Vous êtes sur la page 1sur 21

Wind Forcing of Geostrophic Currents

Aneesh C. S.

October 22, 2008

1 Introduction
The Antarctic Circumpolar Current (ACC) is a unique oceanographic region in the world
ocean. It is the only truly global ocean current, linking the three major ocean basins of
the Southern Hemisphere. Knowledge about the transport in the ACC is important to our
understanding of exchanges of water masses between the world oceans. In Drake Passage,
the three main fronts in the ACC account for three-quarters of the transport of the current
while they account only for 19% of the width of the Passage ( Nowlin Jr and Clifford
[1982]).
There have been some studies about the barotropic transport in the ACC dominating
the variability in the current in the long time periods ( Hughes et al. [1999]). This study
indicates that the ACC has a dominant zonal barotropic mode and at there are regions
in the Drake Passage where the barotropic zonal transport is highly correlated with the
eastward wind stress in these regions.
We intend to explore this relation between the winds in the Southern Ocean and the
zonal transport across the Drake Passage further. The methods and results of this analysis
is explained in the following sections.

2 Wind Stress and zonal transport in the Antarctic Circum-


polar Current
The balance of forces in the Antarctic Circumpolar Current has been simplified to be a
balance between the tendency of the transport and wind stress with a linear drag on the
transport.

 x 
δU τ
∝ − bU (1)
δt Hρo
From this relationship, we can infer that the wind stress drives the net transport in
the zonal direction in the Antarctic Circumpolar current and is only curtailed by the drag

1
on the current by frictional dissipation. We can explore the relationship between the wind
stress and transport better in the frequency domain. Fourier Transform of equation (1)
would give us:

−iω Û ∝ τ̂ − bU
b (2)
x
where Û and τ̂ are the fourier transforms of U and ρτo H respectively. We could quantify
the phase relationship and also how closely the wind stress match the transport at different
frequency bands by using statistical tools. A statistical analysis by computing the coherence
between the two time series would tell us how close the wind stress matches the transport
and also the phase lag between the two timeseries.
Coherence is defined as:


c∗ >
< τ̂ U
C= (3)

p
< τ̂ τˆ∗ >< U
bUc∗ >

where the asterisk indicate the complex conjugate of the quantity and the brackets
denotes the mean quantity ( von Storch and Zwiers [1999]). C is a measure of coherence
between the wind stress τ and transport U. High values of C indicate that the wind stress
and transport vary consistently with each other. Low values of coherence would indicate
that the two timeseries are independent and vary randomly with respect to each other.
The phase lag between the two signals is given by :

" #
Im(< τ̂ Uˆ∗ >)
φ = tan−1 (4)
R(< τ̂ Uˆ∗ >)

If the predominant force balance in the momentum equation is the balance between the
wind stress and tendency of the transport, then from the phase relationship between the
wind stress and transport, we have −iω Ub ≈ τ̂ . The phase lag between the wind stress and
the transport will be given by:

h −ω i π
φ = tan−1 =− (5)
0 2
If the dominant force balance is between the wind stress and the linear drag term in
equation (1), then there is no obvious way to analyze the phase difference between the
wind stress and transport in ACC from equation (1). The phase lag in time would be 0 in
this case, as there would be no means to introduce a phase lag if the predominant balance
is not between the wind stress and transport. Hence the derived phase lag between the

2
wind stress and the transport would depend on the dominant force balance. In nature, the
force balance would be a combination of the different forces acting in the system, such as
the wind stress, transport tendency, frictional drag, stress from the boundary and viscous
dissipation. Hence, we would expect the phase relation to lie between the wind stress and
the transport to be a value between 0 and π/2.

3 Data
3.1 Wind Stress
NCEP reanalysis winds for the time period from 1992 to 1997 are used for this study.
The wind stress data from 18 latitudes and local winds at 5 different stations at which
bottom pressure guages were present are locally averaged. The NCEP winds are based on
reanalysis of historical data assimilated into an atmospheric model Kalnay et al. [1996].
Time series of the Wind Stress and Wind stress curl are shown in Figure (1) and Figure
(2).

3.2 Bottom Pressure


Several bottom pressure recorders were repeatedly deployed in Drake Passage during the
1990s. These time series of data are created by extracting data from the northern and
southern locations and concatenating annual records to make a long time series from 1992
to 1997. There are some time periods in both the North and South bottom pressure time
series when the data is missing. At these time periods, the data was interpolated using
a spline interpolation to fill the missing data. This might not be the best strategy to fill
missing data but was practical and the errors incurred due to this interpolation did not
affect this study significantly.
Time series of bottom pressure at the north and south moorings are shown in Figure (3
). The raw data shows that the bottom pressure varies in an annual scale at low frequency
but also has a very high variability at high frequencies.
The pressure gradient across the Drake passage can be interpreted using the geostrophic
relationship

δp
ρf u = − (6)
δy
Integrating this equation in the meridional direction for a zonal average transport will
give us:

yN
(pyN − pyS )
Z
f udy = (7)
yS ρ

3
Hence, given the north and south bottom pressures across the Drake passage, we can
treat the transport per unit depth at the depth of the bottom pressure guages as pro-
portional to the difference in the pressures between the North and South guages. Most
variability in the Drake passage was reported to be barotropic (Whitworth III [1983]). If
we assume that the variability in the vertical of the zonally averaged transport is negligible
due to the baroclinic flow being a small fraction of the variability in the transport, we can
use the pressure difference data from the North and South pressure guages as a proxy for
the variability in the barotropic transport in the Drake passage.

4 Coherence of Wind and ACC Transport


4.1 Power Spectral Density
The five year time series of wind and ocean bottom pressure were analyzed using coherence
analysis. The 5 year time series is divided into segments of 128 time points and fourier
transformed to compute the power spectra of both the wind stress and the bottom pressure
difference.
The power spectral density of the Wind Stress and the Bottom pressures were computed
and are shown in Figures 4 and 5. The power spectral densities show a typical Red spectra
slope as with most meteorological variables.
Spectral density is computed as the product of the fourier series of the time domain with
its conjugate scaled by the number of time points used to compute the fourier transform.
It is also defined as the complex fourier transform of the autocorrelation of the time series.

Z +∞
S(f ) = R(τ )e−i2πf τ dτ (8)
−∞

where R(τ ) is the autocorrelation function and the time series should be a wide-sense
stationary random process.
The uncertainity in the computed power spectral density is formally given by:

P (χ2ν,1−α/2 < ν fˆ(ω)/f (ω) < χ2 ν,α/2 ) = 1 − α (9)

Practically, these upper and lower limits of the errors are computed as:

ν/χ2ν,1−α/2 < S(f ) < ν/χ2ν,α/2 (10)

where S(f) is the power spectra. The error bars are plotted on the power spectra plotted
to show how significant the peaks in the spectra are. It is evident from the spectra that

4
there is a lot of power in the low frequencies and it doesn’t became white until very high
frequencies.
After the power spectra of the two time series are analyzed for the amount of energy
in the different frequencies, a coherence spectra is computed in the frequency domain of
the two time series using the form of equation (3). Figure 6 shows the coherence between
the bottom pressure difference across the Drake Passage and zonally averaged NCEP wind
stress. For significant coherence, we require that the squared coherence should exceed
0.2081 as computed from the formula for uncertainity in coherence:

β = 1 − α1/(nd −1) (11)

where α signifies the significance level. If α = 0.05, it means that there is a 5% chance
that random noise can give us the same coherence value as computed. Also, the phase
between the difference in the bottom pressure and average wind stress across the Drake
passage was computed according to the equation (4). Uncertainity in this phase calculation
is given by:

" s #
2
1 − Cxy
−1
δφ = sin tα,2nd 2
(12)
2nd Cxy

where Cxy is the value of the coherence at that frequency and the significance is com-
puted from a t-distribution test.
It can be noted from this coherence analysis that a large number of the lower frequencies
have a relatively higher coherence. It could mean that for the long time period variability,
the wind stress could be driving the mean barotropic transport across the Drake Passage.
But if this coherence also means a force balance between the wind stress and the barotropic
transport across the Drake passage, we should see a phase difference of π/2 between the
wind stress and the bottom pressure difference.
All the phase angles at different frequencies are plotted in the phase difference Figure
7. To see and understand the phase lag information in a better manner, the phase angles
are plotted in a rose histogram in Figure 8. From this histogram we can deduce that most
of the angles in the phase lag plot at different frequencies lie in the range between 15 and
25 deg . This is comparable to the results in Gille et al. [2001]

5 Summary and Conclusions


We tried to examine the relationship between the wind stress and the barotropic zonal
transport across the Drake Passge using coherence analysis. The bottom pressure data
from moorings north and south of the Drake Passage are used as proxies for the transport

5
across the basin. The Bottom Pressure time series shows significant coherence with the
NCEP wind stress data in this region. Most of the high coherences occur in low frequency
oscillations. Phase lag between the wind stress and the bottom pressure shows that the
wind stress leads the ocean transport by about 1/20th of a cycle. Since this lag is uniform
about a large band of frequencies, we would expect the time lag of these frequencies differ.
From our hypothesis, if the wind stress was being balanced by the tendency term of the
transport, then the phase lag between the wind stress and the transport would be π/2
but if the wind stress was being balanced by the linear drag term, then the phase lag
should be zero. From our analysis, we conclude that the phase lag has a value in between
the two expected values, suggesting that the force balance in the Drake Passage is some
combination of the wind stress being balanced by linear drag and also a net zonal transport
across the passage.

6
References
S.T. Gille, D.P. Stevens, R.T. Tokmakian, and K.J. Heywood. Antarctic Circumpolar
Current response to zonally averaged winds. Journal of Geophysical Research, 106(C2):
2, 2001.

C.W. Hughes, M.P. Meredith, and K.J. Heywood. Wind-Driven Transport Fluctuations
through Drake Passage: A Southern Mode. Journal of Physical Oceanography, 29(8):
1971–1992, 1999.

E. Kalnay, M. Kanamitsu, R. Kistler, W. Collins, D. Deaven, L. Gandin, M. Iredell, S. Saha,


G. White, J. Woollen, et al. The NCEP/NCAR 40-Year Reanalysis Project. Bulletin of
the American Meteorological Society, 77(3):437–471, 1996.

WD Nowlin Jr and M. Clifford. The kinematic and thermohaline zonation of the Antarctic
Circumpolar Current at Drake Passage. J. Mar. Res, 40:481–507, 1982.

H. von Storch and F.W. Zwiers. Statistical Analysis in Climate Research. Cambridge
University Press, 1999.

T. Whitworth III. Monitoring the Transport of the Antarctic Circumpolar Current at


Drake Passage. Journal of Physical Oceanography, 13(11):2045–2057, 1983.

7
Figure 1: Time series of the NCEP wind stress zonally averaged over the Drake passage
from 23 independent time series of wind stress

8
Figure 2: Time series of the NCEP wind stress curl zonally averaged over the Drake passage
from 23 independent time series of wind stress

9
Figure 3: Time series of the Bottom pressure anomalies at the north and south moor-
ings relative to their time mean and the pressure difference between the north and south
moorings

10
Figure 4: Power Spectral Density of the NCEP Wind Stress.

11
Figure 5: Power Spectral Density of the Bottom Pressure.

12
Figure 6: Coherence of Bottom pressure difference in the North and South of Drake passage
versus wind stress. Coherence is considered significant if it is greater than 0.2081

13
Figure 7: Phase Difference between the NCEP Wind Stress and Bottom Pressure.

14
Figure 8: Rose histogram of the phase difference.

15
clear all;
close all;

% Read the regrided data from input file

nc = netcdf(’data_regrid.nc’);

%Read variables from netcdf file

north_bpr_raw = nc{’NORTH_BPR’}(:); % North bottom pressure


south_bpr_raw = nc{’SOUTH_BPR’}(:); % South bottom pressure
time_old = nc{’TIME2’}(:); % Time
nt_start = floor(time_old(1,1)); % Start time
nt_end = floor(time_old(end,1)); % End time
time_axis = [nt_start:1:nt_end]; % Define time axis

time = [1:1:length(time_axis)];

stress = nc{’STRESS_REGRID’}(:);
curl = nc{’CURL_REGRID’}(:);
stress_mean = mean(stress,2);
curl_mean = mean(curl,2);

south_bpr = interp1(south_bpr_raw(:,1),time,’cubic’); % To fill missing data


north_bpr = interp1(north_bpr_raw(:,1),time,’cubic’); % To fill missing data
stress_mean = interp1(stress_mean(:,1),time,’cubic’);
curl_mean = interp1(curl_mean(:,1),time,’cubic’);

nt = length(stress_mean);

nts = 128; % Number of time points (days) in each time segment


freq = ((1:nts)-1)/nts;
tsave = [1:nts];

% convert to cycles/day

dtsave = mean(diff(tsave));
freqd=freq/dtsave;
period = freq.^(-1);
periodd = period*dtsave;

16
%diff_bpr_old = south_bpr(1:nt)-north_bpr(1:nt);
%x=[1:1:size(diff_bpr_old)];
%diff_bpr=interp1(diff_bpr_old,x,’cubic’);

diff_bpr = south_bpr - north_bpr;

stress_mean=stress_mean’;
diff_bpr=detrend(diff_bpr - mean(diff_bpr));
stress_mean=detrend(stress_mean - mean(stress_mean));

N = 14 % Number of segments the timeseries is divided into for computing spectra

tft_bpr = zeros(N,128);
tft_stress = zeros(N,128);

for i=1:N
tft_bpr(i,:)=fftshift(fft(diff_bpr(1+(128*(i-1)):128*i)’,[],1));
tft_stress(i,:)=fftshift(fft(stress_mean(1+(128*(i-1)):128*i),[],1));
end

inyq = (nts/2)+1;
ifplt = 2:inyq ;
ifont = 20;

tft_bpr_psp = sum(tft_bpr.*conj(tft_bpr)/(128^2),1);
tft_stress_psp = sum(tft_stress.*conj(tft_stress)/(128^2),1);

err_low = N/chi2inv(0.05/2,N);
err_high = N/chi2inv(1-(0.05/2),N);

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% To Plot the power spectral density of Bottom Pressure
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

%semilogy(freqd(ifplt),fliplr(tft_bpr_psp(ifplt)),[32 32],[err_low err_high],’linewidth’,1.


loglog(freqd(ifplt),fliplr(tft_bpr_psp(ifplt)),[0.05 0.05],[err_low err_high],’linewidth’
set(gca,’fontname’,’helvetica’,’fontsize’,ifont)
set(gcf,’color’,’w’)
axis tight
title([’averaged power spectum of Bottom Pressure’])

17
xlabel(’frequency (cyc/day)’)
ylabel(’power spectral density’)

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% To Plot PSD of Wind Stress data
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

figure
%semilogy(freqd(ifplt),fliplr(tft_stress_psp(ifplt)),’linewidth’,1.5)
loglog(freqd(ifplt),fliplr(tft_stress_psp(ifplt)),[0.05 0.05],[err_low/10000 err_high/100
set(gca,’fontname’,’helvetica’,’fontsize’,ifont)
set(gcf,’color’,’w’)
axis tight
title([’averaged power spectum of NCEP Wind Stress’])
xlabel(’frequency (cyc/day)’)
ylabel(’power spectral density’)

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% To plot autocorrelation
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

for i=1:700
lag=i;
[R,P]=corrcoef(diff_bpr(1+lag:end),diff_bpr(1:end-lag));
AutoCorC(i)=R(1,2);
ErrBar(i)=P(1,2);
end

figure
plot(AutoCorC)
set(gca,’fontname’,’helvetica’,’fontsize’,ifont)
set(gcf,’color’,’w’)
axis tight
title([’Lag Correlation of bottom pressure to NCEP Wind Stress’])
xlabel(’Lag period’)
ylabel(’Correlation Coefficient’)

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% To plot the correlation between wind stress and bottom pressure
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

18
for i=1:60
lag=i;
[R,P]=corrcoef((diff_bpr(1:end-lag)),(stress_mean(1+lag:end)));
CorC(i)=R(1,2);
ErrBar(i)=P(1,2);
end

figure
plot(CorC)
set(gca,’fontname’,’helvetica’,’fontsize’,ifont)
set(gcf,’color’,’w’)
axis tight
title([’Lag Correlation of bottom pressure to NCEP Wind Stress’])
xlabel(’Lag period’)
ylabel(’Correlation Coefficient’)

figure
plot(ErrBar)
set(gca,’fontname’,’helvetica’,’fontsize’,ifont)
set(gcf,’color’,’w’)
axis tight
title([’P-value for Correlation of bottom pressure to NCEP Wind Stress’])
xlabel(’Lag period’)
ylabel(’Correlation Coefficient’)

for i=1:60
lag=i;
[R,P]=corrcoef((diff_bpr(1+lag:end)),(stress_mean(1:end-lag)));
CorC(i)=R(1,2);
ErrBar(i)=P(1,2);
end

figure
plot(CorC)
set(gca,’fontname’,’helvetica’,’fontsize’,ifont)
set(gcf,’color’,’w’)
axis tight
title([’Lag Correlation of NCEP Wind Stress to Bottom Pressure’])
xlabel(’Lag period’)
ylabel(’Correlation Coefficient’)

19
figure
plot(ErrBar)
set(gca,’fontname’,’helvetica’,’fontsize’,ifont)
set(gcf,’color’,’w’)
axis tight
title([’P-value for Lag Correlation of NCEP Wind Stress to bottom pressure’])
xlabel(’Lag period’)
ylabel(’Correlation Coefficient’)

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% To plot Coherence for the two datasets
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

tft_bpr_psp = sum(tft_bpr.*conj(tft_bpr),1)/N;
tft_stress_psp = sum(tft_stress.*conj(tft_stress),1)/N;
cospec = sum(tft_bpr.*conj(tft_stress),1)/N;

nd = 14;
alpha = 0.05;

Coherence = abs(cospec)./sqrt(tft_bpr_psp.*tft_stress_psp);
coh_error = 1 - alpha^(1/(nd - 1));

coh_err = coh_error*ones(1,length(Coherence));

phase = atan2(-imag(cospec),real(cospec));
delta_phase = asin((1-betainc(2*nd/(2*nd+alpha^2),nd,.5))*sqrt((1-Coherence.^2)./(2*nd*Cohe

figure;
semilogx(freqd(ifplt),fliplr(Coherence(ifplt)),’linewidth’,1.5)
hold on
semilogx(freqd(ifplt),fliplr(coh_err(ifplt)),’linewidth’,1.5,’color’,’g’)
set(gca,’fontname’,’helvetica’,’fontsize’,ifont)
set(gcf,’color’,’w’)
axis tight
title([’Coherence spectum of NCEP Wind Stress and Bottom Pressure’])
xlabel(’frequency (cyc/day)’)
ylabel(’Coherence’)

20
figure;
semilogx(freqd(ifplt),fliplr(phase(ifplt)),’linewidth’,1.5)
hold on
%semilogx(freqd(ifplt),fliplr(phase(ifplt)-delta_phase(ifplt)),’linewidth’,1.5,’linestyle’,
%semilogx(freqd(ifplt),fliplr(phase(ifplt)+delta_phase(ifplt)),’linewidth’,1.5,’linestyle’,
errorbar(freqd(ifplt),fliplr(phase(ifplt)),delta_phase(ifplt));
errorbarlogx(0.03);
set(gca,’fontname’,’helvetica’,’fontsize’,ifont)
set(gcf,’color’,’w’)
axis tight
title([’Phase Difference of NCEP Wind Stress and Bottom Pressure’])
set(gca,’fontname’,’helvetica’,’fontsize’,ifont)
xlabel(’frequency (cyc/day)’)
ylabel(’Phase’)

figure;
rose(phase+pi)
set(gca,’fontname’,’helvetica’,’fontsize’,ifont)
set(gcf,’color’,’w’)
axis tight
%title([’Rose Histogram of Phases for NCEP Wind Stress and Bottom Pressure’])
set(gca,’fontname’,’helvetica’,’fontsize’,ifont)

21

Vous aimerez peut-être aussi