Multivariate statistical forecasting modelling to predict Poaceae pollen critical concentrations by meteoclimatic data

Forecasting pollen concentrations in the short term is a topic of major importance in aerobiology. Forecasting models proposed in the literature are numerous and increasingly complex, but they fail in at least 25 % of cases and are not available for all botanical species. This work makes it possible to build a forecast model from meteorological data for estimating pollen concentration over a certain threshold of Poaceae, an allergenic family. In Italy, about 25 % of the population suffer from allergies, these in 80 % of cases being caused by airborne allergens, including taxa of agricultural interest such as Poaceae. The pollen dispersion in air is determined by both the phenological stage of plants and the meteorological conditions; the pollen presence varies according to the year, month and even the time of the day. There is a correlation between environmental factors, pollen concentrations and pollinosis. A partial least squares discriminant analysis approach was used in order to predict the presence of Poaceae pollen in the atmosphere with a time lag of 3, 5, 7 days, on the basis of a data set of 14 meteorological and pollen variables over a period of 14 years (1997–2010). The results show a high accuracy in predicting pollen critical concentrations, with values ranging from 85.4 to 88.0 %. This study is hopefully a positive first step in the use of a statistical approach that in the next future could have clinical applications.