PEST Version 11.8 for the PC (November 2008)

PEST version 11.8 can be downloaded from the PEST version 11 page. The PEST Groundwater and Surface Water Utility suites can be downloaded from the utilities page.

New features in PEST version 11.8 include the following:

·         Improved regularization functionality – including “null-space enhanced” Tikhonov regularization which is unique to PEST. This provides another safeguard against numerical instability when undertaking highly parameterized inversion.

·         Linear predictive uncertainty analysis is now much easier with the GENLINPRED utility. This allows a user to calculate the following quantities (and more):

o        parameter identifiability;

o        parameter contributions to predictive uncertainty and predictive error variance;

o        the comparative worth of different existing or yet-to-be gathered data in terms of its ability to reduce predictive uncertainty.

·         Calculation of “super observations”. These are orthogonal linear functions of the data that are uniquely informative of orthogonal linear functions of the model parameterization.

Features in PEST version 11 that were added to version 10 of PEST include the following:

• A bug fix, and some improvements, to the Null Space Monte Carlo utilities RANDPAR and PNULPAR.

• Capabilities to detect and combat bad model derivatives.

• POSTJACTEST and MULJCOSEN – utilities that allow analysis of the status of derivatives computed using finite-difference-based model runs.

• IDENTPAR – a utility that allows a user to readily compute the “identifiability” of each parameter employed in the parameter estimation process.

• SUPCALC – suggests the number of super parameters that should be employed in SVD-assisted parameter estimation.

• Two global optimisers, one implementing the powerful CMA algorithm and the other implementing the popular SCE algorithm. These can be used interchangeably with PEST. Both are enhanced versions of these algorithms and are fully parallelised so that they can be used with complex models with large run times. Where local optima, a high degrees of model nonlinearity, or non-differentiability of model outputs with respect to adjustable parameters limit the use of PEST, these can be used instead.

• A utility named JACTEST which can be used to check the integrity of model-calculated derivatives.

• Utilities to implement “null space Monte Carlo” nonlinear analysis of predictive error variance.

• Enhancements to the TSPROC time series processor supplied with the PEST surface water utilities. This utility automates PEST setup in surface water model calibration contexts, allowing a user to readily incorporate complex multi-component objective functions into the calibration process.

• Additions to the Groundwater Data Utility suite which expedite use of PEST with the RSM model. New SVD-assist options for super parameter definition and estimation.

• Introduction of the LSQR solver for highly parameterized problems.

• New Groundwater Data Utilities for use with the adjoint version of MODFLOW produced by Tom Clemo of Boise State University. (Email us or Tom to find out more; extremely efficient inversion based on thousands of parameters is now easy.)

• New utilities for analysis of predictive error variance and model predictive uncertainty for use in the regularised inversion context (where the “null space term” – mostly the biggest contributor to predictive error variance) is fully accommodated.

• More utilities through which the observation and parameter contributions to model predictive error and uncertainty can be computed. These can be used as a basis for optimisation of future data acquisition. (Note that because they rely on sensitivities rather than actual parameter and observation values, a model need not even be properly calibrated in order to employ these utilities as a basis for future data acquisition.)

• Implementation of Broyden’s Jacobian matrix upgrade methodology, this leading to improved model-run efficiency on just about every occasion of PEST usage.

• Many new utilities to assist in various PEST pre- and post-processing tasks. These include Jacobian matrix manipulation. In many instances a Jacobian matrix can be created for a new PEST run based on sensitivities computed during an old PEST run, saving the need for model run repetition.

Note: Make sure that the directory holding older versions of PEST (including version 9) are no longer in the PATH string, as old versions of some PEST utilities (including those employed by SVD-assist) are not compatible with the latest PEST.

See addendum.pdf for documentation of all the latest features of PEST.

PEST v11.3 for UNIX (August 2007)

PEST (including PEST, SENSAN, Parallel PEST and PEST utilities) is available for UNIX.

To download Version 11.3 of PEST for UNIX, click download UNIX.