Sentinel–1 intensity processing chain

**Figure 2.** The intensity processing chain applied automatically to the Sentinel–1 data in the CreoDIAS cloud environment. The numbers below show how the time series of seven images are reduced to one layer for each area in which the model is applied.

#### 2.3.2. Coherence Processing

For the coherence processing, the interferometric coherence was calculated for each consecutively acquired Sentinel–1 image pair in SLC format. For a time series of seven images therefore, six coherence images were obtained. These were then averaged to reduce clutter and better distinguish roads from the surrounding sand.

The steps in the coherence generation workflow began with the calculation of precise orbits. The three subswaths of each pair were then split to enable back–geocoding, coherence generation and TOPSAR–debursting to be carried out per subswath. These were then merged, before the coherence for each full scene pair was terrain corrected, to the same map system as used for the amplitude data processing. All terrain corrected coherences were co–registered, using a cross correlation, and averaged by taking the mean coherence for each pixel. The coherence average was finally resampled to the same pixel spacing as the amplitude average for each area, to enable the simultaneous (amplitude and coherence) input to the U–Net model. Figure 3 shows the main steps of the processing chain applied automatically to the time series of Sentinel–1 SLC data.
