• Balmaseda, M. A., M. K. Davey, and D. L. T. Anderson, 1995: Decadal and seasonal dependence of ENSO prediction skill. J. Climate,8, 2705–2715.

  • Barnston, A. G., and Coauthors, 1994: Long-lead seasonal forecasts—Where do we stand? Bull. Amer. Meteor. Soc.,75, 2097–2114.

  • Battisti, D. S., and A. C. Hirst, 1988: Interannual variability in the tropical atmosphere-ocean system: Influence of the basic state and ocean geometry. J. Atmos. Sci.,45, 1687–1712.

  • Chang, C.-P., J. E. Erickson, and K. M. Lau, 1979: Northeasterly cold surges and near-equatorial disturbances over the winter MONEX area during December 1974. Part 1: Synoptic aspects. Mon. Wea. Rev.,107, 812–829.

  • Chao, Y, and S. G. H. Philander, 1993: On the structure of the Southern Oscillation. J. Climate,6, 450–469.

  • Cox, M. D., 1984: A primitive equation, 3-dimensional model of the ocean. GFDL Ocean Group Tech Rep. 1, 108 pp. [Available from Geophys. Fluid Dyn. Lab./NOAA, Princeton, NJ 08540.].

  • Cullen, M. J. P., 1993: The unified forecast/climate model. Meteor. Mag.,122, 81–94.

  • Davey, M. K., S. Ineson, and M. A. Balmaseda, 1994: Simulation and hindcasts of tropical Pacific Ocean interannual variability. Tellus,46A, 433–447.

  • Gent, P. R., and J. J. Tribbia, 1993: Simulation and predictability in a coupled TOGA model. J. Climate,6, 1843–1858.

  • Goldenberg, S. D., and J. J. O’Brien, 1981: Time and space variability of tropical Pacific wind stress. Mon. Wea. Rev.,109, 1190–1207.

  • Hasselmann, K., 1988: PIPs and POPs: The reduction of complex dynamical systems using principal interaction and oscillation patterns. J. Geophys. Res.,93(D9), 11 015–11 021.

  • Hastenrath, S., 1991: Climate Dynamics of the Tropics. Kluwer Academic Publishers, 488 pp.

  • Hellerman, S., and M. Rosenstein, 1983: Normal monthly windstress over the World Ocean with error estimates. J. Phys. Oceanogr.,13, 1093–1104.

  • Hirst, A. C., 1988: Slow instabilities in tropical ocean basin–global atmosphere models. J. Atmos. Sci.,45, 830–852.

  • Ineson, S., and C. Gordon, 1989: Parametrization of the upper ocean mixed layer in a tropical ocean GCM. Dynamical Climatology Tech. Note No. 74, 20 pp. [Available from Hadley Centre, Meteorological Office, London, Rd., Bracknell, Berkshire RG12 254, U.K.].

  • Kraus, E. B., and J. S. Turner, 1967: A one-dimensional model of the seasonal thermocline. Part II. Tellus,19, 98–105.

  • Latif, M., A. Sterl, E. Maier–Reimer, and M. M. Junge, 1993a: Climate variability in a coupled GCM. Part I: The tropical Pacific. J. Climate,6, 1–21.

  • ——, ——, ——, and ——, 1993b: Structure and predictability of the El Niño/Southern Oscillation phenomenon in a coupled ocean–atmosphere general circulation model. J. Climate,6, 700–708.

  • Legates, D. R., and C. J. Willmott, 1990: Mean seasonal and spatial variability in gauge-corrected, global precipitation. Int. J. Climatol.,10, 111–127.

  • Luther, D. S., D. E. Harrison, and R. A. Knox, 1983: Zonal winds in the central equatorial Pacific and El Niño. Science,222, 327–330.

  • McCreary, J. P., and D. L. T. Anderson, 1991: An overview of coupled ocean–atmosphere models of El Niño and the Southern Oscillation. J. Geophys. Res.,96, 3125–3150.

  • McPhaden, M. J., H. P. Freitag, S. P. Hayes, B. A. Taft, Z. Chen, and K. Wyrtki, 1988: The response of the equatorial Pacific Ocean to a westerly wind burst in May 1986. J. Geophys. Res.,93(C9), 10 589–10 603.

  • Mechoso, C. R., and Coauthors, 1995: The seasonal cycle over the tropical Pacific in general circulation models. Mon. Wea. Rev.,123, 2825–2838.

  • Moum, J. N., and D. R. Caldwell, 1994: Experiment explores the dynamics of ocean mixing. Eos Trans. Amer., Geophys., Union,75, 489–495.

  • Neelin, J. D., 1991: The slow sea surface temperature mode and the fast wave limit: Analytic theory for tropical interannual oscillations and experiments in a hybrid coupled model. J. Atmos. Sci.,48, 584–606.

  • ——, and Coauthors, 1992: Tropical air–sea interaction in general circulation models. Climate Dyn.,7, 73–104.

  • ——, M. Latif, and F.-F. Jin, 1994: Dynamics of coupled ocean–atmosphere models. Annu. Rev. Fluid. Mech.,26, 617–659.

  • Oberhuber, J. M., 1988: An atlas based on the COADS dataset. Max-Planck-Institut für Meteorologie Report No. 15, 20 pp. [Available from Max-Planck-Institut für Meteorologie, Bundestrasse 55, D-20146 Hamburg, Germany.].

  • Pacanowski, R., and S. G. H. Philander, 1981: Parametrization of vertical mixing in numerical models of the tropical oceans. J. Phys. Oceanogr.,11, 1443–1451.

  • Palmer, T. N., and D. L. T. Anderson, 1994: The prospects for seasonal forecasting—A review paper. Quart. J. Roy. Meteor. Soc.,120, 755–793.

  • Philander, S. G. H., 1990: El Niño, La Niña, and the Southern Oscillation. Academic Press, 293 pp.

  • ——, R. C. Pacanowski, N.-C. Lau, and M. J. Nath, 1992: Simulation of ENSO with a global atmospheric GCM coupled to a high-resolution tropical Pacific Ocean GCM. J. Climate,5, 308–329.

  • Reverdin, G., C. Frankignoul, E. Kestenare, and M. J. McPhaden, 1994: Seasonal variability in the surface currents of the equatorial Pacific. J. Geophys. Res.,99, 20 323–20 344.

  • Robertson, A. W., C.-C. Ma, M. Ghil, and C. R. Mechoso, 1995: Simulation of the tropical Pacific climate with a coupled ocean–atmosphere general circulation model. Part II: Interannual variabilty. J. Climate,8, 1199–1216.

  • Ropelewski, C. F., and M. S. Halpert, 1987: Global and regional scale precipitation patterns associated with the El Nino/Southern Oscillation. Mon. Wea. Rev.,115, 1606–1626.

  • ——, and ——, 1989: Precipitation patterns associated with the high index phase of the Southern Oscillation. J. Climate,2, 268–284.

  • Smith, T. M., and M. Chelliah, 1995: The annual cycle in the tropical Pacific Ocean based on assimilated ocean data from 1983 to 1992. J. Climate,8, 1600–1614.

  • Stockdale, T., D. Anderson, M. Davey, P. Delecluse, A. Kattenberg, Y. Kitamura, M. Latif, and T. Yamagata, 1993: Intercomparison of tropical ocean GCMs. WCRP-79, WMO/TD-No. 545, 90 pp. [Available from WMO, Publication Sales Unit, Case Postale 2300, CH-1211 Geneva 2, Switzerland.].

  • ——, M. Latif, G. Burgers, and J.-O. Wolff, 1994: Some sensitivities of a coupled ocean–atmosphere GCM. Tellus,46A, 367–380.

  • Tett, S., 1995: Simulation of El Niño–Southern Oscillation-like variability in a global AOGCM and its response to CO2 increase. J. Climate,8, 1473–1502.

  • Trenberth, K. E., and D. J. Shea, 1987: On the evolution of the Southern Oscillation. Mon. Wea. Rev.,115, 3078–3096.

  • von Storch, H., G. Bürger, R. Schnur, and J.-S. von Storch, 1995: Principal oscillation patterns: A review. J. Climate,8, 377–400.

  • Ward, M. N., and C. K. Folland, 1991: Prediction of seasonal rainfall in the north Nordeste of Brazil using eigenvectors of sea-surface temperature. Int. J. Climatol.,11, 711–743.

  • Woodruff, S. D., R. J. Slutz, R. L. Jenne, and P. M. Steurer, 1987: A comprehensive ocean–atmosphere data set. Bull. Amer. Meteor. Soc.,68, 1239–1250.

  • Wyrtki, K., and B. Kilonsky, 1984: Mean water and current structure during the Hawaii to Tahiti shuttle experiment. J. Phys. Oceanogr.,14, 242–254.

  • View in gallery

    Monthly mean sea surface temperature from COADS climatology for (a) March and (b) September and from the CGCM simulation for (c) March and (d) September. Contour interval is 1°C, temperatures greater than 28°C are shaded.

  • View in gallery

    Annual mean equatorial longitude–depth section of ocean temperature from the CGCM. Contour interval is 1°C.

  • View in gallery

    Monthly mean wind stress from the Florida State University climatology (1966–85) for (a) March and (b) September and from the CGCM simulation for (c) March and (d) September. Contour interval is 0.025 N m−2.

  • View in gallery

    Monthly mean 850-hPa winds from the UKMO operational archive climatology (1986–94) for (a) JJA and (b) DJF and from the CGCM simulation for (c) JJA and (d) DJF. Contour interval is 5 m s−1.

  • View in gallery

    Monthly mean CGCM tropical precipitation for (a) March and (b) September. Contour interval is 2.5 mm day−1, regions of precipitation greater than 10 mm day−1 are shaded.

  • View in gallery

    Monthly mean CGCM ocean surface currents for (a) March and (b) September. Contours are at 5, 10, 20, and thereafter every 20 cm s−1.

  • View in gallery

    Annual mean latitude–depth section through 150°W of zonal current for the CGCM. Contours are at 0, ±5, ±10, and thereafter every 10 cm s−1.

  • View in gallery

    Annual mean net heat flux from (a) the climatology of Oberhuber and (b) the CGCM. Contour interval is 25 W m−2.

  • View in gallery

    Correlations (×10) of CGCM annual mean sea level pressures with the Darwin values. Correlations are greater than 0.4 in the shaded regions and less than −0.4 in the hatched regions.

  • View in gallery

    CGCM time series of (a) sea surface temperature anomalies (°C) averaged over the Nino-3 area (5°N–5°S, 150°–90°W), and (b) zonal wind stress anomalies (N m−2) averaged over the central Pacific (5°N–5°S, 165°E–135°W).

  • View in gallery

    Time–longitude section of equatorial model sea surface temperature anomalies for years 8–31 of the CGCM simulation. Contour interval is 1°C, negative values are shaded.

  • View in gallery

    Time–longitude section of model vertically averaged temperature anomalies (top 360 m) for years 8–19 of the CGCM simulation at (a) the equator and (b) 6°N. Contour interval is 0.25°C, negative values are shaded.

  • View in gallery

    Time–longitude section of model zonal wind stress anomalies for years 8–31 of the CGCM simulation. Contours are at 0, ±0.01, ±0.02, and thereafter every 0.02 N m−2, negative values are shaded.

  • View in gallery

    Time series of the real (solid) and imaginary (dashed) components of the ENSO-related CGCM POP mode, obtained using time series for five EOFs of the combined SST′–HC′–TAUX′ fields.

  • View in gallery

    Spatial patterns of the ENSO-related CGCM POP mode obtained using five EOFs for combined SST′–HC′–TAUX′ fields. SSTI, HCI, and TAUXI (panels a, c, e) are associated with onset of an ENSO event, while SSTR, HCR, and TAUXR (b, d, f) are associated with the peak ENSO phase (warm). The patterns are components of the normalized combined patterns popR and popI, multiplied by scales ΓR and ΓI. Contour interval is 5 units, negative contours dashed.

  • View in gallery

    CGCM precipitation anomaly patterns associated with the model surface temperature anomaly in the Nino-3 area (Fig. 10a). Contours are at 0, ±5, ±10, ±25, and ±50 mm day−1. Values are greater than 5 mm day−1 in shaded regions and less than −5 mm day−1 in hatched regions.

  • View in gallery

    CGCM 850-hPa winds in year 17 for selected areas for (a) 1 December, (b) 6 December, (c) 11 December, and (d) 3 November. Contour interval is 5 m s−1.

  • View in gallery

    Equatorial time–longitude sections from the CGCM for October year 17 to March year 18 of (a) zonal wind stress (contour interval is 0.05 N m−2); (b) vertically averaged temperature (top 360 m) (contour interval is 0.5°C); and (c) sea surface temperature (contour interval is 1°C, temperatures greater than 28°C are shaded).

  • View in gallery

    Time series of CGCM sea surface temperature anomalies (°C) averaged over the Nino-3 area (5°N–5°S, 150°–90°W) for the control simulation (solid line) and for two ensembles (dashed lines). A solid circle indicates the monthly mean temperature anomaly of each ensemble member after 6 months of integration; the label indicates the month when each ensemble was initiated.

  • View in gallery

    Average positive and negative model sea surface temperature anomalies (°C) for the Nino-3 area (5°N–5°S, 150°–90°W) for each month-of-year, from the 25-yr CGCM simulation.

  • View in gallery

    Time series of CGCM precipitation (solid line) and precipitation for each ensemble member (solid circles) at the end of the hindcast period for (a) the central Pacific (5°N–5°S, 180°–240°E); (b) the India monsoon region (10°–25°N, 70°–95°E); and (c) southeastern Africa (10°–27.5°S, 25°–50°E). The CGCM seasonal cycle of precipition (dashed line) and observations (dotted line) are shown for each area. Units are millimeters per day.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 8 8 1
PDF Downloads 1 1 0

Interannual Climate Simulation and Predictability in a Coupled TOGA GCM

View More View Less
  • 1 Hadley Centre, U.K. Meteorological Office, London Road, Bracknell, United Kingdom
© Get Permissions
Full access

Abstract

A Pacific Ocean–global atmosphere general circulation model is used to simulate the climatic mean state and variability in the Tropics, up to interannual timescales. For this model no long-term trend in climate occurs, but there are systematic differences between the model mean state and observations: in particular, the east equatorial Pacific sea surface temperature is too high by several degrees. Along the equator the seasonal variability in sea surface temperature is good although some features of the seasonal cycle are unrealistic: for example, the east Pacific convergence zone crosses the equator twice a year, residing in the summer hemisphere.

Despite some deficiencies in the simulation of the mean state, there is substantial interannual variability, with irregular oscillations dominated by a 2-yr cycle. A principal oscillation pattern analysis shows that the interannual anomalies are typically generated in the west Pacific and move eastward along the equator, with closely connected oceanic and atmospheric components. The patterns are similar to those associated with observed El Niño events. Rainfall anomalies associated with the model El Niño events also have several realistic features.

Idealized seasonal prediction experiments were made by slightly perturbing the atmospheric component: three 6-month hindcasts were thus made for each of several start times spread through an El Niño cycle. Predictability of central Pacific sea surface temperature anomalies was best for hindcasts starting near a warm El Niño peak. Generally, hindcasts starting in September and December were more accurate, with less spread, than those starting in March and June. The behavior and predictability of seasonal rainfall in several regions was also analyzed. For example, a warm model El Niño produces enhanced rainfall in the central equatorial Pacific and reduced rainfall in the Indian region, which is reproduced consistently in the hindcasts.

The model also shows variability on shorter timescales, and an example is presented of a spontaneous westerly wind burst in the west Pacific and its oceanic impact.

Corresponding author address: Sarah Ineson, Hadley Centre for Climate Prediction and Research, Meteorological Office, London Road, Bracknell, Berkshire RG12 2SY United Kingdom.

Email: sineson@email.meto.gov.uk

Abstract

A Pacific Ocean–global atmosphere general circulation model is used to simulate the climatic mean state and variability in the Tropics, up to interannual timescales. For this model no long-term trend in climate occurs, but there are systematic differences between the model mean state and observations: in particular, the east equatorial Pacific sea surface temperature is too high by several degrees. Along the equator the seasonal variability in sea surface temperature is good although some features of the seasonal cycle are unrealistic: for example, the east Pacific convergence zone crosses the equator twice a year, residing in the summer hemisphere.

Despite some deficiencies in the simulation of the mean state, there is substantial interannual variability, with irregular oscillations dominated by a 2-yr cycle. A principal oscillation pattern analysis shows that the interannual anomalies are typically generated in the west Pacific and move eastward along the equator, with closely connected oceanic and atmospheric components. The patterns are similar to those associated with observed El Niño events. Rainfall anomalies associated with the model El Niño events also have several realistic features.

Idealized seasonal prediction experiments were made by slightly perturbing the atmospheric component: three 6-month hindcasts were thus made for each of several start times spread through an El Niño cycle. Predictability of central Pacific sea surface temperature anomalies was best for hindcasts starting near a warm El Niño peak. Generally, hindcasts starting in September and December were more accurate, with less spread, than those starting in March and June. The behavior and predictability of seasonal rainfall in several regions was also analyzed. For example, a warm model El Niño produces enhanced rainfall in the central equatorial Pacific and reduced rainfall in the Indian region, which is reproduced consistently in the hindcasts.

The model also shows variability on shorter timescales, and an example is presented of a spontaneous westerly wind burst in the west Pacific and its oceanic impact.

Corresponding author address: Sarah Ineson, Hadley Centre for Climate Prediction and Research, Meteorological Office, London Road, Bracknell, Berkshire RG12 2SY United Kingdom.

Email: sineson@email.meto.gov.uk

1. Introduction

Variability in the Tropics is observed on a wide range of time and space scales. The largest interannual variations occur in the tropical Pacific region and are associated with strong interaction between the atmosphere and the ocean. Philander (1990) gives a comprehensive description of the El Niño–Southern Oscillation (ENSO) phenomenon. The impact of these variations affects not only the Tropics but also extratropical regions, and considerable observational and modeling effort has been made in trying to understand, simulate, and predict this important variability. A wide range of coupled ocean–atmosphere models of the interannual ENSO activity have been developed (McCreary and Anderson 1991). The use of simplified coupled models has led to an understanding of a variety of ENSO mechanisms (Neelin et al. 1994). Useful predictive skill at lead times of many months has been demonstrated for several models (see, e.g., reviews by Barnston et al. 1994; Palmer and Anderson 1994).

It is generally assumed that the best representation of climate and its variability will eventually be made by those models that attempt to describe the physical and dynamical processes fully, that is, coupled general circulation models (CGCMs). Progress has been gradual, due to the large complexity and computational cost of such models. Substantial climate drift can easily occur, as described in the intercomparison of 17 models by Neelin et al. (1992). The tropical Pacific behavior of 11 CGCMs recently summarized by Mechoso et al. (1995) shows that although considerable progress has been made, CGCMs generally still have difficulty in accurately representing some basic features of the observed mean state. Despite these difficulties, several CGCMs produce ENSO-like interannual behavior (see, e.g., Neelin et al. 1992; Stockdale et al. 1994; Latif et al. 1993a; Philander et al. 1992; Robertson et al. 1995), and the CGCMs have been applied to ENSO prediction (e.g., Latif et al. 1993b).

In this paper, results are presented from a relatively high resolution CGCM that consists of a global atmosphere model coupled to a tropical Pacific Ocean model, as described in section 2. Features of the model seasonal cycle (section 3) and interannual variability (section 4) are discussed. On shorter timescales, an example of an internally generated westerly wind burst and its impact on the ocean is given in section 5. An assessment of the potential predictability of sea surface temperature anomalies and associated rainfall anomalies has been made by making a series of hindcasts, produced by perturbing the control integration slightly for a range of start times (section 6).

2. Model description

The CGCM used in this experiment is a Tropical Ocean Global Atmosphere (TOGA) configuration of the United Kingdom Meteorological Office (UKMO) Unified Model. An overview of the Unified Model is given by Cullen (1993). In this version the Hadley Centre climate resolution atmospheric general circulation model (AGCM) is coupled to a limited area, tropical Pacific Ocean general circulation model (OGCM).

The atmosphere model has a 2.5° latitude by 3.75° longitude resolution and 19 levels on a hybrid coordinate vertical grid. A split-explicit integration scheme is used and cloud water and ice are included as prognostic variables. A comprehensive physics package is used, including a stability dependent cloudy boundary layer scheme, a land surface hydrology scheme, and a radiation scheme with interactive optical properties. The mass flux convection scheme used to represent shallow, deep, and midlevel convection includes deep convective downdrafts. For further details refer to Cullen (1993).

The ocean model is a Bryan–Cox-type general circulation model (Cox 1984) on a variable spatial grid. The domain is limited to the tropical Pacific Ocean with a meridional grid spacing of 1/3° at the equator increasing poleward to 1° at the open boundaries at 30°N and 30°S. Zonally, the spacing is 1.5° over most of the ocean, decreasing to 0.5° near the closed eastern and western boundaries. There are 16 levels in the vertical with a concentration of levels near the ocean surface. The barotropic mode is neglected and there is no variation in model bottom topography. The model has an active salinity field. The Richardson number–dependent mixing scheme of Pacanowski and Philander (1981) is included to represent shear mixing, and an embedded mixed layer model (Kraus and Turner 1967) is used to parameterize the effects of surface generated turbulence on the mixed layer (Ineson and Gordon 1989). A seasonal cycle simulation using this model is described in detail in Stockdale et al. (1993), who discuss results from an intercomparison study of a number of tropical Pacific OGCMs. The model simulation of interannual oceanic variability obtained when forced with observed wind stresses is described in Davey et al. (1994).

The component models are coupled via wind stress, heat, and freshwater fluxes, which are accumulated and exchanged every 5 days. No flux correction techniques are used. Outside the ocean domain the atmosphere model is forced with climatological sea surface temperatures. The length of the coupled model integration was 32 years. During the first seven years of the simulation, changes were made to the atmospheric component, so the results presented here are from the last 25 years. Results are mostly presented in terms of monthly climatology (25-yr average for the CGCM) and anomalies (departures from monthly climatology, denoted by primes). Further details and results can be found in the CGCM intercomparison by Mechoso et al. (1995).

Results from a previous experiment, in which the same high-resolution OGCM was coupled to a substantially different AGCM (the U.K. Meteorological Office 11-layer climate model), are contained in Neelin et al. (1992). In that version, interannual variability was very weak and there was a gradual cooling trend in Pacific SST. The interannual behavior of another version of the CGCM, using a lower resolution global OGCM, has been described by Tett (1995): he found moderate tropical Pacific interannual variability.

3. Simulation of the seasonal cycle

In this section several aspects of the model climatology are presented, for the atmospheric and oceanic components, to illustrate some of the CGCM strengths and weaknesses.

a. Sea surface temperature

The SST plays a fundamental role in the complicated interactions that occur between the ocean and the atmosphere. Its evolution is determined by both oceanographic and air–sea interaction processes and various feedback processes can lead to the amplification of errors. Achieving a good seasonal simulation is important as the associated convective activity that occurs over the warmest tropical waters involves large latent heat release, driving large-scale motions that impact not only on the Tropics but also on the midlatitudes.

Model and observed (COADS; Woodruff et al. 1987) tropical Pacific SST climatology for March and September is shown in Fig. 1. In the west Pacific the maximum model SST is close to (slightly below) climatological values; like the observations, the annual range of model SST is small and the region of largest SST follows the sun seasonally across the equator. In the east Pacific, however, the model SST is generally too warm, and the zonal gradient in the central Pacific is weaker than observed. This defect is associated with weak easterly surface wind stress (see below), and is induced at least partly by an underestimation of the observed marine stratocumulus sheets, which occur over the Pacific near the South American coast. This cloud type is poorly represented by this version of the AGCM. In March the east Pacific warm bias south of the equator is sufficient to allow the formation of a continuous band of SST >28°C that extends from the west into the east Pacific (Fig. 1c), contrary to observations (Fig. 1a). In September (Figs. 1b,d), when the observed cold tongue in the east Pacific is well developed on the equator, the model cold tongue is also prominent. However, it is too warm compared to observations and has become separated from the coast by even warmer water. The model SST close to the South American coast is too warm all year round, with a maximum error at 10°S reaching 9°C: the model completely fails to reproduce the observed cool SST there.

For the annual mean along the equator, the SST is too warm in the east Pacific, and the east–west gradient is weaker than observed (see Fig. 1 in Mechoso et al. 1995). The SST seasonal cycle along the equator [departures from annual mean, Fig. 2, Mechoso et al. (1995)] is quite good, with realistic amplitude and timing despite the mean warm bias; however, the observed westward phase propagation in the east Pacific is not reproduced.

b. Ocean thermocline

A vertical section of annual mean model ocean temperature is shown in Fig. 2. Compared with observations (e.g., the ocean analysis by Smith and Chelliah 1995), the thermocline is too diffuse and the slope from west to east is too gradual in the central Pacific. In the west Pacific, the pool of water warmer than 28°C is shallower than observed, but the 20°C isotherm is close to the observed depth. In the east Pacific, the thermocline is well defined at about 25-m depth, but the water is generally too warm in the top 200 m. The 20°C isotherm is near 80 m, about twice the observed depth.

c. Surface wind stress

The mean observed and model surface wind stress for March and September in the tropical Pacific region are given in Fig. 3. [The observed field is the Florida State University 1966–85 pseudostress (Goldenberg and O’Brien 1981) with a drag coefficient of 0.75 × 1.535 × 10−3.] The model and observed stress patterns are generally similar. Along the equator the CGCM stresses are weaker than observed. In March the model stresses in the central equatorial Pacific are small and directed across the equator, and are even turning westerly near 10°S, 150°W. The equatorial behavior is consistent with the underlying weak zonal SST gradient (leading to a weak zonal Walker circulation) and with surface winds directed toward the warm SST extension south of the equator. Off the equator the model apparently overestimates the maximum values in the easterly trades by about 25%. However, the drag coefficient stated above was chosen to give wind stresses of realistic strength along the equator: the model trade maxima are generally similar in magnitude to the Hellerman climatology (Hellerman and Rosenstein 1983).

d. Monsoon winds

The observed and model 850-hPa winds over the Indian–west Pacific region for June–July–August (JJA) and December–January–February (DJF) means are provided in Fig. 4. (The observed winds are averages taken from the UKMO operational archive for 1986–94.) In JJA the CGCM winds are in very good agreement with those observed, in both strength and direction. In DJF the agreement is good for the main jets, but over Indonesia the model does not reproduce the westerly winds south of the equator well.

e. Precipitation

The model precipitation fields for March and September are shown in Fig. 5 for the whole Tropics. The impact of the errors in the model SST are evident. In particular, during the early calendar months the South Pacific convergence zone (SPCZ) extends too far east from the west Pacific, over the band with SST > 28°C. In the east Pacific this anomalous extension blends into the intertropical convergence zone (ITCZ). The observed climatological ITCZ stays north of the equator all year in the east Pacific, but in the CGCM the ITCZ crosses the equator to give maximum rainfall south of the equator in March. [The seasonal cycle in the east Pacific is illustrated in Fig. 5, Mechoso et al. (1995).]

f. Ocean currents

The mean March and September Pacific Ocean surface currents (model level 1, depth 5 m) are shown in Fig. 6. They can be compared with the currents analyzed by Reverdin et al. (1994) from observations for the period 1987–92. In March the strongest currents are westward, on the equator near the dateline. Flow is generally westward north of the equator: the observed North Equatorial Countercurrent (NECC) is missing. In the east Pacific, flow is weakly eastward on the equator. In September, the strongest westward currents have moved to the eastern equatorial Pacific, and the NECC has developed around 5°N in the east-central Pacific.

Figure 7 shows a latitude–depth section of the annual mean upper-ocean currents at longitude 150°W. This section can be compared with the observations from the Hawaii–Tahiti shuttle experiment (Wyrtki and Kilonsky 1984). The most prominent feature is the eastward equatorial undercurrent, which is well developed, reaching over 0.7 m s−1 (about 0.9 m s−1 observed), despite the thermocline slope being flatter than observed. The westward surface flow (South Equatorial Current, SEC) has a maximum north of the equator, but it is weaker (0.3 m s−1) than that observed (>0.6 m s−1). The eastward NECC at this longitude has a subsurface maximum (as observed), but mean surface flow is westward rather than eastward, as Fig. 6 suggests.

g. Net surface heat flux

The annual mean net downward surface heat flux for the Pacific region is provided in Fig. 8 for the model and as estimated from observations (Oberhuber 1988). The main feature in each field is the strong maximum (over 100 W m−2) in the equatorial east Pacific, where the model matches the observed estimate. The region of large equatorial fluxes extends farther west in the model, but the fluxes are weak in the west Pacific in both model and observation.

4. Interannual variability

The main objective in developing this CGCM was to obtain a realistic representation of tropical Pacific interannual variability. In this section the strong interannual variability of our CGCM is described and analyzed in various ways.

a. Anomaly fields

Interannual variability in the atmosphere is characterized by the Southern Oscillation, a redistribution of atmospheric mass between the Indonesia–west Pacific and the eastern Pacific regions. Figure 9 shows correlations r of model annual average sea level pressure with the model grid point closest to Darwin, Australia, for the 25 years of model integration. As in observations, the model shows two main centers with substantial correlations: one in the west Pacific–Indonesian area r > 0.6, and one in the central east Pacific with r < −0.4. Comparison with observational data (Trenberth and Shea 1987) shows that the model captures the essence of the signal well over the Pacific and east Indian Oceans with a good positioning of the centers, although the coherence is not as strong as is observed, with a maximum anticorrelation in the east Pacific of −0.6 rather than −0.8. Contrary to observations, the center with r < −0.4 extends into the Gulf of Mexico in the CGCM, and there is high correlation between northeast Canada and Darwin.

Atmospheric variability on interannual timescales is accompanied by large-scale oceanic variations in the tropical Pacific. Figure 10a illustrates the time series of model SST′ for the model simulation averaged over the Nino-3 region (5°N–5°S, 150°–90°W) (prime denotes departures from monthly mean climatology). Strong interannual variability is evident, with a maximum amplitude of 2°C. The variability is irregular but with a prominent 2-yr cycle. A number of ENSO-like events are followed by a quiet period of several years, then ENSO activity is resumed. Zonal wind stress anomalies averaged over the central equatorial Pacific (5°N–5°S, 165°E–135°W) are shown in Fig. 10b. Warm episodes are associated with anomalous westerlies and cold episodes by increased easterlies. There is also considerable variability on shorter timescales than ENSO.

The time–longitude diagram of model equatorial SST′ is shown in Fig. 11. Warm and cold ENSO-like episodes occur, with a tendency for SST anomalies to appear in the west Pacific and spread eastward. (Occasional westward movement is also apparent.) The eastward movement is relatively slow in the west Pacific, accelerating in the central Pacific. The largest amplitude tends to occur in the central Pacific region, whereas observed interannual SST variability is largest in the east Pacific.

Interannual variability of the upper ocean heat content is also substantial. Heat content anomalies are closely related to changes in thermocline depth and hence are strongly influenced by equatorial waves. We use vertically averaged temperature (VAT) for the top 360 m of the ocean to measure this activity. Figure 12 shows time–longitude diagrams of VAT′ along the equator, and at 6°N, for years 8 to 19. Eastward propagation is evident along the equator, and equatorial VAT′ and SST′ are clearly closely connected. Off the equator, propagation is westward. Such behavior has been found in other models (e.g., Chao and Philander 1993).

Zonal wind stress anomalies TAUX′ (Fig. 13) also behave like SST′ and VAT′, with eastward propagation along the equator. The wind stress is more noisy, and several episodes with enhanced westerlies are apparent (e.g., December year 17 near 160°W).

b. ENSO principal oscillation patterns

To investigate the structure of the model ENSO activity, a principal oscillation pattern (POP) analysis (Hasselmann 1988; von Storch et al. 1995) of combined SST′ and VAT′ and TAUX′ was carried out. POP analysis is a method used to represent the data in terms of temporally oscillating and decaying modes (appendix A). Each oscillating mode has a characteristic period and for a free oscillating POP mode the spatial patterns popR and popI appear in the sequence
IRIRI
at quarter-period intervals. For this analysis, SST′, VAT′, and TAUX′ monthly fields were separately normalized by their standard deviations (0.56°C, 0.38°C, and 0.016 N m−2, respectively), then empirical orthogonal functions (EOFs) of the combined fields were obtained, ranked in order of decreasing explained variance. POP analysis was then applied to the time series of EOF coefficients.

Results from the POP analysis using five EOFs are described below. (The first five EOFs explain 29%, 12%, 4.2%, 3.8%, and 3.4%, respectively, of the total variance of the combined fields.) The period of the dominant oscillating mode is about 28 months, with a decay timescale of 18 months, and is clearly related to the model ENSO. Other POP modes generally had much more rapid decay timescales. The dominant POP ENSO mode is robust and differs little when more EOFs are used in the analysis.

Figure 14 shows time series TR and TI for the ENSO-related mode, with corresponding spatial patterns in Fig. 15. (Here, popR and popI are quite similar in structure to EOF patterns 1 and 2, respectively.) The spatial patterns have been multiplied by scaling factors ΓR = 1495 and ΓI = 1446. The positive and negative extrema in TR correspond well to warm and cold extremes, respectively, in Nino-3 SST′. The spatial patterns popR represent conditions near a warm event peak (and –popR represents the cold phase of the ENSO cycle), while popI corresponds to conditions at the onset of a warm phase of the ENSO cycle. From the time series in Fig. 14, we see that the magnitude of this POP mode is relatively large during the main model ENSO events, when the patterns generally follow the sequence (1). At other times, magnitude is small and phase is not as clearly ordered.

For SST′ the onset pattern SSTI has a positive (warm) maximum on the equator near the dateline, with negative values along the equator in the east Pacific that extend along the South American coast. Weak negative values are also found in the southwest Pacific sector. This onset pattern develops a quarter period later into the peak pattern SSTR: the warm patch has amplified and shifted eastward to give large and positive SST′ in the equatorial central-east Pacific, with warm anomalies extending poleward and eastward. Farther south, a substantial negative anomaly also develops along 20°S. SSTR then evolves to −SSTI: the eastern features in SSTI can be regarded as the remnants of a preceding peak phase. Overall, SST′ has a substantial eastward propagating component.

Onset pattern VATI has a big warm patch east of the date line that includes a warm maximum south of the equator. (This southern maximum may be related to downwelling induced by preceding −TAUXR, see below.) Elsewhere, VATI is weakly warm all along the equator, with weak cool spots off the equator near the coasts. A quarter cycle later, the equatorial warm patch has moved to the east Pacific, spreading poleward along the east coast, while the west Pacific has cooled with large negative maxima appearing off the equator with substantial zonal extent. Thus, the peak VATR pattern has strong zonal contrast.

At onset, the main feature in TAUXI is a positive (westerly) patch on the equator in the west Pacific, located to the west of the warm equatorial anomaly in SSTI. There are weaker easterly anomalies south of this westerly patch and weak westerlies in the southeast Pacific. The peak pattern TAUXR has a patch of strong westerlies in the central equatorial Pacific, again located to the west of the warm equatorial anomaly in SSTR, with a maximum slightly south of the equator. There is a belt of substantial easterlies farther south, where SSTR is cool. This pattern implies large positive wind stress curl around 10°S that would cause upwelling: this could explain the subsequent cooling in that region, as seen in −VATI.

The overall impression is of eastward propagation along the equator, with wind, heat content, and SST acting in unison, consistent with the equatorial time–longitude diagrams. Note that the VAT′ patterns have positive (warm) equatorial maxima that occur to the east of the corresponding SST′ equatorial warm maxima, in regions where TAUX′ is weak. Combined with eastward propagation, this suggests that SST changes are strongly influenced by equatorial waves.

These CGCM POP patterns can be compared with other ENSO-related POP patterns. Latif et al. (1993b) describe observed and CGCM POPs, and Davey et al. (1994) have POP-analyzed a Pacific Ocean simulation with the same OGCM as here, forced by observed winds. (Please note: in Davey et al. the peak pattern is shown as popI, whereas here it is popR; the choice is arbitrary.) The overall picture is similar in each case: there is a dominant interannual mode with strong related features in the SST, VAT, and TAUX fields, with evident eastward propagation along the equator. One main difference is that this CGCM POP mode has a substantial eastward propagating SST′ signal, whereas observations and simulations have an SST′ signal that is more like a standing oscillation. The ENSO-related activity south of the equator in our CGCM is not found in the observations or other simulations: this is a coupled model artifact that is probably related to the systematic warm ocean error there (see Fig. 1). An encouraging feature is that central-east SST′ is less confined to the equator in the CGCM than in the OGCM simulation, and more in line with observed behavior. This may be due to the more physically consistent treatment of surface heat flux in the CGCM.

c. ENSO associated rainfall

To examine the rainfall changes associated with the model ENSO cycle, the associated pattern AP(xm) is formed:
i1520-0493-125-5-721-eq1
where NINO3′(tn) is the normalized time series for SST′ in the Nino-3 region at n time points tn and PPTN′(xm, tn) is the precipition anomaly time series at m spatial points xm and n time points. The resulting spatial pattern is shown in Fig. 16. (A very similar pattern is obtained if the TR ENSO POP time series is used in place of NINO3′.) The largest precipitation signal is over the central equatorial Pacific, where positive values indicate increased (decreased) precipitation at a warm (cold) peak of the ENSO cycle. The signal has opposite sign to the north and south of the equatorial maximum. A decrease in precipitation over northeastern South America and the Indian region is associated with warm ENSO. These features are similar to those found in the analysis of global- and regional-scale precipitation anomalies associated with ENSO by Ropelewski and Halpert (1987, 1989).

d. Coupling and oscillation mechanism

An important component of the coupled system is the effect of local SST′ on local TAUX′. In the CGCM, we find that warm (cool) SST′ on the equator is associated with westerly (easterly) TAUX′ in a region immediately to the west of the SST′. By examining scatterplots of SST′ in 20° longitude by 10° latitude boxes versus TAUX′ in the similar 20° × 10° box immediately to the west, we find that the relation is remarkably linear. The coupling strength increases substantially from east to west: 0.011 N m−2 °C−1 near 220°E, 0.017 near the date line, and 0.033 near 160°E. Thus, in the CGCM only a small SST anomaly is required to produce a substantial TAUX anomaly in the west Pacific. The reason for this behavior is not clear, but sensitivity of atmospheric convection to SST is likely to be a factor.

In turn, the equatorial TAUX′ generate ocean current and thermocline depth anomalies, with westerly TAUX′ causing deepening to the east (the Kelvin response) and shallowing to the west (the Rossby response) of the TAUX′ location. Experiments with long local wind bursts applied to the OGCM in a realistic state (i.e., forced by observed winds) confirm this picture and also show that, in terms of equatorial thermocline movement, the Kelvin response is substantially larger than the Rossby response. Such experiments also suggest that western boundary reflection of equatorial waves in the OGCM is very inefficient, for reasons that are not yet clear. These wind effects (Kelvin and Rossby response, and very weak western reflection) are also apparent in lag correlations calculated using the CGCM TAUX′ and VAT′ fields. (The lag correlations are noisy and show no obvious patterns if monthly values are used but give clear and consistent signals when a 7-month running mean is applied to the data. Note that eastward propagation in the equatorial fields is emphasized when the running mean is used.) Local lag correlations also reveal a close relation between equatorial SST′ and VAT′, with VAT′ leading SST′ by 2–3 months in the central Pacific and by 1 month in the east Pacific where the thermocline is shallower, with correlations r > 0.8. In the west Pacific the correlation of SST′ and VAT′ is weak.

The interannual warm and cold episodes typically develop in the west-central Pacific and move eastward as coupled modes, as shown by the POP analysis in section 4b. For example, in the third quarter of year 12 westerly TAUX′ amplifies near 160°W (where SST′ is weakly warm): this causes positive (negative) VAT′ to the east (west), with consequent positive (negative) SST′ development to the east (west) of the wind anomaly (see Figs. 11, 12, and 13). The SST′ amplifies, and the system propagates eastward.

There is also occasional westward propagation from the east Pacific in the CGCM. For example, in years 18 and 27, cold SST′ moves westward from 100°W, accompanied by easterly TAUX′, with a weak negative VAT′ effect. Westward movement is less intense and short-lived compared to the eastward event.

These facts suggest that the dominant oscillation mechanism is associated with thermocline movement, and broadly the coupled Kelvin mode (Hirst 1988) is operating in the CGCM. In such a mode, local SST′ drives local TAUX′, and consequent thermocline changes, as described above. The thermocline deepening (for warm SST′) to the east of TAUX′ then causes SST′ (and VAT′) to gradually increase there, provided that the local thermodynamic balance is dominated by changes in vertical temperature advection. Likewise, SST′ and VAT′ to the west gradually decrease. The net effect is that the initial warm SST′ region can amplify (local positive feedback, which is limited in practice by nonlinear effects) and shift eastward. To the west, the decrease in SST′ leads to more easterly TAUX′ and hence to a cold anomaly that in turn amplifies and follows the warm anomaly eastward.

The occasional signs of westward propagation in the CGCM suggest that ocean surface layer mechanisms are also active. For example, to the west of a cool SST anomaly easterly TAUX′ in an easterly background can cause increased upwelling into the surface well-mixed layer, leading locally to cooler SST′ and thus westward propagation of the cool anomaly.

The analysis by Neelin (1991) shows how different surface and subsurface mechanisms can compete or cooperate to generate a range of eastward, westward, and stationary coupled ocean–atmosphere modes. In effect, the CGCM contains several such mechanisms, and its interannual variability is determined by an ever-changing balance, with changes in thermocline depth being a major factor. The absence of evidence for western-boundary wave reflection in longitude–time sections or in lag correlations suggests that the delayed oscillator mechanism (cf. Battisti and Hirst 1988) involving such reflection does not operate in this CGCM.

5. Westerly windburst

Atmospheric variability in the west Pacific on timescales of one or two weeks is characterized by westerly wind bursts. These wind events are most likely to occur from the period November to March (Luther et al. 1983) and during warm ENSO events. Moum and Caldwell (1994) document a wind burst episode that took place in late December 1992 during the intensive operations period of the TOGA COARE experiment. In our CGCM experiment, several strong wind bursts occurred spontaneously at locations from 150°E to 160°W. There was a particularly large burst in December, year 17, and in this section this individual synoptic event will be described in detail.

In the model, toward the end of October year 17, light westerly winds occurred in the west Pacific. Figures 17a–c show instantaneous 850-hPa winds for a time sequence corresponding to the major westerly burst in December year 17. On 1 December, a cyclonic feature is apparent to the south of the equator. By 6 December, the disturbance has become organized into a twin cyclone pair centered almost on the equator with maximum westerly wind speeds reaching 25 m s−1. Six days later the southern cyclone has moved to the southeast and shortly thereafter the system breaks down.

This burst may be related to previous activity over southeast Asia. An instantaneous picture of the 850-hPa winds for 3 November year 17 is shown in Fig. 17d. During the early part of November, cyclonic development took place to the east of the Philippines and strong northeasterly winds occurred to the north of the Philippines and over the South China sea. The wintertime cold surface high over Siberia and north China is a source area for cold-air outbreaks and the penetration of cold air into the Tropics is believed to enhance convective activity in the equatorial belt (Hastenrath 1991; Chang et al. 1979). We are not able to connect the two events more closely, unfortunately, as only a limited amount of daily information was saved during the CGCM integration.

In the model simulation, mean wind stresses in the west Pacific are light, generally less than 0.02 N m−2. Figure 18a shows the zonal component of the equatorial wind stress for the model simulation period October year 17 to April year 18. During this particular year westerly winds occur in the west Pacific from late October through to early January. There is an apparent eastward propagation of a disturbance with occasional development giving strengthened westerly bursts. The most intense westerly burst, which corresponds to the 850-hPa wind sequence in Figs. 17a–c, takes place in early December and has a maximum magnitude of 0.15 N m−2. Figure 18b shows VAT for the same period. An eastward propagating signal is seen that can be followed right across to the eastern boundary. The signal first appears during November and has a propagation speed that remains more or less constant at 1.7 m s−1 and for a limited time the westward flowing South Equatorial Current becomes an eastward current. Amplification of the VAT signal (but not a significant change in phase speed) seems to coincide with the occurrence of the major windburst in December, and for a short period eastward zonal currents reach a maximum speed of 2.6 m s−1. Corresponding equatorial sea surface temperatures are shown in Fig. 18c. The passage of the VAT signal seems to relate to an increase in temperature, especially so on arrival of the signal at the eastern boundary. However, a major warming was not triggered. This was also the case in the observed windburst event of May 1986 (McPhaden et al. 1988) when the oceanic effects of the wind burst were clear but short lived.

Examination of the time series of integrated zonal wind stress anomaly in Fig. 10b reveals that the CGCM has considerable variability on shorter timescales than ENSO. This activity is not obviously related to ENSO: during the period spanning years 14–21, there is little interannual variability but there are several peaks associated with westerly wind anomalies. Although they apparently occur spontaneously, the bursts may influence the ENSO events: for example, in Fig. 13 a westerly wind burst can be seen at 150°E at the end of year 11. The burst was strong enough to reverse the normally easterly winds there, and a fast Kelvin wave response is evident in VAT′ in Fig. 12. This burst may have hastened the end of the year 11 cold event, and aided the warm event onset in year 12, but the impact is not obvious.

6. Predictability

An experiment has been made in which the sensitivity of the model to very small changes in initial conditions is examined. In this way the potential predictability of various features can be assessed by using a “perfect” model with “near-perfect” initial conditions. The integration described above is used as a control run. At eight different times in the control integration ensembles are generated, starting from 1 December of year 10 when the model ENSO is active and following at 3-month intervals. In each member of the ensemble, the ocean initial conditions are identical to the control while the atmosphere initial conditions are perturbed slightly by taking a few atmosphere time steps with sea surface temperatures held constant. In a similar experiment, Stockdale et al. (1994) made several “twin” integrations and found their model to be rather sensitive to small perturbations. Sensitivity was also found by Gent and Tribbia (1993) whose “twin” experiments were initialized by making small changes to the ocean temperature field: they found a rapid decrease in correlation of monthly sea surface temperature anomalies during the first six months. In our experiment, each ensemble consists of three members and every member is integrated for a period of six months.

a. Sea surface temperature

Figure 19 shows SST′ for the Nino-3 region for each member of each ensemble at the end of the 6-month period, the superscript date indicating the month in which the ensemble was initiated. The control run is plotted to give an additional realization for each ensemble and to provide a reference to the phase of the ENSO cycle in which the ensembles were generated. The trajectories of individual members for two ensembles are plotted. The ensemble initiated in December year 10 shows very little spread with all members indicating a movement from warm conditions to cold. In contrast, the ensemble initiated in June year 12 exhibits a marked divergence: in comparison to the warming at this time in the control run, the ensembles give an augmented and a diminished warming and a slight cooling. In general, the ensembles initiated in September and December are more accurate and show less spread than those starting in March and June.

Figure 20 shows the CGCM average SST′ plotted as a function of month-of-year for the Nino-3 region. The model ENSO signal is closely related to the seasonal cycle with the minimum in SST′ amplitude occurring during boreal spring indicating transition between warm/cold conditions. (The 2-yr period chosen for the prediction experiment conforms to this behavior.) This behavior is similar to that observed in the 1970s, when ENSO was strongly related to the seasonal cycle (Balmaseda et al. 1995). The spread of the CGCM ensembles is greater when starting from initial conditions that occur during or shortly after this transition phase in the ENSO cycle. This is also consistent with the prediction results of Balmaseda et al. (1995), who find that during the 1970s, when there is a minimum in spring variance of sea surface temperature, there is also a pronounced spring barrier in predictive skill.

b. Precipitation

The behavior and predictability of rainfall in several selected areas (Table 1) has likewise been investigated. The chaotic nature of the atmosphere in the midlatitude regions would imply that large ensembles are required to identify areas of potential predictability. However, in tropical regions rainfall anomalies are thought to be largely related to the slowly varying boundary conditions. The areas described here lie mostly within the tropical belt and were chosen with regard to observed ENSO teleconnections, and the model ENSO-associated rainfall (Fig. 16). Note that simple rectangular areas are used that generally cover land and ocean. For each area the model climatology is compared to observations (Legates and Willmott 1990) and the model precipitation at the end of each hindcast period is examined with respect to the control to see if any indication of skill is apparent. To reduce the noise in the rainfall data, a 3-month running mean has been applied.

In the central equatorial Pacific (region 1, Fig. 21a) the amplitude of the model seasonal cycle is moderately well simulated and the timing of the December–January–February maximum is good; however, there is no model June maximum. The control indicates that there is more (less) rain for warm (cold) ENSO phase, in agreement with observed behavior. The hindcasts generally show little scatter and are consistent with the control. The rainfall ensemble least well hindcast (October year 12) corresponds to large scatter in SST′ when the model was trying to hindcast warm conditions from the transition phase. Large scatter in SST′ also occurred when hindcasting cold conditions from the transition phase, but in this case (October year 11) all hindcasts predicted the minimum in rainfall well. This presumably is a reflection on the nonlinear relationship of convection to sea surface temperature.

Region 2 contains the two AGCM grid points that are located in the Nordeste area (northeast Brazil). In this region there is evidence that sea surface temperature contains a large amount of the predictive signal and seasonal rainfall predictions have useful skill in practice (Ward and Folland 1991). The CGCM seasonal cycle is good with a wet season in February–March–April although maximum rainfall is 30% less than observed. Wet season precipitation in the control is slightly more (less) following a cold (warm) event; however, the hindcast behavior is not consistent. In practice, variability in this region is more related to Atlantic conditions, although Pacific SST does have an influence, the observed tendency being that warm (cold) ENSO relates to dry (wet) conditions in northeast Brazil. This latter relationship is captured by the CGCM as illustrated in Fig. 16.

In the Australian monsoon area (region 3) the timing of the seasonal cycle is good, but again the peak January–February–March rainfall is underestimated. The ENSO connection is unclear in the model and the hindcasts, on average, show considerable spread.

The CGCM seasonal rainfall cycle is well simulated for the Indian monsoon region (region 4, Fig. 21b). Other climatological features (e.g., summer monsoon 850-hPa winds, Fig. 4c) also are in good agreement with observations. The region is slightly wetter (drier) during cold (warm) ENSO phases, in agreement with observed tendencies. The hindcasts are generally consistent, although (e.g., July year 12) opposite signed anomalies can occur in the hindcasts. As was previously noted, wet season rainfall related to cold SST′ has apparently been more reliably hindcast than that corresponding to warm SST′.

In the Sahel (region 5), the CGCM seasonal cycle has a rainy season that is too early (May–June–July instead of July–August–September), and too weak (less than 50% of the observed maximum). The CGCM variability has no clear ENSO connection: again, in practice the rainfall in the wet season is only weakly related to Pacific conditions.

For Southern Africa (region 6, Fig. 21c) the CGCM seasonal cycle timing is good (December–January–February peak) but there is a large dry bias. The hindcasts have little spread but can differ consistently from the control, suggesting that a sample of 3 is insufficient. Taking the hindcasts into account, there is not a clear ENSO connection in the model.

7. Summary

Results from a 25-yr simulation of climate with a coupled general circulation model (tropical Pacific Ocean and global atmosphere with moderately high resolution) have been presented. The ocean and atmosphere interact freely and completely in the tropical Pacific, and climatological SST is prescribed outside that region. Ocean temperature and salinity relax to climatological values near 30°N and 30°S. With regard to the model climatology, the results are encouraging. There is no apparent climate drift, in contrast to previous versions of the CGCM, so recent major changes to the AGCM component have been beneficial.

The CGCM broadly simulates the observed seasonal cycle quite well. However, several systematic errors are also apparent. The most serious is the warm bias in the east Pacific upper-ocean temperature: this leads to reduced zonal SST and thermocline gradients across the Pacific, weakened equatorial zonal wind stress, and an ITCZ that crosses the equator in the east Pacific during the seasonal cycle. The reason for this failure is not clear, and further experiments to investigate it are needed. A contributing factor is the poor representation of marine stratus in the AGCM: this causes increased east Pacific solar heating that can warm the ocean and thus weaken the winds, which in turn causes further ocean warming in a coupled feedback interaction. The parameterization of ocean mixing may also be incorrect: in OGCM simulations with observed winds (but simple parameterized heat flux) the thermocline also tends to be too deep in the eastern Pacific.

The CGCM produces variability on a wide range of timescales, and is able to simulate realistically several important features of climate variability in the Tropics. On the interannual timescale, there are substantial and irregular coupled oscillations similar to observed ENSO events in the Pacific region. For example, Nino-3 SST′ has an amplitude of about 2°C. The coupled anomalies tend to develop in the west Pacific, then propagate eastward, with VAT′ leading SST′ and TAUX′ closely related to SST′. The dominant model ENSO timescale is about 2 years, however, which is faster than the observed 3–5-yr scale, and the model activity is unrealistically strong south of the equator. A POP analysis of SST, VAT, and TAUX reveals spatial patterns grossly similar to those observed. Rainfall patterns associated with the ENSO variability also have features like those observed. The effect of the systematic errors on the model ENSO is uncertain, but it is likely that the warm east Pacific SST bias makes it easier for interannual anomalies to penetrate from the central to the east Pacific.

The main spatial and temporal relationship between SST′, VAT′, and TAUX′ shows that wind-driven thermocline movement is an important factor in the interannual oscillation mechanism. The eastward propagation and central-west Pacific regeneration suggest that a basic coupled Kelvin mode as in Hirst (1988) is operating. However, episodes of westward propagation show that other surface-layer processes are also important, and the interannual variability is likely to depend on a varying balance of mechanisms, with mixed modes like those described in Neelin (1991). No clear evidence of western-boundary wave reflection was found in the CGCM, so a delayed oscillator mechanism involving reflection (Battisti and Hirst 1988) is unlikely.

The AGCM is capable of generating high-frequency tropical activity such as spontaneous westerly wind bursts. In the CGCM these have an oceanic impact similar to that observed: a strong Kelvin wave response that causes significant but short-lived SST increases along the equator. The model westerly bursts occur independently of the interannual variability, and do not obviously trigger ENSO events.

Several “near-perfect” hindcast experiments were carried out, at a time of strong interannual variability. With regard to Nino-3 SST′, the results suggest that predictability is best for hindcasts starting in the second half of the calendar year (as found in several ENSO prediction schemes in practice) and best when starting near a warm ENSO peak. Although the number of hindcasts is too small for the results to be very reliable, an investigation of rainfall predictability in several regions showed that some realistic features could be reproduced.

REFERENCES

  • Balmaseda, M. A., M. K. Davey, and D. L. T. Anderson, 1995: Decadal and seasonal dependence of ENSO prediction skill. J. Climate,8, 2705–2715.

  • Barnston, A. G., and Coauthors, 1994: Long-lead seasonal forecasts—Where do we stand? Bull. Amer. Meteor. Soc.,75, 2097–2114.

  • Battisti, D. S., and A. C. Hirst, 1988: Interannual variability in the tropical atmosphere-ocean system: Influence of the basic state and ocean geometry. J. Atmos. Sci.,45, 1687–1712.

  • Chang, C.-P., J. E. Erickson, and K. M. Lau, 1979: Northeasterly cold surges and near-equatorial disturbances over the winter MONEX area during December 1974. Part 1: Synoptic aspects. Mon. Wea. Rev.,107, 812–829.

  • Chao, Y, and S. G. H. Philander, 1993: On the structure of the Southern Oscillation. J. Climate,6, 450–469.

  • Cox, M. D., 1984: A primitive equation, 3-dimensional model of the ocean. GFDL Ocean Group Tech Rep. 1, 108 pp. [Available from Geophys. Fluid Dyn. Lab./NOAA, Princeton, NJ 08540.].

  • Cullen, M. J. P., 1993: The unified forecast/climate model. Meteor. Mag.,122, 81–94.

  • Davey, M. K., S. Ineson, and M. A. Balmaseda, 1994: Simulation and hindcasts of tropical Pacific Ocean interannual variability. Tellus,46A, 433–447.

  • Gent, P. R., and J. J. Tribbia, 1993: Simulation and predictability in a coupled TOGA model. J. Climate,6, 1843–1858.

  • Goldenberg, S. D., and J. J. O’Brien, 1981: Time and space variability of tropical Pacific wind stress. Mon. Wea. Rev.,109, 1190–1207.

  • Hasselmann, K., 1988: PIPs and POPs: The reduction of complex dynamical systems using principal interaction and oscillation patterns. J. Geophys. Res.,93(D9), 11 015–11 021.

  • Hastenrath, S., 1991: Climate Dynamics of the Tropics. Kluwer Academic Publishers, 488 pp.

  • Hellerman, S., and M. Rosenstein, 1983: Normal monthly windstress over the World Ocean with error estimates. J. Phys. Oceanogr.,13, 1093–1104.

  • Hirst, A. C., 1988: Slow instabilities in tropical ocean basin–global atmosphere models. J. Atmos. Sci.,45, 830–852.

  • Ineson, S., and C. Gordon, 1989: Parametrization of the upper ocean mixed layer in a tropical ocean GCM. Dynamical Climatology Tech. Note No. 74, 20 pp. [Available from Hadley Centre, Meteorological Office, London, Rd., Bracknell, Berkshire RG12 254, U.K.].

  • Kraus, E. B., and J. S. Turner, 1967: A one-dimensional model of the seasonal thermocline. Part II. Tellus,19, 98–105.

  • Latif, M., A. Sterl, E. Maier–Reimer, and M. M. Junge, 1993a: Climate variability in a coupled GCM. Part I: The tropical Pacific. J. Climate,6, 1–21.

  • ——, ——, ——, and ——, 1993b: Structure and predictability of the El Niño/Southern Oscillation phenomenon in a coupled ocean–atmosphere general circulation model. J. Climate,6, 700–708.

  • Legates, D. R., and C. J. Willmott, 1990: Mean seasonal and spatial variability in gauge-corrected, global precipitation. Int. J. Climatol.,10, 111–127.

  • Luther, D. S., D. E. Harrison, and R. A. Knox, 1983: Zonal winds in the central equatorial Pacific and El Niño. Science,222, 327–330.

  • McCreary, J. P., and D. L. T. Anderson, 1991: An overview of coupled ocean–atmosphere models of El Niño and the Southern Oscillation. J. Geophys. Res.,96, 3125–3150.

  • McPhaden, M. J., H. P. Freitag, S. P. Hayes, B. A. Taft, Z. Chen, and K. Wyrtki, 1988: The response of the equatorial Pacific Ocean to a westerly wind burst in May 1986. J. Geophys. Res.,93(C9), 10 589–10 603.

  • Mechoso, C. R., and Coauthors, 1995: The seasonal cycle over the tropical Pacific in general circulation models. Mon. Wea. Rev.,123, 2825–2838.

  • Moum, J. N., and D. R. Caldwell, 1994: Experiment explores the dynamics of ocean mixing. Eos Trans. Amer., Geophys., Union,75, 489–495.

  • Neelin, J. D., 1991: The slow sea surface temperature mode and the fast wave limit: Analytic theory for tropical interannual oscillations and experiments in a hybrid coupled model. J. Atmos. Sci.,48, 584–606.

  • ——, and Coauthors, 1992: Tropical air–sea interaction in general circulation models. Climate Dyn.,7, 73–104.

  • ——, M. Latif, and F.-F. Jin, 1994: Dynamics of coupled ocean–atmosphere models. Annu. Rev. Fluid. Mech.,26, 617–659.

  • Oberhuber, J. M., 1988: An atlas based on the COADS dataset. Max-Planck-Institut für Meteorologie Report No. 15, 20 pp. [Available from Max-Planck-Institut für Meteorologie, Bundestrasse 55, D-20146 Hamburg, Germany.].

  • Pacanowski, R., and S. G. H. Philander, 1981: Parametrization of vertical mixing in numerical models of the tropical oceans. J. Phys. Oceanogr.,11, 1443–1451.

  • Palmer, T. N., and D. L. T. Anderson, 1994: The prospects for seasonal forecasting—A review paper. Quart. J. Roy. Meteor. Soc.,120, 755–793.

  • Philander, S. G. H., 1990: El Niño, La Niña, and the Southern Oscillation. Academic Press, 293 pp.

  • ——, R. C. Pacanowski, N.-C. Lau, and M. J. Nath, 1992: Simulation of ENSO with a global atmospheric GCM coupled to a high-resolution tropical Pacific Ocean GCM. J. Climate,5, 308–329.

  • Reverdin, G., C. Frankignoul, E. Kestenare, and M. J. McPhaden, 1994: Seasonal variability in the surface currents of the equatorial Pacific. J. Geophys. Res.,99, 20 323–20 344.

  • Robertson, A. W., C.-C. Ma, M. Ghil, and C. R. Mechoso, 1995: Simulation of the tropical Pacific climate with a coupled ocean–atmosphere general circulation model. Part II: Interannual variabilty. J. Climate,8, 1199–1216.

  • Ropelewski, C. F., and M. S. Halpert, 1987: Global and regional scale precipitation patterns associated with the El Nino/Southern Oscillation. Mon. Wea. Rev.,115, 1606–1626.

  • ——, and ——, 1989: Precipitation patterns associated with the high index phase of the Southern Oscillation. J. Climate,2, 268–284.

  • Smith, T. M., and M. Chelliah, 1995: The annual cycle in the tropical Pacific Ocean based on assimilated ocean data from 1983 to 1992. J. Climate,8, 1600–1614.

  • Stockdale, T., D. Anderson, M. Davey, P. Delecluse, A. Kattenberg, Y. Kitamura, M. Latif, and T. Yamagata, 1993: Intercomparison of tropical ocean GCMs. WCRP-79, WMO/TD-No. 545, 90 pp. [Available from WMO, Publication Sales Unit, Case Postale 2300, CH-1211 Geneva 2, Switzerland.].

  • ——, M. Latif, G. Burgers, and J.-O. Wolff, 1994: Some sensitivities of a coupled ocean–atmosphere GCM. Tellus,46A, 367–380.

  • Tett, S., 1995: Simulation of El Niño–Southern Oscillation-like variability in a global AOGCM and its response to CO2 increase. J. Climate,8, 1473–1502.

  • Trenberth, K. E., and D. J. Shea, 1987: On the evolution of the Southern Oscillation. Mon. Wea. Rev.,115, 3078–3096.

  • von Storch, H., G. Bürger, R. Schnur, and J.-S. von Storch, 1995: Principal oscillation patterns: A review. J. Climate,8, 377–400.

  • Ward, M. N., and C. K. Folland, 1991: Prediction of seasonal rainfall in the north Nordeste of Brazil using eigenvectors of sea-surface temperature. Int. J. Climatol.,11, 711–743.

  • Woodruff, S. D., R. J. Slutz, R. L. Jenne, and P. M. Steurer, 1987: A comprehensive ocean–atmosphere data set. Bull. Amer. Meteor. Soc.,68, 1239–1250.

  • Wyrtki, K., and B. Kilonsky, 1984: Mean water and current structure during the Hawaii to Tahiti shuttle experiment. J. Phys. Oceanogr.,14, 242–254.

APPENDIX

POP Analysis

Let F(xm, tn) denote the combined field at M spatial points xm and N time points tn, as represented by K EOFs. Then in terms of POP modes these data can be expressed as
i1520-0493-125-5-721-ea1
where popR and popI are the (real and imaginary) spatial patterns for different phases of the oscillating modes; popD denotes the pattern of a purely decaying mode; and TR, TI, and TD denote corresponding time series for the amplitudes of the patterns. The time series and spatial patterns are normalized; in particular,
i1520-0493-125-5-721-ea2
The constants Γ contain scaling information. Further, real and imaginary pairs of the combined spatial patterns are here chosen to be orthogonal. (Note, however, that patterns from different oscillating pairs are generally not orthogonal.)
Fig. 1.
Fig. 1.

Monthly mean sea surface temperature from COADS climatology for (a) March and (b) September and from the CGCM simulation for (c) March and (d) September. Contour interval is 1°C, temperatures greater than 28°C are shaded.

Citation: Monthly Weather Review 125, 5; 10.1175/1520-0493(1997)125<0721:ICSAPI>2.0.CO;2

Fig. 2.
Fig. 2.

Annual mean equatorial longitude–depth section of ocean temperature from the CGCM. Contour interval is 1°C.

Citation: Monthly Weather Review 125, 5; 10.1175/1520-0493(1997)125<0721:ICSAPI>2.0.CO;2

Fig. 3.
Fig. 3.

Monthly mean wind stress from the Florida State University climatology (1966–85) for (a) March and (b) September and from the CGCM simulation for (c) March and (d) September. Contour interval is 0.025 N m−2.

Citation: Monthly Weather Review 125, 5; 10.1175/1520-0493(1997)125<0721:ICSAPI>2.0.CO;2

Fig. 4.
Fig. 4.

Monthly mean 850-hPa winds from the UKMO operational archive climatology (1986–94) for (a) JJA and (b) DJF and from the CGCM simulation for (c) JJA and (d) DJF. Contour interval is 5 m s−1.

Citation: Monthly Weather Review 125, 5; 10.1175/1520-0493(1997)125<0721:ICSAPI>2.0.CO;2

Fig. 5.
Fig. 5.

Monthly mean CGCM tropical precipitation for (a) March and (b) September. Contour interval is 2.5 mm day−1, regions of precipitation greater than 10 mm day−1 are shaded.

Citation: Monthly Weather Review 125, 5; 10.1175/1520-0493(1997)125<0721:ICSAPI>2.0.CO;2

Fig. 6.
Fig. 6.

Monthly mean CGCM ocean surface currents for (a) March and (b) September. Contours are at 5, 10, 20, and thereafter every 20 cm s−1.

Citation: Monthly Weather Review 125, 5; 10.1175/1520-0493(1997)125<0721:ICSAPI>2.0.CO;2

Fig. 7.
Fig. 7.

Annual mean latitude–depth section through 150°W of zonal current for the CGCM. Contours are at 0, ±5, ±10, and thereafter every 10 cm s−1.

Citation: Monthly Weather Review 125, 5; 10.1175/1520-0493(1997)125<0721:ICSAPI>2.0.CO;2

Fig. 8.
Fig. 8.

Annual mean net heat flux from (a) the climatology of Oberhuber and (b) the CGCM. Contour interval is 25 W m−2.

Citation: Monthly Weather Review 125, 5; 10.1175/1520-0493(1997)125<0721:ICSAPI>2.0.CO;2

Fig. 9.
Fig. 9.

Correlations (×10) of CGCM annual mean sea level pressures with the Darwin values. Correlations are greater than 0.4 in the shaded regions and less than −0.4 in the hatched regions.

Citation: Monthly Weather Review 125, 5; 10.1175/1520-0493(1997)125<0721:ICSAPI>2.0.CO;2

Fig. 10.
Fig. 10.

CGCM time series of (a) sea surface temperature anomalies (°C) averaged over the Nino-3 area (5°N–5°S, 150°–90°W), and (b) zonal wind stress anomalies (N m−2) averaged over the central Pacific (5°N–5°S, 165°E–135°W).

Citation: Monthly Weather Review 125, 5; 10.1175/1520-0493(1997)125<0721:ICSAPI>2.0.CO;2

Fig. 11.
Fig. 11.

Time–longitude section of equatorial model sea surface temperature anomalies for years 8–31 of the CGCM simulation. Contour interval is 1°C, negative values are shaded.

Citation: Monthly Weather Review 125, 5; 10.1175/1520-0493(1997)125<0721:ICSAPI>2.0.CO;2

Fig. 12.
Fig. 12.

Time–longitude section of model vertically averaged temperature anomalies (top 360 m) for years 8–19 of the CGCM simulation at (a) the equator and (b) 6°N. Contour interval is 0.25°C, negative values are shaded.

Citation: Monthly Weather Review 125, 5; 10.1175/1520-0493(1997)125<0721:ICSAPI>2.0.CO;2

Fig. 13.
Fig. 13.

Time–longitude section of model zonal wind stress anomalies for years 8–31 of the CGCM simulation. Contours are at 0, ±0.01, ±0.02, and thereafter every 0.02 N m−2, negative values are shaded.

Citation: Monthly Weather Review 125, 5; 10.1175/1520-0493(1997)125<0721:ICSAPI>2.0.CO;2

Fig. 14.
Fig. 14.

Time series of the real (solid) and imaginary (dashed) components of the ENSO-related CGCM POP mode, obtained using time series for five EOFs of the combined SST′–HC′–TAUX′ fields.

Citation: Monthly Weather Review 125, 5; 10.1175/1520-0493(1997)125<0721:ICSAPI>2.0.CO;2

Fig. 15.
Fig. 15.

Spatial patterns of the ENSO-related CGCM POP mode obtained using five EOFs for combined SST′–HC′–TAUX′ fields. SSTI, HCI, and TAUXI (panels a, c, e) are associated with onset of an ENSO event, while SSTR, HCR, and TAUXR (b, d, f) are associated with the peak ENSO phase (warm). The patterns are components of the normalized combined patterns popR and popI, multiplied by scales ΓR and ΓI. Contour interval is 5 units, negative contours dashed.

Citation: Monthly Weather Review 125, 5; 10.1175/1520-0493(1997)125<0721:ICSAPI>2.0.CO;2

Fig. 16.
Fig. 16.

CGCM precipitation anomaly patterns associated with the model surface temperature anomaly in the Nino-3 area (Fig. 10a). Contours are at 0, ±5, ±10, ±25, and ±50 mm day−1. Values are greater than 5 mm day−1 in shaded regions and less than −5 mm day−1 in hatched regions.

Citation: Monthly Weather Review 125, 5; 10.1175/1520-0493(1997)125<0721:ICSAPI>2.0.CO;2

Fig. 17.
Fig. 17.

CGCM 850-hPa winds in year 17 for selected areas for (a) 1 December, (b) 6 December, (c) 11 December, and (d) 3 November. Contour interval is 5 m s−1.

Citation: Monthly Weather Review 125, 5; 10.1175/1520-0493(1997)125<0721:ICSAPI>2.0.CO;2

Fig. 18.
Fig. 18.

Equatorial time–longitude sections from the CGCM for October year 17 to March year 18 of (a) zonal wind stress (contour interval is 0.05 N m−2); (b) vertically averaged temperature (top 360 m) (contour interval is 0.5°C); and (c) sea surface temperature (contour interval is 1°C, temperatures greater than 28°C are shaded).

Citation: Monthly Weather Review 125, 5; 10.1175/1520-0493(1997)125<0721:ICSAPI>2.0.CO;2

Fig. 19.
Fig. 19.

Time series of CGCM sea surface temperature anomalies (°C) averaged over the Nino-3 area (5°N–5°S, 150°–90°W) for the control simulation (solid line) and for two ensembles (dashed lines). A solid circle indicates the monthly mean temperature anomaly of each ensemble member after 6 months of integration; the label indicates the month when each ensemble was initiated.

Citation: Monthly Weather Review 125, 5; 10.1175/1520-0493(1997)125<0721:ICSAPI>2.0.CO;2

Fig. 20.
Fig. 20.

Average positive and negative model sea surface temperature anomalies (°C) for the Nino-3 area (5°N–5°S, 150°–90°W) for each month-of-year, from the 25-yr CGCM simulation.

Citation: Monthly Weather Review 125, 5; 10.1175/1520-0493(1997)125<0721:ICSAPI>2.0.CO;2

Fig. 21.
Fig. 21.

Time series of CGCM precipitation (solid line) and precipitation for each ensemble member (solid circles) at the end of the hindcast period for (a) the central Pacific (5°N–5°S, 180°–240°E); (b) the India monsoon region (10°–25°N, 70°–95°E); and (c) southeastern Africa (10°–27.5°S, 25°–50°E). The CGCM seasonal cycle of precipition (dashed line) and observations (dotted line) are shown for each area. Units are millimeters per day.

Citation: Monthly Weather Review 125, 5; 10.1175/1520-0493(1997)125<0721:ICSAPI>2.0.CO;2

Table 1.

Selected precipitation areas.

Table 1.
Save