What is the best way to use PEST to do model predictive uncertainty analysis?
There are a number of alternatives available here. For over-determined inverse problems (i.e. for problems in which there is no null space and a unique parameter set can be found because the steps necessary to ensure parameter uniqueness through parameter parsimony have been taken), PEST can be run in predictive analysis mode. When used in this mode PEST maximizes or minimizes a user-specified prediction while maintaining the model in a calibrated state. Formulas are available for defining an objective function at which a model is no longer deemed to be calibrated. However this is often best done visually because where model-to-measurement misfit is dominated by structural noise, traditional statistical analysis is no longer applicable. An alternative to nonlinear analysis implemented through constrained maximization/minimization in this fashion is linear analysis. This can be implemented in the over-determined context with the help of the PCOV2MAT and MATQUAD utilities. In the under-d
