First evaluation of the GEMS formaldehyde retrieval algorithm against TROPOMI and ground-based column measurements during the in-orbit test period

Abstract. The Geostationary Environment Monitoring Spectrometer (GEMS) onboard GEO-KOMPSAT 2B was successfully launched in February 2020 and has monitored Asia. We present the first evaluation of the operational GEMS formaldehyde (HCHO) vertical column densities (VCDs) during the in-orbit test period (IOT) (August–October 2020) and onward by comparing them with the products from Tropospheric Monitoring Instrument (TROPOMI), Fourier-Transform Infrared (FTIR), and Multi-Axis Differential Optical Absorption Spectroscopy (MAX-DOAS) instruments. During the in-orbit test period, the GEMS HCHO VCDs reproduced the observed spatial pattern of TROPOMI VCDs over the whole domain (r=0.62) with high biases (10–16 %). In the afternoon, GEMS VCDs were too high over the west side of the tropics. We corrected this issue by adding polarization sensitivity vectors of the GEMS instrument as an additional fitting parameter in the retrieval algorithm. Using observed radiances from clear-sky pixels as the reference spectrum in the spectral fitting significantly contributed to reducing artifacts in radiance references, resulting in 10–40 % lower HCHO VCDs over the latitude including cloudy areas in the updated GEMS product. We find that the agreement between the two is much higher in Northeast Asia (r=0.90), including the Korean peninsula and East China. GEMS HCHO VCDs well captured the seasonal variation of HCHO mainly driven by biogenic emissions and photochemical activities but showed larger variations than those of TROPOMI over coastal regions (Kuala Lumpur, Singapore, Shanghai, and Busan). In addition, GEMS HCHO VCDs showed consistent hourly variations with MAX-DOAS (r=0.79) and FTIR (r=0.85) but were lower by 30–40 % relative to the ground-based observations. Different vertical sensitivities between GEMS and ground-based instruments caused these systematic biases. The use of averaging kernel smoothing method reduces the low biases by about 10 to 15 % (NMB: -48.5 % to -32.4 %, -39.1 % to -27.3 % for MAX-DOAS and FTIR, respectively). The remaining discrepancies are due to multiple factors, including spatial colocation and different instrumental sensitivities, which need further investigation using inter-comparable datasets.