Time Series Forecasting of Air Quality: A Case Study of Sofia City

Abstract

Air pollution has a significant impact on human health and the environment, causing cardiovascular disease, respiratory infections, lung cancer and other diseases. Understanding the behavior of air pollutants is essential for adequate decisions that can lead to a better quality of life for citizens. Air quality forecasting is a reliable method for taking preventive and regulatory actions. Time series analysis produces forecasting models, which study the characteristics of the data points over time to extrapolate them in the future. This study explores the trends of air pollution at five air quality stations in Sofia, Bulgaria. The data collected between 2015 and 2019 is analyzed applying time series forecasting. Since the time series analysis works on complete data, imputation techniques are used to deal with missing values of pollutants. The data is aggregated by granularity periods of 3 h, 6 h, 12 h, 24 h (1 day). The AutoRegressive Integrated Moving Average (ARIMA) method is employed to create statistical analysis models for the prediction of pollutants’ levels at each air quality station and for each granularity, including carbon oxide (CO), nitrogen dioxide (NO2), ozone (O3) and fine particles (PM2.5). In addition, the method allows us to find out whether the pollutants’ levels exceed the limits prescribed by the World Health Organization (WHO), as well as to investigate the correlation between levels of a given pollutant measured in different air quality stations.

Machine learning for hours-ahead forecasts of urban air concentrations of oxides of nitrogen from univariate data exploiting trend attributes

“…Another alternative short-term prediction method that is widely used to generate forecasts from univariate NOx time series is the ARIMA method. 22,23 Table 9 displays the results of an ARIMA(1,0,0) model for the Coventry and Lincoln sites for NOx t 0, t + 1, and t + 3 forecasts for 2020 and 2021 compared to the XGB model results. Higher-order ARIMA models with p ≥ 1, d ≥ 0, and q ≥ 1 ( p adjusts the autoregressive element, d adjusts the seasonal-differencing element, and q adjusts the moving average element) failed to converge for any of the city datasets studied due to the “spikiness” of the time series.…”

“…21 Applying autoregressive integrated moving average (ARIMA) models to univariate NOx time series can avoid the use of additional input variables and provide short-term predictions achieving moderate accuracy. 22–24 Typically, ARIMA predictions can be improved upon by applying ML and/or deep learning methods. 25 Another approach is to apply signal decomposition to the univariate NOx time series.…”

A Time Series Approach to Smart City Transformation: The Problem of Air Pollution in Brescia

“…In the study conducted by Marinov et al [17], the temporal trends of air pollution were investigated at five air-quality monitoring stations in Sofia, Bulgaria. Data collected between 2015 and 2019 were examined.…”

Prediction of Particulate Matter (PM10) during High Particulate Event in Peninsular Malaysia using Novel Hybrid Model

“…These missing data will be treated by using Linear Interpolation (LI) method using IBM SPSS Software Version 26. It is important to fill in the missing data before any analysis because the success of the modelling depends on the quality of the dataset [9,10].…”