Seasonal and interannual variability of currentsand temperature off central Chile

Gary Shaffer,1 Samuel Hormazabal,2,3 Oscar Pizarro,4,5,6 and Sergio Salinas7

Abstract. Results are reported for continuous, 6 year records of current and temperaturefrom a continental slope station at 308S off Chile. These observations for November 1991to November 1997 span the moderate, weak, and strong El Nino events of 1991–1992,1994, and 1997 and the La Nina event of 1995–1996. Results for current and temperatureare analyzed together with observations of wind along the South American coast from theERS-1 and -2 satellite scatterometers, wind from a local coastal site, sea level fromnorthern and central Chile, sea surface temperature anomalies from the South Pacificregion, and traditional El Nino indices. Mean poleward flow of 12.8 cm s21 was observedfor the 6 year period in the core of the Peru-Chile Undercurrent (220 m). Mean flow inthe depth range of Antarctic Intermediate Water (750 m) was equatorward at 1.1 cm s21.Intraseasonal-scale variability associated with coastal trapped waves was weaker during theaustral winter: greatest during El Nino events and weakest during the La Nina event.Winds adjacent to the coast showed a strong annual cycle with a southward progression ofmaximum equatorward (upwelling favorable) winds. Alongshore flow over the slopeexhibited strong semiannual and weaker annual variations. Poleward flow in theundercurrent was strongest in the austral spring and fall and weakest in the winter. Nearthe bottom (750 m), seasonal-scale variability was less intense but led that higher up inthe water column (220 m) by ;1 month. Strong interannual variability of alongshorecurrent and temperature was found at all depths over the slope, whereby changes at depthled changes higher up. A rapid warming event was observed at the end of 1996–beginningof 1997 over the slope. This event, also seen in sea level and sea surface temperature offChile, led El Nino indices by several months and may be best explained by local windanomalies coupled to atmospheric teleconnections from the western tropical Pacific at theonset of El Nino. Some aspects of our current and temperature observations areconsistent with results from eastern boundary current models. Our results indicate thatboth ocean and atmosphere pathways from the equatorial Pacific contribute significantlyto interannual-scale variability of alongshore flow and thermocline depth off Chile.

1. Introduction

Much remains to be learned about ocean circulation in theeastern South Pacific and the variability of this circulation onintraseasonal, seasonal, interannual, and decadal timescales.Until recently, direct observations of currents needed to ad-dress these points were lacking in this vast and remote region.The first such observations in the eastern boundary current off

Chile, the Peru-Chile Current, were reported recently [Shafferet al., 1995, 1997] (the latter hereinafter referred to as CH2).These observations at 308S during the moderate 1991–1992 ElNino event showed strong intraseasonal (;50 day period) vari-ability over the slope as free coastal trapped waves. Alongshorecurrent variability in this region then was indeed dominated bysuch waves. Examples of this were several reversals of thepoleward Peru-Chile Undercurrent, which during 1991–1992flowed with a mean speed of 12 cm s21 at ;200 m depth overthe slope. It was shown that the coastal trapped waves origi-nated as equatorial Kelvin waves forced by a series of westerlywind bursts of similar period near the dateline in the equatorialPacific (CH2). These wind bursts were apparently an expres-sion of strong Madden-Julian oscillations in the atmosphere[Madden and Julian, 1971] during the onset of the 1991–1992El Nino event [Kessler and McPhaden, 1995].

The observations considered above spanned ;1 year duringan El Nino event. Clearly, in a region like the Peru-ChileCurrent under the influence of strong interannual variability,observations over a much longer period are required to assessaccurately the mean circulation, the seasonal circulation cycle,and possible modulation of intraseasonal variability on sea-sonal and interannual timescales. Such long-term observationsare obviously needed to capture interannual variability ofalongshore flow and nearshore hydrography associated with El

1Danish Center for Earth System Science, Niels Bohr Institute forAstronomy, Physics and Geophysics, University of Copenhagen,Copenhagen, Denmark.

2Regional Program in Physical Oceanography and Climate, Univer-sity of Concepcion, Concepcion, Chile.

3Now at Danish Center for Earth System Science, Niels Bohr Insti-tute for Astronomy, Physics and Geophysics, University of Copenha-gen, Copenhagen, Denmark.

4Department of Earth Sciences, University of Goteborg, Goteborg,Sweden.

5Now at Regional Program in Physical Oceanography and Climate,University of Concepcion, Concepcion, Chile.

6Also at School of Marine Sciences, Catholic University of Val-paraiso, Valparaiso, Chile.

7School of Marine Sciences, Catholic University of Valparaiso, Val-paraiso, Chile.

Copyright 1999 by the American Geophysical Union.

Paper number 1999JC900253.0148-0227/99/1999JC900253$09.00

JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 104, NO. C12, PAGES 29,951–29,961, DECEMBER 15, 1999

29,951

Nino-La Nina cycles. Huyer et al. [1991] discuss some of thesepoints on the basis of more than 3 years of moored current andtemperature data on the shelf and slope off Peru in 1981–1985.They found that the Peru-Chile Undercurrent showed littleseasonal variation and was not modulated much by the strong1982–1983 El Nino event. On the other hand, they foundintraseasonal variability to be significantly modulated by theevent.

Here we use a 6 year data set of moored current and tem-perature observations over the slope at 308S to address theseasonal and interannual variability of alongshore flow andstratification in the Peru-Chile Current off Chile. These obser-vations are a direct extension of the measurement seriesstarted at this site in 1991 and discussed by CH2. They areapparently the longest essentially continuous current recordsfrom the eastern South Pacific Ocean (among the longest suchrecords from any eastern boundary current region) and span arange of weak to strong El Nino events and a La Nina event.To help in the interpretation of our slope observations, we alsoconsider wind and sea level data from the west coast of SouthAmerica as well as sea surface temperature anomaly data fromthe South Pacific region. In addition, we discuss our results inthe context of modeling work on wind-driven and remotelyforced eastern boundary current circulation and El Nino/Southern Oscillation (ENSO) coastal currents.

2. Observations and MethodsBetween November 4, 1991, and November 3, 1997, a re-

cording current meter (RCM) mooring was maintained on thecontinental slope off Chile at 30819.09S and 71847.09W in awater depth of ;875 m, 10 km off the coast and 70 km shore-ward from the Peru-Chile Trench (circled dot in Figure 1).Depths of more than 500 m extend to within a few kilometersof the coast locally. The mooring was recovered and rede-

ployed approximately twice yearly; the resulting 11 observationperiods are listed in Table 1. There were gaps of almost 3months between periods 2 and 3 and of 2–3 weeks betweenperiods 5 and 6 and periods 8 and 9. Otherwise, the mooringwas redeployed within 24 hours after recovery. The mooringhad three Aanderaa RCM 7 current meters at nominal depthsof 220, 485, and 750 m. Small variations in the mooring posi-tion among the deployment periods led to changes in thesedepths of up to 30 m. However, vertical separations betweenthe meters remained constant with the deepest RCM at 125 mover the bottom. All RCMs had speed, direction, and temper-ature sensors, and the shallowest one had a pressure sensorduring some observation periods. No attempt was made tocorrect the data for “knockdown,” although pressure dataand/or mooring motion simulations using observed currentsshowed vertical displacements of up to 100 m at the shallowestRCM in extreme flow events. Simulations indicate knockdownat the intermediate and deep RCMs of up to 30 and 10 m,respectively.

RCM temperature data were calibrated with conductivity-temperature-depth (CTD) casts made at the mooring site. Theoriginal hourly RCM current and temperature data were low-pass filtered with a 181-point Cosine-Lanczos filter with its halfamplitude point at 60 hours. Three month running means wereapplied twice to the data to suppress intraseasonal variability.All time series treated this way are referred to below as low lowpassed. Significance levels for correlations between time seriespairs were based on integral timescales [Davis, 1976]. Windspeed and direction were measured throughout the study pe-riod on an automatic weather station at Lengua de Vaca Pointeast of the mooring site (triangle in Figure 1, data courtesy ofJ. Rutllant, Universidad de Chile). Data gaps during the firstyear were filled with data from another automatic station 60km to the south (CH2). Gaps of up to 5 days in 1993–1995 (but8 days from August 26 to September 3, 1995) were filled usingautoregression analysis. A large data gap from September 3 toNovember 30, 1997, was filled by data from a third automaticstation 2 km south of Coquimbo (Figure 1). Wind stress wascalculated from the hourly data using a drag coefficient of1.3 3 1023 and an air density of 1.2 kg m23, and the resultswere low-pass and low low-pass filtered as above. These ob-servations were part of a six year, international study of thePeru-Chile Current system in association with the Joint GlobalOcean Flux Study.

Monthly averaged data of alongshore wind stress fromERS-1 and ERS-2 satellite scatterometer observations werecalculated for 40 18 3 18 grids off the west coast of SouthAmerica between the equator and 408S from August 1991 toFebruary 1998. These data are distributed (on CD) by Depar-tement d’Oceanographie Spatiale, Institut Francais de Recher-che pour l’Exploration de la Mer (IFREMER), France (http://www.ifremer.fr/droos) as a World Ocean CirculationExperiment (WOCE) contribution. Data processing and vali-dation details are found in the Scientific User Manual on theCD. The ERS scatterometers provide wind speed and direc-tion at 50 km resolution with a separation of 25 km across a 500km swath. The number of observations used in formingmonthly means varied in time and space but consisted typicallyof 12 to 16 18 3 18 observations. Validation of ERS scatterom-eter winds with buoy wind measurements yielded rms errors of1.20 m s21 for speed and 248 for direction [Graber et al., 1996].For each scatterometer wind vector observation, eastward(northward) wind stress components were taken to be propor-

Figure 1. The study area and locations of moored currentmeter observations over the slope (open circle) and of windobservations at Lengua de Vaca point (triangle). Also shownare locations of sea level and air pressure observations (Aricaand Valparaiso) and of the near-shore 18 3 18 grid monthlywind data along the west coast of South America from theequator to 408S (crosses) on the basis of the ERS-1 and -2satellite scatterometer observations. The shaded box in Figure1a marks the location of the local 18 3 18 grid monthly winddata from the ERS observations.

SHAFFER ET AL.: SEASONAL AND INTERANNUAL VARIABILITY OFF CHILE29,952

tional to the product of the wind speed and eastward (north-ward) wind components. The calculations used a wind speed–dependent drag coefficient [Smith, 1988]. A 1-month gap inFebruary 1992 was filled by linear interpolation between Jan-uary and March. The closest grids to the coast with valid datawere identified (crosses in Figure 1) and mean alongshoredirection for each grid was estimated by fitting a straight line toa piece of adjacent coastline determined from the ETOP05 593 59 data set (www.ngdc.noaa.gov/mgg/global/seltopo.html).The chosen length of the coastal piece depended upon localcoastline complexity but was typically several degrees in north-south extension. The time series in each grid was low low-passfiltered as above.

Sea level data for the study period were obtained from tidegauges at Arica and Valparaiso (Figure 1). These data asmonthly mean “research quality” values were obtained fromthe Sea Level Center, University of Hawaii (M. Merrifield,Director). Quality control (including compensation for changein zero level in this seismically active region) and data process-ing details can be found at http://www.soest.hawaii.edu/kilonsky/uhslc.html. Atmospheric pressure data (three obser-

vations daily) were obtained from the Arica airport (DirectionMeteorologica de Chile) and from Punta Angeles (near Val-paraiso; Servicio Meteorologico de la Armada de Chile).Monthly means were calculated from this data and were sub-sequently added to the monthly sea level data (scale factor 1cm mbar21) to form adjusted sea level (ASL), a measure oftotal subsurface pressure. The ASL time series were then lowlow passed as above. Monthly mean sea surface anomaly datafor the South Pacific region in January, April, and July 1997were obtained from Integrated Global Ocean Services System(Y. Tourre, Scientific Editor, http://ingrid.ldeo.columbia.edu/SOURCES/.IGOSS/.).

3. Seasonal and Interannual ForcingModels of eastern boundary current circulation on seasonal

to interannual timescales indicate that observed flow in ourstudy region may depend upon wind forcing along the westcoast of South America north of 308S and upon remotelyforced current fluctuations originating in the equatorial PacificOcean [McCreary et al., 1987; Clarke and Van Gorder, 1994]. In

Table 1. Means, Variances, and Eddy Kinetic Energies of Low-Pass Current Component Records From the Slope Mooring

Observation Periodu# ,

cm s21v# ,

cm s21uV umaxcm s21

u92,cm2 s22

v92,cm2 s22

u9v9,cm2 s22

Kc,cm2 s22

220 m911105–920420 (1) 1.4 212.1 68.9 58.5 469.7 38.6 264.1920421–921129 (2) z z z z z z z z z z z z z z z z z z z z z930126–930722 (3) 20.6 212.6 47.3 59.4 335.5 21.2 197.5930723–940125 (4) 21.2 29.5 36.6 28.8 180.7 14.5 104.7940126–940905 (5) 21.1 211.1 50.5 35.6 214.2 27.6 124.9940927–950129 (6) 2.2 215.6 47.3 20.0 293.5 30.6 156.8950130–950604 (7) 21.9 214.8 52.4 16.5 220.4 1.7 118.4950604–960113 (8) 21.2 25.9 43.9 42.8 206.4 7.1 124.6960127–960720 (9) 2.0 216.9 40.7 9.3 189.5 21.4 99.4960721–970210 (10) 21.4 217.0 46.3 20.2 209.7 19.4 114.9970211–971103 (11) 0.2 214.9 50.5 44.7 356.0 1.2 200.4Period mean 20.3 212.8 68.9 36.8 279.4 13.4 158.1

485 m1 0.8 1.1 34.6 5.0 189.1 7.6 97.02 20.5 23.6 26.3 4.8 80.0 5.4 42.43 20.3 22.4 29.2 7.3 118.2 3.4 62.74 z z z z z z z z z z z z z z z z z z z z z5 20.8 20.4 23.2 3.1 114.3 20.6 58.76 0.5 21.8 22.2 6.2 82.0 25.1 44.17 20.5 23.6 27.4 3.7 70.4 6.9 37.08 20.3 20.4 21.5 3.5 57.2 21.9 30.39 z z z z z z z z z z z z z z z z z z z z z

10 0.1 23.9 27.4 4.2 79.5 4.5 41.911 20.2 21.6 42.8 6.8 186.2 0.7 96.5

Period mean 20.2 21.8 42.8 5.1 114.3 2.4 59.7

750 m1 20.3 2.1 35.5 2.0 167.1 23.4 84.52 20.3 21.8 33.1 2.2 89.2 24.2 45.73 z z z z z z z z z z z z z z z z z z z z z4 20.4 20.2 18.0 1.8 70.3 23.9 36.05 20.5 4.1 37.6 3.4 122.8 29.7 63.16 0.3 3.2 24.3 5.4 35.4 24.6 20.47 0.0 21.7 15.2 1.8 39.6 21.6 20.78 0.2 0.4 15.0 1.9 15.0 21.1 8.59 0.4 1.9 13.4 1.9 18.4 21.3 10.1

10 0.3 21.0 18.0 1.7 30.8 22.9 16.211 0.5 3.0 42.5 4.0 143.3 28.8 73.6

Period mean 0.0 1.1 42.5 2.7 81.8 24.4 42.3

Positive cross-shore flow u and alongshore flow v are toward 1058 and 0158, respectively. The eddy kinetic energy is half the sum of the currentcomponent variances. uV umax is the maximum speed observed in the filtered records. Also included are total study period means at eachobservation depth (maximum for total period in the case of uV umax).

29,953SHAFFER ET AL.: SEASONAL AND INTERANNUAL VARIABILITY OFF CHILE

these models, alongshore flow over the shelf and slope is sen-sitive to alongshore winds at the coast equatorward of a localsite, through Ekman divergence at the coast, and to local windstress curl over the slope and shelf, through Ekman pumpingand resulting modification of the local Sverdrup transport.

Alongshore winds along the west coast of South Americawere equatorward (coastal upwelling favorable) north of 378Sover the study period with maxima near 158 and 308S and aminima near 208S (Figure 2). Maximum equatorward windsoccur in the austral winter north of 208S. The seasonal maxi-mum occurs progressively later to the south, reaching thespring near 308S and the summer south of ;358S. Polewardpropagation from spring through summer of the equatorwardwind stress maximum also occurs off California and Oregon[Strub et al., 1987]. In the last half of 1996 a stronger-than-usual, more simultaneous maximum developed along the Chilecoast south of ;258S with maximum winds shifted south of308S. This event collapsed into weaker-than-usual equatorwardwinds by early 1997. The collapse occurred at the end of the1996 La Nina event but before the strong 1997 El Nino eventas defined by sea surface temperature from the Nino 3.4 sectorin the eastern equatorial Pacific and the Southern OscillationIndex (Figure 3). The transition from the southward propagat-ing mode to the more simultaneous mode of wind variabilityhad occurred already at the onset of 1996 during peak La Ninaconditions (Figure 2). There is a tendency for more simulta-neous structure during the 1991–1992 El Nino event as well.Finally, note that winds were equatorward between 0 and 408Sthroughout all of 1996 and that upwelling favorable windsnorth of 208S were stronger than usual during El Nino events.

Low low-pass ASL from Arica at Chile’s northern tip (Fig-ure 1) shows little seasonal variation but tracks interannualvariability of the El Nino indices (Figure 3). ASL from Val-paraiso, 1600 km farther south (Figure 1), is also characterizedby interannual variability but only during the last two recordyears do both ASL curves coincide: The strong 1997 El Ninodominates ASL at both sites, whereas the 1991–1992 eventstrongly influences the Arica record but not the Valparaiso

Figure 2. Low low-pass (twice filtered with 3 month running means) alongshore wind stress for the 1991–1997 study period along the west coast of South America from the equator to 408S as estimated for nearshore18 3 18 grids (crosses in Figure 1) from the ERS-1 and -2 satellite scatterometer observations (see also text).

Figure 3. (a) Low low-pass (twice filtered with 3 month run-ning means) anomalies of sea surface temperature for the Nino3.4 region (58N–58S, 1208–1708W) for the 1991–1997 studyperiod (sea surface temperature (SST) Nino 3.4) and low low-pass (as above) Southern Oscillation Index (SOI). The SOI isunitless (proportional to the difference between Darwin andTahiti sea level pressures) and is multiplied by 21 here. (b)Low low-pass (as above) adjusted sea level (ASL) series frommonthly mean ASL at Arica and Valparaiso (Figure 1) for the1991–1997 study period. (c) Low low-pass (as above) series ofalongshore wind stress from monthly mean wind stress basedon wind observations at Lengua de Vaca (Figure 1) and on theERS-1 and -2 satellite scatterometer observations from a 18 318 grid in the study area (shaded box in the Figure 1).

SHAFFER ET AL.: SEASONAL AND INTERANNUAL VARIABILITY OFF CHILE29,954

record. Low low-pass wind stress from the Lengua de Vacacoastal site (triangle in Figure 1) shows upwelling favorablewinds throughout the period and a quite regular annual cyclewith an austral spring maximum. Both annual mean wind stressand annual cycle amplitude decrease throughout the first 5years of the record there. This is also seen in the low low-passERS alongshore wind stress (Figure 3) from the adjacent 18 318 square ;100 km farther seaward (shaded square in Figure1). Taken together, local ERS and Lengua de Vaca wind stressdata indicate that equatorward wind stress and therefore sea-ward Ekman transport may be stronger at the offshore sitethan at the coastal site. There is a strong exception to this ruleat the onset of 1991–1992 and 1997 El Nino events whenequatorward winds greatly weaken at the offshore site but notat the coastal site.

Above results show that ASL at Arica and Valparaiso as wellas offshore wind stress moved rapidly into El Nino-like condi-tions at the end of 1996-beginning of 1997. This was severalmonths before the strong 1997 El Nino appeared in the tradi-tional indices from the equatorial Pacific (Figure 3). Low low-pass sea surface temperature (SST) at Nino 3.4 is highly cor-related (to 99% significance level) with low low-pass ASL atArica during the period 1995–1997, whereby ASL off Chileleads SST in the eastern equatorial Pacific by 2 months atmaximum correlation.

4. Moored Current and Temperature Results4.1. Current Structure and Statistics

The low-pass current records from the three observationdepths at the slope site are shown in Figure 4 as successivevectors. Flow in the Peru-Chile Undercurrent characterizes therecord at 220 m, while mean flow is weak at the other twodepths. Significant low-frequency fluctuations, in particular,the ;50 day period coastal trapped waves, pervade the recordsat all depths. These results show the waves to be weakest and

less frequent during the austral winter and strongest at theonset of El Nino events. This is consistent with the equatorialorigin of these waves (CH2) since equatorial Kelvin waves ofthis period are strongest during event onsets [Kessler andMcPhaden, 1995]. Another notable feature, most visible in therecord at 750 m, is weak, low-frequency variability during the1995–1996 La Nina event.

Means, variances, covariances, and eddy kinetic energies forthe low-pass RCM records at the slope station are shown inTable 1. Current meters malfunctioned during period 2 at220 m, periods 4 and 9 at 485 m, and period 3 at 750 m, butotherwise, good data were collected. Cross-shore (u) andalongshore (v) current components are taken to be positivetoward 1058 and 0158, respectively, as by CH2. Mean polewardflow was observed at 220 m depth throughout the study period,yielding a 6 year mean speed of 12.8 cm s21 in the core of thePeru-Chile Undercurrent (compare to 12.0 cm s21 by CH2).This tends to confirm the CH2 transport estimate for theundercurrent off central Chile of ;1 Sv. Similar transportestimates have been obtained for the undercurrent off Peru[Huyer et al., 1991] and the poleward undercurrent off centraland northern California [Huyer et al., 1989]. Mean alongshoreflow below the undercurrent core was weak and poleward at485 m at our slope station (1.8 cm s21 compared to 1.6 cm s21

by CH2). At 750 m depth, near the core of Antarctic Interme-diate Water in the study region [Shaffer et al., 1995], meanalongshore flow was weak and equatorward (1.1 cm s21 com-pared to 0.0 cm s21 by CH2). Flow near the surface is generallyequatorward in the study region as indicated by geostrophiccalculations [Shaffer et al., 1995].

Mean cross-slope flow was weak at all three depths withslight seaward flow of 20.3 cm s21 at 220 m depth and 20.2 cms21 at 485 m depth. These values are not significantly differentfrom zero and are quite sensitive to coordinate choice, partic-ularly at 220 m in the presence of strong alongshore flow.Mean observed equatorward wind stress in the study region of

Figure 4. Low-pass time series of vector current (upward 3 08) observed at 220, 485, and 750 m depth atthe slope station (Figure 1, water depth 875 m) over the 1991–1997 study period. Original hourly data werelow-pass filtered with a 181-point Cosine-Lanczos filter with a half amplitude point at 60 hours and resampledevery 2 days.

29,955SHAFFER ET AL.: SEASONAL AND INTERANNUAL VARIABILITY OFF CHILE

;0.05 N m22 (Figure 2) would lead to a mean seaward Ekmantransport of ;0.7 m2 s21 in the surface layer there. A subsur-face, shoreward flow to balance this transport would have amean speed of 0.7 or 0.35 cm s21 if spread over 100 or 200 m,respectively. Taken together, these results imply that suchcompensation flow takes place above 220 m in the study area.Even our long records are not sufficient to resolve mean cross-slope flow at our observation depths in the presence of strongseasonal and interannual variability. Other approaches wouldbe needed to resolve this flow [cf. Shaffer, 1986].

Alongshore current variance and eddy kinetic energy exhibitstrong interannual variability with the greatest variance during

moderate to strong El Nino events (observation periods 1 and11). However, there appears to be no simple relationship be-tween the strength of the variability and the intensity of the ElNino events: intraseasonal, coastal trapped waves were appar-ently as strong or stronger during the 1991–1992 El Nino asduring the strong 1997 event (see also Figure 4). Alongshorecurrent variability was very low at 485 and 750 m (but not at220 m) during the 1995–1996 La Nina event.

4.2. Seasonal Variability

Figure 5 shows low low-pass alongshore current and tem-perature records from the slope station over the study period.The Peru-Chile Undercurrent (220 m) varies considerably onseasonal and interannual timescales. Semiannual fluctuationscharacterize the seasonal-scale, current variability there, asconfirmed by harmonic analysis (Table 2). This contrasts mark-edly with the wind stress considered above, which is dominatedby the annual cycle, in particular at the coastal site (Table 2).The low low-pass temperature record at 220 m shows seasonalvariability and a large interannual fluctuation during the last2–2.5 years of the record. For temperature, seasonal-scale vari-ability has a relatively strong annual component (Table 2),probably related to the annual heating cycle. Such variability isconsiderably weaker at 485 m, but the large interannual tem-perature fluctuation persists at this depth. At 750 m the inter-annual component of alongshore flow dominates over theweak seasonal component (Table 2). The low low-pass tem-perature time series there contains weak seasonal and inter-annual variations, but the interannual fluctuation is also ob-served at this depth.

We combined amplitudes and phases of the semiannual andannual harmonics of low low-pass alongshore flow at 220 m(Table 2) and the mean poleward flow of 12.8 cm s21 there forthe study period to estimate an average yearly cycle in thePeru-Chile Undercurrent. This cycle features maxima in pole-ward flow at the end of April (15.8 cm s21) and the beginningof November (15.6 cm s21), a strong minimum of polewardflow at the beginning of August (8.2 cm s21), and a weakerminimum at the beginning of February (11.8 cm s21). A similarcalculation for current data from 750 m at our slope station(Tables 1 and 2) yields equatorward flow maxima at the end ofJanuary (1.6 cm s21) and at the June–July transition (3.0 cms21), a relative minimum in such flow at the March–April

Figure 5. Low low-pass time series of alongshore currentand temperature observed at 220, 485, and 750 m depth at theslope station over the 1991–1997 study period. Original hourlydata were low-pass filtered as for Figure 4 and twice filteredwith a 3 month running mean applied in 10 day steps over thetime series. Positive alongshore flow is toward 0158. Thestraight, slanted lines mark temperature minima and maximaof the La Nina-El Nino transition in 1996–1997.

Table 2. Amplitude, Phase, and Variance Explained for Annual and Semiannual Harmonicsof Low Low-Pass Alongshore Current, Wind Stress, and Temperature Records

Annual Semiannual

Amplitude Phase % Var Amplitude Phase % Var

ERS alongshore wind stress 1.41 308 16.9 0.67 96,278 3.8LdeV alongshore wind stress 1.17 315 54.5 0.18 113,295 1.3Alongshore current (220 m) 1.79 218 7.4 2.78 33,215 17.8Alongshore current (750 m) 1.15 143 10.1 1.09 11,193 9.1Temperature (220 m) 0.31 121 19.4 0.14 15,197 3.8Temperature (750 m) 0.04 56 3.8 0.06 162,344 7.3

Amplitude is in 1022 N m22 for wind stress, cm s21 for currents, and 8C for temperature. Phase is innumber of days after January 1 for maxima of equatorward flow (wind stress and alongshore current) ortemperature. The % Var is variance explained in the combined seasonal-interannual band. The analysisperiod is September 1993 to August 1997, the longest, common, low low-pass record period for currentobservations at 220 and 750 m depth. Low low-pass alongshore wind stress records derive from monthlymean values calculated from the ERS-1 and -2 satellite wind data (ERS; averaged over the shaded 18 318 square in Figure 1) and from continuous wind observations at Lengua de Vaca point (LdeV; trianglein Figure 1).

SHAFFER ET AL.: SEASONAL AND INTERANNUAL VARIABILITY OFF CHILE29,956

transition (0.8 cm s21), and an absolute minimum in the mid-dle of October (20.9 cm s21, i.e., poleward flow). Table 2results and the discussion above show that mean yearly cyclesat 220 and 750 m are similar in form. For example, yearly cyclesat both depths exhibit a strong transition from equatorward topoleward current perturbation during the late austral winter–early austral spring. These results also show that the averageyearly cycle at 750 m leads that at 220 m by ;1 month.

4.3. Interannual Variability

Low low-pass temperature data from the moored currentmeters show that stratification over the slope is strongly mod-ulated by El Nino-La Nina cycles (Figure 5). Temperatures arerelatively high at all depths during the weak 1994 event and fallgradually through 1995 to the last half of 1996 during La Ninaconditions. In response to the strong 1997 El Nino, tempera-tures rise relatively sharply at all three observations depths toexceed 1994 levels. The temperature increases from minima in1996 to maxima in 1997 were 2.18, 2.358, and 0.558C at 220, 485,and 750 m, respectively. From our CTD casts, vertical temper-ature gradients are slightly weaker at 220 m than at 485 mthere (;0.0148C m21 compared to ;0.0178C m21) since 220 mis near the core of equatorial subsurface water in the Under-current. The vertical temperature gradient at 750 m is weakerstill (;0.0058C m21). From the above results, temperature(and density) surfaces deepened during the sharp La Nina-ElNino transition in 1996–1997 by ;150, ;140, and ;110 m at220, 485, and 750 m depth, respectively.

A closer look at the low low-pass temperature data in Figure5 during the last La Nina-El Nino cycle shows significant phasedifferences among the data at individual depths and amongthis data and low low-pass El Nino indices during this period.Temperature data at 220 and 750 m then were highly corre-lated (to 99% significance level), whereby the interannual fluc-tuation at 750 m led that at 220 m by ;2 months at maximumcorrelation. Likewise, deeper layers appear to lead shallowerones during the La Nina-El Nino transition in 1996–1997. Thisis shown by two straight but slanted lines on the right-hand sideof Figure 5 that pass through temperature minima and maximaof the transition (the slope of the lines derives from the lagdetermined in the correlation analysis above). These lines alsopass through corresponding minima and maxima at 485 m thatunderline the vertical continuity of these upward propagatingsignals. Our results also show that deepening of the thermo-cline was underway at all depths over the slope by the end of1996 while La Nina conditions persisted in the eastern equa-torial Pacific (Figure 3). Indeed, the low low-pass temperaturerecord from 220 m there was highly correlated (to 99% signif-icance level) with low low-pass Nino 3.4 SST for the 1995–1997period. The slope temperature record off Chile led easternequatorial Pacific SST by 1–2 months at maximum correlation.

Large seasonal variability of alongshore flow over the slopepresents problems for isolating the interannual signal of along-shore flow there, given continuous pieces of low low-pass dataof only 4–5 years length (Figure 5). Conventional filtering tosuppress the seasonal signal would leave ,2 years of data foranalysis of the interannual signal. Instead, we choose to re-move the semiannual and annual harmonics (Table 2) from thelow low-pass time series from Figure 5. The “interannual” timeseries that result (Figure 6) contain seasonal contamination, inparticular, a 10 month period “wave” during the first half of1996, but still allow a useful discussion of interannual variabil-ity in alongshore flow.

Poleward flow tends to be stronger at 220 m, and equator-ward flow tends to be stronger at 750 m during El Nino eventsof 1994 and 1997. However, there is no simple relation be-tween event strength and alongshore flow: poleward flow at220 m was not stronger during the strong 1997 event thanduring the weak 1994 event. Observed increase in verticalshear of alongshore flow (220–750 m in Figure 6) during ElNino events is expected from geostrophy given a deeper ther-mocline near the coast (Figure 5). Likewise, weaker shearduring the La Nina event is consistent with flatter temperature(and density) surfaces between the offshore region and thecoast. Vertical temperature sections taken during and at theend of the 1991–1992 El Nino event illustrate these two situ-ations (Figures 3 and 4 in CH2). Interannual-scale events ofequatorward (poleward) flow at 750 m appear to lead poleward(equatorward) flow events at 220 m over the slope (Figure 6)with maximum correlation (significant at the 95% level) for a2-month lead in this sense. Together with the similar resultabove for interannual-scale fluctuations of temperature, thisimplies upward phase propagation on the slope of ;10 m d21

for the El Nino-La Nina signals.

5. DiscussionHere we presented results from moored current and tem-

perature observations at 220, 485, and 750 m depth over theslope at 308S off central Chile for the period 1991–1997 andused these results to address seasonal and interannual variabil-ity of alongshore flow and stratification in the Peru-Chile Cur-rent. We also considered the time-space structure of regionalwind forcing, sea level, and the regional signature of ElNino-La Nina events for this period. Below we present someinterpretations for our results in terms of local and remoteforcing.

Figure 6. (a) Low low-pass anomalies of sea surface temper-ature for the Nino 3.4 region (SST Nino 3.4) and low low-passSOI multiplied by 21. (b) Low low-pass time series of along-shore currents at 220 and 750 m depth as described in Figure5 but with annual and semiannual harmonics removed. Dataare plotted for the longest common period of good data at bothdepths. Also shown is the difference between both time series(220–750 m), which is an estimate of interannual-scale changesin vertical shear over the slope during the study period.

29,957SHAFFER ET AL.: SEASONAL AND INTERANNUAL VARIABILITY OFF CHILE

Both local and regional winds were dominated by an annualcycle with a distinct time-space structure. Maxima of equator-ward, upwelling favorable winds over the slope shift southwardfrom the austral winter off Peru and northern Chile over thespring near 308S to the summer farther south (Figure 2). At thecoast of northern and central Chile, land-sea temperature dif-ferences shift the strongest equatorward winds toward thesummer [Pizarro et al., 1994]. As a consequence, near-shorecyclonic wind stress curl, and thus upward Ekman pumping,may be significant in the austral winter off northern Chiledespite weak winds there. This effect can still be seen at 308Sin spring when ERS wind maxima tend to lead Lengua de Vacawind maxima (Figure 3). Bakun and Nelson [1991] found max-imum cyclonic wind stress curl in the austral winter off Perubut not off northern and central Chile, where, however, theiranalysis rested upon a very sparse database. South of our studyarea, maximum cyclonic wind stress curl may occur in theaustral summer [Bakun and Nelson, 1991].

McCreary et al. [1987] address alongshore flow in the Cali-fornia Current forced by steady and seasonally varying, equa-torward winds. Coastal “trapping” of the currents results fromRossby wave dissipation by vertical mixing in this model. Cur-rent strength and structure are sensitive to winds equatorwardof the site in consideration. Also, through vorticity dynamics(sverdrup balance), local cyclonic wind stress curl leads to localpoleward flow in the model. Upward phase propagation(downward energy propagation) results from Kelvin wave-likeproperties of the solutions. Such model results may help ex-plain some aspects of observed, seasonal variability over theslope off central Chile, like upward phase propagation andmaximum poleward flow in the Peru-Chile Undercurrent inNovember, when wind forcing is at a maximum locally andimmediately equatorward of the study site (Figure 2).

Semiannual variability in subsurface poleward flow has alsobeen observed off California [Chelton, 1984; Hickey, 1998].One maximum of such flow occurs in the boreal summer withstrong equatorward winds in a situation like the one in No-vember at 308S off Chile. The second maximum off Californiaoccurs north of 358N in the boreal winter in association withthe poleward Davidson Current, which extends to the surfaceat a time of weak equatorward or even poleward winds [Chel-ton, 1984; Hickey, 1998]. The second undercurrent maximum at308S off Chile occurs in April at a time of minimum equator-ward winds there (Figure 2). Furthermore, annual sea levelcycles (not shown) calculated from the ASL data (Figure 3)show demeaned sea level at Arica to be ;2 cm higher than atValparaiso in the austral fall (;2 cm lower in the australspring). This is consistent with rather barotropic poleward flowin the austral fall off Chile. There is a large semiannual com-ponent of variability in sea surface pressure and geostrophicwinds south of 308S in the Southern Hemisphere [van Loonand Rogers, 1984]. However, information on wind changes istransmitted poleward as coastal trapped waves and seaward asRossby waves. Thus circulation over the slope in the study areawould not be expected to be sensitive to winds poleward of308S and seaward of the slope.

Sea level anomalies at interannual periods have been foundto propagate poleward from the equator along the Pacific coastof the Americas with a speed of ;0.4–0.9 m s21 [e.g., Enfieldand Allen, 1980; Chelton and Davis, 1982]. For such long peri-ods, no poleward propagation would be expected there in theabsence of friction or vertical mixing. Recently, Clarke and VanGorder [1994] included bottom friction in a linear model for

the west coast of North America and were able to reproducethe observed poleward propagation of sea level anomalies atinterannual periods. Coastal trapping of the currents resultsfrom Rossby wave dissipation by bottom friction. The modelpredicts a bottom-trapped, alongshore jet over the slope that isout of phase with near-surface currents over the shelf andslope. Our results do show poleward flow in the near surfacelayers and enhanced equatorward flow near the bottom on theslope in association with El Nino events, but we found consid-erably stronger flow variability at interannual periods at 220 mthan at 750 m near the bottom (Figure 6).

In both the McCreary et al. [1989] and Clarke and Van Gorder[1994] approaches, coastal current structure depends uponRossby and Kelvin/coastal trapped wave modes and their in-teraction with dissipative processes. In both approaches, verysimple assumptions were made about the size and structure ofsmall-scale mixing. Observations of such mixing are generallylacking in eastern boundary current regions. However, onestudy off Peru found the strongest diapycnal mixing in near-surface layers adjacent to the coast, whereby mixing intensitywas more related to current shear than to vertical stratification[Shaffer, 1986]. Better simulations of the structure of mean andvery low frequency currents over the slope in eastern boundarycurrent regions may likely require more realistic treatments ofdissipative processes. A new version of the Clarke and VanGorder model with large friction near the coast but weakbottom friction over the slope simulated our current observa-tions off Chile reasonably well (O. Pizarro et al., El Nino sealevel and currents along the South American coast: Compari-son of observations with theory, submitted to Journal of Phys-ical Oceanography, 1999).

The most dramatic events in our records occurred during thetransition from La Nina to El Nino conditions in 1996 and1997. A strong equatorward wind pulse started to develop inmid-1996 all along the central Chile coast (Figure 3). Thisevent peaked in October when low sea level was observed atthe coast and minimum temperatures were observed in thewater column at our 308S slope station (Figures 3 and 5). Arapid decrease of equatorward winds occurred all along thecoast of central and northern Chile at the end of 1996-beginning of 1997. This decrease coincided with a rapid, quitesimultaneous sea level rise all along this coast, and rapid warm-ing occurred at all depths at our slope station associated withdownward displacement of isotherm surfaces there. In thisprocess, deep layers led shallower ones.

These sequences of events are consistent with coastal oceanresponse off northern and central Chile to the striking, anom-alous wind forcing there in 1996 and 1997. Moreover, duringthe onset and mature phases of the 1997 El Nino event, equa-torward winds greatly weakened at the slope site but less so atthe coastal site at 308S (Figure 3). As a consequence, windstress curl and the Ekman pumping over the slope tended tovanish then. This would result in less poleward flow over theslope, as follows from the discussion above. Our observationsshowed equally strong poleward flow at 220 m during the weak1994 and the strong 1997 El Nino events (Figure 6). Perhapsthe superposition of such local forcing upon the remotelyforced El Nino signal in 1997 limited the poleward flow overthe slope at 308S then.

The results above suggest that not only remote forcing ofequatorial origin by way of the ocean but also local forcing byanomalous winds may contribute to variability of sea level,temperature, and alongshore flow off central Chile in response

SHAFFER ET AL.: SEASONAL AND INTERANNUAL VARIABILITY OFF CHILE29,958

Plate 1. Distributions of monthly mean sea surface temperature anomalies in the South Pacific region in (a)January, (b) April, and (c) July 1997 based on data obtained from Integrated Global Ocean Services Systemimplemented at the Lamont-Doherty Earth Observatory of Columbia University (http://ingrid.ldeo.columbi-a.edu/SOURCES/.IGOSS/).

29,959SHAFFER ET AL.: SEASONAL AND INTERANNUAL VARIABILITY OFF CHILE

to El Nino cycles. These winds probably reflect an atmosphericteleconnection between the eastern South Pacific and thewarm water pool of the western tropical Pacific Ocean (seebelow). This would explain our observation that changes in sealevel and temperature off Chile, apparently forced by thesewinds, lead eastern tropical Pacific SST and the Southern Os-cillation Index by several months at and before the onset of the1997 El Nino event (Figures 3 and 5). In addition, observedupward phase propagation over the slope at and before theonset of the event is consistent with this scenario since suchphase propagation implies downward group velocity as wouldbe expected from surface forcing near or upstream (in thecoastal trapped wave sense) of the study area. Ramp et al.[1997] identified atmospheric as well as oceanic pathways of ElNino signals off central California.

Rasmusson and Carpenter [1982] were the first to identify ElNino precursors in the eastern South Pacific. Their results fora composite of six El Nino events between 1951 and 1972 showwarm SST anomalies several months before the onset of thecomposite event in a band south of the South Pacific Conver-gence Zone and extending from the western Pacific warm poolto the coast of Chile south of ;208S. In the onset phase of theevent these average anomalies exceeded 18C off central Chile.In the same trans-Pacific band, Rasmusson and Carpenter alsofound wind anomalies that off central Chile, lead to a weak-ening of equatorward winds before and during the onset phase.Such weakening was also observed in the first few months of1997 (Figure 2) and led to part of the sea level and watercolumn temperature changes discussed above.

During the onset of the 1997 El Nino event, SST anomaliesin the tropical and South Pacific (Plate 1) reveal an eventsuccession similar to that in the Rasmusson and Carpenter[1982] composite. In January 1997 a crescent-shaped warmanomaly was found between the western tropical Pacific andthe coast of Chile, where anomalies exceeded 28C. At the sametime, La Nina conditions still prevailed in the eastern tropicalPacific. By April 1997, warm anomalies were found all alongthe west coast of South America, and the warm pool in thewestern tropical Pacific had moved eastward. Weaker equato-rial winds along the west coast of South America during thefirst few months of 1997 (Figure 2) could explain part of thewarming there through reduced upwelling and evaporativecooling. By July 1997 a full-fledged event had developed withSST anomalies exceeding 48C along the equator in the centraland eastern tropical Pacific and off Peru. No El Nino modelpredicted well the strong 1997 event when initiated with Jan-uary 1997 conditions [McPhaden, 1999]. The above results sug-gest that SST in the eastern South Pacific, apparently quitesensitive to atmospheric teleconnections at the onset of ElNino events, may be useful in the context of an El Nino “earlywarning system.”

Acknowledgments. We thank Marcel Ramos (University of Con-cepcion) for extensive help in the preparation of this manuscript.Among those involved in the mooring work were the captain (ArturoNakanishi) and crew of Abate Molina, Per-Ingvar Sehlstedt and LeifDjurfeldt (University of Goteborg), Henning Hundahl (University ofCopenhagen), Dierk Hebbeln and Gotz Ruhland (University of Bre-men), Sergio Salinas and Reynaldo Rehoff (Catholic University ofValparaiso), and Jaime Henriquez (University of Concepcion). JoseRutllant (University of Chile) provided us with the Lengua de Vacawind data. The ERS-1 and -2 satellite scatterometer wind stress datawere obtained from Departement d’Oceanographie Spatiale, IF-REMER, France. Sea level data obtained from the Sea Level Center,

University of Hawaii (Mark Merrifield, Director) were collected byServicio Hydrografico y Oceanografico de la Armada de Chile. Atmo-spheric pressure data were supplied by Direccion Meteorologica deChile and Servicio Meteorologico de la Armada, Armada de Chile.Monthly mean SST anomaly data for the South Pacific region wereobtained from Integrated Global Ocean Services System (Yves Tourre,Scientific Editor). Discussions with David Enfield were helpful. Thisresearch was supported by grants from the Swedish Agency for Re-search Cooperation with Developing Countries, from the Danish,Swedish, and Chilean National Research Councils (in part in theFONDAP-Humboldt Current program), and from the Danish Na-tional Research Foundation.

ReferencesBakun, A., and C. S. Nelson, The seasonal cycle of wind-stress curl in

subtropical eastern boundary current regions, J. Phys. Oceanogr., 21,1815–1834, 1991.

Chelton, D. B., Seasonal variability of along-shelf geostrophic velocityoff central California, J. Geophys. Res., 89, 3473–3486, 1984.

Chelton, D. B., and R. E. Davis, Monthly mean sea level variabilityalong the west coast of North America, J. Phys. Oceanogr., 12, 757–784, 1982.

Clarke, A. J., and S. Van Gorder, On ENSO coastal currents and sealevel, J. Phys. Oceanogr., 24, 661–680, 1994.

Davis, R. E., Predictability of sea surface temperature and sea levelpressure anomalies over the North Pacific Ocean, J. Phys. Oceanogr.,6, 249–266, 1976.

Enfield, D. B., and J. S. Allen, On the structure and dynamics ofmonthly mean sea level anomalies along the Pacific coast of Northand South America, J. Phys. Oceanogr., 10, 557–578, 1980.

Graber, H., N. Ebutchi, and R. Vakkayil, Evaluation of ERS-1 scat-terometer winds with wind and wave ocean buoy observations, 78pp., Jet Propul. Lab., Pasadena, Calif., 1996.

Hickey, B. M., Coastal oceanography of western North America fromthe tip of Baja California to Vancouver island, in The Sea, vol. 11,edited by A. R. Robinson and K. H. Brink, pp. 345–393, John Wiley,New York, 1998.

Huyer, A., P. M. Kosro, S. Lentz, and R. C. Beardsley, Poleward flowin the California Current system, in Poleward Flows Along EasternOcean Boundaries, Coastal Estuarine Stud., vol. 34, edited by S.Neshyba et al., pp. 142–156, Springer-Verlag, New York, 1989.

Huyer, A., M. Knoll, T. Paluszkiewicz, and R. L. Smith, The PeruUndercurrent: A study in variability, Deep Sea Res., Part A, 38, suppl.1, 247–279, 1991.

Kessler, W. S., and M. J. McPhaden, The 1991–1993 El Nino in thecentral Pacific, Deep Sea Res., Part I, 42, 295–333, 1995.

Madden, R. A., and P. R. Julian, Detection of a 40–50 day oscillationin the zonal wind field of the tropical Pacific, J. Atmos. Sci., 28,702–708, 1971.

McCreary, J. P., P. K. Kundu, and S.-Y. Chao, On the dynamics of theCalifornia Current system, J. Mar. Res., 45, 1–32, 1987.

McPhaden, M. J., Genesis and evolution of the 1997–1998 El Nino,Science, 283, 950–954, 1999.

Pizarro, O., S. Hormazabal, A. Gonzales and E. Yanez, Variabilidaddel viento, nivel del mar y temperatura en la costa norte de Chile,Invest. Mar., 22, 85–101, 1994.

Ramp, S. R., J. L. McClean, C. A. Collins, A. J. Semtner, and K. A. S.Hays, Observations and modeling of the 1991–1992 El Nino signaloff central California, J. Geophys. Res., 102, 5553–5582, 1997.

Rasmusson, E. M., and T. H. Carpenter, Variations in tropical seasurface temperature and surface wind fields associated with theSouthern Oscillation/El Nino, Mon. Weather Rev., 110, 354–384,1982.

Shaffer, G., On the upwelling circulation over the wide shelf off Peru,2, Vertical velocities, internal mixing and heat balance, J. Mar. Res.,44, 227–266, 1986.

Shaffer, G., S. Salinas, O. Pizarro, A. Vega, and S. Hormazabal, Cur-rents in the deep ocean off Chile (308S), Deep Sea Res., Part I, 42,425–436, 1995.

Shaffer, G., O. Pizarro, L. Djurfeldt, S. Salinas, and J. Rutllant, Cir-culation and low frequency variability near the Chile coast: Remote-ly-forced fluctuations during the 1991–1992 El Nino, J. Phys. Ocean-ogr., 27, 217–235, 1997.

Smith, S. D., Coefficients for sea surface wind stress, heat flux, andwind profiles as a function of wind speed and temperature, J. Geo-phys. Res., 93, 15,467–15,472, 1988.

SHAFFER ET AL.: SEASONAL AND INTERANNUAL VARIABILITY OFF CHILE29,960

Strub, P. T., J. S. Allen, A. Huyer, R. L. Smith, and R. C. Beardsley,Seasonal cycles of currents, temperatures, winds, and sea level overthe northeast Pacific continental shelf: 358N to 488N, J. Geophys.Res., 92, 1507–1526, 1987.

van Loon, H., and J. C. Rogers, Interannual variation in the half-yearlycycle of pressure gradient and zonal wind at sea level in the SouthernHemisphere, Tellus, Ser. A, 36, 76–86, 1984.

S. Hormazabal and G. Shaffer, Danish Center for Earth SystemScience, Niels Bohr Institute for Astronomy, Physics and Geophysics,

University of Copenhagen, Juliane Maries Vej 30, 2100 Copenhagen,Denmark. (gs@dcess.ku.dk)

O. Pizarro, Regional Program in Physical Oceanography and Cli-mate, University of Concepcion, Box 160-C, Concepcion 3, Chile.

S. Salinas, School of Marine Sciences, Catholic University of Val-paraiso, Avenida Altamirano 1480, Box 1020, Valparaiso, Chile.

(Received October 14, 1998; revised July 13, 1999;accepted July 28, 1999.)

29,961SHAFFER ET AL.: SEASONAL AND INTERANNUAL VARIABILITY OFF CHILE

29,962