Assessment of neural networks and time series analysis to forecast airborne Parietaria pollen presence in the Atlantic coastal regions.

Pollen forecasting models are a useful tool with which to predict episodes of type I allergenic risk and other environmental or biological processes. Parietaria is a wind-pollinated perennial herb that is responsible for many cases of severe pollinosis due to its high pollen production, the long persistence of the pollen grains in the atmosphere and the abundant presence of allergens in their cytoplasm and walls. The aim of this paper is to develop artificial neural networks (ANNs) to predict airborne Parietaria pollen concentrations in the northwestern part of Spain using a 19-year data set (1999-2017). The results show a significant increase in the length of time Parietaria pollen is in the air, as well as significant increases in the annual Parietaria pollen integral and mean daily maximum pollen value in the year. The Neural models show the ability to forecast airborne Parietaria pollen concentrations 1, 2, and 3 days ahead. A developed model with five input variables used to predict concentrations of airborne Parietaria pollen 1 day ahead shows determination coefficients between 0.618 and 0.652.