Lichens as an indicator of air pollution – Integration of SEM/EDS and machine learning methods

Thalli of Xanthoria parietina (L.) Th. Fr. lichen collected in locations featuring different pollutant deposit conditions in the urban space of Kielce, including those with intense vehicle traffic, low emission and alkalization from the nearby cement and limestone plant and the open-pit mine were subject to microscopic analyses. The lichen surface had cellular structures with particles characteristic of respective pollutant sources of identified shape and chemical composition (SEM/EDS). The predominant type of particles in the city includes mineral dusts containing silicon and aluminium (natural mineral weathering) and soot with carbon, sulphur and nitrogen (low emission and transport). Sharp-edged structures exceeding 20 µm made of calcium, magnesium and sulphur (cement and lime particles) accompany much smaller, round particles with ø < 5 µm containing iron, aluminium and other heavy metals (industrial fossil fuel combustion). The micrographs taken were used to build a model to create a self-learning pollutant identification system based on the activity of deep neural networks (ResNet). The trained algorithm is able to detect individual items in new micrographs with 71% result. Adding up areas of identified objects (using Euclidean equation) allows identifying their emission sources.