Correlating Particulate Matter Concentrations with Wind Speed and Temperature using Machine Learning Algorithm

Abstract

Almost from the last century, various environmental scientists are trying to correlate particulate matter (PM) concentrations with meteorological factors, for example, wind speed and temperature. This article tries to practically relate those with respect to popular machine learning algorithms to investigate any unusual data distribution in between those. PM concentration is dependent upon various factors, that include source of PM and meteorological factors. This article investigates the one-versus-one relation pattern of PM concentration and meteorological factors of temperature and wind speed. Concentration of PM2.5 and PM10 is collected from continuous ambient air quality monitoring station (CAAQMS) of Talchar coalfield, Odisha, India, along with wind speed and atmospheric temperature for 24 days continuously. Total five different machine learning algorithms are developed to build regression models, for example, Linear Regression, Nearest Neighbour Regression, Support Vector Regression, Decision Tree and Random Forest. The following combination of parameters is evaluated: PM2.5-Wind Speed, PM2.5-Temperature, PM10-Wind Speed and PM10-Temperature. Every one-versus-one relation is trained and tested with each of the five algorithms. In every case, linear regression fetches the best result as it produces the least root mean square error. Also, an overview, descriptive statistical analysis and graphical representation of every parameter considered are illustrated in this article. Main objective of this article is evaluating various regression analysis algorithms that represent relation between meteorological parameters and PM concentration, along with the traditionally studied linear model.