Abstract
In this contribution we review progress on the problem of forecasting chaotic dynamical systems. The nature of chaos was initially illusive, having escaped discovery by several great minds who were in a position to find it. Once the essential features of chaos were uncovered, the concept greatly influenced the thinking of physicists and forecasters, in particular, it brought into focus the limitation of forecasting nonlinear dynamical systems. This resulted in a dichotomy between prediction methods for (nearly) linear systems and nonlinear systems; the important features and distinctions of these two approaches are reviewed. An important issue that arises is the role of model error in forecasting; many methods either ignore or over-simplify the nature of model error, largely because this allows adaption of existing techniques. Finally it is argued that the Isis Programme(Imperfect-model Shadowing and Indistinguishable States) provides a theoretical and practical basis for dealing with all sources of uncertainty, including observational noise and model error.
