Decomposition techniques for modelling the levels of particulate matter PM10 air pollutant in the city of silistra, Bulgaria

摘要

For the studied average monthly values of the levels of the air pollutant PM10 in Silistra in the period 01.2015 - 12.2019, two modern methods for decomposition were used - X-13ARIMA-SEATS and STL. The trend-cycle and seasonal component of the series were estimated in a total of 24 different ways - 8 models with the X-13ARIMA-SEATS approach with two seasonal adjustment options each - X11 and SEATS and 8 with the STL method. A comparative analysis was made between them, both in terms of estimating the components of the decomposition and in terms of the quality of approximation of the predicted values for the first six months of 2020 to the actually observed ones. In 23 out of 24 assessments of the trend-cycle component, a decreasing trend is observed, followed by a slightly increasing trend in the last year and several months of the period 2015-2019. The STL method yielded better forecast results for the first six months of 2020, using the default settings in the corresponding functions of R programming language. The estimated trend-cycle component by STL method is significantly smoother than that by method X-13ARIMA-SEATS.