Forecasting of pollen concentrations in Évora

Abstract

The increase of respiratory diseases related to allergic reactions led the World Health Organization to declare allergies as a public health problem. Changes in pollen seasons, as their intensity or duration, have an impact on the symptoms in people with allergies. This way is important to monitor and forecast the pollen concentrations. The primary objective of this work is to forecast pollen concentra- tion in Évora based on meteorological variables. The methodology utilizes a 22-year daily dataset (2002-2023) from Évora, encompassing pollen concentrations and me- teorological variables, including temperature, precipitation, relative humidity, and wind. Following robust data treatment for handling missing and anomalous values, an in-depth descriptive statistical analysis was performed. This analysis revealed a pronounced seasonality in pollen concentrations, with notable peaks between March and May, and a strong temporal autocorrelation. The modeling approach involves applying and comparing various techniques, ranging from classical statistical time series models (e.g., ARIMA, SARIMA, and Dynamic Regression), to Machine Learn- ing algorithms such as Artificial Neural Networks. Model performance was rigor- ously evaluated using metrics such as the Mean Absolute Error, and Root Mean Square Error, with the use of time-series adapted cross-validation strategies to en- sure their robustness and generalization power. This study aims not only to deepen the understanding of pollen dynamics in Évora but also to provide more accurate forecasting tools that can be directly applied to public health management and support allergic individuals.