• Ambrizzi, T., B. J. Hoskins, and H.-H. Hsu, 1995: Rossby wave propagation and teleconnection patterns in the Austral winter. J. Atmos. Sci.,52, 3661–3672.

  • Bladé, I., 1997: The influence of midlatitude ocean–atmosphere coupling on the low-frequency variability of a GCM. Part I: No tropical SST forcing. J. Climate,10, 2087–2105.

  • Cox, M. D., 1984: A primitive equation, 3-dimensional model of the ocean. GFDL Ocean Group Tech. Rep. 1, Geophysical Fluid Dynamics Laboratory, Princeton, NJ, 141 pp.

  • Delworth, T. L., 1996: North Atlantic interannual variability in a coupled ocean–atmosphere model. J. Climate,9, 2356–2375.

  • Deser, C., and M. S. Timlin, 1997: Atmosphere–ocean interaction on weekly timescales in the North Atlantic and Pacific. J. Climate,10, 393–408.

  • Dix, M. R., and B. G. Hunt, 1995: Chaotic influences and the problem of deterministic seasonal predictions. Int. J. Climatol.,15, 729–752.

  • Drosdowsky, W., 1993a: An analysis of Australian seasonal rainfall anomalies: 1950–87. I: Spatial patterns. Int. J. Climatol.,13, 1–30.

  • ——, 1993b: An analysis of Australian seasonal rainfall anomalies: 1950–87. II: Temporal variability and teleconnection patterns. Int. J. Climatol.,13, 111–149.

  • ——, and M. Williams, 1991: The Southern Oscillation in the Australian region. Part I: Anomalies at the extremes of the oscillation. J. Climate,4, 619–638.

  • Frankignoul, C., 1985: Sea surface temperature anomalies, planetary waves, and air–sea feedback in the middle latitudes. Rev. Geophys.,23, 357–390.

  • Frederiksen, C. S., and R. C. Balgovind, 1994: The influence of the Indian Ocean/Indonesian SST gradient on the Australian winter rainfall and circulation in an atmospheric GCM. Quart. J. Roy. Meteor. Soc.,120, 923–952.

  • ——, and J. S. Frederiksen, 1996: A theoretical model of Australian northwest cloudband disturbances and Southern Hemisphere storm tracks: The role of SST anomalies. J. Atmos. Sci.,53, 1410–1432.

  • ——, D. P. Rowell, R. C. Balgovind, and C. K. Folland, 1999: Multidecadal simulations of Australian rainfall variability: The role of SSTs. J. Climate,12, 357–379.

  • Frederiksen, J. S., and C. S. Frederiksen, 1993: Monsoon disturbances, intraseasonal oscillations, teleconnection patterns, blocking, and storm tracks of the global atmosphere during January 1979: Linear theory. J. Atmos. Sci.,50, 1349–1372.

  • Gent, P. R., and J. C. McWilliams, 1990: Isopycnal mixing in ocean circulation models. J. Phys. Oceanogr.,20, 150–155.

  • Gordon, H. B., and S. P. O’Farrell, 1997: Transient climate change in the CSIRO coupled model with dynamic sea ice. Mon. Wea. Rev.,125, 875–907.

  • Hirst, A. C., S. P. O’Farrell, and H. B. Gordon, 2000: Comparisons of a coupled ocean–atmosphere model with and without oceanic eddy-induced advection. 1. Ocean spin-up and control integrations. J. Climate,13, 139–163.

  • Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis project. Bull. Amer. Meteor. Soc.,77, 437–471.

  • Levitus, S., 1982: Climatological Atlas of the World Ocean. National Oceanic and Atmospheric Administration, 173 pp and 17 microfiche.

  • Madden, R. A., and P. R. Julian, 1994: Observations of the 40–50 day tropical oscillation—A review. Mon. Wea. Rev.,122, 814–837.

  • Nicholls, N., 1988: El Niño–Southern Oscillation and rainfall variability. J. Climate,1, 418–421.

  • ——, 1989: Sea surface temperatures and Australian winter rainfall. J. Climate,2, 965–973.

  • ——, and K. K. Wong, 1990: Dependence of rainfall variability on mean rainfall, latitude and the Southern Oscillation. J. Climate,3, 163–170.

  • Palmer, T. N., and D. A. Mansfield, 1986: A study of wintertime circulation anomalies during past El Niño events using a high resolution general circulation model. II: Variability of the seasonal mean response. Quart. J. Roy. Meteor. Soc,112, 639–660.

  • Parker, D. E., C. K. Folland, A. Bevan, M. N. Ward, M. Jackson, and K. Maskell, 1995: Marine surface data for analysis of climatic fluctuations on interannual to century timescales. Natural Climate Variability on Decade-to-Century Time Scales, D. G. Martinson, K. Bryan, M. Ghil, M. M. Hall, T. R. Karl, E. S. Sarachik, S. Sorooshian, and L. D. Talley, Eds., National Academy Press, 241–250.

  • Richman, M. B., 1986: Rotation of principal components. J. Climatol.,6, 293–335.

  • Rowell, D. P., 1998: Assessing potential seasonal predictability with an ensemble of multidecadal GCM simulations. J. Climate,11, 109–120.

  • Saji, N. H., B. N. Goswami, P. N. Vinayachandran, and T. Yamagata, 1999: A dipole mode in the tropical Indian Ocean. Nature,401, 360–363.

  • Simmonds, I., 1990: A modelling study of winter circulation and precipitation anomalies associated with Australian region ocean temperatures. Aust. Meteor. Mag.,38, 151–162.

  • ——, and A. Rocha, 1991: The association of Australian winter climate with temperatures to the west. J. Climate,4, 1147–1161.

  • ——, ——, and D. Walland, 1992: Consequences of winter tropical pressure anomalies in the Australian region. Int. J. Climatol.,12, 419–434.

  • Smith, I. N., 1994: Indian Ocean sea-surface temperature patterns and Australian winter rainfall. Int. J. Climatol.,14, 287–305.

  • ——, 1995: A GCM simulation of global climate interannual variability: 1950–1988. J. Climate,8, 709–718.

  • ——, M. Dix, and R. J. Allan, 1997: The effect of greenhouse SSTs on ENSO simulations with an AGCM. J. Climate,10, 342–352.

  • Streten, N. A., 1981: Southern Hemisphere sea surface temperature variability and apparent associations with Australian rainfall. J. Geophys. Res.,86, 485–497.

  • Suppiah, R., and X. Wu, 1998: Surges, cross-equatorial flows and their links with the Australian summer monsoon circulation and rainfall. Aust. Meteor. Mag.,47, 113–130.

  • Tapp, R. G., and S. L. Barrell, 1984: The north-west Australian cloud band. J. Climatol.,4, 411–424.

  • Wallace, J. M., C. Smith, and Q. Jiang, 1990: Spatial patterns of atmosphere–ocean interaction in the northern winter. J. Climate,3, 990–998.

  • Walland, D. J., S. B. Power, and A. C. Hirst, 2000: Decadal climate variability simulated in a coupled general circulation model. Climate Dyn.,16, 201–211.

  • Watterson, I. G., 1998: An analysis of the global water cycle of present and doubled CO2 climates simulated by the CSIRO general circulation model. J. Geophys. Res.,103, 23 113–23 129.

  • ——, 2000: Southern midlatitude zonal wind vacillation and its interaction with the ocean in GCM simulations. J. Climate,13, 562–578.

  • ——, S. P. O’Farrell, and M. R. Dix, 1997: Energy and water transport in climates simulated by a general circulation model that includes dynamic sea ice. J. Geophys. Res.,102, 11 027–11 037.

  • ——, M. R. Dix, and R. Colman, 1999: A comparison of present and doubled CO2 climates and feedbacks simulated by three general circulation models. J. Geophys. Res.,104, 1943–1956.

  • Webster, P. J., A. M. Moore, J. P. Loschnigg, and R. R. Leben, 1999:Coupled ocean–atmosphere dynamics in the Indian Ocean during 1997–98. Nature,401, 356–360.

  • Whetton, P. H., 1990: Relationships between monthly anomalies of Australian region sea-surface temperature and Victorian rainfall. Aust. Meteor. Mag.,38, 31–41.

  • Wright, W. J., 1988: The low latitude influence on winter rainfall in Victoria, southeastern Australia. II: Relationships with the SOI and Australian region circulation. J. Climatol.,8, 547–576.

  • Zheng, X., and C. S. Frederiksen, 1999: Validating interannual variability in an ensemble of AGCM simulations. J. Climate,12, 2386–2396.

  • View in gallery

    Mean simulated winds at 800 hPa in the Australian region in (a) Jan and (b) Jul. Shown are the wind vectors (scale: 10° longitude is 18 m s−1, minimum shown 2 m s−1), and their magnitude, shaded (light 5 m s−1 to 10 m s−1, dark above that). The modeled coastline is shown.

  • View in gallery

    Standard deviation of the 800-hPa monthly mean wind components, zonally averaged across the Australian region in (a) Jan and (b) Jul. Shown are the GCM u (solid), NCEP u (long dash), GCM υ (short dash), and NCEP υ (dots).

  • View in gallery

    Rotated principal components of normalized seasonal rainfall in run C, for Jun–Aug, presented as correlations (×100): (a) RPC1 (25.4%) and (b) RPC2 (23.0%).

  • View in gallery

    Two of the RPCs of 800-hPa wind in Jan: (a) RPC1 (28.5% of variance), denoted monsoon surge;(b) RPC4 (6.5%), northerly. Shown are the wind vectors (scale: 10° longitude is 6 m s−1, minimum shown 0.2 m s−1), and their magnitude, shaded with 0.5 m s−1 gradations, first at 0.5 m s−1 last at 4 m s−1.

  • View in gallery

    As in Fig. 4 but for Jul: (a) RPC4 (9.8% of variance), northwesterly; (b) RPC11 (4.2%), northerly.

  • View in gallery

    Regression lifecycle of a monsoon surge (in run M) (from top to bottom); lags −20, 0, +20 days. Surface pressure is shown by contours (long-dashed for negative values, short-dashed for zero), σ = 0.80 winds (interpolated to a 7° grid) by vectors.

  • View in gallery

    Regression life cycle of a Jul northwesterly event (in run M) (from top to bottom); lags −6, −4, −2, 0, and +2 days. Surface pressure is shown by contours (long-dashed for negative values, short-dashed for zero), σ = 0.34 winds (interpolated to a 7° grid) by vectors.

  • View in gallery

    Percentage of variance of lag 0 monthly anomalies of rainfall related to the 12 wind RPCs in (a) Jan and (b) Jul.

  • View in gallery

    Percentage of variance of lag 0 monthly anomalies of surface temperature related to the 12 wind RPCs in (a) Jan and (b) Jul.

  • View in gallery

    Percentage of variance of lag −1 monthly anomalies of surface temperature related to the 12 wind RPCs in (a) Jan and (b) Jul.

  • View in gallery

    Correlation (×100) of lag 0 monthly anomalies with monsoon surge: (a) rainfall and (b) surface temperature. As in following figures, the contours ±10, ±30, ±50, and ±70 are shown, together with some local peaks.

  • View in gallery

    Correlation (×100) of monthly anomalies with Jan northerly: (a) Jan rainfall and (b) Feb surface temperature.

  • View in gallery

    Correlation (×100) of monthly anomalies with Jul northwesterly: (a) Jul rainfall and (b) Aug surface temperature.

  • View in gallery

    Correlation (×100) of monthly anomalies with Jul northerly: (a) Jul rainfall and (b) Aug surface temperature.

  • View in gallery

    Correlations between indices related to four RPCs: (a) wind with rainfall (lagged), (b) wind with SSTA (lagged), (c) rainfall with SSTA (lagged).

  • View in gallery

    Correlation (×100) of Jun–Aug seasonal mean anomalies of (a) rainfall, and (b) surface temperature with the Jun–Aug mean of the rainfall index associated with the Jul northwesterly wind RPC.

  • View in gallery

    As in Fig. 16 but for the Jul northerly RPC.

  • View in gallery

    Correlation (×100) of Dec–Feb seasonal mean anomalies of (a) rainfall and (b) surface temperature with the Dec–Feb mean of the rainfall index associated with the Jan northerly wind RPC.

  • View in gallery

    Correlation (×100) of Jun–Aug rainfall with Jul northwesterly SST index in (a) May and (b) Sep.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 120 120 2
PDF Downloads 7 7 1

Wind-Induced Rainfall and Surface Temperature Anomalies in the Australian Region

View More View Less
  • 1 CSIRO Atmospheric Research and Cooperative Research Centre for Southern Hemisphere Meteorology, Aspendale, Victoria, Australia
© Get Permissions
Full access

Abstract

An extended coupled atmosphere–ocean model simulation has been analyzed to explore the relationship between Australian rainfall and regional surface temperature anomalies. The interannual variability of seasonal rainfall in the model (which has a weak El Niño) was generally similar to that from a simulation by the atmospheric model with SSTs specified to follow a repeated annual cycle, suggesting that much of the variability is generated internally to the atmosphere. Given the expected role of winds, rotated principal component analyses of both January and July regional 800-hPa wind anomalies from the coupled model were performed. In each season, two wind patterns correlated with realistic rainfall patterns: a monsoon surge pattern brought rainfall to northern Australia in January; a northwesterly flow induced winter rainfall along a band extending to the southeast; and a pattern including northerlies along the east coast brought rainfall to eastern Australia, in both winter and summer. The wind patterns also induced significant surface temperature anomalies over both land and sea. Correlations between the wind-related rainfall indices and seasonal SSTs produced regional SST patterns with much in common to those observed. In particular, the northwesterly rainfall coincided with an eastern Indian Ocean dipole pattern. Aside from the monsoon surge, there is little long-term precursor to these wind anomalies in the model, however, and hence little predictability of the rainfall patterns via the SSTs. Clearly, substantial seasonal rainfall–SST relationships do not necessarily result from forcing of wind anomalies by SSTs, and, excepting El Niño, much of the Australian relationship may, evidently, be unforced.

Corresponding author address: Dr. I. Watterson, CSIRO, Division of Atmospheric Research, 107-121 Station St., Aspendale, Victoria 3195, Australia.

Email: ian.watterson@dar.csiro.au

Abstract

An extended coupled atmosphere–ocean model simulation has been analyzed to explore the relationship between Australian rainfall and regional surface temperature anomalies. The interannual variability of seasonal rainfall in the model (which has a weak El Niño) was generally similar to that from a simulation by the atmospheric model with SSTs specified to follow a repeated annual cycle, suggesting that much of the variability is generated internally to the atmosphere. Given the expected role of winds, rotated principal component analyses of both January and July regional 800-hPa wind anomalies from the coupled model were performed. In each season, two wind patterns correlated with realistic rainfall patterns: a monsoon surge pattern brought rainfall to northern Australia in January; a northwesterly flow induced winter rainfall along a band extending to the southeast; and a pattern including northerlies along the east coast brought rainfall to eastern Australia, in both winter and summer. The wind patterns also induced significant surface temperature anomalies over both land and sea. Correlations between the wind-related rainfall indices and seasonal SSTs produced regional SST patterns with much in common to those observed. In particular, the northwesterly rainfall coincided with an eastern Indian Ocean dipole pattern. Aside from the monsoon surge, there is little long-term precursor to these wind anomalies in the model, however, and hence little predictability of the rainfall patterns via the SSTs. Clearly, substantial seasonal rainfall–SST relationships do not necessarily result from forcing of wind anomalies by SSTs, and, excepting El Niño, much of the Australian relationship may, evidently, be unforced.

Corresponding author address: Dr. I. Watterson, CSIRO, Division of Atmospheric Research, 107-121 Station St., Aspendale, Victoria 3195, Australia.

Email: ian.watterson@dar.csiro.au

1. Introduction

The variability of seasonal rainfall over Australia is of significant climatological and economic interest, and the prospect of forecasting such variability has driven much research over many years. In common with many regions of low mean rainfall, the variability, expressed as a coefficient of variation, is large over much of the Australian interior, and furthermore it is clearly influenced by the equatorial sea surface temperature (SST) anomalies of El Niño (e.g., Nicholls 1988). In addition, much of the continent is at low latitudes, which Nicholls and Wong (1990) relate to enhanced interannual variability.

While much remains to be understood of the causes of seasonal rainfall variability, one might surmise that an important aspect is the dominance of “mean-flow” moisture fluxes in feeding the moisture convergence that supports rainfall at low latitudes (e.g., Watterson 1998). This contrasts with the higher latitudes, where ubiquitous poleward fluxes associated with day-to-day weather or “transient-flow” anomalies are important to the seasonal mean convergence. With the close association of the mean-flow moisture flux with the mean winds at the median level of the water column, around 800 hPa (Watterson 1998), substantial spatial and temporal variations occur in the flux, via the wind variation, and hence in rainfall. These include, of course, the monsoons, as well as the Madden–Julian oscillation (MJO; e.g., Madden and Julian 1994) that contributes much variability on the monthly timescale.

Large zonal mean-flow moisture flux lies just to the north of Australia through much of the year, being easterly in June–August (JJA, winter) and westerly in DJF (summer), and relatively minor southward perturbations of this can produce a substantial change in the availability of moisture for condensation over the land. El Niño is an important driver of such perturbations over eastern Australia (Drosdowsky and Williams 1991), through the associated “Southern Oscillation” atmospheric anomaly, and seasonal predictions of rainfall there, based on ENSO, have been made for some years. This influence has been simulated with some success with general circulation models (GCMs) of only moderate resolution (e.g., Smith et al. 1997; Frederiksen et al. 1999). Following Nicholls (1988), Frederiksen et al. (1999) quantified the associated rainfall pattern using a principal component (PC) analysis of normalized seasonal rainfall, and using a Varimax rotation (e.g., Richman 1986) of the first two PCs. The eastern, “ENSO” pattern, is the first rotated PC (RPC1, except in winter when it is RPC2) from an observational dataset over the period 1948–93. Correlation coefficients (r) of the rain pattern amplitude time series with seasonal SST anomalies (SSTAs) reach (at least) −0.4, over the equatorial eastern and central Pacific.

The SSTAs of El Niño are not the only ones correlated with Australian rainfall. Many authors, including Streten (1981) and Whetton (1990), found correlations for SSTs in the western Pacific, Indian Ocean, and seas closer to the continent. In particular, Nicholls (1989) noted a strong relationship between a “dipole” pattern to the northwest of Australia and winter rainfall over the west, center, and south—the first RPC of normalized rain data. Correlations of the rain pattern amplitude time series with seasonal SSTs in the dipole reach ±0.4. The Australian west/south rainfall itself tends to occur during the eastward passage of “northwest cloudbands” (Tapp and Barrell 1984; Wright 1988), and following others we denote the pattern NWCB. Also, Coral Sea SSTAs correlate at 0.4 with the ENSO rainfall in winter.

The significance of these additional SSTAs to predictability remains unclear, however. While closer to Australia, the magnitude of the anomalies are, at best, only half those of El Niño, although Palmer and Mansfield (1986) suggested that the atmosphere may be more sensitive to perturbations in the warm western Pacific, than to those farther east. Drosdowsky (1993b) found SSTAs of +0.4 K to the north of Australia in April–June in years with increased NWCB rain, and precursors in the cool central Indian Ocean branch of the dipole as far back as February–April. He also noted that the seasonal NWCB rainfall was accompanied by northwesterly low-level wind anomalies, consistent with a tropical source of moisture. However, Smith (1994) found little skill in rainfall forecasts based on prior Indian Ocean dipole amplitudes.

Simmonds (1990) and Simmonds and Rocha (1991) showed that very warm Indian Ocean SSTAs could induce northwesterly flow onto Australia in a GCM, and Simmonds et al. (1992) demonstrated that this circulation change led to increased moisture convergence and rainfall over the land. Frederiksen and Balgovind (1994) produced perturbations to the mean flow with an SST dipole with a warm branch peaking in the far western Pacific, at 0°S, 130°E, which Frederiksen and Frederiksen (1996) found led to altered unstable normal modes of the atmosphere. However, these studies reported only modest responses to a specified Indian Ocean SST dipole of more realistic amplitude.

Frederiksen et al. (1999) found statistically significant, but rather weak, correlations of Australian region SSTs with both the NWCB and ENSO rainfall patterns, in ensembles of simulations from two GCMs with global SSTs prescribed to follow those of the Global Sea-Ice and Sea Surface Temperature dataset (GISST) observational record (Parker et al. 1995) for the period 1949–91. However, they noted that there was considerable variation within each ensemble, consistent with anomalies due to internally generated atmospheric variability. Other studies using runs of this type, denoted “G” here (for GISST), such as Smith (1995) and Dix and Hunt (1995), previously noted this effect. Using an analysis of variance (ANOVA) technique (Rowell 1998; Zheng and Frederiksen 1999), Frederiksen et al. (1999) attributed 40% of the variance of local monthly rainfall anomalies, at most, to SST forcing; in winter the value was typically less than 20% (and as low as 1% over land), in the GCMs. Nevertheless, in 5-month means, the models produced correlations of northeast Australian rainfall with the ENSO SST pattern that approached the 0.55 observed value. The correlations with the Southern Oscillation index (SOI, the Tahiti minus Darwin sea level pressure) were even closer to the 0.61 observed result (possibly because SOI variations are not solely forced by SSTAs).

It is generally accepted that much of the SST variation (aside from El Niño) is driven by surface heat fluxes (Frankignoul 1985). Wallace et al. (1990) showed that circulation anomalies in the Northern Hemisphere (NH) are often well correlated with SST tendencies, while Deser and Timlin (1997) found that SST–wind relationships tend to peak when the SSTs are lagged by 2–3 weeks. Whetton (1990), Drosdowsky (1993b), and others suggested that Australian rainfall and regional surface heat fluxes might be perturbed together by wind anomalies. Therefore, some of the apparent SST–rainfall relationship may be the result of winds inducing both anomalies. If these wind anomalies are, in part, generated internally to the atmosphere and not SST forced, then one may need to use fully coupled atmosphere–ocean models to be able to even approximate the observed relationship. Studies by Delworth (1996) and Bladé (1997) of the Northern Hemisphere using coupled models of only moderate resolution found that simulated winds could reproduce realistic SST patterns. Watterson (2000) showed that an important pattern of the Southern Hemisphere wind variability, the midlatitude zonal wind vacillation, perturbed both Australian rainfall and midlatitude SST anomalies in the Commonwealth Scientific and Industrial Research Organisation (CSIRO) model. It is clearly of interest to assess the rainfall–SST relationships in coupled models, although one must acknowledge that the air–sea interaction in models may be affected by the limited vertical resolution of both the atmospheric and oceanic boundary layers, while rainfall processes may be only crudely represented. An advantage of using a model with modest resolution, however, is that datasets covering many more years than have been observed can be generated, so that statistical uncertainty in the results can be virtually removed.

In this paper, we make use of a 1000-yr simulation of the CSIRO coupled model. While this model produces only a weak (around a third the observed amplitude on the 2–5-yr timescale) El Niño (Hirst et al. 2000;Walland et al. 2000), one might hope that the unforced component, at least, of the rainfall–SST relationships might be simulated. While only monthly means are available from all years, an additional 8-hourly dataset from a related model run is used to explore the temporal structure of the anomalies. A 500-yr simulation in which SSTs are prescribed is also considered, when appropriate. In the following section, the GCM and the simulations are briefly described. The mean and variability of the winds across the Australian region are compared with observations, as are SSTs and rainfall.

The approach taken is to target wind-forced anomalies by establishing RPCs of the monthly wind anomalies, relative to the mean annual cycle. The pressure level chosen is 800 hPa, which should relate to moisture fluxes (Watterson 1998), yet also to the near-surface winds that perturb the surface heat fluxes. The principal component analysis of the winds is described in section 3, with basic results in both January and July cases presented. Time-lag regression and composite analyses are used to assess the evolution of both monthly and daily anomalies. The remaining sections focus on four wind RPCs (two in each season) that induce rainfall patterns that are similar to those observed. The associated SST–rainfall relationships are assessed, and the importance of feedbacks of the SSTs is considered. A discussion on the potential implications of the results to seasonal predictability, and the conclusions follow.

2. The datasets

a. Model simulations

The CSIRO Mark 2 GCM used in this study has been described by Gordon and O’Farrell (1997). The atmospheric component of the model has nine sigma levels in the vertical, and a spectral representation with a rhomboidal wavenumber 21 truncation (R21) in the horizontal. It includes a semi-Lagrangian treatment of water vapor transport, dynamic sea ice, and an enhanced land surface scheme, as well as standard parameterizations of radiation, cloud, precipitation, and the atmospheric boundary layer. The oceanic component of the GCM is based on the Bryan–Cox code (Cox 1984) and includes the scheme of Gent and McWilliams (1990) for the transport induced by subgrid-scale eddies. The model has 21 ocean levels and uses the R21 Gaussian latitudes, simplifying the implementation of interaction with the atmosphere through (ocean) surface heat and water fluxes, and surface stresses, acting on the top 25-m deep model layer. Heat is diffused rapidly into a second layer of depth 25 m, producing an effective mixed layer of at least 50 m. Diffusion into the next 30-m layer is much weaker, while convection acts to exchange heat farther down, particularly in the mid- and high latitudes in winter.

The 1000-yr simulation, described by Hirst et al. (2000) and referred to here as run C, is of a climate with constant solar forcing, CO2 (330 ppmv), and ozone concentrations. It exhibits very little climate drift outside the deep ocean, partly due to the use of flux adjustments to the top ocean layer. The adjustments to the surface heat, stress, and freshwater fluxes follow a repeated seasonal cycle, and so do not directly affect interannual variation. Monthly means of a large number of variables are available for analysis, including winds interpolated 6 hourly to pressures corresponding to the model-level sigma values. Winds from the third such level (807 hPa, denoted 800 hPa for convenience) are considered initially.

The 500-yr simulation with specified SSTs, run S, is a continuation of an integration described in Watterson (2000). As the SSTs follow an annual cycle that is repeated each year, the interannual variability is solely due to internal variability of the atmosphere (possibly modified by variation in the land surface, and sea ice).

The 8-hourly dataset, including sigma level winds, used to provide further detail on the short-term temporal evolution of anomalies, is available only from the 1 × CO2 climate run of Watterson et al. (1997), which has been extended to 100 yr. This simulation uses only a simple mixed layer ocean and hence is denoted run M. As the layer depth is 50 m in the lower latitudes, it should interact with the atmosphere rather similarly to the full ocean GCM on the short timescales considered in this analysis.

b. Climatology

The atmospheric climatology of the 1000-yr simulation (run C) is generally similar to that of the mixed-layer ocean model, which has been considered in some detail by Watterson et al. (1999). Here, we consider briefly the key variables of the analysis, the 800-hPa winds, rainfall, and surface temperatures, comparing the simulations with datasets based on observations. The climatological winds for January and July of the 1000-yr run are shown in Fig. 1, over a domain chosen so as to cover the Australian continent and an extended coastal zone, in order to focus on winds that will directly influence the moisture and energy budget of the land and adjacent seas. The seasonal variation of the winds is well modeled, with midlatitude westerlies encroaching on the southern parts, particularly in winter (July). Low-latitude easterlies, in winter, and monsoon westerlies, in summer, pass the north coast. For comparison (not shown) we consider the climatological winds from the National Centers for Environmental Prediction (NCEP) reanalysis dataset (Kalnay et al. 1996), over the years 1958–97, linearly interpolated between the 850- and 700-hPa levels. The model winds in January have a slight westerly bias, particularly in the monsoon and to the southeast of the domain, but the error vector magnitude averages only 1.5 m s−1. In July, the model midlatitude westerlies are a little weak, the northern easterlies too strong, and the mean error is 1.8 m s−1.

The analysis concerns the interannual variability of the monthly means, in particular, and again the model performs very satisfactorily. The standard deviation (SD) of the zonal, u, and meridional, υ, components is somewhat uniform at each latitude in the Australian region, except for an enhancement over some ocean points, and only the zonal means over the domain are shown in Fig. 2. The comparisons with the NCEP data are good, with the main biases being a slightly greater monsoon variation, and too little variation in July around 25°S, 160°E (possibly due to the weak El Niño). The local GCM SDs are generally within 0.5 m s−1 of the observed. Naturally, the standard deviations of the simulated 8-hourly wind anomalies (not shown) are substantially greater than those of the monthly means, typically twice for the u component, and four times for υ. The enhancement is greatest for υ in the midlatitudes in July, when the SD exceeds 8 m s−1. It is worth noting at this point, that while the term interannual variability is used here, a range of timescales actually contribute to the variations between the individual January or summer means, for instance. In fact, there is relatively little persistence of local monthly anomalies from one year to the next, and one can assume that timescales of several months dominate the monthly results, except for SSTs (as discussed in Watterson 2000). A realistic El Niño would contribute to longer-timescale components, of course.

The GCM rainfall (strictly, precipitation) has been compared (not shown) with the seasonal observations presented by Frederiksen et al. (1999). The basic distribution of climatological rainfall is simulated, though, naturally, some coastal and topographic rainfall is not resolved by the R21 grid. The interannual variation of the seasonal rainfall is generally satisfactory. Over most of the continent this is between 8 and 16 mm month−1 in winter, while summer SD increases, as does the mean, from the south (SD and mean around 8 mm month−1) to the north and east (mean to 125 mm month−1, SD to 65 mm month−1). There is too little variability in far eastern Australia in both seasons, consistent with the small El Niño in the model. Nevertheless, the RPC patterns for normalized rainfall, calculated as in Frederiksen et al. (1999), with the commonly used IMSL software, resemble the observational results in both seasons considered here, except that the eastern pattern in summer is RPC2 (with 23% of the variance). The winter patterns are shown in Fig. 3. The summer RPC1 includes more variance in the northwest, than that in Fig. 3a. It is of interest to note that in the 500-yr simulation, S, with specified SSTs, the means and SDs of Australian rainfall are generally similar to those of the coupled model. However, SD was some 30% lower over the adjacent subtropical seas in winter, but 10% higher in the monsoon region in summer. The corresponding RPCs are also similar to those in run C, the main difference being a slight northwestward shift of the RPC2 pattern in each season (but with similar variance fractions). While Fig. 3 serves to illustrate the basic structure of the NWCB and ENSO patterns, given that there is no El Niño at all in run S, it is evident that the eastern pattern in run C, at least, is not solely due to ENSO, even in seasonal means.

SST variability has been compared with the GISST dataset (years 1961–90 were used). While the interannual SDs of monthly temperature anomalies in the central Pacific are typically 40% of the observed, SSTs closer to Australia are only slightly less variable, typically 0.3°–0.6°C. In the Indian Ocean dipole region in winter, both datasets show values around 0.5°C. Watterson (2000) noted that the model with a mixed layer ocean produced slightly greater variability at most latitudes, and including the dipole region. Of course, run S has zero SST interannual variability. Overall, the accuracy of the coupled model, particularly the winds that will be initially analyzed, appears adequate for the present purpose.

3. Wind pattern analysis

a. Method

The broad intention of the analysis was to find representative wind patterns that may induce rainfall and SST anomalies in the Australian region. The principal component analysis was of the (nonnormalized) 800-hPa wind vectors from run C, over the domain shown in Fig. 1. The anomaly fields for the 1000 January months were analyzed separately from those for July. One can anticipate that these months are also representative of adjacent months within the summer and winter seasons. Both u and υ components were included (the squared nature of the analysis leads to the appropriate vector magnitudes). Area weighting was incorporated, simply by first multiplying the anomalies by the weights, then dividing the output patterns by them, to produce the PCs. The first 12 orthogonal PCs of the variation accounted for typically 85% of the wind vector variance at most grid points in both January and July cases. Following Whetton (1990) and many others, a Varimax rotation (including row or Kaiser normalization) of the 12 components was performed, in each case. The resulting RPCs, while no longer strictly orthogonal, tended to be more localized, which may allow easier interpretation. With the original eigenvalues well determined (given the large sample size), mostly nondegenerate, and with no clear jumps in the sequence (cf. Drosdowsky 1993a), the choice of 12 to rotate and analyze was largely dictated by the desire for a convenient number, representing most of the variance. In any case, results for PCs and RPCs calculated from the two 500-yr halves of the dataset were quite similar. While some RPCs have a large amplitude at the edge of the domain, they relate in a continous way, as will be seen, to the surrounding region. The four RPCs that are the focus of discussion match quite well important observed wind patterns, each with a northerly component and onshore to the tropical Australian coast. It is convenient to refer to the spatial fields representing the RPCs as patterns.

Associated with each PC or RPC pattern, say, Pi for i = 1, . . . , 12, is a 1000-point time series, Ai(t), representing the amplitude of the pattern in each year t. The 12 series in each case are uncorrelated with each other, and the scaling is such that each series has an SD of unity. It is worth recalling that each PC is related to its series through the simple projection operation (using a dot product of vectors, here, and evaluated discretely over the domain)
i1520-0442-14-9-1901-e1
where X(t) are the sequence of monthly mean spatial fields, the integral is over the domain, and αij = ∫ PiPj da. For the nonorthogonal RPCs, the other projections also contribute, usually rather weakly, as
i1520-0442-14-9-1901-e2
where the matrix β is the inverse of the matrix α. Conversely, each RPC (or PC) can be reproduced at each grid point, by the (simple, least squares) regression coefficient between the wind time series and the associated amplitude time series.

A feature of the study is the use of regression between the RPC amplitudes and various global quantities lagged in time; for example, January amplitudes correlated with preceding December gridpoint values (the first of the 1000 January months being ignored, here), giving a lag −1 month result. Naturally, such lag −1 results would likely also be applicable to November anomalies relative to winds in December, and so on. Some indication of precursors, and of succeeding, perhaps predictable, anomalies are provided by this approach. Composites of various quantities for high and low amplitudes, averages over cases A > 1 SD and A < −1 SD, respectively, have also been calculated. After normalizing by the mean amplitude, these are, in most cases, very similar to the corresponding regression coefficients, indicating a linear relationship. An exception is rainfall, where weak nonlinearity is seen, due to the nonnegative nature of rainfall. The discussion will focus on the regression results.

In order to examine the evolution of similar patterns in run S, the combined projection calculation of Eq. (2) has been performed with the X representing the 500 January or July 800-hPa wind fields from that run. The resulting time series appear to be of similar character to those of run C, with SD close to unity, within 5% in nearly all cases (featured values will be given shortly). The corresponding wind patterns obtained by regression are also similar, at least for zero lag. Some differences, for other lags and other quantities, will be noted later.

The addition of the other projections in Eq. (2), relative to the straightforward projection of Eq. (1), is appropriate if the variability is of similar character to that of the monthly C dataset. This is not the case in the 8-hourly analysis, and amplitudes of wind-related patterns have been obtained by simple projection, and used in further regression analyses. A similar approach has been used to construct related rainfall and SST pattern amplitudes. In all these analyses, the fields are anomalies relative to the means over all years, calculated separately for each time step through the annual cycle, in the 8-hourly case.

b. The RPC wind patterns

The RPC explaining easily the most variance of wind over the domain in either month, almost 29%, is the January RPC1 shown in Fig. 4a. The pattern is one that is barely changed by the rotation. It relates to the monsoon wind variability, and, for the sign depicted, represents a strengthening and southward movement of the westerlies. A northerly wind component extends across most of northern Australia. The pattern has much in common with an observed monsoon surge, as described by Suppiah and Wu (1998), including having a circulation around the Australian west coast, and the name appears appropriate to use here (although the timescale of the pattern here is longer than in that study). RPC3 (not shown) describes a weaker surge in the northeast of the domain, with limited impact on Australia.

The January RPC2 (11%) essentially represents a meridional shift of the midlatitude westerlies within the sector. A zonally symmetric version of this zonal wind vacillation within the GCM has been studied by Watterson (2000). Three other January RPCs have a similar character, relating largely to midlatitude variation.

The remaining RPCs represent rather localized, tropical circulations. The most significant in terms of rainfall appears to be RPC4 (6.5%, Fig. 4b), which has meridional flow down the Australian east coast, and will be denoted the January northerly (or N) pattern. The structure in eastern Australia is like that found by Drosdowsky and Williams (1991) to relate to the SOI in summer, although the westerly flow to the north is much stronger here.

No single RPC stands out in the July case, with RPC1 explaining only 12.9% of the overall wind variance. The first three patterns, and several others, are located in the midlatitudes. These appear to relate best to local synoptic-scale systems, and while such systems contribute to observed rainfall, particularly near the coast, we will not focus on these here.

The dominant tropical pattern in July is RPC4 (9.8%), shown in Fig. 5a. It features a northwesterly flow across the northwest coast, almost to the southeast coast, and will be denoted the July northwesterly (NW) pattern. This pattern is similar to that found by Drosdowsky (1993b) in observations to be associated with the NWCB rainfall. Four patterns relate to tropical winds, and two of these feature largely zonal flow. Patterns with a large fraction of the weaker meridional flow variance include RPC12, with northwesterly flow across the north coast, and RPC11, the July northerly (N), shown in Fig. 5b, with flow farther east. This pattern is like the SOI-related wind for winter of Drosdowsky and Williams (1991).

For the four featured patterns, the SDs of the corresponding amplitudes in run S, calculated using projection, are given in Table 1 (other quantities there are considered shortly). They are all virtually unchanged from the unit values for C, indicating a negligible influence of SST variation in the raw amplitude variation of these wind patterns. This does not prove a lack of response of the winds, to SSTAs, however. In contrast to the ANOVA approach applied to specified SST (G type) runs, the unforced variability may not be independent of the SST-forced component, as these SSTAs may themselves be partly wind induced. While we are unable to assess the fraction of wind variance that is SST forced, (in run C or in observations) Table 1 suggests that much of these patterns is unforced. A realistic ENSO would presumably result in an enhanced N variability. It is worth noting that the corresponding standard deviations for PCs calculated from the entire Tropics are typically 10% smaller in S than C, hence the model SST interaction can produce differences.

c. Evolution of the patterns

As noted previously, the RPCs explain most of the 800-hPa wind variance across the domain—up to 95% of both the monsoon variance, and of the midlatitude flow in both January and July. They also relate well to winds some distance outside the domain. However, the 12 amplitude time series for 1 month explain rather little of the variance at most locations in the adjacent months, as calculated using time-lagged correlations. An exception is the January case in the Tropics, where 35% of the wind variance in the central Indian Ocean in December is related to the January wind series. The mid-Pacific values in the lag −1 month reach only 15% in both cases. Some 10% of the mid- and high latitude zonal wind variance in December and February is explained, consistent with the moderate persistence of the zonal vacillation noted by Watterson (2000).

Much of the tropical persistence can be related to Madden–Julian oscillation–like (MJO) propagating anomalies, as has been demonstrated using the 100-yr, 8-hourly dataset. Projecting the monsoon surge pattern on the 0.80 sigma level 8-hourly data produced an amplitude at each of the 9300 January steps, the SD of these being 1.4, not much larger than the monthly value. The regression coefficients for winds and pressure with this amplitude series are shown, for an extended domain, in Fig. 6. Here the factor 2.5 × SD is used, to illustrate a large amplitude case, exceeded in 1% of times, assuming normality. The lag 0 wind pattern is very similar to the monthly version in Fig. 4. Regression coefficients for two lags are also shown; these were calculated relative to all January steps, and data from adjacent months were included. Naturally, the peak anomalies decrease with lag of both signs, as differences between individual surge events will increase over time. One might call the evolution that is so depicted, more attractively in computerized animations, a “regression life cycle,” bearing in mind that it is inevitably focused on localized winds at lag 0. In any case, there is a substantial signal even beyond ±1 month, which clearly propagates eastward. Upper-level wind anomalies (on σ = 0.34) have also been calculated, and these tend to be of opposite direction to those below. The northerlies across Australia occur for several weeks prior to lag 0, but turn to weak southerlies after a week. Overall, the pattern has much in common with the observational analysis of Suppiah and Wu (1998), including even weak precursor flow off Asia. The 8-hourly composites at ±28 days are similar in pattern to those of ±1 month from the monthly analysis. Deferring closer examination of the associated timescales to a subsequent paper, the patterns, at least, resemble an MJO.

Turning to the monthly data from the prescribed SST run (S), as noted previously, the SD of the amplitude series for January obtained by projection (Table 1) is almost unchanged from run C. However, the related lagged regression patterns in run S show less propagation, with northerlies persisting over Australia into the lag +1 month. This apparent influence of a coupled ocean (the SSTAs will be presented shortly) warrants further examination than is possible here.

The January northerly (RPC4, not shown) has weaker persistence and propagation to the east than RPC1. Interestingly, over the extended domain, it is very similar to the quasi-stationary monsoon disturbance instability mode calculated for a realistic January basic state by Frederiksen and Frederiksen (1993), including having a reversed upper-level flow (see their Fig. 5). This suggests that such northerly flows in eastern Australia can be generated by internal instability of the atmosphere.

The net wind variance in both the June and August months related to the July wind amplitudes is less than 10%, except for equatorial values up to 17%. Clearly, none of these wind anomalies has significant precursors in the monthly mean wind, and none persists into the following monthly mean with any amplitude. The 8-hourly analysis shows, however, that the winds have interesting time evolutions. Simple projection of the July northwesterly (RPC4) results in an amplitude with a large SD of 2.7, due to enhancement of the circulation in the midlatitudes, where daily variability is large. The lag 0 result (Fig. 7) is rather similar to typical synoptic patterns of northwest cloudbands (e.g., Tapp and Barrell 1984), with a deep low surface pressure anomaly in the Australian Bight, although the fronts typically seen cannot be resolved by the model. (They may disappear from such a regression result, in any case.) Animations of the lagged regression winds and cloud fraction depict an apparently realistic eastward movement of cloud and winds. Upper-level winds (Fig. 7) have a similar pattern, although shifted a little west to those of the pressure and low-level winds, at each time. The regression lifetime of the pattern extends over several weeks but features dramatic propagation. Weak precursors appear to emerge initially in the far South Atlantic, evolving to the wave train structure seen at lag −6 days in Fig. 7. This structure focuses on Australia, then moves farther eastward. Over northwest Australia, the low-level winds are always close to northwesterly during the life cycle, in contrast to midlatitude points. Taking time means of the 8-hourly lagged patterns enhances the subtropical flow, hence supporting this depiction of synoptic flows that contribute to monthly anomalies. We defer consideration of other fields.

The regression life cycle for the July northerly (RPC11, not shown) has much in common with the northwesterly case, but with the peak winds shifted eastward. Evidently, such wave trains sweeping across the region strongly contribute to the various July monthly patterns, depending on the paths. Ambrizzi et al. (1995) show similar wave trains in observations of the region. Naturally, some individual wave train events will overlap 2 months, contributing to a somewhat artificial persistence in monthly means, but this is a rather small effect in these cases. As a component of the eastward-propagating weather patterns in the westerlies, such wave trains are, presumably, ultimately linked to internally generated, baroclinic instability (e.g., Frederiksen and Frederiksen 1996) in the higher latitudes.

There is also some contribution to the July subtropical flow patterns by tropical anomalies. For the northwesterly case, in the lag 0 monthly regression there is a band of low-level westerly anomalies along 5°N that peaks at 85°E with similar amplitude to those in the Australian domain (Fig. 5). There are also smaller wave trains extending across the Pacific with barotropic structure, and greater amplitude at 340 hPa. However, there are only weak precursors, scattered around the hemisphere, in the lag −1 month regression. These anomalies appear to be of less importance than the eastward-propagating wave trains, but the similarity of the Pacific anomalies, in particular, with the slow, unstable barotropic modes evaluated by Frederiksen and Frederiksen (1996), supports their suggestion that instability may contribute directly to the northwesterly pattern. A realistic tropical contribution from ENSO would presumably enhance some monthly wind patterns, in particular the July northerly.

4. Relationships to rainfall and surface temperature

Before focusing on the influence of the four featured RPC patterns in run C, it is worth considering the net variance of monthly rainfall and surface temperature statistically related to the 12 wind patterns in each case. The net related variance in rainfall in both (lag 0) months, over the domain grid points, is shown in Fig. 8. Up to 80% of the monsoon rain variance is related, while over 70% of Southern Ocean variance is related in both January and July. In contrast, over some ocean points almost none is related. At most, half of the July land rainfall variance is related. Presumably, much of the rainfall is related to synoptic systems, but not necessarily to monthly mean anomalies, while some may not relate consistently with the 800-hPa wind at all. One might expect stronger correlations for seasonal means, as short-term fluctuations are averaged out. As in the case of the winds, the net related lagged variance (not shown) is much reduced. Winds in January are related to 30% of the rainfall variance in December and February around 10°N, but otherwise less than 10%, and likewise for winter months.

Land surface temperature monthly means relate well to the winds, particularly in the north where the net variance related to the 12 wind RPCs reaches 70% in both January and July, as seen in Fig. 9. Temperatures in adjacent seas are moderately related to winds in the same month, particularly in January and in the equatorial band. The 1-month lagged correlations (not shown) are larger over most ocean points, with the net related variance reaching 64% in the northwest in February, and 41% there in August. It is the surface heat flux (the net of the evaporative flux, sensible heat, shortwave, and longwave radiative fluxes) that is strongly influenced by the winds, particularly in the tropical ocean where the fractions reach 80%. With the slow response time of the ocean mixed layer to the fluxes, the SSTAs will typically peak around the end of a month of anomalous flux. If the forcing stops, there will then be only a slow damping of the SSTA in the model (particularly in winter, when the effective mixed layer depth is larger). It is evident, then, that the values in Fig. 9 underrepresent the importance of the wind. Over land the response is quicker, and the persistence of anomalies smaller, so that typically 10% of variance is related to the wind patterns in the following month.

The lag −1 results are also of interest, particularly over ocean, and are shown in Fig. 10. Over both land and ocean the net variance in June related to winds in July reaches 4% only at a few, mostly tropical points. It peaks in the mid-Pacific, where it is similar to the July result, indicative of a weak El Niño influence. January values are little better, except for some equatorial regions. Hence, there can be no substantial SST precursors to any of the monthly RPC wind patterns, particularly in July. Ocean surface current anomalies (at lag 0) also relate well to the winds, as anticipated due to the importance of Ekman drift. Both current vector components give fractions over the domain typically 65% in January, and 55% in July.

It is worth noting that midlatitude surface pressure anomalies relate almost as well to the 12 800-hPa wind RPC amplitudes, as do the winds themselves. There is less pressure variance explained in the Tropics, however. Likewise, an initial RPC analysis performed with the pressure field alone also resulted in less rainfall variance explained in the Tropics, although the related land surface temperature variance was similar. There is evidently an advantage in using 800-hPa winds rather than pressure in the component analysis, here.

a. Monsoon surge

Correlations between the January RPC1 wind amplitudes and the rainfall and surface temperature anomalies in the same month (of all 1000 yr) are shown in Fig. 11. The surge brings heavier rainfall and cools the land, particularly in the northwest. SSTs are cooler to the north, due to the enhanced evaporation following the strengthened westerlies. Lag −1 correlations (not shown) of rainfall peak at 0.4 over the ocean to the northwest, and 0.3 over adjacent land. Monthly SST to the northwest correlates at 0.3, the positive values persisting into the lag 0 month, as shown. There are also negative correlations in the equatorial Indian and Pacific Oceans, and positive correlations around 15°N, 130°E. These lag −1 anomalies are partly associated with precursor winds to the west, but there also appears to be a weak ENSO pattern projecting onto the broadscale MJO-like precursor to the monsoon surge. At lag +1 month, the cooler temperatures persist, particularly the SSTAs, while rainfall is negatively correlated in the northern land under southerlies. The evolution is similar in the 8-hourly analysis over lags ±28 days. However, in the S run monthly analysis, the reduced wind propagation results in northern rainfall persisting into February. This may cause the slightly larger SD there in S, noted in section 2 and Table 1.

b. Summer northerly

Rainfall associated with the January northerly wind is again substantial, as seen in Fig. 12a. The rainfall pattern is similar to both the observed eastern (ENSO) pattern (Frederiksen et al. 1999) and the C rainfall RPC2, although the peak is a little too northward, and the southwest is drier. Land temperature anomalies correspond to the rain, and persist into February (Fig. 12b). SSTs are only weakly affected, and there is little December precursor in temperature (not shown), with correlations reaching −0.14 in the mid-Pacific.

c. Winter northwesterly

The rainfall correlations with July RPC4, Fig. 13a, are similar in structure to the NWCB rainfall pattern in observations and Fig. 3a, although there is no peak in southern Australia, here. There is little anomaly in either lag ±1 month regressions. At lag 0, land temperature anomalies are positive due to the warmer air, except in the southwest. Correlations for the specific humidity at 800 hPa (not shown) reach 0.6 over central Australia, consistent with the influx of moister, subtropical air. The SSTA pattern associated with the wind is a dipole, which is stronger in August (Fig. 13b). Surface heat fluxes (not shown) are consistent with the development of the SST pattern, given the effective mixed layer depth of 50 m there in the model. As originally noted by Drosdowsky (1993b), when added to 800-hPa mean wind (Fig. 1b) the wind anomaly (Fig. 5a) augments the wind speed in the Timor Sea, but decreases it farther southwest. Given coherence of the lower-tropospheric winds, and the influence of these on the surface flux, the dipole structure is produced. There is, however, little SST precursor of the NW winds in the monthly analysis–lagged regression coefficients of SSTAs with the wind index of 0.05 K at most.

The evolution of the NW-related winds in run S is very similar to that in C, as are those of rainfall and temperature over land. However, SST anomalies are absent in S, and the C and S lag 1 rain rates over the dipole (in C) are clearly different, by around 0.2 mm day−1 over 0.2 K anomalies.

The 8-hourly analysis shows a similar pattern of rainfall at lag 0 as that from the monthly analysis from C. Ocean mixed layer temperature anomalies develop during the passage of the wave train, around a week. The dipole pattern remaining in the northwest afterward closely resembles that of the lag 1 month result from run C. Elsewhere, the large variation in wind direction during the passage limits the net SST change. These similarities provide further support for the contention that both the rainfall and SST dipole monthly anomalies in the model are largely forced by northwesterly wind anomalies, as the winds need not be fixed in location during the whole month. We will return to this point in section 6.

d. Winter northerly

The July northerly winds also relate to monthly rainfall, Fig. 14a, resembling (although weaker than) the eastern seasonal rainfall pattern of Fig. 3b. Land temperatures are reduced, particularly in the west under southerlies. The reduction persists into the lag 1 month, in the east. SSTs rise significantly to the northeast, where the wind anomaly tends to counter the mean winds (Fig. 1b). The persistence of the rainfall anomaly is again confined to this ocean region. Similar results were obtained from the S run, except over the oceans. There are only weak mid-Pacific SST precursors to the wind in run C.

5. Rainfall–SST relationships

We have seen that some of the 800-hPa wind RPCs of both seasons correlate well with regional rainfall during the life of the wind anomaly and with SSTAs after it. Perturbations of moisture advection and surface heat fluxes induced by the winds appear to provide the dominant physical mechanisms producing these rainfall and SST anomaly patterns. It is of interest to consider further how the amplitudes of such patterns of rainfall and SST directly relate to those of the winds that appear to induce them. Since not all the rainfall is related to such winds, it is necessary to consider also how the rainfall and SST patterns relate to each other. The projection operation, Eq. (1), will be used to calculate an amplitude or index of each wind-related pattern, for both the lag 0 month, and also adjacent months. One should be mindful that such patterns are not orthogonal to each other, so that, for instance, a rainfall index nominally associated with one wind RPC may vary, partly, due to other RPCs.

For rainfall, the January or July lag 0 regression coefficients (with each RPC wind amplitude series) for all Australian land grid points are used as the pattern P in the projection. The resulting indices in January and July (12 cases in each) have an interannual SD always larger than one (and hence the RPC SDs), due to variation not related to the winds. For the featured patterns, the values given in Table 1 range only between 1.2 and 1.7, suggesting that these are rather well related to the RPCs used to define them. Correlating each rain index series at various lags with the associated RPC amplitude series produced the results shown in Fig. 15a. As anticipated, the r values for lag 0 are moderately high. At other lags they are negligible, although some monsoon rainfall anomaly prior to the peak wind surge is indicated by the lag −1 value of 0.27.

For SST, the difference in regression coefficients, lag 1 minus lag −1, is used in order to focus on changes induced by the lag 0 winds. In fact, only for the MJO-related patterns is lag −1 important in this difference. Given some persistence in the wind, this SST change would be greater than a single-month wind anomaly of unit amplitude is likely to produce. Nevertheless, the SST indices also have SDs of at least unity, as shown for the featured cases in Table 1. Watterson (2000) suggests that, to first order, a Markov model for such an SST index may be applicable, in which case the SD could be quite large, given weak damping of the SSTA. It is of interest to consider the autocorrelation for these indices, by correlating the lag 0 series with the lagged results for each case. For the July cases, lag ±1 values are 0.8 to 0.9, and r = 0.5 is reached only at lag ±3. For January, the timescales are typically shorter, except for the midlatitude SST perturbations. The contrast between seasons was also seen in observations by Drosdowsky (1993b), and is consistent with variations in mixed layer depths (Levitus 1982). The propagating nature of the monthly tropical wind patterns, here, will tend to produce wind-induced reversals of surface flux, in summer, reducing persistence. In any case, the relationships between wind and SST indices remain strong, for three of the four featured cases (Fig. 15b), persisting for several months. The January northerly produces little SSTA (Fig. 12b), so there is a weak relationship here. In each case, the lag 1 value is the largest. Except for the monsoon case, there is little precursor.

The correlations between the rainfall indices in the lag 0 month and lagged SST indices are shown in Fig. 15c. Again, in the monsoon case and the July NW and N cases, the rainfall best relates to the positive lagged SSTs. These three cases are the most consistently and strongly related of the 24 cases. In July, the weaker RPC12 produces related rainfall and SST anomalies in the Tropics. In January, RPC8, a wind surge similar to the July northwesterly, actually produces an SST dipole reversed in sign due to the seasonally differing mean winds. The January midlatitude zonal wind (RPC2) cools the surface at 40°S and lowers rain in southeast Australia, as noted by Watterson (2000).

The featured July cases have rainfall patterns resembling the first two observed winter RPCs, NWCB, and ENSO of Nicholls (1989) and Frederiksen et al. (1999), while the January northerly result is also like the observed summer ENSO pattern. It is worthwhile, then, presenting correlations in the same way as these authors, but using the wind-related rain indices, initially. Taking seasonal averages of these monthly indices, over June–August or December–February, we then correlate with averages of both rainfall and surface temperatures at each grid point, for the same season. The results are shown, for an extended domain, in Figs. 16, 17, and 18.

The NW rainfall index produces seasonal rainfall correlations greater than those of the original monthly result (Fig. 13a), but with a similar pattern. The magnitudes (Fig. 16a) are now similar to both the observed NWCB and modeled rain RPC1 (Fig. 3a) seasonal results. A remaining difference is the lack of a peak in the south. While the model may not resolve the rainfall there, particularly if it relates to fronts in reality, it is evident that a wind criterion based only on northwesterlies is inadequate, here. Nevertheless, the seasonal index correlates well with the rain RPC1 amplitude, with r = 0.89. The SST correlations for the former index (Fig. 16b) are also rather similar to the observed, with an east Indian Ocean dipole, and positive values to the north of Australia. Land temperatures are also perturbed. Correlating the seasonal NW rain index with SSTs in past months produces a very weak dipole in May, and only correlations of 0.1 in the Timor Sea in April. Thus the negative central Indian Ocean SST precursors to NWCB rainfall found by Drosdowsky (1993b) are not present. The SST pattern for the model rain RPC1 amplitude is very similar to Fig. 16b, but with a slight shift to the southwest, consistent with winds bringing moisture a little farther south.

The northerly wind rainfall pattern (Fig. 17a) for winter is, again, an intensification of the monthly wind result of Fig. 14a. The SST pattern consists of a moderate peak over the Coral Sea, in the location of the wind-induced warming. Other anomalies occur in bands, rather similar to the rainfall bands. Again these have much in common with the observed values for eastern (ENSO) rainfall in Frederiksen et al. (1999, their Fig. 11). Notably absent, though, is the central and eastern Pacific anomaly associated with El Niño itself, presumably due to the weakness of the El Niño in the GCM, and perhaps also to an incorrect teleconnection to Australia. In both winter cases, regression between each rainfall index and the 800-hPa winds produces patterns very similar to the original RPCs, confirming the role of these wind patterns in generating the rainfall patterns. The model rain RPC2 amplitude correlates at 0.87 with the wind-related rainfall index. Its SST pattern is similar to Fig. 17b, but with reduced Indian Ocean bands, which are evidently associated with the negative western rainfall in Fig. 17a.

The northerly wind rainfall pattern for summer has correlations peaking at 0.83 (Fig. 18a), and decreasing to zero in the southwest, again as in the observed eastern rainfall result. The SST pattern is weak in the Coral Sea, consistent with Fig. 12b, but is surprisingly like that observed (Frederiksen et al. 1999, their Fig. 13) even along the equator; negative in the Indian Ocean, positive around 140°E, and turning negative to a peak of −0.3 at 174°E. Much of the model’s pattern appears to be related to the monsoon surge, particularly given the large correlations in the Timor Sea, relative to those in the mid-Pacific. This suggests that there is some projection of that surge rain as it moves eastward onto the eastern pattern. Indeed, the rainfall pattern here is somewhat shifted to the north, and the southwest negative values in Fig. 12 appear offset by some monsoon component. Regression between the rainfall index and the 800-hPa winds supports this, as the winds appear as a merging of the two patterns shown in Fig. 4. The equatorial flows seen at lag 0 in Fig. 6 are also weakly present. Presumably these contribute to the forcing of the SSTAs. There is evidently some forcing by the SSTs, also, and the weakness of the central Pacific correlation is consistent with the halved amplitude of the SST interannual variation in January, at 174°E, compared to the observed. While it is unclear how realistic the MJO-like patterns and their influences are, it is conceivable that some of the observed eastern rainfall–SST pattern is due to such variability. In any case, the (eastern) rain RPC2 amplitudes are closely correlated with the N rain index, with r = 0.96.

Turning to run S, briefly, we consider the SDs of the rainfall projection time series, obtained using the same C Australian rainfall patterns. In fact, there is typically 10% difference between C and S, at most, in all of these. For the featured patterns (Table 1), the inclusion of SST variation in C apparently decreases the rainfall SD by 5% in the monsoon case, less in the others. The similarities are consistent with those in the rain RPCs, mentioned in section 2b.

6. Discussion

We have seen that in three important cases, the spatial pattern of correlations for seasonal means of rainfall and SST, with Australian rainfall indices derived using the 800-hPa wind principal component analysis, are closely related to direct correlations of the winds with rainfall anomalies and with SSTAs lagged by 1 month. The similarity of these results with the correlations of observed anomalies with ENSO rainfall in both summer and winter, and with NWCB rainfall in winter, provides strong, although circumstantial, evidence that low-level winds induce a large part of the observed relationship between regional rainfall and SSTs in those cases, and potentially in other seasons not considered here. The lag between the peak in wind-induced SSTAs and that of wind-induced rainfall is well masked by taking seasonal means, so that the anomalies appear coincident. The generally small precursors in the SST to the wind anomalies, and the unchanged amplitude of the wind anomalies in the S run with zero SSTAs, suggests that they are largely generated internally to the atmosphere in this model. Only in the monsoon surge case, which appears related to an MJO-like pattern, is there clearly a significant influence of SSTs.

It is of interest to compare the present rainfall–SST relationship (Figs. 16–18), and the similar results for rain RPCs, with the results for such RPCs presented by Frederiksen et al. (1999) for the Hadley Centre of the British Meteorological Office and the Australian Bureau of Meteorology Research Centre (BMRC) GCMs. Overall, the Australian rainfall correlations from those G runs were very similar to those for the present C run, and to the observations. However, the related SST correlations were mostly smaller for G than either C or observations, an exception being the small central and eastern tropical Pacific C values, particularly in Fig. 17b. (Indeed, the G results were shown with a contour interval halved over that used in the observed case.) The SST correlation pattern for winter NWCB was a broad positive region north and west of Australia, reaching a peak around 0.3 in the BMRC result, even less in the Hadley result. Likewise, in the winter ENSO case, there was a broad weak positive region to the north and east, but reaching only around 0.2 in the Coral Sea, in both models. The summer ENSO case from BMRC is actually quite similar to Fig. 18b, except for the eastern Pacific, and for smaller values to the northwest of Australia. The Hadley result was very weak everywhere (although it was quite realistic in September–November). For the cases considered, and over the Australian region, the present C results appear rather more like the observed than either G model.

In fact, the component of the rainfall–SST relationship that is induced by winds generated internally to the atmosphere, which appears to dominate the relationship in the present case, particularly in winter, will be entirely absent in any specified SST run. These internally generated wind anomalies and the wind-induced rainfall patterns can be still simulated in a G or S run, but the SSTAs cannot. This point was apparently not considered by Frederiksen et al. (1999) (and, no doubt, others), who suggested that the apparent deficiencies were partly due to the models having too much internally generated variability of rainfall, which might dilute the forced signal. This is apparently the case for the central east of Australia in these models, as the relationship of rainfall there with the SOI was weak, and this could reduce the correlations of the ENSO rainfall pattern with central Pacific SSTs. The deficiencies in the northwest SST dipole–NWCB relationship were not explained. Simmonds and Rocha (1991) and Frederiksen and Balgovind (1994) suggested that their atmospheric GCMs were insufficiently sensitive to a prescribed northwest dipole. The present results provide an alternative explanation for some of these limitations. The G models may be deficient simply because their oceans cannot respond to surface flux perturbations.

What, then, is the role of SSTAs in producing Australian rainfall anomalies? Clearly, El Niño influences wind patterns over Australia, and hence forces part (the longer period component) of the eastern seasonal rainfall variation. It is worth noting that significant correlations between seasonal SOI and Australian rainfall indices do not necessarily indicate such an influence. The correlations between the model SOI for the present rainfall indices are 0.34 for the monsoon surge, 0.30 for January N, 0.29 for July NW, and 0.14 for July N, all largely due to the Darwin pressure anomalies. The winds forced by El Niño will force the local SSTAs also, and this should also be present in the rainfall–SST relationship of G models. The present results suggest that the (weak) Coral Sea anomaly associated with ENSO rainfall in winter in the Frederiksen et al. (1999) models, may be due to this effect. Recent studies by Webster et al. (1999) and Saji et al. (1999) suggest that a coupled mode of variability may exist in the tropical Indian Ocean, and this may have some influence on Australia. In the C run, at least, MJO-like tropical variability in the Indian Ocean, and extending to the Pacific on a monthly to seasonal timescale, appears to be substantially modified by evolving SSTAs, and also influences Australian rainfall. Further analysis of these topics, using a coupled model generating a more realistic El Niño, is warranted.

Subtropical and midlatitude SSTAs will no doubt have some influence on the atmosphere, but establishing cause and effect in an evolving coupled system is not simple, and the result may depend on the case in question. For the winter northwesterly and the associated NWCB-like rainfall, the similarities between runs C and S (Table 1) suggest that these are largely internally generated. However, there were clear differences in rainfall over the persisting SST dipole. The effect has also been demonstrated by use of multiple linear regression. Regressing local rainfall anomalies in run C with both the NW wind amplitude and the related SST index, all quantities in July, increases the net explained rainfall variance over much of the ocean above that explained by simple regression with the wind. The regression coefficients for the SST index form a dipole pattern (not shown), with locally warmer SSTs giving more rain. However, over land points these coefficients are always small, indicating little direct influence of the dipole (which, here, is mostly forced by winds in prior months that are almost unrelated to the July NW index). A similar result holds for the Coral Sea in the July N case.

The weakly positive SST correlations to the northwest and north seen in the BMRC G result for the NWCB rain pattern, and in the other prescribed SST experiments, suggest that there may be some influence. Simply correlating rainfall, wind, or surface pressure, with the SST dipole index produces the familiar patterns from C, but does not establish cause. However, correlating gridpoint pressures in August with the July SST index does produce a signal that is distinct from the weak residual NW pattern, as the SSTs persist. Pressures tend to be lower over warmer SSTs, and higher over cooler SSTs. This will tend to weakly shift the lower atmosphere wind anomaly, rather than amplify it, however. Focusing on June–August mean rainfall, correlating with the SST index in May, July, and September, in turn, produces contrasts that may also indicate an SST influence. Values for 2 months are given in Fig. 19. The September result is similar to that for July, which appears to indicate a persistence of the dipole that is forced by NW winds during winter. However, the May SST index relates only weakly to winter rainfall over land, although the SSTAs do persist, as is indicated by the ocean rainfall. A further such regression analysis with an SST index formed from anomalies around 0°N, 130°E (following Frederiksen and Balgovind 1994) indicates that warmer temperatures to the north do precede increased winter rain in central Australia, as well as northwesterlies at 800 hPa. However, the seasonal correlations of both these are barely 0.2, and the contribution of the apparent SST forcing to the pattern in Fig. 16 would be minor. Data on the midlatitude storm track, which Frederiksen and Balgovind (1994) suggest might be perturbed, are unavailable, but the enhancement of southern Australian rainfall that might follow is not evident, here, or in the model’s rain RPC1-SST result. Qualitatively similar results regarding SST precursors are seen for the July N case.

The small correlation of May SSTs with winter rainfall, seen in Fig. 19, casts doubt on predictability based on a dipole pattern amplitude. One should not be too discouraged on the existence of useful predictability, in general, though. The approach of this study, beginning with winds, will inevitably focus on wind-induced, and largely lagging, SSTAs. Further, in addition to predictability associated with ENSO and similar long-term coupled anomalies, there is clearly predictability on the monthly timescale that is associated with MJO-like anomalies in the model, at least. Even the winter wave trains that relate to much of the winter rainfall provide some short-term predictability. As an example, the σ = 0.34 patterns from each day of the regression life cycle for the NW case (Fig. 7) have been projected on all July 8-hourly fields over the hemisphere. The resulting indices for negative lags can be used as predictors of the lag 0 pattern, with some success. The correlation skill of a 6-day forecast of lag 0, based on the lag −6 day pattern, is 0.44. No doubt, weather forecast systems properly assimilating the complete pattern can do better. In any case, it is possible that this GCM atmosphere, as with others perhaps, is insufficiently sensitive to subtropical SST anomalies. In addition, the limited vertical resolution in the upper ocean will have an influence. The SST-forced component of the rainfall variability, including that due to feedback from initially wind-induced SSTAs, may be too small.

7. Conclusions

A 1000-yr simulation using a coupled atmosphere–ocean GCM has been analyzed, to explore the relationship between Australian rainfall and SST anomalies that has been found in observations. Comparison of the interannual variability of seasonal rainfall in the GCM with that from a simulation by the atmospheric GCM with specified SSTs, showed rather little difference, supporting the finding of Frederiksen et al. (1999) that much of the variability (not related to El Niño) is generated internally to the atmosphere. First, rotated principal component analyses of both January and July 800-hPa wind anomalies over the Australian region from the coupled model were performed. Four of these wind patterns were found to be strongly related, using linear regression, to realistic rainfall patterns: a monsoon surge pattern brought rainfall to northern Australia in January; a northwesterly flow induced winter rainfall over the south and west, approximating a northwest cloudband pattern; and northerlies along the east coast brought eastern rainfall, in the pattern similar to that associated with observed ENSO, in both winter and summer.

In the coupled model run, these wind patterns also induced significant SST anomalies in the Australian region. Regression between seasonal SSTAs and the wind-related rainfall indices produced patterns with much in common to those observed (an exception being the central Pacific, where the weakness of the model’s El Niño was evident). In particular, the NWCB-like winter rainfall pattern was related to an eastern Indian Ocean SST dipole, and the winter ENSO-like rainfall related to a banded pattern, peaking in the Coral Sea. The lag in the peak SSTA relative to the rainfall in monthly means, however, appears to limit the predictability of these rainfall patterns from SSTAs, in the model at least. It is not clear whether the GCM atmosphere responds realistically to such SSTAs. Aside from the monsoon surge wind pattern, which appears related to the MJO, there is little long-term precursor to the Australian regional wind anomalies in the model. The winter wind patterns are strongly related to barotropic wave trains sweeping around the hemisphere, but not to SST precursors.

The present study shows that substantial seasonal rainfall–SST relationships do not necessarily result from the forcing of wind anomalies by SSTs. One implication of this is that atmospheric models in which realistic SSTs are specified may not be able to simulate the full statistical relationship, even if the atmosphere responds realistically. The implications of the unforced component of rainfall variability for seasonal predictability need to be further assessed in simulations with a more realistic El Niño, and naturally, in observations.

Acknowledgments

My thanks go to members of the climate modeling group of CSIRO Atmospheric Research, and in particular Tony Hirst, who contributed to the development of the coupled GCM and performed the simulation analyzed here. Particularly helpful discussions with Ian Smith and Peter Whetton are gratefully acknowledged. Mark Collier and David Jones provided valuable assistance in the initial stages of the principal component analysis. Valuable comments were received from Ian Simmonds and an anonymous reviewer. This work contributes to the CSIRO Climate Change Research Program and is, in part, funded through Australia’s National Greenhouse Research Program. Information on the IMSL software is available from the National Institute of Standards and Technology internet address /http://math.nist.gov, package IMSLS.

REFERENCES

  • Ambrizzi, T., B. J. Hoskins, and H.-H. Hsu, 1995: Rossby wave propagation and teleconnection patterns in the Austral winter. J. Atmos. Sci.,52, 3661–3672.

  • Bladé, I., 1997: The influence of midlatitude ocean–atmosphere coupling on the low-frequency variability of a GCM. Part I: No tropical SST forcing. J. Climate,10, 2087–2105.

  • Cox, M. D., 1984: A primitive equation, 3-dimensional model of the ocean. GFDL Ocean Group Tech. Rep. 1, Geophysical Fluid Dynamics Laboratory, Princeton, NJ, 141 pp.

  • Delworth, T. L., 1996: North Atlantic interannual variability in a coupled ocean–atmosphere model. J. Climate,9, 2356–2375.

  • Deser, C., and M. S. Timlin, 1997: Atmosphere–ocean interaction on weekly timescales in the North Atlantic and Pacific. J. Climate,10, 393–408.

  • Dix, M. R., and B. G. Hunt, 1995: Chaotic influences and the problem of deterministic seasonal predictions. Int. J. Climatol.,15, 729–752.

  • Drosdowsky, W., 1993a: An analysis of Australian seasonal rainfall anomalies: 1950–87. I: Spatial patterns. Int. J. Climatol.,13, 1–30.

  • ——, 1993b: An analysis of Australian seasonal rainfall anomalies: 1950–87. II: Temporal variability and teleconnection patterns. Int. J. Climatol.,13, 111–149.

  • ——, and M. Williams, 1991: The Southern Oscillation in the Australian region. Part I: Anomalies at the extremes of the oscillation. J. Climate,4, 619–638.

  • Frankignoul, C., 1985: Sea surface temperature anomalies, planetary waves, and air–sea feedback in the middle latitudes. Rev. Geophys.,23, 357–390.

  • Frederiksen, C. S., and R. C. Balgovind, 1994: The influence of the Indian Ocean/Indonesian SST gradient on the Australian winter rainfall and circulation in an atmospheric GCM. Quart. J. Roy. Meteor. Soc.,120, 923–952.

  • ——, and J. S. Frederiksen, 1996: A theoretical model of Australian northwest cloudband disturbances and Southern Hemisphere storm tracks: The role of SST anomalies. J. Atmos. Sci.,53, 1410–1432.

  • ——, D. P. Rowell, R. C. Balgovind, and C. K. Folland, 1999: Multidecadal simulations of Australian rainfall variability: The role of SSTs. J. Climate,12, 357–379.

  • Frederiksen, J. S., and C. S. Frederiksen, 1993: Monsoon disturbances, intraseasonal oscillations, teleconnection patterns, blocking, and storm tracks of the global atmosphere during January 1979: Linear theory. J. Atmos. Sci.,50, 1349–1372.

  • Gent, P. R., and J. C. McWilliams, 1990: Isopycnal mixing in ocean circulation models. J. Phys. Oceanogr.,20, 150–155.

  • Gordon, H. B., and S. P. O’Farrell, 1997: Transient climate change in the CSIRO coupled model with dynamic sea ice. Mon. Wea. Rev.,125, 875–907.

  • Hirst, A. C., S. P. O’Farrell, and H. B. Gordon, 2000: Comparisons of a coupled ocean–atmosphere model with and without oceanic eddy-induced advection. 1. Ocean spin-up and control integrations. J. Climate,13, 139–163.

  • Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis project. Bull. Amer. Meteor. Soc.,77, 437–471.

  • Levitus, S., 1982: Climatological Atlas of the World Ocean. National Oceanic and Atmospheric Administration, 173 pp and 17 microfiche.

  • Madden, R. A., and P. R. Julian, 1994: Observations of the 40–50 day tropical oscillation—A review. Mon. Wea. Rev.,122, 814–837.

  • Nicholls, N., 1988: El Niño–Southern Oscillation and rainfall variability. J. Climate,1, 418–421.

  • ——, 1989: Sea surface temperatures and Australian winter rainfall. J. Climate,2, 965–973.

  • ——, and K. K. Wong, 1990: Dependence of rainfall variability on mean rainfall, latitude and the Southern Oscillation. J. Climate,3, 163–170.

  • Palmer, T. N., and D. A. Mansfield, 1986: A study of wintertime circulation anomalies during past El Niño events using a high resolution general circulation model. II: Variability of the seasonal mean response. Quart. J. Roy. Meteor. Soc,112, 639–660.

  • Parker, D. E., C. K. Folland, A. Bevan, M. N. Ward, M. Jackson, and K. Maskell, 1995: Marine surface data for analysis of climatic fluctuations on interannual to century timescales. Natural Climate Variability on Decade-to-Century Time Scales, D. G. Martinson, K. Bryan, M. Ghil, M. M. Hall, T. R. Karl, E. S. Sarachik, S. Sorooshian, and L. D. Talley, Eds., National Academy Press, 241–250.

  • Richman, M. B., 1986: Rotation of principal components. J. Climatol.,6, 293–335.

  • Rowell, D. P., 1998: Assessing potential seasonal predictability with an ensemble of multidecadal GCM simulations. J. Climate,11, 109–120.

  • Saji, N. H., B. N. Goswami, P. N. Vinayachandran, and T. Yamagata, 1999: A dipole mode in the tropical Indian Ocean. Nature,401, 360–363.

  • Simmonds, I., 1990: A modelling study of winter circulation and precipitation anomalies associated with Australian region ocean temperatures. Aust. Meteor. Mag.,38, 151–162.

  • ——, and A. Rocha, 1991: The association of Australian winter climate with temperatures to the west. J. Climate,4, 1147–1161.

  • ——, ——, and D. Walland, 1992: Consequences of winter tropical pressure anomalies in the Australian region. Int. J. Climatol.,12, 419–434.

  • Smith, I. N., 1994: Indian Ocean sea-surface temperature patterns and Australian winter rainfall. Int. J. Climatol.,14, 287–305.

  • ——, 1995: A GCM simulation of global climate interannual variability: 1950–1988. J. Climate,8, 709–718.

  • ——, M. Dix, and R. J. Allan, 1997: The effect of greenhouse SSTs on ENSO simulations with an AGCM. J. Climate,10, 342–352.

  • Streten, N. A., 1981: Southern Hemisphere sea surface temperature variability and apparent associations with Australian rainfall. J. Geophys. Res.,86, 485–497.

  • Suppiah, R., and X. Wu, 1998: Surges, cross-equatorial flows and their links with the Australian summer monsoon circulation and rainfall. Aust. Meteor. Mag.,47, 113–130.

  • Tapp, R. G., and S. L. Barrell, 1984: The north-west Australian cloud band. J. Climatol.,4, 411–424.

  • Wallace, J. M., C. Smith, and Q. Jiang, 1990: Spatial patterns of atmosphere–ocean interaction in the northern winter. J. Climate,3, 990–998.

  • Walland, D. J., S. B. Power, and A. C. Hirst, 2000: Decadal climate variability simulated in a coupled general circulation model. Climate Dyn.,16, 201–211.

  • Watterson, I. G., 1998: An analysis of the global water cycle of present and doubled CO2 climates simulated by the CSIRO general circulation model. J. Geophys. Res.,103, 23 113–23 129.

  • ——, 2000: Southern midlatitude zonal wind vacillation and its interaction with the ocean in GCM simulations. J. Climate,13, 562–578.

  • ——, S. P. O’Farrell, and M. R. Dix, 1997: Energy and water transport in climates simulated by a general circulation model that includes dynamic sea ice. J. Geophys. Res.,102, 11 027–11 037.

  • ——, M. R. Dix, and R. Colman, 1999: A comparison of present and doubled CO2 climates and feedbacks simulated by three general circulation models. J. Geophys. Res.,104, 1943–1956.

  • Webster, P. J., A. M. Moore, J. P. Loschnigg, and R. R. Leben, 1999:Coupled ocean–atmosphere dynamics in the Indian Ocean during 1997–98. Nature,401, 356–360.

  • Whetton, P. H., 1990: Relationships between monthly anomalies of Australian region sea-surface temperature and Victorian rainfall. Aust. Meteor. Mag.,38, 31–41.

  • Wright, W. J., 1988: The low latitude influence on winter rainfall in Victoria, southeastern Australia. II: Relationships with the SOI and Australian region circulation. J. Climatol.,8, 547–576.

  • Zheng, X., and C. S. Frederiksen, 1999: Validating interannual variability in an ensemble of AGCM simulations. J. Climate,12, 2386–2396.

Fig. 1.
Fig. 1.

Mean simulated winds at 800 hPa in the Australian region in (a) Jan and (b) Jul. Shown are the wind vectors (scale: 10° longitude is 18 m s−1, minimum shown 2 m s−1), and their magnitude, shaded (light 5 m s−1 to 10 m s−1, dark above that). The modeled coastline is shown.

Citation: Journal of Climate 14, 9; 10.1175/1520-0442(2001)014<1901:WIRAST>2.0.CO;2

Fig. 2.
Fig. 2.

Standard deviation of the 800-hPa monthly mean wind components, zonally averaged across the Australian region in (a) Jan and (b) Jul. Shown are the GCM u (solid), NCEP u (long dash), GCM υ (short dash), and NCEP υ (dots).

Citation: Journal of Climate 14, 9; 10.1175/1520-0442(2001)014<1901:WIRAST>2.0.CO;2

Fig. 3.
Fig. 3.

Rotated principal components of normalized seasonal rainfall in run C, for Jun–Aug, presented as correlations (×100): (a) RPC1 (25.4%) and (b) RPC2 (23.0%).

Citation: Journal of Climate 14, 9; 10.1175/1520-0442(2001)014<1901:WIRAST>2.0.CO;2

Fig. 4.
Fig. 4.

Two of the RPCs of 800-hPa wind in Jan: (a) RPC1 (28.5% of variance), denoted monsoon surge;(b) RPC4 (6.5%), northerly. Shown are the wind vectors (scale: 10° longitude is 6 m s−1, minimum shown 0.2 m s−1), and their magnitude, shaded with 0.5 m s−1 gradations, first at 0.5 m s−1 last at 4 m s−1.

Citation: Journal of Climate 14, 9; 10.1175/1520-0442(2001)014<1901:WIRAST>2.0.CO;2

Fig. 5.
Fig. 5.

As in Fig. 4 but for Jul: (a) RPC4 (9.8% of variance), northwesterly; (b) RPC11 (4.2%), northerly.

Citation: Journal of Climate 14, 9; 10.1175/1520-0442(2001)014<1901:WIRAST>2.0.CO;2

Fig. 6.
Fig. 6.

Regression lifecycle of a monsoon surge (in run M) (from top to bottom); lags −20, 0, +20 days. Surface pressure is shown by contours (long-dashed for negative values, short-dashed for zero), σ = 0.80 winds (interpolated to a 7° grid) by vectors.

Citation: Journal of Climate 14, 9; 10.1175/1520-0442(2001)014<1901:WIRAST>2.0.CO;2

Fig. 7.
Fig. 7.

Regression life cycle of a Jul northwesterly event (in run M) (from top to bottom); lags −6, −4, −2, 0, and +2 days. Surface pressure is shown by contours (long-dashed for negative values, short-dashed for zero), σ = 0.34 winds (interpolated to a 7° grid) by vectors.

Citation: Journal of Climate 14, 9; 10.1175/1520-0442(2001)014<1901:WIRAST>2.0.CO;2

Fig. 8.
Fig. 8.

Percentage of variance of lag 0 monthly anomalies of rainfall related to the 12 wind RPCs in (a) Jan and (b) Jul.

Citation: Journal of Climate 14, 9; 10.1175/1520-0442(2001)014<1901:WIRAST>2.0.CO;2

Fig. 9.
Fig. 9.

Percentage of variance of lag 0 monthly anomalies of surface temperature related to the 12 wind RPCs in (a) Jan and (b) Jul.

Citation: Journal of Climate 14, 9; 10.1175/1520-0442(2001)014<1901:WIRAST>2.0.CO;2

Fig. 10.
Fig. 10.

Percentage of variance of lag −1 monthly anomalies of surface temperature related to the 12 wind RPCs in (a) Jan and (b) Jul.

Citation: Journal of Climate 14, 9; 10.1175/1520-0442(2001)014<1901:WIRAST>2.0.CO;2

Fig. 11.
Fig. 11.

Correlation (×100) of lag 0 monthly anomalies with monsoon surge: (a) rainfall and (b) surface temperature. As in following figures, the contours ±10, ±30, ±50, and ±70 are shown, together with some local peaks.

Citation: Journal of Climate 14, 9; 10.1175/1520-0442(2001)014<1901:WIRAST>2.0.CO;2

Fig. 12.
Fig. 12.

Correlation (×100) of monthly anomalies with Jan northerly: (a) Jan rainfall and (b) Feb surface temperature.

Citation: Journal of Climate 14, 9; 10.1175/1520-0442(2001)014<1901:WIRAST>2.0.CO;2

Fig. 13.
Fig. 13.

Correlation (×100) of monthly anomalies with Jul northwesterly: (a) Jul rainfall and (b) Aug surface temperature.

Citation: Journal of Climate 14, 9; 10.1175/1520-0442(2001)014<1901:WIRAST>2.0.CO;2

Fig. 14.
Fig. 14.

Correlation (×100) of monthly anomalies with Jul northerly: (a) Jul rainfall and (b) Aug surface temperature.

Citation: Journal of Climate 14, 9; 10.1175/1520-0442(2001)014<1901:WIRAST>2.0.CO;2

Fig. 15.
Fig. 15.

Correlations between indices related to four RPCs: (a) wind with rainfall (lagged), (b) wind with SSTA (lagged), (c) rainfall with SSTA (lagged).

Citation: Journal of Climate 14, 9; 10.1175/1520-0442(2001)014<1901:WIRAST>2.0.CO;2

Fig. 16.
Fig. 16.

Correlation (×100) of Jun–Aug seasonal mean anomalies of (a) rainfall, and (b) surface temperature with the Jun–Aug mean of the rainfall index associated with the Jul northwesterly wind RPC.

Citation: Journal of Climate 14, 9; 10.1175/1520-0442(2001)014<1901:WIRAST>2.0.CO;2

Fig. 17.
Fig. 17.

As in Fig. 16 but for the Jul northerly RPC.

Citation: Journal of Climate 14, 9; 10.1175/1520-0442(2001)014<1901:WIRAST>2.0.CO;2

Fig. 18.
Fig. 18.

Correlation (×100) of Dec–Feb seasonal mean anomalies of (a) rainfall and (b) surface temperature with the Dec–Feb mean of the rainfall index associated with the Jan northerly wind RPC.

Citation: Journal of Climate 14, 9; 10.1175/1520-0442(2001)014<1901:WIRAST>2.0.CO;2

Fig. 19.
Fig. 19.

Correlation (×100) of Jun–Aug rainfall with Jul northwesterly SST index in (a) May and (b) Sep.

Citation: Journal of Climate 14, 9; 10.1175/1520-0442(2001)014<1901:WIRAST>2.0.CO;2

Table 1.

Standard deviation, over all years, of pattern amplitudes, in the lag 0 month, for two runs (C, the coupled model, and S, the specified SST model) and three quantities. Values are given for four featured 800-hPa wind patterns: Jan RPC1, denoted Mon S (monsoon surge); Jan RPC4, Jan N; Jul RPC4, Jul NW; and Jul RPC11, Jul N. Values other than wind in C were obtained by projection, as explained in the text. The SST variation in S is zero.

Table 1.
Save