External forcings and predictability in Lorenz model: An analysis via neural network modelling

What's about predictability in future climate scenarios? At present, we have no answer to this question in realistic climate models, due to the need of a difficult and time-consuming analysis. So, in the present paper an investigation of this situation has been performed through low-dimensional models, by considering unforced and forced Lorenz systems as toy-models. By coupling dynamical and neural network analyses, some clear results are achieved: for instance, an increase of mean predictability in forced situations (which simply mimic the actual increase of anthropogenic forcings in the real system) is discovered. In particular, the application of neural network modelling to this problem supplies us with some "surplus" information and opens new prospects as far as the operational assessment of predictability is concerned.