Learning conditional probabilities in event bushes with temporal labels

The event bush is a recently developed formalism for knowledge representation optimizedspecifically for Earth science knowledge. It graphically represents logical and/or probabilisticdependencies between its nodes (events). Numerical data in geosciences often come inthe form of time series: therefore, it is a very common situation to have tables with time-relatednumerical data for the variables which correspond to the events represented in an event bush. Inthis work, we present a way to use these numerical data to learn conditional probabilities that haveto be specified in the intermediate event bush created on the basis of an event bush with temporallabels. The same idea works also if the bush also has spatial labels.