Using full-polarimetric SAR data to characterize the

Remote Sensing of Environment 162 (2015) 11–28

Contents lists available at ScienceDirect

Remote Sensing of Environment journal homepage: www.elsevier.com/locate/rse

Using full-polarimetric SAR data to characterize the surface sediments in desert areas: A case study in El-Gallaba Plain, Egypt Ahmed Gaber a,⁎, Farouk Soliman b, Magaly Koch c, Farouk El-Baz c a b c

Port-Said University, Dept. of Geology, Port-Said, Egypt Suez Canal University, Dept. of Geology, Ismailia, Egypt Boston University, Center for Remote Sensing, Boston, MA, USA

a r t i c l e

i n f o

Article history: Received 20 September 2014 Received in revised form 24 January 2015 Accepted 26 January 2015 Available online xxxx Keywords: Surface sediment Texture and grain size Full-polarimetric Radarsat-2 Scattering response GPR & field validation Desert environment

a b s t r a c t Mapping and characterizing surface sediment of desert environments in terms of surface texture and grain sizes provide important information on the geomorphology and depositional history to assess their potential for economic development. Accurate maps of surface sediment are often not available at the desired scales using conventional field surveys. Optical and radar remote sensors are increasingly used as an alternative method for rapid delineation and classification of surface sediments. However, each sensor provides different levels of information and accuracy. In this research work, six datasets (ALOS/AVNIR-2, ASTER, AVNIR-2/PALSAR, ASTER/ PALSAR, quad polarized ALOS/PALSAR and quad polarized Radarsat-2) were prepared to run unsupervised and supervised classifications to compare their accuracy for mapping surface sediment in the non-vegetated ElGallaba Plain of the Western Desert of Egypt, particularly in terms of grain size and surface roughness. The accuracy level of all classification outputs, as well as the official geologic map, in comparison with the reference field information is empirically evaluated using the standard accuracy assessment method (confusion matrices). The best overall accuracy and dataset is the supervised classification using the complex coherence matrix [T3] of the full polarimetric Radarsat-2 which is 96.43% with Kappa coefficient of 0.929. Therefore, the zonal statistics method is applied between the final supervised classification map with its five classes and the generated Entropy (H), Alpha (α°), Anisotropy (A) and the Shuttle Radar Topography Mission (SRTM) to extract information regarding the polarimetric response as well as the elevation of each sediment class. In addition, the polarization signatures of these five classes are extracted in order to examine their scattering mechanism and to distinguish the differences between them in terms of differences in their surface roughness and grain sizes. In addition, the low frequency ALOS/PALSAR data reveal a set of linear features that are hidden under windblown sand and are not visible in the field or in other datasets. These hidden lineament features were confirmed in the field by a ground penetrating radar survey using 100 MHz shielded GSSI antenna. The GPR profiles show very clear discontinuities in the subsurface layers, which are covered by a thin sand sheet. Furthermore, the complicated depositional history of these surface sediments is proposed based on field work and remote sensing analyses. Consequently, the results show that El-Gallaba Plain has high potential for agricultural production using its groundwater resources. © 2015 Elsevier Inc. All rights reserved.

1. Introduction Characterizing desert surface sediments and their spatial distribution in terms of surface roughness and soil grain/rock fragment size can reveal information about the depositional history and environment as well as active processes (Abrahams & Parsons, 1994; Glennie, 1970). Accurate maps of surface sediments in non-vegetated, large, flat and almost featureless desert areas are often not available at the desired scales (1:25,000 or less) because producing them using conventional field survey techniques is challenging, time consuming and very expensive. ⁎ Corresponding author at: Geology Department, Faculty of Science Port Said University 23 December Street, Port Said 42522, Egypt. E-mail address: [email protected] (A. Gaber).

http://dx.doi.org/10.1016/j.rse.2015.01.024 0034-4257/© 2015 Elsevier Inc. All rights reserved.

Space-borne remote sensing with its bird's view capability is being used as an alternative method for rapidly delineating and classifying surface sediments in arid environments (Bartholdy & Folving, 1986; Gaber, Koch, & El-Baz, 2010; Koch & El-Baz, 2000; Mather, Brandt, & Koch, 1998). Currently there are many satellite sensors operating in space and each one of them provides different levels of information about the illuminated targets. However, the optical/thermal and synthetic aperture radar (SAR) sensors are most useful for mapping and classifying the surface sediments. Usually the reflected/emitted signals of optical/thermal satellites convey information about the color, chemical and thermo-physical properties of the top most part of the ground surface, unlike the backscattered SAR waves which provide information about the physical properties of the objects (Elachi et al., 1990; Evans, Farr, Ford, Thompson, & Werner, 1986; Gerstl, 1990). In addition, the

12

A. Gaber et al. / Remote Sensing of Environment 162 (2015) 11–28

low frequency radar sensors have the capability to penetrate the relatively flat, dry and homogeneous desert sediments and provide information about the near-surface structures and geomorphology (Dierking, 1999; Evans, Farr, & van Zyl, 1992; Gaber, Koch, Helmi, & Sato, 2011; Gaber, Koch, Helmi, Sato, & El-Baz, 2013; Gaddis, Mouginis-Mark, & Hayashi, 1990; McCauley et al., 1982; Schaber, McCauley, Breed, & Olhoeft, 1986). Hybrid images of optical and single polarization SAR satellites have been used to perform multisource data fusion and texture analysis, in order to classify the surface deposits based on grain size distribution and predominant rock composition (Gaber et al., 2010; Koch & El-Baz, 2000; Saraf, 1999; Schistad, Jain, & Taxt, 1994; Yang, Cauneau, Paris, & Ranchin, 2000). However, the sediment classification based on the intensity of the backscattered energy only, which is called radar crosssection (σ°) of the calibrated single polarized SAR data, may lead to inaccurate information about the surface roughness and grain size. The amount of backscattered SAR waves are affected by two different parameters, namely the surface roughness (root mean square height) and dielectric constant of the sediments (ε0) which shows a linear relationship (Laur et al., 2002; Melsheimer, Tanck, Gade, & Aplers, 1999). The degree of surface roughness is controlled by the general topography and grain size of the surficial sediments, while the value of dielectric constant is based on the volumetric moisture contents (θv) and the conductivity (homogeneity) of the sediments (Baghdadi, Gaultier, & King, 2002; Baghdadi, Holah, & Zribi, 2006; Mattia & Le Toan, 1999; Moran, Hymer, Qi, & Sano, 2000; Moran, Peters-Lidard, Watts, & McElroy, 2004). Increased homogeneity means less conductive and low dielectric constant and low backscattered SAR waves, and vice versa. This complicated relationship between the amount of backscattered SAR waves and the target's properties increases the ambiguity in the interpretations using single polarization SAR data (Hugenholtz & Van der Sanden, 2001; Zebker, van Zyl, & Held, 1987). Single polarization SAR data provides a single value of reflectivity for each pixel in a scene, whereas full-polarimetric SAR data provide additional information such as the dependence of reflectivity on polarization for each pixel in the scene by measuring both the intensity of the backscattered waves and changes in its phase (ϕ) (Dobson, Pierce, & Ulaby, 1997). The radar polarization signature of a target permits stronger inferences of the physical scattering mechanism than singlepolarization measurements through identification and characterization of the dominant scattering mechanism; thus the solution for geometric shape and dielectric constant of an object is less ambiguous (Boerner et al., 1998; Zebker et al., 1987). Thus such imaging radar polarimetry can be used for soil discrimination in terms of surface roughness and terrain, grain size, volumetric moisture content and homogeneity (van Zyl, 1989; van Zyl & Zebker, 1990; Zebker & van Zyl, 1991). Recently, the fully polarimetric information (HH, HV, VH and VV) of SAR data have been used to map and distinguish between different geomorphic targets based on their polarization response (Cloude & Pottier, 1997; Lee et al., 1999; Papathanassiou & Buchroithner, 1993; Sarah et al., 2011; van Zyl, 1989; van Zyl, Zebker, & Elachi, 1987; Zhang et al., 2011). Thus the main objective of this study is to test and evaluate the performance of full-polarimetric SAR data (L-band, 24 cm and C-band, 5.6 cm) in discriminating and characterizing surface sediments in desert areas in terms of grain size. This may provide important information for more detailed soil mapping and reveal information about the depositional history and environment, as well as the processes actively shaping that environment. 2. Study area The study area is located north west of Aswan city, in the Western Desert between latitude 24° to 25° N and longitude 31° 30′ to 33° E, and comprises the El-Gallaba Plain, the western part of the Kom Umbo basin, and Wadi El-Kubanyia (Fig. 1). It is characterized by an arid climate with desert-like conditions. Although rainfall is not significant

throughout the year, some rare and irregular storms take place over scattered localities during the winter season. El-Gallaba Plain is composed of five main types of landforms. These are: (1) the young alluvial plain of the Nile. (2) The old alluvial plains of the Nile which comprise the terraces found at various heights on both the eastern and western sides of the Nile Valley. (3) The calcareous structural plateau and its bounding slopes. (4) The structural plains which occupy the flat portion of the area. These are essentially underlain by the Nubian Sandstone, and (5) the desert hydrographic basins which include the dry drainage channels leading to the Nile Basin and traversing the structural plains and the calcareous plateau. The tectonic framework of El-Gallaba Plain is related to the Last African Orogenic belt (Abd El-Razik & Razavaliaev, 1972; Said, 1962). The entire Nile Valley in Egypt is controlled by wrench faults that are generally parallel either to the Gulf of Suez or the Gulf of Aqaba directions (Youssef, 1968). The stratigraphic sequence of the study area ranges in age from PreCambrian to Quaternary. The Pre-Cambrian rocks consist mainly of igneous and metamorphic rocks. The sedimentary section overlying the basement complex ranges in age from Paleozoic to Recent. Thus, the study area has been affected by the same structural deformation processes that formed the Nile Valley and shaped the Kom Umbo basin that lies east of the Nile River. The study area lies in a relatively large basin (Kom Umbo basin) which received a significant amount of surface runoff during the rainy season from the eastern Red Sea Mountains range and western calcareous plateau as well (Gaber et al., 2011; Koch, Gaber, Burkholder, & Mohamed, 2012, Koch et al., 2013). The top sediments in El-Gallaba Plain as described in the official geological map of Aswan, Egypt (sheet NG-36 B with scale 1:250 000) are typically characterized as sand, mix of gravel/sand and gravel sediments of fluvial and aeolian deposits (EGSMA et al., 2005) (Fig. 2). These surficial sediments originated from different sources and different directions. 3. Data and methods In this research, different surface sediment classification methods (unsupervised and supervised) using optical and different radar space-borne datasets are performed and compared. This is to figure out which technique and satellite information is the most proper one and accurate in terms of sediment characterization and information inferred about their depositional history in the non-vegetated desert areas. These unsupervised and supervised classification methods are described in details in the sections below. Moreover, the remote sensing extracted information is complemented by a field campaign carried out in March 2014 to obtain detailed in situ information and examine the image processing results. Such field visit served mainly to collect field reference points for supervised classification and validate the different unsupervised classification results, estimate the depositional direction of the different surface sediments and acquire radar profiles by means of ground penetrating radar (GPR) survey at some selected sites. 3.1. Dataset characteristics Images from four sensors (two optical and two SAR) were utilized in this work. These include two scenes of ALOS/AVNIR-2 which cover the visible and near-infrared range of the electromagnetic spectrum (4bands ranging from 0.42 to 0.89 μm) and acquired on 23-May-2009 with a spatial resolution of 10 m. Also two scenes of ASTER which cover the visible and short wave near-infrared range of the electromagnetic spectrum (9-bands ranging from 0.52 to 2.43 μm) acquired on 5-Feb-2008 with a spatial resolution of 15 and 30 m respectively, represent the used optical data (Table 1). The SAR dataset is composed of (1) two ALOS/PALSAR (L-band, single HH-polarization, ascending orbit) scenes acquired on 25-Feb-2008, with a spatial resolution of 6.25 m and an incidence angle of 34.3°. (2) Also ALOS/PALSAR (L-band, full polarimetric, ascending orbit) scenes acquired on 29-Nov-2009,

A. Gaber et al. / Remote Sensing of Environment 162 (2015) 11–28

13

Fig. 1. Landsat ETM image of the study area west of Kom Umbo.

with a spatial resolution of 3.75 m (Az.) × 9.36 m (Ra.) and an incidence angle of 25.6°, and (3) six Radarsat-2 (C-band, full polarimetric, beam mode SQ3 and SQ8, ascending orbit) scenes acquired on 6 and 13July-2014, with a spatial resolution of 4.77 m (Az.) × 7.98 m (Ra.) and an incidence angle of 26.9°. Table 1 shows the detailed descriptions of the satellite data used in this research work. 3.2. Image processing methods and field validation Classification algorithms are generally grouped into unsupervised and supervised methods. In the unsupervised case, the algorithm has no prior information of the scene content or of the terrain classes present in it, while in the supervised case, the algorithm has prior information (Congalton & Kass, 1999). Both methods represent well established and widely used image processing techniques. However, their use with full polarimetric SAR data for surface sediment classification in arid land is still rare, although other more sophisticated textural (e.g., gray-level co-occurrence matrix) and spectral (e.g., multi-layer feed-forward neural network) analysis methods exist (Mather et al., 1998). In this

research work, six different datasets were prepared to run the unsupervised and supervised classifications for mapping the surface sediment in the non-vegetated El-Gallaba Plain in terms of grain size and surface roughness. The accuracy and uncertainty of each method using these datasets have been calculated using field-based information. These six datasets are the optical ALOS/AVNIR-2 and ASTER, the fused AVNIR-2/ PALSAR and ASTER/PALSAR and the full polarimetric ALOS/PALSAR and full polarimetric Radarsat-2 (Fig. 3). These datasets have been clipped to a small common area to fairly and accurately compare their classification accuracy. The most accurate method and dataset has been used later to repeat the classification for the entire study area. 3.2.1. Unsupervised classification 3.2.1.1. Optical data. Two scenes of ALOS/AVNIR-2 optical data were geocoded, atmospherically corrected, mosaicked, clipped and used to generate an unsupervised classification map. This unsupervised classification map was performed using K-means clustering algorithm (with 12 initial classes and 10 maximum iterations). In this work, the selection

14

A. Gaber et al. / Remote Sensing of Environment 162 (2015) 11–28

Fig. 2. Surface sediments in El-Gallaba Plain. Modified after EGSMA et al. (2005).

Table 1 Descriptions of the satellite datasets used in this study. Configurations

Multi-spectral optical data

Synthetic aperture radar (SAR) data

ALOS/AVNIR-2

ASTER

Acquisition date Wavelength Spatial resolution

May 23, 2009 0.42 to 0.89 μm (4-bands) 10 m

Level product Incidence angle at scene center Orbit pass Noise equivalent (NE σo) Absolute geo-location accuracy Absolute radiometric accuracy

O1B2R NA NA NA b114.2 m ≤5%

Feb. 5, 2008 0.52 to 2.43 μm (9-bands) VNIR: 15 m SWIR: 30 m AST_L1B NA NA NA b50 m ≤4%

Single HH-polarization

Full-polarimetric SAR (HH, HV, VH & VV)

ALOS/PALSAR

ALOS/PALSAR

RADARSAT-2

Feb. 25, 2008 24 cm, 1.27 GHz (L-band) 6.25 m

Nov. 29, 2009 24 cm, 1.27 GHz (L-band) Az.: 3.75 m Ra.: 9.36 m P1.1 25.6° Ascending −30 ~ −31 dB b200 m b1.5 dB

July 6–13, 2014 5.6 cm, 5.4 GHz (C-band) Az.: 4.77 m Ra.: 7.98 m SLC_1.1 26.9° Ascending −40 ± 3 dB b20 m b1 dB

P1.5 34.3° Ascending −24 ~ −27 dB b200 m b1.5 dB

A. Gaber et al. / Remote Sensing of Environment 162 (2015) 11–28

15

Fig. 3. Show the used satellite data (a) optical AVNIR-2, (b) optical ASTER, (c) hybrid AVNIR-2/ PALSAR (HH), (d) hybrid ASTER/PALSAR (HH), (e) full-polarimetric ALOS/PALSAR-and (f) full polarimetric Radarsat-2 (the road and power line appears as bright lines in e & f).

of initial number of classes was based on the spectral resolution of the respective sensor and desired number of output classes, where we used to set the initial classes to be three times the used spectral bands. In order to obtain an unsupervised classification map for the surficial sediments only in El-Gallaba Plain, the undesired classes were assigned as unclassified. Three initial classes were chosen to classify the three main classes which are reported in the official geologic map (Fig. 2). In addition, the application of a majority averaging filter with a 7 × 7 kernel size was performed to the AVNIR-2 classified images to improve the results by changing spurious pixels within a large single class to that class, thus, reducing the variance within the resulting clusters (Tomas, 1980; Townshend, 1986; Walter, 2004). The VNIR and SWIR bands of the ASTER data were geocoded, atmospherically corrected, stacked, clipped to the common area and used to run the same unsupervised classification algorithm that was mentioned in previous section using the K-means clustering algorithm (with 30 initial classes and 10 maximum iterations). These 30 initial classes were

later combined and assigned into only three main classes after eliminating the undesired classes (highway, power line, vegetation, urban and rocky areas). In addition the majority averaging filter with a 7 × 7 kernel size was performed (Fig. 4b).

3.2.1.2. Fused optical/SAR data. Multisource data fusion was performed using optical and microwave data sets, i.e., ALOS/AVNIR-2 with ALOS/ PALSAR and ASTER with ALOS/PALSAR data. The main idea behind this fused method is classifying the surface sediments in El-Gallaba Plain based on the extracted properties from the multi-spectral information of optical data as well as the SAR data. The data fusion method used in this study is the Principal Component (PC) spectral sharpening algorithm (Varma, 2003; Vrabel, 1996; Welch & Ahlers, 1987). It is a popular data fusion method and normally used to sharpen a low resolution multi-band image (optical bands) with a high resolution panchromatic band (calibrated HH-ALOS/PALSAR).

16

A. Gaber et al. / Remote Sensing of Environment 162 (2015) 11–28

Fig. 4. Buried lineaments appear in ALOS/PALSAR L-band (a) and the unsupervised map (b).

The calibrated radar cross-section (backscattered coefficient (σ°)) is a response to both surface and system parameters (Hugenholtz & Van der Sanden, 2001). Surface parameters include surface roughness (i.e., centimeter to meter scale) and soil dielectric properties attributable to mineralogy, grain size, soil moisture, and soil structure. System parameters include wavelength (λ), incident angle (θ°), and polarization. However the shorter wavelength of SAR data is much more sensitive than the longer one to differentiate between surface sediment based on their grain size, where the coarse grains have a higher radar cross section (σo) (more backscatter energy) than the fine grains (Bartholdy & Folving, 1986; Gaber et al., 2010; Koch et al., 2012; Papathanassiou & Buchroithner, 1993; Tanck, Alpers, & Gade, 1999; van Zyl et al., 1987). In this work, two ALOS/PALSAR (L-band, HH-polarization, ascending orbit, processing level 1.5) scenes were acquired on February 25th,

2008, with a high spatial resolution of 6.25 m and an incidence angle of 34.3°and were filtered using refined Lee filter with window size 7 × 7 (Lee, 1981), calibrated (Gray, Vachon, Livingstone, & Lukowski, 1990; Shimada, 2005; Shimada, Isoguchi, Tadono, & Isono, 2009) and mosaicked to estimate the backscattered coefficient (σ°) in dB scale for each pixel in the scene. This calibrated and filtered HH-ALOS/PALSAR data were used to generate two fused images (AVNIR-2/PALSAR and ASTER/PALSAR) with the geocoded and atmospherically corrected optical data. The same unsupervised classification algorithm, steps and parameters which were followed before were performed on the fused data. 3.2.1.3. Full polarimetric SAR data. Full polarimetric SAR data have the advantage over single polarization data that the 2 × 2 scattering matrix can be used to get parameters that provide reasonable class

A. Gaber et al. / Remote Sensing of Environment 162 (2015) 11–28

separation (Cloude, Pottier, & Boerner, 2002; Dong, Milne, & Forster, 2001; Evans, Farr, van Zyl, & Zebker, 1988; Hugenholtz & Van der Sanden, 2001; van Zyl, 1989). Moreover, radar interaction with surficial deposits can occur at two levels: (1) according to the roughness of the illuminated ground and (2) signal scattering interactions with large individual objects, from pebbles to boulders, whose size is much greater than the radar wavelength (Boerner et al., 1998). Linear co-polarized microwaves (HH and VV) are generally more sensitive to the orientation of the target relative to the transmitted polarization whereas cross-polarized waves (HV and VH) are more sensitive to multiple scatter at the surface and subsurface if penetration occurs (Hugenholtz & Van der Sanden, 2001). Thus, the crosspolarized waves (HV and VH) increase by increasing the grain size and surface roughness of the desert sediments. In this study, the fully polarimetric (HH, HV, VH and VV) ALOS/ PALSAR L-band and Radarsat-2 C-band data have been used to classify the surface sediments along El-Gallaba Plain using the radar extracted information from the power and scattering mechanism of signals (amplitude and phase). Thus, two scenes of quad polarization ALOS/PALSAR L-band (1.27 GHz, 24 cm) and six scenes of standard quad polarization Radarsat-2 C-band (5.4 GHz, 5.6 cm) were used. Polarimetric SAR is a technology that exploits the polarized nature of electromagnetic waves in order to extrapolate multi-dimensional information from imaged targets (Boerner et al., 1998). The polarization of a wave is the description of the spatial orientation of the electric vector for a given wave (Hugenholtz & Van der Sanden, 2001). In single-polarization, the electric field of SAR data is recorded as either H- or V-polarized wave or a single scattering coefficient is measured for thousands of points in the scene. In this configuration SARs only measure a scalar quantity and lose the polarization properties of the reflected wave. In fully polarimetric SARs the polarization of the reflected wave is measured as a vector quantity and the polarization signature of every resolution cell that can be recorded (Boerner et al., 1998). Fully polarimetric SAR data provides the 2 × 2 target scattering matrix, also referred to as the Sinclair scattering matrix:
S¼

SHH SVH

 SHV : SVV

T 11  ½T 3  ¼ 4 T 12  T 13

T 12 T 22  T 23

K 3p ¼ ½S11 þ S22 S11 −S22 S12 þ S21 

17

ð4Þ

where b N denotes an eventual incoherent averaging operation. The [T3] matrix of each quad SAR scene was imported into NEST ESA SAR tool box and its new metadata files were replaced with the corresponding metadata files of the original raw quad SAR data, in order to recover the geo-reference information of each pixel in the newly transformed T3 image. Consequently, the new T3 matrices of both ALOS/ PALSAR and Radarsat-2 were corrected by applying the multilook operator to reduce the inherent speckle appearance and produce images with nominal pixel size. In addition, the terrain correction process was performed on each multi-looked T3 SAR data for geocoding in order to compare with the others unsupervised maps. The same k-mean unsupervised classification method used in previous sections was performed on the new T3 matrix of ALOS/PALSAR and Radarsat-2, but this time using the NEST software. 3.2.2. Field validation and Ground Penetrating Radar (GPR) survey Further investigations were carried out in the field to validate the different unsupervised classification maps and collect field-based information to assess their classification accuracy. Field measurements were carried out in March 2014 and consisted of describing the texture and composition of the surface sediments in El-Gallaba Plain, measuring the depositional directions to know from where these sediments were derived from and conducting ground penetrating radar (GPR) survey to validate the linear features identified in the full polarimetric PALSAR dataset, which might be related to subsurface structures (Fig. 4).

ð2Þ

ð3Þ

3.2.2.2. GPR Survey. GPR is an effective technique for the non-invasive investigation of the subsurface (Brunzell, 1997; Daniels, 1996; Olhoeft, 2000). Consequently, the GPR method was used to verify the existence and nature of the linear features that appears only on the full polarimetric

ð1Þ

3 T 13 T 23 5 T 33

†

⇒½T 3  ¼ K 3p :K 3p

E

3.2.2.1. Field validation. More than 97 sites were checked out and described in the field to figure out their sedimentary compositions, grain size, moisture contents, surface roughness, local topography and homogeneity (Fig. 5). All these soil parameters affect the amount and mechanism of the optical and radar signal reflected and scattered back to the satellite sensors. In addition, the naturally cropped-out sediments and quarry sites along the study area were visited and inspected to accurately measure the depositional directions of the topmost (~2 m) fluvial and aeolian deposits in order to define the depositional direction of these sediments and establish the paleo-current. This was done by selecting the wellpreserved and represented depositional layers and measuring the directions of the elongate gravel contents using the compass (Fig. 5a). This information helps to understand the paleo-depositional history (source, transport agent and paleo-topography) of the topmost part of the sedimentary cap in El-Gallaba plain. In the field, we found that the surface sediments in El-Gallaba Plain mainly consist of five different sedimentary classes, which are (1) gravel-1 which is located mainly in the east of the study area, with a thick layer of well-sorted gravels, mostly flint, reddish in color because of high content of iron oxides; (2) gravel-2 which is located mainly in the north-west of the study area and consists of ill-sorted gravels from calcareous rock fragments with white and desert-varnished color; (3) gravel/sand-1 which surrounds the low areas west of gravel-1 class and is composed of mixture of reddish gravel (mostly flint) and sand with minor silt content; (4) gravel/sand-2, which is located in the western most part of the study area and at the footwall of the western calcareous Sin El-Kadab Plateau, and consists of a mixture of desert pavement sediments, calcareous rock fragments and wind-blown sand; and (5) sand distributed throughout the study area as well-sorted, wind-blown loose sand filling low areas (Fig. 5).

Knowledge of the scattering matrix permits calculation of the received power for any combination of transmits and receives antennas and is referred to as polarization synthesis (van Zyl & Zebker, 1990). Reciprocity dictates that HV = VH for monostatic polarimeters where the target is not affected by Faraday rotation (Ulaby & Elachi, 1990). In this work, the quad polarization ALOS/PALSAR and Radarsat-2 were used to classify the surficial sediments of El-Gallaba Plain based on the energy and scattering mechanism (degree of randomness) of radar signals that will be scattered back from different soil characteristics. Each quad polarization image was extracted using PolSARpro open source software and transformed from (2 × 2) Sinclair Matrix into (3 × 3) complex Coherence matrix [T3]. The (3 × 3) Coherency matrix [T3] is constructed from a three-element unitary target vector, obtained from the projection of a Sinclair matrix onto a reduced and modified Pauli spin matrix set (Boerner, 2007). It is an incoherent polarimetric representation relating to second order statistics of scattering matrix elements. This matrix is hermitian semi-definite positive (ESA, 2008): 2

D

18

A. Gaber et al. / Remote Sensing of Environment 162 (2015) 11–28

Fig. 5. Measuring the direction of paleo-current (a), GPR survey (b), measuring the soil moisture using TDR (c), and the field recognized five sedimentary classes of gravel-1 (d), gravel-2 (e), gravel/sand-1 (f), gravel/sand-2 (g) and sand (h).

ALOS/PALSAR data (Fig. 4). Such kind of information helps to interpret and know more about the spatial distribution of the surface sediments in El-Gallaba Plain and its depositional history. A total of 5 GPR profiles were acquired using the commercial GSSI 3000 Digital GPR unit with a 100 MHz shielded antenna in monostatic mode (Fig. 5b). The GPR profiles were carried out perpendicular to the linear features with a total length of each profile about 730 m, except profile no. 4 and 5 which totaled 1300 m and 1500 m respectively. All profiles were processed by applying dewow, zero-time correction, manual gain and a background removal. A dielectric constant of 3 was chosen for the dry sand surface material to convert the time range setting into penetration depth.

classification method, which is implemented in the PolSARPro 4.2 software (Cloude & Pottier, 1995), was used to apply the supervised classification for the full polarimetric ALOS/PALSAR and Radarsat-2 datasets. To apply supervised classification to the full polarimetric SAR data, a total number of 80 reference points, which represent three classes, were converted into pixel numbers for both ALOS/PALSAR and Radarsat-2 datasets using the backward geocoding techniques (Bara, Broquetas, Scheiber, & Horn, 2000). These pixels numbers were used to draw training areas in PolSARPro to apply the Wishart supervised classification.

3.3. Classification accuracy assessment 3.2.3. Supervised classification A total number of 97 field points representing the five recognized sedimentary classes (23 points for gravel-1, 38 points for gravel-2, 13 points for gravel/sand-1, 9 points for gravel/sand-2 and 14 points for sand) were used to run the supervised classification method using the aforementioned six images. All these reference points were chosen from the field visit. However, to fairly compare the accuracy of the official geologic map, the generated unsupervised maps and supervised ones, the five sedimentary classes which are recognized in the field were combined to three classes only after merging the gravel-1 with gravel-2 to be assigned as gravel, and merging the gravel/sand-1 with gravel/sand-2 to be assigned as gravel/sand, and the third class being sand. The minimum distance supervised classification technique, which is based on the minimum distance decision rule that calculates the distance between the measurement vector for the candidate pixel and the mean vector for each sample, was applied on the optical and fused satellite datasets. This assigns the candidate pixel to the class having the minimum distance (James & Daniel, 2002). The Wishart supervised

The current standard method for assessing the classification accuracy focuses on the error or confusion matrix, which summarizes the comparison of map class labels with reference field data labels (Congalton, 1991; Congalton & Kass, 1999; Rosenfield & Fitzpatrick-Lins, 1986). Because available reference field data only partially covered the study area, pixels within each map were not equally likely to be selected for sampling. Thus, the expected classification accuracy and level of confidence obtained from a particular accuracy assessment approach depend mainly on the number of samples of each class involved in the analysis (Congalton & Kass, 1999). Two important types of information can be derived from the error matrix: errors of omission, or producer's accuracy; and errors of commission, or user's accuracy (Cambell, 2002; Lillesand & Kiefer, 1994; Story & Congalton, 1986). User's accuracy for a specific class is the ratio of the correctly classified samples to the total number of samples selected in that class. Likewise, producer's accuracy for a class is derived by taking the ratio of correctly defined pixels to the total number of pixels selected for that class in the reference data. A kappa coefficient is commonly used to evaluate the confusion matrices

A. Gaber et al. / Remote Sensing of Environment 162 (2015) 11–28

and measure the map accuracy (Cohen, 1960, 1968; Congalton & Kass, 1999; Hudson & Ramm, 1987; Skidmore, 1999). The number of sample points is usually calculated using the following equation based on binomial probability theory for producing a reliable confusion matrix for accuracy assessment (FitzpatrickLins, 1981): N¼

Z2 pq E2

ð5Þ

where N = number of samples, p = expected or calculated accuracy (%), q = 100 − p, E = allowable error, and Z = standard normal deviate for the 95% two-tail confidence level = 1.96. For the lowest expected map accuracy of 60% with an allowable error of 5%, 369 sample points were required. Classes of small area would not be sampled sufficiently to detect classification errors (van Genderen & Lock, 1977). Work by Congalton (1991) and Congalton and Kass (1999) suggest that sample sizes derived from multinomial theory are appropriate for comparing class accuracies, with minimum sample points of 80 per class. A total number of 800 reference points, which are totally different than the points that used for supervised classifications and representing the five recognized sedimentary classes (160 points for each sedimentary class) were uses for assessing the classification accuracy. Most of these reference points were chosen from the field visit and the remaining from the Google Earth. Thus, for classification accuracy assessment purposes, 160 reference points per class are acceptable in evaluating large area such as El-Gallaba Plain. After evaluating the selected points in each reference data set, an error matrix was constructed, comparing map-class labels to reference data labels for each unsupervised and supervised classification map. Overall map accuracy and class-specific user and producer accuracies were calculated for each class. 4. Results and discussions The surface sediments of El-Gallaba Plain in the Western Desert of Egypt were characterized in the official geological map in 2005 as sand, mix of gravel/sand and gravel (EGSMA et al., 2005). These sediments have a longitudinal distribution pattern with almost straight boundaries (Fig. 2). However, such official geological map of El-Gallaba Plain with its coarse spatial scale (1:250 000) does not show details of each sedimentary class. We believe that the boundaries between these classes are not so sharp in reality and not located correctly from a geographical point of view. Moreover, the windblown sand of the desert environment usually masks the substrate sediments even if it is a very thin veneer of sand sheet. This result in hiding the paleo-fluvial deposits and lead to misinterpreting the depositional history. Near-surface information on the sediments is needed for accurately mapping and identifying their spatial distribution. All these challenges affect the mapping accuracy and lead to wrong or limited interpretation of the depositional history in El-Gallaba Plain. Thus, by accurately mapping and classifying the surface sediments of El-Gallaba Plain based on grain sizes will help not only in understanding the paleo-fluvial activities, but also in adding information regarding the soil condition for potential agriculture development in the study area. 4.1. Classification results and its accuracy assessment The resulting three unsupervised classes using the optical satellite images were assigned as gravel, gravel/sand and sand based on the decreasing DN (reflectance) values under the fact that the small size sand sediments will be more sorted (bright) than the larger size gravel sediments which have different kinds of rock (dark) fragments (Fig. 6a). While the resulting hybrid classification map combines two very different types of information by simultaneously classifying information on surface reflectance properties extracted from the optical sensor with backscattered information extracted from the SAR data. The three

19

resulting classes were overlain on the previously geocoded and calibrated ALOS/PALSAR data (HH) to determine the range of backscatter coefficient values (dB) per class. A similar procedure was applied in the study of alluvial wadi deposits in Sinai Peninsula (Gaber et al., 2010, 2013), and demonstrated that the backscatter coefficient (σ°) values of the sand and gravel fraction ranges from −22.0 to −17.6 dB as extracted from PALSAR L-band data. The three resulting classes of the hybrid images in this work have coefficient values (σ°) of −23.5, −20 and −16.0 dB for the sand, sand/gravel and gravel fractions, respectively (Fig. 6c & d). Since the estimated soil moisture content as measured with the time domain reflectometry (TDR) along El-Gallaba Plain is almost zero (Gaber et al., 2011) and under the assumption that the moisture contents does not change so much in such arid desert environment, the σ° of El-Gallaba Plain surficial sediments will be a response mainly to variations in grain size, mineralogical contents (conductivity) and surface roughness. Therefore, based on these properties the three classes were labeled as sand, sand/gravel and gravel and are listed in the figure legend in descending order from high σ°(gravel) and low σ°(sand) (Fig. 4c & d). In addition, the results of unsupervised classification maps using the new T3 matrix of ALOS/PALSAR and Radarsat-2 were presented in Fig. 6e & f, respectively. While Fig. 7 shows the results of the supervised classification of the six satellite datasets. These unsupervised and supervised classification maps which were generated from the optical and different SAR data products show different distribution pattern for the surface sediments in El-Gallaba Plain with different levels of accuracy. Thus, the accuracy level of all unsupervised and supervised classification outputs and the official geologic map, in terms of agreement between them and the reference field datasets, was evaluated empirically using the standard accuracy assessment method (confusion matrices) and presented in Tables 2, 3 and 4. The best overall accuracy method and dataset was the supervised classification using the full polarimetric Radarsat-2 which was 96.43% with Kappa coefficient of 0.929 (Table 4). This indicates very good agreement between such supervised classification map and the field reference data. The second overall accuracy method was the supervised classification using the full polarimetric ALOS/PALSAR dataset which was 76.92% with Kappa coefficient of 0.692. For instance, the geologic map (EGSMA et al., 2005) showed overall accuracy of 64.69% with very low Kappa coefficient of 0.253 (Table 2). The negative Kappa's coefficient which appears only in the unsupervised classification method using the fused AVNIR-2/PALSAR dataset (Table 3) indicates that the proportion of expected agreement by chance exceeds the proportion of obtained agreement (Cohen, 1960, 1968; Congalton & Kass, 1999; Hudson & Ramm, 1987; Skidmore, 1999). However, the accuracy of unsupervised and supervised classifications with less than 85% are below the acceptable level and the standard of digital image classification recommended by Paul (1991) and Jansen, Bagnoli, and Focacci (2008). Thus the supervised classification based on the backscattered signals of the full polarimetric Radarsat-2 dataset is the best and most accurate in terms of surface sediment discrimination based on their grain sizes and local terrains. This is because of the fact that the cross-polarized signals (HV and VH) of full polarimetric SAR data are very sensitive to volume scattering, which adds more information for better classifications. In addition, the classifications results using the full polarimetric Radarsat-2 are much better than using the ALOS/PALSAR, because of the technical configuration differences between both sensors. Radarsat-2 has many technical advantages over the ALOS/PALSAR in terms of accurate surface sediment classification. These technical issues are: the shorter wavelength (5.6 cm, which means higher spatial resolution), the larger incident angle, the much improved noise equivalent cross-section (NEσ°) which is about − 40 dB and the absolute geo-location accuracy which is b 20 m (Table 1). All these technical parameters are crucial to accurately mapping the surface sediments, since these surface sediments are dry, small in size compare to the radar wavelength and cover relatively flat areas and thus they are likely to have a low backscattering return.

20

A. Gaber et al. / Remote Sensing of Environment 162 (2015) 11–28

Fig. 6. The results of the unsupervised classification methods using (a) AVNIR-2, (2) ASTER, (c) hybrid AVNIR-2/PALSAR, (d) hybrid ASTER/PALSAR, (e) full-polarimetric ALOS/PALSAR and (f) full polarimetric Radarsat-2.

Thus, the full polarimetric Radarsat-2 dataset, which were acquired on 6 and 13 July 2014, were used to run the supervised classification method, for mapping the surface sediments in the entire El-Gallaba Plain as well as generate the images of Entropy (H), Alpha (α°), Anisotropy (A) and Pauli RGB to investigate the dominant scattering response of each sedimentary class. Fig. 8 shows the Pauli RGB image of ElGallaba Plain and the final supervised classification map. The Pauli RGB gives information about the dominant scattering mechanism along the study area, where the green color means volume scattering, the red means double scattering, and blue means surface scattering (Fig. 8). The black color in Pauli RGB image means Bragg scattering or the signals were totally attenuated in the case of thick loose sand sheet (Fig. 8).

Since, the Wishart supervised classification is extracted from the raw complex SAR data, these lose the phase information after geocoding, thus a total 80 reference points which represent the five recognized field sedimentary classes (gravel-1, gravel-2, gravel/sand-1, gravel/ sand-2 and sand) were converted into pixel numbers using the backward geocoding techniques (Bara et al., 2000) to generate the training sites for running the Wishart supervised classification in PolSARPro (Fig. 8). Moreover, the different scattering mechanisms Entropy (H), Alpha (α°), and Anisotropy (A) were polarimetrically generated from the mosaicked T3 matrix. All these images were then imported and geocoded in ENVI. The zonal statistics method was applied to the final supervised classification map and the Entropy (H), Alpha (α°), Anisotropy (A) and the

A. Gaber et al. / Remote Sensing of Environment 162 (2015) 11–28

21

Fig. 7. The results of the supervised classification methods using (a) AVNIR-2, (2) ASTER, (c) hybrid AVNIR-2/PALSAR, (d) hybrid ASTER/PALSAR, (e) full-polarimetric ALOS/PALSAR and (f) full polarimetric Radarsat-2.

SRTM to extract information on the polarimetric response (dominant scattering mechanism) as well as the mean elevation of each sedimentary class. Table 2 User's and producer's accuracies and Kappa coefficients for the official geologic map (EGSMA et al., 2005). Classes

Gravel Gravel/sand Sand Overall accuracy (%) Kappa coefficient

Accuracy (%) Producer's

User's

94.17 0.00 21.62 64.69 0.153

71.01 0.00 25.26

4.2. Polarimetric response of the surface sediments in El-Gallaba Plain 4.2.1. H/A/α° analysis The amount of backscattered SAR signals and its degree of randomness has a positive relationship with the standard deviation of the surface height variation (hrms) (Fig. 9). Consequently the relative variations between the surface sediments in terms of grain sizes can be extracted by investigating the power and mechanism of the backscattered SAR signals after using the proper wavelength (λ) and incidence angle (θinc ). Fig. 9 shows the effect of wavelength

22

A. Gaber et al. / Remote Sensing of Environment 162 (2015) 11–28

Table 3 User's and producer's accuracies and Kappa coefficients for the generated unsupervised classification maps. Space-borne data

Classes

Gravel Gravel/sand Sand Overall accuracy (%) Kappa coefficient

Multi-spectral optical data

Fused optical and SAR data

Full-polarimetric SAR data

ALOS/AVNIR-2

ASTER

AVNIR-2/PALSAR

ASTER/PALSAR

ALOS/PALSAR

RADARSAT-2

Accuracy (%)

Accuracy (%)

Accuracy (%)

Accuracy (%)

Accuracy (%)

Accuracy (%)

Producer's

User's

Producer's

User's

Producer's

User's

Producer's

User's

Producer's

User's

Producer's

User's

66.14 26.41 8.05 48.31 0.016

64.53 19.06 19.81

82.28 17.50 10.53 59.74 0.095

71.04 21.21 22.22

55.59 19.08 19.31 40.01 −0.019

56.99 22.01 15.98

46.21 56.94 9.83 42.68 0.069

68.90 23.36 26.84

67.93 41.67 55.56 60.75 0.343

87.98 45.45 29.17

89.27 74.36 7.41 75.06 0.474

82.92 66.67 21.05

and incident angle of radar signals in discriminating the surface sediments (Peake & Oliver, 1971). Therefore, the H/A/α° parameters were used to investigate the dominant scattering response of each sedimentary class in the study area. These three parameters (H/A/α°) have physical interpretations as described in Cloude and Pottier (1995). Entropy (H) measures the randomness of the scattering process and ranges from 0 to 1 (0 ≤ H ≤ 1). The zero value of Entropy (H) means isotropic scattering. All five classes have entropy mean values less than 0.5 and range from 0.27 to 0.37 which means low random signals compare to the rocky and vegetated areas which have high random signals. However, the differences between these classes, in terms of randomness, were matched with the slight differences between their grain sizes and local terrain. The classes were in descending order based on grain size and local terrain as: gravel-2, gravel-1, gravel/sand-2, gravel/sand-1 and sand with entropy mean values of 0.37, 0.35, 0.32, 0.28 and 0.27, respectively with standard error equal ± 0.04 (Table 5). Thus, the poorly sorted class is gravel-2 while the well sorted class is the sand and the other classes lie in between. These results are well matched with the field observations regarding the sorting degree of these five classes. The anisotropy (A) is a parameter complementary to the entropy and indicates the distribution of the two less significant eigenvalues. Anisotropy becomes 0 if both scattering mechanisms are of an equal proportion while values of A N 0 indicates increasing amount of anisotropic scattering. Thus, the anisotropy (A) defines the relation between the second and the third eigenvalues, and is a measure for the difference of the secondary scattering mechanisms. A can be zero even for rough surfaces. For surfaces characterized by intermediate entropy values, a high anisotropy indicates the presence of only one strong secondary scattering process, while a low anisotropy indicates the appearance of two equally strong scattering processes. The mean anisotropy values of gravel-1, gravel-2, gravel/sand-1, gravel/sand-2 and sand are 0.29, 0.26, 0.34, 0.28 and 0.43, respectively with standard error ± 0.06 (Table 5). From these anisotropy values and the Pauli RGB image (Fig. 8), the secondary scattering mechanism of the gravel-1, gravel-2, gravel/sand-1, gravel/sand-2 and sand can be recognized as double, volume, volume, single and Bragg scattering, respectively. The standard error of the entropy and anisotropy values is very low, which means the differences between these values are significant.

Alpha angle (α°) is associated with the type of dominant scattering mechanism of each pixel, and yields direct information about the scattering mechanism represented by each eigenvector. It ranges from 0° to 90° (0 ≤ α° ≤ 90). Alpha equal 0° means surface scattering, Alpha equal 45° means highly anisotropic or dipole scattering and Alpha equal 90° means dihedral scattering. The mean alpha angles of all five classes range from 10.01° to 13.88° which means the dominant scattering mechanism of all these five sedimentary classes in El-Gallaba Plain have mainly surface scattering, so this is why they all appear close to blue and black in color in the Pauli RGB image (Fig. 8). However class gravel-2 has the highest alpha angle with a mean value of 13.88°, which indicates that this gravel-2 class has the highest anisotropic scattering among the all five classes. Also, the elevation minimum, maximum and mean of each sedimentary class indicate that gravel-2, gravel/sand-1, and gravel/sand-2 have almost the same mean elevation (~ 148 m), while the gravel-1 class has the highest mean elevation (~ 150.5 m), unlike the sand class which has the lowest mean elevation (~ 146.6 m) (Table 5). These slightly differences in the average elevation of each sedimentary class indicate its small effect on the backscattered SAR signals. 4.2.2. Polarization signature The polarization signature technique shows the backscatter response of the target in all combinations of transmit and receive polarizations which are represented as either co-polarized or cross-polarized (Ray, Farr, & van Zyl, 1992; van Zyl et al., 1987). Polarization signatures are extracted from the scattering matrix data for identifying pixels representing specific dielectric constant or surface roughness. However, increasing surface roughness and grains size lead to increase in the pedestal height of the 3D representation of the polarization signature and makes a shift in the peak from the VV toward the HH polarization in the co-polarized signature shape (Evans et al., 1988; Zebker et al., 1987). The differences between the dielectric constants of the targets affect only on the backscattered power values (z axis) of the 3D graph of the polarization signature. Therefore, the polarization signatures of the five classes of the supervised classification map were extracted in order to examine their scattering response and distinguish the differences between these classes in terms of differences in their surface roughness and grain sizes. The

Table 4 User's and producer's accuracies and Kappa coefficients for the generated supervised classification maps. Space-borne data

Classes

Gravel Gravel/Sand Sand Overall accuracy (%) Kappa coefficient

Multi-SPECTRAL OPTICAL DATA

Fused optical and SAR data

Full-polarimetric SAR data

ALOS/AVNIR-2

ASTER

AVNIR-2/PALSAR

ASTER/PALSAR

ALOS/PALSAR

Accuracy (%)

Accuracy (%)

Accuracy (%)

Accuracy (%)

Accuracy (%)

RADARSAT-2 Accuracy (%)

Producer's

User's

Producer's

User's

Producer's

User's

Producer's

User's

Producer's

User's

Producer's

User's

47.81 29.67 19.16 39.32 0.015

63.16 28.01 12.89

52.53 50.00 15.79 46.18 0.087

71.55 19.23 60.00

58.50 37.46 0.00 44.40 0.006

61.68 23.23 0.00

66.02 16.61 27.26 49.25 0.055

64.05 27.94 22.02

63.64 100.00 50.00 76.92 0.692

94.31 75.00 69.11

97.32 100.00 87.04 96.43 0.929

98.45 96.30 88.68

A. Gaber et al. / Remote Sensing of Environment 162 (2015) 11–28

23

Fig. 8. The Pauli RGB image of El-Gallaba Plain (left) and the supervised classification map (right). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

processed, multi-looked and mosaicked T3 matrix of Radarsat-2 data were exported from NEST software as Generic Binary data with new dimensions and pixel numbers in order to ingest it again in PolSARpro software. Several geographic coordinate points representing each class of Fig. 8 were converted into pixel numbers

using the backward geocoding techniques (Bara et al., 2000) in order to extract the polarization signature of each sedimentary class. These polarization signatures show clearly the differences between the five classes in term of their grain size and surface roughness.

Fig. 9. Shows the relationship between the surface height variation (h), SAR wavelength (λ) and incidence angle (θinc) and the scattering mechanism. Modified after Peake and Oliver (1971).

24

A. Gaber et al. / Remote Sensing of Environment 162 (2015) 11–28

Table 5 Scattering response of Radarsat-2 of the surface sediments in El-Gallaba Plain. Sedimentary classes

Gravel (1) Gravel (2) Gravel/Sand (1) Gravel/Sand (2) Sand

SAR scattering mechanism (Mean)

Elevations information (m), a.s.l.

Alpha (α°) (±0.7)

Entropy (H) (±0.04)

Anisotropy (A) (±0.06)

Min.

Max.

Mean (±0.9)

12.66 13.88 10.17 12.41 10.01

0.35 0.37 0.28 0.32 0.27

0.29 0.26 0.34 0.28 0.43

112 113 112 107 112

183 175 173 174 167

150.57 148.45 148.46 148.03 146.62

The co-polarized (HH and VV) signatures of each sedimentary class were extracted (Fig. 10). The pedestal height of the co-polarized signature increases together with a change in polarization from VV to HH as one moves from the sand (low random and low power) to the gravel-2 (high random and high power) class, which confirms that the surface roughness of the classified sediments increases as their grain sizes and local topography increase. The interpretation of the polarization signatures reveals the nature of individual sediment classes (dielectric constant and surface roughness) in the supervised classification map

using the full polarimetric Radarsat-2 data in El-Gallaba Plain. Therefore, polarization signatures showed the differences between the five sedimentary classes and validated the labeling of these classes in the supervised map. On the other hand, the unsupervised classification map using the full polarimetric ALOS/PALSAR data revealed a set of linear features that are hidden under the windblown sand (sand class) and are not visible in the field or in the other sets of classification maps (Fig. 4). This is due to the fact that the low frequency L-band SAR data can penetrate deeper than

Fig. 10. The polarization signatures of each sedimentary class.

A. Gaber et al. / Remote Sensing of Environment 162 (2015) 11–28

the C-band and imaging of near-surface targets (Grandjean et al., 2001). These hidden linear features might be a pattern of near-surface structures and it is worthy to validate the existence of these structures. The existence of these hidden lineament features was confirmed in the field by acquiring ground penetrating radar survey using 100 MHz shielded GSSI antenna. The GPR profiles show very clear discontinuities within the first top meter in the subsurface layers which are covered by thin veneer of windblown sands (Fig. 11). This is confirmed that each wavelength of SAR dataset can give different information about the surface sediment distribution.

4.3. Depositional history of the surface sediments in El-Gallaba Plain El-Gallaba Pain has a long history of fluvial processes, and most recently aeolian processes as well. El-Gallaba Plain has been receiving a significant amount of fluvial deposits from different sources and directions. These sources were; (1) the old Gilf System from east to west, (2) the middle Qena System from north to south and (3) the youngest and current Nile System from south to north (Issawi & McCauley, 1992). Moreover, according to Hinz et al. (2003), Thurmond et al. (2004) and Roden, Abdelsalam, Atekwana, El-Qady, and Tarabees (2011) Wadi El-Kubanyia (Fig. 1) was the western continuation of Wadi Abu Subeira (east of the Nile) before the drainage system was fundamentally reorganized in Middle Pleistocene time, reversing its flow direction. While Wadi El-Kubanyia currently flows SE into the Nile, in late-Pliocene and early-Pleistocene time this drainage system may have been flowing into the opposite direction. In addition, four riverine systems may have deposited sand, silt and gravels next to or on top of El-Gallaba Plain making the sequence of depositions more difficult to reconstruct (Issawi & El-Hinnawi, 1980). Based on our field and remote sensing work, the surface sediments in El-Gallaba Plain were well characterized and mapped as five main sedimentary classes. These five classes were investigated and descripted as follows: (1) the gravel-1 class consists of large amounts of flints and igneous rock fragments which probably derived from the Red Sea Mountains in the east (Fig. 12). These sediments were deposited by braided channels systems and plunged 25°, (2) the gravel-2 class consists of poorly sorted calcareous fragments most probably derived

25

from the western calcareous Sin El-Kadab Plateau, (3) the gravel/ sand-1 class consists of reddish gravel layer intercalated with reddish farraginous sandstone with 50 cm average thickness which most probably derived by Wadi El-Kubanyia, (4) the gravel/sand-2 class consists of a mixture of gravel derived from the western calcareous Sin ElKadab Plateau and desert pavement with sand grains, and (5) the windblown sand class derived from the NW–SE direction and masking the old fluvial deposits and the desert pavements as well (Fig. 12). Gravel2 class seems to be the oldest class because it has been dissected by other classes and it might be related to the old north–south Qena fluvial system. Gravel-1 and gravel/sand-1 classes are most probably related to the Wadi E-Kubanyia fluvial system which came from the Red Sea Mountain in the east. Moreover, the geo-structural setting of the western Kom Umbo basin where the El-Gallaba Plain lies is not well known. Recently, this area has drawn the attention of the oil and gas industry with the discovery of oil-producing wells at the Al Baraka Oil Field. Thus for now, it is very difficult to confirm or assign these surface sediments to any of previous mentioned fluvial systems, unless we fully understand the geo-structural setting of El-Gallaba Plain and have core samples which represent each sedimentary class and measure their relative age using age dating technique. Although, such detailed geo-structural and age dating work is not dealt with in this article, we will investigate it in more detail in the future. In addition, there are many in situ and local deposits of clay, mud and shale deposits along El-Gallaba Plain. These sediments might be deposited within small and stagnant fresh water pools between the old fluvial channels and on the flood plains of meandering system as well. These shale sediments can be used for soil enhancement in agriculture activities in El-Gallaba Plain. Such long history of fluvial processes has increased the potentiality of ElGallaba Plain to be a potential site for agricultural activity using its groundwater resources.

5. Conclusion El-Gallaba Pain in the Western Desert of Egypt has a long history of fluvial processes and most recently with aeolian processes as well, which makes the sequence of deposition difficult to reconstruct. Several

Fig. 11. GPR profiles (4 & 5) show the subsurface layers discontinuities along the dashed lines.

26

A. Gaber et al. / Remote Sensing of Environment 162 (2015) 11–28

Fig. 12. Shows the depositional directions of the surface sediments in El-Gallaba Plain.

optical and radar datasets were investigated and used for mapping the surface sediment in El-Gallaba Plain in terms of grain size and surface roughness. The best overall accuracy method and dataset was the supervised classification using the full polarimetric Radarsat-2. In addition, the polarimetric response (entropy, anisotropy and alpha), polarization signature and mean elevation information of each sedimentary class were extracted and investigated to distinguish the differences between them in terms of differences in their surface roughness and grain sizes. Moreover the full polarimetric ALOS/PALSAR data revealed a set of parallel linear features hidden below the windblown sand. The existence of these hidden linear features was confirmed by acquiring (GPR) survey using 100 MHz shielded antenna. Finally the depositional history of

these surface sediments has been proposed based on the field and remote sensing works. The El-Gallaba Plain has a potential for agriculture activity using its groundwater storage. Acknowledgments The authors would like to thank the Japan Aerospace Exploration Agency (JAXA) for providing the ALOS data as part of the ALOS user agreements (ALOS-RA-81&ALOS-2 RA-1389), the USGS Earth Resources Observation and Science (EROS) for providing the ASTER data, and the Canadian Space Agency (CSA) and MDA for providing the RADARSAT2 data as part of the SOAR-EI project no. 5138. Also, we would like to

A. Gaber et al. / Remote Sensing of Environment 162 (2015) 11–28

thank Professor Mohamed Helmi Greish of the Suez Canal University for joining us and helping in the field work. This work is being funded by the U.S.–Egypt Science and Technology Joint Fund in cooperation with NSF and STDF under Project Award # 1004283 and # 1975, respectively.

References Abd El-Razik, T., & Razavaliaev, A. (1972). On the tectonic origin of the Nile Valley between Idfu and Qena. Egypt Journal of Geology, 16(2), 235–245. Abrahams, A.D., & Parsons, A.J. (Eds.). (1994). Geomorphology of desert environments. London: Chapman Hall. Baghdadi, N., Gaultier, S., & King, C. (2002). Retrieving surface roughness and soil moisture from synthetic aperture radar (SAR) data using neural networks. Canadian Journal of Remote Sensing, 28, 701–711. Baghdadi, N., Holah, N., & Zribi, M. (2006). Soil moisture estimation using multi-incidence and multi-polarization ASAR data. International Journal of Remote Sensing, 27, 1907–1920. Bara, M., Broquetas, A., Scheiber, R., & Horn, R. (2000). Geocoding techniques for interferometric and polarimetric airborne SAR data. Proceedings of SPIE, 4173, 279–290. Bartholdy, J., & Folving, S. (1986). Sediment classification and surface type mapping in the Danish Wadden Sea by remote sensing. Netherlands Journal of Sea Research, 20, 337–345. Boerner, W.M. (2007). Introduction to synthetic aperture radar (SAR) polarimetry. New York: Wexford College (ch. 1). Boerner, W., Mott, H., Luneburg, E., Livingstone, C., Brisco, B., Brown, R., & Paterson, S. (1998a). Polarimetry in radar remote sensing: Basic and applied concepts. In F. Henderson, & A. Lewis (Eds.), (3rd Edition ). Manual of Remote Sensing, Vol. 2. (pp. 271–357). Boerner, W.M., Mott, H., Luneburg, E., Livingstone, C., Brisco, B., Brown, R.J., & Paterson, S. (1998b). Polarimetry in radar remote sensing: Basic and applied concepts. In F.M. Henderson, & A.J. Lewis (Eds.), (3rd Edition ). Manual of remote sensing, Vol. 2. (pp. 271–357). Brunzell, H. (1997). Detection of shallowly buried objects using impulse radar. IEEE Transactions on Geoscience and Remote Sensing, 37(2), 875–886. Cambell, J.B. (2002). Introduction to remote sensing (3rd ed.). New York: The Guilford Press. Cloude, S.R., & Pottier, E. (1995). A review of target decomposition theorems in radar polarimetry. IEEE Transactions on Geoscience and Remote Sensing, 34(2), 498–518. Cloude, S.R., & Pottier, E. (1997). An entropy based classification scheme for land applications of polarimetric SAR image data. IEEE Transactions on Geoscience and Remote Sensing, 35, 68–78. Cloude, S.R., Pottier, E., & Boerner, W. -M. (2002). Unsupervised image classification using the entropy/alpha/anisotropy method in radar polarimetry. Proc. JPL AIRSAR Symposium, Pasadena, California. Cohen, J. (1960). A coefficient of agreement for nominal scales. Educational and Psychological Measurement, 20, 37–46. Cohen, J. (1968). Weighted kappa: Nominal scale agreement with provision for scaled disagreement or partial credit. Psychological Bulletin, 70, 426–443. Congalton, R.G. (1991). A review of assessing the accuracy of classifications of remotely sensed data. Remote Sensing of Environment, 37, 35–46. Congalton, R.G., & Kass, G. (1999). Assessing of remotely sensed data: Principles and practices. Boca Rotan, FL: Lewis Publishers, 34–46. Daniels, D.J. (1996). Surface penetrating radar. Institution of Electrical Engineers Radar Series No. 6London: ERA Technology. Dierking, W. (1999). Quantitative roughness characterization of geological surfaces and implications for radar signature analysis. IEEE Transactions on Geoscience and Remote Sensing, 37(5), 2397–2412. Dobson, M.C., Pierce, L.E., & Ulaby, F.T. (1997). The role of frequency and polarization in terrain classification using SAR data. IEEE Transactions on Geoscience and Remote Sensing, 35(4), 1621–1623. Dong, Y., Milne, A.K., & Forster, B.C. (2001). Segmentation and classification of vegetated areas using polarimetric SAR image data. IEEE Transactions on Geoscience and Remote Sensing, 39(2), 321–329. EGSMA, NARSS, UNDP, & UNESCO (2005). “Geological map of Aswan, Quadrangle, Egypt”. Sheet NG-36 B with scale 1:250 000. Elachi, C., Kuga, Y., McDonald, K.C., Sarabandi, K., Senior, T.B.A., Ulaby, F.T., van Zyl, J.J., Whitt, M.W., & Zebker, H.A. (1990). Radar polarimetry for geoscience applications. Artech House, Inc. ESA (2008). PolSARpro_Data Format. http://earth.eo.esa.int/polsarpro/Manuals/ Evans, D.L., Farr, T.G., Ford, J.P., Thompson, T.W., & Werner, C.L. (1986). Multi-polarization radar images for geologic mapping and vegetation discrimination. IEEE Transactions on Geoscience and Remote Sensing, GE-24(2), 246–257. Evans, D.L., Farr, T.G., & van Zyl, J.J. (1992). Estimates of surface roughnessderived from synthetic aperture radar (SAR) data. IEEE Transactions on Geoscience and Remote Sensing, 30, 382–389. Evans, D.L., Farr, T.G., van Zyl, J.J., & Zebker, H.A. (1988). Radar polarimetry: Analysis tools and applications. IEEE Transactions on Geoscience and Remote Sensing, 26, 774–789. Fitzpatrick-Lins, K. (1981). Comparison of sampling procedures and data analysis for a land-use and land-cover map. Photogrammetric Engineering & Remote Sensing, 47, 343–351. Gaber, A., Koch, M., & El-Baz, F. (2010). Textural and compositional characterization of Wadi Feiran Deposits, Sinai Peninsula, Egypt, using Radarsat-1, PALSAR, SRTM and ETM+ data. Journal of Remote Sensing, 2, 52–75.

27

Gaber, A., Koch, M., Helmi, M., & Sato, M. (2011). SAR remote sensing of buried faults: Implications for groundwater exploration in the Western Desert of Egypt. Sensing and Imaging: An International Journal, 12(¾), 133–151. Gaber, A., Koch, M., Helmi, M., Sato, M., & El-Baz, F. (2013). Near-surface imaging of a buried foundation in the Western Desert, Egypt, using space-borne and ground penetrating radar. Journal of Archaeological Science, 40(4), 1946–1955. Gaddis, L.R., Mouginis-Mark, P.J., & Hayashi, J.N. (1990). Lava flowsurface textures: SIR-B radar image texture, field observations, and terrain measurements. Photogrammetric Engineering & Remote Sensing, 56, 211–224. Gerstl, S.A. (1990). Physics concepts of optical and radar reflectance signatures. International Journal of Remote Sensing, 11(7), 1109–1117. Glennie, K.W. (1970). Desert sedimentary environments. Amsterdam: Elsevier. Grandjean, G., Paillou, P., Dubois, P., August-Bernex, T., Baghdadi, N., & Achache, J. (2001). Subsurface structures detection by combining L-band polarimetric SAR and GPR data: Example of the Pyla dune, France. IEEE Transactions on Geoscience and Remote Sensing, 39, 1245–1258. Gray, A.L., Vachon, P.W., Livingstone, C.E., & Lukowski, T.I. (1990). Synthetic aperture radar calibration using reference reflectors. IEEE Transactions on Geoscience and Remote Sensing, 28(3), 374–383. Hinz, E.A., Stern, R.J., Thurmond, A.K., Abdelsalam, M.G., & Abdeen, M.M. (2003). When did the Nile begin? Remote sensing analysis of paleo-drainages near KomOmbo, Upper Egypt. American Geophysical Union, Fall Meeting, Abstract #H52A-1182, San Francisco, CA, USA. Hudson, W., & Ramm, C. (1987). Correct formula of the Kappa coefficient of agreement. Photogrammetric Engineering & Remote Sensing, 53, 421–422. Hugenholtz, C., & Van der Sanden, J. (2001). Polarimetric SAR for geomorphic mapping in the intertidal zone, Minas Basin, Bay of Fundy, Nova Scotia. 24 (Preprint Canada Centre for Remote Sensing). Issawi, B., & El-Hinnawi, M. (1980). Contribution to the geology of the plain west of the Nile between Aswan and Kom Ombo. In Loaves, & Fishes (Eds.), The prehistory of Wadi Kubbaniya. Department of Anthropology South Methodists University. Issawi, B., & McCauley, J.F. (1992). The Cenozoic rivers of Egypt: The Nile problem. In B. Adams, & R. Friedman (Eds.), The followers of Horus: Studies dedicated to Michael Allen Hoffman 1944–1990. Egyptian Studies Association Publication, 2. (pp. 121–138). Oxford: Oxbow Press. James, A.S., & Daniel, B.C. (2002). A comparison of classification methods for large imagery data sets. JSM 2002 Statistics in an ERA of Technological Change—Statistical computing section, New York City, 3205–3207. Jansen, J.M., Bagnoli, M., & Focacci, M. (2008). Analysis of land-cover/use change dynamic in Manica Province in Mozambique in a period of transition (1990–2004). Forest Ecology and Management, 254, 308–326. Koch, M., & El-Baz, F. (2000). Spectral and textural classification of active wadi systems in arid lands by Landsat TM and Radarsat data. CDROM-Proceedings of the ASPRS Annual Conference, Washington, DC, USA (pp. 10). Koch, M., Gaber, A., Burkholder, B., & Mohamed, H.G. (2012). Development of new water resources in Egypt with earth observation data: Opportunities and challenges. International Journal of Environment and Sustainability, 1(3), 1–11 (ISSN 1927‐9566). Koch, M., Gaber, A., Mohamed, H.G., El-Sayed, Z., Sayed, M.A., & Mostafa, A. (2013). Multisensor characterization of subsurface structures in a Desert Plain Area in Egypt with implications for groundwater exploration. (in proceeding of SPIE, Germany, V. 8887 888712-2). Laur, H., Bally, P., Meadows, P., Sanchez, J., Schaettler, B., Lopinto, E., & Esteban, D. (2002). “ERS SAR calibration derivation of the backscattering coefficient σ0 in ESA ERS SAR PRI products”, Document ES-TN-RS-PM-HL09. URL http://earth.esa.int/ESC2/.ESA Lee, J.S. (1981). Refined filtering of image noise using local statistics. Computer Graphics and Image Processing, 15(4), 380–389. Lee, J.S., Grunes, M.R., Ainsworth, T.L., Li-Jen, Du., Dale, L.S., & Cloude, S.R. (1999). Unsupervised classification using polarimetric decomposition and the complex Wishart classifier. IEEE Transactions on Geoscience and Remote Sensing, 37(5), 2249–2258. Lillesand, T.M., & Kiefer, R.W. (1994). Remote sensing and image interpretation (4th ed.). Hoboken, NJ: Wiley, 568–578. Mather, P.M., Brandt, T., & Koch, M. (1998). An evaluation of Landsat TM spectral data and SAR-derived textural information for lithological discrimination in the Red Sea Hills, Sudan. International Journal of Remote Sensing, 19, 587–604. Mattia, F., & Le Toan, T. (1999). Backscattering properties of multi-scale rough surfaces. Journal of Electromagnetic Waves and Applications, 13, 493–527. McCauley, J.F., Schaber, G.G., Breed, C.S., Grolier, M.J., Haynes, C.V., & Issawi, B. (1982). Subsurface valleys and geoarchaelology of Egypt and Sudan revealed by radar. Science, 218, 1004–1020. Melsheimer, C., Tanck, G., Gade, M., & Aplers, W. (1999). Imaging of tidal flats by multifrequency/multi-polarization synthetic aperture radar. In G.J.A. Nieuwenhuis, R.A. Vaughan, & M. Molenaar (Eds.), Operational remote sensing for sustainable development (pp. 189–192). Rotterdam: Balkema. Moran, M.S., Hymer, D.C., Qi, J.G., & Sano, E.E. (2000). Soil moisture evaluation using multi-temporal synthetic aperture radar (SAR) in semiarid rangeland. Agricultural and Forest Meteorology, 105, 69–80. Moran, M.S., Peters-Lidard, C.D., Watts, J.M., & McElroy, S. (2004). Estimating soil moisture at the watershed scale with satellite-based radar and land surface models. Canadian Journal of Remote Sensing, 30, 805–826. Olhoeft, G.R. (2000). Maximizing the information return from ground penetrating radar. Journal of Applied Geophysics, 43, 175–187. Papathanassiou, K.P., & Buchroithner, M.F. (1993). Signature analysis of multifrequency polarimetric NASA DC-8 Airsar data of alpine geo-applications. EARSeL Advances in Remote Sensing, 2(1-I), 287–299. Paul, M. (1991). Computer processing of remotely sensed images: An introduction. UK: Biddley Limited Publication, 352.

28

A. Gaber et al. / Remote Sensing of Environment 162 (2015) 11–28

Peake, W.H., & Oliver, T.L. (1971). The response of terrestrial surfaces at microwaves frequencies. Technical Report 2770-7. Columbus, OH, USA: Ohio State University. Ray, T.W., Farr, T.G., & van Zyl, J.J. (1992). Polarization signatures for abandoned agricultural fields in the Manix Basin area of the Mojave Desert: Can polarimetric SAR detect desertification? Proceedings of the International Geoscience and Remote Sensing Symposium (IGARSS'92) (pp. 947–949). Houston: Geoscience & Remote Sensing Society. Roden, J., Abdelsalam, M.G., Atekwana, E., El-Qady, & Tarabees, E.A. (2011). Structural influence on the evolution of the pre-Eonile drainage system of southern Egypt: Insights from magnetotelluric and gravity data. Journal of African Earth Sciences, 61, 358–368. Rosenfield, G., & Fitzpatrick-Lins, K. (1986). A coefficient of agreement as a measure of thematic classification accuracy. Photogrammetric Engineering & Remote Sensing, 52, 223–227. Said, R. (1962). The geology of Egypt. Amsterdam, New York: Elsevier, 377. Saraf, A.K. (1999). IRS-IC-LISS-III and PAN data fusion: An approach to improve remote sensing based mapping techniques. International Journal of Remote Sensing, 20(10), 1929–1934. Sarah, N., Douglas, J., Amine, M., Jason, D., & Steve, S. (2011). Assessing Radarsat-2 polarimetric SAR for mapping Shoreline Cleanup and Assessment Technique (SCAT) classes in the Canadian Arctic. 32nd Canadian Symposium on Remote Sensing (pp. 13–16). Schaber, G.G., McCauley, J.F., Breed, C.S., & Olhoeft, G.R. (1986). Shuttle imaging radar: Physical controls on signal penetration and subsurface scattering in the eastern Sahara. IEEE Transactions on Geoscience and Remote Sensing, GE-24(4), 603–623. Schistad, A., Jain, A., & Taxt, T. (1994). Multisource classification of remotely sensed data: fusion of Landsat TM and SAR images. IEEE Transactions on Geoscience and Remote Sensing, 32(4), 768–778. Shimada, M. (2005). Long-term stability of L-band normalized radar cross section of Amazon rainforest using the JERS-1 SAR. Canadian Journal of Remote Sensing, 31(1), 132–137. Shimada, M., Isoguchi, O., Tadono, T., & Isono, K. (2009). PALSAR radiometric and geometric calibration. IEEE Transactions on Geoscience and Remote Sensing, 47(12), 3915–3932. Skidmore, A.K. (1999). Accuracy assessment of spatial information. In A. Stein, F. van der Meer, & B.G.F. Gorte (Eds.), Spatial statistics for remote sensing (pp. 197–209). Dordrecht: Kluwer Academic. Story, M., & Congalton, R.G. (1986). Accuracy assessment: A user's perspective. Photogrammetric Engineering & Remote Sensing, 52, 397–399. Tanck, G., Alpers, W., & Gade, M. (1999). Determination of surface roughness parameters of tidal flats from SIR-C/X-SAR 3-frequency SAR data. IGARSS'99, 1677–1679.

Thurmond, A.K., Stern, R.J., Abdelsalam, M.G., Nielsen, K.C., Abdeen, M.M., & Hinz, E. (2004). The Nubian Swell. Journal of African Earth Sciences, 39, 401–407. Tomas, I.L. (1980). Spatial postprocessing of spectrally classified Landsat data. Photogrammetric Engineering & Remote Sensing, 46, 1201–1206. Townshend, J.R.G. (1986). The enhancement of computer classification by logical smoothing. Photogrammetric Engineering & Remote Sensing, 52, 213–221. Ulaby, F.T., & Elachi, C. (1990). Radar polarimetry for geoscience applications. Norwood, Massachusetts: Artech House Inc. van Genderen, J.L., & Lock, B.F. (1977). Testing land use map accuracy. Photogrammetric Engineering & Remote Sensing, 43, 1135–1137. van Zyl, J.J. (1989). Unsupervised classification of scattering behavior using radar polarimetry data. IEEE Transactions on Geoscience and Remote Sensing, 27(1), 36–45. van Zyl, J.J., & Zebker, H.A. (1990). Imaging radar polarimetry. In J.A. Kong (Ed.), Progress in Electromagnetics Research. Pier, 3. (pp. 277–370). van Zyl, J.J., Zebker, H.A., & Elachi, C. (1987). Imaging radar polarization signatures: Theory and observation. Radio Science, 22, 529–543. Varma, H. (2003). Confusion in data fusion. International Journal of Remote Sensing, 24(4), 627–636. Vrabel, J. (1996). Multispectral imagery band sharpening study. Photogrammetric Engineering & Remote Sensing, 62, 1075–1083. Walter, V. (2004). Object-based classification of remote sensing data for change detection. ISPRS Journal of Photogrammetry and Remote Sensing, 58, 225–238. Welch, R., & Ahlers, W. (1987). Merging multi-resolution SPOT HRV and Landsat TM data. Photogrammetric Engineering & Remote Sensing, 53, 301–303. Yang, W., Cauneau, F., Paris, J. -P., & Ranchin, T. (2000). Fusion of SAR and SPOT images for the detection ofgeological features over the Three Gorges Dam site, China. In Thierry Ranchin, & Lucien Wald (Eds.), Proceedings of the Third Conference “Fusion of Earth data: Merging Point Measurements, Raster Maps and Remotely Sensed Images”, Sophia Antipolis, France (pp. 137–142). Nice, France: SEE/URISCA. Youssef, M.I. (1968). Structural pattern of Egypt and its interpretation. American Association of Petroleum Geologists Bulletin, 52(4), 601–614. Zebker, H.A., & van Zyl, J.J. (1991). Imaging radar polarimetry: A review. Proceeding of the IEEE, 79(11), 1583–1606. Zebker, H.A., van Zyl, J.J., & Held, D.N. (1987). Imaging radar polarimetry from wave synthesis. Journal of Geophysical Research, 92(B1), 683–701. Zhang, Fengli, Maosong, Xu., Xie, Chou, Xia, Zhongsheng, Li, Kun, AiminCai, Yun Shao, Wang, Xuejun, & Touzi, Ridha (2011). Polarimetric signature and the temporal variation analysis for deforestation mapping in Southwest China. PIERS, 7(7), 613–616.