hidden layer, creating new nodes where required and
destroying existing nodes when they become useless.
Parameters stored in each node characterise the associated
structure in the rain map. This is called the synthesis phase.
Parameter evolution of structures in time along a rain map
sequence is then analysed, in order to try and forecast their
behaviour in the near future. Some parameters, showing
heavily non-linear trend in time, are used to train a different
set of RBF NNs, expected to learn their typical trends.
Future values of those parameters will be determined by
submitting latest observed parameters to the trained NN.
The other parameters are simply forecasted using linear
extrapolation in time. This is called the forecast phase.
Forecasted parameters are finally turned into a forecasted
rain map.