Impact of Meteorological Factors on Air Pollution Prediction Based on Multiple Linear Regression with R Program: A Case Study with Comparable-Pollutant Parameters and Fragmented Seasonal Datasets from a Heavy-Urban-Traffic Monitoring Station
TL;DR
The values of air pollutants to meteorological factors with prediction models created with multiple linear regression (MLR) in the R program relates the values of air pollutants to meteorological factors with prediction models created with multiple linear regression (MLR) in the R program.
Abstract
Air pollution data processing is a relevant aspect of urban-life-quality monitoring. This study relates the values of air pollutants to meteorological factors with prediction models created with multiple linear regression (MLR) in the R program. Fragmented data were obtained in the years 2023/24 from the heaviest-urban-traffic location in Belgrade, Serbia. A Serbian-comparable list of air pollutants (PM10, PM2.5, SO2, NO2, CO, and O3) was created according to an analysis of sensor availability at monitoring stations. With the prediction models and seasonally clustered data (spring and summer data clusters were incomplete), each of these pollutants (except O3) was related to all measurable meteorological factors (atmospheric pressure, wind speed, relative humidity, and temperature). All independent (meteorological) factors showed a moderate level (Multiple R2 < 0.7) of impact on the prediction of air pollutant values (dependent factors). The second group of prediction models related pollutant triplets, and fourteen of the created models reached R2 > 0.7. The prediction model with the highest relevant R2 value (CO~PM10 + NO2) was used for CO pollutant prediction and in-sample fit analysis. The obtained mean absolute error is ~0.13 mg/m3. This case study contributes to the examination of the role of meteorological factors in urban-traffic air-quality monitoring within a seasonal context.