Building statistical hypotheses from aerobiological data sets

Aerobiology, as all fundamental scientific fields, has always relied on scientific hypotheses to mark its advances. This has been nicely synthesized by the presentation of the hierarchical aerobiological hypotheses at the first AFAR workshop in 1992. On the other hand, in most aerobiological publications, conclusions are drawn from the probabilistic acceptance of an alternative statistical hypothesis (H1). This paper will present the elements that can help unify aerobiological and statistical hypotheses. Most statistical textbooks deal with predetermined probability models, such as the Normal distribution, and most researchers will try to fit the data to these models. However, the aerobiological data set has its own parameters, mostly because of its predominant time and space determinants, and because of the "cloud" dispersal of biological particles which lead to an heavy representation of low values. So, it should be the data which should tell us about the appropriate model, i.e. that we should fit the model to the data. Moreover, since process uncertainty propagates in time, but not observation uncertainty, it is of prime importance that the real objective of our studies be well understood before deciding on a statistical approach. Finally the Bayesian approach, as it can be applied to aerobiology will be discuss.