Synergy between deep learning and numerical modeling in estimating NOx emissions at a fine spatiotemporal resolution

This study focused on the remarkable applicability of deep learning (DL) together with numerical modeling in estimating NOx emissions at a fine spatiotemporal resolution in the summer of 2017 over the contiguous United States (CONUS). We leveraged the partial convolutional neural network (PCNN) and the deep neural network (DNN) to impute gaps in the OMI tropospheric NO2 column and estimate the daily complete surface NO2 map at a spatial resolution of 10 km × 10 km, showing high capability with a strong correspondence (R: 0.92, IOA: 0.96, MAE: 1.43). We then used the Community Multi-scale Air Quality (CMAQ) model at 12 km grid spacing to conduct an inversion of NOx emissions that allowed us to promote a comprehensive understanding of the chemical evolution. Compared to the prior emissions, the inversion suggested 3.21 ± 3.34 times higher NOx emissions over CONUS, significantly mitigating the underestimation of surface NO2 concentrations with the prior emissions. The results displayed the primary benefits of incorporating DL-estimated daily complete surface NO2 map, which in turn greatly reduced bias (-1.53 ppb to 0.26 ppb) and enhanced daily variability with higher correspondence (0.84 to 0.92) and lower error (0.48 ppb to 0.10 ppb) over the CONUS.