High Dimensional Multiproxy Paleoclimate Reconstructions: New Perspectives
Abstract
The need to understanding centennial-scale climate variability is very compelling and depends on the availability of datasets that are accurate and have both broad temporal and spatial coverage. Due to the limited period over which instrumental measurements are available, temperatures of the past are generally reconstructed by using information from paleoclimate proxy records. To this end, various climate reconstructions methods have been proposed in the literature in order to infer past climate from multiproxy networks, each with their respective strengths and weaknesses. In this talk, we investigate new theoretical tools for high dimensional multiproxy paleoclimate reconstructions. By exploiting recent advances in high dimensional statistical inference, we propose computationally efficient methods that tackle the reconstruction problem in a statistically optimal way. By framing the reconstruction problem in a mathematically rigorous way, we determine which of the proposed methods in the literature are optimal in a well-defined sense. This allows us to put forward different and flexible variants of existing methods, each being optimal under different assumptions. We also proceed to develop theoretical techniques in order to undertake uncertainty quantification for the aforementioned reconstruction approaches. Finally, the methodology is applied to both real and pseudoproxy data to illustrate its performance.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2012
- Bibcode:
- 2012AGUFMPP12B..08R
- Keywords:
-
- 1942 INFORMATICS / Machine learning;
- 1952 INFORMATICS / Modeling;
- 1986 INFORMATICS / Statistical methods: Inferential;
- 1990 INFORMATICS / Uncertainty