Detail of partially buried road

**Figure 10.** Close–up of the buried road segmen<sup>t</sup> shown in the very high resolution (VHR) optical data available on Google Earth Pro. The area corresponds with that shown by the red arrow in Figures 7–9. The imagery date is reported to be 5 May 2010.

#### *3.2. Grand Erg Oriental*

Figure 11 shows the roads detected by the model over a part of the Grand Erg Oriental AOI. The area includes the location of the randomly selected subset (0.2 × 0.2 degree area) in which a more accurate performance evaluation was carried out. This subset is shown in more detail in Figure 12. Figure 13 displays the Sentinel–1 input (VV backscatter) of the same area, and Figure 14 the Sentinel–2 image with the location of OSM roads overlaid. The Sentinel–2 image was acquired on 23 July 2019, which is roughly in the middle of the Sentinel–1 time series (see Table 1).

The confusion matrix for the accuracy assessment is shown in Table 6. Table 7 shows the values of various accuracy indices. The average Jaccard similarity coefficient calculated is 84% and the rank distance is 76%. As with the North Sinai evaluation, there were a few false positives, but many more false negatives. However, the evaluation subset area contains infrastructure in addition to roads. A large segmen<sup>t</sup> of missed detections, for example, includes a road running parallel with a large infrastructure installation (see Figure 15). Given the width of the structure, in particular as it appears on the Sentinel–1 data (Figure 13), the model may have misinterpreted it as a natural feature.

As with the North Sinai area, the VH backscatter over the entire area was much lower, with road features less clearly defined, while the coherence layer had greater speckle. These may be the reasons why the results were better with the VV backscatter alone.

**Table 6.** Confusion matrix for true and detected roads calculated for same area as in Figures 12–14.


**Table 7.** Accuracy indices calculated for the Grand Erg Oriental results.


**Figure 11.** Detected roads for part of the Grand Erg Oriental AOI. The yellow rectangle shows a 0.2 × 0.2 degree subset over which roads were manually digitised and a performance evaluation carried out. This area is shown in more detail in Figure 12.

Detected roads for randomly selected subset of Grand Erg Oriental AOI 

**Figure 12.** Detected roads for a randomly selected 0.2 × 0.2 degree subset over the Grand Erg Oriental AOI. White lines correspond with detected roads. The red arrow points to an example of a missed detection, perhaps due to mixed infrastructure.

Sentinel–1 VV backscsatter average input to Grand Erg Oriental model 

**Figure 13.** The Sentinel–1 average VV backscatter used as an input to the model for the Grand Erg Oriental. The area is the same as that of Figure 12. Contains modified Copernicus Sentinel–1 data 2020.

**Figure 14.** Sentinel–2 image of the same area as in Figure 12. The image was acquired on 23 July 2019 (roughly in the middle of the Sentinel–1 time series). It is displayed in true colour, bands 4,3,2 as red,

green, and blue, respectively. Overlaid in yellow are the available OSM roads for this area. Contains modified Copernicus Sentinel data 2020.
