<< Click to Display Table of Contents >> Navigation:  »No topics above this level« Observations

 

This section discusses observations. It also discusses related topics such as the objective function and observation weights.

When an inverse problem is solved, information is extracted from data. The success of this venture depends on how the inverse problem is formulated. Creativity is required. Much of this creativity must be focussed on definition of observations, and formulation of an appropriate objective function from them.

Model calibration in particular, and history-matching in general, requires so much more than inserting field data into a PEST control file, and then telling PEST to "match that". A modeller must ask him/herself "where does information reside in this observation dataset, and how can I present these data to PEST in a way that allows PEST to access this information?" Often, data have to be prepared for information extraction. This may involve some form of processing.

Model outputs may also require processing before being matched to similarly processed field data. Simulation is imperfect. Some aspects of simulation are more imperfect than others. Matching processed/filtered model outputs to similarly processed/filtered field data may reduce damage inflicted on the inversion process by model imperfections. It may thereby reduce the chances of history-match-induced predictive bias. This is an indispensable component of decomposing the problem.