CISL Annual Report 2005 > CISL research accomplishments

Geophysical Statistics Project accomplishments

The Geophysical Statistics Project (GSP) encourages the application of statistical analysis and the development of new statistical methods in the geophysical sciences. To fulfill this goal GSP also engages in basic statistical research including mathematical statistics and probability theory (see current NSF proposal). Most of the statistical methods developed in GSP are distributed to the community in the form of open source packages in R, a software environment for statistical computing and graphics.

Currently GSP has a strong training component, supporting between four to six statistics postdoctoral visiting scientists and several statistics graduate students during the summer (see GSP staff).

A highlighted project for this year is a method for extrapolating precipitation extremes to areas where station data are not available. Besides facilitating a revision of the U.S. Weather Service precipitation atlas, these Bayesian methods also allow for the analysis of extremes in regional climate model experiments.

The fourth report of the Intergovernmental Panel for Climate Change has generated a set of climate model experiments from more than 20 different atmospheric ocean global circulation models (AOGCMs). GSP has several projects that support the analysis of these numerical data. A statistical framework has been developed to combine the different model results and produce an integrated estimate of climate change at regional scales. Also, the methods used to compare model simulations of the 20th century with observations has been improved using better statistical estimates of the model variability. Finally, as a complement to the limited runs of high-resolution AOGCMs, GSP has worked with intermediate climate models to estimate ranges for key climate parameters such as the climate sensitivity (the increase in temperature due to a doubling of atmospheric carbon dioxide).

Statistical methods for a precipitation atlas Using Bayesian hierarchical models and extreme value theory, a practical method has been developed to estimate the return levels of extreme precipitation from irregularly observed station data. This method improves on existing techniques used by the U.S. Weather Service by providing measures of uncertainty in the maps of extremes and also avoiding some subjective choices in the analysis. The figure illustrates a pilot study for the Colorado Front Range for the 25-year return level for daily precipitation (the level of 24-hour precipitation that has a probability of 1/25 of occurring in any year.) One new feature in this approach is the spatial analysis of the observed data under transformed climate coordinates that incorporate the influence of topography on precipitation. The right plot indicates how stations cluster with respect to elevation and total annual precipitation, and the left figure is a map of the estimated 25-year return level (in cm) for daily precipitation when the analysis is mapped back into geographical coordinates.