Choosing the right PEST operation mode

Immediately after a new FePEST project has been created, the Problem Settings dialog is open automatically. As a first task, the FePEST Operation Mode has to be selected. The table below contains the list of available modes.

 

Operation mode Description
Estimation Use this option for classic model calibration. PEST searches for the parameter values using Gauss-Levenberg-Marquardt (GLM) algorithm to find the best fit to the observations (estimation mode) or - if regularization is applied - a sufficient fit to observation with the most plausible parameter values (regularization mode).
PEST-GLM Same as the Estimation mode, but uses the new implementation PEST++ instead of classic PEST. It provides a number of new features (e.g. inequality constraints for observations) but is less tested than the long-proven classic PEST package.
Prediction Use this option to let PEST calibration the model in order to maximise a certain prediction (predictive analysis mode). This mode had been used e.g. for best/worst-case analysis but is hardly used any more in favor of the Pareto Method.
Pareto Use this option to let PEST perform a Pareto Analysis. This allows to evaluate the trade-off function between two competing objectives, usually the Measurement Objective (calibration RMS) and the a priori parameters in regularized calibration or one of the later against extreme predictive values.
Pre-calibration Monte Carlo Use this option to let PEST perform a traditional Monte-Carlo Analysis. A number of FEFLOW runs is undertaken based on user-defined parameter sample sets (generated from statistical distributions of said parameters). Each of these realization can be used to calculate the objective function, Jacobian and to calibrate the model.
Post-calibration Monte Carlo (Null-Space MC) Use this option to let PEST perform a Null-Space Monte-Carlo Analysis. Similar to the pre-calibration analysis, but now here each realization is projected onto the so-called Null-Space. This provides the user a set of realization of "almost" calibrated parameters. Under these conditions, the user can decide to calculate the objective function, Jacobian or calibrate each sample.
Ensemble Smoothing Applies PEST-IES to create an ensemble of a specified size, by randomizing the parameter values according to the specified prior uncertainty range and calibrating each ensemble member. The resulting ensemble yields non-linear uncertainty distributions of the parameters and can be deployed for calculating forecast ensemble runs that covers the uncertainty of the predictions. PEST-IES is also more tolerant to model instabilities and often requires much less computational time then the classic Estimation mode.

 

The PEST operation mode Regularization has been implemented in FePEST in a different manner. Since regularization methods (Tikhonov, subspace and super parameters) can be applied all the PEST operations, the regularization mode in FePEST has been made available not through the Operation Mode page, rather it has its own section in the Problem Settings dialog

 

You can find a complete description of each operation mode in its corresponding section in this help system.

Table of Contents

Index

Glossary

-Search-

Back