*For correspondence. (e-mail: manab@sac.isro.gov.in)
Initial results using RISAT-1 C-band SAR data
Manab Chakraborty
1,*, Sushma Panigrahy
2, A. S. Rajawat
1, Raj Kumar
1, T. V. R. Murthy
1, Dipanwita Haldar
1, Abhisek Chakraborty
1, Tanumi Kumar
1, Sneha Rode
1, Hrishikesh Kumar
1, Manik Mahapatra
1and Sanchayita Kundu
11Space Applications Centre, Indian Space Research Organisation, Ahmedabad 380 015, India
2Formerly with Space Applications Centre, Indian Space Research Organisation, Ahmedabad 380 015, India
Imaging radars provide information that is funda- mentally different from sensors that operate in the visible and infrared portions of the electromagnetic spectrum. The Indian Space Research Organisation (ISRO) has launched a multi-mode, multi-polarization Synthetic Aperture Radar (SAR) on-board Radar Im- aging SATellite-1 (RISAT-1) on 26 April 2012. Vari- ous data products from RISAT-1 SAR are now going through calibration–validation (cal–val) phase and soon will be available for the global users for opera- tional and research purposes. In this regard, algo- rithms are being developed to retrieve various parameters in diverse application areas. This article deals with the in-house algorithm development for studying different resources using initial available data of RISAT-1.
Keywords: Mangrove ecosystem, ocean surface wind, rice acreage, RISAT-1, ship detection, Synthetic Aperture Radar, wave spectra.
Introduction
RICE, a major staple food crop in India, is grown mostly during the kharif season. Extensive cloud coverage hin- ders optical sensors to sense the crop growth, particularly during monsoon season. Thus SAR data remain the only viable option. Previous studies have demonstrated that the signature of rice is unique and dynamic when moni- tored using SAR data due to the presence of water back- ground in most parts of the country throughout majority of its growth phase. Backscatter shows a steady increase with time since transplanting until the heading stage and thereafter it maintains a constant value1. Results using multi-temporal ERS/RADARSAT imagery have con- firmed that C-HH backscatter can detect differences in crop type, crop growth stage and crop indicators. Cross- polarized radar returns (HV or VH) result from multiple reflections within the vegetation volume and thus can add an extra dimension to the existing single polarization mode. RISAT-1, the country’s first indigenously devel- oped SAR satellite, has come up with huge potential to monitor rice acreage. In the present study rice crop iden-
tification and classification was attempted using RISAT-1 C-band two-date data with central incidence angle 37°, HH/HV polarization. During initiation of the national kharif rice monitoring and acreage estimation project, rice signature has been developed from multi-temporal SAR dataset by Panigrahy et al.2. The model is suitably used to study the rice crop of recently launched RISAT-1.
The present study is an attempt to monitor the rice- growing areas with the initial datasets (MRS, Medium Resolution SAR) from this satellite, which will be the SAR workhorse in near future.
Various oceanographic applications have shown the potentiality of SAR images during the past three decades with the successful launch of a range of spaceborne SARs (SEASAT in 1978, ERS-1 in 1995, ERS-2 in 1995, RADARSAT-1 in 1995, ENVironmental SATellite (ENVISAT) in 2002, etc.). The ocean features commonly seen on SAR imagery include surface waves, mesoscale ocean circulation structures such as eddies and currents, oil slicks, ships and wakes, internal waves and coastal bathymetry. The SAR is so sensitive to the interaction of wind with the ocean surface that, in addition to wind speed, patterns and structures within the atmospheric boundary layer produce identifiable surface imprints3. An algorithm has been developed for the retrieval of very high resolution ocean surface winds, ocean wave spectra as well as detection of coastal and deep sea ships using RISAT-1 SAR data.
Sensitivity of radar to water, due to its high dielectric constant, is extremely valuable to the remote sensing of wetlands. It is not only sensitive to soil moisture, but can also differentiate between moist soil and standing water4. High sensitivity to standing water and soil moisture makes radar an efficient tool for determining hydro pat- tern5,6. SAR technology with improved spatial resolution allows regional wetland mapping. In a study by Baghdadi et al.7, it was suggested that C-band data are useful in wetland mapping and monitoring. Studies conducted with Shuttle Imaging Radar (SIR-C) and Japanese Earth Resources Satellite (JERS)-1 L–HH band imagery con- firmed this finding8,9. Slatton10 had shown that polarimet- ric multiband SAR has potential for mapping the major sub-environments associated with coastal herbaceous wetlands. Discrimination of different types of marshes, swamp thickets and swamp forests is also possible with
radar imagery due to varying stem height and density11. The use of multi-temporal radar data combined via an intensity, hue and saturation (IHS) transformation has also been found to improve wetland mapping12. The opportunity to explore the potential of radar data will improve as many SAR satellites are currently being launched or will soon be launched. These extra satellites will not only increase the amount of data available for analysis, but will also increase collection frequency and the variety of polarizations and frequencies available13. Forest inventory and monitoring have been routinely done by satellite remote sensing techniques. While opti- cal remote sensing is widely used for this purpose, it is often affected by sun angle and atmospheric parameters (e.g. clouds, haze and aerosols) and hinders quantitative modelling. In this context, the active radar remote sens- ing, having particular significance for all-weather, day–
night imaging capability and responsiveness to moisture content and structural attributes of vegetation, holds promise for quantitative analysis in studies on forestry.
One of the unique potentials demonstrated by space- borne radar images is their penetration capability through shallow sand cover in arid regions and detecting subsur- face geological and archaeological features, in particular buried river channels and imprints of archaeological sites occurring in the adjoining areas. Images acquired in northeastern Sahara in 1981 during the first NASA Shut- tle Image Radar mission (SIR-A) demonstrated the capa- bility of the L-band (wavelength 24.5 cm) to penetrate 1–2 m of loose sand and return information about geo- logic and geomorphologic features covered by sand14,15. Subsequently, radar images in L-band by Seasat, SIR-A and SIR-B acquired in the Baden-Jaran Desert of China16, in Saudi Arabia17,18 and in the Mojave Desert of Califor- nia19,20 confirmed the benefit of radar images to study bedrock features beneath few metres of loose sand.
Analysis of Seasat, SIR-A and SIR-B images revealed igneous features (dykes) buried as much as 2 m beneath alluvium in the Mojave Desert19. The radar energy has penetrated through the shallow sand cover and subsurface igneous dykes have acted as subsurface rough surfaces and provided high backscatter to be detected by the radar sensor. Ancient drainage pattern cut in the bedrock is not visible in the optical images because of the sand cover.
The buried palaeodrainage network appears as dark fea- tures due to smoother channel fillings and is identified due to contrasting brightness of hard substrate of the adjoining region14,15,21–24
.
Kharif rice acreage using two-date RISAT-1 data Dataset and study area
RISAT-1 C-band (5.35 GHz) dual-polarization (HH, HV) MRS L2 data (18 m pixel spacing and 37° incidence angle)
are used for estimating kharif rice acreage estimation.
Data have been collected for two different dates suffi- ciently spaced in phenology to identify rice fields. Spacing between the data is minimum of 12 days for the Odisha coastal area and maximum 50 days for Allahabad. Detailed information about the dataset used is given in Table 1.
Rice is predominantly grown in the entire Indo- Gangetic plain during monsoon season. Uttar Pradesh, Bihar, Odisha and West Bengal are considered as the rice bowl of the country. Two-date scenes having overlap in major rice-growing regions of these states are selected as the test area (Figure 1).
Methodology
Processing of the RISAT-1 data was carried out using PCI Geomatica (ver. 9.0) software. L2 product is avail- able in .tiff format. HH and HV images are imported to native raster format of PCI Geomatica (.pix) and trans- ferred to a single file as two separate channels. Reprojec- tion (if needed) is done and then the image is filtered by enhanced Lee adaptive filter with kernel size 5 × 5, as established by standard procedures1. Most of the scenes had abrupt brightness variations in the near and far range side and banding at equal interval along the track due to mosaicing of data of beams used to form the MRS data.
These unwanted patterns were removed by marking a transect across the image and adjusting the brightness by plotting the variations in brightness across the pixels. The coefficient and central pixel values are used to correct this pattern. Calibration for HH and HV polarization am- plitude image were carried out using the calibration con- stant in product.xml as:
Value in dB (in 32-bit real channel)
= 20 × log10 (DN) – calibration constant.
The two-date images were co-registered by fitting manual GCPs with about 25 well-distributed points throughout the image scene. A second-order polynomial model was fitted to register the second date image with respect to the first date image.
On this two-date HH–HV image, the vectors in the form of sample-segment or district boundary were over- laid. Classification of the rice area is performed inside these segments to derive the crop proportion and finally district rice acreage. Addition of HV channel takes care of the fallow and river border region otherwise classified as rice using two-date HH data.
Observation
Rice, being a semi-aquatic crop, generates unique tempo- ral backscatter profile unlike other crops. During trans- plantation, backscatter is quite low due to puddled fields in standing water condition which reflects more energy in
Table 1. Details of RISAT-1 scenes used for the study
Incidence Temporal Pixel
Area Date Polarization angle interval (days) spacing
Allahabad, Uttar Pradesh 17 July 2012 HH 35.09 50 18 × 18 5 September 2012 HV 36.58
Baleswar, Odisha 2 August 2012 HH 35.04 12 18 × 18
14 August 2012 HV 37.49
Aurangabad, Bihar 18 July 2012 HH 34.30 38 18 × 18
26 August 2012 HV 36.56
Figure 1. Area overlapped by two-date RISAT-1 scenes.
forward direction and minimal in the backscattering direction. Only HH data-based assessment for rice field also includes waterbodies and moist fallow lands as commission error. The overestimated area can be removed by introducing HV data with previous set. In many cases, it has been noticed that signature generated from HH backscattered coefficient (in dB) of rice fields has been mixed with waterbody or moist fallow land in the border- ing fields, whereas HV backscattering for these features differs. HH–HV difference is higher for cropped fields compared to fallow land or waterbodies. Two-date data- set in Allahabad, Aurangabad and Odisha areas can suc- cessfully identify early and late transplanted rice crop with combination of HH and HV data.
In two-date FCC, early transplanted rice fields appear as bright green patches. During first-date data acquisition, as transplantation had taken place, rice fields had suffi- ciently low amount of backscattering which had subse- quently increased in second-date acquisition. For late rice transplanted fields, there is a sudden dip in backscattered coefficient in the second date. So, energy available in green and blue channel is less and it appears as bright red patches in the images (Figure 2).
Figure 2. Two-date RISAT image covering a part of rice-growing area of Aurangabad district, Bihar. a, Overview of two-date overlapped area in FCC (R: First date HH, G: Second date HH, B: Second date HV). b, Rice sample segments. c, Classified rice as early (yellow) and late (magenta) transplanted.
Sample segments are used to find the crop proportion of the study region. Rice is classified within the agricul- tural sample segments. Crop proportion is calculated by taking ratio of total number of pixels and total number of pixels classified as rice. Crop acreage is estimated by multiplying this value with total population of each dis- trict. The procedure can be summarized as follows.
Crop proportion = Pixel classified as rice within sample segments/Total number of pixels within sample segments.
Acreage = Crop proportion × N × Segment area in suitable units.
where N is the total population of agricultural segments.
Crop proportion for early transplanted rice in Auran- gabad area is recorded as 0.094. On the other hand, it is
0.30 for late transplanted rice. Misclassification due to moist soil and waterbodies in two-date single polarization data has been solved by adding HV polarization in both dates. HV is more sensitive to above-ground volume and hence backscattering differs. Due to late onset of mon- soon, most of the area has transplanted rice lately, which clearly showed up in the second-date data.
Dataset for coastal Odisha comprises one ascending pass and another descending with 10 days temporal inter- val. Though the interval is too small to identify early and late transplanted rice, the SAR data successfully picked up rice fields because majority of transplanting opera- tions coincided with these dates. Crop ratio for this area is estimated as 0.10 and 0.31 for early and late trans- planted rice respectively. The FCC composed by two-date HH polarization shows a similar picture of the study area (Figure 4).
In the Allahabad scene (Figure 3), analysis shows that most of the rice-growing areas have early transplanted rice, only few fields have sown it late. Crop proportion of early transplanted rice is estimated as 0.30 and for late transplanted rice it is 0.08. Two-date HH dataset was also analysed and results have been compared. Overestimation of late transplanted rice is clearly identified, which esti- mated crop ratio of late transplanted rice as 0.15. The large data gap (50 days) between the two dates rendered some difficulty in picking up the late transplanted paddy as it was missed in the 10 August scene. The fully-grown crop was picked up in the third date, but critical trans- plantation phase for some parts was missed.
Figure 3. a, classified rice sample segment in Allahabad region. b, classified rice in coastal Odisha.
Figure 4. Comparison between two-date HH FCC of RISAT-1 and Radarsat-2.
The present study is a preliminary attempt to utilize the RISAT-1 data for kharif rice monitoring, which was the largest national project using microwave data from a for- eign satellites till the launch of RISAT-1. The feasibility of RISAT-1 data for monitoring the national rice acreage has been demonstrated in this study. Suitable datasets from RISAT-1 would improve classification of the pre- sent study and also boost the monitoring of other kharif crops in future. The low data cost would be a boon for the microwave remote sensing community globally in all arenas of application.
Oceanographic applications of RISAT-1 SAR
Data
It is well known that for oceanographic applications wide swath (scanSAR)-mode data are best suitable. Hence in-house development of all the retrieval algorithms has been carried out using Wide Swath Medium resolution (Wide Swath Mode, WSM) data from Advanced SAR (ASAR) on-board ENVISAT. The algorithms are tested on Fine Resolution (FRS) and Medium Resolution (MRS)-mode data from RISAT-1 SAR (due to non- availability of wide swath data from RISAT-1 while pre- paring this manuscript). Brief descriptions of the various data used in this study are given below.
(1) ENVISAT–ASAR: ENVISAT was launched by the European Space Agency (ESA) on 1 March 2002 in a Sun-synchronous orbit of altitude 799.8 km with inclina- tion 98.550°, as a successor to ERS-1 and ERS-2. The or- bital period of ENVISAT is 100.59 min with a repeat cycle of 35 days and varying imaging frequency from 1 to 3 days. ASAR is one of the ten payloads carried by ENVISAT. ASAR uses in-phase array with an incidence angle range of 15–45° and is able to operate in five dif- ferent polarization modes (VV, HH, VV/HH, HV/HH, VH/VV) in C-band (5.3 GHz, 5.6 cm). Among the five operating modes of ASAR (Image Mode, Alternating Polarization, WSM, Wave Mode and Global Monitoring), only the WSM, using the scanSAR technique, offers a wide enough swath (around 400 km) with a spatial reso- lution adapted to accurate regional monitoring (with a pixel spacing of 75 m). The incidence angle in each image ranges from 17° to 42°.
(2) RISAT-1 SAR: RISAT-1 was launched on 26 April 2012 by ISRO in a Sun-synchronous dawn–dusk orbit of altitude 536 km with inclination of 97.55° and orbital period of 95.5 min. RISAT-1 carries a C-band (5.35 GHz) SAR as the sole payload. The RISAT-1 SAR is capable of imaging the Earth’s surface in different modes, e.g.
HRS (spotlight scanning, resolution less than 2 m), FRS- 1 (Stripmap scanning, resolution 3 m), FRS-2 (Stripmap Scanning, resolution 6 m, quad-pol capability), MRS
(scanSAR scanning, resolution 25 m) and CRS (scanSAR scanning, resolution 50 m) with variable swath widths ranging from 30 to 240 km.
Methodology
SAR achieves its azimuth resolution by coherent process- ing of radar returns. Due to this coherent processing, noises (speckles) appear on the SAR images. Before the processing of SAR images for any retrieval purposes, such speckles have to be removed. In our methodology, speckle removal is performed using a 5 × 5 gamma MAP filter. In case of coastal SAR images, the land portions of the images are to be masked so as to eliminate land con- tamination in the resultant product. We do this using 30 arcsec global topography (GTOPO30) data available from the US Geological Survey (USGS). At this end the SAR image is ready for retrieval of oceanographic parameters.
Retrieval of ocean surface winds: Over the oceans, the roughness explicitly depends on the winds blowing over the surface. Greater the wind speed, the higher is the sur- face roughness and so is Normalized Radar Cross Section (NRCS)25. The variation in surface roughness causes variation in backscattered power and hence in SAR image intensity, which is directly proportional to the NRCS for a calibrated SAR image26. Both the SAR images from ENVISAT and RISAT-1 are calibrated using the method proposed by Rosich and Meadows27 considering the effect of SAR digital counts, external calibration constant and local incidence angles. The calibrated SAR image is then inverted to obtain the wind speed using a C-band model function (CMOD5) and using auxiliary wind direc- tion information from numerical weather model (National Centre for Environmental Prediction, Global Data Assimilation System, NCEP-GDAS)28. This auxiliary information of wind direction is necessary to invert CMOD5, because single-antenna SAR cannot directly measure wind direction. All the CMODs are developed based on VV polarization SAR images. However, the wind retrievals from HH polarized images are performed using an azimuth and incident angle-dependent para- meterization for the effective polarization ratio as given by Mouche et al.29. Figure 5 shows SAR images from ENVISAT and RISAT-1 and their corresponding retri- eved winds. Figure 6 shows the validation results of retrieved wind speeds from ENVISAT. The validation results in Table 2 show that the accuracy of the output of the algorithm is suited for operational use30.
Retrieval of ocean wave spectra: At typical incidence angle between 20° to 70°, SAR interacts with the ocean surface through Bragg’s resonance31. Within this inci- dence angle range ocean wave spectra are retrieved from SAR images. From the speckle-free SAR intensity
imagery, a single SAR image frame is extracted. In case of RISAT-1 FRS-1 mode, such a frame is comprised of 512 × 512 image pixels. Since each pixel represents a 3 m × 3 m area, the image frame corresponds to 1.5 km ×
Figure 5. Wind retrieval from (a) coastal winds from ENVISAT ASAR (output resolution 975 m) and (b) open ocean winds from RISAT-1 SAR (output resolution 900 m).
Figure 6. Validation of ENVISAT ASAR-derived ocean surface wind speeds with OSCAT, TMI, SSMI and JASON-2. Table 2 shows the respective validation results with OSCAT, TMI, SSMI, Jason-2 and their statistics.
Table 2. Validation results with OSCAT, TMI, SSMI, Jason-2 and their statistics
Data No. of points Bias (m s–1) Standard deviation (m s–1) R
OSCAT/ASAR 629 0.87 1.17 0.94
TMI/ASAR 1641 0.48 1.51 0.89
SSMI/ASAR 634 0.38 0.99 0.93
Jason-2/ASAR 217 0.04 1.33 0.88
(OSCAT + TMI + SSM/I + Jason-2)/ASAR 3121 0.35 1.40 0.91
Figure 7. Retrieval of ocean wave spectra from a subset of RISAT-1 SAR image 952F1_S58_RV.
Figure 8. Ship detection from RISAT-1 SAR image 21343_VV. The adjacent table shows the number of detected ships and their corresponding geo-location for the subset.
1.5 km patch on the ocean surface. The frame size pro- vides a sufficiently large area that at least 10 cycles of very long surface waves, up to 150 m in length, can be included in a single frame. At the same time, the frames are also small enough that the ocean surface can be assumed homogeneous within a frame.
The mean intensity is subtracted from the image frame and the resultant is normalized by dividing by the mean intensity. This image of fractional modulation is then Fourier transformed and squared to produce image inten- sity-variance spectrum as a function of azimuth and range wavenumber. The image is then corrected for stationary response.
Since the resulting image has only two degrees of free- dom, the value of the spectrum at each two-dimensional wavenumber bin is a noisy representation of the underly-
ing spectrum. Hence from the 2D spectrum, 20% of the ensemble average is subtracted after convolving with a Gaussian-shaped kernel to remove the noise32. After this, the 2D spectra are divided by magnitude of modulation transfer function (MTF) that accounts for the modulation of capillary wave by underlying gravity waves33. Figure 7 shows a RISAT-1 FRS-1 image and retrieved wave spectra.
Ship detection: Real-time detection of coastal as well as open ocean ships using SAR images provides a valuable aid for building a space-based surveillance system. From the SAR imagery ships can be detected by means of their intensity contrast relative to the immediate background.
In a SAR image, a 20 × 20 pixels window is selected. If the intensity of a pixel is greater than the threshold (com- puted by the mean and standard deviation of 400 pixels
within a single window and assuming a constant false alarm rate, CFAR) of the window enclosing the particular pixel, the pixel is detected as a ‘potential ship pixel’34. The window is then moved through the whole SAR image searching all the potential ship pixels. When the searching operation is completed, ships are identified based on the ship pixels using a scanning-based ship cluster algorithm.
Figure 8 shows ship detection results from one RISAT-1 image. The algorithm has been validated using informa- tion from Director General, Shipping Corporation, Mum- bai and found to be very sound for operation purposes.
Future scope
This article discusses the algorithm development for retrieval of high-resolution ocean surface winds, ocean wave spectra and ship detection from SAR images for operational utilization. All these algorithms were deve- loped using ENVISAT ASAR data and then implemented on RISAT-1 SAR data. Since RISAT-1 is now commis- sioned for cal–val phase, much of RISAT-1 data could not be used, particularly wide swath data. After the com- pletion of cal–val phase, more data will be available to the users and then extensive validations of these algo- rithms are planned. Also, several new aspects of RISAT- 1 SAR, e.g. circular polarization, etc. can be utilized for further improvement of these algorithms.
Analysis of structural components of Wular Lake – Ramsar Site, India – based on RISAT-1 data
Study area
The study area – Wular Lake, Jammu and Kashmir (J&K) is an internationally important wetland under the Ramsar Convention. The lake, along with the extensive marshes surrounding it, is an important natural habitat for wildlife.
It is also an important habitat for fish, accounting for 60% of the total fish production in J&K. The lake is a source of livelihood for the large human population living along its fringes. Encroachments resulting in the conver- sion of vast catchment areas into agriculture land, pollu- tion from fertilizers and animal wastes, poaching of waterfowl and migratory birds and weed infestation are the main threats to the wetland.
Objectives and data used
The objective is to study application of initial set of SAR (C-band HH/HV) data from RISAT-1 for extraction of structural components of Wular Lake by way of delinea- tion of open water and various vegetation types/densities.
The satellite includes RISAT-1 (C-band HH/HV, MRS with 35.07° incidence angle and 18 m pixel spacing) geo-
referenced data of 12 July 2012. The field data comprises vegetation-related information.
Methodology
The methodology involves a standard approach wherein:
• The data were subjected to speckle removal.
• The GPS-aided ground truth information is converted into a spatial point-layer with the updated attributes of field data.
• Classification of the SAR data after defining the train- ing classes based on ground truth as stored in the spa- tial layer (point layer).
Results
RISAT-1 data comprising C-band HH/HV with 35° inci- dence angle and pixel spacing of 18 m in MRS mode has shown reasonably good geometric fidelity and uniformity in contrast without much speckle (Figure 9).
The Wular Lake has an area of 11,377 ha (Table 3).
The classified image shows that Trapa natans and Trapa bispinosa are dominant in the wetland, occupying an area of 4427 ha. Due to lower density, water beneath Phrag- mites communis could be delineated which accounts for 1206 ha area. Fringes of the wetland are observed to have been planted or have natural vegetation comprising wil- low, of two density classes – medium density (1543 ha) and high density (1767 ha). Paddy fields and fallow are observed on the fringes, which together with willow indi- cate encroachment into the wetland. Details of aerial extents of various classes are given in Table 3.
Preliminary decision rule classification of mangrove ecosystem using single-date VV polarization SAR
Mangrove forests are found in the intertidal zones of tropical and subtropical coastlines, and exist as an eco- system comprising estuaries, lagoons, creeks and
Figure 9. (a) Wular Lake RISAT-1 SAR image (C-band HH/HV) of 12 July 2012 and (b) the corresponding classified image (colour codes with their associated description are given in Table 3).
Table 3. Area under various structural components in Wular Lake Colour
code Class description
Area (ha)
Open water 580
Low density Trapa (Trapa natans and T. bispinosa) 2,249
Medium density Trapa (T. natans and T. bispinosa) 809
High density Trapa (T. natans and T. bispinosa) 1,369
Medium density willow 1,543
High density willow 1,767
Water beneath the emergent vegetation (mainly Phragmites communis) 1,206
Habitation 947
Paddy fields 168
Fallow 739
Total 11,377
Figure 10. LISS III false colour composite (FCC) showing Indian Sundarbans as the dark red zone. Yellow colour depicts the boundaries of the districts. The hollow rectangle overlaid on the FCC shows the study area.
intertidal mudflats35. The mangrove forests consist of spreading trees with numerous arched aerial roots and pneumatophores. The system is sensitive to changes in the local hydrological environment, and the changes are typically manifested through alterations in their spe- cies/community composition, structure and biomass.
Essentially, the backscatter coefficient of a mangrove canopy depends upon the interaction of microwaves with leaves, branches, trunks, above-ground roots/pneumato- phores and the underlying mud/water. Thus, apart from sensor parameters (polarization, frequency and incidence angle), the type and structure of mangrove vegetation af- fect the backscatter36.
Studies have already demonstrated that microwave scattering and attenuation in C-band SAR result from inter- actions with tree canopy and small secondary branches, but C-band backscatter from tree trunks is small due to minimal canopy penetration, resulting in a larger amount of signal absorbed and less signal returned37. A summary
of studies on radar remote sensing has been reviewed in the field of mangroves38. Though most of these studies give immense knowledge in understanding radar interac- tion with mangroves, one rarely finds large area applica- tions leading to operational forestry requirements. Studies regarding the characterization of Indian mangrove eco- systems using SAR are minimal. The objective of the present work is to study the usefulness of single-date VV polarization C-band SAR for broad classification of man- grove ecosystem.
Study area and datum used
The study area is Dhanchi Island (Figure 10), lying between 21°36′N–21°43′N lat. and 88°25′E–88°28′E long. and located in the confluence of the Thakuran River in the east, the Jagdal Ganga in the west and the Bay of Bengal in the south. There are two prominent creeks in the island; one originating from the east and the other from the west.
RISAT-1 Fine Resolution Stripmap Beam (FRS) datum having an incidence angle of 17.84° was utilized. Single- date datum acquired on 10 May 2012 of ascending mode (around 1709 h local time) was used. The datum charac- teristics are provided in Table 4. The acquisition time of the image corresponded with low-tide conditions in the island.
Methodology
The processing of the single-date RISAT-1 datum for this study involved downloading, speckle reduction and cali- bration.
Mangrove ecosystem is composed of mangrove forests, creeks/channels and mudflats. Thus, as a first step, one has to discriminate mangrove forests from these associ- ated classes. Mangroves being evergreen forests, the can- opy undergoes changes during the phenological events of flowering and fruiting39. Thus, one can expect little deviation in backscatter signature during a particular sea- son. However, the intra-class variability within mangrove forests is significant, mainly due to species composition, density, age and gradient from tidal influx.
Single polarization (VV) SAR datum was used and the signatures of the different components of mangrove eco- system of the island were noted. Based on the backscatter signatures and ancillary information, a knowledge base was developed which culminated into decision rules that were used for classification. The training sets were selected using the mangrove ecosystem map of National Wetland Inventory and Assessment, showing mangrove forests, creeks and intertidal mudflats40. Considering the intra-class variability within mangrove forests, care was taken to consider training classes covering these variabili- ties. These categories were grouped into sub-classes. For each class training samples were appropriately selected.
A decision rule algorithm was developed based on 65%
of training class signature and validated on the rest of 35% samples. A mask of the mangrove region was cre- ated using available Sundarban maps. Within this mask different ranges of amplitude backscatter values of sin- gle-date VV polarization were used for rule development.
Using macro language of EASI/PACE, the decision rules were generated to classify the single-date image.
Results and discussion
VV image and corresponding backscatter values
In the image intertidal mudflats appeared dark due to low backscatter and creeks/channels appeared bright due to
Table 4. Datum characteristics
Datum characteristics Specification
Product format RISAT-1-GeoTIFF
SAR band and polarization C band, VV polarization
Beam mode FRS1
Sampled pixel spacing (m) 9 × 9 Number of looks (range × azimuth) 2 × 4 Incidence angle (degree) 17.84
Pass direction Ascending
Map projection UTM
Date of acquisition 10 May 2012 Acquisition time (local) Around 17:09 h (IST)
Figure 11. Mean backscatter values with standard deviation bars (of the population means) of VV polarization of different mangrove eco- system components of Dhanchi Island.
high backscatter (Figure 11). The mangrove forests regis- tered backscatter values ranging from –12 to –7.5 dB.
Preliminary decision rule classification
The mangrove ecosystem of the island could be classified into mangrove forests, intertidal mudflat and creeks/
channels (Figure 12). The mangrove forests could be divided into two broad classes, viz. dense and sparse mangrove based on canopy closure (<40% closure for sparse mangrove and >40% for dense mangrove). Thus, dense mangrove corresponded to very dense and moder- ately dense forests, and sparse mangrove to open forests, scrub and non forest zones of the forest cover classifica- tion scheme provided by Forest Survey of India41. Since low-tide conditions prevailed during the image acquisi- tion time, mudflats could be demarcated along both the eastern and the western creeks and several other minor
Figure 12. Preliminary decision rule classification of mangrove eco- system applied on the mask of mangrove region of Dhanchi Island on VV image.
Figure 13. Proportion of different classes of mangrove ecosystem of Dhanchi Island.
Figure 14. Ground photographs of different classes of mangrove ecosystem of Dhanchi Island.
Figure 15. Parts of the palaeochannel of the lost Sarasvati around Anupgarh, northern Rajasthan as seen on (a) RISAT SAR acquired on 12 July 2012; (b) Landsat ETM FCC and (c) FCC of RISAT SAR and Landsat ETM merged product.
creeks and channels. Mudflats could be also discrimi- nated along the eastern coast of the island. The proportion of the different mangrove ecosystem components in the
island was of the order dense mangrove > sparse man- grove > intertidal mudflats > creeks/channels (Figure 13).
Major portions of the sparse mangrove occurred in the area above the eastern creek (Binidar creek) and in the area between the two creeks. Figure 14 shows ground photographs of the different classes.
Potentials of RISAT SAR in detecting palaeo/buried channels
The region of northwestern India (covering the states of Punjab, Haryana, Gujarat and Rajasthan) and floodplains of River Indus and its tributaries in Pakistan are geographically diverse, geologically active and rich in archaeological sites of Harappan Civilization (2500–
500 BC). In the past few decades a large amount of work has been carried out to map palaeo/buried channels in this region using multi-sensor satellite data, including radar data and understand their migration and evolution42–50. These studies have shown evidence of a prominent river system, which has become buried under sand cover of the Thar Desert sometime during late Holocene. This major river has been identified as Sarasvati, a legendary river mentioned in ancient Indian texts. Late Quaternary cli- matic changes and neotectonics have significantly modi- fied the ancient drainage courses in the region and it is difficult to interpret buried/palaeo channels due to pres- ence of vast spread of aeolian sand cover in optical images. Sahai51, and Rajani and Rajawat52 have attempted to understand spatial distribution and number of Harap- pan sites along the identified courses of the River Saras- vati by superposing location of Harappan sites on the palaeochannels of the River Sarasvati in a GIS environ- ment. The study could demonstrate that the clusters of Harappan settlements from the mature period to later period have moved in the same direction as the migration
of the River Sarasvati. The analysis of ERS-1/2 SAR data covering parts of Thar Desert in western Rajasthan led to identification of hitherto unknown buried channels, relict valleys and shallow, sand-covered limestone areas47,49,50,53,54
. Buried channel was identified near an archaeological site, Talakadu (Mysore District, Karna- taka) situated on the banks of the River Cauvery is south- ern India using RADARSAT-1, C band, VV polarization, fine-beam SAR data55,56 of 22, 25 April and 19 May 2008. RISAT SAR data in conjunction with optical images are being analysed to detect buried channels and associ- ated archaeological sites followed by field studies for validation in the Thar Desert. Figures 15 and 16 demon- strate the potential of RISAT SAR data to detect palaeo- channels in parts of Rajasthan.
These are known palaeochannels, and distinct high to moderate backscatter is observed on RISAT SAR images due to higher soil moisture compared to the adjoining sand-dune areas. In addition, lower topography along the palaeochannel is better enhanced due to oblique illumina- tion of SAR. Vegetation along the palaeochannel is dis- tinctly seen on Landsat ETM FCC. Merged RISAT SAR and Landsat ETM FCC integrate the complementary information content of both sensors and the signature of palaeochannel is distinctly enhanced. Work is in progress to detect hitherto unknown palaeochannels in other parts of the Thar Desert.
1. Chakraborty, M., Panigrahy, S. and Sharma, S. A., Discrimination of rice crop grown under different cultural practices using tempo- ral ERS-1 SAR data. ISPRS Photogramm. Remote Sensing, 1997, 52, 183–191.
2. Panigrahy, S., Chakraborty, M., Sharma, S. A., Kundu, N., Ghose, S. C. and Pal, M., Early estimation of rice acre using temporal ERS -1 Synthetic Aperture Radar data – a case study for Howrah and Hughly districts of West Bengal, India. Int. J. Remote Sens- ing, 1997, 18, 1827–1833.
3. Alpers, W. and Brummer, B., Atmospheric boundary layer rolls observed by the synthetic aperture radar aboard the ERS-1 satel- lite. J. Geophys. Res., 1994, 99, 12613–12621.
4. Kasischke, E. and Bourgeau-Chavez, L., Monitoring South Florida wetlands using ERS-1 SAR imagery. Photogramm. Eng. Remote Sensing, 1997, 63, 281–291.
5. Kasischke, E., Bourgeau-Chavez, L., Smith, K., Romanowicz, E.
and Richardson, C., Monitoring hydropatterns in South Florida ecosystems using ERS SAR data. In Proceedings of the Third Symposium on Space at the Service of our Environment, Florence, Italy, 1997, pp. 71–76.
6. Dwivedi, R. S., Rao, B. R. M. and Bhattacharya, S., Mapping wet- land of Sunderban delta and its environs using ERS-1 SAR data.
Int. J. Remote Sensing, 1999, 20, 2235–2247.
7. Baghdadi, N., Bernier, M., Gauthier, R. and Neeson, I., Evaluation of C-band SAR data for wetland mapping. Int. J. Remote Sensing, 2001, 22, 71–88.
8. Hess, L., Melack, J. and Simonett, D., Radar detection of flooding beneath the forest canopy: a review. Int. J. Remote Sensing, 1990, 11, 1313–1325.
9. Hess, L., Melack, J., Filoso, S. and Wang, Y., Delineation of in- undated area and vegetation along the Amazon floodplain with the SIR-C synthetic aperture radar. IEEE Trans. Geosci. Remote Sens- ing, 1995, 33, 896–904.
10. Slatton, K. C., Crawford, M. M., Gibeaut, J. C. and Gutierrez, R., Wetland vegetation using polarimetric SAR. http://www.csr.
utexas.edu/projects/rs/bol1.html (accessed on 19 April 2006).
11. Pope, K., Rey-Benayas, J. and Paris, J., Radar remote sensing of forest and wetland ecosystems in the Central American tropics.
Remote Sensing Environ., 1994, 48, 205–219.
12. Kushwaha, S., Dwivedi, R. S. and Rao, B. R. M., Evaluation of various digital image processing techniques for detection of coastal wetlands using ERS-1 SAR data. Int. J. Remote Sensing, 2000, 21, 565–579.
13. Ramsey, E. and Laine, S., Comparison of Landsat Thematic Map- per and high resolution photographs to identify change in complex coastal wetlands. J. Coast. Res., 1997, 13, 281–292.
14. McCauley, J. F. et al., Subsurface valley and geoarchaeology of eastern Sahara revealed by Shuttle Radar. Science, 1982, 218, 1004–1019.
15. Schaber, G. G., McCauley, J. F., Breed, C. S. and Olhoeft, G. R., Shuttle imaging radar: physical controls on signal penetration and subsurface scattering in the eastern Sahara. IEEE Trans. Geosci.
Remote Sensing, 1986, GE-24, 603–623.
16. Guo, H., Schaber, G. G. and Breed, C. S., Shuttle imaging radar response from sand and subsurface rocks of Alashan Plateau in north-central China. In International Symposium on Remote Sens- ing for Resources Development and Environmental Management, 7th, Proc. ISPRS Comm. VII, Balkema, Boston, 1986, pp. 137–
143.
17. Berlin, G. L., Tarabzouni, M. A., Al-Nasser, A., Sheikho, K. M.
and Larson, R. W., SIR-B subsurface imaging of a sand buried landscape: Al Labbah Plateau, Saudi Arabia. IEEE Trans. Geosci.
Remote Sensing, 1986, GE-24, 595–602.
18. Avery, T. E. and Berlin, G. L., Fundamentals of Remote Sensing and Airphoto Interpretation, MacMillan, New York, 1992, 5th edn, p. 472.
19. Blom, R. G., Crippen, R. E. and Elachi, C., Detection of subsur- face features in SEASAT radar images of Means Valley, Mojave Desert, California. Geology, 1984, 12, 346–349.
20. Farr, T. G., Elachi, C., Hartl, P. H. and Chowdhury, K., Micro- wave penetration and attenuation in desert soil: a field experiment with the Shuttle Imaging Radar. IEEE Trans. Geosci. Remote Sensing, 1986, GE-24, 590–594.
21. Breed, C. S., Schaber, G. G. and McCauley, J. F., Subsurface geology of the Western Desert in Egypt and Sudan revealed by Shuttle Imaging Radar (SIR-A). In First Spaceborne Imaging Radar Symposium, JPL Publ. 83-11, Jet Propulsion Laboratory, Pasadena, CA, 1983, pp. 10–12.
22. Davis, P. A., Breed, C. S., McCauley, J. F. and Schaber, G. G., Surficial geology of the Safsaf region, south-central Egypt, derived from remote-sensing and field data. Remote Sensing Envi- ron., 1993, 46, 183–203.
Figure 16. Palaeochannel detected on RISAT SAR and Landsat ETM merged FCC covering parts of Barmer, Jalor and Pali districts in the southern region of Luni River.
23. Elachi, C. Roth, L. E. and Schaber, G. G., Spaceborne radar sub- surface imaging in hyperarid regions. IEEE Trans. Geosci. Remote Sensing, 1984, GE-22, 383–388.
24. McCauley, J. F. et al., Palaeodrainages of the Eastern Sahara – the radar rivers revisited (SIR-A/B implications for a Mid-Tertiary Trans-African drainage system). IEEE Trans. Geosci. Remote Sensing, 1986, GE-24, 624–648.
25. Brandt, P., Alpers, W. and Backhaus, J. O., Study of the genera- tion and propagation of internal waves in the Strait of Gibraltar using a numerical model and synthetic aperture radar images of the European ERS 1 satellite. J. Geophys. Res., 1996, 101, 14237–
14252.
26. Horstmann, J., Koch, W., Lehner, S. and Tonboe, R., Wind re- trieval over the ocean using synthetic aperture radar with C-band HH polarization. IEEE Trans. Geosci. Remote Sensing, 2000, 38, 2122–2131.
27. Rosich, B. and Meadows, P., Absolute calibration of ASAR level 1 products generated with PF-ASAR. Technical Note, ENVI- CLVL-EOPG-TN-03-0010, European Space Agency, Issue 1 revi- sion 5, 7 October 2004, p. 26.
28. Hersbach, H., Stoffelen, A. and Haan De, S., An improved C-band scatterometer ocean geophysical model function: CMOD5. J.
Geophys. Res., 2007, 112, 1–18; doi:10.1029/2006JC003743.
29. Mouche, A. A., Hauser, D., Daloze, J. F. and Guerin, C., Dual polari- zation measurements at C-band over the oceans: results from airborne radar observations and comparison with ENVISAT ASAR data. IEEE Trans. Geosci. Remote Sensing, 2005, 43, 753–769.
30. Monaldo, F., Thompson, D. R., Beal, R. C., Pichel, W. G. and Clemente-Colon, P., Comparison of SAR-derived wind speed with model predictions and ocean buoy measurements. IEEE Trans.
Geosci. Remote Sensing, 2001, 39, 2587–2600.
31. Wright, J. W., Backscattering from capillary waves with applica- tion to sea clutter. IEEE Trans. Antennas Propag., 1966, AP-14, 749–754.
32. Goldfinger, A. D., Estimation of spectra from speckled images.
IEEE Trans. Aerosp. Electron. Syst., 1982, AES-18, 675–681.
33. Engen, G., Vachon, P. W., Johnsen, H. and Dobson. F. W., Retrieval of ocean wave spectra and RAR MTFs from dual- polarization SAR data. IEEE Trans. Geosci. Remote Sensing, 2000, 38, 391–403.
34. Eldhuset, K., An automatic ship and ship wake detection system for spaceborne SAR images in coastal regions. IEEE Trans. Geo- sci. Remote Sensing, 1996, 34, 1010–1019.
35. Kathiresan, K. and Bingham, B. L., Biology of mangroves and mangrove ecosystems. Adv. Mar. Biol., 2001, 40, 81–251.
36. Lucas, R. M., Mitchell, A. L., Rosenqvist, A., Proisy, C., Melius, A.
and Ticehurst, C., The potential of L-band SAR for quantifying man- grove characteristics and change: case studies from the tropics.
Aquat. Conserv.: Mar. Freshwater Ecosyst., 2007, 17, 245–264.
37. Kasichke, E. S., Melack, J. M. and Dobson, M. C., The use of imaging radars for ecological applications – a review. Remote Sensing Environ., 1997, 59, 141–156.
38. Proisy, C., Mougin, E. and Fromard, F., Radar remote sensing of mangroves: results and perspectives. In Proceedings of IGARSS Conference, Sydney, Australia, 9–13 July 2001.
39. Kumar, T. and Patnaik, C., Characterization of mangrove vegeta- tion using temporal C-band synthetic aperture radar data: a case study of Sundarbans. In Proceedings of International Conference on Geospatial Technologies and Applications ‘Geomatrix 12’, IIT Mumbai, CD-ROM, 2012.
40. National Wetland Atlas. SAC/EPSA/ABHG/NWIA/ATLAS/
34/2011, Space Applications Centre, ISRO, Ahmedabad, 2011.
41. Forest Survey of India, Forest cover mapping. Ministry of Envi- ronment and Forests, FSI, Dehradun, 2009; http://www.fsi.org.
in/scheme_of_classification.html
42. Ghose, B., Kar, A. and Hussain, Z., The lost courses of the Saras- vati River in the Great Indian Desert: new evidence from the Landsat imagery. Geogr. J., London, 1979, 145, 446–451.
43. Ghose, B., Kar, A. and Hussain, Z., Comparative role of Aravalli and the Himalayan river systems in the fluvial sedimentation of the Rajasthan desert. Man Environ., 1980, 4, 8–12.
44. Pal, Y., Sahai Baldev, Sood, R. K. and Agarwal, D. P., Remote sensing of the ‘lost’ Saraswati river. Proc. Indian Acad. Sci.
(Earth Planet. Sci.), 1980, 89, 317–331.
45. Sood, R. K. and Sahai, B., Hydrographic changes in northwestern India. Man Environ., 1983, 7, 166–169.
46. Ramasamy, S. M., Bakliwal, P. C. and Verma, R. P., Remote sens- ing and river migration in western India. Int. J. Remote Sensing, 1991, 12, 2597–2609.
47. Rajawat, A. S. et al., Potentials of radar (ERS-1/2 SAR) and high resolution IRS-1C data in reconstructing palaeodrainage network of western Rajasthan. Mem. Geol. Soc. India, 1999, 42, 245–258.
48. Rajawat, A. S., Sastry, C. V. S. and Narain, A., Application of py- ramidal processing on high resolution IRS-1C data for tracing the migration of the Saraswati river in parts of the Thar desert. Mem.
Geol. Soc. India, 1999, 42, 259–272.
49. Rajawat, A. S., Verma, P. K. and Nayak, S., Reconstruction of pa- laeodrainage network in northwest India: retrospect and prospects of remote sensing based studies. Proc. Indian Natl. Sci. Acad. Part A, 2003, 69, 217–230.
50. Rajawat, A. S., Reconstruction, evolution and significance of palaeodrainage network in western Rajasthan: an investigation based on multisensor satellite data and Geographic Information System (GIS). PhD thesis, Vikram University, Ujjain, 2006, p. 331.
51. Sahai, B., Unraveling of the ‘lost’ Vedic Sarasvati. Mem. Geol.
Soc. India, 1999, 42, 121–141.
52. Rajani, M. B. and Rajawat, A. S., Lost courses of River Sarasvati and spatial distribution of Harappan settlements – observations based on multisensor satellite data. In Remote Sensing for Arche- ology and Cultural Heritage Management (eds Lasaponara, R. and Masini, N.), Rome, Italy, 2008, pp. 383–386.
53. Mehta, N. S., Rajawat, A. S., Bahuguna, I. M., Mehta, D. S. and Shrimal, A. K., Geological potential of ERS-1 SAR data: observa- tions in parts of Aravalli and Thar Desert (western India). In Pro- ceedings of the Second ERS-1 Symposium – Space at the Service of Our Environment, Hamburg, Germany, ESA SP-361, 11–14 October 1993, pp. 931–936.
54. Mehta, N. S., Rajawat, A. S., Bahuguna, I. M., Mehta, D. S., Bhatnagar, P. and Shrimal, A. K., Remote sensing for mineral exploration: geological potentials of ERS-1 SAR data covering parts of Aravalli Hills and Thar Desert (Rajasthan, India). Project Report, SAC/RSA/RSAG/ERS-1/GEO/PR/01/94, Space Applica- tions Centre, ISRO, Ahmedabad, 1994, p. 67.
55. Rajani, M. B., Rajawat, A. S., Krishna Murthy, M. S., Kamini, J.
and Rao, S., Demonstration of the synergy between multi-sensor satellite data, GIS and ground truth to explore the archaeological site in Talakadu region in South India. J. Geomatics, 2012, 6, 37–
42.
56. Rajani, M. B., Bhattacharya, S. and Rajawat, A. S., Synergistic application of optical and radar data for archaeological exploration in the Talakadu region, Karnataka. J. Indian Soc. Remote Sensing, doi: 10.1007/s12524-011-0102-6, published online: 3 June 2011.
ACKNOWLEDGEMENTS. This work was carried out under the RISAT-1 Utilization Programme funded by ISRO. We thank A. S.
Kiran Kumar, Director, Space Applications Centre, ISRO, Ahmedabad and Dr J. S. Parihar, Deputy Director, Earth, Ocean, Atmosphere, Planetary Sciences and Applications Area for their encouragement.
