3.5
Properties of SAR sensors
With the growing number of SAR satellite missions orbiting around the Earth in re- cent years, a robust flood detection algorithm that works on multi-frequency data and independently of the incident angle and the polarization is vital to process images ac- quired by a variety of sensors in different configurations. The availability of various SAR sensors increases the chances of monitoring unforeseen flash floods where the wa- ter tends to recede promptly. Furthermore, when several swaths are joined together, it is possible to track the evolution of large-scale floodings (Boni et al., 2015). But despite the availability of various types of data, certain configurations result in more accurate flood mapping than others. For example, HH-polarization provides the best discrimination between water and land, and is thus the preferred polarization to use for flood detection (Henry et al., 2006). Besides, the mapping of the flood in urban areas requires a fine spatial resolution, which depends itself on the acquisition mode. Finally, the capacity to access the inundated areas in a timely manner is essential to ensure that the acquisition date of the SAR image is as close as possible to the flood peak.
3.5.1 Incidence angle
The incidence angle (θ) is defined as the angle between the emitted radar beam and the local normal to the surface in which the incident electromagnetic (EM) wave and the reflected one both lay. Incidentally, the radar energy hits the Earth’s surface at a point called the point of incidence. The angle between the satellite nadir and where its antenna is pointing is called the look angle (Figure 3.1).
3.5.2 Frequency
The microwave spectrum was divided into multiple frequency bands, each one of them identified by a letter. SAR satellites currently in orbit operate in a single frequency band; TerraSAR-X and COSMO-SkyMed in X-band, Sentinel-1 and Radarsat-2 in C- band, and PALSAR-2 in L-band. The different radar frequencies commonly used in
3.5. Properties of SAR sensors 25
Figure 3.1: Incident angle and look angle (Lillesand et al., 2008).
SAR sensors are shown in Table 3.1 with the corresponding wavelengths. The fre- quency of the SAR sensor determines the remote sensing application and the kind of information that can be extracted from its products (Campbell and Wynne, 2011). It has been proven in Voormansik et al. (2014) that even X-band sensors with their relatively small wavelength can ensure an accurate flood mapping under sparse vege- tation canopies, at least during the leaf-off season. In the same study, L-band sensors, although being the best with respect to canopy penetration (i.e. detecting the flood
Table 3.1: Commonly-used SAR frequency bands and their wavelengths (Campbell and Wynne, 2011). Band Wavelength P-band 77 - 107 cm L-band 15 - 30 cm S-band 7.5 - 15 cm C-band 3.75 - 7.5 cm X-band 2.40 - 3.75 cm
3.5. Properties of SAR sensors 26
under vegetation canopies), were found to be quite inaccurate in the flood extent map- ping. This is due to the fact that sown soil is approximately as smooth as calm open water relatively to L-band’s long wavelength, which makes these two types of land cover appear similarly dark in PALSAR images. Consequently, a relatively smooth soil has greater chances to be misclassified as water (i.e. increasing false-alarms) with L-band sensors, than with shorter-wavelength sensors. Figure 3.2 illustrates how the radar beam has varying degrees of penetration according to its wavelength. To explain this phenomenon on actual SAR images, in Figure 3.3, the same area is perceived dif- ferently by a short-wavelength sensor compared to a long-wavelength one. Ultimately, C-band sensors, thanks to their medium wavelength, reach a good compromise between flood mapping and canopy penetration. It has to be mentioned that with the help of multi-frequency data, the flood extent could be delineated more precisely by avoiding difficulties to interpret complex backscatter variations in certain types of land cover (Boni et al., 2015).
Figure 3.2: Penetration depths of X-band, C-band and L-band frequencies in a vegeta- tion canopy, a dry soil, and a dry snow and ice (Podest, 2017).
3.5. Properties of SAR sensors 27
Figure 3.3: Different depths of penetration in a vegetation canopy with a long (P-band) and a short (X-band) radar wavelength (Campbell and Wynne, 2011). The two SAR magnitude images were taken in Colombia in 2006 by Fugro EarthData’s GEOSAR airborne sensors (Shaffer, 2008). The X-band energy stops at the top layer of the vegetation, while P-band could penetrate the canopy completely to reach the ground (darker areas on the right).
3.5.3 Spatial resolution
The spatial resolution measures the capacity of sensors in general, and SAR in par- ticular, to resolve targets on the ground. The higher the resolution, the finer are the details on the produced SAR image. The resolution of SAR images is given in two directions. The along-track (azimuth) resolution as the name implies is the resolution along the flight direction, and the across-track (range) resolution is perpendicular to the latter. X-band sensors currently in orbit (TerraSAR-X and COSMO-SkyMed) pro- vide the highest spatial resolution among all SAR sensors (Voormansik et al., 2014). COSMO-SkyMed reaches a spatial resolution of 1 m in the SpotLight mode used for civilian applications, while TerraSAR-x is capable of achieving up to 25 cm spatial res- olution in the Staring SpotLight mode. With these resolutions, flood could be detected even in complex scenarios such as urban settlements where streets are relatively narrow (Mason et al., 2014).
3.5. Properties of SAR sensors 28
3.5.4 Acquisition modes
The spatial resolution of the SAR image depends on the acquisition mode, and is generally inversely proportional to the size of the swath captured. In the Spotlight mode for instance (Figure 3.4-a), the radar’s antenna constantly points at the same relatively small patch as the sensor moves around, achieving a fine resolution. In the Stripmap mode (Figure 3.4-b) however, the fixed antenna follows the radar’s trajectory by illuminating the strip of land parallel to the flight direction. The last mode, ScanSAR (Figure 3.4-c), aims to ensure a larger coverage of the area of interest, though with a coarse geometric resolution, by switching each time to a different sub-swath (Ozdemir, 2012).
(a) (b)
(c)
Figure 3.4: SAR acquisition modes (a) Stripmap mode (b) Spotlight mode (c) ScanSAR mode (Ozdemir, 2012).