Comparison Between TROPOMI and MAX-DOAS Tropospheric NO2 Observations in Shihezi: Sensitivity to Spatiotemporal Collocation

Ground-based and satellite observations differ in temporal sampling and spatial representativeness, making collocation strategy important for satellite–ground comparisons. We analyzed MAX-DOAS observations in Shihezi, Xinjiang, China, from 5 June to 31 August 2023 together with Sentinel-5P/TROPOMI tropospheric NO2 vertical column density (VCD) products. Five temporal windows (closest observation, ±30 min, ±1 h, ±1.5 h, and ±2 h) and five spatial schemes (nearest pixel, 3 × 3- and 5 × 5-pixel averages, and across-track and along-track adjacent three-pixel averages) formed 25 collocation configurations. NO2 VCD increased during the morning and gradually decreased in the afternoon, while a representative enhancement episode during 12–16 July was concentrated mainly below 1 km. Across the 25 configurations, R ranged from 0.718 to 0.852, whereas NMB ranged from −42.84% to −32.33%, demonstrating that relatively strong temporal correlation did not imply close agreement in absolute VCD. Wider temporal windows generally reduced RMSE and MAE but did not produce corresponding increases in R. The ±30 min window combined with the along-track three-pixel average produced the highest R among the tested configurations and showed comparatively stable performance, although it was not statistically superior to all alternatives. Applying the TROPOMI tropospheric averaging kernel to the MAX-DOAS profiles increased R from 0.852 to 0.879 and reduced the magnitude of NMB from 33.76% to 5.04%. Nearest-pixel-center distance (0.40–6.81 km) was only weakly associated with the relative absolute difference (R = 0.31). These findings demonstrate that collocation choices materially affect TROPOMI–MAX-DOAS consistency and should be evaluated by jointly considering physical representativeness and multiple statistical metrics.