2.1.

SAR Signal Response over the Ocean Surface

Wind influenced waves are common phenomena over the ocean surfaces as well as nearby beach regions. Depending upon the intensity of the wind blowing, the backscattering range of the sea surface changes, which primarily depends on the wind speed and direction. During the calm wind blowing seasons, a regular pattern of the surface waves is observed synchronized to the direction of the wind flow. In such normal wind blowing seasons, the backscattering nature of the sea surface exhibits constant range, which can be related to the surface pattern of the ripples.^8,9 This regular pattern of the sea surface is well observed in an SAR image when comparing the same SAR image, which is multilooked at 2:4 (azimuth 2; range 4) and 4:8 (azimuth 4; range 8). When there is a cyclonic turbulence, the coastal wave crosses the normal astronomical tide level and becomes a storm surge that inundates the nearby land regions. Such rough ocean leads to temporal irregular patterns over the sea surface and nearby beach regions cause very high backscattering due to the diffuse scattering of the SAR signals.^10,11 Such irregular patterns are also observed in SAR images of large inland water bodies where wind blowing direction changes frequently or when such regions are influenced by monsoon rains. Ulaby et al.¹² studied the relationship between the roughness natures of water surfaces with the incidence angle of an SAR sensor and reported the Bragg scattering, which is a unique scattering occurring over the slightly rough water surfaces with tiny capillary waves and short gravity waves at incidence angles beyond 30 deg. Later Hasselmann et al.¹³ and Hasselmann and Alpers¹⁴ attempted in bringing out the relationship between the wavelength of the periodically spaced surface patterns and the incidence angle of the SAR signal with the help of the Bragg equation. Raney¹⁵ observed that the Bragg scatterer patterns are oriented in such a way that the geometric structures aligned with the phase fronts of the SAR signal illumination and when they are spaced periodically in range direction the backscattering strongly increases through constructive inference at certain incidence angles. Lewis et al.¹⁶ observed enhanced brightness over the oceanic surface of the SAR image, which is in general equal to or higher than the land regions of the SAR image. Using L-band-airborne SAR data Minchew et al. followed polarimetric decomposition techniques mainly to derive the information about the differences between radar backscatter of oil slick regions and adjacent seawater. It has been observed that L-band backscatter from both oil and water consistently shows the backscatter mechanism to be Bragg backscatter from a tilted faceted surface with surface scattering being the dominant type of backscatter. At incidence angles less than 50 deg, oil shows a more dominant single scatterer than water albeit with significantly lower scattered power. Minchem et al.¹⁷ found that the behavior of oil and water differ as a function of incidence angle and volumetric concentration of oil.

2.2.

SAR Signal Response over the Inland Water Bodies

The inland water bodies that have been captured by an SAR sensor during a calm wind period exhibit dark tonal contrast as the smooth surface of the water body acts as a specular reflector that directs the SAR signals away from the sensor direction thus produces dark tone in the SAR image.^18–20 Hence, to map the boundaries of the inland water bodies, the energy interaction and reflection happening over the surrounding land regions of the water body has to be understood, which together produces varying tonal and textural contrast. According to the Rayleigh criterion, such land-water tonal contrast rises with increasing incidence angle.^21–24 At the same time, higher incidence angle images are prone to the occurrences of more radar shadowing.¹³ This is most common in mountainous terrain as well as predominately seen in high-resolution SAR images. Such shadows are generally mixed with the signatures of smooth water surfaces and lead to errors of commission. In addition to this as Henderson observed in steep incidence angle images, due to abundance of layover pixels, the finer level terrain information is lost, which hinders in identifying the narrow water bodies such as the first-order drainage networks or stream networks of the steep slope regions.²⁵ With reference to frequency of the SAR signals, the X-band SAR sensors (2.4 to 3.75 cm) produces cloud shadows in the calibrated SAR image due to the attenuation of the short wavelength of the X-band signals.^26,27 During flood-related disasters, the cloud formation over the disaster region is a parallel event. Hence, higher wavelength SAR data are preferred to map the inundated regions. The next level C-band SAR signals (3.75 to 7.5 cm) has limitations such as low penetration, high-volume scattering as well as enhanced speckle effects that hinders in mapping the inundated pixels of densely vegetated regions as well as forest regions.²⁸ The L-band SAR signals (15 to 30 cm) exhibit more promising results in identifying the inundated status of agricultural regions than the C-band SAR image due to its long wavelength.²⁹ The higher the frequency of SAR signals the higher penetration capability, hence the inundated status of a disaster region can be mapped well when using high-frequency SAR data.

2.3.

SAR Signal Response over Flooded Vegetation

Interpretation of inundated vegetated regions of an SAR image requires the maximum level of understanding as SAR signals are prone to multiple scattering at various levels of vegetation as well as at subsurface or terrain level. These primarily include the scattering happening over the canopy level, from canopy to trunk level, tree stems to surface level, scattering from the rugged floor as well as from inundated water surface. The intensity of such scattering is heterogeneous in nature and varies from one image to another SAR image due to the varying nature of vegetation structures and closures. In addition to understanding such vegetation-related scattering natures, the sensor-related factors such as frequency, incidence angle, and polarization of the SAR signals are also needed to be given parallel importance so maximum possible information of the inundated status of vegetated regions can be extracted. The role and influence of all the above-mentioned parameters and related scattering mechanism happening over the inundated vegetation regions has already been widely studied by many researchers and the noted ones are Ormsby et al., Richards et al., Hess et al., Solomon, Wang et al., Henderson, Kasischke et al., Townsend and Walsh, Rao et al., Alsdorf et al., Townsend, Townsend and Foster, Costa, Lang et al., Hong et al., Voormansik et al., Pope et al., Saatchi et al.^30–51 In line with this, the following paragraph synchronizes and highlights the significant observations of relevant work performed by worldwide researchers.

Ormsby et al. discussed the list of issues related to the capability of using radar in detecting the inundation status of the vegetated regions. The magnitude at which the backscattering range increases due to flooding under vegetation is analyzed with the help of L-band SEASAT SAR, L-band shuttle SIR-A as well as with the airborne X-, C-, and L-band data. With reference to vegetation, Ormsby et al. observed that the L-band SAR data help to separate the deciduous or coniferous vegetation from the shorter, partially submerged grasses and shrubs. The X- and C-band radar data are useful when vegetation of the disaster region is relatively short in nature. With reference signal penetration and mapping the underlying inundated regions, it has been brought out that the X-band energy gets scattered at canopy level due to shorter wavelength hence there is less penetration and thus no enhancement from the water surface below. In the case of short grasslands, some of the C-band energy is reflected from the water surface and then subsequently rescattered back to the receiver producing a brighter return. Whereas the L-band radar able to produce a quantitative change in response between flooded and nonflooded conditions of the vegetated terrain ranging from 3 to 6 dB as this is able to penetrate deep into the trunks and reflect the signals from the below-standing water.³⁰

Richards et al. have clearly defined the four kinds of SAR scattering mechanism, which together contributes to the return signals of flooded forests. Richards et al.’s illustration is shown in Fig. 1.³¹ This includes scattering from the trunk to the forest surface, diffuse scattering from the rugged forest floor, volume scattering from the canopy, and volume to surface scattering happens at the trunk to surface level. Based on Ormsby et al. and Richards et al. studies, it can be firmly concluded that in any flooded SAR image, the combination of all types of the above-mentioned scattering leads to a composite backscattering value over the forested region, which is higher than the nonflooded conditions.^30,31 The same has been confirmed subsequently by Wang et al., Kasischke et al., Townsend, and Hong et al.^37,39,43,48

Fig. 1

Four components of backscatter from forest region, Richards et al.³¹

In addition to the above studies, Hess et al., Townsend and Foster, and Lang et al. have published comprehensive reviews of the interaction of various radar systems with a wide range of vegetation types.^{32–35,45,47} All such studies insist on using the longer wavelength SAR as the longer the wavelength the higher is the capability of the signal to penetrate the vegetation canopy. This is particularly significant when the wavelength is substantially longer than the leaf size.^37,50 Lang et al. found that the increase in the canopy foliage leaf-area index reduces the transmissivity of the crown layer and thus diminishes the amount of microwave energy reaching the forest floor.⁴⁷ In general, over such flooded forest regions, high classification accuracy is obtained when using the high wavelength SAR signals during the leaf-off conditions.⁴⁵ In this regard, the L-band SAR sensors have proven to be more effective in mapping the flooded forest.^30,31,40,52

In case of using the C-band SAR data, publications from Townsend and Walsh, Rao et al., Townsend, Costa, and Lang et al. prove that C-band SAR data still act as a valuable tool in mapping the flood beneath the forest regions where canopy is sparse and short.^{40,41,43,44,46,47} Horritt et al.¹⁹ observed enhanced backscatter of marshland environment in the C-band image rather than in L-band as the C-band signal able to penetrate the sparse canopy cover as well as able to interact with the water surface of the lower part of the flooded vegetative region. Over a lake environment, Alsdorf et al.⁴² reported increase in the C-band backscatter value due to the presence of floating aquatic macrophytes as well as budding shrubs.

While using X-band SAR data, Henderson, Solomon, and Ormsby et al.^30,36,38 observed the prominent occurrence of double-bounce returns from the edges of the sparse flooded forest trees where the penetration of X-band signals is not an issue. Ormsby et al. ³⁰ have brought out the relationship of SAR wavelength over flooded grasses, herbaceous, and timber vegetation, which clearly exhibits the enhanced backscattering of X- and C-band SAR data over the marshland regions. Voormansik et al.⁴⁹ have used high-resolution TerraSAR-X SAR images and found that TerraSAR-X produced higher average backscatter value of 3.2, 4, and 6.2 dB, respectively, over mixed flooded forest areas, flooded coniferous forests regions, and flooded deciduous regions. These results were also compared with 12.5-m spatial resolution L-band ALOS-1 PALSAR and 150-m spatial resolution C-band ENVISAT ASAR images of the same flooding event and confirmed that L-band ALOS-1 PALSAR is superior for detecting floods under dense canopy.

The studies related to using C-, X-, and L-band data over the inundated vegetated regions bring out the fact that over the sparse forest cover region with thin branches and small diameter trunks, brighter backscattering effects occur even with short wavelength signals. The need of L-band SAR is unavoidable over the densely forested regions. In practical situations, temporal monitoring of any regional scale flood events of forested terrain has to make use of the available SAR sensors, which passes through the respective disaster region. Hence, understanding the response of different wavelength SAR signals over the flooded forest region is required to produce accurate mapping results.

2.4.

SAR Signal Response over Flooded Urban Region

In the 1990’s, Giacomelli et al.⁵² and Oberstadler et al.⁵³ used ERS-1 SAR data to map the inundated status of settlements along with the land use and land cover classes of the disaster region. Due to its coarse resolution of ERS-1 SAR images, Oberstadler et al. found difficulties in separating the flooded settlement from nonflooded settlement regions.⁵³ Later in the 2000’s, using an automatic texture-based maximum likelihood classifier algorithm, Solbø and Solheim²⁴ attempted in mapping the urban flood area using ERS-1, VV polarization SAR image, and brought out that in the urban areas it is nearly impossible to do a practical mapping of inundated regions due to high concentration of strong SAR scatter signature as this enhances the backscattering values of the respective disaster locations.

Mason et al.^54–57 made remarkable contributions in analyzing the inundated regions of the urban area using high-resolution TeraSAR-X data. Mason et al. have brought out the possible combinations of SAR backscattering effects of urban streets as shown in Fig. 2, which schematically shows two buildings separated by a flooded street. The dihedral and trihedral reflection from the buildings as well as the presence of metal surfaces lead to enhanced backscatter over the urban part of the image. Moreover, the double-bounce effect between roads and walls of the urban structure along with the radar-specific illumination phenomena over the urban region adversely affect the appearance and ability to detect flooding. In addition to this, considerable urban areas might not be visible due to shadowing (CD) and layover (AB) effects of SAR. The only possibility to observe an inundated region is between B and C due to specular reflection of SAR signals as shown at point Y of Fig. 2.⁵⁴ Hence, while using high-resolution SAR, when the distance between B and C increases, there is a possibility to map the flooded pixels of urban SAR image. However, when the pixels from nonflooded roads and other smooth manmade surfaces are associated with these regions, it is hard to differentiate and separate these smooth surfaces as these regions exhibit specular reflection. Overall, assessment of flooded urban regions using an SAR image is complicated due to multiple backscattering effects that are mainly dominated by the corner reflection of urban buildings and other concrete infrastructures. Later, Mason et al. designed an algorithm for SAR image-based urban inundation mapping, which requires inputs from high-resolution light detection and ranging (LiDAR) data and input from DLR TerraSAR-X simulator to generate the maps of shadow and layover regions of the urban area. Initially, the input SAR scene is logically divided into rural and urban areas and resampled to match the resolution of LiDAR DSM data. Subsequent to this, multistage object segmentation techniques have been used to separate the image water bodies, high land objects, and nonwater objects. In case of the rural part of the image, a threshold has been defined based on the context of the object with the adjacent flooded area, and objects falling to more than this threshold has been classified as flooded. In case of the urban part of the image, seed pixels are identified over the low backscattering regions as training data mainly to classify the pixels into the urban water and nonwater classes. Mason et al. ^55,56 approach involves first detecting flooding in the rural areas and then detecting it in the urban areas using an algorithm guided by the rural flood extent. Hence, the method may not produce promising results in a situation or SAR image where the flood is totally confined within the urban region. Subsequent to this, Mason et al.⁵⁷ enhanced the same flood detection algorithm based on the GO-GO double scattering method of Franceshetti et al.⁵⁸ This approach assumes that the regions of urban flooding detected by their strong double scattering could act as additional clues that could be combined with the already detected flooded urban regions that were not in radar shadow or layover.⁵⁷ For all these urban inundation studies, Manson et al. used the aerial photograph of during flood season, postflood SAR scene of the same SAR sensor of the same incident angle, and LiDAR DSM of the disaster area in enhancing the accuracy of the classification as well as in validating the obtained results.^55–57

Fig. 2

Layover (AB) and shadow (CD) regions in a flooded urban street (AD) between adjacent buildings of height h1 and h2. $\theta$ is the incidence angle, Mason et al.

Giustarini et al. followed a hybrid method of approach, combining calibration of gamma distribution function, radiometric thresholding, region growing, and change detection techniques to map the inundated pixels of during flood and postflood SAR images. Keeping the postflood image as a reference image, the permanent water bodies were extracted and with reference to this, the flood extent of the during flood image is compared to mark the inundated pixels. The results were validated with the help of 0.2-m resolution aerial photographs and found 82% accuracy is possible through this approach.⁵⁹

Pradhan et al. used the TerraSAR-X data of during flood season along with the Landsat optical data of after a flood event in mapping the urban flooded regions. After the initial stage of preprocessing, both these images were segmented as per rule-based classification approach of Taguchi technique. From TerraSAR-X, only flooded and nonflooded pixels are classified. Using Landsat image, vegetation, water, and urban objects are classified. Subsequently, the water bodies extracted from the SAR image and optical image were subtracted to identify the inundated pixels.⁶⁰