Marco van der Kooij , William Hughes and Shinya Sato

Atlantis Scientific Inc, 20 Colonnade Road, Suite 110, Nepean, Ontario, Canada K2E 7M6, Phone: 613-727-1087,

Fax 613-727-5853, e-mail: marco.van_der_kooij@atlantis-scientific.com

Ocean current velocity information is perhaps the single most important type of information required for many applications in the ocean environment. This information is very difficult to collect synoptically and no spaceborne remote sensing technique is currently available that allows robust extraction of quantitative velocity information at spatial resolution scales of the order of 10 km or smaller. This article shows examples of validation and demonstration of a methodology to extract Doppler information from raw SAR imagery. This information allows the generation of current velocity maps at spatial resolutions in the order of 1-2 km and a precision of approximately 0.2-0.3 m/s. Two test areas were selected for this purpose, the Gulf Stream near Cape Hatteras and the northern section of the Gulf of Mexico. The results show that the quality of the measurements is interesting and that there is potential for robust generation of synoptic current velocity information from space. It is expected that this type of information will particularly be interesting in the application of ScanSAR data.

During the processing of raw SAR data the center frequency of the backscattered radar radiation is used to determine the azimuth pointing angle. The spectrum is quite wide due to the variations of the frequency during the integration time of the SAR. It is well-known that the center frequency (Doppler Centroid, D.C.) can be still be estimated with very good precision in particular for relatively homogeneous imagery such as in ocean environments (Bamler, 1991, Jin, 1986). Bamler and Jin reported D.C. standard deviations of approximately 5 Hz for homogeneous imagery using patches of 64 (range) by 256 (azimuth) samples, which typically corresponds to areas of approximately 1 x 1 km on the ground. Previous work (Van der Kooij and Hughes, 1997) involved implementation and optimisation of in-house D.C. estimation algorithms. That study confirmed the potential accuracy of D.C. estimates. The question that formed the basis of the work presented here is whether it is possible to measure the additional frequency shift caused by ocean currents and what the potential and limitations of these measurements are.

The geometry of the Doppler frequency shift is shown in figure 1. The satellite with SAR antenna moves in a direction perpendicular to the paper, the radar incidence angle is q . The Doppler frequency shift D f caused by the current depends on the radar wavelength, the incidence angle and the line-of-sight component of the velocity towards the radar track. Typical Doppler shifts for a 1 m/s current are in the order of 14 Hz for ERS SAR data and approx. 28 Hz for Radarsat at an incidence angle of 45° .

Shuchman (1979) first proposed the idea of using Doppler information for this purpose and investigated it using SEASAT data. The measurements were not considered successful. The reasons for this lack of success are

1. Doppler Centroid estimation technology at that time was not as advanced as it is today.
2. SEASAT had an L-band frequency SAR system. which gives small Doppler shifts caused by current velocity. Some currently available Spaceborne SAR sensors provide data at C-band frequency. This frequency provides larger Doppler shifts.
3. Application might have been more successful if applied for large-scale ocean currents.

Figure 1 Geometry of SAR Doppler-derived current velocity measurements. V_l.o.s.is the SAR line of sight component of the current velocity vector.

Along-track SAR interferometry is a technology that is widely regarded as a promising technique for high resolution current velocity measurements (Goldstein and Zebker, 1987, Graber et. Al. 1996, Campbell et. Al, 1997). The technique requires the use of two antenna phase centers on one platform and is currently only available on airborne platforms. The technique is very similar to the Doppler measurement technique described in this paper in that it measures a phase shift between the antenna phase centers that is a direct measure of the frequency shift caused by ocean currents. The spatial resolution for along track interferometry is higher because it allows measurement at each resolution cell of the SAR data. Graber et al. (1996) showed that the along-track InSAR technique is accurate but that the measurements have to be corrected for (airborne) platform navigation effects and that phase trends due to capillary and small gravity waves and wind-drift effects have to be modelled and removed. They used a wave model (Thompson (1989)) and obtained accuracies in the order of 0.2 m/s after comparison with HF (shore-based) radar measurements. The accuracy was improved to approx. 0.1 m/s by calibrating the data with one externally supplied velocity reference measurement.

The frequency shift technique described in this paper uses a "patch" of data for an area of approximately 1 x 1 km. The advantage of this technique, as opposed to along-track SAR interferometry, is that it requires "regular" SAR data from a one antenna spaceborne platform. The extraction and use of velocity information using state of the art Doppler estimation technology, C-band spaceborne SAR systems and, in particular, application for medium to large scale ocean currents was considered promising and forms the topic of this work.

The purpose of the processing methodology is to create a 2 dimensional array of Doppler centroid estimates using all information available in the SAR data. The following steps can be identified in the process of obtaining a grid (image) of calibrated velocity estimates:

1. Generation of a Single Look Complex (SLC) image.
2. Calculation of a grid of Doppler Centroid estimates for (overlapping) patches of typically 64 (range) by 512 (azimuth) samples. D.C. algorithms used were Madsen (Sine Weighting) and Optimum (non-uniform weighting). Both D.C. estimation algorithms effectively fit a weighted antenna pattern to the observed frequency spectrum where the Sine Weighting is optimum for high signal-to-noise ratios and Non-Uniform weighting is optimum for low signal-to-noise ratios (see Bamler, 1991)).
3. Removal of a model of the Doppler Centroid for non-moving terrain. In this case a simple 2^nd order (range) and 1^st order (azimuth) polynomial model was fitted to the D.C. measurements and subtracted. This approach has its limitations but was considered satisfactory for demonstration and validation at this stage.
4. Conversion of residual D.C. estimates to current velocity values using geometry information and the equation shown in Error! Reference source not found..
5. Additional calibration by using externally supplied reference velocity measurements (e.g. from ship measurements). The component perpendicular to the satellite track of these reference measurements was compared to the SAR derived velocity estimates. A bi-linear trend was fitted to the residuals between SAR and reference measurements. This trend was removed from the SAR derived velocities.

The first test area is in the Gulf stream near Cape Hatteras (North Carolina, U.S.A.). ERS-1 data was collected during the ONR-sponsored High-Resolution Remote Sensing Experiment (High-Res). The experiment involved John Hopkins University and W.H.O.I. (Woods Hole Oceanographic Institute). W.H.O.I. collected A.D.C.P. (Acoustic Doppler Current Profiler) measurements aboard the Research Vessel Iselin. These measurements were collected during or close to the satellite passes at locations close to edge of the Gulf Stream. A.D.C.P. profiles were available for five ERS-1 passes.

Frames from 2 ERS-1 tracks were selected for presentation in this article. The selected frames coincided with the location of the A.D.C.P. measurements. Figure 2 and figure 3 show panels presenting data collected on June 11 and June 14, 1993, respectively. The images in the top left are the radar backscatter intensity images. The top right images show the radar derived velocity image where the intensity is a measure for the velocity perpendicular to the satellite track, which runs approximately from South to North. The range of velocities runs from white (2.5 m/s towards the East) to black (velocity of up to 1 m/s to the west). The red dots and arrows correspond to the location and velocity of ship based current velocity measurements. It is possible to compare both types of measurements by calculating the component of the ship velocity measurements perpendicular to the satellite track. This comparison is visualized in the plot at the bottom right of figure 2 and figure 3.

Figure 2 SAR imagery and Doppler derived current velocity information compared with ship based velocity measurements. The top left image shows an ERS SAR image collected on June 11, 1993. The top right image shows the current velocity

Figure 3 SAR imagery and Doppler derived current velocity information compared with ship based velocity measurements. The top left image shows an ERS SAR image collected on June 14, 1993. The top right image shows the current velocity

The final grid of radar derived velocities was calibrated by removing the offset between radar derived and ship based velocity estimates as described in the processing methodology section.

The comparisons with ship based measurements are favourable. Figure 4 shows a scatterplot that includes measurements obtained from ERS-1 data collected on June 20, 1993. The ERS-1 data collected on June 17, 1993 was rejected because of a low signal to noise level. This resulted in poor D.C. and thus poor velocity estimates. The accuracy appears to be as good as 0.2-0.3 m/s. The comparison suggests that the spatial variations of the current velocity are represented properly in the radar derived measurements. Figure 2 and figure 3 show that the ship measurements were performed close to the edge just North of the Gulf Stream. Several lineaments and current velocity features are visible.

Figure 4 Scatterplot showing the A.D.C.P. (ship) derived current velocities versus the SAR Doppler derived velocity estimates after offset calibration.

The 2^nd test area is in the northern part of the Gulf of Mexico. The off-shore industry in the Gulf of Mexico has an operational need for current velocity information due to the occurrence of the Gulf Stream Loop Current and large energetic eddies. The associated currents can be as large as 1-2 m/s and can have significant impact on oil production activities. The information currently used to detect and predict current velocity variability is derived from AVHRR (ocean temperature), satellite altimeter data and buoys. The AVHRR imagery is generally used as a qualitative tool to understand dynamics but can not be used during the summertime when the temperature contrast in the Gulf of Mexico is low. The disadvantage of the satellite altimeter data is that the spatial resolution is low (~100 km) and the measured velocities are often underestimated. The buoys provide accurate velocity measurements but provide a limited overview and are expensive to use, in particular far from the coast. For validation purposes access was provided to so-called Eddy Watch reports. These reports are produced by Horizon Marine Inc and they are operational products that combine an interpretation of AVHRR, satellite altimeter data and buoys.

It was decided to use RADARSAT data for the demonstration in the Gulf of Mexico. It was considered preferable to use moderate incidence angles for beam modes to be selected. A larger incidence angle provides a high sensitivity (larger Doppler shifts) but selection of a very large incidence angle can cause a very low signal to noise ratio which results in a quality breakdown of the Doppler Centroid estimation. It was decided to use RADARSAT S4 and W2 beam modes.

Figure 5 and figure 6 show strips of RADARSAT S4 imagery collected on March 14, 1998 and May 13, 1998, respectively. The left side shows the SAR imagery with an overlay of the Eddy Watch report. The location of the Fourchon Eddy is visualised as thick lines, the location and 1-week movement of buoys is visualised as thin lines. The right side of this figure shows the Doppler derived current velocities. The velocities are visualised with a grey-scale running from -1 m/s (black, to the West) to +2 m/s (white, to the East). The location of the Fourchon Eddy and associated velocity patterns can be detected and measured from the Doppler-derived velocity imagery. A front visible in the SAR image of figure 5 (left side)at approx. 28° N appears to have no major current velocity change (right side). On the other hand, current velocity changes at the Fourchon Eddy front (26° -27° N) show up very clearly in the Doppler derived velocity image and expressions are also visible in the SAR image (left side).

Figure 5 Strip of RADARSAT S4 imagery (left) and Doppler derived velocity (right). The data was collected on March 14, 1998. An overlay of the Eddy Watch report is shown as visual reference. Fat lines show the location of the loop current estimated from AVHRR data. Thin lines correspond to the movement of buoys. The scaling of the velocity (right) runs from -1 m/s (to the west, black) to +2 m/s (to the East, white).

Figure 6 Strip of RADARSAT S4 imagery (left) and Doppler derived velocity (right). The data was collected on May 13, 1998. An overlay of the Eddy Watch report is shown as visual reference. Thin lines correspond to the movement of buoys. The scaling of the velocity runs from -1 m/s (to the west, black) to +2 m/s (to the East, white).

Figure 7 shows a subsection of Doppler velocity image of figure 6. The vectors derived from buoy data are shown as reference. The visual similarity is interesting. The standard deviation of the differences between buoy and SAR derived velocities appeared to be below 0.2 m/s.

Figure 7 Subsection of Doppler velocity image of figure 6. The vectors derived from buoy data are shown as reference. The visual agreement is striking. The standard deviation of the differences between buoy and SAR derived velocities appeared to be below 0.2 m/s.

It appears that there is useful information in Doppler information derived from spaceborne SAR imagery. Although the resolution is considered low in conventional SAR/radar standards, it could be sufficient as a source of quantitative remotely sensed information at spatial scales of 1-50 km. First quantitative results for ERS and RADARSAT data show that the accuracy of the velocity measurements could be in the order of 0.2 m/s after calibration of a velocity offset using points with known current velocities. We believe that the results presented in this paper provide the first qualitative and quantitative demonstration that Doppler measurements from spaceborne SAR systems can be used to construct current velocity maps. These maps could be very useful for many oceanographic applications. An example of potential application is for the measurement and detection of the Loop Current, eddies and other current velocity features in the Gulf of Mexico. The technique could be used in conjunction with other sources of (qualitative) current velocity information.

A trade-off has to be made in the selection of the incidence angle to be used. A larger incidence angle provides a larger Doppler shift (and thus more sensitivity for velocity) but increases the risk of Doppler Centroid estimation problems due to a low signal to noise ratio. Future SAR systems such as Radarsat-2 (to be launched in 2002) and Envisat (to be launched in 2000) have a vertically polarized mode with a higher signal to noise ratio for ocean surface at large incidence angles. It is expected that future SAR imagery will be more affordable (e.g. Envisat). All this will be beneficial for the application and use of this technique.

Application of this technique looks particularly promising for ScanSAR modes of operation. The wide area coverage (300-500 km) makes it suitable for fully covering medium to large scale ocean current features such as the Gulf of Mexico Loop Current.

It is recommended that the use of a wave model is evaluated to take into account the frequency shifts caused by wind-driven capillary and gravity waves. It is also recommended to consider the creation of a more refined frequency shift model for non-moving surface. Therefore it is necessary to use all aspects of known SAR geometry and use a limited number of D.C. measurements from land surfaces. More thorough and large scale validation efforts are required to determine the full potential and limitations of this technique.

Don Thompson (John Hopkins University-Applied Physics Laboratory) is gratefully acknowledged for helping to provide the ADCP profiles and for stimulating discussions. Jim Edson (Woods Hole Oceanographic Institute) processed the ADCP data. The "Eddy Watch" reports were produced by Jim Feeney of Marine Horizon. The reports were provided to us by Michael Vogel (Shell Oil). He is acknowledged for providing insight into operational requirements for the off-shore industry. Paris Vachon (Canada Center for Remote Sensing) is acknowledged for his support and for providing ERS data. The Radarsat data and the analysis for the Gulf of Mexico were funded through a contract with CSA (Canadian Space Agency). Part of the analysis of the ERS data was funded through a contract with ESA-Esrin. The ERS imagery is © ESA, the RADARSAT imagery is © CSA.

Bamler R. "Doppler Frequency Estimation and the Cramer-Rao Bound", IEEE Trans. Geo and Remote Sensing, V. 29 , May 1991, pp 385-340

Campbell, J.W.M, Gray A.L, Mattar, K.E. Clarke, J.H and Van der Kooij, M.W.A, Ocean Surface Feature Detection with the CCRS Along-Track InSAR, Canadian Journal of Remote Sensing, Vol. 23, No1, March 1997.

Goldstein, R.M, and Zebker, H.A, Interferometric radar measurements of ocean surface currents, Nature, 328, 707-709, 1987.

Graber H.C, Thompson D.R and Carande R.E, Ocean Surface Features and Currents Measured with Synthetic Aperture Radar Interferometry and HF Radar, Journal of Geophysical Research 101, C11, p.25,813-25,832, 1996

Jin M. Y., "Optimal Doppler Centroid Estimation for SAR Data from a Quasi-Homogenous Source", IEEE Trans. Geo and Remote Sensing, V GE-24, November 1986, pp 1022-1025

Shuchman, R, The feasibility of Measurement of Ocean Current Detection Using SAR data, Proc. of the 13th Int. Symp on Remote Sensing of the Environment, Ann Arbor, pp 93-103, 1979)

Thompson, D.R, Doppler spectra from the ocean surface with a time-dependent composite model, Radar Scattering from Modulated Wind Waves, edited by G.J. Komen and W.A. Oost, pp 27-40, Kluwer Acad, Norwell, Mass., 1989.

Van der Kooij, M.W.A and Hughes, W, Measurement of Low Resolution Current Velocity Using RADARSAT SAR Data, Atlantis Scientific report K4502, September 1, 1997