Credible computerized approaches to situation assessment for natural disaster management strongly depend on exploitable expert knowledge. A key problem, however, is to find a suitable knowledge representation method that a) is easy to understand and usable by domain experts of different disciplines, and b) is seamlessly usable by computer-based reasoning techniques. Recent logics-based approaches to situation assessment – ontologies – suffer from the ability to infer new knowledge that is not based on already known propositions (i.e. abnormal eventualities) and are also difficult to use by non-mathematicians or computer scientists. In this paper, we investigate and introduce a promising approach of formal knowledge representation as core building block of a continuous situation assessment component that explicitly supports inherent characteristics of disaster prevention & management situations. We are modelling networks of influence factors for critical situations, derived from domain experts and historical data, by means of probabilistic graphical models. This kind of models offers a very natural and easy to understand support for domain experts and pays tribute to the major aspects of uncertainty & incompleteness of data in disaster situations.