Figure 7. Spatial distributions of color composite (RGB: Bands 4–3–2) images and corresponding AOD retrievals at 550 nm (30-m resolution) for full-scene and zoomed-in core-region views (areas outlined by red boxes) over two typical dark-surface regions: (a-c) Denver, USA and (d-f) Madrid, Spain. Identified unsuitable pixels for aerosol retrievals (e.g., clouds and snow/ice) are in black in the two rightmost mages, and the top and bottom annotations indicate acquisition times (mm/dd/yyyy, where yyyy = year, mm = month, and dd = day) and orbital records (path-row) of the Landsat images.
Figure 8 displays true-color images and corresponding AOD retrievals (at 550 nm) from two regions characterized by bright surfaces: Beijing, China, and the Sahara Desert, on different dates. Beijing, a typical city with a high population density, exhibits complex surface structures and faces substantial air pollution. A comparison of retrievals from different periods clearly indicates significant variations in pollution levels, ranging from clear to highly polluted conditions (Figure 8a-c). Spatially, a noticeable trend of increasing AOD from the northwest to the southeast is observed. This aligns with changes in land-use cover, transitioning from mountainous and forested terrain in the northwest to urban and croplands in the southeast (Figure S6c). The central core urban region experiences much higher air pollution levels than the surrounding areas due to frequent human activities that release a large amount of pollutants. In the Sahara Desert, located in northern Africa, the predominant land surface type is barren, comprising rocky deserts and sand dunes, with low vegetation cover and limited human activities (Figure S6d). The region experiences relatively high aerosol concentrations primarily attributed to frequent dust storms. Our model effectively generates smooth and spatially continuous fine-scale aerosol distributions, even over highly reflective desert surfaces, and the AOD features bear closer resemblance to surface reflectance in Sahara (Figure 8d-f). We also conducted validations of the retrievals against measurements at 8 AERONET and SONET sites in these two regions (Table S7), showing high correlations (R = 0.976 and 0.752). Approximately 52% to 76% of the retrievals met the criteria for adequate EEs. These results underscore the strong robustness of our model in aerosol retrievals over both dark and bright surfaces over land. However, it is important to note that AOD retrievals can be susceptible to interference at cloud edges or in areas with thin clouds due to inaccuracies in Landsat’s official cloud mask. This remains a challenging task for high-resolution satellites with fewer spectral channels, requiring future improvements (J. Wei et al., 2020a).