A relação entre a poluição do ar e as principais variáveis climatológicas na cidade de Canoas, Rio Grande do Sul

Abstract

Oftentimes, air pollution is not noticeable, and there are few cities with proper instruments to measure it. Air pollution has directly affected people’s health and is one of the most severe forms of pollution; especially when it is associated with climate variations, consequences may be softened or worsened. In this context, the present work aims to use multivariate analysis techniques of grouping and factorial to determine the association between air pollution variables and climate variables. Thereafter, the ARMAX model will be employed, using the defined variables in the multivariate analysis as exogenous variables, and the models of ARIMA temporal series, Simple Exponential Smoothing, Biparametric and Holt-Winters, with the purpose to present a comparison of the performance of the models, adjusting air pollution variables. Data were collected along with the National Institute of Meteorology (INMET) (climate variables) and the State Foundation of Environmental Protection (FEPAM) (air pollutants) referring to the city of Canoas-RS, during the year of 2014. In the grouping among the variables, two well defined groups were formed: Cluster 1 puts together the closest variables (O3, Insolation, Evaporation, Minimum and Maximum Temperatures), cluster 2 puts together the closest variables (SO2, CO, PM10). For group 1, prediction models for O3 variable were adjusted, using the other group variables as exogenous to the ARMAX model. In the comparison of the prediction results of the best model, the following criteria were used: MAPE, EQM, MAD and SSE. Results have shown that the SARIMAX model (1,1,1[1,1,1]) (0,1,1), with the exogenous variables inclusion, produced better prediction results in comparison to its opponent ARIMA and the class of exponential smoothing. In group 2, prediction models for the PM10 variable were adjusted, using the other group variables as exogenous to the ARMAX model. In the comparison of the prediction, the ARX model (1,[1,1]), with the inclusion of exogenous variables, produced better prediction results in relation to its opponents ARIMA and the smoothing exponential class. It was possible to conclude that the combined used of multivariate analysis techniques and temporal series provides better prediction results. The vast possibility of usage of combined techniques of multivariate analysis and temporal series stands out, when the intention is to make predictions and describe the temporal process with the usage of explanatory variables, being used as an efficient support tool in the decision making in countless areas.