Abstract

Year-to-year changes are found in Australian precipitation (APP) covarying with those in sea surface temperature (SST) and troposphere moisture flux (MF) over the three oceans surrounding Australia for 40 yr from 1958 to 1997. Australia’s wet (dry) years are associated with warm (cool) SST anomalies surrounding Australia and convergent (divergent) MF anomalies directly overhead. Differences in APP (SST) between wet and dry years can reach 0.75 m (1.2°C) in northeast Australia (subtropical Indian Ocean). Wet (dry) years often occur during La Niña (El Niño), but significant differences in covarying SST, MF, and APP anomalies from one El Niño to the next are found, indicating that regional climate changes also influence APP. Covarying SST and MF anomalies on basin space scales and interannual timescales are found to take 2–3 yr to propagate eastward from Africa to Australia. This propagation occurs in association with the Antarctic Circumpolar Wave (ACW) in the Southern Ocean, the north branch of the ACW in the Indian Ocean, and the global El Niño–Southern Oscillation wave in the tropical ocean. A statistical climate prediction system based upon the slow eastward propagation of SST anomalies and their nearly one-to-one relationship with APP anomalies is constructed, yielding significant hindcast skill for predicting interannual APP anomalies at lead times of 1 and 2 yr. Best hindcast skill for the extratropical portion of Australia derives from the ACW south of Australia and the north branch of the ACW west of Australia. Eastward propagation of SST anomalies in these two oceanic domains is capable of predicting more than 50% of the total interannual variance over Victoria and New South Wales and over Western Australia poleward of 20°S over the 40-yr record. This percentage is much better than expected from chance or persistence, demonstrating the importance of the ACW upon year-to-year changes in APP at these latitudes.

1. Introduction

Australia is located in the latitudinal band 12°–44°S, with the Indian Ocean to the west, the Pacific Ocean to the east, the Southern Ocean to the south, and the tropical ocean to the north, making its climate susceptible to interannual changes in all four oceans. Stone et al. (1996) already found year-to-year changes in Australia winter precipitation significantly influenced by the El Niño–Southern Oscillation (ENSO) in the tropical ocean to the north. On the other hand Nicholls (1989) found year-to-year changes in Australia winter precipitation correlated best with sea surface temperature (SST) changes in the subtropical Indian Ocean to the west. Recently these changes in the Indian Ocean SST have been shown to be influenced by a number of climate phenomena operating on interannual timescales. Tourre and White (1997) found interannual SST and sea level pressure (SLP) anomalies in the tropical Indian Ocean propagating slowly eastward, observed by White and Cayan (2000) to be part of a global climate signal called the global ENSO wave that conducts covarying SST and SLP anomalies across the tropical Indian Ocean and into the tropical Pacific and Atlantic Oceans. Van Loon and Shea (1985) found precursors for ENSO in SLP anomalies south of Australia in the Southern Ocean. A decade later White and Peterson (1996) found these SLP anomalies south of Australia propagating slowly eastward around the Southern Ocean together with SST anomalies as part of a global climate signal called the Antarctic Circumpolar Wave (ACW). Peterson and White (1998) found interannual SST anomalies in the Southern Ocean associated with the ACW propagating slowly equatorward and eastward into the subtropical and tropical Indian Ocean. In this study we answer the question: do these eastward propagating SST anomalies in the tropical ocean north of Australia, in the subtropical Indian Ocean west of Australia, and in the Southern Ocean south of Australia influence year-to-year changes in Australian precipitation (APP)?

The ACW is a nominal 4-yr period climate signal in the high-latitude ocean–atmosphere system, propagating eastward at an average speed of 45° of longitude per year, taking approximately 8 yr to circle the Southern Ocean with global zonal wavenumber 2 (White and Peterson 1996; Jacobs and Mitchell 1996). It displays significant peak spectral energy density rising above the background red noise in zonal wavenumber-frequency spectra of SST anomalies. Qui and Jin (1997) determined that the ACW in the Southern Ocean is independent of ENSO in the tropical Pacific Ocean. The ACW is characterized by a persistent contemporaneous phase relationship between warm (cool) SST anomalies and poleward (equatorward) meridional surface wind (MSW) anomalies. Since SST anomalies are advected eastward by the broadscale flow of the Antarctic Circumpolar Current over most of the Southern Ocean, this indicated to White and Peterson (1996), Jacobs and Mitchell (1996), and Qui and Jin (1997) that these MSW anomalies are in thermodynamical equilibrium with underlying SST anomalies, suggesting that maintenance of the ACW involves coupling between the extratropical ocean and atmosphere. This is reflected in the global spiral pattern in covarying SST and SLP anomalies found in the ACW by White et al. (1998). White et al. (1998) went on to construct a coupled ocean–atmosphere model of the Southern Ocean, which explains the 4-yr quasi-periodicity, propagation speed, zonal wavelength, and global spiral structure of the ACW.

The global ENSO wave is a nominal 4-yr period climate signal in the tropical ocean–atmosphere system, propagating eastward along the annual mean location of the ITCZ across the Indian, Pacific, and Atlantic Oceans. It is composed of global zonal wavenumbers 1 and 2, the former taking approximately 4 yr to circle the globe and the latter taking approximately 6 yr to propagate from sources in the western Indian Ocean to its terminus in the eastern Atlantic Ocean (White and Cayan 2000). It displays significant peak spectral energy density rising above the background red noise in zonal wavenumber-frequency spectra of SST anomalies. The global ENSO wave has low (high) SLP anomalies in thermodynamical equilibrium with warm (cool) SST anomalies, indicating that maintenance of the global ENSO wave against dissipation derives from coupling between the tropical ocean and atmosphere. This slow eastward propagating wave is superimposed upon a global standing wave of the Southern Oscillation in anomalous SLP and surface temperature (ST) deriving from anomalous Walker cell activity in the equatorial plane of the troposphere associated with the El Niño cycle in the eastern equatorial Pacific Ocean (e.g., Webster 1994). As such the global ENSO wave constitutes an additional influence contributing to ENSO variability throughout the global tropical domain, which is more significant when the Southern Oscillation is weak during years between El Niño and La Niña.

Recently Peterson and White (1998) observed covarying SST and SLP anomalies spreading eastward and southward from the subtropical western South Pacific Ocean, driven there by ENSO-induced anomalous Hadley cell activity (see their Fig. 1). These covarying SST and SLP anomalies joined the ACW in the Southern Ocean, taking 2–4 yr to propagate through Drake Passage into the Atlantic and Indian Ocean sectors of the Southern Ocean. There Peterson and White (1998) found covarying SST and SLP anomalies propagating slowly equatorward from the Southern Ocean into both the tropical Atlantic and Indian Oceans from 1982 to 1996. The north branch of the ACW in the Indian Ocean took 2–3 yr to propagate equatorward from south of Madagascar near 50°S to the tropical ocean north of Australia over this period. There it merged with that portion of the global ENSO wave in covarying SST and SLP anomalies taking 1–2 yr to propagate eastward across the tropical Indian Ocean from the vicinity of Madagascar to Indonesia (Tourre and White 1997; White and Cayan 2000). These two independent climate change phenomena (i.e., the ACW and the global ENSO wave) often act in phase to influence Indian Ocean SST anomalies but not always. One of the major questions to be answered in this study is: what is the relative importance of the global ENSO wave in the tropical ocean north of Australia, the north branch of the ACW in the subtropical Indian Ocean west of Australia, and the ACW in the Southern Ocean south of Australia on year-to-year changes in SST around Australia and in precipitation over Australia?

White and Cherry (1999) already answered a similar question for year-to-year changes in winter precipitation over New Zealand, which had been known for some time to be uncorrelated with ENSO activity. The reason for this lay in the influence that the ACW has upon New Zealand winter precipitation. They found wet precipitation anomalies occurring for two reasons. First, when warm (cool) SST and poleward (equatorward) MSW anomalies occur north (south) of New Zealand, then anomalous low-level wind convergence takes place over New Zealand. Second, when warm (cool) SST and poleward (equatorward) MSW anomalies occur east (west) of New Zealand, then anomalous cyclonicity takes place over New Zealand. They utilized the nearly one-to-one relationship between SST anomalies around New Zealand and winter precipitation anomalies over New Zealand to develop a statistical climate prediction system for the latter. This prediction system utilized the eastward propagation of the ACW in the Southern Ocean and Tasman Sea to hindcast and forecast New Zealand winter precipitation, yielding significant skill from one winter to the next over the period of available data from 1982 to 1996.

Presently the prediction of APP is conducted for 60–90-day lead times and is based on the persistence of interannual climate change. The Australia Bureau of Meteorology utilizes empirical orthogonal functions (EOFs) of SST anomalies in the global ocean to derive terciles of precipitation anomalies over Australia (Drosdowsky and Chambers 1998). The Queensland Department of Primary Industries utilizes the phases of the Southern Oscillation index (SOI) to derive precipitation anomalies over northeast Australia (Partridge 1994; Stone et al. 1996). The SOI is the difference between Tahiti and Darwin SLP and is a measure of the intensity of ENSO throughout the tropical Indo-Pacific domain (e.g., Webster 1994). Both of these prediction systems focus on the influence that tropical ENSO has on APP. They largely ignore influences from year-to-year changes in SST variability in the Indian Ocean and in the Southern Ocean. Recently Power et al. (2000) determined that the statistical relationship between the SOI and APP is modulated by interdecadal variability over the past century, reduced to insignificance during decades of global tropical warming. On the other hand R. C. Stone (1999, personal communication) finds little evidence of interdecadal variability degrading his ability to predict precipitation in Queensland and New South Wales using SOI phases.

In the present study we construct a statistical climate prediction system for year-to-year changes in APP similar to that developed by White and Cherry (1999) for New Zealand precipitation. This is made possible by the nearly one-to-one relationship between year-to-year changes in covarying SST and vertical average moisture flux (MF) over the ocean surrounding Australia and the year-to-year changes in APP over Australia. This climate prediction system utilizes the slow eastward propagation of covarying SST and MF anomalies in the Indian and Southern Oceans to predict year-to-year changes in APP. We examine the influence of three separate climate change phenomena that account for this eastward phase propagation; that is, the ACW south of Australia, the north branch of the ACW in the Indian Ocean west of Australia, and the global ENSO wave in the tropical ocean north of Australia. We examine MF anomalies in the troposphere, the convergence of which is dynamically linked to APP and evaporation anomalies through moisture conservation, to understand how SST anomalies influence APP anomalies. Within this context anomalous APP is assumed to be a consequence of the set of coupled ocean–atmosphere–terrestrial thermodynamical mechanisms that are responsible for the slow eastward propagation of the ACW, the north branch of the ACW, and of the global ENSO wave past Australia. We finish by finding that this statistical climate prediction system has significant hindcast skill in predicting APP anomalies at lead times of 1 and 2 yr over the available 40-yr record from 1958 to 1997. Over this record the prediction system is capable of hindcasting more than 50% of the total interannual variability in precipitation over Australia poleward of 20°S. This is much better than can be achieved either by chance or by persistence. It remains to be seen whether these statistics will remain stationary when longer records are considered (Simmonds and Hope 1997; Allan 2000]. But it does demonstrate the influence of the ACW upon year-to-year changes in APP at these extratropical latitudes.

2. Data and methods

We utilize ST, SLP, and vertical average troposphere MF datasets extending from 60°S to 20°N on a 1.875° lat × 1.875° long grid for the 40 yr from 1958 to 1997 from the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalysis (Kalnay et al. 1996). Uncertainty in the SLP reanalysis south of 50°S was doubled for the period 1979–92 due to miscoding of bogus Australian SLP data estimated from satellite data (Kalnay 1996). The ST anomalies over the land are surface air temperature anomalies, while those over the ocean are SST anomalies. The MF anomalies were computed as the integral of mass-weighted moisture flux anomalies at standard pressure levels extending from sea level to 200 hPa (Chen et al. 1996); south of 50°S these MF anomalies may have been corrupted by the greater uncertainty in SLP. We also utilize gridded estimates of land precipitation provided by Hulme and Jones (1993) extending over Australia from 50°S to 0° for the 40 yr from 1958 to 1997 and interpolated from individual station locations onto a 1.83° lat × 1.83° long grid. These gridded precipitation data have recently been updated to 1997 by Hulme and are available from his web site at the University of East Anglia (www.cru.uea.ac.uk/~mikeh). Since the interpolation of precipitation from individual station locations to a regular grid proved to be critical, we interpolated monthly estimates of ST, SLP, and MF onto the same grid as APP, extending this to the entire Southern Hemisphere. This allowed all three datasets to be registered onto the same grid. Then we computed the mean annual cycle for the 40-yr record, subtracting long-term monthly means from individual monthly means to produce monthly anomalies about the mean annual cycle.

Earlier White and Peterson (1996) established that covarying SST and SLP anomalies associated with the ACW in the Southern Ocean at 56°S yielded zonal wavenumber-frequency spectra that contained significant peak spectral energy density at a nominal period of 4 yr and a nominal eastward wavelength of 11 000 km. Similarly White and Cayan (2000) established that covarying SST and SLP anomalies associated with a global ENSO wave contained significant peak spectral energy density at a nominal period of 4 yr and eastward wavelengths >18 000 km. This indicates that in both the global tropical ocean and the Southern Ocean the zonal wavenumber-frequency spectra of SST (SLP) anomalies reveal signals that rise above the background red (white) noise. Thus we isolate this 4-yr period-scale variability from monthly, seasonal, and biennial variability on the high-frequency side and decadal and interdecadal variability on the low-frequency side by bandpassing the time sequences of monthly anomalies using a period admittance window with a half-power points at 2 and 7 yr (Kaylor 1977). This particular admittance window isolates a peak interannual variability of 3–6 yr periods, as observed by White and Peterson (1996) and White and Cayan (2000). To avoid loss of data at the ends of each time sequence due to high-pass filtering, maximum entropy spectral analysis was applied (Andersen 1974) using spectral coefficients to extend the sequences by an amount equal to half the filter width. So for the nominal 40-yr record the bandpass filter yields time records with approximately 20 independent realizations determined by autocorrelation analysis (Emery and Thomson 1996, p. 257), yielding approximately 20 effective degrees of freedom at each grid point.

Distributions of the root-mean-square (rms) of interannual, SST, meridional MF (MMF), zonal MF (ZMF), and APP anomalies for the 40-yr record are displayed in Fig. 1. The largest rms of interannual APP anomalies occurs in northeast Australia, extending from Sydney to Darwin. There it achieves magnitudes ranging from 5 to 15 mm month−1, which is larger than the 4–5 mm month−1 in southwestern Australia. The largest rms of interannual SST anomalies of approximately 0.2°C occurs over the subtropical Indian Ocean between 10° and 30°S, with smaller magnitudes of approximately 0.1°C occurring poleward and equatorward. The largest rms of interannual ZMF anomalies of 20–30 kg m−1 s−1 occurs in the equatorial Indian and Pacific Oceans, with the smallest rms of 10 kg m−1 s−1 occurring over Australia and the subtropical Indian Ocean to the west, where the latter regions are dominated by a subtropical high pressure system (Sturman and Tapper 1996). The largest rms of interannual MMF anomalies of 5–10 kg m−1 s−1 occurs in the tropical Indian and Pacific Oceans, extending poleward along the South Pacific convergence zone in the western South Pacific Ocean east of Australia and in the central Indian Ocean west of Australia. The smallest rms of MMF anomalies with magnitudes <5 kg m−1 s−1 occurs directly over Australia. Where the larger tropical rms of MMF anomalies overlap onto Australia in the northeast, they trace out a pattern similar to that of the rms of interannual APP anomalies. This is consistent with the dynamical relationship between interannual APP anomalies and divergence of interannual MMF anomalies, which together with anomalous surface evaporation must balance in the moisture budget of the troposphere (Chen et al. 1996).

Fig. 1.

Distribution of the root-mean-square (rms) of interannual APP anomalies, interannual SST anomalies, interannual MMF anomalies, and interannual ZMF anomalies in the vicinity of Australia. Contour intervals are 1 mm month−1, 0.02°C, 2.5 kg m−1 s−1, and 2.5 kg m−1 s−1, respectively. Hatching is only for effect.

Fig. 1.

Distribution of the root-mean-square (rms) of interannual APP anomalies, interannual SST anomalies, interannual MMF anomalies, and interannual ZMF anomalies in the vicinity of Australia. Contour intervals are 1 mm month−1, 0.02°C, 2.5 kg m−1 s−1, and 2.5 kg m−1 s−1, respectively. Hatching is only for effect.

Magnitudes in the rms of interannual APP anomalies of 10 mm month−1 seem initially to be relatively small in comparison with what one expects on synoptic timescales, but when considered in their interannual context they are considerable. For example, the rms of interannual APP anomalies is 10 mm month−1 in northeast Australia, but interannual anomalies are positive or negative for over a year and more, amounting to 12 cm yr−1 or so. Furthermore peak variability in wet and dry years will range between ±2–3 standard deviations, yielding peak anomalies of 24–36 cm yr−1. Thus differences in the amount of precipitation between wet and dry years can range from 48 to 72 cm yr−1 or one-half to three-quarters of a meter. This is significant and can have enormous impact upon reservoir, farm, and ranch management.

Magnitudes in the rms of interannual SST anomalies in the subtropical Indian Ocean of 0.2°C seem initially to be relatively small compared to those in the eastern equatorial Pacific Ocean associated with El Niño. Yet Peterson and White (1998) showed the largest rms of interannual SST anomalies in the eastern equatorial Pacific to be 0.6°C, so interannual anomalies in the subtropical Indian Ocean are about one-third of those in the eastern equatorial Pacific Ocean associated with El Niño. Using the same formula as with interannual APP anomalies in the previous paragraph this 0.2°C (0.6°C) in rms magnitude can be seen to be associated with year-to-year changes as large as 0.8°–1.2°C (2.4°–3.6°C). These El Niño SST anomalies are considered capable of influencing climate the world over through their influence upon Hadley and Walker cell activity (e.g., Barnett et al. 1994). On the other hand White et al. (1998) demonstrated that these extratropical SST anomalies are capable of influencing overlying SLP anomalies in the ACW enough so that feedback from the atmosphere to the ocean operating through surface wind anomalies can maintain the SST anomalies against dissipation in their 8-yr odyssey around the Southern Ocean.

3. Placing Australian precipitation within the context of the ACW and the global ENSO wave

We begin by placing climate variability in the region around and over Australia within the context of the ACW in the Southern Ocean (White and Peterson 1996) and the global ENSO wave in the Tropics (White and Cayan 2000). This is possible only after 1982 when satellite SST estimates became available over the entire Southern Ocean, complimenting in situ SLP estimates from First GARP (Global Atmospheric Research Program) Global Experiment (FGGE) buoys, and allowing the ACW to be observed in both SST and SLP anomalies. We accomplish this task by conducting an extended EOF (EEOF) analysis of interannual ST and SLP anomalies extending from 20°N to 60°S and around the globe from 180° to 180° for 16 yr from 1982 to 1997 (Fig. 2). As described in Weare and Nasstrom (1982) EEOF analysis begins by lagging adjacent maps in a time sequence of maps to produce an array of maps in time and temporal lag space upon which EOF analysis can be applied (Preisendorfer and Mobley 1988). The resulting EEOF eigenvectors are modulated by a time sequence of amplitudes as in an usual EOF analysis (top, Fig. 2) with the EEOF eigenvectors themselves comprising a sequence of temporally lagged maps, which in this analysis consists of nine maps extending over 48 months of lag in 6-month increments (bottom, Fig. 2). These lag sequences yield the evolution of interannual ST and SLP variability in time and space for the dominant mode over the 16 yr from 1982 to 1997. Thus each EEOF lag sequence animates most of one complete cycle of interannual variability associated with the ACW and the global ENSO wave. Weights in the central map at 24 lag months can be multiplied by time amplitudes to reconstitute the contribution of this EEOF mode to the original time–latitude–longitude matrix of anomalies.

Fig. 2.

(top) Time sequences of amplitudes associated with the dominant EEOF modes of interannual ST and SLP anomalies for the 12 yr from 1984 to 1995, with 2 yr truncated at the beginning and end of the data record. (bottom) Lag sequences of maps associated with dominant EEOF modes of interannual ST and SLP anomalies, with each map extending over the entire Southern Hemisphere (i.e., from 180° to 180° and from 20°N to 60°S). Each lag sequence extends over 48 lag months where individual positive (yellow–red) and negative (blue) ST and SLP weights can be seen to propagate by following positive and negative anomalies from one map to the next. Contour intervals for spatial weights of both variables are unitless, ranging from ±0.60 on the color bar, and temporal amplitudes range from ±1.0°C and ±2.0 hPa. Magnitudes are recovered by multiplying spatial weights by temporal amplitudes.

Fig. 2.

(top) Time sequences of amplitudes associated with the dominant EEOF modes of interannual ST and SLP anomalies for the 12 yr from 1984 to 1995, with 2 yr truncated at the beginning and end of the data record. (bottom) Lag sequences of maps associated with dominant EEOF modes of interannual ST and SLP anomalies, with each map extending over the entire Southern Hemisphere (i.e., from 180° to 180° and from 20°N to 60°S). Each lag sequence extends over 48 lag months where individual positive (yellow–red) and negative (blue) ST and SLP weights can be seen to propagate by following positive and negative anomalies from one map to the next. Contour intervals for spatial weights of both variables are unitless, ranging from ±0.60 on the color bar, and temporal amplitudes range from ±1.0°C and ±2.0 hPa. Magnitudes are recovered by multiplying spatial weights by temporal amplitudes.

First modes of this EEOF analysis explain about 75% of the total interannual variance for ST and SLP anomalies for the 16-yr period. Time sequences of amplitudes associated with dominant EEOF modes (top, Fig. 2) show peaks in 1987 and 1991 and troughs in 1985, 1989, and 1993 corresponding to the El Niño cycle in the eastern equatorial Pacific Ocean with a nominal 4-yr period scale. Associated lag sequences of maps (bottom, Fig. 2) display some combination of zonal global wavenumbers 1 and 2 dominating covarying ST and SLP variability in both the Tropics and extratropics. The central map of 24 lag months displays positive (warm) ST and negative (low) SLP weights in the eastern equatorial Pacific Ocean during El Niño in mid-1987 and late 1991.

In the tropical domain (20°S–20°N) these lag sequences display warm ST (red) and low SLP (blue) weights propagating eastward together from the eastern Indian and western Pacific Oceans near Indonesia in the lag month 0 to the eastern Atlantic Ocean and western Indian Ocean in lag month 48 over the 4-yr lag period. At lag months 24 and 30, as these anomalies approach the west coast of South America, they follow the ITCZ across Central America between 5° and 15°N, propagating eastward into the tropical North Atlantic Ocean in lag month 36 through the Caribbean Sea. This slow eastward propagation of covarying ST and SLP anomalies is consistent with the global ENSO wave following the annual mean location of the ITCZ around the global tropical ocean (White and Cayan 2000). These lag sequences also display the slow eastward propagation of the ACW in the Southern Ocean (40°–70°S), with warm ST (red) and low SLP (blue) weights in the Atlantic and eastern Pacific sectors of the Southern Ocean, respectively, in the lag month 0 propagating eastward into the western Pacific and Indian Ocean sectors, respectively, in the lag month 48 over the 4-yr lag period.

The impact of the slow eastward propagation of these two global SST waves upon Australia can be understood by tracking warm ST and low SLP weights that surround Australia near the end of the lag sequence (i.e., at lag month 42) back into the western Indian Ocean two lag years earlier. In the subtropical Indian Ocean from 10° to 30°S, warm ST weights off the west coast of Australia in lag month 42 can be seen to have propagated equatorward and eastward from the Southern Ocean south of Madagascar near 40°S in lag month 12, there associated with the slow eastward propagation of the warm ST weights associated with the ACW in the Southern Ocean. This constitutes the north branch of the ACW in the Indian Ocean (Peterson and White 1998). In the Southern Ocean from 30° to 60°S the warm ST weights south of Australia in lag month 42 can also be seen to have propagated eastward from south of Madagascar near 40°S in lag month 12, manifesting the ACW (White and Peterson 1996). In the tropical ocean north of Australia warm ST weights in lag month 42 can be seen to have propagated eastward from the tropical Indian Ocean north of Madagascar at 10°S in lag month 12, associated with the slow eastward propagation of the warm ST weights associated with the global ENSO wave (White and Cayan 2000). The question is whether the three regional climate change phenomena influence year-to-year changes in precipitation over Australia.

4. Spatial phase relationships between SST, MF, and APP anomalies

Now we display maps of interannual SST and MF anomalies side by side with maps of interannual MF and APP anomalies averaged over December–January–February (summer) for five different El Niño years; that is, 1972–73, 1976–77, 1982–83, 1986–87, and 1991–92 (Fig. 3). The MF anomalies are indicated with streamlines that connect anomalous MF vectors, allowing the reader to trace the anomalous transport of moisture in the troposphere. Convergences in these anomalous MF streamlines are related to anomalous APP and anomalous evaporation in order to conserve the anomalous moisture content in the troposphere. Southern Hemisphere summer was chosen because El Niño activity in the eastern equatorial Pacific Ocean achieves peak SST anomalies during this season, projecting its greatest influence to the rest of the globe via Walker and Hadley cell activity (Barnett et al. 1994). We average interannual ZMF and MMF anomalies individually to form summer mean estimates for December–February (DJF) before plotting them as streamlines on the maps displayed in Fig. 3. In this display quantitative information about the magnitude is lost, while qualitative information about the sources of anomalous moisture is enhanced. Absolute magnitude of the anomalous moisture flux vectors can be estimated from the rms of the interannual ZMF and MMF anomalies (Fig. 1). The accuracy of NCEP MF anomalies are suspect over the ocean due the lack of observed temperature, moisture, and wind profiles from rawindsondes, but the availability of observed profiles over Australia should make MF anomalies reliable there and over the surrounding ocean. The NCEP reanalysis conserves heat and moisture in assimilating diverse observations into the reanalysis model (Kalnay et al. 1996).

Fig. 3.

(left) SST anomalies in color over the Indian, Southern, and South Pacific Oceans overlaid with anomalous MF streamlines. (right) Interannual APP anomalies contoured over Australia, overlaid with anomalous MF streamlines. These distributions are displayed during DJF of El Niño years 1972–73, 1976–77, 1982–83, 1986–87, and 1991–92. Contour intervals are given by the color bar, with blue colors indicating negative anomalies and yellow to red colors indicating positive anomalies ranging over ±0.4°C and ±10 mm month−1.

Fig. 3.

(left) SST anomalies in color over the Indian, Southern, and South Pacific Oceans overlaid with anomalous MF streamlines. (right) Interannual APP anomalies contoured over Australia, overlaid with anomalous MF streamlines. These distributions are displayed during DJF of El Niño years 1972–73, 1976–77, 1982–83, 1986–87, and 1991–92. Contour intervals are given by the color bar, with blue colors indicating negative anomalies and yellow to red colors indicating positive anomalies ranging over ±0.4°C and ±10 mm month−1.

Inspection of Fig. 3 reveals a number of salient features. First, El Niño is generally associated with warm SST anomalies in the western tropical Indian Ocean and with cool SST and high SLP anomalies about and over Australia, which is consistent with Tourre and White (1995) and with that observed in the ST lag sequence in Fig. 2. Second, warm (cool) SST anomalies in the extratropics are generally aligned with poleward (equatorward) MF vector anomalies, similar to the alignment observed between warm SST and poleward MSW anomalies in the ACW by White and Peterson (1996). This suggests that anomalous moisture transport on interannual timescales occurs in equilibrium with underlying SST anomalies. Third, El Niño is usually associated with dry APP anomalies in the Northern Territory, Queensland, and Victoria, which is consistent with that observed by Partridge (1994), the exceptions being Northern Territory and Queensland in 1976–77 and Tasmania in 1991–92. And fourth, differences exist from one El Niño year to the next in the distribution of SST, MF streamlines, and APP anomalies around and over Australia. For example in 1976–77 wet APP anomalies occurred in northeast Australia in association with MF anomalies that originated in the Tasman Sea and Southern Ocean, while dry APP anomalies in northeast Australia during the other El Niño years in Fig. 2 occurred in association with MF anomalies originating in the Indian Ocean, either crossing Australia or extending around it to the south. These differences in regional anomaly patterns during different El Niños suggest that regional climate change phenomena influence year-to-year changes in APP and modify influences that may stem directly from El Niño.

To establish the characteristics of regional climate change phenomena around and over Australia, we examine the evolution of standardized interannual SST and MF (both MMF and ZMF) anomalies over the Indian, Southern, and South Pacific Oceans together with standardized interannual APP anomalies. We accomplish this by conducting an EEOF analysis on each variable independently over the nominal 40-yr record, as described previously in connection with Fig. 2. With MMF and ZMF anomalies the EEOF analysis was conducted jointly to yield a true vector representation. Here we extend the analysis to 50°S, expecting adequate coverage of surface marine weather observations from 35° to 50°S from volunteer observing ships transiting New Zealand, Tasmania, Australia, South Africa, and Europe from 1958 to 1997.

We find the first EEOF mode in SST, MMF, and APP explaining from 46% to 57% of the interannual variance over the nominal 40-yr record (Fig. 4). Moreover each EEOF lag sequence (bottom, Fig. 4) has an amplitude time sequence associated with it (top, Fig. 4) that correlates significantly with the others, with SST and APP (MMF and APP) time sequences correlating at 0.90 (0.85). This occurs naturally without having to conduct a joint EEOF analysis among the three variables, indicating that the evolution of APP anomalies on interannual timescales is inextricably linked to the evolution of SST and MMF anomalies over the surrounding ocean. The amplitude time sequences achieve peak amplitude in 1962, 1966, 1969, 1973, 1977, 1983, 1987, and 1991, modulating the central map of 24 lag months in the corresponding lag sequence. These peak years occur during El Niño in the eastern equatorial Pacific Ocean and are in phase with those presented in Fig. 2. At this central map at 24 lag months we see warm SST weights occupying the western and central Indian Ocean, cool SST anomalies surrounding Australia on all four sides, and dry APP weights occupying the coastal domain of Australia. On the other hand, peak dry APP weights can be seen occurring 6 lag months earlier when cool SST weights are more intense in the Indian and Southern Oceans. Peak wet APP weights occur near the end of the lag sequence when warm SST weights surround Australia on all four sides. Weights in the central map can be multiplied by time amplitudes to reconstitute the contribution of this EEOF mode to the original time–latitude–longitude matrix of anomalies.

Fig. 4.

(top) Time sequences of amplitudes associated with the dominant EEOF modes of interannual SST, MMF, and APP anomalies for the nominal 40-yr record. (bottom, left, and middle) Lag sequences of maps associated with the dominant EEOF modes of interannual SST and MMF anomalies, with each map extending over the Southern Hemisphere from 30°E to 180° and from 0° to 50°S. (bottom right) The lag sequence of maps associated with the dominant EEOF mode of interannual APP anomalies over the same record, with each map including Australia extending from 0° to 50°S. Each of these lag sequences extends over 48 lag months. Positive (negative) SST and APP weights are colored yellow to red (blue), and poleward (equatorward) MMF weights are colored yellow to red (blue). Contour intervals for spatial weights of all three variables are given by the color bar; temporal amplitudes range from ±0.3°C, ±30 kg m−1 s−1, and ±30 cm yr−1. Magnitudes are recovered by multiplying spatial weights by temporal amplitudes.

Fig. 4.

(top) Time sequences of amplitudes associated with the dominant EEOF modes of interannual SST, MMF, and APP anomalies for the nominal 40-yr record. (bottom, left, and middle) Lag sequences of maps associated with the dominant EEOF modes of interannual SST and MMF anomalies, with each map extending over the Southern Hemisphere from 30°E to 180° and from 0° to 50°S. (bottom right) The lag sequence of maps associated with the dominant EEOF mode of interannual APP anomalies over the same record, with each map including Australia extending from 0° to 50°S. Each of these lag sequences extends over 48 lag months. Positive (negative) SST and APP weights are colored yellow to red (blue), and poleward (equatorward) MMF weights are colored yellow to red (blue). Contour intervals for spatial weights of all three variables are given by the color bar; temporal amplitudes range from ±0.3°C, ±30 kg m−1 s−1, and ±30 cm yr−1. Magnitudes are recovered by multiplying spatial weights by temporal amplitudes.

Lag sequences in Fig. 4 show SST weights south of Madagascar taking 2–3 yr to propagate eastward and equatorward across the tropical Indian Ocean in association with north branch of the ACW discussed in connection with Fig. 2. These SST weights reach Indonesia north of Australia at the same time as those taking 1–2 yr to propagate eastward across the tropical Indian Ocean from north of Madagascar. This latter eastward propagation was observed earlier by Tourre and White (1997) in covarying SST, MSW, zonal surface wind, and ocean heat storage anomalies between 10°S and 10°N. This was subsequently revealed as the regional manifestation of a global ENSO wave in covarying ST and SLP anomalies propagating eastward across the global tropical ocean (White and Cayan 2000) as discussed in connection with Fig. 2.

A major question is whether SST anomalies influence MF anomalies around Australia and whether the divergence of MF anomalies over Australia yields the corresponding APP anomalies. In the lag sequences in Fig. 4 a general correspondence can be seen between warm (cool) SST weights and poleward (equatorward) MMF weights, both colored yellow to red (blue), the latter displaced eastward of the former in their joint eastward propagation across the Indian Ocean. Notice that SST and MMF weights intensify together in lag months 12 and 36 as they approach Australia from the west. The source of this intensification is unknown, but it is associated with the onset of anomalous dry and wet periods in APP weights over Australia. Anomalous peak dry (wet) periods occur over the next 6–12 lag months (i.e., during lag months 18 and 42) as Australia is surrounded by covarying cool (warm) SST weights and equatorward (poleward) MMF weights.

To establish the significance of this we focus attention on anomalous peak dry and wet periods; that is, lag months 18 and 42 in Fig. 4. During these periods we overplot the streamlines of MF weights on top of SST weights (left, Fig. 5), on top of APP weights over Australia (middle, Fig. 5) and on top of MF convergence weights over Australia (right, Fig. 5). These overplots find poleward (equatorward) MF weights over warm (cool) SST weights in the subtropical Indian and South Pacific Oceans adjacent to Australia associated with convergent (divergent) MF weights over Australia. Streamlines of anomalous MF over Australia during the peak wet period originate from the South Pacific Ocean while those during the peak dry period originate from the Southern and Indian Oceans. Peterson and White (1998) found precipitable water in the air column over cool (warm) SST anomalies to be below (above) normal, indicating that air masses flowing over cool (warm) SST anomalies take up and carry less (more) moisture than normal. Thus oceanic air masses that flow over Australia during the anomalous peak dry (wet) period transport less (more) moisture than normal from their transit over cool (warm) SST anomalies that surround Australia.

Fig. 5.

(top) Two maps of MF streamline weights overlaid (left) on SST weights, (middle) on APP weights, and (right) on MF convergence weights, corresponding to maps at 18 lag months in EEOF lag sequences in Fig. 4. (bottom) Same as above but corresponding to maps at 42 lag months in EEOF lag sequences in Fig. 4. These maps represent anomalous peak dry and wet years over Australia contained in the dominant EEOF mode of APP anomalies. Contour intervals are given by the color bar, with yellow and red (blue) colors indicating positive (negative) weights. These maps display only spatial weights, indicating spatial phasing between variables.

Fig. 5.

(top) Two maps of MF streamline weights overlaid (left) on SST weights, (middle) on APP weights, and (right) on MF convergence weights, corresponding to maps at 18 lag months in EEOF lag sequences in Fig. 4. (bottom) Same as above but corresponding to maps at 42 lag months in EEOF lag sequences in Fig. 4. These maps represent anomalous peak dry and wet years over Australia contained in the dominant EEOF mode of APP anomalies. Contour intervals are given by the color bar, with yellow and red (blue) colors indicating positive (negative) weights. These maps display only spatial weights, indicating spatial phasing between variables.

5. Predicting Australian precipitation from eastward propagation of the ACW in the Southern and Indian Oceans

The APP lag sequence in Fig. 4 explains 57% of the interannual APP variance over the nominal 40-yr record. Yet it contains little propagation, restricting its utility in predicting interannual APP anomalies beyond what can be achieved through persistence. The time sequence of amplitudes modifying this APP lag sequence, however, is correlated at 0.90 with that modifying the SST lag sequence, the latter containing prominent eastward propagation west, north, and south of Australia. This eastward propagation in the SST lag sequence should allow us to predict the behavior of covarying SST and APP anomalies beyond what can be achieved by persistence. For example if we see warm or cool SST anomalies in the western Indian Ocean both north and south of Madagascar, we can expect from the EEOF analyses in Figs. 2–4 that they will propagate into the vicinity of Australia 2–3 yr later and influence APP anomalies in the manner described in Fig. 5. This predictability of both SST and APP can be formalized in a statistical climate prediction system that matches the 4-yr SST lag sequence in Fig. 4 to the observed time sequence of SST anomalies over the Indian Ocean for any 2-yr period, subsequently using the remaining 2 yr of the lag sequence to predict what will happen to SST and APP anomalies in the two succeeding years.

In the construction of this statistical climate prediction system we must train the system; that is, we have to find some portion of the Indian Ocean and Southern Ocean where the best fit (in the least squares sense) between the SST weights in the lag sequence and the SST anomalies in each 2-yr time sequence yields the best hindcast of interannual APP anomalies over Australia on the whole during the nominal 40-yr record. To accomplish this task we plot the cross correlation between the 1-yr hindcast of APP anomalies and observed APP anomalies over the nominal 40-yr record as a function of displacing the fit domain to each SST grid point surrounding Australia (Fig. 6). Through trial and error we find a fit domain of 20° lat and 60° long providing the highest correlations. This process finds the 1-yr hindcast of interannual APP anomalies over Australia to be maximized from two fit domains. One fit domain occurs south of Australia (30°–50°S, 90°–150°E) where the eastward propagation of SST anomalies associated with the ACW in the Southern Ocean allows APP anomalies to be hindcasted (right, Fig. 7). The other fit domain occurs west of Australia (10°–30°S, 60°–120°E) where equatorward and eastward propagation of SST anomalies associated with the north branch of the ACW allows the APP anomalies poleward to be hindcasted (left, Fig. 7). In Fig. 7 correlations over extratropical Australia poleward of 20°S are as high as 0.6–0.7, explaining as much as 36%–49% of the interannual variance on the 1.83° grid over the nominal 40-yr record. Correlations for the 1-yr hindcast are nearly the same as those for the nowcast, with little diminution occurring for the 2-yr hindcast. This is better than with hindcasts computed for New Zealand winter precipitation (White and Cherry 1999), where the 1-yr hindcast was statistically significant but the 2-yr hindcast was not.

Fig. 6.

Distribution of average correlation that fit domains have in hindcasting observed APP anomalies over Australia as a whole at a 1-yr lead. The standard-fit domain constitutes a rectangle extending over 20° of latitude and 60° of longitude. This standard-fit domain was sequentially made to occupy every SST gridpoint location in the geographic area of interest. Correlations greater than 0.4 are hatched, indicating where the statistical climate prediction system yields a significant hindcast of APP anomalies at the 95% confidence level for 20 degrees of freedom over the nominal 40-yr record (Snedecor and Cochran 1980).

Fig. 6.

Distribution of average correlation that fit domains have in hindcasting observed APP anomalies over Australia as a whole at a 1-yr lead. The standard-fit domain constitutes a rectangle extending over 20° of latitude and 60° of longitude. This standard-fit domain was sequentially made to occupy every SST gridpoint location in the geographic area of interest. Correlations greater than 0.4 are hatched, indicating where the statistical climate prediction system yields a significant hindcast of APP anomalies at the 95% confidence level for 20 degrees of freedom over the nominal 40-yr record (Snedecor and Cochran 1980).

Fig. 7.

(left) Distributions of correlation at each grid point over Australia between observed and predicted APP anomalies, the latter estimated from the statistical climate prediction system using the best-fit domain in the Indian Ocean, the location of which is given in the lowest panel. Distributions are given for the nowcast, the 1-yr hindcast, and the 2-yr hindcast. (right) Same as at left but obtained using the best-fit domain in the Southern Ocean, the location of which is given in the lowest panel. Hatched areas indicate where correlations are greater than 0.4, significant at the 95% confidence interval for 20 degrees of freedom over the nominal 40-yr record (Snedecor and Cochran 1980).

Fig. 7.

(left) Distributions of correlation at each grid point over Australia between observed and predicted APP anomalies, the latter estimated from the statistical climate prediction system using the best-fit domain in the Indian Ocean, the location of which is given in the lowest panel. Distributions are given for the nowcast, the 1-yr hindcast, and the 2-yr hindcast. (right) Same as at left but obtained using the best-fit domain in the Southern Ocean, the location of which is given in the lowest panel. Hatched areas indicate where correlations are greater than 0.4, significant at the 95% confidence interval for 20 degrees of freedom over the nominal 40-yr record (Snedecor and Cochran 1980).

Spatial averaging of APP anomalies over larger regions decreases noise in the statistical climate prediction system and produces better hindcasts than on the 1.83° grid in Fig. 7. So we choose four specific regions that represent major population centers in Australia. These regions are defined by a 7.5° lat × 7.5° long box centered on Perth in Western Australia, centered halfway between Melbourne in Victoria and Sydney in New South Wales, centered halfway between Brisbane and Townsville in Queensland, and centered on Darwin in the Northern Territory. This allows us to plot each 2-yr hindcast of APP anomalies on top of the observed time sequence of APP anomalies over the nominal 40-yr record for each region for each fit domain (not shown). We find the 2-yr hindcasts overlaying the observed APP anomalies during most years in the Western Australia region and in the Victoria and New South Wales region for both fit domains. Good hindcast skill is not restricted to El Niño or La Niña years. Less hindcast skill overall is indicated for Northern Territory and Queensland for both fit domains presumably because these tropical portions of Australia are more influenced by ENSO variability from the tropical ocean to the north.

We quantify these hindcast results in two ways. We begin by plotting time sequences of observed regional APP anomalies together with those of 1- and 2-yr hindcasts for each fit domain (Figs. 8a,b). This yields hindcasts for all four regions that correlate significantly (i.e., greater than 0.40 at the 95% confidence level) with observed precipitation anomalies, nearly equally for 1- and 2-yr leads. Correlations for New South Wales and Victoria and for Western Australia range from 0.72 to 0.81 and 0.67 to 0.79, respectively, with hindcasts from the Indian Ocean fit domain doing slightly better than from the Southern Ocean fit domain. These hindcasts explain 45%–65% of the interannual variance over the nominal 40-yr record. Correlations for Queensland and the Northern Territory range from 0.35 to 0.56 and 0.43 to 0.46, respectively, with hindcasts from the Southern Ocean fit domain doing slightly better than from the Indian Ocean fit domain. These hindcasts explain 12%–22% of the interannual variance over the nominal 40-yr record. Magnitudes averaged over each of the four regions range from 5 to 15 cm per year, which is similar to the rms of interannual APP anomalies at individual grid points estimated from Fig. 1 but a factor of 2 or 3 smaller than peak interannual variability expected at individual grid points.

Fig. 8.

(a) Time sequences of 1- and 2-yr hindcasts of APP anomalies conducted each year from 1960 to 1995 superimposed upon the time sequence of observed APP anomalies averaged over four different regions around Australia; that is, for Western Australia, Northern Territory, Victoria and New South Wales, and Queensland. These hindcasts and forecasts are based upon the best-fit domain in the Indian Ocean (see Fig. 7). In each region the heavy line represents observed interannual APP anomalies; light and dashed lines represent 1- and 2-yr hindcasts, respectively. Temporal correlations between hindcasts and observed APP anomalies are displayed in the upper right-hand corner for each region.

Fig. 8.

(a) Time sequences of 1- and 2-yr hindcasts of APP anomalies conducted each year from 1960 to 1995 superimposed upon the time sequence of observed APP anomalies averaged over four different regions around Australia; that is, for Western Australia, Northern Territory, Victoria and New South Wales, and Queensland. These hindcasts and forecasts are based upon the best-fit domain in the Indian Ocean (see Fig. 7). In each region the heavy line represents observed interannual APP anomalies; light and dashed lines represent 1- and 2-yr hindcasts, respectively. Temporal correlations between hindcasts and observed APP anomalies are displayed in the upper right-hand corner for each region.

We also quantify hindcast results by computing the correlation of hindcasted time sequences with those observed at 0-, 6-, 12-, 18-, and 24-month leads over nominal 40-yr record for New South Wales and Victoria, Queensland, Western Australia, and the Northern Territory (Fig. 9). Each plot is superimposed onto that of the correlation between the observed time sequence and that displaced by these same numbers of months, yielding predictability to be gained from persistence. Comparing these plots finds persistence yielding significant correlation (i.e., 0.30 at the 90% confidence level) at leads of 6–9 months, while the statistical climate prediction system extends this to 2 yr and more for all four regions. But correlations at 1- and 2-yr leads are much better for Western Australia and for Victoria and New South Wales regions (i.e., 0.6–0.7) than for Queensland and the Northern Territory (i.e., 0.3–0.5). The former indicates that as much as 50% of the total interannual variance in these former regions can be predicted at 1- and 2-yr leads.

Fig. 9.

Temporal correlation between hindcasted and observed time sequences of winter APP anomalies for Western Australia, Northern Territory, Queensland, and New South Wales and Victoria given in Figs. 8a and 8b at lead times of 0, 6, 12, 18, and 24 months, yielding predictability of the statistical climate prediction system. Also displayed are temporal correlations between time sequences of observed APP anomalies with themselves at these different lead times, yielding predictability attributable to persistence. Dashed lines of 0.30 indicate the 90% confidence level for 20 degrees of freedom over the nominal 40-yr record (Snedecor and Cochran 1980).

Fig. 9.

Temporal correlation between hindcasted and observed time sequences of winter APP anomalies for Western Australia, Northern Territory, Queensland, and New South Wales and Victoria given in Figs. 8a and 8b at lead times of 0, 6, 12, 18, and 24 months, yielding predictability of the statistical climate prediction system. Also displayed are temporal correlations between time sequences of observed APP anomalies with themselves at these different lead times, yielding predictability attributable to persistence. Dashed lines of 0.30 indicate the 90% confidence level for 20 degrees of freedom over the nominal 40-yr record (Snedecor and Cochran 1980).

One can imagine that had we chosen to construct a statistical climate prediction system for interannual APP anomalies over different regions of Australia, rather than over Australia as a whole, then Fig. 6 would yield best-fit domains different from those in Fig. 7. For example had we sought a fit domain that correlates best with year-to-year fluctuations in APP over the Northern Territory, Fig. 6 probably would have shown this best-fit domain located in the tropical ocean north of Australia, indicating that year-to-year fluctuations in the Northern Territory precipitation depend upon the slow eastward propagation of the global ENSO wave.

6. Discussion and conclusions

At present the prediction of precipitation over Australia is being conducted in a variety of ways. The Australia Bureau of Meteorology and the Queensland Department of Primary Industries produce 60–90-day forecasts based upon the persistence of global SST anomaly patterns (Drosdowsky and Chambers 1998) and the phase of the Southern Oscillation index (Stone et al. 1996), respectively. The International Research Institute at the Scripps Institution of Oceanography and Lamont-Doherty Enviromental Observatory relies upon a two-tiered climate prediction system (Barnett et al. 1994): first a forecast of Pacific equatorial SST anomalies associated with El Niño is made with tropical ocean–atmosphere coupled models; this is followed by a nowcast of extratropical atmospheric circulation anomalies over the globe using an atmospheric general circulation model driven by the forecasted equatorial SST anomalies. One inherent difficulty in the two-tiered approach is that it yields little influence over the rest of the globe during years when El Niño and La Niña are absent. The other inherent difficulty is that it ignores competing influences arising from regional climate change phenomena occurring over the global ocean. Examples of the latter include the ACW in the Southern Ocean (White and Peterson 1996), the north branch of the ACW in the Indian Ocean (Peterson and White 1998), and the global ENSO wave in the tropical Indian, Pacific, and Atlantic Oceans (White and Cayan 2000). The propagating nature of covarying SST and SLP anomalies associated with these coupled SST waves depends upon local interactions between ocean and atmosphere operating on interannual timescales (e.g., White et al. 1998). As such they differ from the global standing wave in SLP variability (i.e., the Southern Oscillation) pumped by SST anomalies associated with El Niño and La Niña in the central and eastern equatorial Pacific Ocean (e.g., Webster 1994). These regional climate change phenomena have a significant influence upon year-to-year changes in precipitation over adjacent land areas, as demonstrated by White and Cherry (1999) for New Zealand and demonstrated here for Australia.

First, we demonstrated that interannual APP anomaly patterns are inextricably linked to covarying SST and MMF anomaly patterns over the surrounding oceans. We found peak wet (dry) APP anomalies occurring during years when MF anomalies are convergent (divergent) over Australia. The latter appears to derive from the net transport of anomalous moist (dry) marine air over Australia resulting from its transit over warm (cool) SST anomalies surrounding Australia during these years. Second, we demonstrated that covarying SST and MF anomaly patterns around Australia derive from the slow eastward propagation of the ACW in the Southern Ocean south of Australia (White and Peterson 1996), from the slow equatorward and eastward propagation of the north branch of the ACW in the Indian Ocean west of Australia (Peterson and White 1998), and from the slow eastward propagation of the global ENSO wave in the tropical ocean north of Australia (White and Cayan 2000). We utilized the eastward propagation of interannual SST anomalies and its nearly one-to-one relationship with year-to-year changes in APP in the development of a statistical climate prediction system for the latter. We operated this prediction system in hindcast mode over the nominal 40-yr record, establishing that interannual APP anomaly patterns poleward of 20°S respond about equally to the slow eastward propagation of the ACW in the Southern Ocean between 30° and 50°S and the slow northeastward propagation of the north branch of the ACW in the Indian Ocean between 10° and 30°S. When we examined this predictability by region we found more than 50% of the total interannual APP variance in Victoria and New South Wales and in Western Australia poleward of 20°S able to be hindcasted at 1- and 2-yr leads.

We operated this statistical climate prediction system in hindcast mode assuming the statistics to be stationary over the nominal 40-yr record. Thus data and statistics were not independent. Such independence could have been achieved had we applied the statistics to an independent SST dataset outside the nominal 40-yr record. We were prevented from doing this by the lack of additional data and by the relatively few numbers of degrees of freedom: less than or equal to 20. Inspection of individual hindcasts in Figs. 8a and 8b indicates that the system is relatively robust over the 40-yr record, but this needs to be examined in more detail. Australia is already noted for containing epochs in which the phase relationship between APP anomalies and the SOI alternates between significance and insignificance (Simmonds and Hope 1997; Allan 2000). So the present hindcast study holds out the promise of a useful statistical forecast but does not deliver it; this will require further predictability studies involving monthly and seasonal anomalies and much longer records. The predictability study conducted here on interannual timescales (i.e., 3–6-yr period scales) needs to be extended to biennial, decadal, and interdecadal period scales (White and Cayan 2000). This effort will need to consider the apparent nonlinear modulation of interannual climate change by decadal and interdecadal variability (Power et al. 2000).

Fig. 8.

(Continued) (b) Same as in (a) but for hindcasts based on the best-fit domain in the Southern Ocean (see Fig. 7).

Fig. 8.

(Continued) (b) Same as in (a) but for hindcasts based on the best-fit domain in the Southern Ocean (see Fig. 7).

These results are consistent with the thesis of White and Peterson (1996), Jacobs and Mitchell (1996), Qui and Jin (1997), White et al. (1998), Peterson and White (1998), White and Cherry (1999), and Talley (1999) that covarying SST and SLP anomalies in the Southern Hemisphere are maintained against dissipation on interannual timescales by coupling between extratropical ocean and atmosphere. It follows that covarying SST and SLP anomalies should influence interannual climate change over the adjacent continents through attending atmospheric planetary wave dynamics. Here we demonstrate that APP anomalies arise from the divergence of MF anomalies occurring in response to the propagation of covarying SST and SLP anomalies around Australia by the ACW in the Southern Ocean south of Australia, the north branch of the ACW in the Indian Ocean west of Australia, and the global ENSO wave in the tropical ocean north of Australia. What is missing is understanding of the physics of ocean–atmosphere coupling that explains these regional climate change phenomena and the physics of their interaction with the continental landmass of Australia. This must come from diagnoses of global coupled ocean–atmosphere–terrestrial models that simulate these global and regional climate change phenomena. Ultimately, forecasting APP anomalies must derive from the integration of such a model forward from specified initial conditions, yielding precipitation anomalies over Australia as intrinsic aspects of both global and regional coupling dynamics.

Acknowledgments

This research was supported by the Office of Global Programs of NOAA (NOAA NA37GP0372) in concert with Experimental Climate Prediction Center at Scripps. Warren White is also supported by the National Science Foundation, Division of Ocean Sciences (OCE-9633474), and by the National Aeronautics and Space Administration (NASA) under Contract NA27GPO-539. Our thanks extend to Ted Walker, who conducted the statistical analyses presented in this study, and to Andrea Fincham, who helped construct the figures. We appreciate reviews and comments made by Roger Stone and Ian Simmonds.

REFERENCES

REFERENCES
Allan, R. J., 2000: ENSO and climatic variability in the last 150 years. El Niño and the Southern Oscillation: Multiscale Variability, Global and Regional Impacts, H. F. Diaz and V. Markgraf, Eds., Cambridge University Press, in press.
Andersen, N., 1974: On the calculation of filter coefficients for maximum entropy spectral analysis. Geophysics, 39, 69–72.
Barnett, T. P., and Coauthors, 1994: Forecasting global ENSO-related climate anomalies. Tellus, 46A, 381–397.
Chen, S.-C., C. L. Norris, and J. Roads, 1996: Balancing the atmospheric hydrologic budget. J. Geophys. Res., 101, 7341–7358.
Drosdowsky, W., and L. Chambers, 1998: Near global sea surface temperature anomalies as predictors of Australian seasonal precipitation. BMRC Research Rep. No. 65, Bureau of Meteorology, Australia.
Emery, W. J., and R. E. Thomson, 1996: Data Analysis Methods in Physical Oceanography. Elsevier, 634 pp.
Hulme, M., and P. D. Jones, 1993: A historical monthly precipitation data set for global land areas: Applications for climate monitoring and climate model evolution. In: Analysis methods of precipitation on a global scale. WMO/TD-No. 558, Geneva.
Jacobs, G. A., and J. L. Mitchell, 1996: Ocean circulation variations associated with the Antarctic Circumpolar Wave. Geophys. Res. Lett., 23, 2947–2950.
Kalnay, E., 1996: Letter to Dr. Abraham Oort. NCEP/NCAR CDAS/Reanalysis Project. [Available online at http://wesley.wwb.noaa.gov/reanalysis.html.]
.
——, and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77, 437–471.
Kaylor, R. E., 1977: Filtering and decimation of digital time series. Institute of Phys. Sci. and Tech., University of Maryland, College Park, Tech. Rep. Note BN 850, 14 pp.
Nicholls, N., 1989: Sea surface temperature and Australian winter precipitation. J. Climate, 2, 965–973.
Partridge, I. J., 1994: Will it rain? The effect of the Southern Oscillation and El Niño on Australia. 2d ed. Dept. of Primary Industries, Brisbane, Australia, 56 pp.
Peterson, R. G., and W. B. White, 1998: Slow oceanic teleconnections linking the Antarctic Circumpolar Wave with tropical ENSO. J. Geophys. Res., 103, 24 573–24 583.
Power, S., T. Casey, C. Folland, A. Coleman, and V. Mehta, 2000: Interdecadal modulation of the impact of ENSO upon Australia climate. Climate Dyn., in press.
Preisendorfer, R. W., and C. D Mobley, 1988: Principal Component Analysis in Meteorology and Oceanography. Elsevier, 425 pp.
Qui, B., and F.-F. Jin, 1997: Antarctic Circumpolar Wave: An indication of ocean–atmosphere coupling in the extratropics. Geophys. Res. Lett., 24, 2585–2588.
Simmonds, I., and P. Hope, 1997: Persistence characteristics of Australia rainfall anomalies. Int. J. Climatol., 17, 597–613.
Snedecor, G. W., and W. G. Cochran, 1980: Statistical Methods. Iowa State University Press, 507 pp.
Stone, R. C., G. L. Hammer, and T. Marcussen, 1996: Prediction of global precipitation probabilities using phases of the Southern Oscillation index. Nature, 354, 252–255.
Sturman, A., and N. Tapper, 1996: The Weather and Climate of Australia and New Zealand. Oxford University Press, 476 pp.
Talley, L., 1999: Simple coupled midlatitude climate models. J. Phys. Oceanogr., 8, 2016–2037.
Tourre, Y. M., and W. B. White, 1995: ENSO signals in global upper-ocean temperature. J. Phys. Oceanogr., 25, 1317–1332.
——, and ——, 1997: Evolution of the ENSO signal over the Indo-Pacific domain. J. Phys. Oceanogr., 27, 683–696.
van Loon, H., and D. J. Shea, 1985: The Southern Oscillation. Part IV: The precursors south of 15°S to the extremes of the oscillation. Mon. Wea. Rev., 113, 2063–2074.
Weare, B., and J. Nasstrom, 1982: Examples of extended empirical orthogonal function analysis. Mon. Wea. Rev., 110, 481–485.
Webster, P. J., 1994: The role of hydrological processes in ocean–atmosphere interactions. Rev. Geophys., 32, 397–476.
White, W. B., and R. Peterson, 1996: An Antarctic circumpolar wave in surface pressure, wind, temperature, and sea ice extent. Nature, 380, 699–702.
——, and N. J. Cherry, 1999: Influence of the Antarctic Circumpolar Wave upon New Zealand temperature and precipitation during autumn–winter. J. Climate, 12, 960–976.
——, and D. R. Cayan, 2000: A global ENSO wave in surface temperature, pressure, and wind, and its interdecadal modulation from 1900 to 1999. J. Geophys. Res., in press.
——, S.-C. Chen, and R. Peterson, 1998: The Antarctic Circumpolar Wave: A beta effect in ocean–atmosphere coupling over the Southern Ocean. J. Phys. Oceanogr., 28, 2345–2361.

Footnotes

Corresponding author address: Dr. Warren B. White, Scripps Institution of Oceanography, 9500 Gilman Dr., La Jolla, CA 92093-0230.