Regional climate model animation
An animation ncepjan2000.mp4 (6.4Mb) of one of the NARCCAP regional climate models being forced by the global (and lower resolution) NCEP reanalysis field over the month of January 2000 (courtesy of Stephan Sain). See also NARCCAP for information of the sources of these data and model output.
The NCEP reanalysis is a gridded data product for the atmosphere that uses a weather forecast model with observations to give a physically balanced estimate of the physical variables of the atmopsphere on a 3-d grid. In this animiation the surface precipitation is indicated for NCEP although the actual variables driving (known as forcing) the region model in the interior are moisture, momentum and energy fluxes. This model particular regional model (ECPC) is unusual for a regional model in that it not only uses the fluxes at the domain boundaries but also nudges its interior values to the NCEP field. The finer resolution and more detailed structure in the regional model is considered to be a more accurate simulation of weather processes such as rainfall. The cliamte predicted by the model is found by simulating a long period of weather and averaging the values over time. A numerical simulation where a regional model is forced by observed weather, such as the NCEP reanalysis, is a test of how well the model can reproduced detailed climate under observed conditions. To determine possible changes in climate for the future the NCEP forcings are replaced by the fields from a global climate system that uses future scenarios of greenhouse emissions and other changes to simulate a different set of set of large scale conditions.
Where are statisticians in the Earth System?
An invited talk at JSM08 Denver, August 2008.
Climate Past, Climate Present, Climate Future, A tale
told by a statistician
A public lecture given at the 7th World Congress in Probability and Statistics, Singapore, July 2008.
About the Santa ensemble:
The Santa images are taken from an article: Jason Salavon (2004). Artist Project: 100 Special Moments, Cabinet 15, pp 77- 81.
Is part of a series of art projects, 100 Special Moments by Jason Salavon. His completed work is Kids with Santa , a composite Santa image that I understand to be a weighted sum of mean and median ensemble pixel values. I was surprised that Salavon's web page does not explicitly show the ensemble or give the details of his process. The construction, even though it is fairly complicated, is not referenced and I interpret this to mean that the process of construction and the underlying sample of 100 images is not important to enjoying the work. But it does raise a caution about taking my illustration in the talk too literally. It is not clear how Salavon choose the 100 Santa images, what digital preprocessing he might have done to align images, adjust color levels, etc. So the 100 Santa's may not be the kind of representative and random ensemble one would strive for in a statistical context. The 100 images leading to the santa estimate of the "central tendancy" are reproduced in the article by Salavon in Cabinet Magazine, Issue 15.
The movie: (Quicktime) 32M (First two plots are ensemble mean and sd fields, remaining 6 plots are the first 6 members. The animation brings in each observation sequentially having sorted by location latitude and updates the ensemble. The result of each update is the conditional distribution of the full ozone fields given the observations brought in so far. The first frame are the ensembles draw from an ozone "summer climatology".) The calculations for a large ensemble approximate those needed to update the conditional multivariate normal distribution but instead of keeping track of the conditional covariances and mean the information is propagated by through the discrete distribution represented by the ensemble. Any mean of covariance is determined by the sample statistics based on the ensemble members. The R code available below makes this calculations explicit.
To construct the frames for this movie, transfer the files README.txt
fun.R setup.R from the directory EKFmovie.
Follow the instructions in the
README.txt file. These animations may
seem complicated but there are really just a few tricks (like everything
else I do!). Basically a for loop to write out each frame as a separate
jpeg image. A key step is that I use Quicktime on a mac to
image sequence into a quick time movie, surprisingly this is just a few
clicks. NOTE: These functions require the R statistical environment and
The R statistical environment for data analysis and graphics.
Short overview of R given to CISL exec. August, 2005.
Earlier talks focusing on univariate analyses.
WEBCAST/Climate and Health Summer School, NCAR, July, 2004 (PDF 1.2M)
Stanford, October 2003 (PDF 710K)
Supporting paper in Journal of Climate (PDF) 2.4M ,
Software and examples from this analysis
To reproduce the animations and figures,
browse this talks directory and transfer the
the R source files script.R fun.R, plots.R, setup.R .
( you will also need the
The script.R will run the animation plots.R draws all the figures used in this talk..
Large Spatial Problems
A Multiresolution approach to nonstationary and efficient computing (PDF) 3.6M .
Computational Environmetrics Workshop, Chicago, October 2004.
See also the older talk on wavelet based
covariances that has some data examples. Software for this model in
R will be posted soon!
The matrix reloaded: Computations for large spatial data sets. (PDF 830K)
SAMSI/GSP Workshop on spatiotemporal statistics June, 2003, Boulder, CO
Santa Cruz and Berkley October 2003
Discussion to Michael Stein's Medallion Lecture. JSM Toronto,CA 8/12/2004 (PDF) 174K
Statistical methods for relating mortality to temperature
WEBCAST/Climate and Health Summer School, NCAR, July,2004
NCARs projects for the RIPS student team at the Institute for Pure and Applied Mathematics (PDF 1.7M)
A introduction and case study for extremes
Summer Institute at the Center on Global Change, Duke University, June 15 2004
A shorter talk (PDF 2.4M) given at the Greybill Conference, Fort Collins, June 17,2004
Materials from the software demo at Duke. Load the R data object RDataDEMO and follow the script scriptDEMO
To reproduce the figures and analysis in R
grab the (TEXT) files
(optional: script.r This is an optional batch file that draws all figures and writes them to separate pdf files. ) You also need FIELDS spatial package to run these. In an R session load the source files (you have to do this just once)
library(fields) # load the fields library source("example.setup.r" ) # creates data objects and fitted surfaces source("example.figs.r" ) # loads the figure functions # OK ready to draw the figs make sure the fields library is loaded! fig1() # will draw the first figure etc. ... figoz3() # will draw the last one.You only have to source the two files example.setup.r and example.figs.r once provided you save the workspace. But you will always have to attach the fields library. The numbering of figures may not match the order in the talk. See the interesting figure fig6b not mentioned but related to fitting an anisotropic model based on lon/lat coordinates. figoz3 may also be of interest it does the conditional simluation of the ozone fields.
source("all.source.r"); source("example.r"); source("example2.r")( you will need the FIELDS spatial package to runs these.)
source("setup.r") fig8() fig9()to create and draw the last two figures (in color) from the paper. See the README file for more details.
To reproduce all the figures and analysis used in this talk you can grab the R data workspace .RData (1.6M). Check to make sure this file is named .RData but that it does not overwrite a workspace file that you might already have created. You will also need the FIELDS spatial package to run these. Put the file .RData into a (new) subdirectory, start up R,
library(fields) # load the fields library fig1() # will produce the first figure used in the talkThe data objects colo.elev colo.id colo.loc colo.names colo.tmax colo.tmin have the monthly station data and related information, the time span is 1895-1997. The data are matrices where rows index month and columns index the stations. This is a (small) subset of Observed monthly precipitation, min and max temperatures for the conterminous US 1895-1997
To reproduce the R analysis from these talks browse the directory FHDA ( FHDA: fourth highest daily averge) and transfer the ASCII files
Mpeg animation of Lorenz dynamical system.
Mpeg animation of Community Climate Model, 300 mb stream function. Top plot is the complete field bottom is the deviations from the mean field (anomalies).