Holocene El Niño–Southern Oscillation variability reflected in subtropical Australian precipitation

The La Niña and El Niño phases of the El Niño-Southern Oscillation (ENSO) have major impacts on regional rainfall patterns around the globe, with substantial environmental, societal and economic implications. Long-term perspectives on ENSO behaviour, under changing background conditions, are essential to anticipating how ENSO phases may respond under future climate scenarios. Here, we derive a 7700-year, quantitative precipitation record using carbon isotope ratios from a single species of leaf preserved in lake sediments from subtropical eastern Australia. We find a generally wet (more La Niña-like) mid-Holocene that shifted towards drier and more variable climates after 3200 cal. yr BP, primarily driven by increasing frequency and strength of the El Niño phase. Climate model simulations implicate a progressive orbitally-driven weakening of the Pacific Walker Circulation as contributing to this change. At centennial scales, high rainfall characterised the Little Ice Age (~1450–1850 CE) in subtropical eastern Australia, contrasting with oceanic proxies that suggest El Niño-like conditions prevail during this period. Our data provide a new western Pacific perspective on Holocene ENSO variability and highlight the need to address ENSO reconstruction with a geographically diverse network of sites to characterise how both ENSO, and its impacts, vary in a changing climate.

: Relationship between cumulative rainfall deficit (red) and lake level at Swallow Lagoon (blue). Swallow Lagoon is fringed by a small community of Melaleuca quinquenervia, an evergreen tree native to Papua New Guinea, New Caledonia and the tropical and subtropical east coast of Australia where it fringes freshwater lakes and dominates intermittently inundated coastal floodplains. The time between leaf formation and leaf abscission in M. quinquenervia can vary from 2 to 4 years 1 . As the lake is geologically and hydrologically isolated from other lakes and wetlands on the island, and the next nearest M. quinquenervia stand is at Brown Lake approximately 2 km away, we are confident that the M. quinquenervia leaves picked from the sediment cores originate from this local community. The Swallow Lagoon record is a composite of three sediment cores collected from the deepest part of the lake. As the uppermost 150 cm of sediment had very high water content, soft-sediment push corers were used to maintain stratigraphic integrity of the sediment. To maximise collection of leaves and leaf fragments, a large-diameter soft-sediment core was taken of the top 94 cm (core SLP2). A second small-diameter core (core SLP3) was taken of the top 250 cm to ensure penetration into the firm underlying sediment. Both were extruded at 1 cm intervals in the field. Core SL-2 (150-370 cm) was collected using a Livingstone corer and was correlated to SLP-3 using a conspicuous 7 cm thick band of sand that was evident at a depth of 235 cm in each core.

Macrofossil sampling
The sediment was sampled at contiguous 1 cm-equivalent intervals after correction for compaction. Each sample was sieved and the >500 µm fraction was retained. M. quinquenervia leaf fragments were identified under stereoscope by obvious morphological features, such as thick veins, epidermal colour and parenchymal cells in the spongy mesophyll. The leaf fragments varied in size from several cm 2 to several mm 2 . Leaves of other taxa (e.g., Eucalyptus) were occasionally present though were not used in analyses. Macrofossil samples for radiocarbon dating were also picked from the sieved remnants.
Precipitation reconstruction Carbon isotope discrimination by C 3 plants relative to the atmosphere (Δ leaf ) 2 is significantly related to rainfall at global 3 and regional 4 scales as moisture availability is the primary driver of leaf gas exchange in modern plants 3 . Recently, Schubert and Jahren (ref. 5) have argued that changes in atmospheric CO 2 can also influence Δ leaf . In addition, the strength of the Δ leaf -rainfall relationship is species-specific because of plant phylogeny and leaf traits. Other factors that can affect Δ leaf are location-specific, such as canopy density, degree of shading and the hydrological stress of individual trees 3 . In this study, we can control for inter-species variation as leaves from only one tree species (M. quinquenervia) are analysed. Intra-species variation in M. quinquenervia are accounted for as the tree species only occurs within 4 metres of Swallow Lagoon and, as North Stradbroke Island is a sand island, constant soil type and consistent hydrological stressors exist within this local Melaleuca community. In addition, the broad open canopy precludes any transpiration effects on 'shade' leaves 6 . For these reasons, the species-specific calibration dataset of Tibby et al (ref . 7) has advantages over other global datasets for this study 3,8 Assessing variability Standard statistical methods that might be applied to these data typically focus on estimating the trend in the mean of the response, making the assumption that the variance or other moments are constant across observations. Visual inspection of the δ 13 C calibrated rainfall time series suggested the presence of both a nonlinear trend and an increase in the variance of the observations. A trend in variance would violate any assumption of homogeneity. Furthermore, it is of significant interest to estimate the magnitude of any change in variance of the climate system as recorded at Swallow Lagoon.
Previous approaches for estimating change in variance in (palaeo)environmental time series have adopted a moving window approach (e.g. refs 9 and 10), in which the variance of a set of observations within a window of known width (typically ≤ half the time series length) is computed 11 . The window is shifted forward in time by one observation and the process repeated. Typically, the moving variance is computed using a detrended version of the time series.
There are a number of significant issues with the moving window approach: i) the irregular spacing of samples in time in a typical palaeoenvironmental time series is not easily accommodated and remedying the irregularity via interpolation can induce changes in the distributional properties of the data that we are interesting in estimating (e.g. ref. 12), ii) significance testing of the derived moving variance time series is complex due to strong autocorrelation patterns introduced because a moving window contains all but one sample used in the estimation of the variance in the previous window, iii) it is not possible to simultaneously estimate the mean and variance of a time series using the moving window approach, and detrending the series prior to estimating the moving variance risks conflating changes in variance with changes in the mean, biasing estimates of both, and iv) a significant reduction in power to detect change in variance arises via edge effects because the window width used requires us to observe as many observations as are included in the window before the first estimate of the variance can be calculated (if using a right padded moving window) or observations equal to half the window width if using a centred window.  Table S1: Radiocarbon results and modelled cal. ages at 95% probability. Samples with laboratory numbers commencing with "Beta-" were analysed by Beta Analytics, with "Wk-" by Waikato Radiocarbon Laboratory, and with "OZ" by the Australian Nuclear Science and Technology Organisation 13 * -Post-bomb date; fraction of modern carbon (F 14 C), instead of 14 C age, is reported.