Identification of NO_x hotspots from oversampled TROPOMI NO₂ column based on image segmentation method

Satellite-based measures of NO₂ have become increasingly available for resolving the limitation on insufficient spatial and temporal coverage of ground-level monitoring networks. Oversampled NO₂ column density can obtain more detailed features of NO₂ column with a spatial resolution as high as 2 km × 2 km, while it is still challenging to identify hotspots of NO_x pollution plume in city-scale due to background interference. In this study, we proposed a method for detecting the NO_x hotspot grids from oversampled satellite NO₂ column based on the image segmentation method, and identifying major source types using Term frequency-inverse document frequency (TF-IDF). A fractal model was used to evaluate and eliminate the background portion of the NO₂ column and an adaptive threshold method was adopted to identify the region of interest (ROI) of local hotspot NO₂ column. Hot-grid index, counting the frequency of NO₂ hotspot ROI in each grid, was conducted to identify the hotspot grids. TF-IDF was used to semantically analyze the major source types of NO₂ hotspot grids. Taking Central and Eastern China as the studied domain, the hotspot grids of NO₂ and the relevant major source types were identified based on the proposed method. The major non-road mobile sources (such as Beijing Capital International Airport), industrial areas (such as Caofeidian Industrial Park) and urban areas were clearly distinguished. The power plant, Coke and Iron and Steel were identified as major source types in the whole year in the corresponding NO_x hotspot grids. Notably, the identification of hotspot grids indicated a higher probability of a local high-intensity NO_x pollution plume rather than a quantitative NO_x emission; the key source types were the semantic keywords in hotspot grids, which does not mean there were no other exiting emission sources. This proposed method has strong implications on rapidly identifying the NO_x hotspot grids based on oversampled TROPOMI NO₂ column and the list of industrial enterprises.