Author(s)

Piotr A. Kowalski, Maciej Kusy, Marcin Szwagrzyk & Jan Izydorczyk

Abstract

Air pollution is one of the most dynamically developing problems of the contemporary world. Due to constantly present threat of air pollution, it is essential for the society to be aware of this issue and to be able to trace the individual factors influencing the existence of smog, as well as to predict the state of air quality in the following hours and days. This paper aims to determine the feasibility of cascading prediction of atmospheric boundary layer (ABL) values for several consecutive days, and then use this information to synthesize a prediction procedure for harmful smog particulate matter (PM10) for several days as well. Various prediction methods are used in the current study, among which the linear regression algorithm proves to be the most effective. Herein, the simulations concerning the investigated prediction algorithms are based on real data provided by the Airly company network of pollution measurement stations as well as ABL from the Copernicus Climate Data Sore. Evaluation of the obtained results is carried out using such measures as mean squared error, mean absolute error, Pearson correlation coefficient R, and index of agreement. As a result of the simulation, ABL and then PM10 predictors are synthesized for three consecutive days. The latter is characterized by an average daily mean absolute error in the range of 8-10 µg/m3, and index of agreement 0.88-0.89 depending on the day of the prediction and the variants of the prediction algorithm selected

Keywords

air pollution prediction (forecasting), atmospheric boundary layer, big data, data science, machine learning, particulate matters, regression task, data science.