PEST is accompanied by a plethora of utility programs. These fall into four groups.
You receive these when you download PEST itself. They perform tasks ranging from comprehensive error checking of PEST input datasets, to evaluation of the efficacy of different strategies for acquisition of future data.
These utilities expedite the use of PEST in conjunction with many popular groundwater models, including MODFLOW, MT3D, SEAWAT, FEFLOW and RSM. They promulgate the use of pilot points as a parameterization device, and facilitate application of complex regularization constraints in conjunction with them. They read output files produced by these models, and perform spatial and temporal interpolation of these outputs to the sites and times at which measurements were made.
They perform many other tasks as well, some of which have nothing to do with model calibration, but are just plain useful.
Calibration of surface water models is best served by construction of a multi-component objective function. Ideally, each component should contain a “distillation” of field data on the one hand and its model-generated counterpart on the other hand, that carries information about a different aspect of the system under study. This, combined with regularized inversion, maximizes the extent to which the calibration process can facilitate transfer of information from field data to the model. The result is a calibrated parameter set of minimum error variance, together with an ability to explore the magnitude of that variance.
PEST’s surface water utilities automate the construction of complex PEST input datasets that accomplish these objectives. The size of the calibration dataset and/or the number of parameters featured in the calibration process present no obstacles to their use.
There are occasions when so-called “gradient methods” (such as the Gauss-Marquardt-Levenberg method and singular value decomposition) do not work. This can result from high levels of model nonlinearity, leading to the introduction of local optima to the objective function surface. Or perhaps a model uses a simulation algorithm or numerical solver that introduces discontinuities into relationships between model outputs and parameters.
PEST is accompanied by two so-called “global optimizers”. One of these uses the popular Shuffled Complex Evolution scheme, while the other employs the well-known Covariance Matrix Adaptation scheme. Both can be used interchangeably with PEST. Both can parallelize model runs across a computer network.
C4SF Login |
Web Site Managed via C4Site Factory,™ a trademark of
Echo Valley Graphics, Inc.
©2014 Copyright Echo Valley Graphics, Inc. All Rights Reserved.