Better Water Resources Management: Part 4 (a)
First Step in Learning Calibration of Groundwater Models with PEST - zone based approach
In this course you are going to learn calibration of groundwater models with PEST using zone based approach. This online course is 1st of three parts, in 2nd part (due to be released in a month) calibration using pilot points (highly parameterized inversion with regularization) will be explained and in 3rd par uncertainty analysis of your calibrated groundwater models will be described.
After forming a model, translating it from conceptual to numerical model, state of art in today's environmental and groundwater modeling is finishing the model in two steps - first step groundwater modeller is doing calibration and step two is conducting uncertainty analysis (or error analysis) by the groundwater modeler in order to make right decisions required by project.
In this course we are going to explore and hopefully the participant will learn (we are with you all the time the course) how to calibrate groundwater models with PEST using zone based approach.
In zone based approach model parameters in active area of model are divided into zones of equal values (manual regularization is done using mathemathical language) and PEST is used to calibrate a groundwater model.
This approach is still mostly used in calibrating groundwater models and it is the starting point for more advanced way of model calibration - highly parameter inversion with regularization.
Description of units:
- This unit lasts 50 minutes
- In first 15 minutes water resources management case study in thin (7 meters thick) alluvial aquifer was presented. Model was calibrated with PEST quickly in order participants to get a feeling how the whole process is working.
- Analytic Element feature in Groundwater Vistas for simulating inflow in aquifer was described as part of the case study
- Discussion about vertical and horizontal dimensions in modeling and rationale for choosing 2D approach in this case study using 3D visualization tools
- The whole process of setting up and running PEST calibration was shown in step by step approach so that participants can easily follow it.
- What can happen if we over-fit model?
- This unit lasts 45 minutes
- Three type of files needed for PEST run (created by Groundwater Vistas), opening them in text editor and examining how Groundwater Vistas is making them (plus a little experiment lasting all together 15 minutes)
- The best way to run PEST - from DOS prompt (yes, it is true)
- Description of one PEST optimization iteration (15 minutes):
- Definition and calculation of objective function
- Jacobian matrix with examples
- Marquardt lambda
- Parameters sensitivity in PEST - intro with examples (15 minutes)
- Do small zones in PEST matters?
- Non-linearity of parameter sensitivity during calibration process
- This unit lasts 45 minutes
- How to insure smooth PEST run (5 min)?
- Sensitivity is not enough, let's talk about correlation of estimated parameters in your groundwater model - practical exercise
- Definition of regularization (for now only manual regularization)
- Importance of correlation between estimated parameters for prediction and decision making of your groundwater model
- Multiple correlation
- Parameters identifiablity
- How to implement "soft knowledge" with tied parameters during your calibration process
- Description of PEST control parameters
- What is Marquardt lamda and its role during PEST optimization (calibration) process
- Assigning weights for different calibration types (groundwater heads, discharge, concentration)
- Using springs discharge as calibration target
- Using PEST utilities
- Examination of output PEST files (record file, parameter file, sensitivity file...)
- Before running a PEST you will learn how to check is PEST running correctly with one simple run
- SVD (Singular Value Decomposition) will help you to get rid of insensitive parameters that will make PEST run difficult or impossible
- Testing integrity of finite difference derivative approximation (when to use global optimisers)
Who should attend?
This course is ideally suited for hydrogeologists,
geologists, climate change experts, environmental and agriculture engineers who already have a knowledge of groundwater modeling with MODFLOW.
B.Sc in hydrogeology, geology, environmental engineering, civil engineering or similar...Good idea is to finish Hydrogeology Bundle 1 - Gentle Introduction to Analytic and Numerical Groundwater Modeling
Average 2 - 3 hours per week to finish each Module. We encourage you to spend time with the course each week. Your focused attention will pay off in the end!
If anything seems unclear or you just have a question, then please let me know - I will get back to you as soon as possible, usually less then 24 hours. After each lecture you have place to ask questions. Again, thank you for joining me on this course.
StartUnit 1: Excercise that will show you how to setup and run PEST calibration on real case study (51:41)
StartUnit 2: Three type of files that PEST needs for running, optimization iteration, sensitivities (44:46)
StartUnit 3: Importance of corelation of parameters after calibration, identifiability of parameters and more (52:26)
StartUnit 4: Assigning weighths and rules during calibration proccesswith PEST (51:23)
Dragan Kaludjerovic received his BSc, MSc an PhD at Belgrade University and he is currently at online specialization at The Johns Hopkins University - Environmental Engineering.
In his early career, 1998, Dr. Kaludjerovic focused his research on more complicated numerical models such as unsaturated zone modeling and transport modeling with MT3DMS. Feedback from his presentation at the MODFLOW 98 Conference influenced him to continue to work on mathematical models and their application to hydrogeology engineering which resulted in his MSc defense.
Later, in 2000, he was invited to work at Waterloo Hydrogeologic Inc, (WHI), a world leader at that time in groundwater model software development and consulting. During his time at WHI, he was introduced to innovative modeling techniques, including PEST for parameter estimation and reactive transport modeling with RT3D.
During the next several years, Dr. Kaludjerovic worked on numerous projects including the design of a large groundwater source with 1000 l/s in Macedonia.
In 2005, Dr. Kaludjerovic started work on his PhD thesis which was completed in 2008 and titled "Natural Attenuation of Contamination in Groundwater". The scope of this thesis focused on applications of the RT3D (Reactive Transport) model with multiple electron acceptors to a field case (aquifer contaminated with xylene).
After his PhD thesis, Dr Kaludjerovic continue to conduct research in the area of aerobic and anaerobic bioremediation with emphasis on emulsified oil projects and gas infusion technologies. During his carrier Dr Kaludjerovic published more then 25 papers which can be found below or on his Linkedin page.
In last several years, Dr Kaludjerovic is working with high resolution satellite images (both optical and radar satellites) and its application in environmental science and modeling.