PEST: Two Day Course with Optional Third Day
17, 18 and 19 October 2019
Beach Hotel, Port Elizabeth
Principles and Practice of Model Calibration and Uncertainty Analysis
The course has two purposes. One of these is served on days 1 and 2 of the course. The third is served on the optional third day of the course.
The first two days of the course will be devoted to explaining the principles and algorithms that underpin model calibration and calibration-constrained uncertainty analysis. Important by-products of this exploration are insights into what modelling can and cannot achieve, and what should, and should not, be asked of modelling when undertaken for decision support. The presentation will be informal, with many real-world examples. There will be plenty of time for discussion; however no hands-on modelling exercises will be undertaken. Hence these first two days will prove useful not just for modellers, but for those who rely on models for decision-support, or who are stakeholders in decisions that are made on the basis of modelling.
The third day of the course is for modellers who wish to learn more about using programs of the PEST and PEST++ suites in their modelling work. It will include further details of PEST and PEST++ algorithms and behaviour, as well as advice on practical use of these programs. Participants will install these programs, as well as files for hands-on exercises, on their laptops at the beginning of the day.
The course is presented by John Doherty. John is the 2019 Darcy lecturer. He is also the author of the PEST suite of software. He has worked in the water industry as both a geophysicist and a modeller in a career spanning over 40 years. He has been employed by government, academic and consulting institutions. He presently runs his own company, Watermark Numerical Computing (of which he is the sole employee).
Who Should Attend?
The first part of the course (i.e. days 1 and 2) is not just for modellers. It will also be of interest to those who commission the building of models, and to those who are stakeholders in model-based groundwater management. Participants will become familiar with a range of model-value-adding software. At the same time, they will also gain important knowledge of what modelling can and cannot achieve. This will enable them to explore whether innovative modes of model usage can provide benefits for decision support that are presently unrealized by present-day modelling practice.
The optional second part of the course (i.e. day 3) is targeted at those who actually wish to use PEST and PEST++ software in their modelling work.
Topics covered in this day will include the following:
The role of modelling in the decision-making process;
The need for data-assimilation;
Expert geological knowledge and geostatistics;
Use of Bayes equation to explain pre- and post-calibration model predictive uncertainty;
The Jacobian/sensitivity matrix;
Insights available through singular value decomposition;
The null space;
Calibration as a projection operator;
Why post-calibration parameter and predictive uncertainty can be high;
The need to work in a highly-parameterized space;
The role of regularization;
Parallelization of model runs;
Overview of programs of the PEST and PEST++ suites.
Topics covered on this day will include the following:
The effect of model defects on pre/post-calibration model predictions;
Creative formulation of a multi-component objective function as defence against model defects;
When to calibrate and when not to calibrate;
Pilot points as a parameterization device – benefits and drawbacks;
Uncertainty quantification through iterative ensemble smoothers;
Choice between a simple or complex model;
Direct hypothesis-testing using models;
Calibration of surface water and land use models;
Global sensitivity analysis;
Linear uncertainty, identifiability and data worth analysis;
Decision optimization (under uncertainty);
Day 3 (Optional)
Through practical work undertaken on this day, course participants will gain experience in the following:
Specifications of PEST and PEST++ input datasets;
Utility support for construction of these datasets;
Adding regularization to a PEST input dataset;
The role of covariance matrices in regularisation;
Linear and nonlinear uncertainty analysis using PEST-suite programs;
The PLPROC PEST support utility;
Using PESTPP-IES (the PEST++ iterative ensemble smoother) for uncertainty analysis;
SVD-assist as a mechanism for reducing model runs;
Data space inversion.
Day 1 & 2 : R4000.00 (VAT 15% = R4600)
Day 1 - 3: R5500.00 ( VAT 15% = R 6325)
To register, click here.