U.S. Dept. of Commerce / NOAA/ OAR / PMEL / Publications
Four different datasets of monthly mean near-equatorial Pacific sea surface temperature for 1982-83 are compared, and the space-time regions for which there was consensus that cooling or warming took place, are determined. There was consensus that warming took place east of the date line, averaged over the period July-December 1982, and that the warming progressed eastward from the central Pacific. There was also consensus that weak cooling took place in April 1983, and that substantial cooling occurred in June-July 1983, generally over the central and eastern Pacific. However, the analyses tend to agree on the sign of SST change only in periods of cooling or warming in excess of 1°C/month; quantitative agreement at the level of 0.5°C/month or better is almost never found.
SST changes in five ocean-circulation model hindcasts of the 1982-83 period (differing only in that each used a different analyzed monthly mean surface wind stress field to drive the ocean), are compared with the observations and with each other. There is agreement that net warming occurred in the July-December 1982 period and cooling in mid-1983. The heat budgets of these experiments indicate that the major model central Pacific warmings occurred primarily from anomalous eastward surface advection of warm water. Further, east zonal advection remains significant, but a diminished cooling tendency from meridional advection can also be important; different hindcasts differ on the relative importance of these terms. Surface heat flux changes do not contribute to the warmings. The reduced cooling tendency from meridional advection is consistent with diminished surface Ekman divergence, suggesting that southward transport of warm north equatorial counter current water was not a major factor in the model warmings. The hindcasts do not agree on the relative importance of local or remote forcing of the eastward surface currents; while there is clear evidence of remote forcing in some hindcasts in particular regions, local forcing is also often significant. The main 1983 midocean cooling began because of increased vertical advection of cool water; but once cooling began horizontal advection often contributed. Further cast, where the easterlies generally return later than they do in midocean, upwelling and horizontal advection all can be important. Again no model consensus exists concerning the details of SST evolution.
Because the observations do not agree on the sign of SST change during much of the 1982-83 period, improved SST data is needed in order to document the behavior of the ocean through future ENSO periods. Better forcing data will be needed to carry out improved ocean-model validation studies, and to explore the mechanisms likely responsible for SST change through entire ENSO cycles.
The 1982-83 El Niño-Southern Oscillation (ENSO) episode was among the most intense in the historical record and was the best observed, but it is not well understood. While substantial ocean and atmosphere datasets have been collected and studied, many of the most basic questions about the physical processes responsible for the event cannot yet be answered.
Central to the ENSO problem is the evolution of near-equatorial sea surface temperature (SST) before and during the event. All existing coupled ocean-atmosphere models attribute a central role to SST variation, and atmospheric general circulation models have been able to reproduce many aspects of the low frequency ENSO changes when they are forced by imposing observed monthly mean SST fields. Unfortunately, the ocean-atmosphere coupled interaction processes that control SST evolution are among the most difficult to observe because there is great high frequency and small space-scale structure in the various heat equation term fields and because so few observations of the needed data are available.
Understanding the mechanisms of surface warming and cooling in ENSO events is important, both in its own right, and in order to validate coupled ocean-atmosphere models. Several recent coupled model studies seem to be converging on a mechanism for the existence of ENSO events; we need as much information as possible in order to judge the plausibility of this result. Even the most basic qualitative questions have not been answered so far from the observations. For instance, we are not able to determine from observations the extent to which anomalous eastward advection of warm water was remotely or locally forced; or whether anomalous southward advection of warm water from the NECC played any significant role in the warming; or the degree to which variation in the amount of upward advection of cold water played an important role in the major warming of 1982.
Lacking sufficient data, it is necessary to turn to model systems of the individual fluids themselves and explore possible physical scenarios for the event. The work reported here concerns modes of oceanic near equatorial sea surface temperature evolution during 1982-83, using the best available analyses of the surface wind stress field and estimating the surface heat flux using ocean-surface temperature, wind speed and other specified parameters. The ocean circulation model is that described by Philander and Seigel (1985), and will be described below. Because the surface wind is only imperfectly known over most of the tropical Pacific, and because model SST changes will be seen to be very sensitive to the surface wind in these studies, it is difficult to make definitive inferences about the ocean's behavior from model results like these. However, it will be seen that the largest model changes generally are consistent with the largest observed changes and that there is general model hindcast consensus on the mechanisms of warming and cooling over some regions and times, despite the large differences in surface wind stress fields. The present model results can speak to a number of issues of general interest, but authoritative determination of the SST change mechanisms at work in the ocean will not be possible until surface conditions are better known than they were in 1982-83.
What do we know about how tropical Pacific SST changed during 1982-83? Fields of monthly mean SST have been obtained from Fleet Numerical Ocean Central ("spot obs", made available to us by Scott Woodruff), the NASA THEP (TOGA Heat Exchange Project) fields (made available by Tim Liu), and the NOAA Climate Analysis Center (CAC) "in situ" and "blended" analyses (made available by Dick Reynolds). The FNOC and NMC "in situ" fields were based primarily on surface marine observations, while the NMC "blend" product made considerable use of satellite (MCSST) SST spatial gradient information, together with surface data. The THEP product was based primarily on satellite (SMMR) data. The large-scale pattern of SST evolution during 1982-83, according to the NMC "blend" analysis, has been presented in a number of places (e.g., Rasmusson and Wallace 1983), and is assumed to be familiar to the reader. This event differed from the composite post-1950 ENSO event described by Rasmusson and Carpenter (1982) in several respects: there was no significant warming along the South American coast in Spring 1982; the first substantial warming occurred along the equator in late Summer 1982; substantial warm anomalies appeared along the South American coast late in 1982, and the midocean remained significantly warmer than usual into late Spring 1983. Our interest is to understand the mechanisms responsible for these changes.
The heat equation relates the time rate of change of SST to the various advective, diffusive and forcing processes, so it is necessary to concentrate not on the patterns of SST, but of SST change from month to month. In order to limit this discussion, the equatorial Pacific has been divided into five regions, each extending from 2°N to 2°S, and across 10 degrees of longitude in zonal extent. Region center longitudes are 95°W, 115°W, 135°W, 155°W and 165°E. Forward time differences of SST from each SST dataset averaged over each region are displayed in Fig. 1(a-c; d and e); climatological SST changes, based on the post-1950 surface marine data, are also included for reference.
Figure 1. Monthly average rates of change of SST for the five analysis regions, according to the NMC "in situ," NMC "blend," FNOC spot obs, and NASA THEP SST datasets (see text for more information on these SST fields, and for discussion). Note that the different fields generally differ on the sign of the month-to-month change, except when the rate of change exceeds roughly 1 degree C per month.
In the 95°W region the climatological cycle of cooling between March and September, followed by warming between October and February, is followed roughly during 1982; there is less cooling in May, July and August 1982 than is inicated in the climatology, and there is warming instead of cooling in September 1982. Nineteen hundred and eighty-three was also similar in pattern to climatology until July, when much stronger than normal cooling took place; this cooling was sufficiently strong to bring SST values back to normal. These is consensus among the SST datasets that cooling took place in March, May and June 1982 and April, June and July 1983; there is consensus that warming took place between October 1982 and March 1983.
In the 115°W region, where there are fewer surface observations, the pattern of cooling and warming is similar to that in the 95°W region, but with more month-to-month variability and generally with smaller values. The climatological cycle is also noisier than at 95°W. The strongest warming that is agreed upon occurs in September 1982; the strongest cooling was in May and June 1983. Both are sharp departures from climatological behavior. These departures are consistent with the near surface 110°W mooring data reported by Halpern (1987).
At 135°W the normal seasonal cycle pattern roughly holds until around late August 1982 when there is arguably a two month period of substantial warming. (Two of the analyses show very anomalous August 1982 warming, while the others show no temperature change; all show subsequent warming.) There follow many months of uncertain temperature tendency, until in June-July 1983 there is agreement that the ocean cooled much more strongly than indicated by climatology. Averaged over the period August-November 1982 there is consensus that warming took place. No moored or other densely sampled dataset is available for comparison in this region for this period.
At 155°W there is agreement that there was modest warming in February-May and late July 1982, and cooling in late June. There is no consistent month-to-month pattern or agreement between August 1982 and February 1983, but the average over August-November 1982 is for warming in each analysis. There is consensus that warming took place in late February, early March 1983, and cooling in June-July 1983. The agreed upon substantial departures from climatological behavior are the cooling in late June 1982 and late June 1983; the late July 1982 warming represents less than a 1°C/mo departure.
At 165°E the best agreement is March-April 1982 (warming) followed by a tendency toward cooling. There is little agreement again until late January, early February 1983 (cooling). These patterns are consistent with those reported by Meyers et al. (1986), based on XBT data. The climatological cycle of SST change never involves changes greater than 0.5°C/mo, and contains much more semiannual variability than annual variability. It is difficult to compare the 1982-83 data with the climatological cycle because the differences between the various versions of 1982-83 behavior are larger than the month-to-month climatological changes.
Note that in the easternmost two regions, the analyses generally agree on the sign of the derivative of SST whenever it exceeds 1°C/mo in at least one analysis. When it is smaller than 1°C/mo there may or may not be agreement on sign. In the regions farther west it is much harder to find even agreement on the sign of the time rate of change of SST from month to month. Also worthy of note is that the departures from climatological behavior tend to be smaller than the differences between the different SST datasets, even in a major ENSO event like 1982-83.
Clearly our knowledge of this most basic quantity for the diagnosis of SST change processes, which is fundamental to any model-data comparison of the 1982-83 ENSO, is marginal. Although the FNOC spot obs. dataset often shows the largest month-to-month jumps, careful examination of Fig. 1(a-c; d and e) will reveal that there is no consistent pattern of disagreement between SST datasets. Because of the eruption of El Chichon in the Spring of 1982, satellite SST retrieval algorithms had to be modified to account for the altered atmospheric aerosol loading; this may account for some of the scatter in the analyzed SST fields. But by 1983 these anomalies were of negligible concern. The continued disagreements between datasets suggest that the SST monitoring systems in place in 1982-83 were generally not able to agree unless the SST change signal is at least 1°C/mo. Even in this major ENSO event, analyzed SST changes in the equatorial Pacific (where the SST changes are larger than aywhere else in the mid-ocean) are seldom this large.
In order to understand the hindcast SST changes and change mechanisms, it is useful to have some information about the forcing fields that were used in the different hindcasts. The five basic hindcast experiments make use of five different monthly men wind pseudostress or stress analyses [obtained from the U.S. Navy's Fleet Numerical Ocean Center (FNOC), from the European Centre for Medium Range Forecasting (ECMWF), from the National Meteorological Center (NMC), from the special ship wind analysis effort at Florida State University (FSU) and from the special ship wind/adjusted cloud motion vector analyses of Sadler at U. Hawaii (SADLER)]. Harrison et al. (1988) describe how the various wind stress fields which were used in the hindcasts were produced from the operational and research products. Figures 2 and 3 present the zonal and meridional components of wind stress within 2 degrees of the equator, as they were interpolated in time and used to drive the ocean model. In addition the Hellerman and Rosenstein (1983) climatological stresses, which were used in the model spinup, are plotted for reference.
Figure 2. Monthly average zonal wind stress fields, averaged between 2°N and 2°S, for the different wind stress analyses used in the model hindcast experiments: ECMWF, FNOC, FSU, NMC and SADLER. See text for notation and discussion. The Hellerman and Rosenstein (1983) climatological stress is presented for comparison.
Figure 3. As for Figure 2, except meridional wind stress fields.
The zonal-stress plots (Fig. 2) very clearly show that
large zonal wind stress anomalies occurred in 1982-83, and that the different analyses
generally agree qualitatively on the nature of the ENSO signal: westerly stress appeared
in the western Pacific by mid-1982, ultimately extended east of the date line and
essentially disappeared in December 1982-January 1983 in the central Pacific; much weaker
than normal stresses occurred across the basin through early 1983; easterly stress
returned to at least 0.02 N M
by May-June 1983. Note,
however, that the various analyses differ greatly in the timing of the appearance of the
westerlies, their magnitude, whether there was slow eastward propagation or rather abrupt
extension across many degrees of longitude, and in the eastward extent of the westerlies.
There are also major differences in the far western Pacific in early 1982, in how the
easterly trades behaved east of 160°W in June-December 1982, and in how the
climatological easterlies reappeared in mid-1983. While qualitative agreement is
considerable, quantative differences are large.
The meridional stress plots (Fig. 3) again agree qualitatively that there was unusually strong southerly stress in the western Pacific in mid-1982 and unusually widespread northerly stress in December 1982-Spring 1983. As with the zonal stress comparison, large quantitative differences are found at various times, particularly in the magnitude of the "southerly" anomaly and of the "northerly" anomaly; differences in detail are widespread.
For the following discussion, the needed information can be obtained from Figs. 2 and 3; a more quantative comparison of the differences between the products can be found in Harrison et al. (1988).
The model SST results presented here are from the same hindcasts of the 1982-83 event that have been described in Harrison et al. (1988, 1989). In brief, each hindcast uses a 27 vertical level version of the Cox-Bryan-Semtner primitive equation model that has been developed by Pacanowski and Philander over the last few years. The particular form of the model is that which was used in the pioneering 1982-83 ENSO hindcast of Philander and Seigel (1985); the various model parameters and equations are described there. Monthly mean surface wind stress fields are imposed: surface heat flux is computed from the surface wind, the SST, and an imposed air-sea temperature difference (the sensible and latent heat flux terms are the only ones permitted to vary, and a minimum wind speed is used in the latent computation to ensure that cooling continues to occur from evaporation even when the monthly mean wind is small); and each hindcast is begun from a climatological January initial condition obtained from a calculation of spinup for several years, with Hellerman and Rosenstein (1983) climatological wind stresses from the Levitus density field. As mentioned above, the five basic hindcast experiments make use of five different monthly mean surface wind stress analyses; we label each hindcast by the stress field used to force it, because the hindcasts are in every way identical save for the wind stress fields and the resulting different surface heat flux fields.
Figure 4(a-c; d and e) shows the time rate of change of SST for each hindcast, averaged over the analysis regions. Note that snapshots from the model were stored every three days, so that shorter time scale information is presented in these plots than was possible for the analyzed SST fields; this is necessary in order to determine the mechanisms responsible for SST change.
Figure 4. Rates of change of SST over the analysis regions, according to the five hindcast experiments, based on snapshots three days apart. See text for discussion.
In the 95°W region the hindcasts agree with each other, and with the observations that there was cooling in February-March 1982, but there is disagreement about conditions during April (note that there was disagreement in the observations about April also). FSU agrees with the observed May-June 1982 cooling; the other hindcasts all indicate warming. Averaged over the period July 1982-April 1983, each of the hindcasts indicates net warming (in agreement with the analyses of SST), but there is much more month-to-month variation between warming and cooling in the hindcasts than is seen in the analyses. Further, there is considerable disagreement between hindcasts, month to month. Hindcast consensus that cooling took place begins in June 1983; at this time the analyses also agreed that cooling was occurring. Note that the analyses indicated that some cooling began in April 1983 but differed in behavior during May.
In the 115°W region, consensus is found in the hindcasts on: warming in late March-late April 1982 (roughly in agreement with the analyses); net warming over the period August 1982-December 1982 (good agreement with the analyses); cooling in January 1983 (the analyses are equivocal on this); and major cooling in June-July 1983 (good agreement with the analyses). The hindcasts differ greatly on month-to-month details. Although the analyses indicate that the 95°W and 115°W regions had quite similar SST changes, the hindcasts show some significant differences between the changes in the two regions.
In the 135°W region there is hindcast consensus that warming occurred in March-May 1982 (in agreement with the analyses), that net warming existed averaged over July-December 1982 (generally in agreement with the analyses), that cooling took place in January 1983 (the analyses are equivocal, but none indicate significant cooling) and that cooling was present in June-July 1983 (good agreement with analyses). The August-December 1982 period contains a strong pulse of warming in each hindcast but there are large differences in the timing of this pulse; the ECMWF pulse occurs in August and the FNOC and FSU pulses occur in December. The June-July 1983 cooling is much stronger in FSU and SADLER than is found in the analyses.
In the 155°W region there is consensus on warming in April 1982 (in agreement with the analyses), on net warming averaged over August-November 1982 (generally supported by the analyses), on cooling averaged over mid-December 1982-mid-February 1983 generally supported by the analyses), and on cooling mid March 1983-June 1983 (generally supported by the analyses). The hindcast net August-November 1982 warming results, in each case, from one or more episode of strong warming, with different timing in each hindcast. The SADLER hindcast has two warming pulses while the others have only one each. The timing of the main 1983 cooling also differs by more than a month between hindcasts.
In the 165°E region there are no multimonth periods of hindcast consensus, just as there was little consensus in the analyses. The strongest signal in the analyses, cooling in mid-January-mid-February 1983, has an analogue in the ECMWF, SADLER and FSU hindcasts. Mostly there is simply a lot of month-to-month variability of warming and cooling, typically of about 1°C/mo.
Table 1 summarizes the periods of consensus from both the analyses and the hindcasts. It shows that there was warming in February-March 1982 (as occurs typically in the climatological seasonal cycle at this time), followed by relatively normal cooling in May-June 1982. The ENSO signal appears in the subsequent warming that began in July in the 155°W region and later and later farther east of this region. The consensus periods of ENSO warming ended between late 1982 and early 1983, depending upon the longitude of the region. No consensus is found until the end portion of the event begins, with cooling in May-June-July 1983, depending upon the region. There is often consensus among the hindcasts on the sign of SST change in the periods when there is also consensus among the analyses, though the hindcast consensus sometimes is found only through averaging over several months. In the periods when the analyses do not agree, generally, no hindcast consensus is found.
Table 1. Summary of periods of consensus agreement on surface warming (+) or cooling (-). SST products are summarized in "Data," modem hindcasts results are summarized in "Model." A period marked with a bar and (+) indicates that the average over the period is for warming, but the month-to-month consensus was not found.
The ocean model heat equation for the upper level includes all three temperature advection components--zonal, meridional and vertical--as well as a surface heat flux, a vertical diffusion between the first and second model levels, and a horizontal "eddy" diffusion between adjacent grid boxes; if surface cooling produces a gravitationally unstable situation, convective adjustment as it has been done by Bryan and Cox is used to produce a neutral density stratified upper water column. There is no formal model mixed layer. Because the model vertical mixing to some degree minimizes gradients right at the surface, the roles of vertical diffusion and vertical advectioin must be carefully considered. Any process that increases the vertical temperature gradient between the first and second model levels will tend to increase the downward diffusion of heat; upwelling (upward advection of colder water) into the second vertical grid box, or surface warming will each increase the gradient. Although the vertical mixing coefficient increases with decreasing Richardson Number (Pacanowski and Philander 1981), deeper in the water column it is constant between the first and second levels, thus vertical diffusion of heat is determined there by the vertical temperature gradient. Physically correct interpretation of increased vertical diffusion of heat generally requires detailed examination of the model flow changes as well as of the thermal field changes.
Time series plots of the zonal wind stress (multiplied by ten for clarity), the time
rate of change of SST, zonal advection of SST and meridional advection of SST are
presented for some of the regions and some of the hindcasts, in Figs.
5, 6 and 7. (Note that
the units for stress are dyn cm, not °C/mo.) The text
below will be intelligible without reference to these figures, but they are offered to
support the remarks made here. and especially the comments concerning the relative
importance of local and remote forcing.
Figure 5. Some SST equation terms for the 155°W region, together with the zonal wind stress over this region for the SADLER (a) and FNOC (b) hindcasts. See text for discussion.
Figure 6. As for Figure 5, except the 135°W region and the SADLER (a) and ECMWF (b) hindcasts.
Figure 7. As for Figure 6, except the 115°W region and the SADLER (a) and NMC (b) hindcasts.
In the 155°W and 135°W regions each hindcast produced the major warming at a different time between July and December (Fig. 4(a-c; d and e)), but in each case the predominant warming mechanism was zonal advection. Under normal summer conditions when the surface flow is westward and the zonal SST gradient is negative, the zonal advection SST tendency is to cool; however, at some point in this period of 1982, the zonal advection became strongly positive in each hindcast and warming usually closely paralleled the zonal advection change. Eastward surface currents were the cause of the change in zonal advection tendency.
In none of the hindcasts in either the 155°W or 135°W region was the warming a result of increased surface heat flux. In fact, in almost every instance the surface heat flux was decreasing as the warming took place. Because of the simple parameterization of surface heat flux used in the model no inference concerning the role of surface heat fluxes in the oceanic event can be drawn. The model result does, however, establish that warming comparable to that observed in 1982 can result from purely ocean dynamical processes.
In the 115°W region, the hindcasts do not show as much agreement on the mechanisms of surface warming. In the FNOC hindcast positive zonal advection is the initial warming process, but a substantial reduction in cooling from meridional advection and a reduction in vertical cooling from heat diffusion subsequently contribute to maintain the warming. The meridional advection tendency never becomes positive, just less negative (as would be expected from diminished equatorial Ekman divergence resulting from weakening easterly surface wind stress). In the SADLER, ECMWF, and NMC hindcasts the strongest warming periods are dominated by positive zonal advection, but the initial warming tends to involve each of the three mechanisms described above. In the FSU hindcast, instability waves are present until late October 1982 and no dominant mechanism emerges.
In none of the surface wind stress fields used here does the surface stress become
westerly in the 115°W region during the agreed upon period of warming (Fig. 2). The zonal stress becomes very weak (less than 0.02 N
m) in NMC and ECMWF but typically remains greater than
0.04 N m in SADLER and FSU. These stress differences
and the large local circulation differences that result account for the different warming
mechanisms.
In the 95°W region there is no consensus on a dominant warming mechanism; vertical diffusion changes, diminished meridional-advection cooling tendency, increased surface heat flux and changes in zonal advection contribute to varying degrees in each of the hindcasts. Zonal advection change plays a significant role in each hindcast, but it is seldom the dominant process. Meridional advection changes can be as important as zonal advection changes in some hindcasts and are unimportant in others. The wind stress fields differ greatly in the 95°W region (Figs. 2, 3), so these differences come as no surprise.
The relative importance of remote and local wind forcing in the advective warming episodes observed here is of interest. Remote forcing would be indicated if the flow were to accelerate eastward while the local zonal stress either remained constant or became more negative; it would also be indicated if the surface zonal pressure gradient were to become more positive while the local stress either remained the same or became less negative. Examination of the surface zonal momentum budgets indicates that remote forcing clearly was a significant factor in the warming in some regions of some of the hindcasts, but was not clearly of importance in others, at least relative to the locally induced surface current change. Examples of remotely forced warming are shown in Fig. 6b (ECMWF, 135°W, August-September 1982) and Fig. 7a (SADLER, 115°W, November-December 1982). Figure 5a (SADLER, 155°W, July-August 1982) shows a case where local and remote forcing appear to be comparable. There simply is no hindcast consensus on the relative importance of the two types of forcing.
Results from the SADLER hindcast are included in Figs. 5, 6, 7 so that the reader can contrast the behavior of the warming in three sequential regions. Note that two episodes of warming occur in the 155°W region in July-August 1982, and the other in September-October 1982. In the 135°W region there is a significant episode only in September-October 1982. In the 115°W region there is modest warming in September-October 1982 and substantial warming in November-December 1982. The earliest mid-Pacific pulse of warming was largely cut off west of 140°W because of strong easterly stress east of the date line, with maximum near 150°W in July-August 1982 (Fig. 2). The subsequent pulse of warming originated in forcing west of the date line but is substantially modified by local zonal stress changes east of the date line. Smoothing over several months in time, as has been done typically with historical studies of ENSO behavior (e.g., Rasmusson and Carpenter 1982) is likely to obscure important information about the mechanisms responsible for SST evolution if the ocean behaves at all as the model does.
In the 155°W region there is consensus that cooling begins with increased downward
diffusion of heat. Examination of the zonal wind stress fields reveals that this takes
place concurrently with either the reappearance of easterly stress or with reestablishment
of the easterly stress at roughly 0.02 N m. The
increased downward diffusion results primarily from an increase in the temperature
gradient brought about by subsurface upwelling. Once the cooling has begun both zonal and
meridional advection begin to contribute a cooling tendency (Fig.
5) (as they do in the climatological easterly wind stress regime), and the ocean cools
until roughly climatological temperatures are reached. In eyery hindcast the surface heat
flux is increasingly positive as the cooling takes place.
The situation in the 135°W region is similar, except zonal and meridional advection sometimes enter the budgets more quickly than in the 155°W region because west of this region the easterlies generally return to significant amplitude first; as a result significant SST gradients exist to be advected against as the local easterly stress increases.
In the 115°W region there is hindcast consensus, except for FSU, that increasing easterly stress first caused increased downward heat diffusion, and that cooling from meridional advection followed. The role of zonal advection is not agreed upon. In some hindcasts it also provided cooling, but in others it tended to warm. The different roles of zonal advection result from the initial pattern of increase of easterly stress and the induced equatorial SST gradient; if there is warm water east of this region, zonal advection will provide a warming tendency, and vice versa. The surface heat flux increased throughout the initial cooling phase.
In the 95°W region the hindcasts differ on the relative importance of meridional advection cooling and vertical diffusion cooling as the initial primary mechanism. Once cooling begins, meridional advection cooling becomes important in every case. Zonal advection never provides a cooling mechanism.
The 1982-83 ENSO SST changes were among the largest in the historical record but their detailed structure was not well observed. Comparison of several different monthly averaged SST datasets indicates that only when SST changes exceed 1 °C/mo is agreement on the sign of SST change likely to be found; quantitative agreement at the level of 0.5°C/mo is almost never found. However, the SST fields agree that there was warming east of the date line sometime between July and December 1982; the more eastern the region, the later the warming began, and the later it ended. There is also agreement that there was strong cooling in June-July 1983, which represented the end of this warm event. There is no consistent consensus on the behavior of SST in the two months prior to June 1982, or on the months between January and June 1983; only the major changes have reasonably clear signals.
The ocean circulation model hindcasts produced net warming over the period June-December 1982 in each of the regions east of the date line, but they differed significantly with each other on the timing of the warming. Also, in any particular region, model warming occurred more abruptly than is indicated by the SST analyses, and the total warming was somewhat greater than observed; plausible reasons for these differences will be discussed below. The main model cooling period fell within May-July 1983, in each hindcast and for each region, again in qualitative agreement with observations.
Examination of the model SST equation balances reveals that the central Pacific 1982 warming occurred primarily because the surface currents became eastward, causing the SST tendency due to zonal advection to become positive. The physics of the surface current reversal is not the same in each hindcast, but in general it involves a combination of some remote forcing and some local response to diminishing local easterly trades. Neither surface heat flux changes nor meridional advection changes play a significant role in this part of the warming.
The differences in the timing of the warming in the different hindcasts result clearly from the differences in the wind fields. The spatial character of the appearance of westerly winds in the western Pacific, their subsequent evolution, and the evolution of the central Pacific easterlies all contribute to creating different surface current fields in the model hindcasts. Different surface current fields produce different SST advection tendencies. It should be noted that the magnitude of the eastward surface flows that cause most of the warming is generally consistent in each hindcast with that observed at 160°W by Firing et al. (1983). The rapid model warming results from the fact that the initial model zonal SST gradient is larger than indicated in the SST analyses (Harrison et al. 1989), and that we have presented model rates of change from three-day snapshots, rather than from monthly means.
In the more eastern regions there is less consensus on the mechanisms of warming; meridional advection changes can be nearly comparable in importance to zonal advection changes. However, in every case the zonal advection tendency became positive, while the meridional advection tendency only very seldomly became weakly positive. In general, it remained negative at least through the main period of warming.
Thus the model hindcasts indicate that the very warm surface waters of the NECC were not an important factor in near equatorial warming. When meridional advection does play a significant role, it appears more in the form of reduced negative tendency that results from reduced meridional flow, that in turn results from reduced wind-driven Ekman divergence.
The main 1983 cooling begins in each hindcast in the central Pacific where the
easterlies first return or regain an amplitude of about 0.02 N m, through increased downward diffusion of heat out of the
surface grid box. This increased diffusion results from an increase in the temperature
gradient below the surface resulting from upwelled colder water. Of course, the upwelling
is simply forced by the surface Ekman divergence required by the easterly stress.
Once the surface begins to cool anywhere, significant SST gradients are quickly established and the surface current field subsequently plays a major role in cooling the rest of the central and eastern basin. The relative importance of meridional advection, zonal advection and increased downward diffusion (i.e., increased upwelling caused by increased local easterly stress), varies greatly from hindcast to hindcast.
Little can be said about SST change mechanisms west of the date line; the various SST analyses differ greatly about the patterns of observed SST change, and the hindcast results also show great diversity of behavior. At least from the large-scale perspective adopted here, the SST changes are too small and the differences between wind stress fields too large to permit useful inferences to be made.
As noted in section 5, surface heat fluxes in the hindcasts almost always provide a negative feedback process for SST. The model surface heat flux parameterization is very simple; only variability of the latent and sensible heat flux components is included, and their variability is limited by assuming a fixed relative humidity and air-sea temperature difference. As there are large cloudiness changes in the western and central Pacific during ENSO events, and as cloudiness can substantially alter the short- and longwave radiation flux components, the model parameterization is unlikely to be faithful to reality. In fact, the present parameterization appears to introduce too much heat into the ocean when the winds are light and SST is high; maximum model SST values can be 32°C when observations suggest that 30°C is seldom exceeded. However, it has been shown that the major signals of the 1982-83 event can be reproduced in the model without recourse to more sophisticated surface heat flux parameterizations. The result of Reed (1986), that the surface data available to him suggest that heat fluxes estimated from bulk formulae are either uncorrelated or weakly negatively correlated with SST changes in the central and eastern equatorial Pacific in both the 1972 and 1982 ENSO events, should be noted; this type of behavior is not inconsistent with the behavior found in the model results. Meyers et al. (1986) have suggested that the western Pacific cooling observed from XBT data around 160°E, roughly between July 1982 and February 1983, arose largely from an increase of latent heat flux cooling; we do not find consistently such behavior in the hindcasts, but note that because the model parameterization includes a background level of evaporative cooling it is not well suited to produce such behavior.
Because near equatorial surface current variations are so large during ENSO events and because SST gradients generally are large in the eastern central and eastern Pacific, it should not be a surprise that horizontal advection of SST plays such a dominant role in the warming process. Wherever there are climatological surface easterly stresses, the surface ocean is forced into a state much colder than would otherwise exist, and one which is maintained substantially by vertical and horizontal advective processes. Any diminution of these easterlies will permit the system to warm, whether by local processes (reduced upwelling, reduced zonal and meridional advection cooling) or by remote processes (e.g., Kelvin wave induced warming). Harrison and Schopf (1984) have previously pointed out the possibility that remotely forced anomalous zonal advection could have accounted for much of the observed central and eastern Pacific 1982 warming, while Philander and Seigel (1985) noted that anomalous zonal advection resulting from local wind stress changes appeared to be a major element of the 1982 warming in their hindcast (which used the NMC stresses). The hindcasts presented here suggest that both local and remote forcing played a role in the main 1982-83 warming, but there is much disagreement on the specifics of the warming in any given region; the wind stress differences are so large as to produce quite different surface current fields and, therefore, quite different patterns of horizontal advection of SST.
The problem of detailed model-data comparison with respect to upper ocean temperatures, was considered by Harrison et al. (1988, 1989). None of these hindcasts yielded what they define to be "quantitative hindcast skill" for SST. However, the range of hindcast SST values almost always bracketed the XBT observations that provided the validation dataset. Although the differences are larger than desired, no systematic inconsistency in model behavior is found. From this we infer that the hindcast errors introduced by the uncertainty in the imposed surface wind stress and heat flux fields must be reduced before more thorough model validation can be undertaken.
Of necessity, we have here confined our attention to the large amplitude periods of the 1982-83 ENSO event; the surface data are not adequate for discussion of the smaller amplitude periods. More, and better, SST data are needed if we are to study the small amplitude SST change periods of ENSO events. From this work, it appears that data adequate to get reliable rates of change of SST at the level of 0.5°C/mo are needed to study the small amplitude phases of ENSO events (or, it must be noted, the annual cycle of SST). These periods may be the crucial ones for validation of coupled ocean-atmosphere models because all existing ocean and atmosphere models give similar behavior in response to the anomalies that exist in the large amplitude phases of ENSO events. It is understanding the transitions between normal behavior and ENSO behavior that will bring the ability to predict ENSO events.
Acknowledgments. The programming assistance of Steve Hankin, Mark Verschell and
Jerry Davison is gratefully acknowledged. The hindcasts were performed on the CYBER 205 at
the NTIS/ERL CS
facility. Financial support from the
NOAA/EPOCS program and from Dr. E. Bernard, Director NOAA/PMEL is acknowledged with
thanks. This is PMEL publication No. 1052.
Firing, E., R. Lukas, J. Sadler and K. Wyrtki, 1983: Equatorial un- dercurrent disappears during 1982-83 El Niño. Science, 226, 1069-1071.
Halpern, D., 1987: Observations of the annual and interannual thermal and flow variations along the equator in the eastern Pacific ocean during 1980-1985. J. Geophys. Res., 92, 8303.
Harrison, D.E., and P.S. Schopf, 1984: Kelvin-Wave induced anomalous advection and the onset of surface warming in El Niño events. Mon. Wea. Rev., 112, 913-933.
---, W.S. Kessler and B.S. Giese, 1988: Model-data comparisons for the 1982-83 El Niño: the XBT tracks. NOAA Tech. Memo. ERL PMEL-79, Pacific Marine Environmental Laboratory, 61 pp.
---, --- and B.S. Giese, 1989: Ocean circulation model hindcasts of the 1982-83 El Niño; thermal variability along the ship of opportunity tracks. J. Phys. Oceanogr., 19, 397-418.
Hellerman, S., and M. Rosenstein, 1983: Normal monthly mean wind stress over the world ocean with error estimates. J. Phys. Oceanogr., 4, 145-167.
Meyers, G., J.R. Donguy and R.K. Reed, 1986: Evaporative cooling of the western equatorial Pacific Ocean by anomalous winds. Nature, 323, 523-526.
Pacanowski, R., and S.G.H. Philander, 1981: Parameterization of vertical mixing in numerical models of tropical oceans. J. Phys. Oceanogr., 11, 1443-1451.
Philander, S.G.H., and A.D. Seigel, 1985: Simulation of El Niño of 1982-1983. Coupled Ocean-Atmosphere Models, J. Nihoul, Ed., Elsevier, 517-541.
Rasmusson, E.M., and T.H. Carpenter, 1982: Variations in tropical sea-surface temperature and surface wind fields associated with the Southern Oscillation/El Niño. Mon. Wea. Rev., 110, 354-384.
---, and J.M. Wallace, 1983: Meteorological aspects of the 1982-83 El Niño. Science, 222, 1195.
Reed, R.K., 1986: Effects of surface heat flux during the 1972 and 1982 El Niño episodes. Nature, 322, 449-450.
Return to Abstract.