Groundwater Model Calibration using PEST

Course presentation

• Introduction of the course

• Who are you and why are you attending?

S1-2: Review of fundamental concepts

• Flow equation (where everything starts from)

• Pumping test interpretation (how can it help in a model?)

• Analysis and processing of hydrogeological dataset (to become sounding model targets)

• Semivariogram (exercise with Excel)

• Regionalization of field data (exercise)

• Finite difference solution (exercise with Excel)

• Introduction to MODFLOW-USG and the CVFD solution

S3: Case study – Numerically enhanced conceptual modelling

S4: Overview of calibration and uncertainty analysis using PEST

• Metrics of a “good model”

• What is calibration?

• Modelling for decision-making

• The need to assess risk

• What should be expected of a model and what should not be expected

• Overview of PEST capabilities

• Overview of PEST++ capabilities

• Some real-world examples

 

09:00 - 13:00
S1: Getting familiar with the command line

• The command line... why we still need it

• Windows settings to make things easier

• Execution of general tasks through the prompt line (exercise to check the proper settings of participants’ computers and do a little practice)

S2: Waiting for PEST... 

• Structure of some Modflow input/output files

• PEST input/output files

• Executable files... “Seek, and ye shall find”

• Hybrid use of PEST

14:00 - 18:00
S3: Traditional Parameter Estimation

• Well-posed inverse problem

• The Jacobian matrix – why it is needed and how to calculate it

• How to define an objective function

• How to reduce an objective function

• Observation weights

• Prior information

• Parameter transformations

• Post-calibration parameter uncertainty

• Some important PEST settings

• Serial PEST and Parallel PEST 

• BEOPEST and PEST_HP

• Dealing with problematical models

S4: Workshop (GUI + cmd) PEST, BeoPEST and PEST_HP settings

A model is given, participants will have to set up PEST through the GUI, make a small change to the PEST control file and run PEST from the command line. Parameters to be calibrated will be hydraulic conductivity zones. The same PEST file will be run using BeoPEST in parallelized mode. Aim of the workshop is to make the participants feel comfortable with the command line, while doing useful tasks.

S1: Highly parameterized inversion

• Why do we need regularisation

• Subspace regularisation – singular value decomposition

• Tikhonov regularization

• Measurement and regularization objective functions

• Pilot points as a spatial parameterization device

• Optimal pilot point emplacement

• Pilot points and geostatistical regularization

• Covariance matrices as a regularization device

• Saving model runs with SVD-Assist

• Parameter identifiability

• The resolution matrix

S2: Calibration of groundwater models – some practicalities

• Steps to ensuring good finite-difference derivatives

• Detecting problematical finite-difference derivatives

• PEST groundwater utilities

• The PLPROC utility

• Defense against model defects

• Incorporating temporal and spatial differences into the objective function

S3 Case study with PP calibration

S4: Workshop (GUI + cmd): PEST Calibration with Pilot Points

In this workshop, participants will calibrate the model of the previous workshop using pilot points. PP will be added and modified through the GUI, and strategies for efficient use of pilot points in groundwater model calibration will be discussed. The calibration process will be accelerated by doing model runs in parallel using BeoPEST. Tikhonov regularisation, singular value decomposition and SVD-Assist will be demonstrated. 

09:00 - 13:00
S1: Introduction to uncertainty analysis

• Uncertainty analysis and sensitivity analysis

• Basic statistics – the variogram

• Multiple point geostatistics

• Some other modern geostatistical methodologies

• Predictive uncertainty and spatial parameterization detail

• Monte Carlo methods

S2: Calibration-constrained uncertainty analysis

• Statement of the problem

• Linear uncertainty analysis

• Nonlinear uncertainty analysis (calibration-constrained Monte Carlo)

• Null space Monte Carlo

• Ensemble Kalman smoother

• Direct hypothesis testing

S3 Case study – application of uncertainty analysis

S4: Workshop (GUI + cmd) Uncertainty analysis with Null Space Monte Carlo

The calibrated model of the previous workshop will be used to generate a small number of realizations (i.e. 10 calibrated version of the model, but the same method would apply to 1000). Most of the uncertainty analysis will take place in the command line environment. The purpose of the exercise is to understand the importance of the uncertainty analysis, especially when the modelling is done to support decisions that must be informed about what is known and what is not known.