Satellite instruments are very useful for studying atmospheric composition but they generally provide an estimate that is vertically integrated over the atmosphere. They are typically not capable of giving direct estimates of air quality near the ground. This PhD position has the overall goal of developing algorithms for providing satellite-based surface air quality products. Such data is directly applicable to estimating human exposure to air pollution and thus increases the societal relevance of currently available satellite-based data products on air quality. Generating surface air quality products is accomplished by exploiting synergy effects with other complementary data sources and by utilising machine learning approaches.
In the current reporting period the main focus of the project lied on generating a surface nitrogen dioxide product from the data acquired by the Sentinel-5P/TROPOMI instrument. The approach trains a machine learning model against station observations of nitrogen dioxide and uses additional information from spatial predictor variables such as land cover, boundary layer height, as well as meteorological data. The results indicate that the model is capable of estimating surface nitrogen dioxide with a mean absolute error of less than 7 µg/m3.