Quantitative stratigraphic models are in more common use for predicting stratigraphic attributes in locations where observations and control points are lacking. The general philosophy behind these predictions is that if the models match stratigraphic attributes in places where there are observations for comparison with model output, then model predictions in other places also are probably good.

These quantitative models can be categorized into three groups: (1) geostatistical models; (2) forward models; and (3) syntactic or rule-based models. Although geostatistical models may be conditioned to honor observations, they cannot overcome the fundamental limitation that the simulations are based purely on mathematical descriptions, rather than some understanding and description of the stratigraphic process-response system. With a given observed dataset, a geostatistical model may produce a simulation which is not reasonable geologically, even if mathematically correct. Forward modeling simulates stratigraphic process-response relations directly. However, current versions of forward modeling match observations qualitatively through trial-and-error simulations. Rule-based approaches can provide a more general and flexible framework for stratigraphic simulation. In rule-based models, empirical rules can substitute for mathematical equations when quantitative expressions are unknown, and non-numerical stratigraphic attributes, such as lithology, can be treated symbolically rather than numerically.

By combining the strengths of these model types, we can build an adaptive stratigraphic forward modeling system, which uses a process-response oriented stratigraphic forward model, a syntactic model for comparison and an optimization genetic algorithm that modifies values of forward model parameters to achieve better matches between simulations and observed stratigraphy. The geological process-based forward model produces synthetic stratigraphy as output; the comparison technique measures the difference between the forward model output and the observed dataset; and the optimization genetic algorithm as adaptive simulation technique automatically adjusts the suitable parameters of the forward model, so that the system can produce a output stratigraphy that honors or closely matches the observed dataset.

Experimentation of the method on synthetic stratigraphy as the dataset shows that the system tries to approximate the original parameters, and when converged, it almost recovers them. Application of the method to a real stratigraphic example as the dataset is in progress.

AAPG Search and Discovery Article #90937©1998 AAPG Annual Convention and Exhibition, Salt Lake City, Utah