OCEAN SURFACE DRIFT BY WAVELET TRACKING USING ERS-2 AND ENVISAT SAR IMAGES Antony K. Liu, Yunhe Zhao Ocean Sciences Branch, NASA Goddard Space Flight Center, Greenbelt, Maryland, USA Ming-Kuang Hsu Northern Taiwan Institute of Science and Technology, Taipei, TAIWAN Historically, ocean surface feature tracking analyses have been based on data from a single orbital sensor collected over its revisit interval of a lone, low-earth orbital satellite. Today, ocean surface currents are being derived by performing feature tracking using data from the same type of sensors on different satellites. With allweather, day/night imaging capability, Synthetic Aperture Radar (SAR) penetrates clouds, smoke, haze, and darkness to acquire high quality images of the Earth s surface. The ability of a SAR to provide valuable information on the type, condition, and motion of the sea-ice, ships and surface signatures of swells, wind fronts, oil slicks, and eddies has been amply demonstrated [Liu and Wu, 2001]. This makes SAR the frequent sensor of choice for cloudy coastal regions. At present, there are three major synthetic aperture radars in orbit. Radarsat-1, the first Canadian remote sensing satellite, was launched in November 1995. Radarsat-1 has a ScanSAR mode with a 500 km wide swath and a 100 m resolution. The ERS-2, having a conventional SAR with a swath of 100 km and a resolution of 25 m, was launched in April 1995 by ESA. Envisat-1 with an Advanced SAR (either conventional narrow swath or wide swath of 405 km with 150 m resolution) was also launched in March 2002 by ESA. With repeated coverage, spaceborne SAR instruments provide the most efficient means to monitor and study the changes in important elements of the marine environment. Due to high-resolution of SAR data, the coverage of SAR sensor is always limited, especially for a repeat cycle. With more SAR sensors from various satellites, new data products such as ocean drift can be derived when two SARs tracks overlap in a short time over coastal areas. Currently, there are two SAR sensors on different satellites, ERS-2 and ENVISAT, having acquisition time offset around 30 minutes with almost the exactly same path. That is, ERS-2 is following Envisat with a 30-minutes delay, which will be a good timing for ocean mesosclae feature tracking and new product such as ocean surface drift. Ocean surface backscattering images provided by ERS-2 SAR and Envisat ASAR can be used to derive ocean surface drift. SAR data from Envisat and ERS-2 were collected on April 27, 2005 at 01:54 and 02:22 GMT, respectively, over the southern part of Luzon Strait near Philippines. Figure 1 shows the ERS-2 60 km x 80 km SAR image obtained on April 27, 2005, north of Philippines in the Luzon Strait, and the location map with the SAR image coverage area shown in the big box for reference. A chain of islands in the Luzon Strait can be easily identified in the SAR image. Also, many eddies, oil slick, wave refraction, and fronts around these islands are clearly observed in SAR image as the mesoscale surface features. For further detailed study, a zoomed SAR subscene has been selected from each image. Figure 2 shows the overlaid of these two SAR subscenes of ERS-2 in green and Envisat in red. The central location of these subscenes is 21,15 o N and 121.68 o E, and the size is approximately 28.2 km x 28.2 km. The subscene coverage area is shown in the small green box in Figure 1. The major oceanographic feature, a long oil slick oriented in north-south direction, can be clearly identified. The phase shift of this oil slick in 28 minutes shows the surface drift pattern due to the advection of surface current. To validate the results, wind data from QuikSCAT are compared with the satellite-derived flow field. The qualitative comparison shows a generally consistent pattern. WAVELET ANALYSIS OF SATELLITE IMAGES A two-dimensional wavelet transform is a highly efficient band pass-filter. The two-dimensional Gaussian wavelet (often referred to as a "Mexican hat" wavelet) has been applied to satellite images to separate processes at various scales, including relative phase/location information for coastal monitoring applications [Liu et al., Proceedings of SEASAR 2006, 23-26 January 2006, Frascati, Italy, (ESA SP-613, April 2006)
1997a], and for ice edge and ice floe tracking [Liu et al., 1997b]. The wavelet transform for small-scale features in a satellite image can be used as a band-pass filter, with transforms of various length scales, to separate texture or features; for near real-time "quick look" analyses of satellite data for feature detection; and for data reduction using a binary image. Wavelet analysis of NSCAT (NASA Scatterometer), QuikSCAT backscatter and SSM/I (Special Sensor Microwave/Imager), AMSR (Advanced Microwave Scanning Radiometer) radiation data can also be used to obtain daily sea-ice drift information for both the northern and southern polar regions [Liu et al., 1998; 1999; Zhao et al., 1998; 2002; Liu and Cavalieri, 1998]. Overall, the comparison of scatterometers and radiometers-derived ice motion with Arctic buoy data shows good agreement. Chlorophyll a concentration images provided by MODIS (Moderate Resolution Imaging Spectroradiometer) and SeaWiFS (Moderate Resolution Imaging Spectroradiometer) can also be used to derive surface layer drift in a large scale [Liu et al., 2002]. The surface layer drift has been derived by wavelet tracking, and major oceanographic features, such as the Gulf Stream boundary and a large cold-core cyclonic eddy south of the Gulf Stream, have been clearly identified. To validate the drift results, data from several drifter buoys were compared with the satellite-derived flow field. The qualitative comparison showed a generally consistent pattern over the east coast of the United States. When using multiple SAR data from different satellites to track ocean feature motion, the first step is to transform the full-resolution images to the same map projection. In this case, the ERS-2 SAR and Envisat ASAR subscenes used are both 512 x 512 pixels (with pixel size of 55 m approximately), and the Mexican-hat wavelet transform is applied to filter each image with several length scales. The length scale of the wavelet transform corresponds to the length scale of the Gaussian function and is based on the length scale of the feature of interest. Filtered images, acquired 28 minutes apart, are then examined to find matching features using templates, which are then readily converted to motion vectors and averaged onto a 0.88 km x 0.88 km grid. The choice of 0.88 km as the matching template size corresponds to 16 pixels of satellite re-mapped data. The accuracy of this technique is only limited by the persistence of the features and by the spatial resolution and navigational accuracy of satellite data. A single pixel feature displaced 55 m over 28 minutes will have a maximum velocity uncertainty of 3.3 cm/s due to sensor resolution. The geolocation uncertainty is about 25 m (sensor resolution) for ERS-2 SAR and Envisat narrow swath ASAR. Figure 3 shows the surface drift (green arrows) derived by wavelet analysis of feature tracking from ERS-2 SAR and Envisat ASAR surface roughness backscattering data. A re-mapped Envisat image appears as background to highlight oil slick. The data were collected on April 27, 2005 over the southern part of Luzon Strait near Philippines, separated by 28 minutes. As shown in this figure, the oil slick motion of 1.2 m/s by surface current advection has been well derived and can be clearly identified. Notice that the converging area at the top showing a kink on the oil slick. Also, the shear zone in the middle dilutes and bends the slick near the bottom. Furthermore, on the right-hand side of slick, a small eddy of 10 km size can be identified from their cyclonic circulation flow pattern. The areas lacking drift vectors in the map indicate the regions where filtered features were not matched. COMPARISON WITH WIND DATA The SeaWinds instrument on the QuikSCAT satellite is a specialized microwave radar that measures near-surface wind speed and direction under all weather and cloud conditions over Earth's oceans. Currently, the empirically derived model function has been used to relate normalized radar cross-section with wind speed and direction. The wind map has been distributed with a 25 km grid over the ocean surface. In this study geographical location, three wind vectors have been identified as red arrows in Figure 3. The wind speeds and directions for the three wind vectors in Figure 3 from left to right are listed as follows: (a) 5.74 m/s, 198.5 degree; (b) 5.18m/s, 193 degree; and (c). 3.4 m/s, 297.27 degree, where the wind direction of 0 degree implies a flow toward the north. So, the wind vectors (a) and (b) having a speed of 5 to 6 m/s are coming approximately from the North and that is probably why the oil slick is more or less north-south oriented originally. The wind on the right-hand side is relatively weak (3.4 m/s) and coming approximately from the East. Although the wind data are very limited, the comparison shows a qualitatively consistent pattern between wind data and SAR observation, especially for the oil slick feature. As shown in Figure 3, the oil slick motion of 1.2 m/s in maximum by surface current advection has been well derived and can be clearly identified. Notice that the converging area at the top showing a kink on the oil slick.
Also, the shear zone in the middle dilutes and bends the slick near the bottom. Furthermore, on the right-hand side of slick, a small eddy of 10 km size can be identified from their cyclonic circulation. These results indicate that multiple SAR images overlapped in a short time can be used to derive ocean surface drift, and can help to identify oceanic processes such as currents and eddies. Since there are no in-situ measurements or drifter buoy data in this Philippines coast water, further validation and calibration are definitely warranted for future study. MYSTERY SHIP NEAR EDDY Ship and their wakes can be detected in the high-resolution SAR imagery provided by satellites. In general, ship is a very effective corner reflector, so ship can be easily observed as a bright spot in the SAR image. But, occasionally, the ship in the SAR image remains invisible, and only trailing dark turbulent wakes are seen [Liu et al., 1996]. Figure 4 shows Envisat and ERS-2 28 km x 28 km SAR subscenes obtained on April 27, 2005 north of Philippines in the Luzon Strait. The images cover the area from longitude of 20.61 degree to 20.86 degree, and latitude of 22.09 degree to 122.34 degree. The invisible ship and its wake in the boxes near the eddy can be tracked easily in these figures. Then, the ship speed is estimated from the distance between ship locations in each SAR image and SAR acquisition time interval (28 minutes) to be 5.94 m/s. Very low backscattering of the ship configuration may have hidden the invisible ship from view, or the wake could has been formed, instead, by an underwater vehicle. In this area covered by SAR imaging, many internal tide and nonlinear internal wave packets have been observed in the SAR images [Liu et al., 1998]. Based on satellite observations of internal wave distribution from the last ten years, most of internal waves in the northeast part of South China Sea are propagating westward. The wave crest can be as long as 200 km with amplitude of 150 m, due to strong current from the Kuroshio branching out into the South China Sea. These huge internal waves may be generated in many channels between islands in the Luzon Strait as shown in Figure 1. The surface drift pattern derived from satellites can be very useful to study the sources of internal wave generation area. ACKNOWLEDGMENTS The authors would like to thank Wolfgang Lengert of ESRIN in ESA for encouragement of this research. This work is supported by the Office of Naval Research and Taiwan s National Science Council. All ERS-2 SAR and Envisat ASAR data are copyrighted by ESA. REFERENCES 1. Liu, A. K., C. Y. Peng, and Y.-S. Chang, Mystery ship detected in SAR image, EOS, Trans. AGU, 77, 17-18, 1996. 2. Liu, A. K., C. Y. Peng, and S. Y.-S. Chang, Wavelet analysis of satellite images for coastal watch, IEEE J. Oceanic Eng., 22, 9-17, 1997a. 3. Liu, A. K., S. Martin, and R. Kwok, Tracking of ice edge and ice floes by wavelet analysis of SAR images, J. Atmos. Oceanic Technol., 14, 1187-1198, 1997b. 4. Liu, A. K., and D. J. Cavalieri, Sea-ice drift from wavelet analysis of DMSP SSM/I data, Int. J. Remote Sens., 19, 1415-1423, 1998. 5. Liu, A. K., Y. Zhao, and S. Y. Wu, Arctic sea ice drift from wavelet analysis of NSCAT and special sensor microwave imager data, J. Geophys. Res., 104, 11529-11538, 1999. 6. Liu, A. K., S. Y. Chang, M.-K. Hsu, and N. K. Liang, Evolution of nonlinear internal waves in East and South China Seas, J. Geophys. Res., 103, 7995-8008, 1998. 7. Liu, A. K., Y. Zhao, and W. T. Liu, Sea-ice motion derived from satellite agrees with buoy observations, Eos Trans. AGU, 79, 353-359, 1998. 8. Liu, A. K., and S. Y. Wu, Satellite remote sensing: SAR, Encyclopedia of Ocean Sciences, London: Academic Press, Edited by J. H. Steele, S. A. Thorpe, and K.K Turekian, 5, 2563-2573, 2001. 9. Liu, A. K., Y. Zhao, W. E. Esaias, J. W. Campbell, and T. Moore, Ocean surface layer drift revealed by satellite data, EOS, Trans. AGU, 83, 61-64, 2002. 10. Zhao, Y., A. K. Liu, C. A. Geiger, Arctic sea ice motion from wavelet analysis of SSM/I data, J. Adv. Mar. Sci. Soci., 4, 313-322, 1998. 11. Zhao, Y., and A. K. Liu, Validation of sea ice motion from QuikSCAT with those from SSM/I and buoy, IEEE Trans. Geosci. Remote Sens., 40, 1241-1246, 2002.
Fig. 1. (a) ERS-2 60 km x 80 km SAR image (copyright ESA 2005) obtained on April 27, 2005, north of Philippines in the Luzon Strait, and (b) the location map with the SAR image coverage area shown in the big box.
Fig. 2. Overlaid of two SAR subscenes collected over the Luzon Strait near Philippines from ERS-2 (in green), and Envisat (in red) on April 27, 2005 separated by 28 minutes. The distance units are in pixels (size of 55 m) for both horizontal and vertical coordinates. The subscene coverage area is shown in the small green box in Fig. 1. Fig. 3. Ocean surface drift (green arrows) derived from ERS-2 and Envisat SAR data over the Luzon Strait (Envisat image as background). The surface drift unit of 1 m/s is indicated by a white arrow at the top. The QuikSCAT wind data are shown as red arrows.
Fig. 4. Envisat and ERS-2 28 km x 28 km SAR subscenes (copyright ESA 2005) obtained on April 27, 2005 north of Philippines in the Luzon Strait. The invisible ship and its wake near the eddy can be tracked easily.