STOR 890 Spring 2009

STOR 890, SPRING 2009
ENVIRONMENTAL STATISTICS
Instructor: Richard L. Smith

This page was last updated July 6, 2009.

The actual presentations have been deleted to save disk space, but if anyone would like to see the originals, please email me.

Changryong Baek write-up
Petro Borysov write-up
Chris Cabanski write-up
Jessi Cisewski write-up
Jennifer Clark write-up
Jennifer Clark appendix
Emil Cornea write-up
Scott Hauswirth write-up
Beth Horton write-up
Soyoung Jeon write-up
Sungkyu Jung write-up
Jordan Kern write-up
Valmik Khadke write-up
Xuan Li write-up
Feng Liu write-up
Xin Liu write-up
Eric Lock write-up
Ruiwen Zhang write-up
Su Zhang write-up
Zhitao Zhang write-up
Yingqi Zhao write-up

Summary of Classes 1-3 (Jan 13, 15, 20) and part of Class 4 (Jan 27): In these classes I talked about a forthcoming JASA paper on constructing probabilistic climate projections using ensembles of climate models, and also referred to a book chapter that described the same and related work in more tutorial fashion. In particular, Appendix A of the book chapter discussed basic methods of Bayesian statistics, including the Gibbs sampler and Hastings-Metropolis sampler. I also cited a recent note on climate projections for North Carolina as an example of how these techniques could be applied to answer climate change questions in specific regions, and mentioned the PCMDI website where a large volume of climate model output data is available for public download and analysis. Although I haven't yet mentioned it in class, I should also point towards the CRU website (Climatic Research Unit, University of East Anglia) which holds many observational datasets, as well as providing further links with other websites storing climate data. Two related references are a paper by Reinhard Furrer and co-authors where they discuss application of some similar ideas in the context of a fully spatial-temporal dataset, and a paper by Tebaldi and Sanso on joint modeling of temperature and precipitation.

In class 4 (Jan 27), I started to discuss the NMMAPS data website which is used to examine the relationship between air pollution exposure and various adverse health coutcomes including death. I am including here a link to the 2004 JAMA paper that played a major role in EPA's 2008 revaluation of the ozone standard; I also recommend exploring Francesca Dominici's website (from which the above link to the JAMA paper was taken) for much more information about NMMAPS and the research it has led to. Here is a Simple program to evaluate health outcomes in one city and a more complicated program that is intended to reproduce the individual-city analysis of the 2004 JAMA paper (based partly on code in the NMMAPS data website lined above). The output from that program is stored here (Cols 1-2: parameter estimate and standard error for the "all year" ozone-mortality relationship; cols 3-4: parameter estimate and standard error for the "summer only" ozone-mortality relationship; each computed for all the 108 NMMAPS cities except where ozone data are not available ("NA")). Note that both these programs presume that you have pre-loaded the NMMAPS dataset into R, which can be done either directly from their website, or within R by clicking on "Packages", then "Install package(s)", and scrolling down the list of available packages to "NMMAPSlite". I'm also linking a 2007 seminar in Biostatistics where I discussed some of my own work on these analyses. For the next class, I recommend looking as a paper by Everson and Morris that discusses hierarchical Bayesian analysis for this kind of dataset.

Updates February 3: here are two versions of a sample tlnise program for computing hierarchical estimates. The second program uses this file that indicates in which of the 7 NMMAPS regions each city lies.

Link to paper by Crooks et al.

Link to notes on Air Pollution Epidemiology (updated Feb 5 but still work in progress!)

Beginning Feb 10, we started discussing Detection and Attribution in Climatology. Here is the current version of my presentation on this topic (this version Feb 12 but still being updated!) For the moment, I've left this topic incomplete but will return to it later.

Beginning Feb 17, we started discussing spatial statistics. The first part of the presentation covered basic motivation, background on spatial processes, estimation, kriging, examples (classes of Feb 17, Feb 18, Feb 24, Feb 26).

The classes of March 3 and March 5 we spent talking mostly about packages and computation (see links below).

The classes of March 17, March 26, March 31, April 2 were spent talking about nonstationary processes and then spatial monitor design, covered in this presentation. (We didn't cover the "lattice models" sections.)

All these topics are covered in more detail in my full length notes on environmental statistics

The class of April 7 covered Zhengyuan Zhu's "two-stage" approach to spatial design. I hope to post some notes on that later.

Then, beginning in the last part of the class on April 7 and continuing through the classes of April 9, April 14 and April 16, we covered Extreme Value Theory . Much of this material is covered in RLS review paper (2003) while the actual presentations are here

Notes from Malta short course (thanks to Eric Gilliland - see also "Extremes Toolkit" below):
Main notes
Practice examples
Sample R code

The last two classes (April 21, 23) we returned to the theme of detection and attribution . The presentation from those two classes is here

Papers referred to in class

Cressie (1989), The American Statistician 43, no. 4, 197-202
Handcock and Stein (1993), Technometrics 35, no. 4, 403-410
Holland, de Oliveira, Cox and Smith (2000), Environmetrics 11, 373-393
Holland, Caragea and Smith (2004), Atmospheric Environment 38, 1673-1684
Smith, Kolenikov and Cox (2003), J. Geophys. Res. 108, D24, 9004, doi:10.1029/2002JD002914, 2003