|
Oct. 16 - 18, 2000
Park City, Utah, US
(Immediately following GMS Course)
Oct. 31 - Nov. 2, 2000
University of Virginia
Northern Virginia Center
Falls Church, Virginia, US
(Just Outside Washington, DC)
PEST Course Description
PEST Course Objectives
PEST Instructor
What is PEST?
What is Nonlinear Parameter Estimation?
PEST Course Schedule
Practical Sessions
Cost
Location
Registration
PEST Course Description
This intensive short course will instruct participants in the automated calibration of environmental models and analysis of their predictive certainty. The
principal instructor for this course is the developer of PEST, the most versatile program available for automated model calibration. Recent additions to PEST functionality include quantification of the uncertainty associated
with predictions made by a previously calibrated model. While the course will include a thorough coverage of the theory and applications of the method of nonlinear parameter estimation to the calibration of different types of
models, there will also be a strong practical aspect of the course. Participants will gain hands-on experience in the use of PEST2000, Parallel PEST and the new Visual PEST.
Participants are encouraged to bring their own case studies. Ample time will be available during the evening sessions and on the third day to develop a
customized application of PEST to your case study under the guidance of the course instructor. The model must have ASCII input and output files and run in an MS-DOS environment. Please bring the model executable and all your
case files.
Anyone interested in model calibration, parameter optimization, or analysis of model predictive uncertainty should attend this course. Those new to PEST, as
well as those with previous PEST experience, will all benefit from the course.
To get the most out of the course, participants should have some modeling experience and should feel comfortable working directly with model input and output
files.
PEST Course Objectives
The three day training course on PEST covers the following:
- Theory of nonlinear parameter estimation
- Application of nonlinear parameter estimation to model calibration
- Analysis of uncertainty and non uniqueness in calibrated parameters
- Problem regularization
- Analytical and intuitive analysis of calibration residuals
- The effects of parameter uncertainty on model predictive uncertainty
- Use of the new PEST2000 predictive analyzer
- How to choose an appropriate level of model complexity
PEST Instructor
Dr. John Doherty. John is the author of PEST. He has worked for over 23 years in the water industry, first as a groundwater exploration geophysicist and then as a modeler engaged in the simulation of both groundwater and surface water systems. He has worked in both the public and private sectors and spent three years serving as a Research Fellow at James Book University of North Queensland, Australia. John now works for his own consulting and development firm.
John has had over five years experience in presenting short courses. He uses a wealth of visual aids that make his courses interesting and informative and tries
to ensure that participants have fun while they learn.
What is PEST?
PEST is a model-independent nonlinear parameter estimator. Since its inception five years ago, PEST has become the industry standard in calibration of
environmental models of all kinds. The two cornerstones of PEST's model-independence are:
- PEST communicates with a model through the model's own input and output files. Hence the model does not need to be recompiled to be linked to PEST; it can
be used with PEST exactly as it is.
- Though based on the Gauss-Marquardt-Levenberg method, the nonlinear parameter estimation algorithm used by PEST is uniquely robust and powerful having been
developed specifically for use with complex environmental models.
The model-independent nature of PEST allows modelers to calibrate their own tailor-made "composite models," built through assimilating one or more models
and appropriate pre- and postprocessing software into a single batch file. Enormous creativity can be exercised in the construction and calibration of these models.
For more information on PEST2000, click here.
What is Nonlinear Parameter Estimation?
In simple terms, nonlinear parameter estimation is a methodology whereby the arduous, labor-intensive and distinctly frustrating task of
multi-parameter model calibration can be carried out automatically under the control of a computer. The advantages of computer-aided model calibration include the following:
- the task is accomplished much more rapidly than by using manual methods,
- better parameter estimates are obtained,
- estimates of the uncertainties accompanying optimized parameters are produced as part of the calibration process,
- freed from the drudgery of having to undertake countless model runs "by hand," the modeler is able to inject more initiative into the calibration process
thus making this process a partnership between calibration technology and the art of the modeler,
- a greater understanding of the environmental processes simulated by the model can be gained through viewing model calibration as a sophisticated method of
data interpretation,
- an analysis of the effects of parameter uncertainty on model predictive uncertainty can be undertaken.
PEST Course Schedule
Day 1
Registration 8:30 - 9:00
Welcome/Introduction
- Morning Lecture 1 - Introduction to Nonlinear Parameter Estimation
- Mathematics of Nonlinear Parameter Estimation
- Stochastic Interpretation of Nonlinear Parameter Estimation Theory
- Parameter Nonuniqueness
- Analysis of Residuals
- Nonlinear Parameter Uncertainty Analysis
- Predictive Analysis
- Break
- Morning Lecture 2 - Basics of PEST
- PEST and Model-Independence
- Templates of Model Input Files
- Reading Model Output Files
- PEST Control File
- Requirements of a Model
- Construction of Composite Models
- Parallel PEST
- Visual PEST
- Lunch
- Afternoon Practical Session
- Use of PEST in Calibrating a Basic Storage Model. This exercise will illustrate many problems encountered with environmental modeling.
- Afternoon Lecture - Data Collection and Model Calibration
- Strategies for Data Collection
- Formulation of Observation Groups
- Balancing of the Objective Function
- Dinner Break
- Informal Evening Session
- Participants may introduce their models to course instructors. Calibration strategies for these cases and others will be discussed.
- Review previous material and answer questions.
Day 2
- Morning Lecture 1 - Use of PEST in Groundwater Modeling
- Overcoming MODFLOW Problems (including drying and re-wetting)
- Temporal and Spatial Interpolation of MODFLOW Outputs to Observation Times and Sites
- Calibration of MT3D
- Analysis of Residuals in Groundwater Context
- Assignment of Observation Weights
- Examples of Parameter Nonuniqueness
- Description of Pertinent Utility Software supplied with the course
- Morning Lecture 2 - Use of PEST in surface water modeling
- Problems in Using Nonlinear Parameter Estimation in Surface Water Modeling
- Methods of Overcoming Problems
- Problem Regularization
- Water Quality and Sediment Transport
- Model Suitability
- Description of Pertinent Utility Software Supplied with the Course
- Break
- Practical Session
- Using PEST to Calibrate a MODFLOW Model
- Lunch
- Afternoon Practical Session
- Using PEST to Calibrate HSPF
- Break
- Lecture - Sensitivity Analysis, Predictive Analysis, and Model Complexity
- Sensitivity Analysis
- Linear Propagation of Parameter Uncertainty to Predictive Uncertainty
- Examples of Parameter Nonuniqueness in Ground and Surface Water Modeling
- Effects of Parameter Nonuniqueness on Predictive Nonuniqueness
- Choosing an Appropriate Level of Model Complexity
- "Rapid Fire Re-Calibration" as a Method of Predictive Analysis
- Predictive Analysis as a Complement to the Use of Distributed Parameter Models
- Dinner Break
- Informal Evening Session
- Finish practical session examples
- Work on individual problems with assistance of course leaders
- Informal discussion on questions or issues from the day
Day 3
- Morning and Afternoon
- Participants will be assisted in solving their particular calibration problems. Those who did not bring modeling problems may team up with others, or
work on problems supplied by the instructors.
- Informal lectures to discuss problems raised by the participants.
Practical Sessions
Five practical problems have been prepared for participants to work on during the course. Participants can choose those problems that best satisfy
their own needs or work on a problem of their own under the guidance of the course instructor. Prepared practical problems are as follows.
- Use of PEST in calibrating a basic storage model. Though simple, this exercise will illustrate many of the problems encountered in environmental modeling of all kinds. As they carry out this exercise, students will gain hands-on experience in the use of PEST, Parallel PEST and Visual PEST. They will also learn about parameter correlation, parameter uncertainty and the repercussions of parameter uncertainty on predictive uncertainty.
- Using PEST in the calibration of a MODFLOW model. The model will be used
independently of any commercial graphical user interface. The utility software supplied with the course will be used to demonstrate to participants that, when the occasion demands it, they can extend the calibration
functionality offered by commercial MODFLOW PEST interfaces to far more complex calibration problems. PEST usage in the context of a number of commercial MODFLOW graphical user interfaces will be demonstrated.
- Using PEST to calibrate SWIM. SWIM is a Richards-equation-based
unsaturated zone water movement model written by the Australian CSIRO. By working through this informative practical exercise, participants will learn how to use PEST in the design of a field experiment, and in exploring
the effects of parameter nonuniqueness on predictive nonuniqueness. It will be discovered that the repercussions of parameter uncertainty on predictive uncertainty are not always easy to quantify without the use of PEST's
new predictive analyzer.
- Using PEST to calibrate HSPF. HSPF is a popular hydrologic
quantity/quality simulation model produced by the US EPA. The surface water modeling utility software supplied with the course will be used to facilitate PEST setup for a HSPF model involving hundreds of flow observations.
Participants will learn how to extend the same techniques to the calibration of other surface water models as well.
- Using PEST to calibrate 3-PG. 3-PG is a generalized model of forest
productivity based on simplified concepts of radiation use efficiency, carbon balance and partitioning. Participants will learn how to calibrate 3PG using PEST. A number of 3PG models based on forest stands at different
sites,will be calibrated simultaneously making maximum use of available data to ensure parameter consistency between sites.
Cost
The cost of the 3-day course is $995. Registration includes the following:
- Instruction and hands on use of PEST with a personal computer
- Course notes
- A free copy of PEST2000
Location
Oct. 16 - 18, 2000
Shadow Ridge Hotel
Park City, Utah, US
800-451-3031
A block of rooms has been reserved for training course attendees at a rate of $72 per night.
Oct. 31 - Nov. 2, 2000
University of Virginia
Northern Virginia Center
At the West Falls Church Metro Station
7054 Haycock Road
Falls Church, Virginia, US
(Just Outside Washington, DC)
703-536-1116
Hotels
Virginia:
Holiday Inn at Tyson's Corner
1960 Chain Bridge Rd
McLean, VA 22102
(703) 893-2100
DoubleTree Hotel at Tyson's Corner
7801 Leesburg Pike (Rt 7)
Falls Church, Virginia
(703) 893-1340
Download a map to University of Virginia course (1,170 Kb - pdf format).
Click here for a detailed map and instructions to University of Virginia course (gif format).
Click here for parking instructions (gif format).
PEST Registration
Download Registration Form (12 Kb - pdf format)
|
