Conclusions (Page 26)

Low-dimensional Physics-Based Models offer attractive alternative to attempts to perform first principle modeling of complex nonlinear dynamics.

Feed-forward and Feed-back loops in systems result in bifurcations, chaos, and chaotic attractors that provide intrinsic limits to predictability.

Neural Net works and DSP for prediction filters produced lowest error measures, but lack knowledge of fundamental physics constraints.

Modeling Theory introduces the probability for "Model Validation" over a database. Simpler models yield higher overall performance/acceptance rating within this (Bayesian) theory.