CATs are a pair of transformation between the phase spaces of reference truth and the imperfect model. These inexpensive mappings have the potential to improve forecasts from imperfect models whose states lie on a totally different attractor than the reference truth.
This repository trains and evaluates CATs for the Lorenz'96 chaotic model.
This repository contains both ipython notebooks and python scripts. The notebooks are helpful in understanding the theory and implementation. The python scripts were mostly used for training different flavours of CATs for different lead times.
CATs_L96.ipynb is the main notebook that implements CATs for Lorenz'96 as
discussed in the paper. All notebooks begin with a comment that explains the
objectives achieved in the notebook in more detail.
Lorenz'96 model equations are numerically integrated to obtain the dataset required for training CATs. These are embedded within the notebooks and the python scripts.