Constructing High Resolution Maps of Aerosol Concentration using Machine Learning and Google Earth Engine

The object of research is the data generated by Landsat 8 satellite, meteorologocal data ERA5 generated by ensemble of models and data logs of ECMWF in-situ stations of air quality monitoring. The purpose of the research is PM10 pollution maps spartial resolution increasment in milligrams per cubic meter (mg/m3). PM10 is fine dust with size ≤ 10 µm. PM10 is one of the most widespread pollutant of urban amosphere and it conduces around 9% and 5% of deaths caused by lung cancer and cardiovascular pathologies, respectively. Using Google Earth Engine cloud platform more than 35000 Landsat 8 scenes of European region were analyzed for more than 2 years period, which is around 34 Tb of data. In the research the map of in-situ PM10 concentration was build with spatial resolution of 30 meters per pixel with the help of machine learning based on combined historical data of the Landsat 8 satellite, meteorological data of the ensemble of models ERA5 and logs of in-situ PM10 monitoring stations. The majority of previous researches conducted generated this kind of maps with 1 km resolution, given smaller amounts of data, different data sources with lower spatial resolution and other approached to predict PM10 in-situ concentration. One of the exceptions in the term paper of the second course with adjacent topic formulation, however, that time there were used more basic models and smaller amount of data sources. Knowledge of the distribution of PM10 in the urban environment allows more effective planning of sanitary and hygienic activities to improve air quality. These include control of company emissions, planning green plantations, optimizing the movement of dust collection machines in the urban environment, and other measures.