Identifying coherent patterns of environmental change between multiple, multivariate records: an application to four 1000-year diatom records from Victoria, Australia
Introduction
Our understanding of Southern Hemispheric climate variability on multi-decadal to multi-centennial timescales is limited by a scarcity of quantitative, sub-decadally resolved climate records, a problem which is particularly manifest in Australia. To date the only annually resolved palaeoclimate records from Australia which extend further back in time than the most recent c. 350 years, are located in the latitudinal extremes of Tasmania and the northern tropics (Cook et al., 2000, Neukom and Gergis, 2012, PAGES 2k Consortium, 2013, Haig et al., 2014). By contrast, a number of sedimentary records exist, from shallow coastal/marine areas, lakes, peat bogs and speleothems, some of which span multiple millennia at sub-decadal resolution, but are limited by some degree of chronological uncertainty (Dodson et al., 1994, Mooney, 1997, Saunders et al., 2012, Saunders et al., 2013, Mills et al., 2013, Wilkins et al., 2013, Barr et al., 2014). In most cases, these archives offer indirect records of climate change, via geochemical, sedimentological and palaeoecological properties, limiting the potential for deriving quantitative climate records. However, identification of coherent patterns amongst multiple datasets can provide convincing evidence for regional scale climate or hydrological change. Importantly, by combining datasets, it is possible to separate patterns of externally forced climate variability from the idiosyncrasies which may exist within stand-alone records. Such potential has been highlighted by the use of empirical orthogonal functions (EOFs), derived using principal components analysis (PCA), in order to identify the coherent patterns between multiple records within regional and global palaeoclimate datasets (Clark et al., 2007, Clark et al., 2009, Shakun and Carlson, 2010, Anchukaitis and Tierney, 2012, Clark et al., 2012, Shakun et al., 2012, Tierney et al., 2013).
The use of EOFs is well grounded in climate and palaeoclimate research, but the majority of studies have applied the technique to either instrumental data or annually resolved palaeoclimate records such as tree ring or coral datasets (Weare et al., 1976, Smith et al., 1996, Mann et al., 1998). The extraction of EOFs from sedimentary archives represents a particular challenge due to the uncertainties associated with both dating and climatic interpretation. Recent implementation of Monte Carlo iterative age modelling within EOF analyses represents a valuable step towards dealing with age uncertainty in sediment records (Shakun and Carlson, 2010, Anchukaitis and Tierney, 2012). However, an ongoing issue with sediment based EOFs relates to the multivariate nature of many sedimentary records. Most sedimentary archives contain multiple lines of information, including geochemical parameters, microfossil remains or sedimentological properties. In the usual absence of quantitative palaeoclimate reconstructions (e.g. temperature, rainfall; Saunders et al., 2012, Saunders et al., 2013), the selection of representative variables from each site remains a source of ambiguity. It is impossible to include all of these variables within a regional ordination: doing so would fail the criterion of data independence and potentially bias the analysis to the site which contributes the most variables. However, selecting a single representative timeseries from each site can be subjective and can undermine the explorative element of the analysis. In addition, reducing a detailed matrix to a single variable leads to the loss of potentially relevant information and discards the considerable effort and time invested in collecting the data in the first place.
One particular consequence of reducing multivariate palaeoecological data to single variables prior to EOF analysis is that it undermines the possibility of observing correlations between secondary modes of variability at two or more sites, despite the intuitive likelihood of such correlations existing. Lake ecosystems are subjected to a variety of external and internal forces, which ultimately determine the expression of their sedimentary record (Battarbee, 2000, Smol et al., 2005, Mills et al., 2014, Wigdahl et al., 2014). Neighbouring lake systems exhibit different sensitivities to external forcing: a hypothetical ‘Site A’ might respond dramatically to changes in rainfall, through changes in lake level, whilst the effect of rainfall upon ‘Site B’ might manifest in a more muted physical/ecological response, e.g. through changing nutrient balance or lake water stratification. In such a scenario, even though rainfall changes do affect ‘Site B’, a comparison of the major patterns of ecological change at both sites would reveal limited coherency. Existing approaches to regional data syntheses do not allow for such variable climate–lake interactions. We therefore propose an alternative approach to exploring regional coherency between sedimentary archives using two-tiered nested ordinations. This involves reducing each site-specific multivariate dataset to a series of orthogonal variables using established methods of ecological data dimension reduction (in this case, Detrended Correspondence Analysis; DCA). The DCA sample scores of all site specific ordinations (hitherto termed ‘local DCAs’) are then combined within a ‘regional’ (multiple site) EOF analysis, which is performed using PCA following previous studies. The approach is applied to a suite of four diatom records from volcanic lakes in Western Victoria, Australia, which span the millennium prior to European colonisation (CE 800–1800). Coupled with iterative age modelling to account for age uncertainty similar to Anchukaitis and Tierney (2012), our approach provides a flexible, multi-layered means of exploring coherency between multivariate sediment records which allows for the detection of multiple patterns of change.
Section snippets
Sites and methods
Four sedimentary diatom records were selected from lakes within the western Victorian Volcanic Plains, Australia. Those sites were Lake Purrumbete, Lake Elingamite, Lake Surprise and Tower Hill Main Lake (Fig. 1). Each of the lakes is situated in a late Pleistocene volcanic crater, located within 50 km of the western coastline of Victoria. The climate of this region is Mediterranean in character, with cool, wet winters and mild, dry summers. The four study sites were selected based on their
Results
Age-depth models (Fig. 2) mirror those published elsewhere (Mills et al., 2013, Barr et al., 2014). Lake Elingamite and Lake Surprise both have >5 radiocarbon age constraints younger than 1500 cal. B.P., although both datasets are associated with large uncertainty around the initial age determinations (Fig. 2). Lake Purrumbete and Tower Hill Main Lake have only three radiocarbon dates, spaced at ∼2000 and ∼1000 years respectively. These 14C measurements have smaller uncertainties, and both sets
Methodological considerations
Multi-site data syntheses are becoming an increasingly common feature of palaeoclimate research, providing both a means of exploring large datasets for coherent signals (Clark et al., 2007, Shakun and Carlson, 2010, Anchukaitis and Tierney, 2012, Nicholson et al., 2013) and of testing hypotheses on the nature of and controls behind global scale climate changes (Clark et al., 2009, Shakun et al., 2012, Tierney et al., 2013). The Monte Carlo Empirical Orthogonal Function of Anchukaitis and
Conclusion
Palaeoclimate research is moving away from the traditional single site approach, towards utilising the increasing number of datasets, arranged in space, to pick apart the primary features of climate and environmental change. However, identifying regionally coherent patterns amongst multiple environmental timeseries can be challenging, particularly since such records are usually diverse, age uncertain, non-quantitative and multivariate. Performing prior site specific (local) ordination is one
Acknowledgements
The authors would like to thank Gavin Simpson, Peter Minchin and Bob Clarke for their advice on dimension reduction techniques in ecology, Maarten Blaauw for his advice on age-depth modelling and Kevin Anchukaitis and Jessica Tierney for sharing their Matlab code for Monte Carlo Empirical Orthogonal Functions. John Tibby's comments on an earlier version of this manuscript were much appreciated. This research was funded by two Collaborative Research Network (CRN) fellowships to JJT and KM, an
References (74)
- et al.
Climate variability in south-eastern Australia over the last 1500 years inferred from the high-resolution diatom records of two crater lakes
Quat. Sci. Rev.
(2014) Palaeolimnological approaches to climate change, with special regard to the biological record
Quat. Sci. Rev.
(2000)Methods and code for ‘classical’ age-modelling of radiocarbon sequences
Quat. Geochronol.
(2010)Out of tune: the dangers of aligning proxy archives
Quat. Sci. Rev.
(2012)- et al.
A pluvial episode identified in arid Australia during the Medieval Climatic Anomaly
Quat. Sci. Rev.
(2012) - et al.
Cross-continent comparison of high-resolution Holocene climate records from southern Australia – deciphering the impacts of far-field teleconnections
Earth Sci. Rev.
(2013) - et al.
6000 years of environmental changes recorded in Blue Lake, South Australia, based on ostracod ecology and valve chemistry
Palaeogeogr. Palaeoclimatol. Palaeoecol.
(2010) - et al.
Modelling historical lake levels and recent climate change at three closed lakes, Western Victoria, Australia (c.1840–1990)
J. Hydrol.
(2001) Quantitative reconstructions in palaeolimnology: new paradigm or sick science?
Quat. Sci. Rev.
(2013)- et al.
Holocene lake salinity changes in the Wimmera, southeastern Australia, provide evidence for millennial-scale climate variability
Quat. Res.
(2012)
Holocene dust deposition rates in Australia's Murray-Darling Basin record the interplay between aridity and the position of the mid-latitude westerlies
Quat. Sci. Rev.
Late Holocene changes in precipitation in northwest Tasmania and their potential links to shifts in the Southern Hemisphere westerly winds
Glob. Planet. Change
A global perspective on Last Glacial Maximum to Holocene climate change
Quat. Sci. Rev.
Comparative optical and radiocarbon dating of laminated Holocene sediments in two maar lakes: Lake Keilambete and Lake Gnotuk, south-western Victoria, Australia
Quat. Geochronol.
Reconstructing climate variability on the northeastern Tibetan Plateau since the last Lateglacial – a multi-proxy, dual-site approach comparing terrestrial and aquatic signals
Quat. Sci. Rev.
Identifying coherent spatiotemporal modes in time-uncertain proxy palaeoclimate records
Clim. Dyn.
Lake and catchment response to Holocene environmental change: spatial variability along a climate gradient in southwest Greenland
J. Paleolimnol.
Droughts and Flooding Rains: a Fine Resolution Reconstruction of Climatic Variability in Western Victoria, Australia, over the Last 1500 Years
Determination of the number of zones in a biostratigraphical sequence
New Phytol.
Estimating the amount of compositional change in Late-Quaternary pollen-stratigraphical data
Veg. Hist. Archaeobot.
Tracking Environmental Change Using Lake Sediments
Flexible paleoclimate age-depth models using an autoregressive gamma process
Bayesian Anal.
Late Quaternary stratigraphy and radiocarbon chronology of water level fluctuations in Lake Keilambete, Victoria
Nature
Deposition models for radiocarbon dating
Quat. Sci. Rev.
Cladocera as indicators of trophic state in Irish lakes
J. Paleolimnol.
The Last Glacial Maximum
Science
Mechanisms for an ∼7-kyr Climate and Sea-level Oscillation During Marine Isotope Stage 3, Ocean Circulation: Mechanisms and Impacts
Global climate evolution during the last deglaciation
Proc. Natl. Acad. Sci. U. S. A.
Change in Marine Communities: an Approach to Statistical Analysis and Interpretation
Warm-season temperatures since 1600 BC reconstructed from Tasmanian tree rings and their relationship to large-scale sea surface temperature anomalies
Clim. Dyn.
Palaeolimnological records of shallow lake biodiversity change: exploring the merits of single versus multi-proxy approaches
J. Paleolimnol.
Are the controls of species composition similar for contemporary and sub-fossil cladoceran assemblages? A study of 39 shallow lakes of contrasting trophic status
J. Paleolimnol.
Holocene ostracods, other invertebrates and fish remains from cores of four maar lakes in southeastern Australia
Proc. R. Soc. Vic.
A thousand years of environmental change and human impact in the alpine zone at Mt. Kosciusko, New South Wales
Aust. Geogr.
Chemical and biological trends during lake evolution in recently deglaciated terrain
Nature
Compositional dissimilarity as a robust measure of ecological distance
Plant Ecol.
Theory and Applications of Correspondence Analysis
Cited by (13)
Palaeoclimate perspectives on the Indian Ocean Dipole
2020, Quaternary Science ReviewsCitation Excerpt :In this context, and with the distribution of currently available hydroclimate records, it is not possible to assess how last millennium changes in the IOD may have also impacted rainfall in the Indonesian region. In southeastern Australia, lake sediments provide a history of effective moisture (precipitation minus evaporation) in the region through the last millennium until European settlement (Barr et al., 2014; Tyler et al., 2015). The interval from 1400 to 1880 experienced high effective moisture and low interdecadal variability (Barr et al., 2014).
Australian hydroclimate during Marine Isotope Stage 3: A synthesis and review
2019, Quaternary Science ReviewsPost little ice age drying of eastern Australia conflates understanding of early settlement impacts
2018, Quaternary Science ReviewsCoherent patterns of environmental change at multiple organic spring sites in northwest Australia: Evidence of Indonesian-Australian summer monsoon variability over the last 14,500 years
2018, Quaternary Science ReviewsCitation Excerpt :Empirical orthogonal functions (EOFs) can identify shared coherency between multiple sites and provide evidence for regional-scale climate change compared to local idiosyncrasies (Tyler et al., 2015). Monte Carlo iterative age modelling within EOFs has also been used to take into account age uncertainty which can be inherent in palaeoenvironmental records (e.g. Shakun and Carlson, 2010; Anchukaitis and Tierney, 2012; Tyler et al., 2015). MCEOF analysis therefore provides a means to empirically and objectively assess the synchronicity of changes observed in different time-uncertain palaeoenvironmental records by isolating dominant modes of regional variability (i.e. highlighting trends that are reliably shared between sites).
Millennial-scale variability in south-east Australian hydroclimate between 30,000 and 10,000 years ago
2018, Quaternary Science ReviewsCitation Excerpt :We used the Monte Carlo Empirical Orthogonal Function (MCEOF) approach to identify significant trends common to all records. Our method follows Anchukaitis and Tierney (2012) using code modified from Tyler et al. (2015). To maintain consistency in the chronological control, we updated each age-depth model using the ‘Bacon’ software in R (Blaauw and Christen, 2011).