Mixed-state spatio-temporal auto-models

We consider in this paper a general modelling for mixed-state data. Such data consist of two components of different types: the observations record many zeros, together with continuous real values. They occur in many application fields, like rainfall measures, or motion analysis from image sequences. The aim of this work is to present ad hoc spatio-temporal models for these kinds of data. We present a Markov Chain of Markov fields modelling, the Markovian fields being defined as mixed-state automodels, whose local conditional distributions belong to an exponential family and the observations derive from mixed-states variables. Some specific examples are given as well as some preliminary experiments. ◦ SAMM, Université Paris 1, 90 rue de Tolbiac, 75634 Paris Cedex 13, France • Faculty of Engeneering, University of Buenos Aires, Argentina.