Airborne observation with a low-cost hyperspectral instrument: retrieval of NO2 vertical column densities (VCDs) and the satellite sub-grid variability over industrial point sources

Abstract. High-spatial-resolution NO2 vertical column densities (VCDs) were retrieved from airborne observations using the low-cost hyperspectral imaging sensor (HIS) at three industrial areas (i.e., Chungnam, Jecheon, and Pohang) in South Korea, where point sources (i.e., power plant, petrochemical complex, steel yard, and cement kiln) with significant NO2 emissions are located. An innovative and versatile approach for NO2 VCD retrieval, hereafter referred to as the modified wavelength pair (MWP) method, was developed to overcome the excessively variable radiometric and spectral characteristics of the HIS attributed to the absence of temperature control during the flight. The newly developed MWP method was designed to be insensitive to broadband spectral features, including the spectral dependency of surface and aerosol reflectivity, and can be applied to observations with relatively low spectral resolutions. Moreover, the MWP method can be implemented without requiring precise radiometric calibration of the instrument (i.e., HIS) by utilizing clean-pixel data for non-uniformity corrections and is also less sensitive to the optical properties of the instrument and offers computational cost competitiveness. In the experimental flights using the HIS, NO2 plumes emitted from steel yards were particularly conspicuous among the various NO2 point sources, with peak NO2 VCDs of 2.0 DU (Dobson unit) at Chungnam and 1.8 DU at Pohang. Typical NO2 VCD uncertainties ranged between 0.025–0.075 DU over the land surface and 0.10–0.15 DU over the ocean surface, and the discrepancy can be attributable to the lower signal-to-noise ratio over the ocean and higher sensitivity of the MWP method to surface reflectance uncertainties under low-albedo conditions. The NO2 VCDs retrieved from the HIS with the MWP method showed a good correlation with the collocated Tropospheric Monitoring Instrument (TROPOMI) data (r=0.73, mean absolute error equals 0.106 DU). However, the temporal disparities between the HIS frames and the TROPOMI overpass, their spatial mismatch, and their different observation geometries could limit the correlation. The comparison of TROPOMI and HIS NO2 VCDs further demonstrated that the satellite sub-grid variability could be intensified near the point sources, with more than a 3-fold increase in HIS NO2 VCD variability (e.g., difference between 25th and 75th quantiles) over the TROPOMI footprints with NO2 VCD values exceeding 0.8 DU compared to footprints with NO2 VCD values below 0.6 DU.