8ISG Journal of Geomatics

9

Water Area Extraction Using Geocoded High Resolution Imagery of TerraSAR-X Radar Satellite in Cloud Prone Brahmaputra River Valley

Md. Surabuddin Mondal*, Nayan Sharma*, P K Garg**, Bettina Böhm***, Wolfgang-Albert Flügel ***, R D Garg**, R P Singh*

*Dept. of W R D & M, Indian Institute of Technology. Roorkee, India-247667**Dept. of Civil Engineering, Indian Institute of Technology. Roorkee, India-247667

*** Dept. of Geoinformatics, Hydrology and Modelling, Friedrich-Schiller University (FSU-Jena), Germanymsk.iit@gmail.com, smondal.iit@gmail.com

(Received : 11 November, 2008; in final form 20 June, 2009)

Abstract: In this study, supervised pixel based maximum likelyhood classifier (MLC) was evaluated for differentiating water area from non-water area in a cloud prone river valley region using images obtained from German TerraSAR-X radar satellite. TerraSAR-X satellite microwave data acquired on 8th May 2008 was used for a test site in Brahmputra river valley, India. The performance of traditional supervised (MLC) classification method was evaluated on TerraSAR’s X band strip geo coded high resolution (6 meter) images in VV polarization channel. The 42 training sites for water area and 40 training sites for non-water areas were carefully selected over the entire image where optical Resourcesat - 1 LISS IV multi-spectral image and pre-classification ground truth were used as apriori knowledge. The calculated backscatter coefficient ranges from approximately -24.1 to 6.5 dB where water areas usually have low dB value around -20 dB. Overall classification accuracy was 94.92 %. This study shows that high resolution TerraSAR-X radar satellite image have advantage of weather independence over optical data. Furthermore, the supervised (MLC) classification can be used for extraction of water area from single band high resolution radar images where traditional water area extraction methods i.e., NDVI, NDWI etc., can not be used on radar images.

Keywords: TerraSAR-X Radar Satellite Images, Supervised (MLC) Classifier, Accuracies, Water and Non-water area.

1. Introduction

Optical data are commonly used for land cover mapping. However, the problem of cloud cover is always an obstacle, which restricts the use of optical remotely sensed data (Colwell, 1983). Since the Brahmaputra river basin is overcast with clouds during monsoon, the availability of cloud free optical data is very poor. To overcome this difficulty, use of microwave satellite data is envisaged. Like other radar satellite images TerraSAR-X microwave satellite images have great advantages of weather independence over optical satellite images. NDVI (Normalized Difference Vegetation Index) was introduced by Rouse et al. (1973) to separate green vegetation from non-vegetation area using Landsat MSS optical data. It is expressed as the difference between the near infrared and red bands normalized by the sum of those bands, i.e. NDVI= (NIR-RED) / (NIR+RED). Later on NDWI (Normalized Difference Water Index) was introduced by Gao (1996) to assess water content in a normalized way i.e., NDWI = (NIR – SWIR)/ (NIR + SWIR). These are the traditional normalized differences used differentiating water area from non-water area using multi-band optical remote sensing data. But NDVI, NDWI can not be used on single band radar images. Maximum likelihood classifier (MLC) has been applied to TerraSAR-X microwave satellite high-resolution images for differentiating water and non-water area covering parts of Brahmaputra river basin.

2. Description of TerraSAR-X Radar Satellite Images

TerraSAR-X is a German remote sensing satellite program,

which is the first commercially available radar satellite cloud independent to offer 1 meter to 16 meter resolutions. TerraSAR-X is a satellite with right-side-looking X-band synthetic aperture radar (SAR) based on active phased array antenna technology (DLR, 2006). The active antenna allows not only the conventional StripMap mode but also Spotlight and Scan SAR modes. The following imaging modes are defined for the generation of basic image products. Strip Map mode (SM) in single or dual polarization (3 meter resolution), high resolution Spotlight mode (HS) in single or dual polarization (1/2 meter resolution), spotlight mode (SL) in single or dual polarization (2/4 meter resolution), ScanSAR mode (SC) in single polarization (16 meter resolution) (DLR, 2006). This is the basic SAR imaging mode as known from other radar satellites. The ground swath is illuminated with a continuous sequence of pulses while the antenna beam is pointed to a fixed angle in elevation and azimuth. This results in an image strip with constant image quality in azimuth. The maximum length of an acquisition is limited by battery power, memory and thermal conditions in the sensor (Jörg et al., 2007). The orbital and altitude parameters of TerraSAR-X Radar satellite are listed in Table 1(a).The characteristic parameters of this StripMap mode are listed in Table 1(b). As mentioned in Table 1(b), StripMap can be operated in single or dual polarization mode resulting in one or two image layers, respectively.

3. Data Used

A single date TerraSAR-X VV polarized with StripMap mode (SM) microwave images acquired on 8th May 2008 was used to differentiate water area from non-water area. The ground

© Indian Society of Geomatics

Journal of GeomaticsJournal of Geomatics Vol 3 No.1 April 2009 Vol 3 No.1 April 2009

10

resolution of the radar scene is 6 meter with the pixel spacing at 2.75 meter in TIFF/IMG format. In supervised classification for optical data, the identity and

Table – 1(a): Orbit and Altitude Parameters of TerraSAR-X Radar Satellite

location of some of the land cover types are known a priori through a combination of field visits and study of toposheets. Since the study is limited to identifying within water and non water from radar satellite images, optical Resourcesat -1 LISS IV multispectral images of 16th May 2008 of the same area and ground truth are used as priori knowledge. Generally, Radar satellite sensors are quite sensitive to soil moisture so there are chances of confusion to identify highly soil moisture content area and water area. To avoid this confusion, optical LISS-IV image of the same area is used as prior knowledge. In general optical data i.e., LISS-IV images are not much sensitive to soil moisture contents, like radar data. Land use map of the study area, post classification ground truth and optical Resourcesat -1 LISS IV multispectral images are also used for accuracy assessment.

Table – 1(b): Characteristic Parameters of Strip Map Mode

4. Test Site

Test site is located between latitude 26° 5’ 7” to 26° 29’ 20” North and longitude 92° 00’ 26” to 92° 14’ 53” East. It covered a part of Brahmputra river basin in north-east India, comprising an area of 717.62 km². This part of Brahmputra river basin is characterized as river valley (flood plain) area with some hilly region in southern part (Fig.-1).

5. Methods Used

The supervised maximum likelihood method was used to classify radar image. The classification was performed with

11

6. Results and Analysis

Fig. - 2(a) shows TerraSAR-X radar image of the study area and Fig. - 2(b) provides the supervised classification results from TerraSAR-X radar image. Table – 3 shows the area statistics of results. The resulting classification image produced a classification accuracy with 94.92 % overall accuracy where optical Resourcesat -1 LISS IV multi-spectral images and pre-classification ground truth are used as priori knowledge to classify TerraSAR-X radar image. The accuracy of classification was checked using error matrix methods where stratified random sampling (3x3 windows) method was used for sampling collection. The reference data used to verify the accuracy of results was land use map of the study area and limited post classification ground truth. The overall accuracy was 94.92 % and overall kappa coefficient was 0.84. Table – 4 illustrates the different accuracy of classification results.

improve classification accuracy when radar satellite images (i.e. TerraSAR-X radar satellite images) are used for classification. The information derived from optical data can lead to improvement in classification of microwave data.

3. Supervised classification can be used for single band radar data classification. Supervised (MLC) classification produced the satisfactory results for TerraSAR-X radar image classification.

Journal of GeomaticsJournal of GeomaticsJournal of GeomaticsJournal of Geomatics Vol 3 No.1 April 2009 Vol 3 No.1 April 2009

Source: http://www.infoterra.de

82 training sites. The 42 training sites for water area and 40 training sites for non-water areas were carefully selected over the entire image. An attempt was made to locate specific sites as priori knowledge in the remotely sensed data that represents homogeneous land cover types. These areas are commonly referred as training sites in supervised classification. As stated earlier optical Resourcesat -1 LISS IV multispectral images and pre-classification ground truth were used as apriori knowledge.

Multivariate statistical parameters are calculated for each training site. Every pixel both within and outside these training sites is then evaluated and assigned to a class of which it has the highest likelihood of being a member. Tables - 2(a) & 2(b) show characteristics of the training sites used for image classification.

The calculated backscatter coefficients range from approximately -24.1 to 6.5 dB where water areas usually have low dB value around -20 dB. Backscattering coefficients of the shadow area in high relief and slope regions are very similar to those of the water area (Peng, et al., 2003). This part of Brahmaputra river basin is characterized as mainly by river valley (flood plain) area with small area covered by hill in southern part. The terrain shape i.e. elevation (average 90 metres) and slope (less than 10°) of hilly part are also very gentle and therefore, shadow area has not been found in this area. In this regard, the texture information of TerraSAR-X radar data has been carefully studied (pre-classification), especially for the hilly region of study area.

Table – 2(a): Backscattering coefficient (dB) Statistics of Training Site

Table – 2(b): Characteristics of Training Site(Based on Post Processed DN value)

Table-3: Area Statistics of Classified Image

Table-4: Accuracy of Supervised (MLC) Classification Results

7. Conclusion

The following conclusions were drawn from this study.

1. TerraSAR-X radar satellite’s geocoded high resolution image is a good source of data for differentiating water and non-water area in a cloud cover rainy area. In the study area, it is very difficult to acquire an cloud free optical remotely sensed imagery. TerraSAR-X radar data can be used as a satisfactory alternative, to get information about water and non-water area.

2. The high-resolution optical data (i.e. LISS-IV) as well as pre-classification ground truth can be used as priori knowledge to

Fig. – 2(a): TerraSAR-X Radar Image of 8th May, 2008

Fig.1:Location of study Area

Fig. – 2(b): Classified TerraSAR-X Radar Image(Water and Non-Water Area)

Nominal Orbit Height at the Equator

Orbits/\Day

512 km

152/11

Revisit Time (Orbit Repeat Cycle) 11 Days

Inclination097.44

Ascending Node Equatorial Crossing Time

18.00=0.25 h(local time)

Yaw Steering Yes

Source: http://www.infoterra.de

Parameter Value

Scence Extension (Azimuth) 50 km Standard Max.1650 km

Swath Width (Ground Range)

30 km (single polarization) 15 km (dual polarization

Data Access IncidenceAngel Range

0 015 - 60

Full PerformanceIncidence Angel Range

0 020 -45

Number of ElevationBeams ca. 27

Azimuth Resolution 3 m at 150 and 300 MHz

Ground Range Resolution 0 01.55 - 3.21 m @ 45 - 20incidence angle

PolarizationSingle pol(HH,VV),Dualpol(HH/VV,HH/HV,VV/VH)

Water(dB)

Non Water(dB)

MinMax

MeanStd.Dev.

Number of Samples

-23.95-5.35

-19.081.60

956250 1249541

7.09-8.153.52

-21.14

Water Non Water

MinMax

Mean

Std.Dev.

Number of Samples

34.11

956250 1249541

0

142

10.31 90.47

147.47

58943

Class 2Area in km .

Total

WaterNon Water

131.62585.99717.62

Area in%

18.3481.66100

ClassProducerAccuracy

(%)

UserAccuracy

(%)

ConditionalKappa

Coefficient

Water

Non-Water

84.00

97.57 96.57

89.36 0.87

0.80

Overall Classification Accuracy - 94.92%

Overall Kappa (K ) Statistics - 0.84%>

10

resolution of the radar scene is 6 meter with the pixel spacing at 2.75 meter in TIFF/IMG format. In supervised classification for optical data, the identity and

Table – 1(a): Orbit and Altitude Parameters of TerraSAR-X Radar Satellite

location of some of the land cover types are known a priori through a combination of field visits and study of toposheets. Since the study is limited to identifying within water and non water from radar satellite images, optical Resourcesat -1 LISS IV multispectral images of 16th May 2008 of the same area and ground truth are used as priori knowledge. Generally, Radar satellite sensors are quite sensitive to soil moisture so there are chances of confusion to identify highly soil moisture content area and water area. To avoid this confusion, optical LISS-IV image of the same area is used as prior knowledge. In general optical data i.e., LISS-IV images are not much sensitive to soil moisture contents, like radar data. Land use map of the study area, post classification ground truth and optical Resourcesat -1 LISS IV multispectral images are also used for accuracy assessment.

Table – 1(b): Characteristic Parameters of Strip Map Mode

4. Test Site

Test site is located between latitude 26° 5’ 7” to 26° 29’ 20” North and longitude 92° 00’ 26” to 92° 14’ 53” East. It covered a part of Brahmputra river basin in north-east India, comprising an area of 717.62 km². This part of Brahmputra river basin is characterized as river valley (flood plain) area with some hilly region in southern part (Fig.-1).

5. Methods Used

The supervised maximum likelihood method was used to classify radar image. The classification was performed with

11

6. Results and Analysis

Fig. - 2(a) shows TerraSAR-X radar image of the study area and Fig. - 2(b) provides the supervised classification results from TerraSAR-X radar image. Table – 3 shows the area statistics of results. The resulting classification image produced a classification accuracy with 94.92 % overall accuracy where optical Resourcesat -1 LISS IV multi-spectral images and pre-classification ground truth are used as priori knowledge to classify TerraSAR-X radar image. The accuracy of classification was checked using error matrix methods where stratified random sampling (3x3 windows) method was used for sampling collection. The reference data used to verify the accuracy of results was land use map of the study area and limited post classification ground truth. The overall accuracy was 94.92 % and overall kappa coefficient was 0.84. Table – 4 illustrates the different accuracy of classification results.

improve classification accuracy when radar satellite images (i.e. TerraSAR-X radar satellite images) are used for classification. The information derived from optical data can lead to improvement in classification of microwave data.

3. Supervised classification can be used for single band radar data classification. Supervised (MLC) classification produced the satisfactory results for TerraSAR-X radar image classification.

Journal of GeomaticsJournal of GeomaticsJournal of GeomaticsJournal of Geomatics Vol 3 No.1 April 2009 Vol 3 No.1 April 2009

Source: http://www.infoterra.de

82 training sites. The 42 training sites for water area and 40 training sites for non-water areas were carefully selected over the entire image. An attempt was made to locate specific sites as priori knowledge in the remotely sensed data that represents homogeneous land cover types. These areas are commonly referred as training sites in supervised classification. As stated earlier optical Resourcesat -1 LISS IV multispectral images and pre-classification ground truth were used as apriori knowledge.

Multivariate statistical parameters are calculated for each training site. Every pixel both within and outside these training sites is then evaluated and assigned to a class of which it has the highest likelihood of being a member. Tables - 2(a) & 2(b) show characteristics of the training sites used for image classification.

The calculated backscatter coefficients range from approximately -24.1 to 6.5 dB where water areas usually have low dB value around -20 dB. Backscattering coefficients of the shadow area in high relief and slope regions are very similar to those of the water area (Peng, et al., 2003). This part of Brahmaputra river basin is characterized as mainly by river valley (flood plain) area with small area covered by hill in southern part. The terrain shape i.e. elevation (average 90 metres) and slope (less than 10°) of hilly part are also very gentle and therefore, shadow area has not been found in this area. In this regard, the texture information of TerraSAR-X radar data has been carefully studied (pre-classification), especially for the hilly region of study area.

Table – 2(a): Backscattering coefficient (dB) Statistics of Training Site

Table – 2(b): Characteristics of Training Site(Based on Post Processed DN value)

Table-3: Area Statistics of Classified Image

Table-4: Accuracy of Supervised (MLC) Classification Results

7. Conclusion

The following conclusions were drawn from this study.

1. TerraSAR-X radar satellite’s geocoded high resolution image is a good source of data for differentiating water and non-water area in a cloud cover rainy area. In the study area, it is very difficult to acquire an cloud free optical remotely sensed imagery. TerraSAR-X radar data can be used as a satisfactory alternative, to get information about water and non-water area.

2. The high-resolution optical data (i.e. LISS-IV) as well as pre-classification ground truth can be used as priori knowledge to

Fig. – 2(a): TerraSAR-X Radar Image of 8th May, 2008

Fig.1:Location of study Area

Fig. – 2(b): Classified TerraSAR-X Radar Image(Water and Non-Water Area)

Nominal Orbit Height at the Equator

Orbits/\Day

512 km

152/11

Revisit Time (Orbit Repeat Cycle) 11 Days

Inclination097.44

Ascending Node Equatorial Crossing Time

18.00=0.25 h(local time)

Yaw Steering Yes

Source: http://www.infoterra.de

Parameter Value

Scence Extension (Azimuth) 50 km Standard Max.1650 km

Swath Width (Ground Range)

30 km (single polarization) 15 km (dual polarization

Data Access IncidenceAngel Range

0 015 - 60

Full PerformanceIncidence Angel Range

0 020 -45

Number of ElevationBeams ca. 27

Azimuth Resolution 3 m at 150 and 300 MHz

Ground Range Resolution 0 01.55 - 3.21 m @ 45 - 20incidence angle

PolarizationSingle pol(HH,VV),Dualpol(HH/VV,HH/HV,VV/VH)

Water(dB)

Non Water(dB)

MinMax

MeanStd.Dev.

Number of Samples

-23.95-5.35

-19.081.60

956250 1249541

7.09-8.153.52

-21.14

Water Non Water

MinMax

Mean

Std.Dev.

Number of Samples

34.11

956250 1249541

0

142

10.31 90.47

147.47

58943

Class 2Area in km .

Total

WaterNon Water

131.62585.99717.62

Area in%

18.3481.66100

ClassProducerAccuracy

(%)

UserAccuracy

(%)

ConditionalKappa

Coefficient

Water

Non-Water

84.00

97.57 96.57

89.36 0.87

0.80

Overall Classification Accuracy - 94.92%

Overall Kappa (K ) Statistics - 0.84%>

12

Acknowledgements

The authors are thankful to German Space Agency (DLR) for providing us geocoded TerraSAR-X radar satellite data to carry out this study.

References

Colwell N. Robert, (1983). Manual of Remote Sensing. Vol. -1.2nd Ed. American Society of Remote Sensing. Virginia, pp. 370-426.

DLR Cluster Applied Remote Sensing, (2006). TerrsaSAR-X Ground Segment Basic Product Specification. Issue 1.4 October 6th, 2006, pp. 10-15.

Gao, B.C., (1996). NDWI - normalized difference water index for remote sensing of vegetation liquid water from space. Remote Sensing of Environment, 58, 257-266.

Jörg Herrmann, Alejandra and González Bottero, (2007). TerraSAR-X Mission: The New Generation in High Resolut ion Satel l i tes . Infoterra GmbH, 88039 Friedrichshafen, Germany. Anais XIII Simpósio Brasileiro de Sensoriamento Remoto, Florianopolis, Brasil, 21-26 April 2007, INPE, pp. 7063-7070.

Peng, X., J. Wing, M. Read, and J. Gari, (2003). Land cover mapping from RADARSAT stereo images in a mountainuious area of southern Argentina, Canadian Journal of Remote Sensing, 29(1), 75-87.

Rouse, J. W., R. H. Haas, J. A. Schell, and D. W. Deering, (1973). Monitoring vegetation systems in the Great Plains with ERTS, Third ERTS Symposium, NASA SP-351 I, 309-317.

URL: Basic Image Products at: http://www.infoterra.de

Journal of GeomaticsJournal of Geomatics Vol 3 No.1 April 2009 59Journal of Geomatics

1. Introduction

Landslide has been a common but disastrous problem of the hilly terrains, especially during the monsoon seasons, when the bulk of soils gets detached from its original location and flows down the hills causing loss and disturbances to lives and public properties (Govt. of Sikkim Annual report 2005). Similar to other Himalayan terrains, the Sikkim state of India also has formidable physical features. Being a part of the Himalayan orogenic belt, the natural hazards (landslides, earthquakes) also form an integral part of the state. Major portion of state is covered by Pre-Cambrian rock and is much younger in age (Lama 2001). The Rangeet and the Teesta, which form the main channels of drainage, run nearly North-South. The valleys cut by these rivers and their chief feeders are very deep.

Landslides in Sikkim are triggered both due to natural phenomena (high rainfall, seismicity) and anthropogenic activities (road cutting, deforestration). Commonly observed slope failures include block slide, debris slide and earth creep. Thus, mitigation and management of the landslide hazard in this area is one of the foremost requisites for landuse planners. Factors contributing to slope failures at a specific site are generally specific with respect to conditioning and triggering factors (Harp and Jibson, 1996; Jibson, et.al., 1994; 1998; 1999). Hence, hazard maps representing the

susceptibility of slope failures due to different conditioning and triggering factors (variables) could be a better choice in preparing hazard zonation maps. Generally landslides in Sikkim often occur at the slopes of national highway (NH31 A).The main effected location of many events is on the road side areas and due to this transportation of entire Sikkim gets highly affected.

The goal of this study is to carry out a pilot study to predict the locations for landslide hazards within a small area. Spatial identification of hazard and risk allows for implementation strategies to effectively reduce risks. This study serves as a valuable tool for others. For example, transportation officials, foresters etc can benefit for identifying hazard zones in which landslides might occur.

Several probabilistic methods (for example quantification theory, multiple regression, discriminant analysis, monte-carlo simulation, etc.) have been used (Hayashi, 1952; Carrara,1983; Haruyama and Kitamura, 1984; Kawakami,1984; Yin and Yan,1988; Jade and Sarkar, 1993; Luzi et.al. and Ramakrishnan et al.2005) in the past to derive a probabilistic zonation map for landslide hazard(Gorsevski et al 2000). In this paper an attempt has been made to create a hazard model using a simpler methodology using aggregation method by overlay analysis of ARC GIS. This method is preferable because it can be worked out by Boolean

Application of Remote sensing & GIS for landslides hazard and assessment of their probabilistic occurrence - A case study of NH31A between Rangpo and Singtam

Mousumi Gupta1, MK Ghose2 and LP Sharma31Dept of Computer Science & Engineering,Sikkim Manipal Inst.Tech., Majitar, E Sikkim mousmi_gt@yahoo.co.in2Dept of Computer Science & Engineering,Sikkim Manipal Inst.Tech., Majitar, E Sikkim headcse.smit@gmail.com

3National Informatics Center Gangtok, Sikkim lp.sharma@nic.in

(Received : 19 December, 2008; in final form June 1, 2009)

ISG

© Indian Society of Geomatics

Abstract: Several methodologies using Remote sensing and GIS are cited in the literature for landslide hazard assessment. Most of these methods need extensive mathematical modeling / simulation to evaluate the probability of occurrences of landslides. Proper methodology of assessing the landslide zone and the probability of occurrences of landslides will definitely be instrumental in landslide mitigation problem. In this paper an attempt has been made to assess the landslide hazard using a deterministic method. The study area has been chosen from Rangpo and Singtam along NH 31 A. The identified conditioning factors include soil, geology, forest, and drainage, and triggering factors such as slope and aspects are taken as input to fit into an aggregation model for assessment of landslide hazard. These probabilistic maps are compared with landslide maps generated from Google Earth from recent data (2007) for the accuracy of prediction. The generated hazard maps agree with the observed landslide occurrences. Thus the proposed methodology can be used in landslide hazard zonation prediction.

Keywords: Landslide hazard, Sikkim, Raster, Reclassify, DEM, Overlay Analysis

Vol 3 No.1 April 2009

Page 13Page 14Page 15Page 16