A new method of missing data imputation applied to time series of PM10 concentration

The study of atmospheric pollution, with the emphasis of inhalable particulate matter (PM10), is necessary; given the damage was done for population health, besides other losses. Historical series, used for forecasting data, often have gaps due to several factors, which can detract from the quality of the forecast. The aim of this study was purposed a new method of missing data imputation, and after this, to use a time series model to forecast PM10 concentration. It was obtained the PM10 daily concentration in QUALAR system of CETESB, related to cities of Campinas, Jundiai and Paulinia, all in Sao Paulo State. The method of imputation of missing data, purpose in this study, was called TDEM (Time-Dependent Effect Method). The TDEM method was compared to others two methods ("Mean during month" and "Mean during year") of imputation of missing data, and it presented better results related to correlation coefficient, mean square error and mean absolute deviation. After imputation data of series, the data were analysed in order to forecast future PM10 concentrations. It was used ARIMA and SARIMA for time series models. The more satisfactory results were obtained for SARIMA models, which real data remained within the 95% forecast limits.