PEST - Model-Independent Parameter Estimation and Uncertainty Analysis

PEST News

1st May, 2013. PEST version 13.0

PEST version 13.0 includes the following improvements over version 12.3 of PEST.

  • The SSSTAT utiliy provide linear analysis of observation information content and parameter identifiability.
  • The INFSTAT1 utility extends the use of observation influence statistics such as Cook's D and DFBETAS to the highly parameterized context.
  • SUPOBSPAR1 uses subspace methods to compute combinations of observations which uniquely and entirely inform combinations of parameters taking into account the prior probability distributions of parameters.
  • If PEST is using singular value decomposition, LSQR or SVD-assist for estimation of parameters, parameters can be scaled according to their innate variability as informed by parameter bounds.
  • Some improved efficiencies have been added to parallel run management as undertaken by BEOPEST.
25th May 2012. Latin Hypercube

A set of utility programs has been written which facilitates use of PEST with a Latin hypercube parameter sampling package developed by Sandia National Laboratories. The utility software and the Latin hypercube sampler itself can be accessed through the "Latin hypercube" main menu item.

3rd May, 2012. PEST version 12.3

PEST version 12.3 includes the following improvements over version 12.2 of PEST.

  • A number of minor bug fixes.
  • The ability to use absolute parameter change limits in addition to relative and factor change limits.
  •  Signficantly enhanced BEOPEST run management functionality. This allows better distribution of model runs across platforms with very different speeds, as well as greater capacity to tolerate model run failure or network communication failures.
  • "Observation re-referencing" - allowing a user to employ different models for derivatives calculation and undertaking parameter upgrades. Where the former is a simple surrogate for the latter, this can achieve enormous increases in optimisation efficiency.
  • An ability to forgive model run failure during derivatives calculation resulting from model intolerance of upgraded parameter values.
  • Easier calculation of super parameters and super observations.
28th April, 2012. PEST++ and Genie

PEST++ is a PEST-compatible, object-oriented package that undertakes highly parameterized inversion using a variety of regularization methodologies. GENIE is a model-independent TCP/IP-based parallel run manager. PEST++ was written by Dave Welter, while GENIE was written by Chris Muffels. For more details see "The Next Generation" sub-menu of the "Third Party PEST-Compatible Software" main menu item.

20th Februaray, 2012. PLPROC

PLPROC is a "parameter list processor". It was developed as a generalized spatial model pre-processor. The model can employ a structured or unstructured grid. It can be two or three-dimensional. Use of PLPROC facilitates PEST usage with any such model - for both calibration and uncertainty analysis.

A user controls PLPROC through a set of statements that comprise a type of simple programming language - not unlike PYTHON. Model spatial parameterization can be zone or pilot point based. In the latter case either kriging or radial basis functions (or a combination of the two) can be employed to conduct spatial interpolation between parameterization points and the model grid. Parameters, or whole suites of parameters, can be subjected to arbitrary mathematical manipulation through user-defined mathematical and selection equations.

See the "PLPROC" sub-menu of the "Utility Support Software" main menu item for details.

 

C4SF Login | Web Site Managed via C4Site Factory,™ a trademark of Echo Valley Graphics, Inc.
©2014 Copyright Echo Valley Graphics, Inc. All Rights Reserved.