Hi Sergio,
I think you’ll have to call az.plot_ppc multiple times and organize your plots on different matplotlib axes in the same figure.
Since you’re already using dims and coords to parametrize your model, I think you should make use of the return_inferencedata kwarg in pm.sample and set it to True. That way, you’ll get an ArvIZ InferenceData object directly and will be able to select your dims and coords more intuitively for posterior handling – see the updated radon NB for several examples.
Most ArviZ functions have kwargs to directly use these dims and coords 
Hope this helps 