How does PEST perform in the face of local objective function minima?
Before answering this, it is important to note that local minima are often a function of non-differentiability of model outputs with respect to parameter values. It would not take too much trouble in many cases to eradicate this problem if some small steps were made in model algorithmic design to remove thresholds and discontinuities that are “manufactured” by the model itself. This is well explained in the following article:- Kavetski, D., Kuczera, G. and Franks, S.W. (2006). Calibration of conceptual hydrologic models revisited: 1. Overcoming numerical artefacts. Journal of Hydrology, 320 (1), pp173-186. Having noted this however, it must also be admitted that the problem is real – and would still exist (though to a lesser extent) even if all models were designed in such a way as to avoid artificial thresholds. It is also true that the Gauss Marquardt Levenberg method, on which PEST is based, cannot provide a guarantee that a local minimum will not be found instead of the global obje
