Black Carbon characterization with Raman spectroscopy and machine learning techniques: first results for urban and rural area

Among the chemical substances of Particulate Matter (PM), there is a considerable quantity of black carbon (BC), which is linked to adverse public health effects and climate change. This study aims to develop an innovative method for the source apportionment of BC inside the PM, using Raman spectroscopy and machine learning techniques. Different BC sources, including biomass ashes and vehicle emissions, and different PM samples from air quality monitoring stations have been analyzed with a Raman spectrometer. The PM samples used in the present study are collected from two different locations: an urban environment (Turin, Italy) and an alpine valley context (Oulx, Italy). To each obtained spectrum, which presents the characteristic G and D bands, a five-band fitting has been applied to gather information that can lead to the identification of the different BC sources. Machine learning techniques, including the K-Nearest Neighbors (KNN) algorithm, have been applied to calculate the cluster resolution through a value of accuracy. Finally, the same algorithm, trained on the BC emission sources' data, tries to associate each BC in the PM to its source. In particular, a large amount of BC from diesel engine car exhaust emissions is found in all the considered PM samples.