|
Titel |
Using Bayesian regression to construct proxy time series from palaeoclimate archives |
VerfasserIn |
B. Goswami, J. Heitzig, K. Rehfeld, N. Marwan, J. Kurths |
Konferenz |
EGU General Assembly 2012
|
Medientyp |
Artikel
|
Sprache |
Englisch
|
Digitales Dokument |
PDF |
Erschienen |
In: GRA - Volume 14 (2012) |
Datensatznummer |
250065510
|
|
|
|
Zusammenfassung |
Age-depth observations from palaeoclimatic archives (i.e., peat cores, lake sediments, marine
cores, etc.) consist of a set of dating points that comprise of the calibrated radiocarbon-dated
ages and corresponding depths measured in the archive. However, the actual understanding of
palaeo-climate comes from proxies (such as oxygen isotopes) that are related to various
climatic parameters. Due to limitations of measurement, radiocarbon (i.e., 14C) age-depth
measurements are far fewer in number than the number of proxy-depth measurements. Thus,
the first step in palaeoclimatic studies becomes the construction of an age-depth
relationship that transforms the proxy measurements from the depth domain to a time
series.
However, it still remains to be resolved as to how the errors of radiocarbon
dating be effectively captured in the final proxy time series. Recent advances in
this area have shown an emerging consensus favouring the use of Monte Carlo
interpolation techniques. These methods typically involve approximate probability
distributions that are generated by using thousands of Monte Carlo age-depth models.
Despite their relative success and applicability, these methods have one primary
drawback: they assume that the calibrated ages have a Gaussian distribution. This is an
untenable assumption as the process of calibration - in which the measured 14C age is
related to the actual age using a standard 14C calibration curve - converts the simple
Gaussian error distribution of the 14C measurement into a complicated multimodal
error distribution as a result of the fundamental irregular nature of the calibration
curves.
We present a regression based Bayesian approach to this issue. Our method focuses
on the ultimate goal of arriving at a meaningful proxy time series and not on the
in-between stage of constructing an age-depth model. We suggest to employ the
conditional distributions of the measurements (of both the 14C ages as well as the proxies
with depth) and thereafter construct an estimator based on regression that provides
the distribution of the proxy time series. In this approach, we arrive at meaningful
results - such as the mean and standard deviation of the proxy - without having to
assume that the calibrated ages have Gaussian errors. The method is validated using
simulated data sets where the true values of the age and proxy are known. Moreover,
besides radiocarbon dating, the method is applicable to other dating methods as
well.
This novel approach based on a perspective of regression can open up newer possibilities
of tackling the issue of uncertainties in age-depth relationships and proxy measurements. |
|
|
|
|
|