Elsevier

Quaternary Geochronology

Volume 60, October 2020, 101112
Quaternary Geochronology

An exploration of the utility of speleothem age distributions for palaeoclimate assessment

https://doi.org/10.1016/j.quageo.2020.101112Get rights and content

Highlights

  • Speleothem growth density plots can reveal Quaternary climate shifts.

  • Kernel Density Estimators are preferable to Probability Density Plots.

  • Age uncertainty can be reintroduced into a KDE using Monte-Carlo simulations.

Abstract

Age distributions of speleothem growth frequency are often employed as proxies for past climate variation. However, their interpretation can be influenced by many variables such as the type of probability density estimator used, the number of samples studied, and indeed the nature of the sampling itself. In order to explore the utility of speleothem age distributions in this context, we constructed a synthetic speleothem growth dataset by randomly generating U-Th ages, based on a known climatic forcing to evaluate whether the most basic climatic information could be recovered by subsequent sampling of this dataset under differing conditions. This modelling approach allows us to separately examine the influences of sample size, the type of probability density estimator, and its parameterisation on the shape of the resulting age distribution. Our model demonstrates that periodic Quaternary climate fluctuations can be recovered from growth frequency distributions derived from as few as 120–150 radiometric ages. To construct such distributions, a kernel density estimator may more accurately represent speleothem growth frequency than the commonly employed probability density plot, because peaks in the latter often reflect data precision rather than data density. For U-Th ages, randomizations of the activity ratios can be used to reintroduce chronometric or analytical uncertainty into the Kernel Density Estimator in a Monte-Carlo simulation.

Introduction

Speleothems (secondary cave calcites) are important archives of palaeoclimate information that are both geographically extensive and often well preserved (Hendy and Wilson, 1968). Like marine sediments and ice cores, speleothems can record past climatic changes with high temporal resolution, but they also have the advantage of being highly amenable to precise and accurate radiometric dating (e.g. Harmon et al., 1975; Richards and Dorale, 2003; Schwarcz, 1980). Speleothems form as meteoric water becomes enriched in CO2 by vegetation productivity (e.g. root respiration and decomposition of organic matter) as it travels through soil layers, resulting in an elevated CO2 partial pressure, pCO2 (Hendy, 1971). Through interaction with the carbonate epikarst, the percolation waters dissolve calcium carbonate, increasing its calcium concentration until saturation is reached. Upon entering the cave atmosphere with lower pCO2 values, these bicarbonate-enriched waters become oversaturated, de-gas and precipitate carbonate which accumulates as speleothems (Fairchild and Baker, 2012). Production of soil pCO2 is closely related to temperature (Brook et al., 1983; Domínguez-Villar et al., 2013) and, in water-limited environments, also constrained by the availability of moisture (Brook et al., 1983). The type of vegetation cover is another important influence on soil pCO2 (Brook et al., 1983; Domínguez-Villar et al., 2013; Ford, 1971; Gunn and Trudgill, 1982). Variations in these controls on soil pCO2 are recorded in speleothems through a variety of proxies. One of these is speleothem growth frequency which has been frequently used to reconstruct changes in rainfall and/or regional or global temperature during the Quaternary (e.g. Ayliffe et al., 1998; Baker et al., 1993; Geyh, 1970; Gordon and Smart, 1984; Hennig et al., 1983; Hercman, 2000; Jo et al., 2014; Scroxton et al., 2016; Williams et al., 2010), allowing comparison with other archives such as ice-cores and marine records.

In this approach, peaks and troughs in the relative abundance of ages of (individual) samples from a single cave or cave region are interpreted as intervals of enhanced and reduced speleothem growth, respectively. The climatic interpretation of these intervals depends on which aspect of the regional climate is most likely to limit speleothem growth (Baker et al., 1993; Gordon et al., 1989): in moisture-limited regions (Ayliffe et al., 1998; Vaks et al., 2010), speleothem growth responds primarily to changes in effective precipitation (precipitation minus evaporation), whereas in cold (Lauritzen, 1991, 1993; Spötl et al., 2002; Vaks et al., 2013), temperate (Gordon et al., 1989; Hellstrom et al., 2020; Hercman, 2000; Jo et al., 2014; Williams et al., 2010) and tropical climates (Scroxton et al., 2016), it most likely driven by changes in temperature.

Unfortunately, the construction and interpretation of growth frequency distributions can be non-trivial and is strongly dependent on the precision and accuracy of the analyses (i.e. radiometric ages) feeding into them, the type of probability density function employed (Gordon and Smart, 1984; Gordon et al., 1989), the sampling method (Hercman, 2000; Scroxton et al., 2016) and the sample size (Gordon et al., 1989; Hercman, 2000). It is difficult to address many of these issues objectively using natural datasets, because the underlying genetic controls are essentially unknown. Previous studies have tried to justify which peaks are assumed to be real (i.e. peaks that reflect intervals of speleothem growth, in contrast to artefacts resulting from statistical errors) by calculating 95%-confidence intervals (Baker et al., 1993), by taking a random subset of the dataset (Gordon et al., 1989; Hercman, 2000), or by a randomization process in the speleothem age calculation (Hercman, 2000). Significant ambiguity remains, however, as to what extent growth frequency distributions truly reflect changes in climate.

In order to address this problem, here we present a model in which a synthetic dataset simulating ages of ‘growing stalagmites’ is first constructed using a well-understood palaeoclimatic record as the underlying climate forcing. With this we aim to explore two fundamental first-order issues: whether, in the case of speleothem growth frequency distributions, we can recover trends in major climatic changes from a record based on a well-known and well-understood palaeoclimate forcing and, if so, what is the optimal sample size. We base our model on known relationships between climatic parameters and speleothem growth and, in this example, assume that speleothem growth is enhanced when temperatures are increased. Speleothems are then subsequently sampled in different ways from this artificial dataset to determine how closely the original climate forcing can be reconstructed. The model allows us to directly isolate factors that influence the growth frequency distribution without the ambiguities imposed by interpretation of a natural dataset. The primary goal of this paper is not to provide a plug-in algorithm, but simply to evaluate whether even the most basic of climatic information from a speleothem proxy record can be retrieved using probability density estimation.

Section snippets

Choice of data visualisation tool

Past studies have used a variety of probability density estimators for summarising growth frequency distributions. Density estimation is widely used to illustrate patterns in datasets when the underlying controls on the observational data are unknown. A histogram is the simplest density estimator used for speleothem growth frequency (Ayliffe and Veeh, 1988; Geyh, 1970; Harmon et al., 1977; Hennig et al., 1983). However, as it is defined by the number and width of bins over a given interval, a

Variation with assigned bandwidth and sample size

We evaluated the effect of bandwidth on the growth frequency distribution by stepwise increase in the bandwidth for each model run. Fig. 4 shows nine plots of 1000 KDEs with a constant sample size of n = 250 and bandwidth increasing from 5 to 50 ka. Lower bandwidth resulted in higher resolution of the KDE but might also produce undersmoothing. In contrast, all resolution is lost with bandwidths higher than 30 ka. Therefore, we chose to run the statistical analysis (section 2.3) with bandwidths

Influences on growth frequency distributions

Detailed interpretation of speleothem growth frequency distributions depends on the precise shape of the curve. The shape itself depends on several factors, such as how the frequency distribution is constructed, sample size and several potential biases (Gordon and Smart, 1984; Gordon et al., 1989; Hercman, 2000; Scroxton et al., 2016). Because the underlying climate record is known in our model, it is possible to analyse how each factor influences the growth frequency distribution in detail, as

Conclusions

Through modelling, we have shown that most basic of climatic information can be retrieved from assessments of speleothem growth frequency. While it remains challenging to provide precise and accurate timing for periods of only moderate increase in speleothem growth, the large fluctuations in Quaternary climate can be successfully recovered. We demonstrate that a KDE may be a better statistical way of presenting speleothem growth frequency than the previously used PDP, because it better reflects

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgements

We would like to thank Sandra McLaren for initially suggesting the experiment of producing a synthetic cave model, Cameron Patrick from the Melbourne Statistical Consulting Platform for his statistical advice, and John Idárraga for assistance with programming. We also thank the three anonymous reviewers whose detailed and constructive comments greatly improved the manuscript. This research was facilitated by Australian Research Council grants FL160100028 to JDW and FT130100801 to JH.

References (63)

  • J. Hellstrom

    U-Th dating of speleothems with high initial 230Th using stratigraphical constraint

    Quat. Geochronol.

    (2006)
  • C.H. Hendy

    The isotopic geochemistry of speleothems—I. The calculation of the effects of different modes of formation on the isotopic composition of speleothems and their applicability as palaeoclimatic indicators

    Geochem. Cosmochim. Acta

    (1971)
  • G.J. Hennig et al.

    Speleothems, travertines, and paleoclimates

    Quat. Res.

    (1983)
  • G. Kaufmann

    Stalagmite growth and palaeo-climate: the numerical perspective

    Earth Planet Sci. Lett.

    (2003)
  • G. Kaufmann et al.

    Stalagmite growth and palaeo-climate: an inverse approach

    Earth Planet Sci. Lett.

    (2004)
  • M.S. Sambridge et al.

    Mixture modelling of multi-component data sets with application to ion-probe zircon ages

    Earth Planet Sci. Lett.

    (1994)
  • N. Scroxton et al.

    Natural attrition and growth frequency variations of stalagmites in southwest Sulawesi over the past 530,000 years

    Palaeogeogr. Palaeoclimatol. Palaeoecol.

    (2016)
  • A. Vaks et al.

    Middle-Late quaternary paleoclimate of northern margins of the Saharan-Arabian desert: reconstruction from speleothems of Negev desert, Israel

    Quat. Sci. Rev.

    (2010)
  • P. Vermeesch

    How many grains are needed for a provenance study?

    Earth Planet Sci. Lett.

    (2004)
  • P. Vermeesch

    On the visualisation of detrital age distributions

    Chem. Geol.

    (2012)
  • P. Vermeesch et al.

    Making geological sense of ‘Big Data’ in sedimentary provenance analysis

    Chem. Geol.

    (2015)
  • T. Andersen et al.

    Visualizing, interpreting and comparing detrital zircon age and Hf isotope data in basin analysis–a graphical approach

    Basin Res.

    (2018)
  • L.K. Ayliffe et al.

    500 ka precipitation record from southeastern Australia: evidence for interglacial relative aridity

    Geology

    (1998)
  • Z.I. Botev et al.

    Kernel density estimation via diffusion

    Ann. Stat.

    (2010)
  • A.W. Bowman et al.

    Applied Smoothing Techniques for Data Analysis: the Kernel Approach with S-Plus Illustrations

    (1997)
  • A.P. Boyle et al.

    F-Seq: a feature density estimator for high-throughput sequence tags

    Bioinformatics

    (2008)
  • C. Bronk Ramsey

    Methods for summarizing radiocarbon datasets

    Radiocarbon

    (2017)
  • G.A. Brook et al.

    A world model of soil carbon dioxide

    Earth Surf. Process. Landforms

    (1983)
  • J. DiNardo et al.

    Nonparametric density and regression estimation

    J. Econ. Perspect.

    (2001)
  • W. Dreybrodt

    Processes in Karst Systems

    (1988)
  • I.J. Fairchild et al.

    Speleothem Science: from Process to Past Environments

    (2012)
  • Cited by (0)

    View full text