Generating 250 m-resolution regional NO 2 concentration products first from MODIS retrievals using extreme gradient boosting

Surface Nitrogen dioxide (NO 2 ) is highly related to multiple adverse human health and environmental effects. The current satellite-derived surface NO 2 products generally have a coarse spatial resolution, limiting their applications in evaluating the spatial characteristics of NO 2 in urbans and their central districts, where the NO 2 pollution and population density are relatively high. This study proposes an approach to produce the 250 m surface NO 2 concentrations using an ensemble learning model based on MODIS-derived surface PM 2.5 concentrations. It is the first time to produce such high spatial resolution of NO 2 products from the satellite retrievals with high frequency. The approach was tested over the Yangtze River Delta (YRD) urban agglomeration of China. The model has high accuracy on instantaneous NO 2 estimations with a cross-validated coefficient of determination of 0.82, a root-mean-square error of 8.5 µg/m 3 , and a mean prediction error of 6.2 µg/m 3 , and a mean relative prediction error of 20.8%, respectively. The model accurately captures the fine-detailed distribution of NO 2 concentrations over the YRD region, urbans, and their central districts in YRD under different atmospheric pollutant levels. The ultrahigh spatial resolution NO 2 products have an obvious advantage for producing finer NO 2 distribution, possibly offering a way for locating NO 2 emission sources and monitoring the local NO 2 episodes. The products are useful for air pollution monitoring, controlling, and epidemiological- and environmental-related studies, especially in urbans and their districts. The study enriches the applications of MODIS in environmental studies.