Forecasting Method for PM10 Concentrations in Seoul, with Adjustments for the Count Time Series Distribution and Excess Zeros

This study addresses the problem of monitoring and forecasting of particulate matter (PM) data, focusing, in particular, on high-level PM¹⁰, which is known to adversely impact human mortality and morbidity. We use hourly PM¹⁰ data, collected over a period of 3 months between October 1, 2018, to December 31, 2018, from 40 stations located in the Seoul metropolitan area of South Korea. We model the number of regions corresponding to “bad” or “very bad” categories of the PM¹⁰ density. It is challenging to model the data set, not only because it has excessive zero, the right tail of the distribution is extremely long, but also because the sample autocorrelation function of the series shows the serial correlation. Furthermore, it exhibits heteroscedasticity. Ignoring the zero-inflation and the serial dependence might produce inaccurate results. In this paper, several zero-inflated models with explanatory variables and pure time series models without explanatory variables are used to forecast future values of the aforementioned variable and generate confidence intervals with adjustments for the count time series distribution and excess zeros.