INTEGRATED USE OF SAR AND OPTICAL DATA FOR COASTAL ZONE MANAGEMENT

 ABSTRACT

The usefulness of integration of SAR (ERS) and TM (LANDSAT) data, for coastal zone management has been demonstrated for the eastern coastal areas of Pakistan. The study area includes a part of Indus Delta and Karachi sea port. Different types of mangrove forest and other land- cover features have been classified. Sea pollution and possible sources detected. Different digital filter have been applied to SAR data to ease image interpretation and a comparison between the filtered images have been made to evaluate their performance/usefulness. It has been observed that MAP and LEE filter are among the best for such studies. The Normalised Difference Vegetation Index (NDVI) and the optical brightness have been calculated using LANDSAT TM Data and are combined with the MAP filtered SAR Image. The resultant colour composite have been used to obtain a landcover / landuse map of the area.

Keywords: Satellite Remote Sensing, landcover, mapping, ERS-SAR, LANDSAT-TM

 

INTRODUCTION

 The growing pressure on resources, infrastructure and land in the coastal areas of developing countries as a result of expanding population and increasing commercial, industrial, and other development related activities has underlined the urgent need to manage these coastal areas in an optimal and judicious manner. Timely reliable information of the coastal environment is obviously an essential prerequisite for such efforts aimed at optimising the management of these areas.

Karachi is the largest city of Pakistan and is situated at 67° E 24° 45'N. It is associated with the biggest port of the country and is the trading, industrial and financial capital of the country. About 60% industry of the city is located along the Liari river which has its mouth very close to the sea port. Because of this and of ship traffic it is one of the most polluted areas of the Arabian Sea. This pollution has an adverse effect on mangrove forests and natural habituates. The complexity of the environmental problem is further increased by the nearby delta of Indus River.

 The constant interaction between land, sea and atmosphere makes it an area where diverse processes are always at work. The coastal areas of the region were ever changing in the past, but now human activities as an influence factor has also been added. The complex interaction between natural and man-made processes needs to be well understood to avoid or at least to control and minimise the adverse affects on the environment and ecology of the area.

Accurate and comprehensive information acquired frequently, is obviously an essen-tial pre-requisite for all efforts aimed at optimising the management of such coastal area. The synoptic, repetitive and multi-spectral data available from different Remote Sensing Satellites can provide meaningful information on both natural and man-made processes in the coastal areas. Satellite Remote Sensing data can thus supplement existing efforts based on conventional ground based techniques for monitoring and managing of coastal zones.

METHODOLOGY

 The new generation of Remote Sensing Satellites i.e. ERS-1, ERS-2, JERS, Radarsat etc. are equipped with Synthetic Aperture Radars (SAR) which are very sensitive essentially to roughness and allow to differentiate mangroves forest, of similar optical spectral reflectivity.Moreover, SAR data can also be used to differentiate densitiy of urbanisation. Thus a complimentary use of SAR and Optical data has opened up new application areas for satellite remote sensing. The flow chart to generate landcover map based on the complementary use of ERS SAR and LANDSAT TM data used in this investigation, is shown in Fig. 1.

SAR SPECKLE FILTERING

The main problem associated with the 16 bit SAR data is speckle noise which needs to be reduced to enhance the image interpretability and to improve the classification. Different filters have been reported in the literature [1, 2] to reduce the speckle noise in SAR images. To evaluate their performance, different speckle reducing filters had been applied to 16 bit SAR data of the Karachi port and its surrounding areas. The co-efficient of variance (s = Mean / Standard Deviation), for different landuse/landcover classes using sub images (50 x 50 pixels) for each class have also been calculated and is listed in Table 1. It is observed that the difference between the coefficients of variance (s) for different landuse/landcover categories, after the use of Lee or MAP filters, is more prominent and hence they are regarded as the best speckle reducing filters for such studies.

The speckle reduced output images of the area are shown in Fig. 2a and Fig. 2b. It has also been noted that visual inspection probably provides the best assessment of performance of a particular filter. It is observed that a MAP filter data set is a smoothed image of reduced speckle but still shows all the fine details and hence all meaningful information. It has been chosen for its complementary use with the optical LANDSAT TM data.

A composite image comprising SAR 16 bit (original) (blue), MAP filtered (red) and Lee filtered (green) images of Karachi port area and a part of Indus delta area have been shown in Fig. 3a and Fig 3b. It is observed that the performance of MAP filter is slightly better than that of Lee filter as the colour composite contain quite a few red-dish/magenta spots indicating the information present in original 16 bit SAR image but lost in Lee filtered image. The lee filter, while smoothing out the image, destroyed some information over the buildup areas near karachi port. The better performance of the MAP filter over buildup areas can also be confirmed by low values of s for "urban" in Table 1. Furthermore, a blurring of some narrow channels in the Indus Delta area, can also observed in the Lee filtered image.

Table 1: Coefficient of variance (s = mean / standard deviation) for different land-cover categories after application of filters

Fig.1: Flow chart of the complementary use of SAR and optical data

Fig. 2a. A comparison of output images (Karachi port area) after application of different digital filters. Speckle noise can be observed in the original 16bit SAR image "SAR". The MAP filtered image is smoothed but still showing fine details. Therefore, it is regarded as the best speckle reducing filter for such studies.

Fig. 2b: A comparison of output images (Karachi port area) after application of different digital filters. Speckle noise can be observed in the original 16bit SAR image "SAR" on Fig2a. The MAP filtered image is smoothed but still showing fine details. Therefore, it is regarded as the best speckle reducing filter for such studies.

Fig 3a: Karachi port area. Composites of orginal 16bit SAR image (blue), Lee filtered image (green) and MAP filtered image (red). Magenta spots in the image indicate the information present in both, SAR 16bit data and MAP filtered image, but lost during smoothing/blurring by Lee filter.

Fig 3b: Portion of Indus Delta area.Composites of orginal 16bit SAR image (blue), Lee filtered image (green) and MAP filtered image (red). Magenta spots in the image indicate the information present in both, SAR 16bit data and MAP filtered image, but lost during smoothing/blurring by Lee filter.

SAR/OPTICAL COLOUR COMPOSITE

The MAP filter had been applied to the entire SAR images (16 bit) of both areas (Karachi Port area and the Indus Delta Area), before its reduction to 8 bit pixels and resampling to the new pixel size, equivalent to that of LANDSAT TM data. The final SAR sub-scene images have been merged with the corresponding TM sub scenes. Beside other land-cover features, both the test areas have mangroves forest, thus a Normalised Difference Vegetation Index (NDVI) using band 4 (infrared) and band 3 (red) of TM data [3] have been calculated. Similarly optical brightness have been calculated by averaging bands 2 and 3 of TM data [4]. The band combination of the SAR/optical composite, as shown in Fig. 4a and Fig. 4b is explained in Table 2.

Table 2: Band combination of final SAR / optical colour composites

Fig. 4a: Karachi port area. Composites of Map filtered SAR image (Blue), optical brightness (green) and NDVI (red). Catagories of mangroves can be distinguished by change of colour (red to magenta) depending upon the de3nsity and heights of trees. Other landcover features are also very clear and easy to recognise.

Fig. 4b: Portion of the Indus Delta area. Composites of Map filtered SAR image (blue), optical brightness (green) and NDVI (red). Catagories of mangroves can be distinguished by change of colour (red to magenta) depending upon the density and heights of trees. Other landcover features are also very clear and easy to recognise.

CLASSIFICATION

Based on the colour combination of the final SAR/optical composite, the images have been classified visually to differentiate landcover classes. The colour code for classification is explained in Table 3.

It is also important to note that, because of additional information in SAR images, clouds are dyed in two different colours (cyan to light green) depending upon the information beneath them. Over sea their colour depends upon the sea roughness (mainly wind driven) and of over land it depends upon land roughness due to high mangrove forest or build up area.

Table 3: Colour code for visual interpretation

CONCLUSION

The complementary use of SAR and optical spaceborne data adds a new dimension to remote sensing applications. Two major improvements have been observed:

- Firstly, because, microwave can penetrate through the clouds and thus provides information even in clouded parts of the scene.

- Secondly, because, microwave are highly sensitive to roughness and di-electric properties of the backscattering surface targets, new geo-physical parameters become available.

Thus, complementary use of SAR and optical spaceborne data enables to differentiate some land-cover/land-use classes which are otherwise, not distinguishable. The classification scheme mentioned in the paper is straight forward to understand and simple to apply, however, more complicated digital classification techniques may improve the results.

REFERENCES

[1] Gunter Schreier (Ed), "SAR Geocoding: Data and Systems", Wichmann, Germany, 1993.

[2] Zhenghao Shi and Ko B. Fung, "A Comparison of Digital Speckle Filters",IEEE pp.2129-2133. 1994

[3] J.W. Rouse, et al., "Monitoring Vegetation Systems in the Great Plains with ERTS", 3rd ERTS Symposium, NASA, Vol. 1, pp 309-317, 1973.

[4] J. Lichtenegger, J.F. Dallemand and P. Reichert, "Multi-Sensor Analysis for Land Use Mapping in Tunisia", Earth Observation Quarterly, No 33, 1991

ACKNOWLEDGEMENTS

One of the authors (Dr. Abdul Raouf) wishes to thank United Nations Office for Outer Space Applications (UNOOSA) for his selection as UN/ESA research fellow for the year 1996. He also wishes to thank ESA for providing ERS data of the study area and excellent facilities which enabled him to complete the project. Thanks are also due to SUPARCO who has granted Landsat data of the study area.

Keywords: ESA European Space Agency - Agence spatiale europeenne, observation de la terre, earth observation, satellite remote sensing, teledetection, geophysique, altimetrie, radar, chimique atmospherique, geophysics, altimetry, radar, atmospheric chemistry