Model-Independence

To calibrate a model, its parameters must be adjusted so that its outputs match field measurements as closely as possible. If done manually, this can be tedious in the extreme, particularly if the model has more than a few parameters.

If you calibrate a model by hand, you will never know:

• whether you could have achieved a better fit between model outputs and field data;
• whether any other sets of parameters could have achieved an equally good fit;
• the uncertainty associated with parameters that have just been estimated;
• the uncertainty associated with model predictions which depend on these parameters.

PEST calibrates a model automatically.

You, the user, inform PEST of numbers on model input files that it must adjust. You also inform PEST of numbers on model output files that must be matched to field values; you then provide these values. Then you let PEST run the model as many times as it needs to, adjusting parameters all the while, until the fit between model outputs and their field-measured counterparts is reduced to a minimum in the weighted least squares sense.

With PEST, it is easy to obtain a good fit between model outputs and field measurements.

Before every model run, PEST writes a set of model input files which contain parameter values which PEST would like the model to use on that run. After the model has run to completion, PEST reads from the model output files those numbers which must be matched to field measurements. PEST runs the model through a system call.

PEST interacts with a model through the model's own input and output files.

What does this mean? Well to start with:

• You can link PEST to any model that reads ASCII input files, writes ASCII output files, and can be run from the command line. This makes it totally model-independent.
• You don't have to be a programmer to link PEST to a model.
• The "model" that PEST runs can be as complicated as you like. It can be a batch or script file that runs a few, or many, executable programs in succession. Thus you can:
  • calibrate multiple models simultaneously;
  • pre-process parameters and post-process model outputs (for example undertake temporal and spatial interpolation from model outputs to the times and sites of field measurements);
  • calibrate "chained" models (for example a water quantity model followed by a contaminant transport model);
  • tailor your formulation of the inverse problem to extract maximum information from the calibration dataset.

The model that PEST calibrates can be as complicated as you like, with an unlimited number of parameters cited on an unlimited number of model input files, and an unlimited number of outputs read from an unlimited number of model output files.