What are the most important variables for Poaceae airborne pollen forecasting?

• Published in: The Science of the total environment Vol. 579; pp. 1161 - 1169
• Main Authors: Navares, Ricardo, Aznarte, José Luis
• Format: Journal Article
• Language: English
• Published: Netherlands Elsevier B.V 01.02.2017
• Subjects: Aerobiology; Air Pollutants - analysis; Air Pollution - statistics & numerical data; Allergens - analysis; Environmental Monitoring - methods; Feature selection; Forecasting; Madrid; Nonparametric tests; Poaceae; Pollen; Prediction; Random forests; Spain; Time series; Weather; Spain, Castilla, Madrid; Feature selection; Madrid; Prediction; Time series; Nonparametric tests; Aerobiology; Random forests
• ISSN: 0048-9697; 1879-1026
• DOI: 10.1016/j.scitotenv.2016.11.096

In this paper, the problem of predicting future concentrations of airborne pollen is solved through a computational intelligence data-driven approach. The proposed method is able to identify the most important variables among those considered by other authors (mainly recent pollen concentrations and weather parameters), without any prior assumptions about the phenological relevance of the variables. Furthermore, an inferential procedure based on non-parametric hypothesis testing is presented to provide statistical evidence of the results, which are coherent to the literature and outperform previous proposals in terms of accuracy. The study is built upon Poaceae airborne pollen concentrations recorded in seven different locations across the Spanish province of Madrid. [Display omitted] •A new data-driven approach to forecast airborne pollen concentrations is presented.•The relative importance of variables is estimated and used to build the models.•The assumption-free findings are coherent with previous phenological-based results.•Non-parametric hypothesis testing confirms the statistical validity of the results.•A reduced set of important variables renders more simple and accurate models.