Ten Lectures on Statistical Climatology

                                          

by Doug Nychka, NCAR.

August 6-10, 2012
175 Johnson Hall
University of Washington, Seattle

Local information.

Participant list.

Monday August 6

8:20-8:50 Registration

8:50-9:00 Peter Guttorp, University of Washington: Opening remarks
9:00-10:00 Douglas Nychka, NCAR: Lecture 1. The Earth’s climate system and climate change

File 1. File 2.
This lecture contains basic information about atmospheric circulation, climate, and climate change. The role of greenhouse gases and other forcings is emphasized.

10:00-10:15 Discussion

10:15-10:45 Coffee break

10:45-11:45 Will Kleiber, NCAR: Daily Spatio-Temporal Stochastic Weather Simulation

File.

11:45-12:00 Discussion

12:00-1:30 Lunch break

1:30-2:30 Nychka Lecture 2. Climate models

File 1. File 2. Climate movie. L96movie. L96forced. L96Sens
The basic equations of fluid dynamics are described, and their application to climate systems. The implementation of computer models to solve the equations is described with a view to helping a statistical audience understand the strengths and weaknesses of such physically based models. Scenarios (i.e., guesses of future policies, emissions etc.) are needed to compute projections of future climate. Global climate models are relatively coarse, and in order to make regional projections it is necessary to develop regional models, or use other forms of downscaling global models to a finer resolution.

2:30-2:45 Discussion

2:45-3:15 Coffee break

3:15-4:30 Roundtable discussions

1. How to develop a research plan (discussion leader Jennifer Hoeting)

2. Outreach and policy assistance in statistical climatology (Peter Guttorp)

3. Research collaboration across disciplines–difficulties and advantages (Mikyoung Jun)

4:30-5:30 Poster setup (225 Kane Hall)

5:30-8:30 Poster session and welcome reception (225 Kane Hall)

Bessac. Chong. Haug. Mesquita. Sansom. SchliepWeller.

Tuesday August 7

9:00-10:00 Nychka Lecture 3. Smoothers
File. (updated)

10:00-10:15 Discussion

10:15-10:45 Coffee break

10:45-11:45 Tamara Greasby, NCAR: Variability in Annual Temperature Profiles: A Multivariate Spatial Analysis of Regional Climate Model Output

File.

11:45-12:00 Discussion

12:00-1:30 Lunch break

1:30-2:30 Nychka Lecture 4. Spatial processes and geophysical data
Geophysical data typically come as spatio-temporal fields living on (at least approximately) a rotating sphere. Among the tasks for the statistician is estimation (e.g., linear prediction, also called Kriging) of the field at unobserved locations in space and/or time, assimilating measurements from different sources collected at different spatial and temporal scales, and adjustment for changes in instrumentation, instrument location etc. The processes involved are nonstationary in space and time, and usually space and time are not separable, and our statistical models need to take this into account.

File (updated). R code 1. R code 2. North American data.

2:30-2:45 Discussion

2:45-3:15 Coffee break

3:15-4:30 Roundtable discussions

1. Career choices and challenges (Grace Chiu and Veronica Berrocal, discussion leaders)

2. Have hammer, want nail or the other way around? (Peter Guttorp)

3. Issues in climate data (Michel Mesquita)

Wednesday August 8

9:00-10:00 Nychka Lecture 5. What is a spline?
Common tasks in statistical climatology are estimation of trends in space and time. Linear models are common in climate science, but nonparametric nonlinear smoothers are better descriptions of the complicated functional forms of these trends. There are many classes of splines, many defined as solutions to constrained optimization problems.

File.

10:00-10:15 Discussion

10:15- 10:45 Coffee break

10:45-11:45 Dan Cooley, Colorado State: Modeling tail dependence and performing prediction via the angular measure

File.

11:45-12:00 Discussion

12:00-1:30 Lunch break

1:30-2:30 Nychka Lecture 6. Some large sample theory for Kriging and splines
The large sample theory for Kriging and splines were developed by Stein and Wahba, respectively. The theory makes clear what assumptions are most important in spatial analysis and unifies the methods of kernel smoothing, variational methods and Kriging.

File.

2:30-2:45 Discussion

2:45-3:15 Coffee break

3:15-4:30 Computing exercise 1.

Mesquita slides. R code (see Thursday’s session).

Thursday August 9

9:00-10:00 Nychka Lecture 7. Spatial models for large data sets
For large data sets the statistical computations to estimate parameters, form spatial predictions and to quantify the uncertainty may not be feasible given current computational resources. There are different approaches to deal with this problem, such as developing sparse matrix approximations to the covariances, or using Markov random field approaches. In either case these practical solutions modify the spatial model and it is important to understand how they change the statistical assumptions.

File.

10:00-10:15 Discussion

10:15-10:45 Coffee break

10:45-11:45 Mikyoung Jun, Texas A&M: Nonstationary  cross-covariance models for multivariate processes on a globe

File.

11:45-12:00 Discussion

12:00-1:30 Lunch break

1:30-2:30 Nychka Lecture 8. Functional analysis of regional climate experiments
This first case study deals with what is in essence an analysis of variance for regional climate models, driven by boundary conditions from global models. This example is based on results from the North American Regional Climate Change and Assessment Program (NARCCAP), to date the largest factorial experiment examining the pairing of different regional and global models.

File.

2:30-2:45 Discussion

2:45-3:15 Coffee break

3:15-4:30 Computing exercise 2.

LatticeKrig. R code.

Friday August 10

8:30-9:30 (note different start time) Nychka Lecture 9. Reconstructing paleoclimate with Bayesian hierarchical models
This case study shows how to combine different climate proxies, such as tree rings and ice cores, to reconstruct with appropriate uncertainty measures the prehistorical climate. The original motivation for this work is the much publicized “hockey stick” controversy and this lecture frames those issues as a statistical problem.

File.

9:30-9:45 Discussion

9:45-10:15 Coffee break

10:15-11:15 Veronica Berrocal, University of Michigan: Regional climate model assessment using statistical upscaling and downscaling techniques

File.

11:15-11:30 Discussion

11:30-1:00 Lunch break

1:00-2:00 Nychka Lecture 10. Data assimilation for climate model prediction and tuning
An important role for statistics in climate science is to assist modelers in tuning parameters of models and model predictions using appropriately chosen data. For simple models this uses Bayesian melding and similar approaches, while for complex models statistical emulators are developed.

File.

14:00-14:15 Discussion

14:15-14:45 Coffee break

14:45-15:00 Peter Guttorp, University of Washington: Closing remarks

 

This entry was posted in Past events. Bookmark the permalink.

Leave a Reply