Coastal Wind Field Retrievals from ERS SAR Images

Abstract

A unique series of ERS-1/2 C-band SAR images was obtained off the southern coast of Norway during the COAST WATCH'95 experiment in September 1995. In this paper we carry out a systematic analysis of the mesoscale coastal wind field conditions expressed in the SAR images. Four different categories of phenomena including wind rows, fetch limited seas, wind fronts and oceanographic fronts are examined and discussed. The quantitative retrievals of the wind field are based on examination of the backscatter characteristics and the spectral properties of the SAR image. Results are compared and validated against coincident ship and buoy data providing complementary and independent observations of the oceanographic and meteorological conditions.
Keywords: SAR, Coastal regions, Wind field, Azimuth cut-off, CMOD

1.Introduction

Scatterometer observations over the ocean provide direct estimates of the global wind vector field at spatial resolution of 50 km with an accuracy of 2 m/s in speed, 20° in direction and a directional ambiguity of 180 [Stoffelen and Anderson, 1993]. For some applications, such as in semi-enclosed seas, in straits, in coastal regions, in estuaries, in polynas in ice and along the marginal ice zones, this resolution is, however, too coarse. For monitoring and forecasting in these regions, wind field estimates retrieved from high resolution SAR images can therefore be very useful.

In order to better quantify SAR imaging of upper ocean and atmospheric boundary layer processes, an experiment, COAST WATCH'95 [Johannessen O. M. et al., 1996], was carried out off the southwest coast of Norway during September 1995 as a ERS-1 and ERS-2 tandem announcement of opportunity experiment. The tandem operation provided a unique SAR coverage of the experiment area consisting of 148 SAR images (frames). Meteorological and oceanographical in situ observations were provided from the research vessel R/V Håkon Mosby of University of Bergen, and an advanced metocean buoy from the Naval Postgraduate School, Monterey.

The main study objective reported in this paper is to demonstrate the optimum wind retrieval from SAR images. We will use the procedure outlined in Korsbakken [1996] and Korsbakken and Johannessen [1996]. Section 2 contains a brief review of the two retrieval methods. Application of the methods to the COAST WATCH'95 data is reported in section 3 including a systematic comparison of the separate wind speed estimates from the spectral and radar backscatter properties (CMOD4) of the SAR images as well as a combination of the two methods. In section 4, the main results including the feasibility of the methods are summarized and some future possibilities using new sensors and platforms are outlined.

2. Wind retrieval algorithms

The two SAR wind retrieval algorithms are based on the extraction of the wind field from different parts of the ocean surface wave spectrum, in particular the medium wind wave regime and the small centimeter scale regime. Figure 1 shows a conceptual overview of the wind field estimation as further described below.

Figure 1 Conceptual algorithm for wind field retrievals.

SAR Wind Algorithm (SWA)

In case of a fully developed sea (no fetch limitation) the empirical SWA relation, based on evaluation of 1200 SAR wave-mode imagettes with a central incidence angle of 20.2° is given by Chapron et al. [1995] (see also Kerboal et al., 1996 and Kerboal et al., this issue) as

where U₁₀ is the wind speed at 10 m above the surface and _c is the azimuth cut-off wave length. Assuming a Gaussian shaped low pass filter for the azimuth cut-off the cut-off wavelength can be estimated from the auto covariance function (ACF) derived from an inverse Fourier transform of the SAR image power spectrum.

The CMOD4 model

The CMOD4 wind retrieval model [Stoffelen and Anderson, 1993] is developed for the ERS-1 C-band scatterometer but it is also shown to give good estimates of wind speed when applied to ERS-1 SAR images (e.g. [Vachon et al., 1996], [Wackerman et al., 1996], [Johannessen et al., 1995], [Vachon et al., 1995]). The CMOD4 model provides s₀ values as a function of relative wind direction (=0 for a wind blowing toward the radar), wind speed and incidence angle, expressed as

The coefficients B0, B1 and B2 depend on the local incidence angle of the radar beam and wind speed. The accuracy in the model is 20° in relative wind direction and 2 m/s in wind speed when applied to scatterometer data.

In this work the absolute calibrated ₀ from ERS SAR is derived in accordance with a comprehensive calibration scheme provided by ESA [Laur et al., 1996] except from correction for variance in the replica pulse power. The latter is shown to be negligible using the CMOD4 algorithm for wind speed retrieval in SAR data [Scoon et al., 1995].

In this paper we also emphasize that the wind direction can be estimated from the CMOD4 model for different incidence angles provided the wind speed, derived from the SWA method, can be associated with the corresponding measured radar backscatter (s₀). In such cases we will show that four solutions, i.e. two pairs, each with a 180° ambiguity can be found, except in the cases when the direction is close to upwind (the wind blowing towards the radar) or downwind, for which only one pair is found.

It has also been demonstrated [Johannessen et al., 1995] that wind rows manifested in the SAR images can be used to indicate the near surface wind direction during the SAR integration time. In such cases the number of wind direction solution pairs is also reduced to one (180° ambiguity).

COAST WATCH'95 Analysis

Three case studies from the Coast Watchí95 data base are presented including: a) wind rows under fetch limited conditions; c) wind front; and d) oceanic front (jet) together with a local wind front. A composite overview of these three cases including the SAR image expressions together with the retrieved wind vector maps and a comparison of the derived wind speeds from SWA and CMOD4 is shown in each case as follows:

a) The ERS-1 SAR images are from 17 September 1995 (Figure 2, left) shows a characteristic pattern of wind rows aligned in the wind direction of about 110. In particular, the northernmost area, west of the coast, lacks of expressions features, while the southern areas are recognized with wind rows with an west-northwestward orientation of 110. There is also a gradual increase in the backscatter towards south. The corresponding wind vector map is shown in Figure 2 (center). The wind speed in vicinity of the ship is about 13 m/s, and reveals a clear north west gradient as expected. The obtained wind direction is in good agreement with the isobars from the hindcast model. Wind retrievals from a total of 238 sub-images were compared for the 17 September images. The CMOD4 - SWA wind speed difference was less than 2 m/s for about 90 % of the sub images. The scatter plot in Figure 2 (right) reveals good agreement between the SWA and the CMOD4 derived wind speeds in the range of 3 to 15 m/s for all three images.

b) The SAR images from 23 September 1995 (Figure 3, left) shows a characteristic local wind front appearing as a bright-dark boundary. Wind rows are also present. The corresponding wind vector map shows a northeasterly wind speed gradient and a wind direction of 260-300. The images have moreover been arranged into three areas in which area 1 present the near shore zone, area 2 is immediately north of the wind front and area 3 is out of the front. Weak evidence of wind rows are found in areas 2 and 3. The wind front curves through the image with a width or transition zone of about 15 to 20 km. The largest gradient in ₀ appears to be oriented approximately perpendicular to the wind direction (darkest area in the front as illustrated in Figure 12), while the ₀ gradient becomes weaker as the frontal orientation becomes more closely aligned with the wind direction. A total of 116 sub-images were examined, of which 60 % has a SWA-CMOD4 wind speed difference of less than 2 m/s in the range of winds from 3 to 15 m/s as shown in Figure 3 (right).The change in the CMOD4 wind speed across the front from area 2 to 3 is about 2-5 m/s .

c) The final case study is based on the SAR images from 27 September 1995 (Figure 4, left) which shows a westward flowing coastal jet bounded by two distinct, some places parallel fronts. A local, 5-10 km wide wind feature is also running diagonally across the image, in the second image. The corresponding wind vector map (Figure 4, center) shows a wind speed of about 10 m/s from southwest at about 215. The SAR image is shown to be classified into 5 distinct zones covering zones in vicinity of as well as inside these frontal features. A total of 144 sub images were analyzed for the 27 September image (Figures 5 and 16), of which only 8 % have a wind speed difference of less than 2 m/s in the range from 3 to 15 m/s. From the SAR image power sperctrum (SIPS) analysis a strong peak in the energy occurs between 150 and 300 m wavelength, 30 to 50 to the range axis, in the SIPS. This peak in the SIPS is most likely caused by a corresponding peak in the real ocean wave spectrum suggesting the presence of incoming swell.

Figure 2 Case a). SAR images from 17 September (left) and corresponding wind vector map derived from inverting the CMOD4 model function using the SWA wind speed and the calibrated backscatter (center). Isobars and corresponding surface wind vectors from diagnostics are superimposed on the figure. The position of R/V Håkon Mosby and the ëmetoceaní buoy at satellite overpass is indicated in the map.

CMOD4 wind speed vs. SWA wind speed (left), derived from the SAR images at 17 September 1995. The interval of ± 2 m/s and the in situ measurement (wind speed) from R/V Håkon Mosby and the buoy are indicated. The sub-images containing land are rejected. (Squares is used for the northernmost SAR image, triangles for the center image and diamonds for the southernmost SAR image.)

Figure 3 Case b). SAR images from 23 September (left) and corresponding wind vector map derived from inverting the CMOD4 model function using the SWA wind speed and the calibrated bakscatter (center). Isobars and corresponding surface wind vectors from diagnostics are superimposed on the figure. The position of R/V Håkon Mosby at satellite overpass is indicated in the map. CMOD4 wind speed versus SWA wind speed (left) derived from the SAR images at 23 September 1995 , and classified according to the three areas. The interval of ± 2 m/s and the in situ measurement (wind speed) from R/V Håkon Mosby are indicated.

Figure 4 Case c). SAR images from 27 September (left) and corresponding wind vector map derived from inverting the CMOD4 model function using the SWA wind speed and the calibrated backscatter (right). Isobars and corresponding surface wind vectors from diagnostics are superimposed on the figure. The position of R/V Håkon Mosby at satellite overpass is indicated in the map.

CMOD4 wind speed versus SWA wind speed (low panels) derived from the SAR images at 23 September 1995. The interval of ± 2 m/s and the in situ measurement (wind speed) from R/V Håkon Mosby are indicated.

The swell seems to affect the wind field results, in particular the SWA wind speed and the directional estimates. Due to the large deviation in the SWA and CMOD4 wind speed, the retrievals of wind directions and in turn wind vectors were only possible from the southern parts of the image. A shift of about 1.5 m/s to 2 m/s in the CMOD4 wind speed is observed across the jet and across the wind front between area 4 and 5 as seen in the left plot in the Figure 5 (right). In contrast to CMOD4 wind speeds there is a large spread in the SWA derived wind speed for the areas. However, no evidence of wind shifts are found across the fronts.

Figure 5 (right) shows that the SWA wind speed is up to 10 m/s higher than the CMOD4 in particular for areas 2 and 3. Particularly no correlation is therefore present. In comparison to the previous results, in particular from Case a, the results are poor. The SWA is consistently larger than CMOD4 by at least 4 m/s and completely without any correlation for the conditions encountered in Case c.

Summary

In this paper we have shown that the radar backscatter and spectral signatures of the ocean surface obtained from ERS-1,2 SAR images can provide valuable and quantitative information on near surface wind speed and wind direction. 7 ERS SAR images, from 17, 23 and 27 September 1995 have been examined by studying their ₀ and spectral properties in 498 sub-images. Two different wind retrieval models, SWA and CMOD4, have been applied to the data. We have found by comparing the SWA and CMOD4 wind speed retrievals that about 65 % of the sub images gives a wind speed difference less than 2 m/s.

It is demonstrated that the surface conditions impact on the performances of the different wind field retrieval methods and their corresponding results. The presence of homogenous wind rows are clearly favored. For fetch limited seas and in vicinity of wind fronts, on the other hand, the SWA method underestimates the wind speed. For fetch limited seas the waves are not in equilibrium with the near surface wind speed. Hence, the distribution of the velocity field of surface scatters, as introduced by the orbital velocity of the waves will be narrower than for fully developed seas leading to underestimation. Moreover the relatively large relaxation rate for the longer wind waves (also the medium wavelengths suppressed by the general resolution of the SAR) allows these waves to propagate across a wind front. Hence they will maintain their original equilibrium state over some distance away from the front. In turn, the wind front will not be resolved properly by the SWA method. For the same reason, the presence of swell may cause an increase in the derived wind speed. In contrast the CMOD4 method is not limited by these conditions.

While absolute image calibration is necessary for the CMOD4 method, it is not required for the SWA method since the first method uses radar backscatter values while the latter uses spectral characteristics. But as mentioned in the analysis of the 17 September image, the SWA method is limited to SAR images containing clear wave modulation from which the azimuth cut-off can be derived. In the case of very low wind conditions (lower than approximately 3 m/s) or in presence of slicks the SWA method will therefore breakdown. But these conditions will also effect the CMOD4 since the threshold wind speed for C-band waves is around 3 m/s [Johannessen et al., 1996].

Hence, by combining the SWA and CMOD4 methods, the limitations encountered by each individually, can be suppressed. Moreover, this combination allows the wind direction to be retrieved independently thus offering a system for quantitative estimation of wind speed and direction at a resolution of about 10 km.

Further investigation of these methods is necessary in order to fully understand their limitations and strengths, in particularly in regards to the surface conditions and synoptic weather situation. However, as reported in this and other papers on wind retrievals from SAR the possibility looks promising in regards to the continuation of regular spaceborne SAR observations.

For the different ERS-1,2 SAR operating modes only the low resolution images constrain the SWA method since typical wavelengths of wind seas are not properly resolved in such images. However, since the ERS satellites carry a wind scatterometer which deliver global vector wind field the situation is fairly good. For the ESA ENVISAT-1 satellite, to be launched in 1999, on the other hand, no scatterometer is included. Hence the quality of the wind retrieval algorithms, as described in this paper, will become more important. In table 1 some performance characteristics for the ERS-1,2 modes and the proposed ENVISAT ASAR (Advanced Synthetic Aperture Radar) modes are listed suggesting which modes of ASAR operation are favorable for wind retrievals. It is clear that they closely correspond to those modes wanted for wave retrievals. In the ENVISAT time frame the aim is thus to systematically provide optimum wind and wave retrievals by a combination of wind field retrieval models as presented in this paper, and the directional ocean wave spectrum modeling.

Table 1 Qualitative performance characteristics of the SAW and CMOD4 methods.

Acnowledgements

This work was founded by the Norwegian Research Council as a part of a strategic SAR program at the Nansen Environmental and Remote Sensing Center. The SAR data was provided under the ESA AO2/N108 (announcement of opportunity) program. The ship time for the COAST WATCH'95 experiment was provided by the University of Bergen and funded Following contributions are highly appreciated; K. Davidson and Paul Fredrickson at the Naval Postgraduate School for providing the buoy measurements and M. Reistad at the Norwegian Meteorological Institute for providing the weather diagnostics.

References

Chapron B., Fouhaily T. E. and Kerbaol V. 'Calibration and Validation of ERS Wave Mode Products' IFREMER March 95 Document DRO/OS/95-02, 1995.

Johannessen J.A., Vachon P.W. and Johannessen O.M. 'ERS-1 SAR imaging of marine boundary layer processes' Earth Observation Quarterly, ESA, 1995.

Johannessen O. M., Johannessen J. A., Jenkins A. D., Davidson K., Lyzenga D. R., Shuchman R., Samuel P., Espedal H. A., Knulst J., Dano E. and Reistad M., 'COAST WATCH-95 ERS-1,2 SAR Applications of Mesoscale Upper Ocean and Atmospheric Boundary Layer Processes off the coast of Norway', Proceedings of IGARSS'96 Lincoln, Nebraska, 1996.

Kerbaol V., Chapron B., El Fouhaily T., Garello R., Fetch and Wind Depence of SAR azimuth Cutoff and Higher Order Statistics in a Mistral Case. In proceedings of IGARSS'96, Lincoln, Nebraska, USA, 1996.

Korsbakken E. 'Quantitative Wind Field Retrievals from ERS SAR Images' YGT report. Available at the Ocean and Sea Ice Unit, Earth Sciences Division, ESTEC, ESA, 1996 a.

Korsbakken E and Johannessen J. A. 'Quantitative Wind Field Retrievals from ERS SAR Images' in proceedings of the Third ERS Workshop IFREMER/BREST, June 18-20, 1996 b

Laur H., Bally P., Meadows P., Sanchez J., Schaettler B., Lopinto E., Derivation of the backscattering coefficient 0 in ESA ERS SAR PRI products', document no: ES-TN-RS-PM-HL09, Issue2, Rev. 2, ESA ESRIN, 28 June 1996.

Scoon A., Robinson T. S. and Meadows P. J., 'Demonstration of an improved calibration scheme for ERS-1 SAR imagery using a scatterometer wind model' Int. journal of Remote Sensing, vol.17, no.2, 413-418, 1996.

Stoffelen A. and Anderson D.L.T. 'ERS-1 Scatterometer Data and Characteristics and Wind Retrieval Skills' Proceeding of first ERS-1 Symposium, ESA SP-359, March, 1993.

Vachon P.W., Johannessen J.A and Browne D