This short monograph packs a lot of punch. It starts by solving an almost trivial, nonunique inverse problem. While simple, the problem is typical of many that are encountered on an everyday basis in decision-support groundwater modelling.
Solution of this problem provides important insights into the journey that information takes as it is harvested by the decision-support modelling process and delivered to decision-critical model predictions in order to reduce their uncertainties. This journey is fraught with dangers which express themselves as predictive bias. The only things that can protect information from these dangers are “correctness” of model structure and “correctness” of prior parameter covariance.
As a model becomes more complex however, there comes a point where these assurances cannot be given, and the dangers of transporting information through adjustable parameters become too great. This is particularly the case for predictions that are only partly informed by a calibration dataset, so that their uncertainties are only mildly reduced through history matching. At this point, the uncertainty reduction imperatives of decision-support modelling are better served by data space inversion than by parameter adjustment. The monograph explains why this is so.