Predictive Analysis: Best and worst case scenarios

Predictive Analysis requires the completion of the fundamental setup. It is fur­ther recommended to successfully run a history matching (calibration) pro­cess before starting predictive analysis. The parameter set found resulting from the history matching process is used as the initial parameter values in the current FePEST setup (in the Show results panel).

It is common to have multiple calibrated models, especially for environmental models. However, the predictions made by using these models may vary sig­nificantly despite the fact that they all honour historical data. The prediction attained by a particular model is therefore just one out of the possibly many outcomes.

Predictive analysis is a simple tool in PEST for non-linear model predictive error and uncertainty analysis. It searches for the calibrated model with the maximum or minimum key prediction. This facilitates identification of the worst-case and/or best-case scenarios among the set of calibrated models.

However, the use of PEST in predictive analysis model is restricted to well-posed problems only. This implies that it is applicable only to cases where the variability result from measurement noise. In case of an ill-posed problem, advanced users may choose to use PEST methods such running PEST in Pareto mode.