La Niña de Australia — contemporary and palaeo-hydrology of Lake Eyre

Deep-Sea Research II 50 (2003) 2331–2347

Protracted’ ENSO episodes and their impacts in the Indian Ocean region R.J. Allana,*, C.J.C. Reasonb, J.A. Lindesayc, T.J. Anselld a Met Office, Hadley Centre for Climate Prediction and Research, Bracknell RG12 2SY, UK EGS & Oceanography Depts., University of Cape Town, Private Bag, Rondebosch 7701, South Africa c Education Manager, CRC for Greenhouse Accounting, GPO Box 475, Canberra ACT 2601, Australia d School of Earth Sciences, University of Melbourne, Parkville, Vic. 3052, Australia

b

Abstract Evidence for a quasi-decadal (9–13 yr) signal interacting with the quasi-biennial (QB) (2–2.5 yr) and low-frequency (LF) (2.5–8 yr) components of the El Nin˜o Southern Oscillation (ENSO) phenomenon to produce both ‘protracted’ El Nin˜o and La Nin˜a episodes is examined using historical patterns of climatic variables over the Indian Ocean basin in the global context. Across the Indian Ocean basin, seasonal rainfall correlations on LF ENSO and quasi-decadal timescales show that the lower frequency signal displays ‘ENSO-like’ characteristics. Thus, there is evidence that the quasi-decadal signal alone can provide an important modulation of Indian Ocean climatic patterns. However, examinations of the Indian Ocean situation in the wider global context provide strong evidence that ‘protracted’ El Nin˜o and La Nin˜a episodes result from interactions between QB, LF and quasi-decadal signals in the climate system. The seasonal tropospheric velocity potential correlations show that QB, LF and quasi-decadal changes in mass overturning and convective regimes are of global extent. These mass overturnings are most coherent on LF ENSO timescales, but are also clearly evident in seasonal QB and quasi-decadal signal sequences. The situation in October– December (OND) also suggests that a propensity exists in the climate system for drought or flooding extremes to occur over central-eastern Africa to central Asia on QB to decadal timescales, and that this may be particularly manifest during ‘protracted’ El Nin˜o and La Nin˜a episodes. An investigation of individual ‘protracted’ El Nin˜o and La Nin˜a episodes reveals that the magnitude and phasing of the QB, LF and quasi-decadal signals can lead to periods during ‘protracted’ El Nin˜o (La Nin˜a) episodes, which start to take on ‘La Nin˜a-like’ (‘El Nin˜o-like’) characteristics, before reverting back again. This phasing nature involves opposing combinations of the above signals, especially on QB and LF ENSO timescales. Ultimately, it can lead to modulations of climatic variables that are different from those usually observed over ‘classical’ ENSO-sensitive regions. Such characteristics may have major consequences for global teleconnection patterns, and thus have the potential to impact on both regional and remote climatic signatures. Crown Copyright r 2003 Published by Elsevier Science Ltd. All rights reserved.

*Corresponding author. Tel.: +44-1344-856904. E-mail address: rob.allan@metoffice.com, rjallan@metoffice.com (R.J. Allan). 0967-0645/03/$ - see front matter Crown Copyright r 2003 Published by Elsevier Science Ltd. All rights reserved. doi:10.1016/S0967-0645(03)00059-6

2332

R.J. Allan et al. / Deep-Sea Research II 50 (2003) 2331–2347

1. Introduction Research into natural climatic variability has grown rapidly in the last 5–10 yr in tandem with concerns about climatic change and global warming. There is now growing evidence that natural fluctuations in the climate system, with distinct dynamics, operate on decadal–multidecadal time scales (Navarra, 1999; Reason, 2000; Barlow et al., 2001; Tourre et al., 2001), including across the Indian Ocean region (Allan et al., 1995; Reason and Rouault, 2002). It also has become apparent that interannual climatic phenomena, such as ENSO, respond not only to their own dynamics but may well be themselves modulated by what are being revealed as distinct, lower frequency decadal–multidecadal modes of climatic variability (Gershunov and Barnett, 1998; Power et al., 1999a,b; Navarra, 1999; Tourre et al., 2001). In an earlier paper by Reason et al. (2000), an examination was made of the evolution of strong, sharp El Nin˜o and La Nin˜a events across the Indian Ocean basin in a global context. That study noted the presence of distinct ENSO episodes of longer duration than the more ‘classical’ interannual events found in the climate record. These episodes, extending over three or more years, were equated with ‘protracted’ El Nin˜o and La Nin˜a sequences as described in Allan and D’Arrigo (1999). The role that such long ENSO episodes play in modulating climatic patterns has yet to be quantified fully across the Pacific Ocean basin, let alone the Indian Ocean domain. However, ‘protracted’ El Nin˜o conditions during the first half of the 1990s and the presence of a ‘protracted’ La Nin˜a event from May 1998 to the end of 2001 have heightened scientific efforts to understand this phenomenon and its impacts. The climatic signature of these recent ‘protracted’ El Nin˜o and La Nin˜a episodes indicates that they cannot be characterised as simply longer lived mirror images of the more ‘classical’ interannual El Nin˜o and La Nin˜a events (Allan and D’Arrigo, 1999; Barlow et al., 2002), but involve the presence of a quasidecadal ‘ENSO-like’ signal (Allan, 2000). An understanding of these longer-lived phenomena is thus crucial if we are to incorporate fully and

correctly their nature and dynamics into our existing climate prediction schemes. As a consequence of the above findings, this study aims firstly to examine the lower frequency quasi-decadal signal in the climate system, which seems to play a role in the occurrence and nature of ‘protracted’ El Nin˜o and La Nin˜a episodes, and which has a significant expression and impact across the Indian Ocean basin. Particular emphasis is then focused on the interplay between quasibiennial (QB), ‘classical’ interannual ENSO (LF) and quasi-decadal features of the climate system over the Indian Ocean in the global context. Although the focus of this Special Issue is on the Indian Ocean, it is necessary to consider both the Indian and Pacific oceans in this study given that ENSO and the ‘ENSO-like’ decadal modes are essentially Indo-Pacific phenomena. This broader spatial perspective, using near-global canonical QB, LF and quasi-decadal patterns and correlation fields, is balanced by an examination of the interactions between such signals during each of the individual, strong ‘protracted’ El Nin˜o and La Nin˜a episodes in the historical record.

2. Data and methods A monthly gridded global temperature data set has been created by merging the UK Hadley Centre’s sea-Ice and Sea-Surface Temperature (HadISST) data compilation (Rayner et al., 2003) with the UK Climatic Research Unit’s land-surface air-temperature data set (Parker et al., 1995). This 5 latitude by 5 longitude surface temperature data set is corrected over land and ocean for local changes in variance due to varying data coverage (Jones et al., 2001) and filled using Reduced Space Optimum Interpolation (RSOI) (Kaplan et al., 1997, 2000) to give globally complete coverage from 1870–2000. The RSOI technique uses the characteristic geographical patterns of data anomalies to interpolate areas of missing data, and gives more (less) weight to data with small (large) likely errors. This version of the data is designated as HadCRUTv (OI). The Hadley Centre’s global Mean Sea Level Pressure (MSLP) compilation is also used in this

R.J. Allan et al. / Deep-Sea Research II 50 (2003) 2331–2347

paper. Gridded MSLP data on 5 latitude by 5 longitude resolution are filled using RSOI to produce the HadSLP (OI) (Basnett and Parker, 1997) data compilation on a monthly basis over a regular 5 latitude by 5 longitude grid. The common period of HadCRUTv (OI) and HadSLP (OI) determined for this study is from 1871 to 1998. For ease in examining broad scale climate patterns, both data sets are interpolated onto a regular 10 latitude by 10 longitude grid. These data are normalised with respect to their means and standard deviations for the period 1961–1990. Spatial distributions of land, island and ship observations used to construct HadCRUTv (OI) and HadSLP (OI) are published in the literature (Allan et al., 1996; Rayner et al., 1996). Allan et al. (1995, 1996), Rayner et al. (1996) and Jones et al. (2001) have provided thorough discussions of both the quality of the MSLP and SST data sets and the density of data sampling over this century for the Indian Ocean region, while Harrison and Larkin (1996) and Larkin and Harrison (2001, 2002) have explored data compilations used in the construction of HadCRUTv (OI) and HadSLP (OI) in analyses of both El Nin˜o and La Nin˜a over the periods 1946–1993 and 1946–1995, respectively. Based on these analyses, the above data sets are amongst the best that are currently available. The monthly global land and island rainfall data set on the KNMI Climate Explorer (http:// climexp.knmi.nl/) WWW site, derived originally from Hulme (1992) and updated (1900–1998) on the Climatic Research Unit (CRU), University of East Anglia’s WWW site (http://www.cru.uea. ac.uk/Bmikeh/datasets/global/), constitutes the major precipitation set used in this study. Seasonal correlation analyses are performed on data sets of Hulme precipitation plus 0.995 and 0.2101 sigma level velocity potential (1958–1996) on the KNMI Climate Explorer (http://climexp.knmi.nl/) and NOAA CDC correlation (http://www.cdc.noaa.gov/Correlation/) WWW sites, respectively. ‘Protracted’ El Nin˜o and La Nin˜a episodes can be seen in the trace of the Southern Oscillation Index (SOI) in Fig. 1 (where the data are smoothed with an 11-point running mean) and are tabulated along with the strong ‘classical’ El Nin˜o and La

2333

Nin˜a events detailed in Reason et al. (2000) and in Table 1. This is done in order to provide some continuity between the work, while indicating in a simple and clear manner that ‘protracted’ and ‘classical’ ENSO designations display both similarities and differences. For instance, the basic classification of ‘classical’ ENSO events into El Nin˜o-1 (year before the event), El Nin˜o (year of the event), La Nin˜a-1 (year before the event) and La Nin˜a (year of the event) designations is not readily applicable to ‘protracted’ episodes. Accordingly, composites of ‘protracted’ ENSO episodes referred to later in this study (but not shown) differ from the event composites in Reason et al. (2000), where mean seasonal patterns were produced specifically for each of the El Nin˜o-1, El Nin˜o, La Nin˜a-1 and La Nin˜a designations. An inspection of the spatial and temporal patterns of MSLP and SST during the evolution of various ENSO events and episodes in Allan et al. (1996) also shows these characteristics quite clearly. In order to address the above issues, seasonal anomalies of the data variables discussed earlier (from their various base periods) are constructed by a simple compilation of their monthly values. Seasons are defined as January–March (JFM), April–June (AMJ), July–September (JAS) and October–December (OND). Joint Empirical Orthogonal Function (EOF) time series of HadCRUTv (OI) and HadSLP (OI), band-pass filtered in each of the significant 2–2.5 yr (OB), 2.5–8 yr (LF) and 9–13 yr (quasi-decadal) bands from Allan (2000), are used in correlation analyses to investigate the nature of the ‘protracted’ El Nin˜o and La Nin˜a episodes noted in Table 1. These bands noted above have also been resolved using a multitaper method singular value decomposition (MTM-SVD) technique (Mann and Park, 1999), which generated a joint local fractional variance (LFV) spectrum of the above data sets (see Fig. 2). All spatial correlation fields in Figs. 3 and 4 are presented together with fields of their statistical point significance above the 90% level using the grid point t test on the KNMI Climate Explorer WWW site (http://climexp.knmi.nl/). In Figs. 5 and 6, correlations calculated on the NOAA CDC correlation WWW site (http://www.cdc.noaa.gov/

2334

R.J. Allan et al. / Deep-Sea Research II 50 (2003) 2331–2347

Fig. 1. Monthly SOI values from January 1866 until February 2002 slightly smoothed with an 11-point running average (except for end points). Periods of more El Nin˜o conditions are shaded in red/lightly shaded and of more La Nin˜a conditions in blue/darkly shaded.

Correlation/), with values exceeding + or are locally significant at the 95% level.

0.32

3. Seasonal correlations with the quasi-decadal signal across the Indian Ocean basin The trace of the SOI in Fig. 1 indicates that El Nin˜o and La Nin˜a events vary in magnitude, onset and cessation times, and in duration. This characteristic is observed in all historical indices of ENSO (e.g., Nin˜o 3, 3.4 and 4 region SSTs), indicating that the phenomenon encapsulates a ‘family’ of events including around four major long El Nin˜o (1894–1897, 1911–1914, 1939–1942, 1990–1995) and seven major long La Nin˜a (1878– 1880, 1892–1894, 1909–1911, 1916–1918, 1954–

1956, 1973–1976, 1998–2001) sequences over the 136 yr-period in Fig. 1. As noted earlier, such long episodes have been described in Allan and D’Arrigo (1999) as ‘protracted’ El Nin˜o and La Nin˜a episodes and postulated as being more than just long sequences of ‘classical’ interannual El Nin˜o or La Nin˜a events. The longest ‘protracted’ ENSO episode in the instrumental record is the 1990–1995 El Nin˜o (Fig. 1 or Table 1). In their analysis of palaeo data back to 1706, Allan and D’Arrigo (1999) indicate that the longest ‘protracted’ El Nin˜o and La Nin˜a episodes in such records were of around 5 and 7 yr duration, respectively. Thus, a significant quasi-decadal signal operating in the climate system with ‘ENSO-like’ features and displaying the patterns of SST and atmospheric circulation seen in the

R.J. Allan et al. / Deep-Sea Research II 50 (2003) 2331–2347 Table 1 Years defined as ‘protracted’ ENSO episodes (in italics) (ENSO phase composites examined in Reason et al. (2000)are in bold El Nin˜o-1

El Nin˜o

La Nin˜a-1

La Nin˜a

1877 1888 1896 1899 1902 1905 1911 1913 1914 1918 1925 1930 1940 1941 1957 1963 1965 1972 1982 1986 1990 1991 1992 1993 1994

1878 1889 1897 1900 1903 1906 1912 1914 1915 1919 1926 1931 1941 1942 1958 1964 1966 1973 1983 1987 1991 1992 1993 1994 1995

1879 1886 1889 1892 1909 1916 1917 1924 1933 1938 1942 1949 1954 1955 1970 1973 1975 1988 1998 1999 2000

1880 1887 1890 1893 1910 1917 1918 1925 1934 1939 1943 1950 1955 1956 1971 1974 1976 1989 1999 2000 2001

2335

most recent ‘protracted’ ENSO episodes (Allan, 2000), would be the type of climatic signal most likely to produce such episodes in longer historical and palaeo data. The LFV spectrum generated by the MTM-SVD analyses of historical annual HadCRUTv (OI) and HadSLP (OI) data (Fig. 2), plus the studies detailed earlier of the 1990–1995 El Nin˜o and recent 1998–2001 La Nin˜a sequences, provide the basis for investigations of ‘protracted’ ENSO episodes. Fig. 2 reveals not only QB and LF climatic signals associated with the ‘classical’ interannual ENSO phenomenon over the IndoPacific basin, but also the presence of significant decadal–multidecadal signals in the climate system. Of these low-frequency fluctuations, a quasidecadal signal operating around 9–13 yr is most prominent, and its climatic signatures have been evident during both of the recent ‘protracted’ El Nin˜o and La Nin˜a episodes (Latif et al., 1995, 1997; Barlow et al., 2002). In the following subsection, the impacts and evolution of quasidecadal signal influences are examined through seasonal correlations of the 9–13 yr EOF time series with precipitation over the Indian Ocean domain.

JOINT MTM-SVD LFV SPECTRA of HadCRUTv(OI) & HadSLP(OI): DOMAIN 65N-35S: 1871-1998 PERIOD (YEARS)

16

0.9

8

4

2

ENSO 2.5-8 yr 0.8

QB 2-2.5 yr

64 yr 11.5 yr

0.7 LFV

16.8 yr

0.6

99% 95% 90%

0.5

50%

0.4 0

0.0625

0.125

0.1875

0.25

0.3125

0.375

0.4375

FREQUENCY (CYCLES/YEAR)

Fig. 2. Multitaper frequency-domain Singular Value Decomposition (MTM-SVD) localised variance spectrum (LFV) from a joint analysis of HadCRUTv (OI) and HadSLP (OI) from 1871 to 1998 (relative variance is explained by the first eigenvalue of the SVD as a function of frequency) over the domain 65 N–35 S. The 50%, 90% and 99% statistical confidence limits are shown as horizontal lines, and various significant climatic features in the spectrum are pointed out on the diagram.

2336

R.J. Allan et al. / Deep-Sea Research II 50 (2003) 2331–2347

R.J. Allan et al. / Deep-Sea Research II 50 (2003) 2331–2347

3.1. Seasonal rainfall correlations on LF ENSO and quasi-decadal time scales (1900–1998) Correlations are made firstly between each of the seasonal LF ENSO (2.5–8 yr) and quasidecadal (9–13 yr) ‘ENSO-like’ time series (the joint EOFs from Allan, 2000) with the seasonal Hulme rainfall data set from 1900 to 1998 (Figs. 3 and 4, respectively). Regions with correlations that are statistically significant above the 90% level of significance for the grid point t test are also shown in each figure. The seasonal LF ENSO correlations in Fig. 3 show the well-known and coherent relationships with rainfall in ‘ENSO-sensitive’ areas across the Indian Ocean domain. On quasi-decadal ‘ENSOlike’ time scales (Fig. 4), such seasonal correlations are weaker, but significant relationships are observed in the LF ‘ENSO-sensitive’ regions of Australia, southern and eastern Africa, China and the Indian subcontinent. In JFM in Fig. 4, much of southern Africa, parts of East Africa, the Indian subcontinent, and eastern Australia display significant correlations with the quasi-decadal signal. The southern African response is weaker by AMJ and is also confined to Namibia and western South Africa in this season (Fig. 4). Apart from a small area on the Indian subcontinent, the correlations in the other regions bordering the Indian Ocean during JFM have largely disappeared one season later. By JAS, the southern African correlations have changed sign, and do so again in OND (Fig. 4). The change in sign in the correlations over southern Africa during JAS are of no real consequence since only the far southwestern and extreme south coast regions get significant rainfall during this season. Fig. 4 also shows that some areas of statistically significant correlations have returned to parts of eastern Australia and are now evident over New Guinea in JAS. By OND, most of eastern Australia and areas of East Africa exhibit significant rainfall correlations. The eastern Australian region displays the most spatially

2337

coherent rainfall response to the quasi-decadal mode during any season in Fig. 4.

4. ‘Protracted’ El Nin˜o and La Nin˜a episodes: the Indian Ocean in the global context The distinct ‘sawtooth’ shapes of the SOI traces during a number of the ‘protracted’ ENSO episodes in Fig. 1 suggest that such sequences involve more than just a low frequency modulation of the climate system by the quasi-decadal signal. It would seem that interactions of higher frequency climatic features with the quasi-decadal signal may have the potential to cause the ‘sawtooth’ nature exhibited by a number of the ‘protracted’ episodes in the historical record (Fig. 1). These higher frequency signals may reflect the influence of QB and LF interannual components of ENSO. Accordingly, the basic elements of this interaction are explored in this section through analyses of seasonal correlations of lower and upper tropospheric velocity potential on QB, LF and quasi-decadal time scales. 4.1. Upper and lower tropospheric velocity potential correlations with QB, LF and quasi-decadal signals Seasonal correlations of global lower and upper tropospheric velocity potential fields with QB, LF and quasi-decadal signals (Figs. 5 and 6) all show similar global evolutions of large-scale mass overturning patterns and convective activity. However, there is an inherent change from hemispheric patterns involving fluctuations in deep convection and subsidence between broad Indo-Australasian and Pacific regions during the JFM to AMJ/JAS seasons, to a situation in OND where the IndoAustralasian node is centred on Australasia and a region in-phase with the Pacific occurs over East Africa/western Indian Ocean to central Asia. A similar pattern is found in examinations of the seasonal evolution of SSTs across the Indian

Fig. 3. Correlations, and their grid point significance, across the Indian Ocean domain between each of the seasonal LF ENSO (2.5– 8 yr) EOF band-pass filtered time series in Allan (2000) and the Hulme rainfall data set from 1900 to 1998 for the seasons (a) JFM, (b) AMJ, (c) JAS, and (d) OND. The seasonal significance diagrams show regions where correlations are statistically significant at and above the 90% level from the grid point significance t test on the KNMI Climate Explorer WWW site (http://climexp.knmi.nl/).

2338

R.J. Allan et al. / Deep-Sea Research II 50 (2003) 2331–2347

Fig. 4. As in Fig. 3, except for seasonal correlations between each of the quasi-decadal (9–13 yr) EOF time series in Allan (2000).

R.J. Allan et al. / Deep-Sea Research II 50 (2003) 2331–2347 Fig. 5. Global correlations between each of the seasonal QB, LF ENSO and quasi-decadal EOF time series in Allan (2000) and lower tropospheric velocity potential (1958–1996) from the NOAA CDC correlation WWW site (http://www.cdc.noaa.gov/Correlation/). No statistical significance is shown. 2339

2340 R.J. Allan et al. / Deep-Sea Research II 50 (2003) 2331–2347

Fig. 6. As in Fig. 5, except for upper tropospheric velocity potential.

R.J. Allan et al. / Deep-Sea Research II 50 (2003) 2331–2347

Ocean domain (Reason et al., 2000). Re-examinations of global precipitation and surface air temperature correlations with the Nin˜o 3 SST index on a seasonal basis by Dutch researchers (http://www.knmi.nl/onderzk/oceano/enso/effecten.html) indicate that a distinct precipitation signal emerges over central Asian–East African only in September–November. Although seasons in the Dutch research are not exactly aligned with those used in this paper, the rainfall response is consistent with the mass overturning changes seen in OND relative to the overturning patterns in JFM to JAS seasons (Figs. 5 and 6). This largescale change in overturning patterns is most apparent over the African–Indian Ocean sector, suggesting that there may be a propensity for central Asian–East African drought or flooding extremes to be inherent in the climate system. Such climatic extremes would be magnified during particularly strong interannual ENSO events and temporally extended under the influence of ‘protracted’ El Nin˜o and La Nin˜a episodes. Both Figs. 5 and 6 indicate that the strongest (weakest) mass overturnings associated with the velocity potential patterns occur on the QB time scale in JAS and OND (AMJ), at LF interannual ENSO frequencies throughout all seasons, and on the quasi-decadal time scale in the JFM, AMJ and OND (JAS) seasons. These characteristics suggest that of the various combinations of signals that may arise, there are certain seasons of the year in which there is a greater or lesser propensity for ENSO and ‘ENSO-like’ extremes to be manifest. There is also the possibility that the so-called ENSO ‘predictability’ barrier, which occurs around the boreal spring (austral autumn), may be naturally confined to that time of the year when the weakest interplay between the QB, LF and quasi-decadal signals tends to occur. Conversely, the JAS and OND seasons appear to coincide with the strongest interaction between these signals, and are thus most likely to display robust ENSO and ‘ENSO-like’ activity. These findings also raise a bigger question, which is beyond the scope of the current paper. Is the quasi-decadal signal a separate ‘ENSO-like’ fluctuation that interacts with QB and LF ENSO signals, or is it a part of the ENSO phenomenon itself?

2341

5. Examinations of individual strong ‘protracted’ El Nin˜o and La nin˜a episodes To illustrate aspects of the nature of individual strong ‘protracted’ El Nin˜o and La Nin˜a episodes, an examination is made of each episode’s individual QB, LF and quasi-decadal components generated by the joint EOF analysis detailed in Allan (2000). The diagrams in Figs. 7 and 8 show clearly the range of variations in the joint EOF time series on each of the above time scales during individual ‘protracted’ El Nin˜o (1894–1897, 1911– 1914, 1939–1942, 1990–1995) and La Nin˜a (1878– 1880, 1916–1918, 1954–1956, 1973–1976) episodes, respectively. As the joint EOF analysis above terminates in 1998, the recent 1998–2001 ‘protracted’ La Nin˜a sequence was not included in the first part of the assessment of various episodes. A particular feature of each of the individual ‘protracted’ episodes in Figs. 7 and 8 is that they are all influenced by various combinations and contributions from the three signals involved. Both ‘protracted’ El Nin˜o and La Nin˜a episodes are characterised by times during their existence when they appear to be terminating, only to recover again and often then be followed by several other periods of waxing and waning before they finally end. These intraepisode fluctuations result from periods of strong QB and/or LF signal ‘surges’, when more interannual ENSO characteristics dominate. This characteristic leads to the development of the ‘sawtooth’ pattern or signature often found in time series used to define ‘protracted’ ENSO episodes. The changing nature of the above signals over the historical record can readily be seen in wavelet or MTM-SVD evolutive spectral analyses of various standard ENSO indices or indicators (see Nin˜o 3, Nin˜o 4 and SOI wavelet results on http:// www.cgd.ucar.edu/Btorrence/interdec/wavelet others.html), and MTM-SVD evolutive spectra in Allan, 2000). Such analyses indicate that major ‘protracted’ ENSO episodes during the pre-1880s, 1900–1920s, 1940–1950s and post-1970s epochs occurred during epochs that were influenced by strong QB, LF and quasi-decadal signals, while during the 1920–1930s and 1960s much weaker climatic signals tended to prevail. Thus, strong

R.J. Allan et al. / Deep-Sea Research II 50 (2003) 2331–2347

2342

STRONG PROTRACTED EL NINO EPISODES 1894-1897

1911-1914

1914.8

1914.6

1914.3

1914.1

1913.8

1913.6

1913.3

1913.1

1912.8

9-13 YR

YEARS

YEARS

1990-1995

1939-1942 2-2.5 YR

1912.6

1912.3

2.5-8 YR

1912.1

1911.1

1897.8

1897.6

1897.3

1897.1

1896.8

1896.6

1896.3

-2 1896.1

-2

1895.8

-1

1895.6

-1

1895.3

0

1895.1

0

1894.8

1

1894.6

1

1894.3

2

1894.1

2

1911.8

2-2.5 YR

9-13 YR

1911.6

2.5-8 YR

1911.3

2-2.5 YR

2.5-8 YR

9-13 YR

2-2.5 YR

2

2

1

1

0

0

2.5-8 YR

9-13 YR

-1

-1

1995.6

1995.1

1994.6

1994.1

1993.6

1993.1

1992.6

1992.1

1991.6

1991.1

1990.6

1942.8

1942.6

1942.3

1942.1

1941.8

1941.6

1941.3

1941.1

1940.8

1940.6

1940.3

1940.1

1939.8

1939.6

1939.3

1939.1

YEARS

1990.1

-2

-2

YEARS

Fig. 7. Joint EOF time series of HadCRUTv (OI) and HadSLP (OI) data band pass filtered in each of the significant QB (2–2.5 yr), LF ENSO (2.5–8 yr) and the decadal (9–13 yr) bands for the individual strong ‘protracted’ El Nin˜o episodes in 1894–1897, 1911–1914, 1939–1942 and 1990–1995.

‘protracted’ episodes seem to require significant contributions from each of the QB, LF and quasidecadal signals. This did not occur during the 1920– 1930s and 1960s, and hence these epochs experienced no significantly strong ‘protracted’ ENSO episodes. Thus, it would seem that the general epochs of waxing and waning of QB and interannual ENSO signals in the historical record are also evident on quasi-decadal time scales. In fact, this common characteristic of all of these signals may be a further indication that the quasi-decadal signal is actually a low frequency part of ENSO. Overall, such findings illustrate how crucial it is to understand the various signals contributing to

longer sequences of ‘ENSO-like’ behaviour. This can be further appreciated by an examination of the major impacts of the 1990–1995 ‘protracted’ El Nin˜o and 1998–2001 ‘protracted’ La Nin˜a episodes. 5.1. The 1990–1995 ‘protracted’ El Nin˜o episode The situation in Australia during the 1990–1995 ‘protracted’ El Nin˜o episode, as noted in Allan and D’Arrigo (1999), is illustrative of the contrast between the impact of a protracted episode as opposed to a ‘classical’ El Nin˜o event. Throughout much of the 1990–1995 period, the state of Queensland experienced prolonged drought con-

R.J. Allan et al. / Deep-Sea Research II 50 (2003) 2331–2347

2343

STRONG PROTRACTED LA NINA EPISODES 1916-1918

YEARS

YEARS

1954-1956 2-2.5 YR

1918.6

1916.1

1880.8

1880.6

1880.3

1880.1

1879.8

1879.6

1879.3

-4

1879.1

-3

-4

1878.8

-2

-3

1878.6

-2

1878.3

-1

1878.1

0

-1

1918.3

1

0

1918.1

2

1

1917.8

3

2

1917.3

3

9-13 YR

2.5-8 YR

1917.1

4

1916.8

4

1916.6

2-2.5 YR

1917.6

9-13 YR

2.5-8 YR

1916.3

2-2.5 YR

1918.8

1878-1880

1973-1976 9-13 YR

2.5-8 YR

2-2.5 YR

4

4

3

3

2

2

1

1

0

0

-1

-1

-2

9-13 YR

2.5-8 YR

-2

-3

-3

YEARS

1975.8

1975.6

1975.3

1975.1

1974.8

1974.6

1974.3

1974.1

1973.8

1973.6

1973.3

-4

1973.1

1956.8

1956.6

1956.3

1956.1

1955.8

1955.6

1955.3

1955.1

1954.8

1954.6

1954.3

1954.1

-4

E YARS

Fig. 8. As in Fig. 7, except individual strong ‘protracted’ La Nin˜a episodes in 1878–1880, 1916–1918, 1954–1956 and 1973–1976.

ditions that were much like what would occur if a moderate ‘classical’ El Nin˜o event were made to persist for several years. However, over inland southeastern Australia which, together with inland Queensland, usually also experiences drought during ‘classical’ El Nin˜o events, there were many months in a number of years during the 1990–1995 situation when above average to near record breaking flooding events occurred. Most notable were the August to December 1992 and the September to December 1993 periods. This pattern was a result of the influence of tropical–extratropical cloud band (Tapp and Barrell, 1984) activity and mid-latitude frontal interactions driven by SST anomalies in the Indian Ocean (particularly adjacent and to the south–southwest of Sumatra) which, because of QB, LF and quasi-

decadal interactions, waxed and waned between ‘El Nin˜o-like’ and ‘La Nin˜a-like’ structures (Allan and D’Arrigo, 1999). In the austral summer, above-average rainfall was experienced in northern and eastern South Africa during January–March 1991 and again in December 1993–February 1994. However, this is a region which tends to experience dry conditions during El Nin˜o events. In JFM 1991, these areas received 1–2 standard deviations above average rainfall when there were warm SST anomalies in the neighbouring southwestern Indian Ocean (a situation that is known to be favourable for wetter conditions in this part of South Africa—Walker, 1990; Reason and Mulenga, 1999). Furthermore, NCEP re-analyses indicate stronger near-surface winds in the southwestern Indian Ocean during

2344

R.J. Allan et al. / Deep-Sea Research II 50 (2003) 2331–2347

this period, which would be favourable for increased advection of moist marine air over eastern South Africa. These re-analyses do not show a strong positive MSLP anomaly over the Indian Ocean/southeastern Africa region during JFM 1991, which is consistent with an atypical El Nin˜o impact over eastern South Africa in this season. The Indian summer monsoon seasons of 1990–1993 were notable for near average rainfall (1990 and 1993 were slightly above average and 1991 and 1992 were slightly below average), but significant flooding with enhanced monsoon conditions occurred in 1994, and a slightly below average monsoon rainfall situation was experienced in 1995. During the enhanced monsoon of 1994, warm SST anomalies occurred throughout much of the western Indian Ocean in April and May, i.e. just prior to and around the onset of the monsoon. Large areas of warm SST anomalies persisted in the tropical Indian Ocean south and southwest of India during June–September 1994. NCEP re-analyses indicate enhanced low level southwesterlies in the tropical western Indian Ocean, particularly during June–August 1994. These wind and SST anomalies during boreal summer 1994 are consistent with increased monsoonal rainfall over India. In summary, both South Africa and India experienced anomalously wet rainfall seasons at some point during the ‘protracted’ 1990–1995 episode. Thus, it appears that the most distinct and coherent modulation of rainfall during this ‘protracted’ episode was observed in northeastern Australia. 5.2. The 1998–2001 ‘protracted’ La Nin˜a episode Similar differences in conditions from those seen during ‘classical’ La Nin˜a episodes have marked the 1998–2001 ‘protracted’ La Nin˜a episode. Across Australia, the JAS to OND 1998, SON 1999 to MAM 2000 and SON 2000 to DJF 2001 seasons were marked by above average rainfall across north–northeastern parts of the country, particularly over the inland regions of the state of Queensland. However, marked rainfall deficiencies were experienced in southern and especially southeastern Australia from JAS 1998 to OND 1999, and again from MAM to JAS 2000, and from DJF

to JAS 2001. In central to northern parts of Western Australia, conditions alternated between above average rainfall in JAS 1998 to MAM 1999, SON 1999 to MAM 2000, and in DJF 2001, and very dry periods in JAS 1999, JAS to OND 2000, and MAM to JAS 2001. Over northern and eastern South Africa, rainfall was significantly above average over large areas for only JFM 2000, while during JFM 1999 and 2001 much of this region experienced up to 1.5 standard deviations below average rainfall. As with the JFM 1991 case, regional differences in the circulation anomalies from what is generally expected may account for the rainfall anomalies. During JFM 1999, NCEP re-analyses indicate that there was a substantial anticyclonic anomaly situated in the central-west South Indian Ocean, with little evidence of the cyclonic anomalies over southeastern Africa that are favourable for good austral summer rains. Instead, these anomalies were only evident over the far western regions of southern Africa. The Indian summer monsoon was late in 1998, but above average falls occurred in southern India with late season flooding in Bangladesh and along the Ganges valley. During the monsoon seasons of 1999 and 2000, Indian summer rainfall was markedly below average in western India, and much above average in Bangladesh and eastern India. The monsoon rainfall deficiency continued somewhat in western India and neighbouring countries into the 2001 summer season. Barlow et al. (2002) link the recent severe drought in central and southwestern Asia with this latest ‘protracted’ La Nin˜a episode and this appears to have been its major Indian Ocean impact. Thus, during the course of ‘protracted’ episodes, periods can occur when climatic patterns and signatures in various regions across the Indian Ocean basin can take on a very mixed, and at times opposite, configuration to that usually seen in ‘classical’ El Nin˜o and La Nin˜a events.

6. Conclusion and discussion Decadal–multidecadal climatic fluctuations have been reported in palaeoenvironmental,

R.J. Allan et al. / Deep-Sea Research II 50 (2003) 2331–2347

observational and climate model studies. Of the various modes described, a quasi-decadal ‘ENSOlike’ pattern involving a strong Pacific signature, seems to be important in the generation of ‘protracted’ El Nin˜o and La Nin˜a episodes. Seasonal correlations of both the LF ENSO and quasi-decadal signals with Indian Ocean domain rainfall suggest that the latter signal displays ‘ENSO-like’ properties. Hence the quasi-decadal signal alone can produce considerable low frequency ‘ENSO-like’ modulations of the climatic regime across the Indian Ocean region. However, ‘protracted’ El Nin˜o and La Nin˜a episodes also display higher frequency characteristics, and thus embody more than just the operation of a quasi-decadal feature in the climate system. Seasonal velocity potential correlations with QB, LF ENSO and quasi-decadal signals all show an evolution in their patterns of global mass overturning. There are broad similarities in these overturnings and any interactions between these signals would have the potential to cause various permutations of enhanced or suppressed patterns. A totally in-phase interaction of all signals would lead to not only a major enhancement of the magnitude of ENSO and ‘ENSO-like’ conditions, but would also protract them out in time. There is also a distinct regime shift that is evident in these correlations. The JFM, AMJ and JAS seasons are dominated by an Afro-Indo-Australasian source (sink) and a Pacific sink (source) of atmospheric mass overturning. However, this pattern changes to a regime with an Afro-western Indian Ocean sink (source), an eastern Indian Ocean–Australasian source (sink) and a Pacific sink (source) in OND. This evolution is found to occur to varying degrees on each of the QB, LF and quasi-decadal timescales. Any coincidence of the overturning regimes on all of the above timescales during ‘protracted’ El Nin˜o and La Nin˜a episodes in OND, and to a lesser extent JFM, would have the propensity to cause widespread drought or flooding extremes over central-eastern Africa to central Asia. Evidence from a recent study by Barlow et al. (2002) confirms the above findings. A focus on strong individual ‘protracted’ El Nin˜o and La Nin˜a episodes in the historical record further highlights the range of permutations

2345

resulting from any linear interactions between QB, LF and quasi-decadal signals in the climate system. A detailed analysis of the most recent 1990–1995 ‘protracted’ El Nin˜o and 1998–2001 ‘protracted’ La Nin˜a episodes shows how these particular climatic sequences severely impacted certain regions in countries bordering the Indian Ocean. Most notable were the persistent drought conditions across northeastern Australia, linked to the 1990–1995 episode, and the severe and protracted drought in central and southwestern Asia that resulted from the influence of the 1998– 2001 ‘protracted’ La Nin˜a episode. In summary, this study indicates that ‘protracted’ El Nin˜o and La Nin˜a episodes must be seriously considered in efforts to better predict the impact and influence of climatic fluctuations. This is particularly true for the Indian Ocean region with its complex ocean–atmosphere interactions and ENSO impacts. Climatic forecasting systems need to be developed that take account of not just ‘classical’ ENSO events, but also longer ‘protracted’ episodes and the decadal–multidecadal climatic modes and interactions that underpin them. The latter suggests that it is important to train forecasting schemes using data that extend back through the instrumental record as far as possible. Statistical schemes (e.g. Canonical Correlation Analysis [CCA] or Neural Network approaches) typically make use of SST and MSLP or wind data to make forecasts of future SSTs (for input into Atmospheric General Circulation Models [AGCMs]) or for forecasting rainfall. These methods can potentially use HadSLP and HadISST data extending back through the previous century to capture a range of ‘classical’ and protracted ENSO events and decadal modes. Schemes based on AGCMs or coupled General Circulation Models (GCMs) should as much as possible take into account results from prior integrations with historical SSTs (Folland et al., 2002) to include the influence of previous ‘protracted’ ENSO episodes.

Acknowledgements The first author’s work on this paper was undertaken initially at CSIRO Atmospheric

2346

R.J. Allan et al. / Deep-Sea Research II 50 (2003) 2331–2347

Research, Melbourne, Australia and was supported jointly through the CSIRO ‘Climate Variability and Impacts Programme’ and ‘Climate Change and Impacts Programme’, and funded in part by the Australian Federal Government’s National Greenhouse Research Programme of the Australian Greenhouse Office, and the State Government of Queensland. The paper was completed after the first author moved to the Met Office in the United Kingdom (UK), and is thus partly supported by the UK Government Meteorological Research Contract. The paper is crown copyright through the contribution of the UK author. Partial funding from the Australian Research Council and the South African Water Research Commission is gratefully acknowledged by the second author on this paper.

References Allan, R.J., 2000. ENSO and climatic variability in the last 150 years. In: Diaz, H.F., Markgraf, V. (Eds.), El Nin˜o and the Southern Oscillation: Multiscale Variability, Global and Regional Impacts. Cambridge University Press, Cambridge UK, pp. 3–56 Chapter 1. Allan, R.J., D’Arrigo, R.D., 1999. ‘Persistent’ ENSO sequences: how unusual was the 1990–1995 El Nin˜o? Holocene 9, 101–118. Allan, R.J., Lindesay, J.A, Reason, C.J.C., 1995. Multidecadal variability in the climate system over the Indian Ocean region during the austral summer. Journal of Climate 8, 1853–1873. Allan, R.J., Lindesay, J.A., Parker, D.E., 1996. El Nin˜o Southern Oscillation and Climatic Variability CSIRO Publications, Melbourne, Australia, 405pp. Barlow, M., Nigam, S., Berbery, E.H., 2001. ENSO, pacific decadal variability, and us summertime precipitation, drought, and streamflow. Journal of Climate 14, 2105–2128. Barlow, M., Cullen, H., Lyon, B., 2002. Drought in central and southwest Asia: La Nin˜a, the warm pool, and Indian Ocean precipitation. Journal of Climate 15, 697–700. Basnett, T.A., Parker, D.E., 1997. Development of the Global Mean Sea Level pressure data set GMSLP2. Climate Research Technical Note CRTN79, Hadley Centre, Meteorological Office, Bracknell, UK, 16pp. Folland, C.K., Shukla, J., Kinter, J., Rodwell, M., 2002. The climate of the twentieth century project. CLIVAR Exchanges 7, 37–39. Gershunov, A., Barnett, T.P., 1998. Interdecadal modulation of ENSO teleconnections. Bulletin of the American Meteorological Society 79, 2715–2725.

Harrison, D.E., Larkin, N.K., 1996. The COADS sea level pressure signal: a near-global El Nin˜o composite, time series view, 1946–1993. Journal of Climate 9, 3025–3055. Hulme, M., 1992. A 1951–80 global land precipitation climatology for the evaluation of general circulation models. Climate Dynamics 7, 57–72. Jones, P.D, Osborn, T.J., Briffa, K.R., Folland, C.K., Horton, E.B., Alexander, L.V., Parker, D.E., Rayner, N.A., 2001. Adjusting for sampling density in grid-box land, ocean surface temperature time series. Journal of Geophysical Research 106, 3371–3380. Kaplan, A., Kushnir, Y., Cane, M.A., 2000. Reduced space optimal interpolation of historical marine sea level pressure: 1854–1992. Journal of Climate 13, 2987–3002. Kaplan, A., Kushnir, Y., Cane, M.A., Blumenthal, M.D., 1997. Reduced space optimal interpolation for historical datasets: 136 years of Atlantic sea surface temperatures. Journal of Geophysical Research 102, 27835–27860. Larkin, N.K., Harrison, D.E., 2001. Tropical Pacific ENSO cold events, 1946–95: SST, SLP, and surface wind composite anomalies. Journal of Climate 14, 3904–3931. Larkin, N.K., Harrison, D.E., 2002. ENSO warm (El Nin˜o), cold (La Nin˜a) event life cycles: ocean surface anomaly patterns, their symmetries, asymmetries, and implications. Journal of Climate 15, 1118–1140. Latif, M., Kleeman, R., Eckert, C., 1995. Greenhouse warming, decadal variability, or El Nin˜o? An attempt to understand the anomalous 1990s. Max-Planck-Institut fur Meteorologie Report No. 175, 51pp. Latif, M., Kleeman, R., Eckert, C., 1997. Greenhouse warming, decadal variability, or El Nin˜o? An attempt to understand the anomalous 1990s. Journal of Climate 10, 2221–2239. Mann, M.E., Park, J., 1999. Oscillatory spatiotemporal signal detection in climate studies: a multiple-taper spectral domain approach. Advances in Geophysics 41, 1–131. Navarra, A. (Ed.), 1999. Beyond El Nin˜o. Decadal, Interdecadal Climate Variability. Springer, Berlin, Germany, 374pp. Parker, D.E., Jackson, M., Horton, E.B., 1995. The GISST2.2 sea surface temperature, sea-ice climatology. Climate Research Technical Note CRTN63, Hadley Centre, Meteorological Office, Bracknell, UK, 35pp. Power, S., Tseitkin, F., Mehta, V., Lavery, B., Torok, S., Holbrook, N., 1999a. Decadal climate variability in Australia during the twentieth century. International Journal of Climatology 19, 169–184. Power, S., Casey, T., Folland, C.K., Colman, A., Mehta, V., 1999b. Inter-decadal modulation of the impact of ENSO on Australia. Climate Dynamics 15, 319–323. Rayner, N.A., Horton, E.B., Parker, D.E., Folland, C.K., Hackett, R.B., 1996. Version 2.2 of the Global sea-Ice, Sea Surface Temperature data set, 1903–1994. Climate Research Technical Note CRTN74, Hadley Centre, Meteorological Office, Bracknell, UK, 35pp. Reason, C.J.C., 2000. Multidecadal climate variability in the subtropics/midlatitudes of the Southern Hemisphere Oceans. Tellus 52A, 203–223.

R.J. Allan et al. / Deep-Sea Research II 50 (2003) 2331–2347 Reason, C.J.C., Mulenga, H.M., 1999. Relationships between South African rainfall, SST anomalies in the south west Indian Ocean. International Journal of Climatology 19, 1651–1673. Reason, C.J.C, Rouault, M., 2002. ENSO-like decadal patterns, South African rainfall. Geophysical Research Letters 29(13), 16–1-16–4. Reason, C.J.C., Allan, R.J., Lindesay, J.A, Ansell, T.J., 2000. ENSO, climatic signals across the Indian Ocean basin in the global context: Part 1, interannual composite patterns. International Journal of Climatology 20, 1285–1327.

2347

Tapp, R.G., Barrell, S.L., 1984. The north-west Australian cloud band: climatology, characteristics, factors associated with development. Journal of Climatology 4, 411–424. Tourre, Y.M., Rajagopalan, B., Kushnir, Y., Barlow, M., White, W.B., 2001. Patterns of coherent decadal, interdecadal climate signals in the Pacific basin during the 20th century. Geophysical Research Letters 28, 2069–2072. Walker, N.D., 1990. Links between South African summer rainfall, temperature variability of the Agulhas, Benguela Current systems. Journal of Geophysical Research 95, 3297–3319.