The numerical simulation and a data analysis 494 C4horizon.documentation.ird.fr/.../b_fdi_51-52/010019844.pdf276 G. REVERDIN et al. winds. This resulted in large errors in unassimilated

Prog. Oceanog. Vol. 21, pp. 273-340,1991. printed in Great Brit& All rights reserved.

O079 - 6611191 $0.00 + 50 Q 1991 Pergmon pnss plc

'i .

The near surface tropical Atlantic in 1982-1984: Results from a numerical simulation and a data analysis

494 C4 G. REVERDI"~*, P. DEL&cLUSE', C. LJ%Y1, P. ANDRICH', A2dORLIEREl and

3 O$?,*" $I&& 'Laboratoire &Ockanographie Physique et de Dynamique du Climat (LODYC),

Université Paris VI, 4 PI. Jussieu, 75005 Paris

2Lamont-Doherty Geological Observatory, Palisades, Ny 10964, USA

'ORSTOM, 213 rue Lafqette, 75010 Paris

Abstract - Isotherm vertical displacements within the thermocline and surface currents were investigated in the tropical Atlantic Ocean from 12"N to 12"s in 1982-84, the period of the FOCAL- SEQUAL experiment. The study is based on anumerical simulation of an oceanic general circulation model tuned for the study of the equatorial regions, and on the analysis of the large scale thermocline displacements and currents using obsewed temperahue profiles. Ground truth is provided by temperature and currents from moorings, records from inverted echo sounders and tide gauges as well as from drifting buoys. Comparison of the analysis with the ground truth shows that some important aspects of the low frequency variability are "captured" by the analysis when the data base is large enough.

On large scales, the simulation generaUy resembles the analysis. Along the equator, the upwelling signal propagates eastward. The seasonal set-up of the westerly winds is associated with large westward currents, and a following overshoot of the zonal dynamic topography. Otherwise, the zonal dynamic topography is in near-equilibrium with the winds. The North Equatorial Countercur- rent is portrayed comparably in the analysis and the simulation, where, after starting as a narrow eastward flow near 5"N, it extends northward through the northem summer. In temual variations are found both in the analysis andqthe simulation. In particular, the thermocline flattened early in 1984.

However, the simulation differs in significant respects from the real world the equatorial undercurrent is too weak in the east and the model produces too much variability south of the equator. The 20°C isotherm is too shallow above the core of the thermocline, and the surface layer is too stratified Because the surface layer is where the wind stress, main forcing of the model is applied, major effort will have to be devoted to parameterizing the near-surface downward miXing of momentum, heat and fresh water.

Fonds Documentaire ORSTOM

274 G. MIN et aZ.

CONTENTS

1. Introduction 2. Comparison with Time Series

2.1 2.2 2.3 Vertically integrated measurements 2.4 The Low Frequency Basin Scale Variability 3.1 Monthly fields 3.2 Meridional sections 3.3

Vertical sections of temperature and zonal current The mooring thermocline depth and the surface currents

Discussion of time series comparisons 3.

Basin scale variability of the depth of the 20°C isotherm and the surface currents 3.3.1 20" depth 3.3.2 Surface currents 3.3.3 Decomposition in orthogonal empirical functions

The North Equatorial Countercurrent (NECC) 3.4 Season4 upwelling 3.5

4. Conclusion 5. Acknowledgements 6. References Appendix A The Numerical Model Appendix B: The Analysis

Appendix B.l The algorithm Appendix B.2 The noise Appendix B.3 Data distribution

1. INTRODUCTION

Our aim is to develop a numerical model to describe the seasonal variability in the upper equatorial Atlantic ocean. A validation of the model requires a comparison with oceanic data. The upper ocean average seasonal cycle was analyzed earlier. MERLE (1980) and KATZ (1981) presented evidence from temperature and density profiles showing a seasonal spatial redistribution of the upper ocean heat content by the currents, both zonally near the equator and meridionally north of the equator. MERLE (1983), MERLE and ARNAULT (1985) and GARzOLI and KATZ (1983) refined this analysis and related it to the seasonal meridional displacements of the intertropical convergence zone. The meridional structure of the thermocline vertical displacements near the equator was presented in PICAUT (1983) who focused on the annual harmonic of the vertical displacements. HOUGmON (1983) commented on the impulsive component of this equatorially trapped signal. Both suggested that remote changes in the equatorial zonal winds were the source of the eastern Atlantic vertical displacements. Diagnostic studies with numerical models of the ocean response to the wind forcing in different areas also show that variability in the eastern Atlantic is remotely forced (BUSALACCHI and PICAW, 1983; PHILANDER and PACANOWSKI, 1986). Currents were analysed to a significant extent by using data obtained from ship drifts by RTCHARDSONand McKEE(l984), RICHARDSON and WALSH (1986) and ARNAULT (1987). They exhibit a strong seasonal variability both near the equator and north of the equator. Off the equator, the currents show changes which are expected from the geostrophic currents deduced from surface dynamic topography.

Numerical simulations of the upper ocean with climatological wind forcing have been encouraging. The most salient characteristics of the observed low frequency variability of the surface dynamic height are reproduced in the shallow water simulation of B U S U C C H I and

273 282 282 283 293 295 298 298 303

307 307 309 312 317 318 323 326 327 330 334 334 335 336

Near surface tropical Atlantic 1982-1984 275

PICAUT (1983) and in the multi-mode numerical run of DU PENHOAT and TRÉGUIER (1985). General circulation models (GCMs) with more realistic coast lines and vertical density profiles also reproduced the low frequency variability (PHILANDER and PACANOWSKI, 1986; RICHARDSON and PHILANDER, 1987). These models are now usedin near-real time to describe the upper ocean, both for the Pacific Ocean (LEETMAAand JI, 1989) and the Atlantic Ocean (MORT-@ZW,REV€?RDIN and MERLE, 1989). The hope is that the errors and caveats of the simulations can be corrected by an adequate assimilation of observed data (see also, CARTON and HACKERT, 1989, 1990). However, in simulations of the Atlantic climatological cycle, the particular intensity of some of the model currents or vertical displacements seem to differ significantly from the observations (DUCHENE, 1989; " K I G N O U L , DUCHENE and CANE, 1989). Whether these differences originate from within the models themselves, the wind forcing used, or aliased interannual variability is still a matter of debate.

Differences in the ocean response are expected depending whether the forcing varies smoothly or is more impulsive, according to numerical studies discussed in PHILANDER and PACANOWSKI (1986) and in WEISBERG and TANG (1987). It should be, therefore, more insbxctive to compare the simulations with data collected during well observed years than with the climatological seasonal cycle. The FOCAL-SEQUAL Experiment (Français Oc6an Climat dans l'Atlantique Equatorial-Seasonal Equatorial Atlantic Ocean) sponsored an extensive observational program which took place in 1982-84. This paper will describe the ocean variability in 1982-1984 both in the data and in a wind forced numerical simulation. We will try to find whether the same patterns and characteristics of the variability are present in the two sets. This is a preliminary step for a necessary quantitative estimation of the closeness between model and data, which should be undertaken only if the data sampled the variability well.

Of course there will be some redundancy with earlier presentations of numerical simulations and of observations. The FOCAL-SEQUAL observation array includes time series from a few locations as well as a broader scale sampling of the thermal and current variability (Fig.la). The time series provided valuable information on the seasonal cycle, which generally support the earlier description from the climatology. Assessment of the zonal structure of the seasonal cycle near the equator and investigation of the impulsive forcing of the seasonal cycle were empha- sized. Studies focused on the inverted echo-sounder array (am, 1987a; GARZOLI, 1987), and on equatorial temperature and current moorings (WEISBERG and WEINGAR=, 1986; WEIS- BERG, HICKMAN, TANG and WEINGARTNER, 1987). The meridional structure was investigated along 4"W during the 1983-1984 upwelling seasons (HOUGHTON and COLIN, 1986), and across the North Equatorial Countercurrent at 28"W (RICHARDSON and REVERDIN, 1987; GARZOLI and RICHAFU)SON, 1989). Information on variability from eight seasonal cruises for the North Equatorial Countercurrent along 23"W (HI?" and HISARD, 1984) and along the equator (€€(SARD and HÉ", 1987) was also examined.

These studies suggest that a seasonal cycle was present, both in 1983 and in 1984; but there were important differences between 1983 and 1984 detected both in the time series from the observational-array and in the cruise data. One major difference is that the eastern equatorial Atlantic thermocline deepened more in early 1984 than in early 1983 (KATZ, HIZARD, VER- STRAETEandGARZOLI, 1986; VERSTRAETE and VASSIE, 1990; HISARD andHÉ", 1987). Buoy drifts south of the equator simultaneously showed a relaxation of the South Equatorial Current (REVERDIN and MCPHADEN, 1986). HOUGHTON and COLIN (1986) discussed differences in the timing and intensity of the equatorial upwelling in 1983 and 1984; GARZOLJ and RICHARDSON (1989) discussed differences in the North Equatorial Countercurrent between the two years.

However, the variability analyzed in these earlier studies is far from representative of the whole equatorial Atlantic. Earlier OGCMs numerical simulations for 1982-84 did not provide much information on variability, since they were forced by model winds very unlike the actual

276 G. REVERDIN et al.

winds. This resulted in large errors in unassimilated model runs. In one (MORL-, et al, 198% the slope of the zonal thermocline along the equator is poorly simulated and in another (CARTON and HACKERT, 1989) the off-equatorial surface currents are greatly underestimated. In this paper, these earlier investigations of temperature and current variability for 1982-84 to the basin scale are extended using a more comprehensive data set and a numerical simulation is carried with a carefully constructed wind field.

Our analysis of the observations is based on the FOCAL-SEQUAL mooricg array, temperature and current profiles collected as part of the FOCAL-SEQUAL program (HEIMERDINGER, 1987) and other sources (annual summaries, Fig.1). This extends the data set used by CARTON and HACKERT (1990) for their data assimilation of 1982-1984. Monthly data distributions are presented in Appendix B, Fig.B2, which show that large areas were not sampled for up to 3 months. The worst coverage is in the western Atlantic Ocean for May-July 1984, and for 1982 and early 1983 in the eastern Atlantic. These gaps are large compared to variability scales. They can not be filled by optimal interpolation without a priori knowledge of expected structures. Thirteen empirical functions extracted from the simulation and from earlier analyses of the seasonal cycle of dynamic height are used to define the spatial structure of the signal. The temporal structure of the signal is defined by harmonics of period equal or larger than 3 months. The analysis scheme is an objective function fitting discussed inDAVIs (1985) (see Appendix B). The analysis was checked against the observations in areas where data coverage was good, and this suggested that the analytical procedure has little impact on the data in the time series. Unresolved high frequencies induce an uncertainty in the analysis, but these can be estimated from the statistical characteristics of time series from the moored instruments or from inverted echo sounders. Studies of the high frequencies are presented in GARZOLI (1987), and KATZ (1987b) for equatorial Kelvin waves and in WEISBERG and mINGARTNFR (1988) for the instability waves.

The parameters analyzed include the depth of the 20°C isotherm and dynamic height, computed by combining the temperature profiles with a temperature-salinity relationship extracted from 1982-84 cruise data. The large scale variability of the near surface currents is also analyzed. However, surface current data from the moorings, surface drifters, and the current profilers were insufficient to define the large scale low frequency variability (CARTON and HACKERT, 1989). Following M A U L T (1987), approximations for the surface currents ilTe obtained as a combination of surface geostrophic current and an Ekman surface current estimated from the wind stress. The geostrophic current is computed from the dynamic height referred to 400db. We made two comparisons to evaluate the realism of these proxy currents. Firstly, for an average year, they were compared with the ship drifts (MAULT, 1987), which suggested that they capture the seasonal variability of the currents, except near the equator. Secondly, the numerical simulation for 1982-84 was used to compare the model surface currents with the model proxy currents. According to this model, the variability in the zonal component is well reproduced in the proxy surface currents, including currents close to the equator. Results are less satisfactory for the meridional component with large systematic biases. The proxy surface circulation near the equator is less divergent meridionally and the proxy zonal currents are more westward.

The numerical simulation of 1982-84 was conducted using an updated version of the Ocean Global Circulation Model (OGCM) developed at LODYC (see Appendix A). The model solves the system of the three-dimensional equations of a stratified ocean under the hydrostatic and Boussinesq approximations as in BRYAN (1969). The equations are compartmented into 16 depth levels and a variable horizontal C-grid. The initial condition is a steady ocean with the

, r

- ,G

' b


20N

4 s o 3 4 -

20s 40W 20w

60W longitude a

20N

al a 4 3 0 3 -

20s 20w 0

40w longitude 60W

20N

4 5 $ 0 6 d

20s. 60W 40W 20w 0

longitude

FIG.1. Distribution of the profiles reaching the 20°C isotherm for each of the three years 1982, 1983 and 1984.

278 G. REVWDIN et al.

temperature and salinity based on LEVITUS (1982) climatology with no current. In a procedure similar to that used by SARMIENTO and BRYAN (1982), the temperature and salinity fields are relaxed throughout the run towards the monthly climatology of temperature and the annual climatology for salinity ( L E ~ S , 1982). The relaxation is very small near the equator, but increases at higher latitudes. After three years of spinning up the model with an average seasonal cycle of the extemal forcings, the forcing for the years 1982-84 is applied. As a result of this initialization and the adjustment processes, there is an overall basin-wide cooling below 150m throughout the simulation (Fig.A2 in Appendix A). These trends are of little importance for us, as their meridional scale is larger than the near-equatorial structures investigated and the signal is small in the equatorial thermocline. Real trends with a comparable magnitude could result from decadal or centennial variability in the wind fields, but these trends in merchant ships reports (REVERDIN and DU PENHOAT, 1987) could be an artefact and were ignored in this simulation.

The wind forcing is estimated from 5-day averages of ship reports. The data are spotty and quite noisy (Fig.2a,b), although the sampling is adequate along the major shipping routes. Two examples illustrate a well defined inter-tropical convergence zone (ITCZ) centered at 7"N in July 1983, and one situation in early May 1984 with the ITCZclose to the equator. These illustrate that even on this time scale, there is a band of winds with a westerly component extending over a few degrees of latitude. On the July map, there is aregion of strong shear at the northem edge of this band. Except for this structure, it seems that a 2" resolution will capture the significant patterns in the fields. Wind stress fields are constructed by first composing the individual pseudo-stresses V V (V is the wind vector and V its modulus) into monthly 2"x2" fields. These monthly fields are filtered by m Empirical Orthogonal Function analysis as in DUCHENE (1 989) and a deviation for each 5-day period is then estimated by optimal function fitting. The wind stress is computed using an aerodynamic bulk coefficient estimated from the pseudo-stress and the temperature- humidity ship reports, based on the LIU, KATSAROS and BUSINGER (1979) scheme, where the wind is taken as the square root of the pseudo-stress. The zonal winds or wind stresses have been compared by DUCHENE (1989) and HOUGHTON (1989) to the monthly winds and wind stresses derived from moored anemometer data (COLIN and GARzOLI, 1987). The comparisons suggest that the analysis captures a large part of the seasonal variability in 1983 and 1984. Whether the analysis is also representative of the higher frequencies is difficult to determine.

Data distribution is usually good (Fig2a,b) near the Sa0 Pedro e Paolo Peíïedos anemometer. Comparison between the anemometer record and the ship reports is presented on Fig.2cYd). The fluctuations of the five-day averages from the ship winds have the right magnitude, but are not always coherent. It is important to retain wind stress high frequency variability to stir the model near surface layer, so we have retained these noisy sub-monthly fluctuations. Of course the 5-day averaging can alias energetic higher frequencies (COLIN and GARZOLI, 1987), so not much attention should be given to the high frequencies in the model. The main surprise in the anemometer-ship reports comparison is an unexplained systematic difference in the wind direction: the ship winds point more to the north, which results in a larger meridional pseudo- stress and a weaker zonal pseudo-stress. This could originate from problems in the ship winds, but the in situ anemometer winds could also be faulty; this creates an unexpected difficulty in validating the wind product.

According to earlier model studies, the zonal equatorial wind stress is very important in &he forcing of the equatorial ocean. A seasonal variability is obvious at al l longitudes (Fig.2e), but its amplitude is larger west of 25"W. There is no major transition near 30"W a transition found in the wind product used in PHILANDw and PACANOWSKI (1986) was probably the result of excessive smoothing. To some extent, the equatorial wind stress variability can be summarized


a) 1269 wind messages 26-30 July 1983

10s -

60W 50W 40W 30W 20W 1OW O l ong i tude

b) 1086 wind messages 01-05 May 1984 20N ice i I I 1 I I . . . I I I I I . +I -

1 ON

ar a 5 0 % m cl .-(

10s

20s 60W 50W 40W 30W 20W 1OW O

longi tude

FIG.2. (above and overleaf). Wind data (a), @): distribution of wind messages for two five day periods (26-30 July 1983 and 1-5 May 1984) (the domain where the wind has a westerly component is outlined). Zonal (c) and meridional (d) components of the five-day averaged pseudo-stress V V near Sa0 Pedro e Paulo Rock l0N, 29"W. (V is the wind vector and V its modulus.) The full line is for the anemometer measurements, as the dashed line corresponds to the analysis from shipof- opportunity data. (e) 15-day smoothed equatorial zonal wind stress in 1982-84 from ship reports.

The areas with a stress larger than 1O4mzs-' are shaded.

280 G. -IN et al.

anemometer: -

1983 1984

Sao Pedro e Paolo Peñedos (Oo55'N, 29'21'W)

m2/s2 d)

meridional pseudo-stress analysis: ...-..- anemometer: - 60

50

40

30

20

10

O -1 o -20

-30

-40

1

Near surface tropical Atlantic 1982-1984' 281

e) zonal equatorial wind stress (1 .e-6 m2/s2) (15 day smoothing)

a z O.

a h vl

; = f < A h

E 9 L Q

E r& h

O

Q E

a L h

z O

* < m c r a h

E Q E

vl

A

L cr

---5 a

7 0 - 0

7 O

c

a

40'w 30-w 20-w 1 O'W O' 1 O'E Longitude

as a combination of a period with weak winds early in the year, and after a rapid intensification, more intense winds lasting until December. This is comparable to what was used by WFJISBERG and TANG (1987) in their model study. There are also interannual differences; the wind increase occurred earlier and was greater in 1983 than in the two other years (the ITCZ did not move as far south in 1983).

(1) Comparison of numerical simulation with available time series from moored instruments. (2) The large scale variability simulation and analysis with particular emphasis on the thermocline depth (depth of the 20°C isotherm) and the surface currents. We are particularly keen to reproduce these two variables with the goal of simulating sea surface temperature. We show fields for particular months and present the variability along meridional sections. (3) Discussion of the fields, the average, their variability and a principal component analysis for the years 1982-84. (4) The spatial structure of the near-equatorial variability and the NECC.

.!&

2 u This paper will discuss the following:


2. COMPARISON WIT73 TIME SERIES

Time series from moorings provide an interesting insight into variations of the vertical structure of temperature and the currents to compare with the simulation and large scale analysis of temperature profiles. We focus particularly on the time series recorded by the surface moorings from the FOCAL-SEQUAL, experiment at O", 28"W, O", 4"W and 6"N, 28"W these not only provide the best vertical resolution and the longest records, but also represent very different areas of the equatorial Atlantic. The vertical structure of temperature and zonal current and surface currents and thermocline depth will be investigated. The data were presented in WEISBERG and WINGARTNER (1986), WEISBEXG et al (1987), WEISBERG and COLIN (1986) and RICHARDSON and REVERDIN (1987). We will present time series of vertically integrated data on travel time from inverted echo-sounders and sea level from tide gauges, which we will compare with the simulation and the analysis of the thermal profiles. Finally, we will compare these results with earlier attempts to simulate the equatorial Atlantic upper ocean in 1982-84 and discuss whether a joint presentation of the simulation and analysis of the temperature profiles will yield additional information for 1982-84.

2.1 Vertical sections of temperature and zonal current

In the mooring data at O" 28"W, the thermocline is always well defined with the steepest vertical temperature gradients between 16°C and 25°C forming a layer with a thickness of 50- 75m (Fig.3a). It is usually overlain by a well mixed layer which is particularly thick when the surface temperature is the coldest and the thermocline is the deepest in July to December. In the simulation (Fig.3b), the thermocline also moves as a unit with steepest gradients between 15°C and 20"C, but between July and December the surface mixed layer is a4Om thick - thinner than in the mooring data. In the model this layer is not as sharp as in the mooring observations. Both the mooring data and the model run (averages of 5 and 6 days respectively) exhibit a well defined seasonal variability in the temperature structure above 150m.

In both the observations (Fig.3~) and the numerical simulation (Fig.3d), there is an intense subsurface eastward current (the Equatorial Undercurrent, EUC), which moves up and down with anearly constant thickness. Its core is the deepest in October 1983, and is the shallowest in March- April 1983 and 1984. In the mooring data, WEISBERG et al (1987) found that the vertical displacements of the EUC were similar to those of the thermocline and this is also true for the simulation. WEISBERG et al (1987) did not find a pronounced seasonal cycle of the core velocity of the EUC. Figure 3c suggests that there was a decrease in January-June 1984, but the coarse vertical sampling by the current meters makes the core velocity hard to track. This variability is certainly more pronounced in the simulation when the model EUC slackened considerably in May-June 1984.

In both the simulation and the mooring data, the seasonal variations of the surface currents seem to be more in phase with the depth of the EUC (and of the thermocline) than with the EUC core velocity. In our simulation, when the undercurrent and the thermocline are the shallowest, the westward momentum at the surface is the smallest. The simulation and the observations differ in that the simulated currents have a negative shift (westward) with respect to the observed currents, reaching 20cm s-I from the surface to the core of the EUC. At the surface, this difference is consistent with the model mixed layer being thinner, which implies that the momentum input from the westward wind stress is distributed through a thinner layer in the model than in the real ocean. Below 20Om (not shown), the current meter data are poorly simulated.


9

3

At O" 4"W, the mooring records show the thermocline usually moves in phase between 15°C and 25°C (Fig.4a). During the upwelling season (between May and July 1983, and June-July 1984) the top of the thermocline got very close to the surface. So then it is likely that isotherm depths for temperatures larger than 15°C were not only related to vertical motions, but also to diabatic processes and vertical mixing. The salinity changes from cruise data suggest that vertical mixing occurred at these temperatures. This may explain why the maximum uplift of the 15°C isotherm occurred later than that of the 20°C isotherm (in 1983, it occurred in July, and in 1984, in August). The mixed layer at 4"W was much thinner than at 28"W, exceeding a thickness >25m only in October 1983 and September 1984. A detailed study of the variations of the mixed layer depth is presented in HOUGHTON (1989).

Most of these characteristics are also found in the simulation (Fig.4b) but during the upwelling season the upper isotherms remain at more constant depths than in the observational data. During this season, the isotherms 14°C and 15°C also shoal considerably more in the simulation than in the observations. During the remaining part of the yeaar, the main difference is in the maximum vertical gradient being less steep in thesimulation than in the mooring data (i.e. near the isotherm 20°C).

The mooring at O" 4"W showed that there was alarge subsdace eastward current (EUC) with velocities >40-50cm s-' most of the time. This EUC was weaker than at 28"W and underlay an intense westward current (the South Equatorial Current). The depth of the core of the EUC experienced seasonal variations, especially in 1984 (deepest in early 1984). These changes were less than the changes in the depth of the thermocline (Fig.4a), so that the minimum temperature of the core occurred during the upwelling season. In the simulation (Fig.4d) there is also an eastward subsurface current, except in June 1983, but it is much weaker than the observed current only exceeding 50cm s-l in January and September 1983. We find no well defined seasonal cycle either in the simulated core velocity, or in its depth. As at 28"W, the simulated currents are also more negative (westward) than the observed structure.

At 5"N, 28"W the mooring data and the other temperature observations show there was a thicker thermocline with coherent vertical motions over the upper 15Om (Fig.Sa,b). In the simulation, however, the thermocline appears closer to the surface so that the mixed layer is much thinner. Compared to the equatorial records, the subseasonal variability is shifted at lower frequencies. This subseasonal variability is not reproduced in the simulation. We do not show the vertical zonal current structure at 6"N 28OW, which was very simple. There is a strong seasonal cycle throughout the upper layer both in the observations and in the simulation, with currents varying in phase to 150m.

2.2 The mooring thermocline depth and the su#ace currents

Mooring records show, the 20°C isotherm depth characterised the vertical displacements of the thermocline in most areas, except possibly during the upwelling season in the Gulf of Guinea. There, the simulated 20°C isotherm is so close to the surface that, on FigSaillustrating the zonal equatorial slope in July, the simulation exhibits no zonal slope of the 20°C isotherm depth east of 15"W. In the model Gulf of Guinea, the 20°C isotherm is too shallow and has smaller seasonal displacements than the deeper isotherms. We found that the simulated maximum temperature vertical gradient is not located at the correct depth and that the mixed layer is poorly reproduced in the model run. However, below 5Om, the isotherm vertical displacements are usually well reproduced in the simulation. The zonal slope of the 20°C isotherm is well reproduced both in February and July 1984, two seasons which have a different structure.

284 G. RWWDIN et al.

(a) In situ temperature at (Oo, 28OW)

' A O

25 25

.50 50 A

E 75 75 A

100 A

125

n

W

5 100 Q U 125 U

150 150 A

175 1 75

200 200 A J F M A M J J A S O N D J F M A M J J A S O N D

1983 1984

(b) Simulated temperature at (Oo, 2S0W)

O O A A

25 25 A A

50 50 A A

A

n

E 75 75

5 100 100 A

v

Q

U 125 125 A

150 150 A

175 175 A

200 200

1983 1984 A

HG.3. (above and right) Timedepth section at O", 28"W. (a) the temperature mooring data averaged by 5 days (on the right margin, the dots indicate the levels most commonly sensed; (b) the temperature numerical simulation sampled once every 6 days (vertical grid points are given in Appendix B). (c), (d), id. (a), (b) but for the currents (currents lower than Ocm s.' or larger than 75cm s-l are shaded). The levels most commonly sampled are indicated on the right of the figure.

'Y .

L,

h -

i r


(c) In siru zonal current at (O", 28OW)

n . "

25 - 50 -

n

E 75-

5 100- v

Q Q) 125- U

150-

175-

200 -- 1 J F M A M J J A s o N D'J F M A M J J A s o N D

1983 1984

A

25

50 A

75 A

100 A

125

150 A

175

200 A

(d) Simulated zonal c m n t at (Oo, 28OW)

o , A

A 25 A

50 A

75 A

A

100 A

1 2 5 ~

150 A

175 A

200 J F M A M J J A s o N D'J F M A M J J A s o N D

1983 1984 A

286 G. REVERDIN et aZ.

n

E 75-

5 100- W

Q (3 125- U

150-

175-

(a) In situ temperature at (O', 4"W)

A

- 25 : - 50 A

-75 A - * - l o o A

- L - 125

- 150

- 175

- I I "

I 200" IV

,2001 I .I , 1 1 1 I I 1 ' I I I I 1 , , I I I

J F M A M J J A S O N D J F M A M J J A S O N D

1983 1984

(b) Simulated temperature at (O", 4'W)

A A

25 25 : 50 50 A

L

A

- 75 E 75

5 100 100 W

a U

125 125 A

150 150A

175 175"

200 200 J F M A M J J A S O N D J F M A M J J A S O N D A - . +

1983 1984 T

i r

FIG.4. (above and right) Time-depth section at O", 4"W. (a) the temperature mooring data averaged by 5 days (on the right margin, the dots indicate the levels most commonly sensed; (b) the temperature numerical simulation sampled once every 6 days (vertical grid points are given in Appendix B). (c), (d), id. (a), (b) but for the currents (currents lower than Ocm s1 or larger than 75cm s-I are shaded). The levels most commonly sampled are indicated on the right of the figure.


(c) In situ zonal current at (Oo, 4OW)

- I I "

A

U 1504

1754

200 J F M A M J J A S O N D J F M A M J J A S O N D

1983 1984

(d) Simulated zonal c m n t at (Oo, 4"W)

25

50 n

E 75

r, 100 W

El CD 125 U

150

1 75

200

O A A

25 A A

50 A

A 75

A

looA

1 2 5 ~

150A

175A

200 J F M A M J J A s o N D' J F M A M J J A s o N D

1983 1984 A


O

25

50 n

E 75

5 100 W

Q 0 125 U

150

175

200

(a) In siru temperature at (6'N, 28'W) O

25 A

50 A

75 A

1 O0

125

150"

175


1983 1984

(b) Simulated temperature at (6'N, 28OW)

' A A

25 A A

50 A

75 A

100 , 125,

'150,

175,


1983 1984

FIG.5. (above) Timedepth section at 6"N, 28"W. (a) the temperature mooring data averaged by 5 days (on the right margin, the dots indicate the levels most commonly sensed; (b) the temperature numerical simulation sampled once every 6 days (vertical grid points are given in Appendix B).

FIG.6. (right) Equatorial time series of temperature in the thermocline (a), zonal profiles of the depth of the 20°C isotherm in February 1984 (broken h e ) and July 1984 (full line). The heavy line corresponds to the analysis, the thin line to the simulation. (b): at 0" SOW, extracted from Fig.3, at 9% (c) at O", 4"N, extracted fromFig.4 at 7Om. The heavy line corresponds to the mooring data,

and the thin line to the simulation.


" I I I 1 I I a)

50 -

e - -

-- -

---.-e E N

20°C isotherm depth at the equator simulation analysis

150 February 1984: - - July 1984: -

1 I I I I 40° 30" 200 1oow. 00 10'E.

Longitude

temperature at 90 m (O", 28"W) mooring: - simulation: -

t 23

-18

- 17

-16

15 16

1 5 1 I I I I I


1983 1984

temperature at 70 m (O", 4"W) mooring: - simulation: - 23

22

21

20

19

18

17

16

15

14

13

JPO 27:3/4-C 1983 1984


The simulation of temperature in the thermocline is more successful than for the depth of the 20°C isotherm. As shown for the two moorings at O", 28"W (Fig.6b) and at O", 4"W (Fig.6~) simulated temperature variability is very similar to the observations. At 28"W there is a slight tendency for the simulated variability to anticipate the observed variability in late 1983 and early 1984.

Surface currents agreement (Fig.7a7b,c) is less satisfactory. In particular, at O", 28"W the simulated currents are much more westward than at the mooring, and there is no significant correlation between the two time series. At O", 4"W, the two time series are similar with a

January-February 1984 as well as in August-September 1983 and August-October 1984. This variability is also seen in the seasonal cycle of ship drifts (RICHARDSON and WALSH, 1986; ARNAULT, 1987), as in the numerical simulation for an average year (PHILAMIER and PACANOW- SKI, 1986). At 6"N, 28"W, the two time series have less in common (especially from November to April) where again the simulated currents are more westward than the observed. But the set- up of the eastward flowing North Equatorial Countercurrent (NECC) is simulated even if subseasonal frequencies are dissimilar in the two time series. There are other shorter mooringrecords collected in 1984 along 28"W at l"47'N and 3"N, which show comparable differences with the model simulation and analysis (REVERDIN, RUAL, DU PENHOAT and GOURTOU, 1991).

pronounced seasonal cycle. The currents are particularly weak in January-February 1983 and ~ 9 '

-

zonal current at 10 m (Oo, 28"W) mooring: - simulation: - inn . , inn

754 a)

-1 O0

J F M A M J J A S O N D J F M A M J J A S O N D * Y

1983 1984

mG.7. (above and right) Zonal c m n t at 1Om. (a) at O", 28"W, extractedfromRg.3 (b) at Oo,4OW, extracted from Fig.+ (c) at 6"N, 2IoW, extracted from the mooring .The heavy line corresponds to

the mooring records, and the light line corresponds to the numerical simulation.

Near surface tropicd Atlantic 1982-1984 291

zonal current at 10 m (Oo, 4"W) mooring: - simulation: - , l o o 100 ,

- 25 T-

25

E o -0 o 3 -25

-50

-75

- -25 W

- -50

- -75

I -100 J F M A M J J A S O N D J F M A M J J A S O N D

1983 1984

zonal current at 20 m (6"N, 28"W) mooring: - simulation: - , 100 100 I

-501 -75

I -100 - 1 0 0 / 1 1 I I I I I I I I I I I I I I I I I I I I I


1983 1984

292 G. RevwDIN et al.

I I

90 a)

o 70

Sao Pedro e Paulo (l0N, 29"W) dynamic height (0/400 db) and sea level

analysis: - simulation: - tide gauge: -. -

60 1

Sao Tome (Oo, 6 O E ) dynamic height (0/400 db) and sea level

analysis: - simulation: - tide gauge: -. -

I I I

JFMAMJJASONDJFMAMJJASONDJFMAMJJASOND I I I 1 l I I I I I I I I l I I l I I I I I l I I I I I I I I l I I

1982 1983 1984

FIG.8. Low-passed time series of the surface dynamic height referred to 4OOdb (cut-off at 2 months for the tide gauge and 18 day averages for the simulation). (a) Near Sa0 Pedro e Paulo, l0N, 28"W, the ti& gauge records (mixed line), the analysis (heavy line) and the simulation (thin line); 0) near Oo, 6"E (Sa0 Tom€). The tide gauge records are pressure records at a shallow depth adjusted to a climatological dynamical height (differences with surface sea level resulting from seasonal changes of the density of the water between the tide gauge and the surface are of the order of lcm peak-to-

Peak).


2.3 Vertically integrated measurements

In this section, we analyze various measurements which have been tentatively related either to dynamic height of the upper ocean or thermocline depth; viz. the travel time records from bottom-mounted inverted echo sounders (ES) (KATZ, 1987a; GARZOLI, 1987), and the near sea level pressure from tide gauges (CARTWRIGHT. SPENCER and VASSIE, 1987; VERSTRAETE and VASSIE, 1990). These records provide a complementary coverage of the ocean with which to test the model, particularly in the eastem equatorial Atlantic and off the equator along 28"W. The relevance of comparing such diverse data sets has been discussed in GARZOLI and KATZ (1983), and more recently in KATZ, HISARD, VERS- and GARZOLI (1986). REVERDIN, RUAL, DU PENHOAT and GOURIOU (1991) argue that there could be differences related to the deep variability. However, it is useful to compare these in situ data to model data of the upper ocean. The model product used is the dynamic height at the surface referred to 400db, which we can compare to the dynamic height from the temperature profiles. This will also provide an independent check for the hypothesis that the acoustic travel time from IES and sea level records are related to the changes in the upper ocean temperature.

There are two locations where tide gauge pressure records can be compared to the surface dynamic height from both the simulation and the analysis of the temperature profiles (Fig.8). There are shifts between the curves on which we will not comment, beyond pointing out that the model does not use the same equation of state as the analysis, and the tide gauge records are adjusted to a climatological mean.

Near 1"N, 29"W, the analysis and the simulation show a similar variability (Fig.8a). A larger signal in dynamic height is expected from the thermocline vertical displacements displayed on Fig.3, but the temperature variations at the surface compensate some of the deeper fluctuations. Interestingly, these two time series differ somewhat from the tide gauge record (especially the low values in May-June 1983) although the differences are not large (<5cm), and the signal is weak (10cm peak-to-peak).

At O", 6"E in the Gulf of Guinea (FigAb), there is large variability in all three curves (peak- to-peak, 22cm for the tide gauge), with differences between them on the order of 5cm after correcting for the mean shifts. In 1982, the differences with the analysis are larger but few data are included in the analysis. From October 1983 to January 1984, although the analysis seems to reproduce poorly the variability, it is still within the error bars of the analysis. The analysis and the simulation are similar enough to ascertain whether some of the differences with the tide gauge record are systematic.

The E S signal is often presented as a proxy for dynamic height (KATZ et al, 1986) or for the depth of the thermocline (GARZOLI and KATZ, 1983). Three records will be compared to the analysis and the simulation (Fig.9). They are converted as a depth of the 20°C isotherm, following GARZOLI and RICHARDSON (1989). Some aspects of the variability are portrayed in the three curves. The two upwelling seasons in the Gulf of Guinea (Fig.9a) and at 28"W (Fig.9b) are well rendered. The signal has a larger amplitude at 3"N than at O" along 28"W (a feature captured by the E S and the analysis, but not by the simulation).

However, there are differences between the curves which far exceed the uncertainties from random errors. The mooring records (Fig.4a), simulation and analysis (Fig.9a) in the equatorial Gulf of Guinea all suggest there was a considerable deepening of the thermocline in early 1984. This change is only wemy expressed in the Sa0 Tom6 tide gauge record (Fig.8a), and is undetectable in the closest E S record (Fig.9a). At 3"N, 28"W, the maximum thermocline depth occurred significantly later in the analysis than in the simulation or the E S records. Some of the

'2 *

4 >

' .. c - '

294 G. -IN et al.

90

1 0 0 1 1 1 1 1 , 1 1 1 1 1 1 ~ 1 1 1 1 1 1 1 1 1 1 1 ~ 1 , 1 1 1 1 1 1 , 1 1

20°C depth analysis: - simulation: - at (O", 4"W) IES at (O", 1O"W): .. .. at (O", 1"W): -. -

- 90

1 O0

n

E v

g 120-

CV 130-

140 -

5 a a U

- 60

- 70 - 80 - 90

.- 100

- 110

- 120

- 130

- 140 " *

20 -

30 -

40 -

50 - 60 -

70-

80-

L O

~ i'

20°C depth at (O", 28"W) analysis: - simulation: - IES: - . - 50 I 1 50

FTG.9. (above and right) Comparison between monthly averaged inverted echo sounders (ES) travel time records (thin line) and the 20°C isotherm depth from nearby observations (heavy line). The conversion from travel time to isotherm depth is the one used in KATZ (1987a). (a) O", low, except in October 1983 - September 1984, when the two records at low and 1O"W are shown; (b)

O", 28"W; (c) 3ON. 28"W.


20°C depth at (3"N, 28"W) analysis: - simulation: - E S : - ._ i 50

- . .i *

60

70 80 I 90

g 120-1 I 1 2 0

130 ,,,: I I I I I I I I I I I I I I I I I I I I I I I I I I , I I I I I 1 ;:: 150 150

JFMAMJJASONDJFMAMJJASONDJFMAMJJASOND

1982 1983 1984

discrepancies between the analysis and the E S can be attributed to the vertical structure of the temperature fluctuation which are integrated in the E S travel time. This was more thoroughly investigated in JXEVERD~N et al (199 1) based on longer temperature time series from expandable bathy-thermograms. They showed that at 3"N, 28"W, the effect is large and can easily explain the lag between the curves (the deeper cycle precedes the near surface cycle, Fig.9c). The Gulf of Guinea evidence is not as good, but the similarity between analysis and simulation suggests something similar has happened there.

2.4 Discussion of time series comparisons

The comparisons at the mooring sites show that the 2OoC isotherm variability is reasonably well simulated near the equator. Earlier simulations of the seasonal variability were no better ( P J " D E R and PACANOWSKI, 1986). The first attempts to simulate 1982-84 with OGCMs produced results which were clearly not as good. For example, in CARTON and HACKERT (1989) the unassimilated run did not show much variability in the thermocline depth at 28"W, and although the undercurrent core had vertical displacements, it was too shallow. Inadequacies of a similar magnitude are found in the runs without assimilation of MOKlÈRFJ, DEL&cLUSE,

,. - ANDRICHandCAMUSAT(1989)andMO~~,REVERDINandMERW?, 1989; Fig.ll),evenifthe proper thermocline stratification is used as initial condition (the dotted line in Fig. 11). This suggests that the improvement in the results presented here is related to the more realistic wind forcing in the model, and possibly also to the more realistic initialization and the inclusion of salinity.

Even in this simulation, the mixed layer is not very realistic and the depth of the maximum vertical temperature gradient is not always good. Probably for similar reasons, the equatorial currents are not very well reproduced. To simulate the current correctly at the equator is a very demanding task, since the momentum balance is expected top be complicated with a momentum flux from upwelling and vertical mixing (WACONGNE, 1989). At 3"N, 28"W, the large discrep-

ci


ancy of our simulated 20°C isotherm could be related to the model inadequately reproducing the vertical structure of the observed variability which indicates that there is a large advance in the low frequency phase at depth.

Comparison at other sites is clearly needed to validate the model. The hope that the E S would provide a complementary data set was confounded at 3"N 28"W by the signal of the E S being strongly influenced by the temperature variability in the water column below the upper thermocline. No other ground truth is available, but in the following section the description of the variability will be complemented by the less accurate but larger-scale gridded analysis of the temperature profiles. In many areas this analysis is adequate for determining the 20°C isotherm depth with estimated errors which are much less than the seasonal cycle (Fig.10). This parameter in the analysis compares well with the moorings at O" 28"W and O" 4"W. The rms difference is of the order of 3m, after applying a 3-month low-pass filter to the mooring time series (the mooring temperatures are not included in the analysis). In those areas, the analysis can be used as ground-truth to validate the model (see Fig.9c at 3"N, 28"W).

In other areas the high frequencies are badly aliased in the analysis even using the low frequency filter (a cut-off period of 3 months) because the data distribution does not resolve them (FigB2). It can hardly be considered as ground-truth and we will both consider the analysis and simulation to describe low-frequency variability. In the following section, the simulation high frequencies are partially filtered out by an 1 8-day running average in time and a box average over 1" in latitude and 5" in longitude on the same grid as the analysis. For the currents, we will also present in situ measurements on the same grid whenever possible.

- . '

-1

20°C isotherm depth at (3"N, 28"W) analysis: - simulation: - . - -50 -50

-60 -60

-70 -70 v E -80 -80

g-1 O0 -1 O0 'W

g -120 -i 20

r, -90 -90

-u -110 -110

r C Cd -130 -130

-1 40 -i40

-150 -1 50 JFMAMJJASONDJFMAMJJASONDJFMAMJJASOND

1982 1983 1984

FIG.10.2O"C isotherm depth at 3"N, 28"W. The heavy lines delineate the 95% confidence interval for the analysis (assuming a Gaussian distribution of the error, see DAVIS, 1985) and the light line

corresponds to the simulation.

- * * .


20°C isotherm depth at (O", 4"W) analysis: - simulation: -

20°C isotherm depth at (O", 28"W) analysis: - simulation:-

- :;:] E -80 W

5 -90

-a -110

$4 O0

g -120

-130

-1404

. - . . . . .

- -60

- -70 - -80 - -90

- -1 O0

--110

- -1 20

--130

--140

FIG.11. The 20°C isotherm depth in 1983-84 for different simulations (a) at O", 4"W, @) at O", 28"W. Heavy line for analysis, light line for the simulation discussed in this paper. Other lines are for earlier simulations: dashed line and mixed line are from two n m s presented in Mon-, DEZJkLUSE, ANDRICH and CAMUSAT (1989), and the dotted line corresponds to another run of the same model with initialization from the data (MORL.IÈRE, REVERDIN and MERLE 1989).


3. THE LOW FREQUENCY BASIN SCALE VARTABILllT

Initially the variability is described by showing spatial fields of the 20°C isotherm depth and currents for two contrasting seasons. Historical data (ship drifts) suggest that July and March are months during which the contrast in currents is maximum, so we selected July 1983 and March 1984 for our comparison. The currents in August 1985 for which we have a particularly large set of in situ measurements, will be illustrated.

In the following section, the time variability will be illustrated in more detail along two meridional sections: 30"W for the central-western Atlantic, and 4"W for the eastern Atlantic, which lie close to sections described in historical data. In the third section, basin scale low- frequencies are summarized in a few time series and their characteristics defined, both in analysis and simulation. In the fourth, equatorial upwelling and associated currents are investigated in more detail, and finally, the seasonal cycle of the NECC is discussed.

I ~ - . - . :

3.1 Monthly fields

We found systematic differences at the mooring sites in the depth of the 20" isotherm, especially during the upwelling season in the east, but which are also found basin-wide (Fig.12). There are discrepancies between observations and the simulation; the mean structure south of 7"s is poorly reproduced in the simulation and the average structure is less in the western NECC region. However, a number of changes between July 1983 and March 1984 are found both in the analysis and in the simulation. In July 1983, there is a depth minimum in the 20" isotherm near the equator and a well defined set of trough and ridge north of the equator in the westem and central Atlantic extending at least to 15"W. The trough north of the equator extends further east than in the climatology (m, 1983). There is also a zonal gradient near the equator (see also Fig.6a). In March 1984, these structures are much weaker east of 30"W. In particular there is very little zonal gradient along the equator, and the depth of the 20°C isotherm is maximum at the equator. West of 30"W, some of the July structure persists in the March analysis (less so in the simulation). The difference may be caused by data scarcity in the area seen in Fig.B2.

For each of the three months presented (Fig. 13), three sets of currents (simulated, analyzed and observed) show similar spatial structures, in particular for the zonal component away from the Brazil coast and more than 2" off the equator. Close to Brazil, the analysis with a 5" zonal resolution cannot capture the boundary current and also near the equator is too uncertain. In July 1983, the most salient features are the intense westward South Equatorial Current with a minimum at the equator west of 15"W, and the eastward NECC and Guinea Current (near 4"N in the Gulf of Guinea). Current observations are quite noisy as aresult of the intense fluctuations near the equator described in WEISBERG and FVEINGARTNER (1988). In March 1984, the currents were generally weaker, except for the westward currents west of 30"W at 4"N and north of 10"N. In the eastern equatorial Atlantic, the simulation and the equatorial mooring suggest the existence of a westward current which was not detected by the analysis. The situation in August 1984 was relatively similar to July 1983, except that the westward current north of the equator was less obvious in the central Atlantic.

The meridional velocity component shows differences among the three fields. In particular, there is little meridional velocity south of the equator in the analysis. However, this is very sensitive to the way in which we estimate the currents. If proxy model currents, instead of the simulated currents, were taken using the winds and the model surface dynamic height, little meridional drift would remain south of the equator. This suggests we should not use the meridional velocity component of the analysis to validate the model surface currents.

' -

,* b

T il

analyzed 20 C isotherm depth in July 1983 a)

O' 10'E W W W W 40'W 30'W 2O'W 10'W Longitude

simulated 20 C isotherm depth in July 1983

W W W W 40'w 30'w "W 10'w O' 10'E Longitude

analyzed 20 C isotherm depth in March 1984 b)

W W W W 40'W 30'W 20'W 1 O W O' 10'E Longitude

simulated 20 C isotherm depth in March 1984

ww W W 4o'w 30'w m w IOW 0' 10'E Longitude

FIG. 12.- Fields of the analysis and the simulation (20 day averages) 2OoC isotherm depth (a) in mid- July 1983, and @) mid-MsCh 1984.

I

300 G. REvwDnv et al.

a) simulated current July 1983

. - . 1 ON

a

4 s o A

c . . e . . . . r . r

10s

j/ 4-

60W 50W 40W 30W 201 1OW O longitude

analysis current July 1983 r , I I , I I u 1 ) ' ' I I ' ' I

608 50W 4QW 30W 20W IOW Q longitude

current data July 1983

I , , I I I I 1 ) ' 1 4 ' '

lONC

low O , " L I , , , , , , , I

10s -

BOW 50W 40W 37znngitui:W

mG.13. (above, right and overleaf) The aments from simulation (at 5m), analysis (as in AR- NAULT, 1987) and direct observations (moorings, drifters and profiler data) for three months. (a)

mid-July 1983; (b) mid-Mmh 1984; (c) mid-AupSt 1984.

1 .

o

1 0 N

..p.

c Y -


b, simulated current March 1984

l ' ' ' ' I ' '

c

+ i- c c e

.. Y

c L L e c

c r Y

60W 50W 4 0 W 3 0 W ZOW 1 O W O longitude

analysis current March 1984

) " ' ' ' I

BOW 50W 40W

c

s +

.. f- h c c

c c c

c +

e t

e

c .. +

30W 20W low O longitude

302 G. m m et al.

simulated current August 1984 cl

C C * . . C r . - . . & - . . . . . . * - -

O 60W 50W 40W 30W 20W 1OW longitude

analysis current August 1984

i) ,

c 4- c

-t

3 <

4-

..

c

c

$- .. h t-

L

h h .. c I

"( , , I O 60W 50W 40W 30W 20W lOW

longitude

current data August 1984

lost I I , , " I , , , J

60W 50W 40W 3i)oWngitu2W low o

. .

:*


The situation depicted in these fields is reasonably similar to that of the historical seasonal cycle both for temperature and currents deduced either from ship drifts or the geostrophic analysis of the currents. The currents appear structured in zonally oriented bands which extend almost across the basin. The depth of the 20" isotherm is tilted zonally and the changes in this tilt between the fields are the most obvious.

3.2 Meridional sections

The time sequences along two sections transecting the main currents presented in Fig.14 define the meridional structure of the variability. The two sections selected are along 30"W and 5"W for which data are most plentiful, and characterize the central equatorial Atlantic and the Gulf of Guinea respectively. For the 20" isotherm the average depth over the three years is removed at each latitude for clarity.

Along 30"W both the analysis and the simulation of the depth of the 20°C isotherm show a variability nearly in phase between 8"N and 6"s. At 10"N and further north, the variability has the opposite sign. The variability has a well defined annual harmonic also noticed both in the climatology (m, 1983; GARZOLI and KATZ, 1983) and from 7 years of XBT data near this longitude (REWERDIN et al, 1991; HOUGHTON, 1991). There are year-to-year differences which are portrayed similarly in the analysis and the simulations. The shallow thermocline in April-May is not as pronounced in 1983 as in the two other years, and the deep thermocline seen in mid and late 1983 has no equivalent in 1984. There is, however, amajor discrepancy between the analysis and the simulation: the analysis exhibits maximum variability near 4-5"N, and less variability south of the equator than in the simulation, in which the variability is meridionally more symmetric, and is less near 4-5"N than in the analysis.

In the Gulf of Guinea, the depth of the 20°C isotherm (Fig.15) showed a large variability associated with the equatorial upwelling trapped within 2" of the equator, which was very different from that along 30"W. These features are also clearly present in the climatology (HOUGHTON, 1983; PICAUT, 1983). Data coverage was poor in early 1982, and details in the analysis for this period should be disregarded. The amplitudes of upwelling for the three years differed between the analysis and the simulation; the simulated 20°C isotherm is too shallow during this season, and differs from that of the mooring. The phase of upwelling at the equator and near the coast is better documented in the analysis of HOUGHTON and COLIN (1 986). The most salient feature in the interannual variability is the very large equatorial deepening observed in early 1984 which was later also pronounced near the coast at 4-5"N. In the analysis, the cycle near 8-1 O'S differs from that near the equator, and does not seem to have undergone much interannual variability. This cycle is not reproduced in the simulation, but was observed further east in XBT data from a ship-of-opportunity programme (HOUGHTON, 1991).

Within 6" of the equator along 30"W, the zonal currents in the analysis (Fig.16a) varied less to the south of the equator than to the north (the 1.5"N - 1.5"s band is not presented). This trend also appeared in the average cycle of the ship drift data (RICHARDSON and WALSH, 1986; MAULT, 1987). However, in the simulation (Fig.16b) the current variability was comparable north and south of the equator, a discrepancy similar to the simulation of the depth of the 20°C isotherm. Early in its cycle around May, the analyzed NECC had its core (AOcm s1 in 1983) near 5"N. (The current developed a month later in 1984 than in 1983.) It then moved northwards as the current widens, which also happened in the simulation. In the analysis the southern boundary of the current was displaced to the north, but in the simulation, displaced slightly southward. However, this difference is at the limit of the resolution of our analysis and of the analysis of GARZOLI and RICHARDSON (1989).

u

.- -

*

304 G. MIN et al.

z m

z o

a, U æ.

.?i O tn d

I" m

z m

z L

a, U 3.

.?i O I" U

I" m

b) simulation anomalies along 30W


1982 1983 1984

t

FIG.14. Latitudinal time sections of the depth of the 20°C isotherm along 30"W (at each latitude, the average for 1982-84 is removed). (a) analysis; (b) simulation (contour interval of lOm, with Om

contour omitted).

- Y -


a) analysis anomalies along 5W

z ‘m

tI I I I 1 1 I ’ 1 - 1 1 1 I 3 1 . 1 I l II . -


1982 1983 1984

b) simulation anomalies along 5W


1982 1983 1984

mG.15. Latitudinal time sections of the depth of the 20°C isotherm along 5”W (at each latitude, the average for 1982-84 is removed). (a) analysis; (b) simulation (contour interval of 1h, with Om

contour omitted).

G. REVWDIN et al. 306

analysis along 30W zonal velocity (cm/s)

10 10

5 5

a U .- 2 o j Y t o id -

-5 -5

-I o -1 o JFMAMJJASONDJFMAMJJASONDJFMAMJJASOND

1982 1983 1984

simulation along 30W zonal velocity (cm/s)

10

5

-5

-1 o JFMAMJJASONDJFMAMJJASONDJFMAMJJASOND

1982 1983 1984

..

. .-

C I I

.. -*

RG.16. Latitudinal time sections of the zonal current along 30"W. (a) analysis results except within 1.5O of the equator; @) simulated nurents at 5m (velocities larger than 10cm s-' and larger than

30cm s1 are shaded).

Near surface aopicel Atlantic 1982-1984 307

Both in the analysis and the simulation, there was a strong semi-annual cycle close to the equator, extending a little further north in the analysis, where it was more intense. At the equator (where only the simulation is available supported by the mooring data) the semi-annual variability was less intense. In the simulation, an equatorial eastward current only occurs except in April 1984. On the equator, the westward equatorial current is usually less strong than at other nearby latitudes, a feature most pronounced west of 30"W. This equatorial minimum is also present in the ship drifts (RICHARDSON and WAJSH, 1986) and is more pronounced at the equatorial mooring (Fig.7a and WEISBERG et al, 1987).

Along 5"W there are no strong eastward drifts in the simulationnear the equator (Fig. 17b) (the analysis close to the equator is too noisy to be reliable). However, each year in January-February, there was a slight eastward drift this way; more pronounced in 1982 and 1983 than in 1984. Further south, there was an anomaly in January 1984 with an eastward drift between 3"s and 6"s. This small drift was also found during one cruise of the RV Capricorne (HI!" and HISARD, 1984,1987) and on buoy drifts (REVERDIN and MCPHADEN, 1986).

North of the equator, there was an intense annual cycle in the Guinea current. Its core was often located at 4"N, 1" off the Ivory Coast, both in the analysis and in the simulation (the analysis in 1982 is too noisy). The core of the Guinea current was fastest when the current shifted further to the south (August-September in the simulation, August in the analysis when the westward near- equatorial drift had already decreased). Ship drifts also suggest these features (MAULT, 1987) and a synthesis of the sections which sampled this current in 1982-84 has been presented in COLIN (1988). This paper suggests that the strongest currents are to be found in June-July when the current is close to the shore, also a secondary maximum occurs in January-February, a feature we failed to find either in the analysis or in the simulation.

. " .

CI_ '

.

.

3.3 Basin scale variability of the depth of the 20°C isotherm and the su@ace currents

This section attempts to extend the description of variability in the thermocline depth and surface currents to the whole basin. We will f ist present averages and standard deviations of these parameters, then discuss the correlation coefficient at zero lag between the simulation and the analysis. A few characteristic patterns are then extracted to characterize the variability with time series. Finally, a more detailed description is given of the equatorial band and of the NECC variability.

3.3.1. 20°C depth. The rms variability in this 20°C isotherm depth has a different spatial pattern in the simulation and the analysis (Fig.l8a,b). This is not surprising in the eastem Gulf of Guinea, where in the simulation during the upwelling season the 20°C isotherm depth is much closer to the surface than was observed. In the analysis, the largest variability in the west was found off the south American coast near 4"N-5"N, and the domain of large variability extended far to the east (close to 20"W).

Some of the spatial structures included in the analysis are extracted from the model dynamic height simulation, so that similarities in the variability pattem between the simulation and the analysis should be viewed with caution. We tried the analysis with functions from the model 20°C isotherm fields but results were poor, and a large share of the signal (close to 50% of the variance) was not portrayed in this set of functions. The spatial pattems from the model simulation of dynamic height were complemented by orthogonal functions from the climatological seasonal cycle and ad hoc functions (Appendix B) gave more degrees of freedom to the analysis.

The spatial distribution of the analyzed variance shows similarities to the model dynamic height field and the map bears resemblance to the seasonal variance in the average seasonal cycle presented in MERI.E (1983), at least for the eastern and western parts. However, in the centre of

-.

-? ..

308 G. REveRDnv et al.

analysis along 5W zonal velocity (cm/s) a)

- .- 2 o j Y to

' 3 - -5 -5

-1 o -1 o JFMAMJJASONDJFMAMJJASONDJFMAMJJASOND

1982 1983 1984

simulation along 5W zonal velocity (cm/s)

10 10

b)

5 5

a, -u 2 0 O .- - 3

-5 -5

-1 o -1 o ? -

JFMAMJJASONDJFMAMJJASONDJFMAMJJASOND 1982 1983 1984

I --

m~.17. Latitudinal time sections of the zonal current along 5"W. (a) analysis results except within 1.5" of the equator; @) simulated currents at 5m (velocities larger than Ocm s.I and larger than

3Ckm s-' are shaded).


the basin, near 4"N-5"NY there is more variability in the analysis than in the simulation dynamic height. There is also more variability than either in the climatology of the 20°C isotherm depth of MERLE (1983), or in the dynamic height presented inMWLEand M A U L T (1985). A similar discrepancy with climatology was also found when analyzing the 1979 profiles (REVERDIN, MOLJNARI and DU PENHOAT, 1985).

The nns difference between analysis and simulation of the depth of the 20°C isotherm (Fig. 18c) is weaker than the one associated to each of the two individual fields, except near 5"s- 6"s and 6"N-8"N, where variability in all instances was not especially large. Consideration of the correlation between the two time series shows that this is where the correlation coefficient at zero lag is weak. On the other hand, the correlation coefficient exceeds 0.8 within the equatorial band, as well as north of ZOON, in the westem Atlantic (the correlation of 0.6 is non zero at the 95% confidence level, if the number of degrees of freedom is 8 as estimated from the time series at O" , 28"W). Lagged correlations show that the higher correlation coefficient is found for lags of 1 month or less, with the maximum centered about a lag of 15 days between the simulation and the analysis.

3.3.2 Suflace Currents. The results for the currents are less significant, but still exhibit interesting features. First consider the average currents (Fig.l9aYb). In the western Atlantic the simulation generates a minimum in the westward currents at the equator associated with a meridional divergence. Along the coasts of South America there is a strong northwestward flow parallel to the coast with a weak eastward flowing branch at 5"N. In the Gulf of Guinea, the largest westward currents are found at the equator, and there is an intense average Guinea current at 4"N flowing to the east.

The analysis averaged currents have a similar pattem (no currents are estimated at the equator). Part of the differences at 1" or 2" off the equator or poleward of 6" are related to the formula used to derive the surface currents from the estimated dynamic height and the surface winds, which could introduce a bias with respect to the real currents. However, some differences in the zonal currents are systematic and are very coherent zonally. They are often corroborated by more direct estimates of the current: the FOCAL cruises HISARD and PlTON, 1987) or the buoy drifts (RICHARDSON and REVERDIN, 1987). These should be attributed to defaults in the simulation except along the coast of South America and close to Africa, where the large scale analysis fails to resolve the coastal currents. The current is too westward near 3"s and not westward enough near 3"N. Near 6"N-8"N the average NECC is underestimated. From mooring data, we see that the simulated current in the central Atlantic is also too westward at the equator, so that the observed equatorial minimum is ill defined. The differences in the meridional velocity seem to be related to some extent to the analysis method and will not be discussed.

Along the equator, the rms variability of the simulated zonal current (Fig.19~) is particularly large between 25"W and O" and is weaker further west. West of 30"W there is also a local maximum of variability near 5ON-8"N. Although this is a sensible pattem according to ship drifts (RICHARDSON and WALSH, 1986; M A U L T , 1987), it is hard to see how this fits reality. At the equator the simulated currents were reasonably similar to the observed currents at 4"W, but less so at 28"W. Near the equator the analysis variability is strongly dependent on the method used to estimate the surface current. It is also very noisy at 1.5"s as a result of errors in the geostrophic currents; this may explain why the analysis variability is larger than the simulated. Somewhat unexpectedly, the correlation between the analysis and the simulation (Fig.19e) is on the verge of being significant at the 95% correlation level (over 0.60, in particular near 2ON-3"N and 2"s- 3"s east of 30"W, as well as in the NECC).

b .

I .

- - a -

e *

-. -

simulation a) analysis b)

20 N 20 N

10 10

O O

10 10

20 s 20 s 6 0 W 50 40 30 20 10 O 10 ZOE 60W 50 40 30 20 10 O 10 20E

C) analysis - simulation

20N I d) 100 * correlation coefficient

20 N

10 10

O O

10 10

20 s 20 s

rms 1982-84 (m)

60 W 50 40 30 20 10 O 10 ZOE 60W 50 40 30 20 10 O 10 20E

RG.18. Comparison of the 2OoC isotherm depth derived íÌom analysis and simulation (labelled OPA). (a) the average simulated depth; @) average difference between simulation and analysis; (c) rms standard deviation is presented for the simulation; (d) correlation coefficient at zero lag between

the two fields. 11

+ 1 L ' ..

average 1982-84

10

O

10

20s : I I I I I I I

60W SO 40 30 20 10 O 10 20E

simulation I

20N I

rms 1982-84 (em/sj

10

O

10

20 s

4 100 * correlation coefficient

D . ' 1'

I

b) simulation - analysis - 0.3 m/s 20N I

1 average 1982-84 10

O

10

20s I I I 1 I 1 I l

4 simulation - analysis

6OW 50 40 30 20 10 O 10 20E

I

) rms 1982-84 km/s 20N

10

O

10

20 s

I

1982-84 20N I

10

O

10

-

I I I 1 1 I I

60W 50 40 30 20 10 O 10 20E

FIG.19. Comparisonbeh7veenthezonalcurrents inthe analysis and derived from the simulation @abelled OPA). (a) average simulated current; (b) average difference between the simulation and the analysis; (c) rms standard deviation for the simulation; (d) rms difference between the two fields; (e) correlation

coefficient at zero lag between the two fields.

\ I

20s / I l I I I I 1

6 0 W 50 40 30 20 10 O 10 20E


The simulated meridional drift has a maximum of variability within 2" of the equator and near South America. In most areas there is little correlation between the simulation and the analysis, but is very sensitive to how the surface analyzed drifts are estimated. The seasonal cycle east of 25"W between 5"N and 10"N is a robust feature in both fields.

3.3.3. Decomposition in orthogonal empirical functions. We selected spatial patterns characteristic of the large scales in order to present a few time series characteristic of the largest coherent features. Since the total time series were only three years long, patterns estimated from an empirical orthogonal function decomposition (EOF) are unstable. Removal of one of the three years changes the patterns drastically, except for the first two EOFs. We selected the common EOFs loadings which maxiiiiized the common variability in the fields of the analysis and used the same simulation as DuCHENE (1989). The patterns obtained independently for each of the two fields are close, and accounted for similar portions of the variance.

The patterns for the depth of the 20'C isotherm (Fig.20) are similar to those derived from the climatology of surface dynamic height (DELCROIX, 1984; DUCHENE, 1989). The first pattern corresponds to a large scale "in phase" redistribution of mass, zonally across 1O"W and meridionally across 7"N. It is reminiscent of the annual harmonic-amplitude map in (1983) except for a weaker eastern amplitude. The second pattern is more characteristic of equatorially trapped phenomena with a large part of its variance within 2' of the equator. It is reminiscent of the upwelling structure (HOUGHTON, 1983; PICAUT, 1983). Therefore a complex EOF analysis would have been more conducive to retain the timing differences of the upwelling signal with longitude (WEISBERG and TANG, 1987; REVERDIN and DU PENHOAT, 1987). (See section 3.5.)

The time series associated with the EOFs for the analysis and the simulation of the depth of the 20°C isotherm are indistinguishable for the first two EOFs. Together they include wer 65% of the variance (Fig.21), the 95% confidence interval estimated from data distribution is shown in the analysis). Both EOFs have a strong annual cycle with weaker amplitudes in 1983 than in the other two years. The highest values in the first EOF are found in April 1984 and the lowest values of the second EOF in February-March 1984. This period corresponds to the flattest thermocline both meridionally and zonally and is associated with the particularly strong relaxation of the equatorial wind stress in early 1984. If the second EOF is interpreted as a zonally integrated index of the equatorial upwelling, 1982 is the year when upwelling was strongest; it is also when equatorial sea surface temperatures were coldest (SERVAIN, SEVA, LUKAS and ROUGIE!R, 1987).

A scalar EOF analysis was also carried out for the simulated currents (with no analyzed currents at the equator we did not perform a similar EOF decomposition for the analysis). A vector analysis could also have been used but RICHARDSON and PHILANDER (1987) showed that the energy associated with the vector rotation component is small for the climatology. The first component (Fig.23a) associates the South Equatorial Current with the NECC. The second component (Fig.23b) is an equatorially trapped 4' wide current structure with stronger values in the Gulf of Guinea. These two structures are closely associated off the equator with the first two EOFs of model dynamic height (as geostrophic currents would do) and closely compare with EOF structures for the seasonal cycle of the surface drifts (RICHARDSON and PHILANDER, 1987). The two components include 55% of the total variance. The two time series (Fig.22b) have a strong seasonal cycle in quadrature with each other. There seems to be less interannual variability than for the depth of the thermocline. Possibly, the peak in May 1984 on the second EOF may correspond to a more intense westward transport of upper ocean water out of the Gulf of Guinea when the upwelling sets up. The lower values in 1984 for the first EOF could correspond to a weaker NECC.

. -

_- -.

?.

. __

spatial structure for the simulated 20°C isotherm depth

20 N 20 N

10 10

O O

10 10

20 s 6OW 50 40 30 20 10 O 10 20E 60W 50 40 30 20 10 O 10 20E

20 s

spatial structure for the analyzed 20°C isotherm depth

20 N 20 N

10 10

O O

10 10

20 s O 10 ZOE 20 s

OW 50 40 30 20 10 O 10 ZOE

FIG.20. The first and second common empirical orthogonal functions (EOR) for the analyzed and simulated fields of 2OoC isotherm depth. (a) spatial loadings for the simulated fields; (b) spatial

loadings for the analysis (in meters).

- 20°C isotherm depth - E.O.F. 1 analysis: S E 2 simulation: --.

10 - E r,

o -5

- 5

% o U

O Cu

-1 o

15

10 n

E 5

o -5

- 5

E o U

O cv

-1 o

-1 5

- 10

- 5

- 0

- -5

--IO

I I I I I I I I I I I I I I I I I I I [ ( I I Í I ( 1 [ [ I 1 1 1 ( 1

1982 1983 1984 JFMAMJJASONDJFMAMJJASONDJFMAMJJASONI

'II

20°C isotherm depth - E.O.F. 2 analysis: simulation: -

- 1 5 1 , , 1 1 , I I [ I ~ 1 ~ 1 1 1 1 1 1 1 1 1 1 1 ~ 1 1 1 1 I I I I I

15

10

5

O

-5

-1 o

-1 5

-15 w

. c

' - I I 1.J

FIG.21. The time series associated with the first and second EOFs. The light line corresponds to the EOFs of the simulation, the heavy line is associated to the EOFs of the analysis (the 95% confidence

interval is plotted), and the dashed line corresponds to the common EOFs analysis (in meters).


b)

1 EOF1(32%) I """ I

10- - cn

o >

\

E 0 - v

, u - * -10-

u

20s I I I I I I I I I 60W 50 40 30 20 10 O 10 20E

-10

- 0

--io

20N ' I

EOF 2 (22%) I

20s 1 \ I I

I I I I I I

60W 50 40 30 20 10 O 10 20E

-.. - 2 0 1 1 I I I I I I l I l l ) I 1 1 1 1 1 1 1 1 1 1 / 1 1 1 1 1 1 1 1 1 1 1 1 -20 JFMAMJJASONDJFMAMJJASONDJFMAMJJASOND

1982 1983 1984

FIG.22. The principal component dekomposition of the simulated currents at 5m (in cm 8'). (a) the spatial structures for the first and second EOFs; (b) the associated time series (the first EOF, light

line; the second EOF, heavy line).


1984

1983

1982

a) analyzed equatorial dynamic height (h(0/400)-75cm} in 1982-1984

1984

1983

1982

5O'W 40'W 30'W 20-W 1o'w o' 1 WE Longitude

b) simulated equatorial dynamic height {h(0/400)-70 cm) in 1982-1 984

50'W 40'W 30'W 20'W 1 O'W O' 1 O'E Longitude

FIG.23. (above and right) Time-zonal plots for the surface dynamic height referred to 4OOm. (a) analysis; (b) the simulation; (c) time series of the simulated zonal dynamic height gradient (light line) between 35"W and lOoW, an estimate of the zonal dynamic height gradient h m IES between 35"W and 10"W (mixed line), and z"/ph, where z" is the zonal wind stress averaged between 35OW

and lOoW and h is taken as 5Om (heavy line).


I I I

the equatorial zonal dynamic height gradient 35"W-1O"W i.e.s.: - . - simulated -

the equatorial zonal wind stress: - x10-6 Ids2

O I I

" '

3.4 Seasonal upwelling

Some of the remaining questions on the upper equatorial Atlantic Ocean depend on how slow or fast the winds set up and whether or not the ocean response is a near-equilibrium to the wind stress. Because the winds used are not heavily low-pass filtered and because of the observation density in the equatorial band, the analysis and the simulation may enhance the information.

Both in the analysis and simulation the maximum isotherm uplift (andlowest surface dynamic height) occurred earlier at 30"W than in the Gulf of Guinea (Fig.23a,b). This feature is also found if monthly averaged winds are used which produce very similar fields (MO" and DUCHENE, 1990). It is also present in a 20-year run with a multi-mode linear model forced with monthly winds (REVERDIN and DU PENHOAT, 1987).

It is noteworthy that in the Gulf of Guinea, the shallowest 20°C isotherm depth (or the lowest surface dynamic height) occurred roughly a month earlier in 1983 than in 1984 (also discussed in HOUGIFION and COLIN, 1986). Although the seasonal cycle is striking in these time series, the strong anomaly of the zonal slope in early 1984 is remarkable both in the analysis and in the simulation (Fig.lOc, Fig.23). The slope is at a minimum first in February in the east, then the effect is seen in the west (March-April). West of 30"W the zonal slope remains strong in the analysis, but this is more uncertain as few data points are available.

PHILANDER and PACANOWSKI( 1986) commented that these changes are nearly in phase with the local wind stress. Indeed, the zonal wind stress on Fig.2e also shows some of these features. In early 1984, the zonal wind stress weakened first in the east, then later in the west. However, there is no indication in the winds of the slow eastward phase propagation of the upwelling signal. WEISBERG and TANG (1987) showed that this can be related to the zonal structure of the

2'

U "

--

318 G. -IN et al.

intensification of the wind, and happens even if the changes in the winds are in phase zonally. They also attributed some of the changes later in the year in the Gulf of Guinea to the impulsive nature of the wind set up.

In simulation and data analysis for 1982-84, changes in the east are also not inphase with the local winds, and the zonal dynamic slope responds to remote forcing. Even in central Atlantic Ocean, the zonal gradient variations differ from those in the zonal wind stress (only the simulation slope is shown on Fig.23~). Phase differences can reach 1 month. There is a major overshoot of the pressure force each year in June-July after the wind intensifies. The cycles of the zonal pressure force and of the wind stress fit best if it is assumed that the stress is distributed over alayer 40-50m thick, which is roughly the depth of the mixed layer in the central equatorial Atlantic. However, defining amixed layer at the equator is a delicate task and this balance may be very non- linear. We have also plotted the zonal gradient estimated from the E S (as in KATZ, 1987a). In line with what was said for the time series (Fig.9a), this estimate of the zonal gradient shares similarities with that of the model.

The equatorial zonal currents result from the differences between the zonal wind stress and the pressure forces. From Fig.23~ it is not surprising that the currents are not in phase with the wind stress, as was discussed in PHILANDER and PACANOWSKI (1986). To illustrate the currents associated with the Set-up of the upwelling, we show the domain for each year where westward currents in the simulation are larger than 50cm s 1 (Fig.24). In these fields the strong westward currents near 1O"W-15"W on the equator set up in April in 1982 and 1983 and a little earlier in 1984 (although the westerly winds are still very weak). This was already suggested by the time series of the second EOF for the currents (Fig.22b) and the westward currents occur before the westerly winds reach their maximum. The area of the strong westward drifts then extends both eastward and westward (especially near 3"N and 3"s) during the next two months.

The largest drifts last for two months. The moorings in 1984 along 28"W at l"47'N and 3"N detected these large drifts in June-July. At 15"W and 4"W, a strong westward flow was already present in late March. It is not clear from the moorings or other direct information whether the westward extension of the large westward drifts also occurs in 1982 and 1983.

There is no eastward jet at the equator comparable in magnitude to these westward currents. We expected to find a large anomaly in the surface currents associated with the relaxation of the pressure force in early 1984 (also suggested in HISARD's (1986) analysis). However, the weak eastward drifts found in early 1984 at the equator were almost normal for the season according to the simulation. This suggests, together with the zonal sea level slope (Fig.23~) that the slackening of the winds at the end of 1983 and in early 1984 was slow so that the oceanic response did not generate large surface equatorial currents. On the other hand, in the central and eastem Atlantic nearZ"S-5"S aweakeastward currentwas found fromDecember 1983 to February 1984; an anomaly which was also seen in buoy drifts ( R E m m and MCPHADEN, 1986).

3.5 The North Equatorial Countercurrent (NECC)

The other area with strong seasonal changes is further north, where the NECC (and the Guinea Current further east) sets up seasonally. The climatology of ship drifts suggest that the location of the NECC varies seasonally, being closest to the equator in April-May. There have been earlier studies of the NECC during 1982-84 between fixed latitudes. KATZ (1987a) studied the slope between 3"N and 9"N from E S records in 1983 and in 1984. He explained its seasonal variability with a linear wave model forced by the curl of the wind stress. More recently, GARZOU and RICHARDSON (1989) have noticed that this latitude band in the central Atlantic (28"W) includes not only the NECC but also part of the South Equatorial Current. RICHARDSON and REVWDIN (1 987) also focussed attention on the vicinity of 6"N.


1982 u < -0.5 m/s 20 N 20 N

10 - - 10

0 - - 0

10 - - 10

20 s 20 s I I I I I I I

60W 50 40 30 20 10 O 10 ZOE

1983 u < -0.5 m/s

10 -

60 W 50 40 30 20 10 O 10 ZOE

1984 u < -0.5 m/s

10 -

60 W 50 40 30 20 10 O 10 ZOE

pIG.24. The areas with simulated westward drifts larger than 0.5m S' between April and June for each of the three years 1982 to 1984 (months indicated by number, contour in N1 line for April and

May, in dashed line later).

320 G. RBvwDIN et al.

The shifts in the position of the NECC can be seen both in observations and in the simulation by delineating the domain where the zonal current is eastward (Fig.25). The shifts of the eastward Guinea current were described in Fig. 17. As there are meridional drifts off westem Africa, there is not always continuity between the NECC and the Guinea Current. However, in May-June, when the NECC sets up the simulation indicated there was continuous flow across the Atlantic over 60" of longitude. In the west, the meridional displacements are rather weak, and the southern boundary of the eastward drift is close to 3"N (at 38"W an IES was located at this latitude). Indeed, in the simulation 38"W corresponds to the location of anear-stationary southward meander of the

East of 30"W, there was large seasonal variability of the northern boundary of the current. It starts in a southem position as a narrow current in April or May (a little earlier in the analysis), then its core shifts to the north, simultaneous to a widening of the current. Later, the current velocity decreases. Furthermore, the shifts of the southern boundary are much weaker, but do not happen in the same way in the analysis and in the simulation. The analysis is close to the seasonal cycle of the ship drifts presented inRICHARDSON and W m H (1986) and buoy drifts (RICHARDSON and REvF", 1987).

One consequence of the meridional shifts of the NECC in the central Atlantic is that it is barely possible to make deductions of its intensity from measurements made as faraway as 3"N and 9"N (inverted echo sounders available for earlier studies). The current is weaker at 28"W than it is in the westem Atlantic, but, because it is wider, its transport (Fig.27) per unit depth is not as different from that at 38" Was might be expected from the inverted echo sounder data at 3"N and 9"N. KATZ (1987a) suggested adifference of afactor 2. The meridionally integrated transports in the analysis and the simulation compare well; there were larger transports per unit depth at all longitudes in 1983 than in the other two years. A significant difference between the analysis and simulation is that the analyzed current persists longer than the simulated current.

The vertically integrated transport probably changes zonally more than surface transports per unit depth because the thermocline shallows from west to east (CARTON and KATZ, 1990). This change in the seasonal cycle penetration is apparent in the simulation, but the mooring at 6"N, 28"W (RICHARDSON and RE", 1987) suggested that the seasonal cycle still extended very deep (large changes werepresent at 300m). This question was reviewed b y ~ ~ ~ ~ ~ ~ ~ e t a Z ( l 9 9 1 ) who concluded, based on XBT profiles, that near this longitude the deep variability of the transport in the seasonal cycle is of the order of 4 or 5 Sverdrups. This is much lower than the transport assumed by IUCHARDSON and REVERDIN (1987) based on the mooring data of 36 106 Sv below 75m), but greater than in CARTON and KATZ (1990) who neglected it.

. NECC (see also CARTON and KATZ, 1990). - ß

__ _.

“ f b I l ,

1983 analysis 20N

I

07-09 I

10 ii r i I t 2 0 s fi

6 0 W 50 40 30 20 10 O 10 20E

1983 simulation 20N ’ I

07-09 I

60W 50 40 30 20 10 O 10 20E

1984 simulation 1984 analysis 20 N

20 N

10 - / _ A - - . - -- lo

0 - O

10 - 10

- 20 s 1 1 I 1 1 1 I

6 0 W 50 40 30 20 10 O 10 20E 20 s

J3G.25. Areas with an eastward NECC largerthan O.lm s-I in 1983 and in 1984, both in the analysis and in the simulation (months indicated by number; contour in fulI line for April and May, in dashed

line later for June-July, and dotted line for September).

G. REVERDIN et al. 322

200 200

175 175

- 150 150

in 125 125

o

n T-

E 100 1 O0 W

> 75 75

50 50

25 25

O O JFMAMJJASONDJFMAMJJASONDJFMAMJJASOND

1982 1983 1984

transport of NECC at 30W (u>l Ocm/s) OPA simulation and analysis

n , 200 200 , 175

150

125

1 O0

75

50

25

O JFMAMJJASONDJFMAMJJASONDJFMAMJJASOND

1982 1983 1984

RG.26. (above and right) Meridionally integrated zonal mass transport between 3"N and 12"N at different longitudes (only eastward speeds larger than O.lm s-* are included). The scale 100 corresponds to 10 Sverdrups for a transport distributed over a lOOm thick layer (lightline, the

simulation; heavy line, the analysis). (a) at &OW, (b) at 30"W. (c) at 20"W.


transport of NECC at 20W (u>l Ocm/s) OPA simulation and analysis

200 I '7 200

1754

'1504 n

6 100 - 1 > 75

C)

251

J F M A M J J A S ONDJFMAMJJASONDJFMAMJJASOI 1982 1983 1984

4. CONCLUSION

ND

Although the seasonal cycle is more repetitive in the equatorial Atlantic than in the Pacific Ocean, individual years contain more variability than in the average seasonal cycle. In this instance, we benefitted from the marked differences between 1983 and 1984. Describing and simulating years 1982-84 had been an exercise which had outlined features not found in earlier descriptions of the average seasonal cycle. For example, the climatological simulations of BUSALACCHI and PICAW (1983) and PHILANDER and PACANOWSKI (1987) missed the eastward propagation of the upwelling that was observed and simulated in the central Atlantic Ocean during at least 2 of the 3 years of our study. The meridional shifts of the NECC in the central Atlantic have also been poorly described in earlier simulations of the climatology, including some of the earlier simulations of 1982-84. This may have resulted from the input of wind data being

Linear models forced by winds as realistic as the ones we used, also effectively simulate some aspects of the seasonal cycle. For example, the simulation of DUPENHOAT and GOURIOU (1987) reproduces important aspects of the variability of the thermocline along the equator and across the countercurrent. Linear models aimed at isolating one aspect of the variabiIity have shown some success, for example in predicting the intensity of the North Equatorial Countercurrent (KATZ, 1987a), and in investigating the near-equatorial dynamics (WEISBERG and TANG, 1987). In this study the wind was idealized to show the influence of the sudden April or May increase in the trade winds near the equator. The solution presented in Fig.27a shows some resemblance with the OGCM simulation (Fig.27b) and the observations (Fig.23a). Some inadequacies are also obvious, such as the timing of the upwelling in the central and eastern Atlantic, and the too complete relaxation of the slope in the east.

3 * inadequate. I d .. +- *

324 G. RBvwDIN et al.

a) equatorial dynamic height (Weisberg and Tang, 1987)

1984

1983

1984

1983

50'W W W 30'w 20'W 1 O'W o' 1 VE Longitude

b) simulated equatorial dynamic height {h(0/400)-70 cm) in i 982-i 984

u

"W 4O'W 30'W 20'W 1o'w O' 1 O'E Longitude

FIG.27. Time zonal plots along the equator of the surface pressure; (a) ikom the reduced gravity model presented in WEISBWG and TANG (1987) (a scaled version of the surface elevation); and

@) from the simulation dynamic height referred to 400".

Near surface tropical Atlantic 1982-1984. 325

- 28

- 27

- 26

- 25

- 24

- 23

- 22

The aim of this tropical OGCM is to understand how the sea surface temperature evolves. Arguably, for this purpose, an OGCM should ultimately be more successful than simpler linear models. The way this simulation is conducted (see Appendix A) means that only the near- equatorial interannual and seasonal variability can be simulated (fluxes off the equator force the simulation toward the climatology). Comparison with the mooring data is shown on Fig.28. The overall time evolution of SST is simulated, in particular at 4"W where the warm anomalies of 1984 are well reproduced. However, at O", 28"W, the simulated SST is too low, suggesting that the influence of upwelling is too strong in the simulation. Away from the equator at 28"W, 6"N, the variability is not strong either in the simulation or the mooring data. These comparisons are promising, but show that systematic differences are still present which in a coupled ocean- atmosphere model could influence the air-sea interactions.

In the comparison with the moorings and with the large scale fields, we have noticed that the vertical density structure is not very well rendered in the simulation. The 20°C isotherm is too shallow and the near surface layer is too stratified. The seasonal cycle of the currents and of the thermocline south of the equator is also too large compared to observations (the same problem is found in the climatological simulation presented in RICHARDSON and -ER, 1987). Because the driving of the model near the equator depends crucially on how the momentum input of the wind stress is vertically distributed, the problem of how better to simulate the density structure near the sea surface should be addressed in further studies. This could involve improving the numerical scheme, the way vertical mixing is parameterized and how fluxes are introduced at the sea surface.

To test whether changes will result in further improvements, a sophisticated validation approach should be applied. It is not clear whether the data distribution for 1982-84 will be sufficient for this purpose even though in sim data are more numerous than in most other years. Observations are still lacking from large parts of the basin for entire seasons and this will obviously limit what can be deduced from the comparison with a large scale analysis.

. A ,

7

+.

21 -

temperature at 10 m (O", 4"W) mooring: - simulation: - 30 I . I 30

- 21

29

28

27

-2 26 ín

2 24

k 23

22

- a 25

c 29 I


temperature at 10 m (Oo, 28”W) mooring: - simulation: - 30 I - - ~ 3 0

- 29

- 28

- 27

- 26

- 25

- 24

- 23

v

22i 21

20

J F M A M J J A S O N D J F M A M J J A S O N D 1983 1984

temperature at 20 m (6”, 28”W) mooring: - simulation: - 30 30

29 29

28 28

v) 27 27

.2 26 26 u) a, 25 25

2 24 24

23 23

-

22 22

21 21

20 20

1983 1984 J F M A M J J A S O N D J F M A M J J A S O N D

n

5. ACKNOWLEIDGEMENTS m - æ

* We are particularly grateful to all the French and American scientists ’who contributed to the collection and validation of the oceanographic data set from the tropical Atlantic Ocean during the years 1982 to 1984. The encouragement provided by Jacques Merle to canyon this comparison was greatly appreciated, and comments on the manuscript by Steve Zebiak, reviewers and Rose-Marie Cline were very helpful. Discussions on modeldata comparison and wind data sets with Christine Duchêne and Claude Frankignoul have been quite stimulating. We ive grateful for the continuous financial support to pursue the analysis and the numerical study which was provided by various French agencies under the supervision of the “Programme National pour 1’Etude de la dynamique du Climat”. Computer resources were provided by the “Centre de Calcul Vectoriel pour la Recherche”, and computations were done on the CRAY II of this centre. Gilles Reverdins stay at Lamont was made possible by the generous gifts of the G. Lenger Vetlesen foundation and members of IpIEcA; a graut jointly funded by the National Science Foundation and NOM, OCE90-00127; and a N O M grant, NNO-AADAC502.

* -*


ANDRICH, P. (1989) Développement de mod2les numériquespour l'&de du couplage ockan-atmosph2re en zone

ARAKAWA, A. (1972) Design of the UCLA general circulation model. Numerical simulation of weather and

M A U L T , S. (1987) Tropical Atlantic geostrophic currents and ship drifts. Journal ofGeophysica1 Research,

ASSELI", R. (1972) Frequency filter for time integration. Monthly Weather Review, 100,487490. BRETHERTON, F., M. McPHADEN and E. KRAUS (1984). Design studies for climatological measurement of

BRYAN, K. (1969) A numerical method for the study of the world ocean. Journal of Computafional Physics, 4,

BUSALACCHI, T. and J. PICAUT (1983) Seasonal variability from a model of the tropical Atlantic Ocean. Journal of Physical Oceanography, 13, 1564-1587.

CARTON, J.A. and E.C. HACKERT (1989) Application of multi-variate statistical objective analysis to the circulation in the tropical Atlantic Ocean. Dynamics of Atmosphere and Oceans, 13,491-515.

CARTON, J.A. and E.C. HACKERT (1990) Data assimilation applied to the temperature and circulation in the tropical Atlantic, 1983-84. Journal of Physical Oceanography, 20,1150-1 165.

CARTON, J.A. and E. KATZ (1990) Estimates of the zonal slope and seasonal transport of the Atlantic North Equatorial Countercurrent. Journal of Geophysical Research, 95,3091-3100.

CARTWRIGHT, D.E., R. SF'ENCER and J.M. VASSIE (1987) Pressure variations on the Atlantic equator. Journal of Geophysical Research, 92,725-741.

COLIN, C. (1988) Coastal upwelling events in front of Ivory Coast during the FOCAL program. Oceanologica Acta, 11, 125-138.

COLIN, C. and S. GARZOLI (1987) In situ wind measurements and the Ocean response in the equatorial Atlantic during the Programme Français Ocean et Climat dans l'Atlantique &patorial and Seasonal response of the Atlantic Ocean experiment. Journal of Geophysical Research, 92,3741-3750.

DAVIS, R. (1985) Objective mapping by least squares fitting. Journal of Geophysical Research, 90,47734775. DELCROIX, T. (1984) EOF analysis of the thermocline depth in the tropical Atlantic Ocean. Unpublished

Manuscript. Tropical Ocean and Atmosphere Newsletters, Sept 1984. 18-19. DUCHENE, C. (1989) Test statistique de la validité des mo&les nudriques équatoriaux d l'aide d'observations.

Thbse de Doctorat 2 l'Universit6 Paris VI, defended on June 2 1989. ECKART, C. (1958) Propedes of water. Part II: The equation of water and sea water at low temperatures and

pressures. Journal of the Atmospheric Sciences, 256,225-240. ESBENSEN, S.K. and Y. KUSHNJR (1981) The heat budget of the Global Ocean: an Atlas based on estimates

from marine surface observations. Climatic Research Institution, Report 29, Oregon State University, Corvallis, 27pp, 188fig.

FRANKIGNOUL, C., C. DU- and M. CANE (1989) A statistical approach to testing the equatorial ocean models with observed data. Journal of Physical Oceanography, 19,1191-1208.

tropicale, Thhse de Doctorat de l'Universit6 Paris VI, defended on May 19 1989.

climate. Department of Meteorology, University of California, Rep.7,116pp.

92,5076-5088.

heat storage. Journal of Physical Oceanography, 14,318-337.

347-376.

~

- 2

GARZOLI, S. (1987) Forced oscillations on the equatorial Atlantic basin during the Seasonal Response of the Equatorial Atlantic Program (1983-1984). Journal of Geophysical Research, 92,5089-5100.

GARZOLI, S. and E. KATZ (1983) The forced annual reversal of the Atlantic North Equatorial Countercurrent. Journal of Physical Oceanography, 13,2082-2090.

GARZOLI, S. and P. RICHARDSON (1989) Low frequency meandering of the Atlantic North Equatorial countercurrent. Journal of Geophysical Research, 94.2079-2090.

-, S. and M. R O S E " (1983) Normal monthly wind stmss over the world Ocean with error estimates. Journal of Physical Oceanography, 13, 1093-1104.

HÉ", C. and P. HISARD (1984) Surface equatorial current system along 23"W (July 1982-January 1984) Geophysical Research Letters, 11,765-768.

HÉ", C. andP. HISARD (1987) TheNorthEquatorial Countemurent observed during theProgrammeFrançais Oc& Climat dans l'Atlantique Equatorial Experiment in the Atlantic Ocean, July 1982-August 1984. Journal of Geophysical Research, 92,3751-3758.

HÉ", C. P. HISARD and B. PlTON (1987) Observations hydrologiques dans l'ocban A h t i q u e c?quatorial (juillet 1982-aotlt 1984). Travaux et Documents ORSTOM, skrie FOCAL., 1 9 6 , 1 9 1 ~ ~ .

LI 4

~


HEIMERDINGER, G. (1987) Inventory ofoceanographic datu in the equatorial Atlantic, 1980-1984. A SEQUAL working document. February 1987.

HISARD, P. (1986) El Niño response in the tropical Atlantic Ocean during the 1984 year. International Symposium of Long Tem Changes, in Marine Fish Populations, 273-290.

HISARD, P. and C. HÉ” (1987) Response of the equatorial Atlantic ocean to the 1983-1984 wind from the Programme Français Oc& Climat dans l’Atlantique Equatorial Experiment cruise data set. Journal of Geophysical Research, 92,3759-3768.

HOUGHTON, R. (1983) Seasonal variations of the subsurface themal structure in the Gulf of Guinea. Journal of Physical Oceanography, 13,2070-2081.

HOUGHTON, R. (1989) Influence of local wind forcing in the Gulf of Guinea. Journal of Geophysical Research, 94,4816-4828.

HOUGHTON, R. (1991) The relationship of SST to thermocline depth at annual and intemual time scales in the tropical Atlantic Ocean. Journal of Geophysical Research, (in press).

HOUGHTON, R. and C. COLIN (1986) Thermal structure along 4OW in the Gulf d Guinea during 1983-1984. Journal of Geophysical Research, 91, 11727-11739.

KATZ, EJ. (1981) Dynamic topography of the sea surface in the equatorial Atlantic. Journal ofMarine Research, 39, 53-63.

KATZ, EJ. (1987a) Seasonal response of the sea surface to the wind in the equatorial Atlantic. Journal of

KATZ, E.I. (1987b) Equatorial Kelvin waves in the Atlantic. Jouml of Geophysical Research, 92,1894-1898. KATZ, EJ., P. HISARD, J.M. VERSTRAETE and S. GARZOLI (1986) Annual change of sea surface slope along

the equator of the Atlantic Ocean in 1983 and 19%. Nature, 322,245-248. LEETMAA, A. and M. JI (1989) Operational hindcasting of the tropical Pacific. Dynamics of Amtosphere and

Oceans, 13,465490. L E W S , S. (1982) Climatological Atlas of the world ocean. N O M professional paper 13, US Department of

Commerce, Rockville, MD, 174pp. LIU, T., K. KATSAROS and J. BUSINGER (1979) Bulk parameterization of &-sea exchanges of heat and water

vapor, including the molecular constraints at the interface. Journal of thedtmspheric Sciencex, 36,1722- 1735.

MERLE, J. (1980) Seasonal heat budget in the equatorial Atlantic Ocean. Journal of Physical oceanography, 10, 464-469.

hERLE, J. (1983) Seasonal variability of the subsurface thermal structure in the tropical Atlantic Ocean. In: Hydrodynamics of the Equatorial Ocean, J.C.J. NIHOU+ editor, Elsevier, N.Y., 3149.

hERLE, J. and S. M A U L T (1985) Seasonal variability of the surface dynamic topography in the tropical Atlantic 0cean;Journal of Marine Research, 43,267-288.

MORLIÈRE, A. P. DELECLUSE, P. ANDRICHand B. CAMUSAT (1989) Une 6valuation des champs thermiques simul6s par un modhle de circulation g6n6rale dans l‘Atlantique tropical. Olceanologica Acta, 12,9-22.

MO&, A., G. FEVERDIN and J. MERLE (1989) Assimilation of temperature profles in ageneral circulation model of the tropical Atlantic. Journal of Physical Oceanography, 19,1892-2899.

MORL-, A. and C. DUCHENE (1990) Evaluation des courants de sHace d’un m&le de circulation g&nkrule de l’Atlantique tropical. Intemal Report, LODYC, May 1990.

PACANOWSKI, R and G. PHILANDER (1981) Parameterization of vertical mixing in numerical models of tropical oceans. Journal of Physical Oceanography, 11, 1443-1451.

PENHOAT, Y. DU and A.M. TRÉGUIER (1985) The seasonal linear response of the tropical Atlantic Ocean. Journal of Physical oceanography, 15,3 16-329.

PENHOAT, Y. DU and Y. GOWOU (1987) Hindcasts of equatorial Sea surface dynamic height in the Atlantic in 1982-1984. Journal of Geophysical Research, 92,3729-3740.

PHILANDER, S.G. andR.C. PACANOWSKI (1986) A model of the seasonal cycle in the tropical Atlantic Ocean. Journal of Geophysical Research, 91,14192-14206.

PICAUT, J. (1983) Propagation of the seasonal upwelling in the eastem equatorial Atlantic. Journal of Physical Oceanography, 13, 18-37.

REVERDIN, G. and M. MCPHADEN (1986) Near surface current and temperature variability observed in the equatorial Atlantic from drifting buoys. Journal of Geophysical Research, 91,6569-6581.

-IN, G. and Y. DU PENHOAT (1987) Modeled surface dynamic height in 1964-1984: An effort to assess

Geophysical Research, 92,1885-1893.

h

E

- *

hL

- - 1L - I )


how well the low frequencies in the equatorial Atlantic were sampled in 1982-1984. Journal ofGeophysica1 Research, 92,1899-1913.

REVERDIN, G., R. MOLINART and Y. DU PE!NHOAT (1985) Objective analysis of thermocline depth distribution obtained in the tropical Atlantic ocean during FGGE, 1979. Deep-sea Research, 33,4343.

REVERDIN. G., P. RUAL, Y. DU PENJ3OAT and Y. GOURIOU (1991) Vertical structure of the seasonal cycle in the central equatorial Atlantic Ocean: XBT sections from 1980-88. Journal ofphysical Oceanography,

RICHARDSON, P. and T. McKEE 1984) Average seasonal variation of the Atlantic equatorial currents from historical shipdrifts. Journal of Physical Oceanography, 14,1126-1 138.

RICHARDSON, P. and S.G. PHILAMlW (1987) The seasonal variations of surface currents in the tropical Atlantic Ocean: A comparison of ship drift data with results from a general circulation model. Journal of Geophysical Research, 92,715-724.

RICHARDSON, P. and G. REVERDIN (1987) Seasonal cycle of velocity in the Atlantic North Equatorial Countercurrent as measured by surface drifters, current meters, and ship drifts. Journal of Geophysical Research, 92,3691-3708.

RICHARTSON, P. and D. WALSH (1986) Mapping climatological seasonal variations of surface currents in the tropical Atlantic using ship drifts. Journal of Geophysical Research, 91,10537-10550.

SADOURNY, R. (1975) The dynamics of finitedifference models of the shallow water equations. Journal of the Atmospheric Sciences, 4, 680-689.

SARMIENTO, J.L. and K. BRYAN (1982) An ocean transport model for the North Atlantic. Journal of Geophysical Research, 87,394-408.

SEAGW, R (1988) Modelling tropical Pacific sea surface temperature: 1970-1987. Journal of Physical Oceanography, 19,419-434.

SERVAIN, J., I. SEVA, S. LUKAS and G. ROUGlER (1987) Climatic atlas of the tropical Atlantic wind stress and sea surface temperature 1980-1984. Ocean-Air-Znteraction, 1. 109-182.

WSTRAElE, J.M. and J. VASSE (1990) The vertical structure of the zonal and meridional pressure gradients in the equatorial Atlantic during 1983 and 1984. Deep-sea Research, 37,l-26.

WACONGNE, S. (1989) Dynamical regime of a fully non-linear stratified model of the Atlantic equatorial undercurrent. Journal of Geophysical Research, 94,4801-4815.

WEISBERG, R and C. COLIN (1986) Upper ocean temperature and current variations along the equator in the Atlantic ocean during 1983-1984. Nature, 322,240-243.

WEISBERG, R. and T.J. WEINGARTNW (1986) On the baroclinic response of the zonal pressure gradient in the equatorial Atlantic ocean. Journal of Geophysical Research, 91,11717-11726.

WEISBERG, RH. and T.J. WEINGAR'I'NER (1988) Instability waves in the equatorial Atlantic Ocean. Journal of Physical Oceanography, 18, 1641-1657.

WEISBERG, R.H. and T.Y. TANG (1987) Further studies on the response of the equatorial thermocline in the Atlantic Ocean to the seasonally varying tyrade winds. Journal of Geophysical Research, 92,3709-3727.

WEISBERG, R.H., J.H. HICKMAN, T.Y. TANG and T.N. WEINGARTNER (1987) Velocity and temperature observations during the seasonal response of the equatorial Atlantic Experiment at O". 28OW. Journal of Geophysical Research, 92,5061-5075.

21,277-291.

a

U

MI"

d -


t -

APPENDJX A THE NUMEIUCAL MODE%

The oceanic model solves the system formed by the primitive equations with adequate boundary conditions for a stratified Atlantic ocean of a constant depth H (500Om). Classical assumptions are made on the equations: Boussinesq and hydrostatic approximations, incom- pressibility. The rigid lid assumption is made to filter the external barotropic waves. Density is computed from the state equation for the sea-water proposed by ECKART (1958). The set of equations is the following:

3, T + V.(T U) = F(T)

3,s + V.(S U) = F(S)

V.(U) = o k.V x (1/H ' 7 3 , ~ ) = k . V x G

where: G(x,y) is the second member vertical integral of the momentum equation (1); U is the vector current U(x,y,z)=(u,v,w); U, is the horizontal current U,=(u,v,O); U',=(U,-l/H.kJ,&), the baroclinic component; w (x,y,t) is the stream function associated to the vertical integral of the currents; F(U), F(T) and F(S) parameterize the effects of subgrid processes, and are expressed as:

F(U) = v h (vh(v.uh) - VA(vAUh)) + az(vvazu h)

F(T) = K h v.(vhT) adK, a,T> The horizontal heat diffusion coefficient 5 and the horizontal dissipation coefficient vh are equal to 1000m2s-1 which is half the value chosen by PHIL,ANr)m and PACANOWSKI( 1986). The vertical diffusion coefficients K, and 'U, are parameterized as a function of the gradient Richardson number R., similarly to the one developed in PACANOWSKI and PHILANDER (198 1):

K, = "i.' ( l + a R J + K b

2 2 2 -'y n+ vb with n = 2, vo= 100 cm /s, a = 5, v b = 0.0134 m /s, K ~ = 0.00134 m /S vv - ( I + a R J

During the computation, vertical static stability is instantly restored in the water column, and between the two upper levels at 5m and 15m, 'U, is forced to be larger than lOcm s-'. The model is a grid point model written with finite differences. The fields are distributed om a staggered grid of the type C, following ARAKAWA'S classification (1972). The discretization is made such that enstrophy is conserved, following SADOURNY (1975). For the temporal integuation, a leapfrog scheme is used to compute the currents, the salinity and the temperature fields, and, at each time step (one hour) a ASSELIN'S (1972) filter is applied.


The mesh on which the equations are compartmentalized (Fig.Al) varies smoothly in latitude and longitude. In latitude, the resolution is 0.33" at the equator and increases until 1.5" at 50"N. In longitude, the resolution is about 0.75'. The equations are written on this mesh with respect to the vertical operators using a curvilinear formalism. At 50"N and 40°S, the ocean is limited by rigid boundaries where no-slip boundary conditions and a linear damping on momentum are applied. The model has 17 vertical temperature and current levels with high resolution in the surface layers: 1Om for the first three layers and increasing thereafter (Fig.A2).

, A

t t h e model mesh in the tropical a t lant ic 25

20

60W 50 40 30 20 10 O 10 20E longitude

J?IG.Al . The tropical portion of the model grid mesh.

The boundary conditions are the following: (a) there is no flux of momentum nor heat across the lateral boundaries, the bottom and the coastlines; and (b) at the surface the momentum, salinity and heat fluxes are prescribed by the following equations:

@ -

2 :

-7 *

I ) ~ ~ , U ~ = z/p (z wind stress in Pa)

K,~,S = (E-P) S (E-P evaporation minus precipitation)

K ~ ~ ~ T = Q/(p C> (Q heat flux in W m-2)

The surface heat flux is written as: Q = F, - FLw - F, - FL, where Fm and FLw are respectively the short and long wave contributions to the net radiation at the sea surface estimated from the ESBENSEN and KUSHNIR (198 1) climatology. The sensible (F,) and the latent (F,) heat fluxes are diagnosed from the sea surface temperature computed in the model:


FL = pacJV,IL.(es(TJ - r el(TJ).(0.622/PJ

where ra is the air density (1.2kg m-3), c, is the drag coefficient (1.4 IV) is the modulus of the wind at 1Om; in the bulk formula, IV; is imposed to be always greater than 3.85 m s-I in order to take account of the high frequency variability at the surface; cp is the specific heat coefficient (1004.lkg m"), L is the latent heat (2.5 106J kg'), Ta is the air temperature, To is the sea surface temperature, Pa is the sea level pressure (taken to 1.013 105 Pa), r i s the relative humidity taken constant (O.@, es the saturated vapor pressure is estimated as eJT) = 10(9~mi-u53~ (T in O K ) .

This formulation, where the air temperature is specified from climatology (ESBENSEN and KUSHNIR, 198l), contributes to the introduction of a negative feedback toward an equilibrium seasonal cycle. However, a comparison done by ANDRICH (1989) with a simpler model, suggested that the time evolution is not too different from the one obtained by specifying atmospheric humidity to 80% of the saturation humidity at the sea surface temperature (as done in SEAGEiR, 1988). The wind stresses are computed from ship drifts collected between 30"N and 20°S, east of 60"W. Interpolation is made outside this domain to fit the Emu!" and ROSENSTEIN's (1983) monthly mean wind stress.

There is no direct parameterization for the evaporation minus precipitation budget at the surface. Instead of specifying azero flux (which can produce unrealistic behaviour of the surface salinity in the area under the influence of equatorial upwelling of high salinity water as pointed

- out by-WACONGNE, 1989), the surface salinity is relaxed to its climatological value with a6month time scale.

The ocean is initially at rest, with prescribed temperature and salinity profiles from m s (1982) climatology. During the computation, the temperature { T} and the salinity { S) not only evolve according to the equations presented above, but also via arelaxation (Newtonian damping toward their climatological values T,, S,) in a way comparable to that used in SARMIENTO and BRYAN (1982). The restoring coefficients increase with increasing latitude (A in degrees), and decreases with depth, as (for temperature):

K*(T,-T)*(&+l-cos(A)) where K = k,+(lfi-k,)*exp(-z/H)

with k,=1/720 day-', 4=1/4day', H=200m and &=0.001 day-'. At the surface, the relaxation term starts to be significant 30" away from the equator.

The computation is first forced for one year by the annual mean wind stress data for the period 1982-84; it is then forced for another two years with the average seasonal cycle 1982-84 and finally the period 1982 through 1984 is iintegrated with 6 day averaged wind stress. The first two seasonal cycles appear very similar; there is, however, a trend in the simulated fields under the thermocline (Fig.A2 for a similar longer simulation with a monthly average wind stress). This trend lasts throughout the whole run. It has little spatial structure between 20"N and 20"S, and at any given location near the equator, it is weak in the upper 20Om compared to the seasonal and the interannual variability.


T

25°C

20°C

15OC

10°C

5 O C

20"N - 20"s average temperature at model levels

I ' I I I I I I 1

1982 1983 1984 1985 1986 1987 1988 1989

N 1 2

3

4

5

6

7

8

9

10 11

12

13

14

15

16

Depth (m)

5 15

25

36.5

51.5

70

90

110

130

150

172

217

325

700

1625

3000

FIG.A2. Time evolution of the basin average temperature (2OON to 20's) for some of the 17 model levels for a longer model simulation forced with monthly averaged winds.

334 G. REWERDIN et al.

AF'PEWDIX B: THE ANALYSIS

B.1. The algorithm

The problem is, given a particular set of observations { yr(x,>} unevenly distributed in time and space, how to establish the most likely projection of the observations on a set of prescribed orthogonal functions {FJx)} defining the signal space. It should be obvious that the selection of these basic functions constitutes a major proportion of the problem.

The technique that we use, is the adaptation of the optimal analysis technique described in BRJTHERTON, MCRHADEN and KRAUS (1984) similar to DAVIS (1985). This requires statistics on the noise (the space which complements the signal space), but does not require apriori statistics on the signal. If these noise statistics are known, DAVIS (1985) showed that, in aleast square sense, it would be more appropriate to perform amore classical optimal analysis with this prescribedin- formation. But for the large scale variability in our investigation, we have little confidence in our ability to define these statistics; also, in this particular case, we do not want to bias the signal, which would be underestimated in an optimal analysis.

The analysis principle is the following: we wish to estimate the signal

'u(x) = CbmFm(x) m

from the observations, as a combination of the basic functions: h

'u(x> = Z P ( x , xn) v(xn> n

This is obtained by least square minimisation of: h h

<(Y(x) - Y(x>)~> subject to the no-bias condition <Y(x) - Y(x)> = O

where the brackets c.> represent a statistical average of many realizations of the same space-time distribution of data. The main difference with a statistical optimal analysis is contained in the second condition. For the projection on the space functions of the signal, this implies

that m

<gp = <b>

To solve this minimization, a covariance function of the noise Eb(x$) is specified. The resulting formulae are presented in DAVIS (1983, and give the set of { bm} as well as the error covariance matrix estimated for the solution in the function space for the signal functions (Eqp}. To!estjmate the error covariance matrix in physical space, we just project:

E(x, X ' ) = C Fm(x) Em p Fp(x') m, P

To simplify the computation, we have selected a set of functions which are orthogonal. We have separated the time and space functions, with the larger scale structures as the lower orders in the set of spatial functions, and in time, sine and cosine harmonics with periods larger or equal


to a period of three months. In space, 7 functions are extracted from an analysis of empirical orthogonal functions of the simulated fields. This set is complemented by 3 functions extracted from the seasonal cycle of dynamic height (DUCHENE, 1989) constructed from data assembled and discussed in MERLE and ARNAULT (1985) and ARNAULT (1987). Some systematic residuals were still present, so we added three other functions defined arbitrarily with large amplitudes near the equator in the Gulf of Guinea and in the eastem North Equatorial Countercurrent. Then, to simplify the computation, we define anew set of orthogonal functions by removing the part which projects onto the previous functions. Initially, the functions were calculated from the simulated depth of the 20°C isotherm. However, because the structure in the eastem Gulf of Guinea during the summer upwelling season is poorly reproduced (see also Fig.4), the analysis showed large residuals from the individual data in the Gulf of Guinea. A much better analysis is obtained when the empirical orthogonal functions are extracted from the field of dynamic height. The analysis is carried out for 2 year segments, which are then patched together (very little difference, except for the two months at each end).

L C

r

B.2. The noise

If there were enough data, the noise statistics could be estimated from the residuals of the data from the analysis {yr(xn) - Y?(x,)}. However, the available irregular and rather scarce data distribution provide insufficient clues on the noise error statistics. We have, however, used these residuals to estimate the noise variance. It varies for the depth of the 20°C isotherm between 50m' near the equator and in the Gulf of Guinea to 150m2 near 5"N - 10"N in the western equatorial Atlantic. To obtain an estimate of the noise covariance functions, we have mainly used either time series from mooring data or the XBT data collected with a dense enough resolution by merchant ships and cruise vessels during the FOCAL-SEQUAL field program. For that we have assumed that the zonal, meridional dependency, and time dependency could be separated and would have a universal form. That is:

For each estimate (either time series or sections), the data are projected first on the basic signal functions, and this signal is removed. First, the time correlation is estimated for the time series (piece wise averaging for all 30 day pieces). The correlation for the different moorings are indistinguishable at the high frequency, with a zero crossing after 5 days. An example is shown on Fig.Bl at (O'N, 28"W). We have therefore taken the same correlation function everywhere, * a but have removed the secondary negative or positive extrema. Then, we estimate the correlation for the residuals on the meridional (or zonal) sections, first taking into account the effect of time 3 : decorrelation. These different estimates are averaged (a westem estimate crossing the equator at

%,- '* 31"W, acentralestimatenear20"W andaneastemestimatenear9"Wattheequatorallcontainover 15 individual realizations). Although this estimate is noisy, it suggests that near the equator the meridional correlation goes to 0 for aseparation of 2". However, it is not obvious that it is the same everywhere, and although we have made this approximation (0 at 2" in latitude and 10" in longitude, 0.2 at 1" in latitude and 5" in longitude), it is possible that the small correlations neglected at large lags in space or time could indeed contribute to the error in the analysis in a noticeable way. However, usually the sampling is not so dense that it would be a very serious problem.


...*... - I'- ". ....*,.,,.* ,.'" *-., ,*' --... ,d .-.. -... ....-..... .

5 days 10days 20 days .30 days

FIG.Bl. Time auto-correlation of the depth of the 20°C isotherm estimated by hear interpolation between the levels of the mooring records at O", 28OW. The low frequencies defining the signal space have been removed before computing the correlation. The dotted lines give the 95% confidence

interval.

B.3. Data distribution

As seen for the annual summaries on Fig.1 and presented in more detail on Fig.B2, the coverage is usually sparse, but data are collected in large quantities along ship routes, so that there is a need to take into account the possible correlation of noise between the data. This is done by retaining all the data, but increasing its variance as:

2 (XJ 11 + CC(X, xn3I

n;tn

Then, the data can be considered as independent realizations of the noise, and the objective filtering is applied. Of course, because the functions are non local, there can be a very remote influence on data, especially in the area which are coarsely sampled. An example is presented in Fig.B3 where the standard analysis for the depth of the 20°C isotherm is compared with an analysis where data west of 48"W are not included. This area close to South America has a large noise level because of eddies, and has not been sampled homogeneously in time. The surprising feature is to find differences of up to Sm rms not only in this area, but also in the east close to Africa afew degrees off the equator. This area, of course, was also not well sampled. We have also found large differences in poorly sampled areas when profiles not reaching 400db (the analysis of dynamic height) were omitted. This suggests that the data distribution for those years did not oversample the large scale ocean variability.

rr

337 Near surface tropical Atlantic 1982-1984

FIG.B2. (above and overleaf) Monthly distributions of the temperature profiles usedin the analysis.


601 401 P O 1 O longitude

Near surface tropical Atlantic 1982-1984'


influence of data west of 48W

20N ’ I

rms di& (m) I 10

O

10

20 s 60W 50 40 30 20 10 O 10 ZOE

low O

10

20 s 60W 50 40 30 20 10 O 10 ZOE

RG. B3. RNIS difference in metres for the 2OoC isotherm depth in 1982-1984 between the standard analysis, and an analysis in which the data west of 48’W has been removed.

c