1977), changes in chlorophyll-specific absorption (a*; 300%
variability at different light-limited growth rates; Mitchell and Kiefer, 1988).
These ¥ estimates are higher, and more variable, than those of Falkowski
(1981) and Platt (1986). The high values were more similar to those observed by
Yoder et al. (1985) and Campbell and O'Reilly (1988). ¥ showed relatively little
systematic variance with distance offshore, while integral production, chlorophyll
and nitrate showed strong offshore gradients. The 3-fold variability in ^ , with the
highest values in the late fall and winter, was similar to the pattern observed by
Campbell and O'Reilly (1988; highest range in the early winter), but they
described overall variability of 6-8 X. Claustre et al. (1997) measured photosynthetic efficiencies at the Long-Term Ecological Research site near Palmer Station,
Antarctica, from 1991 to 1994. They measured the water column photosynthetic
cross-section, ¥*, which is essentially identical to ¥ except that the photosynthesis in equation (2) is converted to equivalent chemical energy stored in photosynthesis by multiplying by 39 kJ (g C fixed)"1. They, too, observed that ^* was
~6x higher in the southern hemisphere winter than the summer. They argued
1852
Vegetation gradients in the sea
o
en 1000-
Us-
I
Q.
1
CD
|
+
1001
r
i-
10
100
1000
Integral chlorophyll (mg Chi m"2)
Fig. 13. Plot of integral primary production versus integral chlorophyll for all CalCOFI stations occupied during 1984. The line represents a least-squares fit, and is represented by the following equation:
log integral production = log integral chlorophyll x 1.049 + 1.37. The r2 of this fit is 0.77 and is based
on a sample size of stations 133.
that this was associated with variability in seasonal irradiance and also taxonomic
dependency (e.g. due to differences in quantities of photoprotective pigments).
^P* increased when diatoms dominated the assemblage, as opposed to cryptophytes. We agree with the above scenario. It should nevertheless be emphasized
that even with all the variability in ty, there was still strong covariance between
integral chlorophyll and integral production (Figure 13). In terms of an algorithm,
an average function needs to be derived which can be used as a way to estimate
^ roughly in temperate and tropical regions. Claustre et al. (1997) provide such
a relationship for the Antarctic, but temperate and tropical regions appear to
suffer from higher variability in ty. The best application for using "V data in temperate and tropical regions is as a constraint to verify that predicted production
values, when taken along with integral biomass and light, produce ^ values within
the expected factor of three.
It was completely unexpected that baroclinicity would be more useful in predicting gradients in integral primary production than for predicting gradients in
integral chlorophyll. This is especially true when one considers the errors inherent in the respective methods. A possible reason for this is that baroclinicity
directly impacts the rate at which nitrate is mixed upwards into the euphotic zone.
Along-isopycnal mixing simultaneously enriches surface, nutrient-poor water,
causing faster growth while simultaneously removing algal cells. The model is
analogous to how the dilution rate of a chemostat directly controls the growth
rate of the phytoplankton. As long as the inflowing water has a nitrate concentration greater than the nitrate half-saturation coefficient, then the nutrients in
the reactor will be used up, and the growth rate will be a direct function of nutrient input, while the biomass remains constant over the dilution rate (except at
washout). In other words, our interpretation is that integral biomass gradients
should be less variable than integral primary production gradients as a function
of isopycnal slope (nutrient supply).
1853
WJH.Bakh, B.CBowier and CF.Byrne
It seems likely that grazing would also have affected the vegetation gradients.
One scenario is that grazing removed accumulated standing stock, but this, in
turn, would allow nutrients to penetrate further into the euphotic zone, thus
maintaining primary production gradients that correlate with 8557-. Moreover,
recycled production fueled by the grazing might maintain the gradients of integral
primary production, even after the gradient in standing stock had been destroyed.
The fact that the relationships of gradients in SST or integral biological fields
versus baroclinicity improved with distance simply reaffirms the importance of
large-scale physics in driving biological patterns in the sea (Sverdrup, 1955). The
interaction that we observed, however, adds support to the Stommel diagram for
biological pattern, published previously by Haury et al. (1978). This diagram
changed how we view biological patterns in the sea, in that it (i) defined biological
variability as a function of space and time, and (ii) directly connected the time
and space scales of biological phenomena to the relevant physical time and space
scales. The results presented here quantify how the connection (i.e correlation)
between physics and biology in the sea improves with increasing spatial scales.
This provides insight as to why pixel-by-pixel estimates of integral primary production rarely account for even a majority of the variance in sea-truth data (Balch
et al., 1992). J.W.Campbell (personal communication) has shown that in a comparison of 15 different empirical and semi-analytical primary production algorithms, a maximum of -45% of the variance in integral primary production could
be accounted for by the models. These are primarily based on light input terms,
whereas, as cited earlier, neglect of nutrients in attempts to model plant primary
production can be devastating to the algorithm's predictive ability. [See also the
discussion of Sosik and Mitchell (1995) in which they show that the depth of the
nitracline affects the chlorophyll-specific absorption coefficient.] By incorporating information on the large-scale vegetation gradients of primary production, we
hope to bridge the gap between the short-term response of phytoplankton to light
versus the large-scale response of phytoplankton to nutrients. There remain some
distinct limitations and advantages to the baroclinicity approach.
Limitations of the baroclinicity approach
Our CalCOFI results suggest that the SST-baroclinicity relationships are
strongest in deep waters, where topographical effects are minimized, but that the
topographical effects typically accounted for <10% of the explained variance.
The SST-baroclinicity relationships have been successfully applied to the California Current, which (i) is an eastern boundary current, where gradients are not
as strong as in a western boundary current, (ii) has a very narrow continental
shelf, so topographical effects are minimized, and (iii) has a relatively straight
coastline. The approach has yet to be applied to a western boundary current
region such as off the continental USA. It would be expected that in regions such
as Georges Bank, SST-baroclinicity relationships would be significantly modified, as would baroclinicity/primary productivity relationships. For example, the
major hydrographic boundary associated with Georges Bank occurs at the 60 m
isobath (Holligan et al., 1984; Flagg, 1987). The typical slope of the isopycnals
1854
Vegetation gradients in the sea
crossing this front is 2.5 X 10~3 vertical meters per horizontal meter, which is five
times bigger than the largest baroclinicity observed in the California Current!
Given that there is an increased interest in describing productivity in such coastal
margin zones, it is essential to test the technique in more complicated environments, such as the Gulf of Maine, if it is to be of use in remote sensing.
Another limitation is related to error propagation in the calculations. For
example, using AVHRR imagery, the RMS error for measuring SST in surface
waters is about ±0.4-0.6°C (or a relative error of <5% at temperatures of 15°C).
Assuming that distance can be measured to within one pixel on an image, then
the distance error is ~1 km (or a maximum relative error of -0.5% at the 200 km
length scale). Using standard error propagation, the relative error in an SST
gradient determination should be -6%. In practice, we compared the SST
measurements with sea-truth values from the same time period, and found errors
of >2°C. We had insufficient data to determine whether the error was consistent
throughout the image, which would allow us to determine SST gradients with
better accuracy. The linear model relating SST gradient to the gradient in integral
production had an r2 = 0.55 at length scales of 400 km. Nevertheless, the relative
error of the slope of this relationship was 80%; such an error would propagate to
the final estimate of the productivity gradient. As already stated, since the relative error decreases as the size scale increases, then the baroclinicity approach
will have the most relevance at large size scales.
Advantages of the baroclinicity approach
The statistical approach described here provides baroclinicity estimates down to
depths of 100-150 m, much deeper than the depths observed by ocean color
sensors. This is essential for understanding the nutrient influx into the bottom of
the euphotic zone. A bonus of this approach is that it yields information on gradients of primary production, such that with SST information on either side of a
cloud-obstructed area, it is possible to predict statistically the productivity gradient under the cloudy area. Given the horizontal scale dependence, another
advantage is that baroclinicity information provides improved production estimates over larger basin and mesoscale areas rather than on a pixel-by-pixel basis.
This is important for enhancing regional productivity extrapolations. A technique which melds the more traditional pixel-by-pixel productivity prediction
with the regional 'baroclinicity' approach described here will probably be the
best way to increase our knowledge of vegetation gradients in the sea, as well as
increasing the explained variance of integral primary production, as measured
from space.
Acknowledgements
Jim Bishop kindly provided us with the ISCCP irradiance data to match with the
CalCOFI station data. Klaus Ruth (University of Miami) provided stimulating
discussion on mixing and phytoplankton growth. Support for this work was kindly
provided by NASA (NAGW 2426; NAS5-31363), ONR (N00014-91-J-1048) and
1855
W.M.Balch, RCBowler and CF.Byrae
NSF (OCE-9596167) to W.M.B. We dedicate this paper to C.S.Yentsch, whose
earlier discussions and papers stimulated the research presented here. This is
Bigelow Laboratory contribution number 97006.
References
Balch.W.M. and Byrne,C.F. (1994) Factors affecting the estimate of primary production from space.
/. Gcophys. Res. Oceans, 99, 7555-7570.
Balch.W.M., Eppley,W. and Abbott, M. (1989) Remote sensing of primary production II. Introduction of semi-analytical algorithm. Deep Sea Res., 36, 1201-1217.
Balch.W.M., Evans,R., BrownJ., Feldman.G., McClain.C. and Esais.W. (1992) The remote sensing of
ocean primary productivity—use of a new data compilation to test satellite algorithms. /. Geophys.
Res., 97, 2279-2293.
BarcelonaJvU., Cummings.L.C, Lieberman,S.H., Fastenau,H.S. and North.WJ. (1982) Marine
fanning the coastal zone: Chemical and hydrographic considerations. Calif. Coop. Oceanic Fish.
Invest. Rep., 23,180-187.
BehrenfeldjrfJ. and Falkowslci.P.G. (1997) Photosynthetic rates derived from satellite-based chlorophyll concentration. Limnol. Oceanogr., in press.
Berger.W.H. (1989) Global Maps of Ocean Productivity. In Berger.W.H., Smetacek,V.S. and Wefer.G.
(eds), Productivity of the Ocean: Present and Past, John Wiley and Sons Limited, pp. 429-455.
Bidigare^R.R., Prezelin.B.B. and Smith^R.C. (1992) Bio-optical models and the problems of scaling.
In Falkowski,P.G. and Woodhead^A.D. (eds), Primary Productivity and Biogeochemical Cycles in
the Sea. Plenum Press, New York, pp. 175-212.
BishopJ.K.B. and Rossow.W.B. (1991) Spatial and temporal variability of global surface solar irradiance. / Geophys. Res., 96,16839-16858.
CampbellJ.W. and O'ReillyJ.E. (1988) Role of satellites in estimating primary productivity on the
northwest Atlantic continental shelf. Com Shelf. Res., 8,179-204.
ChurginJ. and Halminski,SJ. (1974) Temperature, Salinity, Oxygen and Phosphate in Waters Off
United States—VoL III. Eastern North Pacific Key to Oceanographic Records Documentation No.
2. National Oceanographic Data Center, NOAA, Washington, DC.
Claustre,H., Moline,M.A. and Prezelin3-B. (1997) Sources of variability in the column photosynthetic cross section for Antarctic coastal waters. /. Geophys. Res., in press.
CullenJJ. (1991) Hypotheses to explain high-nutrient conditions in the open sea. In Chisholm,S.W.
and Morel .F.M.M. (eds). What controls phytoplankton production in nutrient-rich areas of the open
sea? UmnoL Oceanogr., 36,1507-1965.
Eppley.R.W., Harrison.W.G., Chisholm.S. and StewartJE. (1977) Paniculate organic matter in surface
waters off southern California and its relationship to phytoplankton. /. Mar. Res., 35, 671-696.
Eppley.R.W. and Peterson^- (1979) Paniculate organic matter flux and planktonic new production
in the deep ocean. Nature, 282, 677-680.
Eppley.R.W., Renger^E.H. and Harnson.W.G. (1979) Nitrate and phytoplankton production in
southern California coastal waters. UmnoL Oceanogr., 24, 291-301.
Falkowski,P. (1981) Light-shade adaptation and assimilation numbers./ Plankton Res., 3, 203-217.
Flagg.C.N. (1987) Hydrographic structure and variability. In Backus,R.H. (ed.), Georges Bank. MIT
Press, Cambridge, MA, pp. 108-124.
Fleming.R.H. (1957) General features of the ocean. In HedgpethJ.W. (ed.), Treatise on Marine
Ecology and Paleoecology. Geological Society of America Memoir 67, 87-107.
Haury, Loren R., McGowanJ.A. and Wiebe,P.H. (1978) Patterns and processes in the time-space
scales of plankton distributions. In SteeleJ.H. (ed.), Spatial Patterns in Plankton Communities.
Plenum Press, New York, pp. 277-327.
Haury,L.R., VenrickJE.L., Fey.C.L., McGowanJ.A. and Niiler.P.P. (1993) The Ensenada front: July
1985. Calif. Coop. Oceanic Fish. Invest Rep., 34,69-88.
HerbIand,A. el at (1985) Size structure of phytoplankton biomass in the equatorial Atlantic Ocean.
Deep-Sea Res., 32, 819-836.
Hickey3-M. (1979) The California Current system—hypotheses and facts. Prog. Oceanogr., 8,191-279.
Holligan,P.M., Balch.W.M. and Yentsch.C.M. (1984) The significance of subsurface chlorophyll,
nitrite and ammonium maxima in relation to nitrogen for phytoplaniton growth in stratified waters
of the Gulf of Maine. /. Mar. Res., 42,1051-1073.
Jackson.G. A. (1986) Physical oceanography of the Southern California Bight. In Eppley.R.W. (ed.),
Lecture Notes on Coastal and Estuarine Studies, 15. Springer-Verlag, New York, pp. 13-52.
1856
Vegetation gradients in the sea
KamykowskiJ). and Zentara,SJ. (1986) Predicting plant nutrient concentrations from temperature
and sigma-t in the upper kilometer of the world ocean. Deep-Sea Res., 33, 89-105.
Kiefer.D.A. and MitcheU,BG- (1983) A simple steady state description of phytoplankton growth
based on absorption cross-section and quantum efficiency. Limnol. Oceanogr., 28, 770-776.
Koblentz-Mishke.O.I., Volkovinsky.V.V. and KabanovaJ.G. (1970) Plankton primary production of
the world ocean. In Wooster,W. (ed.), Scientific Exploration of the South Pacific, National
Acadamy of Sciences, Washington,D.C., pp. 183-193.
Lewis,M.R., Harrison.W.G., Oakey,N.S., Herbert.D. and Platt,T. (1986) Vertical nitrate fluxes in the
oligotrophic ocean, Science, 234, 870-873.
Lieth,H. and Whittaker.R.H. (1975) Primary Productivity of the Biosphere. Springer-Verlag, New
York, 339 pp.
Lynn,RJ. and SimpsonJJ. (1987) The California Current system: the seasonal variability of its
physical characteristics. /. Geophys. Res., 92,12947-12966.
Manryla,A.W., Venrick.E.L., and Hayward,T.L. (1995) Primary production and chlorophyll relationships, derived from ten years of CalCOFI Measurements. Calif. Coop. Oceanic Fish. Invest. Rep.,
36,159-166
Martin^I. and Fitzwater,S.E. (1988) Iron deficiency limits phytoplankton growth in the north-east
Pacific Subarctic. Nature, 331, 341-343.
McEwen.G.F. 1948. The dynamics of large horizontal eddies (axes vertical in the ocean off southern
California. /. Mar. Res., 7,188-216.
Mitchell,B.G. and Kiefer,D.A. (1988) Chlorophyll a specific absorption and fluorescence excitation
spectra for light-limited phytoplankton. Deep-Sea Res., 35, 639-663.
Morel^A. and Bricaud,A. (1981) Theoretical results concerning light absorption in a discrete medium,
and application to specific absorption of phytoplankton. Deep-Sea Res., 28,1375-1393.
Platt.T. (1986) Primary production of the ocean water column as a function of surface light intensity:
Algorithms for remote sensing. Deep-Sea Res., 33,149-163.
Platt,T. and Subba Rao J).V. (1975) Primary production of marine microphytes. In Cooper J.P. (ed.),
Photosynthesis and productivity in different environments. International Biological Programme.
Cambridge Univ. Press, Cambridge, England, 3, 249-279.
Redfield.A.C. (1936) An ecological aspect of the Gulf Stream. Nature, 138,1013.
Reid J.L. (1961) On the geostrophicflowat the surface of the Pacific Ocean with respect to the 1,000decibar surface. Tellus XIII, 4,489-502.
ReidJ.L.Jr., Roden.G.I. and WyllieJ.G. (1958) Studies of the California Current system. Calif. Coop.
Oceanic Fish Invest. Rep. 6,27-56.
ReidJ.L.Jr. (1988) Physical oceanography, 1947-1987. CalCOFI Rep., 29, 42-65.
RytherJ.H. (1969) Photosynthesis and fish production in the sea. Science, 166, 72-76.
Scripps Institution of Oceanography (1984a) Physical, chemical and biological data, CalCOFI Cruise
8401, 4-27 January, 1984; SIO ref. 84-18, University of California San Diego.
Scripps Institution of Oceanography (1984b) Physical, chemical and biological data, CalCOFI Cruise
8402-3, 7 February-30 March, 1984; SIO ref. 84-23, University of California San Diego.
Scripps Institution of Oceanography (1984c) Physical, chemical and biological data, CalCOFI Cruise
8404,9 ApriW May, 1984; cruise 8405,14 May-2 June 1984; cruise 8406, 2-24 June, 1984. SIO ref.
84-25, University of California San Diego.
Scripps Institution of Oceanography (1984d) Physical, chemical and biological data, CalCOFI Cruise
8407, 5-31 July, 1984; SIO ref. 84-30, University of California San Diego.
Scripps Institution of Oceanography (1984e) Physical, chemical and biological data, CalCOFI Cruise
8410, 18 October-11 November, 1984; SIO ref. 85-1, University of California San Diego.
Sosik^H.M. and MitcheU3-G. (1995) Light absorption by phytoplankton, photosynthetic pigments
and detritus in the California Current System. Deep-Sea Res., 42,1717-1748.
Steele J.H. and Baird.I.E. (1962) Further relations between primary production, chlorophyll and particulate carbon. Umnol Oceanogr, 7, 42-47.
Steemann Nielsen^E. (1954) On organic production in the oceans. Conseil Perm. Intern. Explor. Mer,
J. du Conseil, 19, 309-328.
Steeman NielsenJE. and Jensen,E.R. (1957) Primary oceanic production, the autotrophic production
of organic matter in the oceans. Galathea Rep., 1,1-136.
Sverdrup,H. (1955) The place of physical oceanography in oceanographic research. /. Mar. Res., 14,
287-294.
Sverdrup,H-H.U. and Fleming,R.H. (1941) The waters off the coast of southern California,
March-July 1937. Scripps Insl Oceanogr. BulL, 4, 261-387.
Tibby,R.B. (1943) Oceanographic results from the E.W.Scripps cruise VIII, May 10 to July 10,1939.
Records of Observations, Scripps Inst. Oceanogr. 1,67-78.
1857
W.MBaldh, RGBowler and CF.Byrne
Yentsch,C.S. (1974) The influence of geostrophy on primary production. Tethys, 6,111-118.
Yentsch.C.S. (1988) Why is primary productivity so high during the southwest monsoon? In
ThompsonJd.G. and Termizi,N.M. (eds), Marine Science of the Arabian Sea. American Institute of
Biological Science, Washington, DC, pp. 349-358.
Yentsch.C.S. (1993) CZCS: Its role in the study of the growth of oceanic phytoplankton. In Barale.V.
and Schlitttenhardt.P.M. (eds), Ocean Colour: Theory and Applications in a Decade of CZCS
Experience. KJuwer Academic Publishers, Norwell, MA, pp. 17-32.
Yentsch.C.S. and Phinney.D.A. (1992) The effect of wind direction and velocity on the distribution
of phytoplankton chlorophyll in the western Arabian Sea. In Desai,BN. (ed.), Oceanography of
the Indian Ocean. Oxford & IBH Publishing Co. PVT. Ltd, New Delhi, pp. 57-66.
Yoder^.A., Atkinson,L.P., Bishop.S.S., BlantonJ.O., Lee/T.N. and Pietrafesa.LJ. (1985) Phytoplankton dynamics within Gulf Stream intrusions on the southeastern United States continental
shelf during summer 1981. Com. Shelf Res., 4, 611-635.
ZarJ.H. (1974) Biostatistical Analysis. Prentice-Hall Inc., Englewood Cliffs, NJ, 620 pp.
Received on May 5, 1997; accepted on July 22, 1997
1858