*3.2. Coherent and Incoherent DDM Observations*

The coherent and incoherent observation is determined by the threshold of the classification estimator of NIDW-derived TES. Figure 6 shows the average SM values from 9 km EASE-Grid 2.0 SMAP level-3 product, average CYGNSS gridded coherent and incoherent land surface reflectivity with the same projection grid in January 2018. Land coherent DDM can be detected in the entire footprint of the CYGNSS mission, whereas incoherent observations are more likely to occur in high altitude mountainous and hilly terrain.

**Figure 6.** The global distribution of monthly average Soil Moisture Active Passive (SAMP) soil moisture (SM) (**a**), coherent reflectivity (**b**), and incoherent observations (**c**) in January 2018.

According to the classification results, the range of coherent reflectivity is from −44 dB to −3 dB; the strongest coherent reflection indeed comes from the inland open water surface, while the area of the tropical rainforest and the arid mountainous area has the lowest reflectivity. It is worthy to note that the GNSS-R reflectivity over tropical dense forest areas is lower than the barren/desert area, which is consistent with previous studies [37]. Nevertheless, compared to the distribution of SMAP SM in Figure 6a, it can be found that the SM values in corresponding areas are high. Meanwhile, incoherent scattering rarely occurs in dense vegetation-covered areas, as Figure 6c shown. However, part of the reason is the applying of QC, which excludes some noisy DDM with lower DDN SNR. However, even we ignore the influence of QC, the count of involved incoherent observations for spatial averaging is still less than four in most of the grids, which is much smaller than

the total number of coherent measurements in the same grid. In terms of the International Geosphere-Biosphere Program (IGBP) land cover type parameters provided in the SMAP products, the statistical results also show that the proportion of coherent and incoherent GNSS-R observations over different land cover types is almost the same before and after QC. It confirms that even the dense upwelling vegetation cannot change the scattering mechanism of the land surface, but dense forest canopies will generate a strong attenuation effect on the GNSS-R coherent scattering signals, which may be attributed to vegetation volume scattering. Moreover, many coherent and incoherent overlapped areas can be found in Figure 6b,c; we speculate that the main reason is the spatial distribution of the surface roughness is different within the projected grid, so the coherent and incoherent observations can be collected simultaneously in a grid.
