Douglas Nychka

Voice (303) 497-1711, FAX (303) 497-2483 Cell (business hours) (303) 725-3199 Email nychka "at" ucar "dot" edu
CV Everything: html (pdf) (LaTeX) NSF style (but topical): html (pdf) (LaTeX) A one pager: (pdf) (latex)

I am the Director of the Institute for Mathematics Applied to Geosciences (November, 2004 - present) (IMAGe) and also a Senior Scientist in Geophysical Statistics Project (GSP). My main task is to enrich the scientific and educational activity at NCAR through mathematical methods and models. Also, I use the large scientific projects at NCAR to engage the mathematical science communities in new applications and to motivate new mathematics. Before directing IMAGe I was the GSP project leader (August,1997 - June, 2006)

Some current personal research interests are nonparametric regression (mostly splines), statistical computing, spatial statistics and spatial designs. I was a Math/Physics undergrad at Duke, with Robert Wolpert as my advisor, and received my Ph D ('83) from the University of Wisconsin under Grace Wahba. Although my original thesis work was on splines and inverse problems, over time I have become more interested in spatial statistics and Bayesian methods for curve and surface fitting. This change of course came about by the chance to work at the National Institute of Statistical Sciences (1994-1996) on an EPA grant. During this time I wrote the first version of the Krig function (at about 11PM the night before an EPA presentation) in the S language. Krig was first part of the funfits package, coauthored with Perry Haaland and Barb Bailey and it is now in the fields package -- completely different code but still with the same name. I guess this is like my career!

I came to GSP/NCAR in 1997 after spending 14 enjoyable years as a faculty member in the Statistics Department at North Carolina State University. This was supposed to be an interesting leave of absence for just two years ... Well, people will not dispute that time moves differently in Boulder. My most recent research interests are a mathematical statistics project with Eva Furrer on the large sample properties of geostatistics estimators and, with Bo Li and Caspar Ammann, on an application of inverse methods and hierarchical models to the reconstruction of past climate.

My wife Helen Nychka is a speech language pathologist serving autistic children. We have three kids: Allie (19), Everett (16) and Grace (13). Some interests outside work are skiing the trees next to Little Ten at Mary Jane, trail running -- and remodeling bathrooms (the one who dies with the most tools wins)! Everett recommends that you check out New Schoolers for the latest in what can be done with skis. Stayed tuned for the Mary Jane Project to be offically launched later this season.

Come visit us at the Mesa Lab

NARCCAP Companion

Useful datasets and source for interpreting the NARCCAP model output..

Daily Surface Ozone Eastern US 1995-1999

An example of a nonstationary spatial process and spatial extremes .

A case study dataset on Colorado climate:

Regional climate model and observed precipitation data for the Colorado Front Range.

Some beta R packages

fields Fitting curves and surfaces to data

fields is a collection of programs based in R for curve and function fitting with an emphasis on spatial data and flexible covariance functions for Kriging.

The major methods implemented as R functions include:
Tps: Thin Plate spline regression
Krig: Spatial process estimate (Kriging)
All Kriging functions allow you to supply a covariance function that is written in R code.
mKrig: micro Krig Fast spatial prediction that can take advantage of compactly supported covariance functions
cover.design: Finds a space filling design
as.image, image.plot, quilt.plot: Some useful functions for working with data on 2-d grids.
sreg, qsreg : 1-d smoothing splines and 1-d quantile splines

There are also generic functions that support these methods such as:
plot diagnostic plots of fit
summary statistical summary of fit
surface graphical display of fitted surface
predict, predict.se evaluation fit and prediction error at arbitrary points

Beta Version for the Field Guide:
fields_4.3.tar.gz 22APR2008
This is a working version that will be the basis of a book in the UseR! series with Reinhard Furrer and Steve Sain. This may be more recent than the version posted to CRAN.

Estimated 20 year return levels for daily rainfall in the Colorado Front Range.

nnreg Neural Networks Package

Estimates a function using a single hidden layer neural network by nonlinear least squares. The fitting algorithm is both robust and accurate. Has supporting functions for diagnostics, GCV and graphing.
Doug Nychka (contact), Stephen Ellner and Barbara Bailey nnreg_1.1.tar.gz (59K)

LENNS Lyapunov Exponents fit by Neural Networks

Fits nonlinear autoregressive maps to multivariate time series data and estimates global and local Lyapunov Exponents. Doug Nychka (contact), Stephen Ellner and Barbara Bailey lenns_1.0.tar.gz (25K)