Rev. bras. oceanogr.. 48(2):93-105. 2000

The annual cycle of satellite derived sea surface temperature on thewestern South Atlantic shelf

Carlos A. D. Lentini1; Edmo J. D. Campos2& Guillermo G. Podestá1

lRosenstiel School ofMarine and Atmospheric ScienceDivision ofMeteorology and Physical Oceanography

(4600 Rickenbacker Causeway, Miami, FL 33149-1098, USA)Phone: (305) 361-4628 Fax: (305) 361-4696

E-mail: clentini@rsmas.miami.edu

2Instituto Oceanográfico da Universidade de São PauloDepartamento de Oceanografia Física

(Caixa Postal 66149, 05315-970 São Paulo, SP, Brasil)

. Abstract: ln this article, thirteen years of weekly sea surface temperature (SST) fields derivedtrom NOAA Advanced Very High Resolution Radiometer global area coverage intrared sateIlitedata, trom January 1982 to December 1994, are used to investigate spatial and temporalvariabilities of SST seasonal cycle in the Southwest Atlantic Oceano This work addresses largescale variations over the eastern South American continental shelf and slope regions limitedoffshore by the 1000-m isobath, between 42° and 22°S. SST time series are fit with annual andsemi-annual harmonics to describe the annual variation of sea surface temperatures. The annualharmonic explains a large proportion of the SST variability. The coefficient of determination ishighest (> 90%) on the continental shelf, decreasing offshore. The estimated amplitude of theseasonal cycle ranges between 4° and 13°e throughout the study area, with minima in August-September and maxima in February-March. After the identification and removal of the dominantannual components of SST variability, models such as the one presented here are an attractive toolto study interannual SST variability.

. Resumo: Neste artigo, treze anos de imagens semanais da temperatura da superficie do mar(TSM) obtidas através do sensor intravermelho Advanced Very High Resolution Radiometer abordo dos satélites NOAA, de janeiro de 1982 a dezembro de 1994, são utlilizadas para investigaras variabilidades espacial e temporal do ciclo sazonal de TSM no Oceano Atlântico Sudoeste. Estetrabalho objetiva as variações de larga escala sobre a plataforma continental e o talude leste daAmérica do Sul limitados ao largo pela isóbata de 1000 metros, entre 42°S e 22°S. As sériestemporais de TSM são ajustadas aos harmônicos anual e semi-anual para descrever a variaçãoanual das temperaturas da superficie do mar. O harmônico anual explica a maior parte davariabilidade da TSM. O coeficiente de determinação é alto (> 90%) sobre a plataformacontinental, decrescendo em direção ao largo. A amplitude estimada do ciclo sazonal varia entre4° e l30e na região de estudo, atingindo mínimas temperaturas em agosto-setembro e máximasem fevereiro-março. Após identificação e remoção das componentes dominantes da variabilidadeda TSM, modelos como o apresentado aqui são uma ferramenta atrativa para o estudo davariabilidade inter-anual da TSM.

. Descriptors: AVHRR, SST, Annual variability, South Atlantic.

. Descritores: AVHRR, TSM, Variabilidade anual. Atlântico Sul.

Cantr. no. 834 do Inst. aceanagr. da Usp.

94 Rev. bras. oceanogr.. 48(2). 2000

Introduction

Sea surface temperature (SST) is one of themajor components for studies in air-sea interactionsand cIimatc changc. Knowledge of thc annual SSTcycle, a pattern that nearly repeats itself every 12months, is also important to understand the timing oflocal fishery resources. Indeed, a description of theSST annual cycle allows the dominant seasonal signalto be removed, thus, providing a way to evaluateintcrannual SST variabiIity. Such an evaluation isexpected to provide a background for empiricalinvestigations into annual and interannual cycles fedby atmosphere-ocean dynamics.

This paper describes the annual cycle of SSTon the Westem South Atlantic (WSA) continentalshelf using thirteen years (January 1982-December1994) of infTared satellite observations. Similarstudies have concentrated on other parts of the worldocean (Wyrtki, 1965; Mer1e et ai, 1980; Hore1,1982;Gacic et ai, 1997). In the Brazil-Ma1vinasConfluence (BMC) region there are also someprevious studies (e.g., Podestá et ai, 1991; Provost etai, 1992). Neverthe1ess,none of these investigationshave focused on the WSA continental she1f.Repeatedmeasurements by satellite-based sensors now permitan improved definition of the timing and amplitude ofthe SST cycle in this area, historically undersampledby conventiona1 p1atforms such as ship and buoyobservations.

The area of study (Fig. 1) encompasses twodifferent hydrographic regimes. The northward flowofco1d Sub-Antarctic Waters ofthe Ma1vinasCurrent(MC) a10ng the shelf break, and the southward flowofTropica1 Water fTomthe Brazil Current (BC) a10ngthe continental margin (Emílsson, 1961; Castro Filhoet ai, 1987) (Fig. 2). The Brazil and Ma1vinascurrents roeet over the continental slope off theArgentinean basin, near 36° S, creating a strongfrontal zone (Gordon, 1989). Both currents then floweastward closing the subtropical gyre. Recentdescriptions of the hydrography and the circu1ationinthis area have been provided by 01son et ai. (1988),Garzoli & Garraffo (1989), Campos et aI. (1995),and Pio1a et ai. (2000).

Data and methods

Satellite Data

The satellite-derived SST estimates areobtained fTomthe Advanced Very High Reso1utionRadiometer (AVHRR) onboard the NOAA-N polarorbiting satellites. A detai1ed description of thespectral bands of the AVHRR can be found inSchwa1b(1978) and Kidwell (1991).

40"560'W 55"W 50"W 45"W

Fig. 1. Area of study. The dots indicate the 344 points forwhich SST was extracted and the seasonal cyclewas estimated. Points labeled A through H arerepresentative of the main hydrographic regimes inthe region. The three thin black tines indicate the50-m, 200-m, and l000-m isobath.

Fig. 2. Schematic representation of the Brazil Currentdomain. The focus here is placed on the BrazilCurrent and Malvinas Current (referred here as theFalkland Current) regimes. Bathymetric contoursare at 1000-m intervals; depths less than 3000meters are shaded. Adapted fi'om Peterson &Stramma (1991).

LENTINI et a!.: The annual cycle of SST in the WS AtIantic 95

SSTs are derived fTom Global AreaCoverage (GAC) data. The spatial resolution of theAVHRR GAC data is 4 km at nadir. Global SSTfields are produced at the University of Miami'sRosenstiel School of Marine and AtmosphericScience (RSMAS) using the SST estimates providedby NOAAINESDIS (National Oceanic andAtmospheric Administration/National EnvironrnentalSatellite and Data lnformation Service). Sea surfacetemperature is computed using a multichannelalgorithm (MCSST) for AVHRR infTaredchannels 4and 5 (McClain et a!., 1985).

The MCSST is derived fTom a multiple-regression analysis using coincident satellite anddrifting buoy data (Barton, 1995). The root-mean-square (rms) errar for global estimates of SST usingchannels 4 and 5 is typically 0.6°C (Barton, 1995;Walton et a!., 1998). This value is simply the rmserror obtained in the linear regression analysis used toderive the MCSST algorithm.

Ocean infTared measurements may includeother errar sources, such as instrumental noise,residual cloud contamination, residual calibrationerrors, etc. Moreover, the satellite-derived SST datainclude discrepancies between the ocean skintemperature measured by the AVHRR and theunderlying mixed layer or bulk temperature measuredby the drifting buoys (Walton et aI., 1998). A currentstatus on the accuracy of satellite-derived SST can befound in Casey & Cornillon (1999). SST fields arethen mapped to a fixed earth-based grid using acylindrical equi-rectangular projection. This grid,with element size of 18 x 18 km, has dimensions of2048 (longitude) x 1024 (latitude) (Olson et aI.,1988). Daytime-only cloud-fTee SST retrievals arepooled at weekly intervals. Cloudiness associatedwith warm westem boundary currents is a majorproblem for infTaredSST determination, either by notallowing SST retrievals when pixels are totally cloud-covered ar by introducing negative bias in SSTestimates for cloud-contaminated pixels (Podestá eta!., 1991). Pooling them at weekly intervals increasesthe number of cloud-fTeepixels on an image and, atthe same time, reduces the likelihood of negative biasdue to cloud contamination. Description and details ofthe method can be found in Olson et a!. (1988). Theweekly SST fields are the basis of ali subsequentanalysis. The data used here is fTom radiometersonboard NOAA-7 to NOAA-13.

Aualysis

Sixteen transects defined over the area ofstudy are considered. Out of these transects, 13 arecross-shelf and 3 along-shelf. For each transect thegrid points are spaced by 25 km. The transects areoriented fTomsouth to north, and fTomonshore to the1000-m isobath. SST values are extracted at each gridpoint. After the extractions, 344 individual time series

are obtained. The SST time ser'ies analyzed span 13years, fTomJanuary 1982to Oecember 1994.

The central premise of ali subsequentanalysis is that each weekly SST series, Y(t), can beconsidered as the sum of several components that aredistinguished by the way they vary in time. Followingthe nomenclature proposed by Cleveland et a!.(1983), SST series can be mathematically expressedas

(1)

for (t = 1.. .N), and where: Y(t) is the observed SSTvalue at day t, Tt is the trend component, associatedwith the low-fTequencybehavior in the leveI of Y(t)series, St is the seasonal component, a pattern thatnearly repeats itself every 12 months, and it. theirregular component which describes the remainingvariation. This last component can be a misnomer,since it may contain regular pattems not captured bythe other two components.

The seasonal component in each SST timeseries is mode1ed following the methodologydescribed by Wyrtki (1965) and Podestá et aI.(1991). The methodology combines the use of ageneral linear model to explicitly model the annualseasonal component (Cleveland et a!., 1983). Theseasonal cycle of SST at each grid point was, then,modeled as a sum of sine and cosine termsrepresenting the contribution of annual andsemiannual harmonics plus a mean value. This modelcan be written as

Y(t) = ao + Ia}cos(hjt/365.25) + b,sin (hift/365.25)] + e(t).)=\

(2)

where: Y(t) is the SST value at day t (expressed indays after 1 January 1982), ao is the estimatedtemporal mean value, aj and bj are the estimatedregression coefficients for the annual harmonic, a]and b] are the coefficients for the semiannualharmonic, and e(t) is the error term, assumed to beindependent and normal1ydistributed.

The model includes only annual andsemiannual harmonics because both are found to beof importance in other studies of temperature cyclesin tropical (Wyrtki, 1965; Merle et aI., 1980) andsubtropical (Podestá et a!., 1991; Provost et a!.,1992) oceans. Some advantages of this model can bestated as follows: (i) data are implicitly a ftrnction oftime; (ii) SSTs do not have to be observed at regulartemporal intervals; and (iii) the coefficients a(haj ...b)(for j=1,2) can be easily estimated by least squaresmultiple regression. In addition, the selected modelhas a physical justification. Periodic fluctuations in

96 Rev. bras. oceanogr., 48(2), 2000

the amount of solar irradiance on the sea surface,which are known to play the major role in forcing theannual SST cycle mainly in coastal and shelf seas,have a sinusoidal pattem (Seckel & Beaudry, 1973).

To evaluate the adequacy of the model, theF-statistic is calculated. For a review of the method,the reader is referred to Brook & Amold (1985). Thegoodness of the model can be expressed through thecoefficient of determination (,-2 ), which, in turn,measures the proportion of the variation in y that isexplained by the predictor variables (x).

Results

Due to the large amount of regressionsperformed, it is decided to select eight points asrepresentatives of the main oceanographic features forfurther analyses. Point A (J8°20'S, 54°27'W) jsinfluenced by the offshore extensions of both Braziland Malvinas currents, Point B (35°26'S, 54°W) isapproximately on the midshelf off Rio de Ia Plata,

Point C (34°41'S, 5001O'W) is located in BrazilCurrent waters, Points D (31°16'S, 50'W) and F(26°31'S, 4TO 1'w) are on the mid-outer shelf, PointsE (29°41'S, 49°W) and H (24~3' S, 45°08'W) are onthe midshelf, and finalIy Point G (25°56'S, 45°20'W),which is influenced offshore by the meanderingpartem ofBrazil Current. The estimated SST seasonalcycle tracks the observed SST quite welI, suggestingthat the sinusoidal model (Eq. 2) is adequate (Figs 3and 4).

Overall significance of the fitting

For the 344 time series, the regression modelis significant. In every case, the probability of theoveralI F-test is <0.01. The high significance of theregression model suggests that the selected modelaccounts for most of the variability in SSTdistributions. As an index of the model fit, the rms ofthe residuaIs for alI grid points is l.l7°C with a rangeofO.80° to 2.69°C (Fig. 5).

Lat 38°20'8 Lon 54°27'W (Point A)

22

2:18f- 14(f)(f) 10

1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994

24

~ 20

tn 16(f)

12

1982 1983 1984 1985 1986 1987 1988 '989 1990 1991 1992 1993 1994

28Lat 32°41 '8 Lon. 50°1 O'W (Point C)

1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994

28

Lat 31°16'8 Lon 500W (Point O)

~ 24

tn 20(f)

16

1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994

Years

Fig. 3. Satellite-derived SSTs (dots) and estimated annual SST cycle (line) forpoints A through D. The estimated annual cycle is approximatelysymmetric with maxima and minima SSTs observed about six monthsapart reflecting a pattern typical of midlatitudes.

LENTINI et ai.: The annual cycIe of SST in the WS AtIantic 97

28

Lat 29°41 '8 Lon 49°W (Point E)

~ 24

~ 20(j)

16

1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994

30

Lat 26°31'8 Lon 47°01'W (Point F)

- 26

~I- 22(j)(j)

18

1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994

28

Lat 25°56'8 Lon 45°20'W (Point G)

~I- 24(j)(j)

20

1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994

30

Lat 24°23'8 Lon 45°0S'W (Point H)

- 26

~I- 22(j)(j) , -

18

1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994

Years

Fig. 4. Satellite-derived SSTs (dots) and estimated annual SST cycIe (line) forpoints E through H. The estimated annual cycIe is approximatelysymmetric with maxima and minima SSTs observed about six monthsapart reflecting a pattem typical of midlatitudes.

Variance in SST values explained by thesinusoidal fit is quantified by the coefficient ofdetermination (?). The spatial distribution of ? isshown in Figure 8.

Although the r2 values are probablyslightly overestimated due to serial correlation,emphasis should be placed on the spatial pattemand the relative magnitudes as discussed byChelton (1983) and Podestá et aI. (1991). Fromsouth of Rio de Ia Plata estuary up to 28° S, thecoefficient of determination is highest (?>90%) onthe continental shelf. In the BMC region and northof25°S, low values (? < 80%) are observed offshoreand onshore, respectively, reflecting large rmsvalues(~ 2°C). These large values are probably dueto advective processes related to meridionaldisplacement and eddy shedding ffom the BC inthe BMC, and the strong meandering of the BCnorth of 25°S. These processes introduce SSTvariabilities that are not directly associated withthe seasonal cycle. Points A, G, and H arerepresentative of these two hydrographic features,clearly seen in the residual plots by the scatteringof residuaIs around the adjusted values (Figs 6and 7).

The closeness of point H to the well-knownupwelling region south of Cabo Frio (RJ) (CastroFilho & Miranda, 1998) is also reflected in the SSTresidual plot. Negative SST residuaIs mostly observedduring austral spring-summer season, whenupwelling favorable conditions prevail (Valentin etai., 1987), are suggestive of this scenario.

The seasonal cycle

The reiative contribution of each component(i.e., annual and semi-annual harmonics) of theestimated annual SST cycle can be evaluatedthrough the absolute magnitude of its respectiveregression coefficients. For the annual component,the coefficients are generally one or two orders ofmagnitude larger than those for the semi-annualharmonics, indicating the much greater importanceof the annual harmonics. Given the highsignificance of the annual coefficients, generalconclusions conceming their influence are probablynot invalidated by the presence of serialcorrelation. On the other hanei, if realistic tests takinginto account autocorrelation were applied, the semi-annual component might be insignificant (Podestá etaI., 1991).

98 Rev. bras. oceanogr., 48(2), 2000

Fig. 5. Contours 01"root-mean-square (rrns) (in 0c) ofthe regression fits of annual cycle. Higher rrnsvalues (~ 2°C) are observed south of 36°S in theBrazil-Malvinas Confluence region. North of theRio de Ia Plata up to Rio de Janeiro, values lowerthan 1°Ccan be seen.

Lat 38°20'8 Lon 54°27'W (Point A)

6

t:; 4

(f) 2

~ o~ .2Q) -4

tI: -6

, .,', ';.: ".".

..' ..'. .:~'.:.:. :.;' ..;.-; -,

. .~ ::. . .&;.-. .:,' .. ~

o'.. ............

.'

'. .'

1982 1983 1984 1985 '986 1987 1988 1989 1990 1991 1992 1993 1994

Lat 35°26'8 Lon 54°W (Point B)

.:-::;-- =.::;., :>:- ~~ .:../.,~ ..;...." : ~

.: .~~ :'.'/ ::. .!:::."-::-:: : ,.,.,....

.,

1982 1983 1964 1985 1986 1987 1988 1989 1990 1991 '"IQ2 1993 1994

Lat 32°41'8 Lon 50°1 o'W (Point C)

6t:; 4(f) 2

~ o

~~~

tI: -6

I :"'::.::' .~_:...~. .::.~~~,: \:::~:~.::~? : ~;~;..,~:~,:::' ::~~. ': ~:~:.-'.'

1982 1983 -1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994

Lat 31 °16'8 Lon 500W (Point D)

6t:; 4(f) 2

~ o

~~~

tI: -6

........... _.;> 0.° .:.~: "._0 ",:: n

::_:.',' ' 1;. ':" ::.....

1982- 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994

Years

r'ig. 6. Time sequence ofresiduals (dots) ftom estimated SST annual cyclefor points A through D. The largest non-seasonal variability isobserved for DOintsA and B.

LENTlNI et aI.: The ann~ cycle of SST in the WS Atlantic 99

The resulting distribution of the meanestimated SST for the austral summer (JFM) andthe austral winter (JAS) shows the basic large-scaletemperature structure for the region of study (Figs9 and 10). Estimated surface water temperaturesduring summer show small cross-shelf thermalgradients. The whole shelf has essentiallyhomogeneous temperatures. From Rio de Janeiroto the mean latitude of separation of the BrazilCurrent ttom the continental margin (Olson et ai.,1988), the region is dominated by the southwardflow of BC warm waters, with surface temperatureshigher than 20°C. Although local heating is knownto be an important factor, the orienmtion of theisotherms suggests that an increase in surfacetemperature is primarily due to the advection ofwarm waters ttom lower latitudes. In winter,however, minimum temperatures tend to be parallelto the coast and a strong thermal gradient devt<lopsoffshore (Fig. 10). In this season, subantarctic coldwaters reach their northernmost position, occupyingmost of the continental shelf portion south of thePatos Lagoon, with surface temperatures ranging

between 7°C and 14°C. North of 32°S, a strongthermal contrast (> 3° C) over the shelf-breakcan be observed. The southernmost limit ofBC istypically marked by average temperatures higherthan 18°C, extending ttom shelf to slope regions.However, one has to keep in mind that this isonly a surface feature, not totally coincident withthe sub-surface and deeper circulations.

Amplitude af the Seasonal Cycle

The estimated coefficients for the seasonalmodel are used to compute the highest and thelowest SST values for each grid point and the dayof the year at which they will occur. Theamplitude of the annual cycle is calculated as themaximum minus the minimum predicted SST.Generally, the annual amplitude (in 0c) of thesinusoidal model (Eq. 2) is maximum on theshelf. The general tendency is an amplitudedecrease ttom south to north and &om inshore tooffshore, reaching values higher than 7°C on the shelf(Fig. 11).

Lat 29°41'8 Lon 49°W (Point E)

1982 1983 1984 1985 1986 1987 1988- 1989 1990 1991 1992 1993 1994

Lat 26°31'8 Lon 47°01'W (Point F)

6

ti:; 4

(/) 2

~ o-o -2.~ -4a: -6

1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994

Lat 25°56'8 Lon.45°20'W (Point G)6

ti:; 4(/) 2

~ o

~~~

a: -6

1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994

Lat 24°23'8 Lon 45°0S'W (Point H)

...

1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994

Years

Fig. 7. Time sequence ofresiduals (dots) ttom estimated SST annual cyclefor points E through H.

100 Rev. bras. oceanogr. 48(2), 2000

(Podestá et aI., 1991). Small variations in SSTamplitude occur further north and away from theshelf-break (> 7"C), where the region is mostlydominated by warm waters of the Brazil Currentthroughout the year.

40"860"W 55"W 50"W 45"W

Fig. 8. Contours of the coefficient of determination (r2 x]00) of the regression fits of annual cycle. Valuesare expressed as percentages. High r2 values (> 90%)are near the coast, where the adjusted model tracksbetter the observed SSTs.

35"8

40"860"W 50"W 45"W55"W

Fig.9. Estimated maxima annual SSTs at each grid point(in "C) for austral summer (Jan-Feb-Mar).

The largest annual amplitudes (12-13"C)occur otl Rio de Ia Plata estuary, where depthsrange from 20 to 200 meters. On the shelf,amplitudes range from I I"C south of 33"S to 6"Cnorth of 30"S. Offshore, maximum amplitudes (10-II "C) are observed in the area of MC recirculation

35"8

40"860"W 55"W 50"W 45"W

Fig. 10. Estimaled minima annual SSTs aI each grid point(in "c) for austral winter (Jul-Aug-Sep).

30"8

35"8

40"860"W 50"W 45"W

Fig. 13. AmpJitude of lhe annual SST cycle, estimated asthe difference between the highest and lowestpredicted SSTs at each grid point. The amplitudedecreases offshore, and the highest amplitude isobserved off Rio de Ia Plata.

30"8 , /1:25"8

1I

35"8_-._._m=m --". - 30"8-

LENTINI et ai.: The annual cycle of SST in the WS Atlantic 101

Timing of the seasonal cycle

Figures 12 and 13 show the day of the yearwhen maxima and minima predicted SSTs occur.On the shelf, the general tendency for the phaseof the maximum annual SST referred to January Iis between 30 days (at the mouth of Rio de IaPlata estuary) and 70 days (in the South BrazilBight and in the BC flow region). Most of thearea, however, experiences the highesttemperatures between days 50 and 70, whichcorresponds to February (19) and March (11),respectively. In Figure 12, the timing of maximumpredicted SST is between days 30-40 south of35°S and wet of 55°W at the mouth of Rio de Ia

Plata estuary, and is between 60-70 days north of34°S and east of 50oW, extending from the coast to1000 m offshore.

Generally, south of Cabo de Santa Martathe region warms up quicker than north of thisboundary, reaching higher predicted temperaturesin February than the beginning of March. Thisimplies that the highest summer surfacetemperatures can be reached approximately 30 daysearlier in the southern portion and inshore of BCdomain.

Fig. 12. Day of the year (starting from January jSt) in whichmaximum SST is predicted to occur. Most of thearea experiences the highest surface temperaturesbetween 50 and 70 days, which corresponds toFebruary (19) and March (lI), respectively.

Fig. 13. Day of the year (starting from January jSt) in whichminimum SST is predicted to occur. The predicteddays of minima SSTs range from the beginning ofAugust (8) to the beginning of September (7).

The predicted days of minimum temperaturerange from the beginning of August (day8) to thebeginning of September (day 7). These valuescorrespond to two regions: north of the Patos Lagoonand on the shelf; off the 200-m isobath and south of

the Patos Lagoon. The temporal lag for minimumpredicted SSTs is less than that for maximumpredicted SSTs.

Discussion

The linear model provides a compactdescription of the annual SST cycle. For the 344 SSTtime series the annual component is dominant,addressing high surface thermal variability to theseasonal timescale. Annual harmonics are generallyone to two orders of magnitude larger than those forthe semiannual component, indicating the importanceof annual harmonic. Strub et aI. (1987) observed thatlimiting the harmonic fits fo annual and semiannualcomponents reduced the effect introduced by high-frequency fluctuations in short time series, which didnot happen in longer time series. Although statisticalsignificance of the regression coefficients can beinflated by serial correlation, the sinusoidal model isprobably not invalidated by the presence of serialcorrelation due to the high significance of theregression coefficients (Podestá et aI., 1991).

Indeed, the first two harmonics (i.e., annualand semiannual) plus the mean value are sufficient to

102 Rev. bras. oceanogr., 48(2), 2000

represent the SST annual cycle. The seasonal cycleaccounts for more than 90% of total SST variability.SST variations within a year clearly show theexistence of distinct seasons. The adjusted annualSST cycle can be considered as approximatelysymmetric with minima and maxima surfacetemperatures observed about six months apart. Thissinusoidal pattem, typical of midlatitudes, suggeststhat SST is forced mostly by seasonal fluctuationsin solar radiation (Seckel & Beaudry, 1973). Fortropical and subpolar latitudes, however, thesemiannual component is as important as the annualcomponent, giving an asymmetrical pattem to theseasonal SST cycle (Merle et a/., 1980; Provost elaI., 1992).

The model, thus, allows the reconstruction ofan SST field for any day of the year by using theregression coefficients from the sinusoidal model (Eq.2) with an accuracy greater than 1.0°C on the innerand mid-shelf regions. Small variances of thepredicted SSTs are near the coast, where the adjustedmodel tracks better the observed SSTs (see Fig. 6).The small variance at points where residuaIs arelarge, like in the frontal region (e.g., point A),indicates that although signals with other periods areimportant there, the annual cycle is undoubtedlyclearly defined.

Although a sinusoidal cycle with a singleannual frequency'Seems to explain accurately most ofthe temporal variations of the SSTs, one can easilydistinguish other periodic signals in the residual timeseries. The characterization of the SST annual cycleestimated here may be affected by long-term trends,obviously not completely resolved by thirteen yearsof data. For instance, analysis of the SST residuais(not shown here) do not suggest any significantupward trend as the one reported by Strong (1989).Instead, the oscillations of the residuaIs from theannual cycle about the zero line are more indicativeof a low-frequency component of SST variability thanany other long-term trend. Indeed, such interannualvariability is known to exist, as reported by recent insitu and satellite-based observations in the WSA(Campos et a/., 1996; Diaz et a/., 1998; Stevensonet a/., 1998; Campos et a/., 1999; Lentini et a/.,2001). These studies describe the occurrence ofimportant non-seasonal SST features appearingwithin a few years of each other. Using the samedataset described here, after the removal of theseasonal cycle Lentini et aI. (op. cit.) observe a totalof thirteen cold and seven warm SST anomaliesduring and right after ENSO onset in the area ofstudy.

As a consequence, these SST anomaliesradically change the local physical (Campos et a!.,1996; Lentini et a/., op. cit.) and biological dynamics(Bakun, 1993; Stevenson et aI., 1998; Sunye &Servain, 1998) in the WSA continental shelf For acompact description of the occurrence of these SST

anomalies in the area of study the reader should referto Lentini et aI. (op. cit.).

Podestá et aI. (1991) compared theestimated annual cycle with the COADS climatologyfor the Southwestem Atlantic. Even though thepredicted SST map was on aio square grid, thespatial structure of the westem boundary currentshad a much more realistic appearance than theone derived from COADS dataset. Exceptions maybe observed in the frontal region of BMC zoneand the region affected by the Rio de Ia Platadynamics. In the confluence region, uncertaintiesincrease to 2.69°C offshore, probably associated tothe energetic eddy field made up ofhigh amplitudemeanders and detached rings and eddies (Olson eta!., 1988).

Displacement of these features may causerelatively large SST changes at a particular locationnot directly related to the annual cycle. Off Rio de IaPlata the model has a relatively low accuracy of1.5°C, probably as a consequence of the temperaturedifferences between freshwater runoff and surfacecooling to adjacent open sea surface (Piola et a/.,2000).

, The estimated amplitude of the annual SSTcycle ranges between 4°C and 13°C throughout thestudy area, where a southward alongshelf gradient isbasically established. During summer, an increase oftransport by the BC advects warm waters furthersouth of its mean separation latitude (Olson et aI.,1988; Matano, 1993). This seasonal increase adds upwarm waters to the regular annual cycle contributing,thus, for the large amplitudes observed south offPatos Lagoon and off Rio de Ia Plata. During winter,however, the stronger Malvinas Current, 70 Sv(Peterson, 1992) against 20 Sv (1 Sv = 106m3s'1)(Gordon & Greengrove, 1986; Campos et a/., 1995),essentially pushes the Brazil Current northward up to300S and offshore. Coastal waters are apparentlymodified by surface heat fluxes over the Argentineancontinental shelf and by freshwater discharge fromRio de Ia Plata (Piola et a!., 2000). South of 40° S,the shelf is dominated by excess evaporation overprecipitation and continental runoff (Bunker, 1988).The region exhibits large seasonal variationsassociated with air-sea heat fluxes, which, in tum,drives large variations in density fields at seasonal tohigher frequencies. Indeed, the combination of thenorth-south seasonal displacement and the largecontrast between air and water temperatures (Provostet aI., 1992) is certainly responsible for the largeamplitude values observed.

The estimated timing of the annualmaximum SST shows a south-north lagoAt the timeof the annual maximum, most of the study area isvertically stratified. For shallow depths, smallchanges in heat input will have similar effectsthroughout a large area due to seawater thermalinertia.

LENTINI et a!.: The annual cycle of SST in the WS Atlantic 103

The region represented by point H shows a10-day temporal lag to its surrounding. This can a1sobe observed in Figure 6, where r2 values are lowerthan 85%. This local maximum cou1d be associatedwith the meandering pattem, frontal vortices pinchedoff the Brazil Current around Cabo Frio (RJ), anel/orupwelling-favorab1e conditions south of Cabo Frio(RJ). The spatial pattem of the predicted times ofminimum SST is much more uniform than that of themaximum, which a south-north lag is observed again.

As stated by Podestá et aI. (1991), thetiming for maximum and minimum SST values issensitive to the regression coefficients used in theprediction. As the coefficients should be inflated byserial correlation, the confidence intervals should benarrow and the estimated dates would be slightlyhigher. The spatial pattems for both minima andmaxima temperatures show a good estimative for thephase, since each data point in the grid is extraçtedfrom weekly images.

The annual minimum SST is estimated tooccur in August-September, whereas the annualmaximum SST is predicted to take place in February-March. Oceanographic/meteoro10gical conditions areknown to control fish distribution and abundance(e.g., Lima & Castello, 1995; Sunye & Servain,1998). Estimates of the timing of the annual cycle,thus, can be useful to understand, the occurrence,distribution, and migration of local economic fishstocks. For example, the optimum temperature for thedistribution of sardine along the Brazilian coastline isbetween 19°-26°C (Matsuura et aI., 1991; Saccardo& Rossi-Wongtschowski, 1991). Its spawning activityreaches a peak in austral summer, being essentiallyconfined within the South Brazil Bight (SBB)(Bakun, 1993). Therefore, latitudinal variations in seasurface temperatures, which may serve as a goodindicator of water column temperature, may beresponsible for sardine migration in the SBB. Asanother example, the Brazilian anchovy and theArgentine hake are spatially segregated bytemperature (Bakun, op. cit.). Both anchovy and hakemigrations occur fTom late winter to early summer(Lima & Castello, 1995; Podestá, 1990), at the timewhen shelf-break waters begin to warm up.

Conclusions

1n retrospect, this paper provides a detaileddescription of the seasonal SST cycle in the WestemSouth Atlantic continental shelf. Emphasis has beenplaced on the spatial distribution of the timing andamplitude of the annual SST cycle. The lack of longtime series of oceanographic data in the southemhemisphere makes satellite-derived SST time seriesover the area of study a particularly valuab1edata set.The high spatial resolution and quasi-synoptic

coverage of satellite-derived SST observations mayallow meaningfu1 estimates of spatia1 and temporalpattems of SST anomalies particular1y near oceanboundaries, where their closeness to 1andmakes theirinfluences more direct and strong. Coastal watersstore the sun's heat in summer and release it to theatmosphere in winter, he1pingto moderate the climateof littoral regions. AIso, as SST anoma1ies stronglyinfluence the coup1ing between ocean andatmosphere, the methodology presented here maycontribute to an understanding of interannua1variability related to climate change. Such anapproach can be seen in Campos et aI. (1999) andLentini et aI. (2001), where after removal ofthe SSTseasonal cycle, a deep investigation of interannua1SST variability is done.

Acknowledgements

This work.is a result of efforts supported bythe Inter-American Institute for Global ChangeResearch (IAI), through the SACC's CRN and ISP-IProjects, by the Conselho Nacional deDesenvolvimento Científico e Tecno1ógico (CNPq)(Proc. no. 201443/96-1), and by the Fundação deAmparo a Pesquisa do Estado de São Paulo(FAPESP) (Proc. 96/4060-0). The authors expresstheir gratitude to O. Brown and D. 01son, fTomRSMAS/Univ. of Miami, who provided the data, partof the financia1 support and guidance for the firstauthor during a visit to the USA, when a pre-ana1ysisof the AVHRR data set was done. We extend ourappreciation to the reviewers for their valuab1ecomments.

References

Bakun, A. 1993. The Califomia Current, BenguelaCurrent, and southwestem Atlantic shelfecosystems: a comparative approach to identifYingfactors regulating biomass yields. In: Sherman,K.; Alexander, L. M. & gold, B. D. eds Stressmigration and preservation of large marineecosystems. Washington, AAAS. p.199-221.

Barton, L 1. 1995. Satellite-derived sea surfacetemperatures: Current status. 1. geophys. Res.,1OO(C5):8777-8790.

Brook, R. 1. & Amold, G. C. 1985. Appliedregression analysis and experimental designoNewYork, M. Dekker. 237p.

Brown, O. B.; Brown, 1. W. & Evans, R. H. 1985.Calibration of advanced very high resolutionradiometer infTaredobservations. 1.geophys. Res.,90(C6):11667-11677.

104 Rev. bras. oceanogr., 48(2), 2000

Bunker, A. F. 1988. Surface energy fluxes of SouthAtlantic Oceano Mon. weath. Rev., 116(4):809-823.

Campos, E. 1. D.; Lentini, C. A. D.; MiIler, 1. L. &Pio1a, A. R. 1999. Interannua1 variabi1ityofthe sea surface temperature in the SouthBrazi1 Bight. Geophys. Res. Letts, 26(14):2061-2064.

Campos, E. 1. D.; Ikeda, Y; Castro Filho, B. M.;Gaeta, S. A.; Lorenzzetti, J. A. & Stevenson, M.R. 1996. Experiment studies circu1ation in thewestern South Atlantic. EOS Trans. Am. geophys.Un., 77(27):253-259.

Campos, E. 1. D.; Gonçalves, 1. E. & Ikeda, Y. 1995.Water mass characteristics and geostrophiccircu1ation in the South Brazi1 Bight-Summer of 1991. 1. geophys. Res.,100(Cll):18537-18550.

Casey, K. S. & Comillon, P. 1999. A comparison ofsatellite and in-situ-based sea surfacetemperature c1imatologies.1. Climate, 12(6):1848-1863.

Castro Filho, B. M. & Miranda, L. B. 1998. Physicaloceanography of the western Atlanticcontinental shelf located between 4°N and 34°Scostal segment (4'W). In: Robinson, A. R. &Brink, K. H. The Sea. Oxford, John Wiley &Sons. p.209-25I.

Castro Filho, B. M.; Miranda, L. B. & Miyao, S. Y1987. Condições hidrográficas na plataformacontinental ao largo de Ubatuba: variaçõessazonais e em média escala. Bolm Inst. oceanogr.,S Paulo, 35(2):135-151.

Chelton, D. B. 1983. Effects of samp1ing errors instatistical estimation. Deep-Sea Res., 30(10):1083-1103.

Cleveland, W. S.; Freeny, A. E. & Graedel, T. E.1983. The seasonal component of atmosphericCOz - Information IToma new approaches to thedecomposition of seasonal time series. J.geophys. Res., 88(CI5):10934-10946.

Diaz, A. F., Studzinski, C. D. & Mechoso, C. R.1998. Relationships between precipitationanomalies in Uruguay and Southern Brazil and seasurface temperature in the Pacific and AtlanticOceans. 1. Climate, II (2):251-271.

Emilsson, r. 1961. The shelf and coastal waters offsouthem Brazil. Bolm Inst. oceanogr., S Paulo,11:101-112.

Gacic, M.; Marullo, S.; Santoleri, R. & Bergamasco,A. 1997. Analysis of the seasonal andinterannual variabi1ity of the sea surfacetemperature field in the Adriatic Sea fi-omAVHRR data (1984-1992). J. geophys. Res.,102(C 10):22937-22946.

Garzo1i, S. L. & Garraffo, Z. 1989. Transports,Frontal Motions, and Eddies at the Brazil-Malvinas Currents Confluence. Deep-Sea Res.,36(5A):681-703.

Gordon, A. L. 1989. Brazil-Malvinas confluence1984. Deep-Sea Res., 36(3A):359-384.

Gordon, A. L. & Greengrove, C. L. 1986.Geostrophic circulation of the Brazil-Falk1andConfluence. Deep-Sea Res., 33(5A):573-585.

Hore1,1. D. 1982. On the annual cyc1eof the tropicalPacific atmosphere and oceano Mon. Weath. Rev.,110(12):1863-1878.

Kidwell, K. B. 1991. NOAA Polar Orbiter DataUser's Guide (TIROS-N, NOAA-6, NOAA-7,NOAA-8, NOAA-9, NOAA-10, NOAA-ll, andNOAA-12). Washington, NOAA/NESDISSatellite Data Services Division, De.

Lentini, C. A. D.; Podestá, G. P.; Campos, E. 1. D. &0lson, D. B. 2001. Sea surface temperatureanomalies on the westem South Atlantic fi-om1982 to 1994. Continent. Shelf Res. 21(I ):89-112.

Lima, r. D. & Castello, 1. P. 1995. Distribution andabundance of South-west Atlantic anchovyspawners (Engraulis anchoita) in relation tooceanographic processes in the southem Brazilianshelf. Fish. Oceanogr., 4(1):1-16.

Matano, R. P. 1993. On the separation of the BrazilCurrent fi-om the coast. 1. phys. Oceanogr.,23(1):70-90.

Matsuura, Y.; Spach, H. L. & Katsuragawa, M. 1991.Comparison of spawning pattems of the Braziliansardine (Sardinella brasiliensis) and anchoita(Engraulis anchoita) in Ubatuba Region, SouthernBrazil during 1985 through 1989. ICES C. M.,H22. 25p.

McClain, E. P.; PicheI, W. G. & Walton, C. C. 1985.Comparative performance of AVHRR-basedmultichannel sea surface temperatures. 1.geophys. Res., 90(C6):11587-11601.

Merle, J.; Fieux, M. & Hisard, P. 1980. Annualsignal and interannual anomalies of seasurface temperature in the Eastem EquatorialAtlantic Oceano Deep-Sea Res. II, Part A,26(2):77-101.

LENTINI et a!.: The annual cycIe af SST in the WS Atlantic 105

Olsan, D. 8.; Podestá, G. P.; Evans, R H. & Brown,O. 8. 1988. Temporal variations in the separationof Brazil and Malvinas Currents. Deep-Sea Res.,35(12): 1971-]990.

Peterson, R. G. ]992. The boundary currents in thewestem Argentine Basin. Deep-Sea Res., 39(3-4A):623-644.

Peterson, R G. & Stramma, L. ]991. Upper-Ievelcirculation in the South Atlantic. Ocean Prog.Oceanogr.,26(]):]-73.

Piola, A. R.; Campos, E. J. D.; Mõller, O. O.; Charro,M. & Martinez, C. 2000. The subtropical shelfffont off eastem South America. J. geophys. Res.,]05(C3):6565-6578.

Podestá, G. P.; O. 8. Brown & Evans, R H. ]991.The annual cycle of satellite-derived sea surfacetemperature in the Southwestem Atlantic OceanoJ. Climate, 4(4):457-467.

Podestá, G. P. ]990. Migratory pattem of Argentinehake (Merluccius hubbisi) and oceanic processesin the southwestem Atlantic Oceano Fish. BulI.,88(1):]67-]77.

Provost, c.; Garcia, O. & Garçon, V. 1992. Analysisof satellite sea surface temperature time series inthe Brazil-Malvinas Current Confluence region:dominance of the annual and semiannual periods.J. geophys. Res., 97(C] ]):1784]-]7858.

Saccardo, S. A. & Rossi-Wongtschowski, C. L. D. 8.]991. Biologia e avaliação do estoque de sardinhaSardinella brasiliensis: uma compilação.Atlântica, Rio Grande, 13(1):29-43.

Schwalb, A. ]978. The TIROS-NINOAA A-Gsatellite series. NOAA Tech. Memo. NESS 95,p.75.

Seckel, G. R. & Beaudry, F. H. ]973. The radiationffom sun and sky over the North Pacific OceanoTrans. Am. geophys. Un., 54(1]):]] ]4.

Stevenson, M. R; Dias-Brito, D. D.; Stech, J. L. &Kampel, M. ]998. How cold water biota arrive intropical bay near Rio de Janeiro, Brazil?Continent. ShelfRes., ]8(13):]595-1612.

Strong, A E. ]989. Greater global warming revealedby satellite-derived sea surface temperaturetrends. Nature, 338(62] 7):642-645.

Strub, P. T.; Allen, J. S.; Huyer, A; Smith, R L. &Beardsley, R C. 1987. Seasonal cycles ofcurrents, temperatures, winds, and sea levei overthe northeast Pacific continental shelf: 35"N to48°N. J. geophys. Res., 92(C2):1507-]526.

Sunye, P. S. & Servain, J. ]998. Effects of seasona]variations in meteorology and oceanography onthe Brazilian sardine fishery. Fish. Oceanogr.,7(2):89-] 00.

Valentin, J. L.; Andre, D. L. & Jacobs, S. A ]987.Hydrobiology in the Cabo Frio (Brazi]) upwelling:two-dimensional structure and variability during awind cycle. Continent. She]f Res., 7(1):77-88.

Walton, C. c.; Pichei, W. G.; Sapper, J. F. & May, D.A. ]998. The development and operationa]application of nonlinear algorithms for themeasurement of sea surface temperatures with theNOAA polar-orbiting environmental satellites. J.geophys. Res., ]03(C]2):27999-280]2.

Wyrtki, K. 1965. The annual and semiannualvariation of sea surface temperature in the NorthPacific Oceano Limnol. Oceanogr., ]0(3):307-3 ]3.

(Manuscript received 18 January 2000; revised03 March 2000; accepted 21 September 2000)