Modelling aerobiological time series. Application to Urticaceae

This paper deals with the characterisation of aerobiological time series. The series discussed are annual series of mean daily concentration of Urticaceae pollen, obtained by means of Hirst traps on six sites of the Aerobiological Network of Catalonia (XAC). First, some alternatives for fitting a smooth trend to one of these series are discussed, and two satisfactory methods are presented, a Gaussian mixture model (parametric) and the Friedman supersmoother (non-parametric). Dividing daily concentrations by trend values, a series of residuals is obtained. The residual series can be characterised with simple conventional time series models, and some common patterns emerge from the parameter estimates. The long-term patterns of the pollen dynamics can be thus characterised by the trends, and the short-term behaviour by the residuals. The Spearman correlation is used in this paper as a measure of the similarity of two series.