![Hier klicken, um den Treffer aus der Auswahl zu entfernen](images/unchecked.gif) |
Titel |
Bayesian hierarchical models for regional climate reconstructions of the last glacial maximum |
VerfasserIn |
Nils Weitzel, Andreas Hense, Christian Ohlwein |
Konferenz |
EGU General Assembly 2017
|
Medientyp |
Artikel
|
Sprache |
en
|
Digitales Dokument |
PDF |
Erschienen |
In: GRA - Volume 19 (2017) |
Datensatznummer |
250143169
|
Publikation (Nr.) |
EGU/EGU2017-6870.pdf |
|
|
|
Zusammenfassung |
Spatio-temporal reconstructions of past climate are important for the understanding of the
long term behavior of the climate system and the sensitivity to forcing changes.
Unfortunately, they are subject to large uncertainties, have to deal with a complex
proxy-climate structure, and a physically reasonable interpolation between the sparse proxy
observations is difficult. Bayesian Hierarchical Models (BHMs) are a class of statistical
models that is well suited for spatio-temporal reconstructions of past climate because they
permit the inclusion of multiple sources of information (e.g. records from different proxy
types, uncertain age information, output from climate simulations) and quantify uncertainties
in a statistically rigorous way.
BHMs in paleoclimatology typically consist of three stages which are modeled
individually and are combined using Bayesian inference techniques. The data stage models
the proxy-climate relation (often named transfer function), the process stage models the
spatio-temporal distribution of the climate variables of interest, and the prior stage consists of
prior distributions of the model parameters. For our BHMs, we translate well-known
proxy-climate transfer functions for pollen to a Bayesian framework. In addition, we can
include Gaussian distributed local climate information from preprocessed proxy records. The
process stage combines physically reasonable spatial structures from prior distributions with
proxy records which leads to a multivariate posterior probability distribution for the
reconstructed climate variables. The prior distributions that constrain the possible
spatial structure of the climate variables are calculated from climate simulation
output.
We present results from pseudoproxy tests as well as new regional reconstructions of
temperatures for the last glacial maximum (LGM, ∼ 21,000 years BP). These reconstructions
combine proxy data syntheses with information from climate simulations for the LGM that
were performed in the PMIP3 project. The proxy data syntheses consist either of
raw pollen data or of normally distributed climate data from preprocessed proxy
records.
Future extensions of our method contain the inclusion of other proxy types (transfer
functions), the implementation of other spatial interpolation techniques, the use of
age uncertainties, and the extension to spatio-temporal reconstructions of the last
deglaciation.
Our work is part of the PalMod project funded by the German Federal Ministry of
Education and Science (BMBF). |
|
|
|
|
|