Sack Lunch Seminar (SLS)

Reconstructing pre-historic temperatures from natural proxies:
statistical methods in paleoclimate research



Reconstructing the space-time evolution of a climate field from
incomplete instrumental and proxy time series poses both scientific
and statistical challenges. Over the last two decades, the statistics
community has made major advances in the modeling and analysis of
space-time processes. Many of these advances have not yet been applied
to the paleoclimate reconstruction problem, and doing so has the
potential to improve our understanding of past climate. I begin by
outlining both the scientific and statistical challenges involved in
reconstructing past climate, and discuss shortcomings of popular
approaches that have been used to overcome them. In particular, I
address the incongruous results presented in a forthcoming Annals of
Applied Statistics paper by McShane and Wyner. I then present a
unifying, hierarchical space-time modeling framework for the
paleoclimate reconstruction problem, and indicate how modern
statistical expertise can be brought to bear upon the problem. Within
this framework, the modeling assumptions made by a number of published
methods can be understood as special cases, and the distinction
between modeling assumptions and analysis or inference choices becomes
more transparent.



I finish by discussing how the length of the reference interval
influences the calculation of climate anomalies. As climate data sets
are generally composed of time series of differing lengths, the
reference interval is often shorter than the length of the data set,
and the choice of reference interval can affect the second moment
properties (i.e., spatial variance) of the data set. I present a
Bayesian multi-factor ANOVA model for the standardization problem, and
apply it to the Climate Research Unit's gridded temperature anomaly
product.