Great Barrier Reef coral luminescence reveals rainfall variability over northeastern Australia since the 17th century



[1] Northeast tropical Queensland rainfall is concentrated in the summer half year and characterized by high interannual variability, partly related to El Niño–Southern Oscillation (ENSO) events. This results in highly variable river flows affecting nearshore coral reefs of the Great Barrier Reef, Australia. Freshwater flood events are recorded in long-lived, annually banded massive coral skeletons as luminescent lines. Quantitative measurements of luminescence intensity were made for 20 Porites coral cores from nearshore reef sites between 11°S and 23°S. Seventeen of the coral luminescence series were significantly correlated with an instrumental record of northeast Queensland summer rainfall and were used to develop seven significantly calibrated and verified rainfall reconstructions based on between 17 (starting 1891) and 1 (starting 1639) coral series. The longest reconstruction, based on more than one coral, provides insights into northeast Queensland rainfall variability from the late 17th century. Comparisons with various independent climate proxies are equivocal: the magnitude and significance of relationships with, for example, a proxy ENSO index vary through time. An extended drier period reconstructed from approximately the 1760s to the 1850s is associated with lower interannual rainfall variability. Since the late 19th century average rainfall and its variability have significantly increased, with wet and dry extremes becoming more frequent than in earlier centuries. This suggests that a warming global climate maybe associated with more variable tropical Queensland rainfall.

1. Introduction

[2] Reliable high-resolution proxy climate records considerably enhance our understanding of the nature, dynamics and causes of past and current interannual, decadal and longer-time-scale climate variability and change. The distribution of such records for the past several centuries is still, however, biased toward Northern Hemisphere land areas with a relative paucity of information from tropical and Southern Hemisphere regions [Jones et al., 2009; Mann et al., 2009]. This imbalance makes it difficult to fully understand how global climate has varied over recent millennia and possible regional climate variations (PAGES 2K Network:

[3] Long-lived, annually banded, massive coral skeletons contain a range of paleoclimatic tracers and make a significant contribution to understanding past tropical climate variability [Gagan et al., 2000; Correge, 2006; Grottoli and Eakin, 2007; Lough, 2010]. One of these tracers, the occurrence and intensity of luminescent lines when a coral slice is illuminated by ultraviolet (UV) light, provides information about past river flow and rainfall. Since their discovery in inshore corals of the Great Barrier Reef (GBR), Australia [Isdale, 1984], several studies have demonstrated a strong link between luminescence intensity and freshwater river flows and rainfall in adjacent catchments from various tropical coral reef locations [e.g., Scoffin et al., 1989; Fang and Chou, 1992; Lough et al., 2002; Nyberg, 2002; Peng et al., 2002; Jupiter et al., 2008; Grove et al., 2010]. Several long-term reconstructions of tropical river flow, rainfall and hurricane activity have been developed from coral luminescence records [e.g., Smith et al., 1989; Isdale et al., 1998; Hendy et al., 2003; Lough, 2007; Nyberg et al., 2007]. Such long-term reconstructions of tropical hydroclimate are now being incorporated into multiproxy reconstructions of larger-scale regional and hemispheric climate indices covering the past several centuries [D'Arrigo et al., 2008; Braganza et al., 2009; Mann et al., 2009; McGregor et al., 2010].

[4] One of these earlier studies [Lough, 2007] used simple visual assessment of the occurrence and intensity of luminescent lines in up to 25 long coral cores from nearshore reefs of the GBR to reconstruct freshwater flow into the GBR and Queensland rainfall back to the 17th century. Although based on a simple and potentially subjective assessment (a four-point scale of intensity), significantly calibrated and verified climate reconstructions were obtained [see also Hendy et al., 2003]. The purpose of the present study is to develop new reconstructions of NE tropical Queensland rainfall based on quantitative measurements of luminescence intensity using fluorescence spectroscopy [Barnes et al., 2003] in multiple coral cores from the GBR.

[5] The annual luminescence range (measured as the difference between summer peak luminescence and the preceding winter minimum luminescence) is used here as a measure of annual luminescence intensity. Several published hydroclimate reconstructions from measured annual average or annual maximum coral luminescence show declining trends with time [e.g., Smith et al., 1989; Isdale et al., 1998; Nyberg et al., 2007]. Recent analyses [Grove et al., 2010; Lough, 2011] provide evidence that these declining trends may be a coral age artifact possibly linked to a similar long-term declining trends noted for average skeletal density [Lough, 2008]. This apparent age artifact does not affect the measured luminescence range [Lough, 2011] or when luminescence measurements take into account skeletal density changes [Grove et al., 2010].

[6] The target for the luminescence-based reconstructions is NE tropical Queensland summer rainfall. Rainfall across northern Australia and Queensland is concentrated in the summer half year (Figure 1), shows large-scale coherence of anomalies and exhibits high interannual and decadal variability [Lough, 1991; Verdon et al., 2004]. This variability is closely linked with El Niño–Southern Oscillation (ENSO) events with above average rainfall during La Niña events and drier conditions during El Niño events [Lough, 1994; Risbey et al., 2009]. The central Pacific Modoki ENSO [Ashok et al., 2007] appears, however, to have a greater effect on rainfall over northwestern and northern Australia than eastern Australia [Taschetto and England, 2009a; Taschetto et al., 2009]. Highly seasonal and variable rainfall also results in highly seasonal and variable river flows and hence freshwater inputs to the GBR, which are concentrated in the summer half year [Finlayson and McMahon, 1988; Lough, 1994]. River discharge in eastern Australia is particularly sensitive to ENSO events [Ward et al., 2010]. The strength of the teleconnections between ENSO events and northeastern Australian climate is also modulated on multidecadal time scales by the Pacific Decadal Oscillation (PDO) [Mantua et al., 1997; Power et al., 1999]. During PDO cool phases the teleconnections between ENSO and eastern Australian rainfall tend to be stronger, with more coherent rainfall anomalies and higher rainfall variability compared to PDO warm phases [Kiem et al., 2003; Verdon et al., 2004; Meinke et al., 2005].

Figure 1.

Locations of 20 coral sites on Great Barrier Reef and major Queensland rivers draining eastward. Inset shows seasonality of annual rainfall across northern and eastern tropical Australia (adapted from Australian Bureau of Meteorology,

[7] An extended rainfall record will also allow assessment of current rainfall variations in a longer-term context. Widely reported declines in eastern Australian rainfall [e.g., Commonwealth Scientific and Industrial Research Organisation (CSIRO), 2007] appear, at least for Queensland, to be largely confined to the southeastern part of the state and only become apparent when records from the late 20th century are considered [Smith, 2004; Taschetto and England, 2009b]. Long-term changes are also of interest in the context of how NE Queensland summer rainfall is likely to change with continued global warming. Although the Australian summer monsoon is projected to strengthen [Meehl et al., 2007], regional-scale projections over this century range from drier to wetter conditions with variations comparable in magnitude to observed decadal variability [CSIRO, 2007; Allison et al., 2009]. Most models do, however, project increases in tropical rainfall variability.

[8] This study aims to develop reliable reconstructions of NE Queensland summer rainfall that significantly extend the instrumental record. Also considered is whether there have been significant changes in rainfall and its variability through time and whether the reconstructions are consistent with independent proxy climate records.

2. Methods

2.1. Coral Samples and Luminescence Data

[9] Slices from 20 coral cores were selected from the Australian Institute of Marine Science's archive of long coral cores from the GBR [Lough et al., 1999]. Choice of cores was based on (1) source reef located in nearshore waters likely to be affected by freshwater flood plumes [King et al., 2001; Lough et al., 2002], (2) previous analyses of annual coral growth records indicated record started prior to 20th century [Lough and Barnes, 2000; De'ath et al., 2009], and (3) the sample sites extended along much of the 2000 km length of the GBR. The 20 selected cores were from reefs located between 0 and 30 km from the mainland between 11°S–23°S (Figure 1 and Table 1).

Table 1. Details of Coral Cores Used in Reconstructions
ReefAIMS Core IDLatitude (°S)Longitude (°E)Start YearEnd YearR Rainfalla
  • a

    Correlation coefficient between the coral series and instrumental northeast Queensland summer rainfall over the period 1901–1981 (bold values significant at 5% level).

Red Wallis IslandRWI01B10.83142.17185819850.08
Conical RocksCNR01B15.13145.32188819830.46
Normanby IslandNOR01B17.20146.10187419840.69
Dunk IslandDUN02A17.95146.17186819870.47
Coombe IslandCOO01E18.03146.18184419850.67
Brook IslandBRO01A18.15146.28178319840.45
Great Palm IslandGPI02A18.68146.58188119840.39
Havannah IslandHAV01A18.85146.55163919850.67
Acheron IslandACH01A18.95146.65187619840.64
Magnetic IslandMAG01D19.15146.87181419850.48
Hook IslandHKO01B20.07148.95166119840.44
North Molle IslandNMI01B20.23148.80188219830.57
Cid HarbourCID01A20.27148.93182519900.42
South Molle IslandSMI01C20.27148.83185519810.59
Humpy IslandHMP01B23.20151.00168519950.42

[10] Coral luminescence (excitation k (nm)/emission k (nm) = 390/490) was measured at 0.25 mm intervals with a 2 mm diameter beam using optical luminometry along a major growth axis of each coral core slice [Barnes et al., 2003]. Luminescence and reflectance were measured against background values (to allow for possible drift within and between sample runs) at the start and end of each run. This produced relative luminescence values versus distance along each coral core slice. Annual luminescence peaks were dated backward from the date of collection and in conjunction with X-ray positive images of the core slices, UV photos of the core slices and skeletal density along the slices measured by gamma densitometry [Chalker and Barnes, 1990; Lough et al., 1999]. The annual luminescence range (difference between annual (summer) luminescence maximum and preceding (winter) minimum) was then extracted for each year; this variable is significantly correlated with the visual luminescence indices used by Lough [2007] and does not appear to be subject to a potential coral age artifact found in annual average or annual maximum luminescence that may be related to a similar apparent artifact in skeletal density [Lough, 2011]. The earliest dated record (Havannah Island) started in 1639 and the latest, 1891 (Pandora Reef), with 1981 being the last dated year common to all 20 records (Table 1).

2.2. Instrumental Climate Data

[11] Monthly instrumental rainfall was available for the period 1900–2008 from the Australian Bureau of Meteorology (gridded monthly rainfall data,, accessed February 2010). Monthly data were averaged for the region of NE tropical Queensland whose rivers drain eastward into the GBR (∼11°–23°S, 144°–151°E). Annual summer (October–March) rainfall was calculated for this region for the period, 1901–2008. Summer rainfall averaged 1052 mm, representing ∼70% of the annual total rainfall. The summer rainfall index was also significantly correlated with summer flows of the major Queensland rivers draining eastward into the GBR (for the period 1922–2003, the correlations were: Barron River = 0.38, Herbert River = 0.58, Burdekin River = 0.60, Pioneer River = 0.51, and Fitzroy River = 0.55).

[12] Average summer sea surface temperatures (SST) for the Niño 3.4 region, 1871–2009, ( were used as an index of ENSO events [Rayner et al., 2003]. The summer rainfall index was significantly inversely correlated with the summer Niño 3.4 index, r = −0.56 (1901–2008) and the correlation was relatively stable over time (r = −0.51, 1901–1954, and r = −0.60, 1955–2008).

2.3. Proxy Climate Data

[13] For comparison with the new rainfall reconstructions, 21 annually resolved proxy climate series were selected from the NOAA Paleoclimatology database ( Selection of the series was based on the reconstruction extending back to at least 1685 (the start year of the rainfall reconstruction based on three coral records; see section 2.4), and was, with one exception, related to tropical or Southern Hemisphere climate. The 21 series consisted of: three ENSO proxies [Braganza et al., 2009; Mann et al., 2009; McGregor et al., 2010]; six Pacific Decadal Oscillation (PDO) proxies [Biondi et al., 2001; D'Arrigo and Wilson, 2006; MacDonald and Case, 2005; Shen et al., 2006; Linsley et al., 2008; Mann et al., 2009]; global average SST and Northern Hemisphere temperatures [Mann et al., 2009]; Tasmanian temperatures [Cook et al., 2000]; southwestern Australian rainfall [Cullen and Grierson, 2009]; three coral δ18O series which reflect SST and/or sea surface salinity (SSS) [Quinn et al., 1998; Druffel and Griffin, 1999; Zinke et al., 2004]; South American summer rainfall and temperature [Neukom et al., 2010a, 2010b]; and an Indonesian region winter Palmer Drought Severity Index (PDSI) extracted from the Monsoon Asia Drought Atlas [Cook et al., 2010]. For the latter index, the first principal component of the reconstructed PDSI for 28 boxes within the area ∼96°–116°E, 9°N–9°S was calculated. This accounted for 56.5% of the variance and was significantly correlated with the Niño 3.4 index (r = −0.61) and NE Queensland rainfall in the following summer (r = 0.32), 1901–2005. These relationships reflect a tendency for drier/wetter conditions in NE Australia and the Indonesian region during El Niño/La Niña events, and some continuity between the strength of the Northern Hemisphere summer (JJA) monsoon and that of the Australian summer season [McBride et al., 2003; Meehl et al., 2003]. It should be noted that several of these proxy climate series are not independent as they share several common data series.

2.4. Reconstruction Methodology

[14] Initial screening of the coral luminescence records (“R rainfall” column in Table 1) showed that, with three exceptions, all coral series were significantly correlated with instrumental summer rainfall over the period 1901–1981. The three exceptions (Red Wallis Island, Pascoe and Jeannie cores) were all from the most northerly reefs. The reason for these nonsignificant relationships was unclear and these three series were excluded from further analyses.

[15] The 17 coral luminescence series started at different times between the early 17th and late 19th centuries (Table 1). A simple approach, based on Principal Component Analysis (PCA), was used to develop a set of reconstructions with successively fewer series included as the reconstruction extended further back in time. The first principal component (PC1) was extracted for six groups of corals varying from 17 corals for the period 1891–1981 to three corals for the period 1685–1981 [Lough, 2004]. Each PC1 was then linearly regressed against the NE Queensland summer rainfall index over the 1901–1981 calibration period. A seventh reconstruction was based on the single Havannah (HAV) coral luminescence time series which extended back to 1639 (Table 2).

Table 2. Reconstruction Data Over the Calibration Period 1901–1981
SeriesReconstruction IDPeriodPC1 Variancea (%)R2 1901–1981 (%)R2 1901–1941 (%)R2 1942–1981 (%)R2 1901–1981 Decadalb (%)
  • a

    Variance explained by PC1 of each set of corals.

  • b

    R2 between each reconstruction and instrumental record using 10 year Gaussian filtered data.

17 coresR171891–198145.660.852.668.965.6
13 coresR131876–198146.260.352.268.767.2
9 coresR91844–198139.955.544.166.865.6
6 coresR61820–198142.149.028.870.460.8
4 coresR41783–198146.448.134.460.865.6
3 coresR31685–198146.046.543.349.947.6
1 coreHAV1639–1981na44.933.955.037.2

3. Results

3.1. Reconstructions

[16] The six PC1s explained similar percentages of the total variance (∼40–46%; Table 2), comparable to 49% explained in a PCA of Queensland summer rainfall based on 39 instrumental rainfall series [Lough, 1991]. All linear regressions (Table S1 in the auxiliary material) with NE Queensland summer rainfall were significant at the 5% level and the explained variance, although decreasing with decreasing number of coral series, varied from ∼61% (with all 17 cores) to 45% (with only the Havannah core). All regressions were also significant when calculated for two independent subperiods, 1901–1941 and 1942–1981 but showed greater explained variance during the later subperiod (Table 2). All reconstructions retained a significant (allowing for reduced degrees of freedom) proportion of decadal rainfall variability (as measured by the common variance of 10 year Gaussian filtered instrumental and reconstructed time series) (Table 2).

[17] The instrumental rainfall record (Figure 2a) was characterized by high interannual variability superimposed on decadal variations with, noticeably, wetter conditions in the 1950s and 1970s. The instrumental rainfall series for NE Queensland did not show any significant linear trends toward either wetter or drier conditions when examined over the whole length of record, 1901–2008 nor when examined over periods starting in 1931, 1951 and 1971. These features were retained in the various reconstructions (Figures 2b2d and Figures S1a–S1d in the auxiliary material) but with new features emerging as the length of the respective reconstruction extended further back in time. Only two of the reconstructions had a significant long-term trend. These were R6 (starting in 1820) and R4 (starting in 1783) and the trend was toward wetter conditions.

Figure 2.

Northeast Queensland summer rainfall: (a) instrumental 1901–2008, (b) R17 1891–1981, (c) R6 1820–1981, and (d) R3 1685–1981. Reconstructions based on 1901–1981 calibration period. Series presented as anomalies (millimeters) from respective 1901–1981 average (note different y axis scale for instrumental series). Thick blue line is 10 year Gaussian filtered series.

[18] With the exception of the single-core reconstruction (HAV), the 1973–1974 Queensland summer wet season was identified as the wettest on record (Table 3). 1974 has also been noted as the wettest year across Australia over the instrumental record period, 1901–2002 [Smith, 2004]. Although the decade 1970–1979 was variously identified as one of the wettest on record, including data from the late 18th century (R4, R3 and HAV) placed the 1890s as the wettest on record. With the longer records, a marked drier period emerged in late 18th to early 19th centuries (Table 3).

Table 3. Summary Statistics for Instrumental and Reconstructed NE Queensland Summer Rainfall, Based on 1901–1981 Calibration Perioda
  • a

    Bold indicates subperiod averages significantly different (at 5% level) from long-term mean.

Mean (mm)10521075108310641042103310491040
Coefficient of Variation (%)28%23%23%21%20%18%16%17%
Median (mm)1053104810441031100299910211015
Annual Maxima and Minima
Max 119741974197419741974197419741896
Max 219101896189619181870189618961950
Max 319111918191818961918191819101974
Min 119021961189218921969182118231751
Min 219151964196419641821196119611969
Min 319421892196619071845179919641961
10 Year Maxima and Minima
Max 11970–19791970–19791970–19791970–19791970–19791890–18991890–18991890–1899
Max 21950–19591949–19581890–18991890–18991890–18991968–19771949–19581949–1958
Max 31905–19141894–19031949–19581949–19581949–19581949–19581699–17081873–1882
Min 11915–19241960–19691960–19691960–19691845–18541845–18541961–19701849–1858
Min 21986–19951930–19391930–19391845–18541961–19701791–18001845–18541926–1935
Min 31961–19701920–19291920–19291930–19391820–18291961–19701770–17791803–1812
30 Year Maxima and Minima
Max 11950–19791950–19791890–19191890–19191890–19191889–19181889–19181889–1918
Min 11919–19481919–19481919–19481919–19481827–18561827–18561770–17991833–1862

3.2. Comparison With Previous Reconstruction

[19] The present reconstructions were based on quantitative luminescence measurements from up to 17 coral cores. The contributing cores were similar, though not identical, to the suite assessed for visual luminescence indices and used to reconstruct indices of October–September freshwater flow into the GBR and a different index of Queensland rainfall to that used here [Lough, 2007]. Despite the different methodologies and slightly different suite of cores, the two sets of reconstructions would be expected to have a high degree of common information, given the coherent and regional-scale nature of summer rainfall in tropical Queensland [Lough, 1991]. This was indeed the case, with the present set of reconstructions having between ∼55% (HAV, R3) and ∼80% (R13, R17) variance in common with the earlier reconstruction of Queensland rainfall.

[20] Although interannual and decadal variability of the old and new reconstructions were very similar, there were differences in the relative magnitude of rainfall anomalies from the two reconstructions. This is illustrated for the R3 reconstruction (Figure 3) which reconstructs drier conditions from the mid-18th to early 19th centuries and wetter conditions in the late 19th century compared to the old rainfall reconstruction. The presumed better identification of drier and wetter conditions in the new reconstructions is likely due to the simple visual assessment technique not being able to capture the nuances of faint and intense luminescent lines [Lough, 2011]. The new reconstructions, based on direct measurements are, therefore, considered more reliable.

Figure 3.

Differences between reconstructed summer rainfall (R3) and earlier reconstructed October through September rainfall, 1685–1981 [Lough, 2007]. Both series were standardized with respect to the mean and standard deviation over the period 1901–1981. Only differences greater than ±0.5 standard deviation are plotted.

3.3. Evidence for Changes in Rainfall Variability

[21] Earlier reconstructions of river flow and rainfall from GBR coral luminescence [Hendy et al., 2003; Lough, 2007] remarked on an apparent increase in variability of NE Queensland hydrological regime during the 20th century with more very wet and very dry extremes compared to earlier centuries. Such an increase in tropical rainfall variability is one projected consequence of global warming [Meehl et al., 2007]. McGregor et al. [2010] also note a 20th century increase in variance of their Unified ENSO Proxy (UEP). This record extends back to 1650 and was developed from 10 ENSO proxies (only one of which [Braganza et al., 2009] includes, in a total of 8 proxies, the Hendy et al. [2003] luminescence record from the GBR, relatively independent of the present reconstructions). Braganza et al. [2009] also provide evidence for variations in the amplitude and frequency of ENSO over the past four centuries. They suggest that high-frequency (∼2–4 years) ENSO variability increased over the most recent 200 years (up to 1982) compared with the preceding 250 years of their reconstruction.

[22] The R3 reconstruction was examined for evidence of changes in NE Queensland rainfall variability through time. The new reconstruction also showed increased variability from the late 19th century compared to earlier periods (Figure 4a). There was also an extended period of reduced rainfall variability from the late 18th to early 19th centuries that corresponded with the period of below average rainfall (Figure 2d). This increase in variability since the late 19th century of NE Queensland rainfall roughly coincided with the increased variability reconstructed in the UEP [McGregor et al., 2010] though the latter does not so strongly identify the late 18th and early 19th centuries with reduced variability (Figure 4b). The PDSI-Indonesia also showed changes in variability with minima ∼1730s, 1770s and 1830s roughly coinciding with minima in reconstructed NE Queensland rainfall (Figure 4c).

Figure 4.

Running 21 year standard deviations (expressed as departures from the overall average SD) for (a) R3 reconstruction, (b) UEP [McGregor et al., 2010], and (c) MADA-Indonesia PDSI [Cook et al., 2010].

[23] Changes in variability were also evident through analyses over three ∼100 year subperiods of the frequency of the most extreme reconstructed rainfall years, based on the 90th and 10th percentiles over the period 1685–1981 (Table 4). Both the wettest and driest years were markedly less frequent in the period 1785–1884, with very wet years only occurring about every 25 years and very dry years every 14 years. This represents a decrease in frequency of extremes from the preceding 100 years when very wet years occurred about twice as often. The most recent period from that late 19th century showed an increase in frequency of both very wet and very dry years. Overall, the frequency of both wet and dry year extremes varied from about once every 5 years in the first 100 years of the reconstruction, once every 9 years in the middle century and increased to once every 3 years in the most recent period. This increased frequency of both wet and dry extremes again supports the evidence for increased variability of NE Queensland rainfall, and possibly tropical climate, since the late 19th century.

Table 4. Return Periods of Reconstructed (R3) Rainfall Extremes for Three Subperiods
PeriodVery Wet, >90th Percentile (years)Very Dry, <10th Percentile (years)Total (years)

3.4. Comparison With Other Proxy Climate Records

[24] ENSO events significantly influence summer rainfall variability in NE Queensland and the correlation between the instrumental rainfall series and the Niño 3.4 index over the period 1901–1981, was −0.55 (significant at the 5% level). The seven reconstructions were also significantly correlated (at the 5% level), though with reduced magnitude, with the Niño 3.4 index, 1901–1981, with correlations of −0.31 (HAV), −0.39 (R3), −0.37 (R4), −0.35 (R6), −0.37 (R9), −0.38 (R13) and −0.37 (R17). Instrumental NE Queensland summer rainfall was also significantly correlated with summer SSTs, especially in the eastern equatorial Pacific and tropical Indian Ocean (Figure 5, top) with higher rainfall associated with cooler waters in these regions and lower rainfall with warmer waters. This pattern, which is characteristic of the ENSO signature in tropical SSTs [e.g., Eakin et al., 2009] was also evident for the reconstructed (R3) rainfall index (Figure 5, bottom).

Figure 5.

Correlation coefficients between summer season SSTs (HadISST) [Rayner et al., 2003], 1901–1981, and (top) instrumental northeast Queensland summer rainfall and (bottom) reconstructed (R3) northeast Queensland summer rainfall (the ±0.2 contour approximates the 5% significance level).

[25] The earlier rainfall and river flow reconstructions [Lough, 2007] were weakly but significantly correlated with the Niño 3.4 reconstruction from Mann et al. [2000], r = −0.26 over the independent period 1650–1923. The new reconstruction (R3) was also significantly correlated with the UEP index [McGregor et al., 2010] over the period 1901–1977 (r = −0.38) and for the independent period 1685–1900 (r = −0.26), that is, same magnitude correlation as found for the earlier reconstruction. The magnitude of the correlation between the rainfall reconstruction and the UEP also varied through time (Figure 6a) with the first half of the 18th century showing relatively high correlations and again from the mid-19th century but interspersed with a period from the late 18th through mid-19th century of weak or even reverse sign of the correlations. The correlations between the rainfall reconstruction and the Indonesian PDSI (Figure 6b) were high during the 20th century but weak and even of reverse sign prior to then. Again, however, this period of reduced strength of ENSO teleconnections coincides with the period of reconstructed below average rainfall, reduced rainfall variability and relatively infrequent rainfall extremes.

Figure 6.

Running 21 year correlation coefficients between R3 reconstruction and (a) UEP [McGregor et al., 2010] and (b) MADA-Indonesia PDSI [Cook et al., 2010].

[26] The reconstructed (R3) rainfall and UEP series were also compared by examining the most extreme years identified in the two series. The rationale is that as all proxy climate records are imperfect and may not capture the continuum of climate variations, particularly with close to average conditions, but they may better identify the most extreme climate anomalies. Over the period 1685–1977, using the ±0.5 SD criteria of McGregor et al. [2010], 88 El Niño and 104 La Niña years were identified from the UEP. The 104 wettest years and 88 driest years were then identified from the R3 reconstruction. If both were responding consistently to ENSO with drier conditions in NE Queensland during El Niños and wetter conditions during La Niñas, then we would expect a good correspondence between the two independent proxies. The results (Table S2) identified 33 years over the 293 year period when both proxies had values characteristic of El Niño events and 42 years when both proxies had values characteristic of La Niña events (Figure 7a). There were, however, many years of disagreement. First, when the two proxies identified totally opposite extremes; R3 identified 23 very wet years which corresponded to UEP identifying expected dry El Niño conditions and R3 identified 24 years which were very dry, yet UEP suggested La Niña conditions (Figure 7b). Second, there were 40 years that R3 indicated as very wet but UEP did not show a marked anomaly and, conversely there were 31 years identified in R3 as very dry but no marked UEP anomaly (Figure 7c). Finally, there were 32 years that UEP indicated were El Niño years, but R3 showed no marked anomalies and, conversely, there were 37 years that UEP indicated were La Niña years but R3 indicated no marked anomaly (Figure 7d). In addition, the years of agreement and disagreement did not show any clear pattern through time. One might, for example, expect the level of agreement to increase and the level of disagreement to decrease with time toward the present. This was not found. The 75 years within the period 1685–1977 when both proxies agreed about rainfall extremes and ENSO status were, however, highly likely to have been ENSO years (Figure 7a).

Figure 7.

Agreement and disagreement between UEP ENSO years and wet and dry reconstructed NE Queensland rainfall, 1685–1977: (a) agreement wet and La Niña, dry and El Niño; (b) total disagreement wet and El Niño, dry and La Niña; (c) disagreement wet and non-ENSO, dry and non-ENSO; and (d) disagreement La Niña and average rainfall, El Niño and average rainfall.

[27] Finally, the various proxy climate records were examined for three subperiods identified in the 1685–1981 (R3) NE Queensland summer rainfall reconstruction (Figure 2d). These were 1685–1764, 1765–1855 and 1856–1900. The average rainfall anomaly (from the 1901–1981 mean) and SD for these three periods were: +2.2 ± 145.2 mm, −79.5 ± 109.6 mm, and +46.0 ± 166.0 mm, respectively. The changes in average rainfall and rainfall variability between the successive periods were all significant. Do other independently derived climate proxies show significant changes across these time periods? In particular, one might expect reconstructions of the PDO to highlight the 1765–1855 period of lower NE Queensland rainfall and rainfall variability as more warm PDO phase and the period 1856–1900 when rainfall increased and became more variable as more cold PDO phase [Kiem et al., 2003; Verdon et al., 2004; Meinke et al., 2005]. The results (Table 5) were equivocal. Only one of the PDO reconstructions [MacDonald and Case, 2005] showed a significant change to warm phase ∼1765 and two of the PDO reconstructions [D'Arrigo and Wilson, 2006; Mann et al., 2009] showed a significant change to warm PDO phase coinciding with NE Queensland rainfall becoming wetter and more variable (i.e., the opposite of expected). All three coral δ18O records showed significant changes across the three periods toward warmer and/or fresher conditions which suggests a long-term, ongoing trend. South American summer rainfall also became significantly wetter across the three periods. Cross-correlations between the various series and the R3 rainfall reconstruction were variable and inconsistent across the subperiods (Table S3). The R3 reconstruction was only significantly correlated with the two ENSO series of Braganza et al. [2009] and McGregor et al. [2010] over the recent period. Many of the significant correlations among the 21 climate proxies for the earlier time periods are likely to have arisen from several of these reconstructions drawing from the same pool of high-resolution proxy climate series. This makes it difficult to independently assess proxy climate reliability in the past.

Table 5. Differences Between Climate Reconstructions Averaged for Three Subperiods Between 1685 and 1900a
SiteProxyVariable(1765–1855) Minus (1685–1764)(1856–1900) Minus (1765–1855)Source
R3coral luminescenceraindrierWetterthis study
UEPmultiproxyENSO  McGregor et al. [2010]
ENSO5multiproxyENSO more La Niña–likeBraganza et al. [2009]
Nino3multiproxyENSO more El Niño–likeMann et al. [2009]
PDOtree ringPDO  Biondi et al. [2001]
PDOtree ringPDO more warm phaseD'Arrigo and Wilson [2006]
PDOtree ringPDOmore warm phase MacDonald and Case [2005]
PDOdocumentaryPDO  Shen et al. [2006]
PDOcoral δ18OPDO  Linsley et al. [2008]
PDOmultiproxyPDO more warm phaseMann et al. [2009]
GlobalmultiproxySST WarmerMann et al. [2009]
NHmultiproxytemp WarmerMann et al. [2009]
Tasmaniatree ringtemp  Cook et al. [2000]
Lake Tay, WAtree ringrain  Cullen and Grierson [2009]
Abraham S GBRcoral δ18OSST/SSSwarmer and/or fresherwarmer and/or fresherDruffel and Griffin [1999]
Ifaty, Madagascarcoral δ18OSST/SSSwarmer and/or fresherwarmer and/or fresherZinke et al. [2004]
New Caledoniacoral δ18OSST/SSSwarmer and/or fresherwarmer and/or fresherQuinn et al. [1998]
S AmericamultiproxyDJF rainWetterWetterNeukom et al. [2010a]
S AmericamultiproxyDJF temp CoolerNeukom et al. [2010b]
Indonesiatree ringJJA PDSI  Cook et al. [2010]

4. Summary and Conclusions

[28] New reconstructions of NE Queensland summer rainfall were developed back to the 18th century. Reconstructions were based on the measured annual luminescence range in up to 17 coral cores and did not exhibit declining trends toward the present [e.g., Isdale et al., 1998; Nyberg et al., 2007] which may be a possible age artifact in measured annual average luminescence possibly associated with a similar trend in average skeletal density [Grove et al., 2010; Lough, 2011]. Improved understanding of the nature and causes of these apparent age-related changes in average skeletal density and average luminescence could lead to even more robust measures of annual luminescence intensity in coral skeletons and hence further refine the reconstructions presented here. The reconstructions were significantly calibrated and verified and also capture decadal variability characteristic of the instrumental record. The longest reconstruction of NE Queensland summer rainfall from 1685 based on more than one coral (R3) supersedes earlier reconstructions for the region based on simple visual assessment of annual luminescence intensity.

[29] This new reconstruction identified 1823 as the driest year and the 1974 summer as the wettest, at least since 1685. This year has also been shown to be the wettest across Australia during the instrumental record period [Smith, 2004]. It may, however, be rivaled by the 2010–2011 summer, the first half of which (October–December) experienced highest rainfall totals on record across large parts of Queensland resulting in significant and devastating flooding over much of the state ( = colourandtime = latestandstep = 0andmap = decileandperiod = 3monthandarea = nat, accessed January 2011). Although there was no significant linear trend toward either wetter or drier conditions from the 17th to 20th centuries, the new reconstruction was characterized by an extended period from ∼1760s to ∼1850s of drier and less variable summer rainfall compared to earlier and later periods. From ∼1850s average rainfall and its variability increased and both wet and dry extremes occurred more frequently. This increased variability from the late 19th century is also evident in two ENSO climate proxies [Braganza et al., 2009; McGregor et al., 2010] and there is, therefore, mounting evidence for a significant change in tropical climate variability at this time. More extreme tropical rainfall variations are one projected consequence of continued global warming. Reconstructed drier and less variable summer rainfall from the 1760s to 1850s would be characterized by a warmer phase of the PDO, though only one independent PDO reconstruction [MacDonald and Case, 2005] showed this change, a result that could have arisen by chance given the lack of agreement among different PDO reconstructions prior to the 20th century [D'Arrigo et al., 2005; Lough, 2007].

[30] Northeast Queensland summer rainfall is modulated by ENSO events and the reconstruction was significantly correlated with the instrumental ENSO index. Agreement between reconstructed rainfall and reconstructed ENSO extremes was mixed. Over the period 1685–1977, for example, 192 years were identified by the independent UEP [McGregor et al., 2010] as either El Niño or La Niña events for which NE Queensland summer rainfall would be expected to be drier and wetter, respectively. For 39% of these years, the reconstructed rainfall and proxy ENSO series agreed, representing 75 years back to 1685 which were highly likely to have been ENSO extremes. The two proxies disagree totally for 24% of years and for 36% of years there were no marked reconstructed rainfall anomaly.

[31] The sign and significance of levels of agreement between a suite of climate proxies, largely from tropical oceans and Southern Hemisphere region, were variable since 1685. The growing use of the same suite of proxy climate records to develop different larger-scale climate reconstructions makes it very difficult to properly assess whether different records are truly independent or are based on the same fundamental data.

[32] Finally, measured luminescence range was demonstrated in this study (with appropriate replication) to be a robust proxy for rainfall (and river flow) in the highly seasonal and highly variable hydroclimatic regime of NE tropical Queensland. Application of this approach to long-lived massive corals from other locations would considerably enhance our understanding of tropical climate variability over the past several centuries.


[33] The long coral cores used in this study were collected by Bruce Parker and Peter Isdale; skilled technical support was provided by Monty Devereux and Eric Matson. The 20 annual luminescence range series and seven reconstructions are lodged with AIMS Data Centre and the NOAA Paleoclimatology Data Center. The author is very grateful for the constructive comments provided by three anonymous reviewers.