Wind Fields from ERS SAR, Compared with a Mesoscale Atmospheric Model Near to the Coast

Abstract

The synthetic aperture radar (SAR) aboard the European remote sensing satellites (ERS-1 and ERS-2) aquire images which are used in this study to derive ocean surface winds. The wind speed is computed by the C-band model (CMOD4), which was developed for the wind scatterometer (SCAT) operating at the same wave length. Wind fields from ERS-1/2 SAR images are presented for several geographical locations and compared to ground truth measurements. The ERS-1 SAR image of the island Rügen in the Baltic Sea is compared to results of the mesoscale wind model GESIMA. Errors due to uncertainty in wind direction are computed for various wind conditions. The error in wind speed caused by saturation of the analogue to digital convertor leading to a power loss of the ERS SAR data are estimated. Problems of the recalibration algorithm in case of inhomogeneous images, for instance in coastal zones, are presented.

This work presents first results of the ESA SARPAK project (AO2D113).

Keywords: calibration, coastal oceanography, power loss, SAR, surface wind, wind models

Introduction

For several years it has been possible to measure wind fields over the ocean with the wind scatterometer (SCAT) aboard the European Remote Sensing Satellites (ERS-1 and ERS-2). The ERS synthetic aperture radar (SAR) is an instrument operating at the same wave length as ERS SCAT and therefore can use the same models for wind derivation. The SCAT can access wind direction in addition to the velocities by utilizing its three antennas looking from different angles and subsequences at the same area. In most SAR images wind streaks or shadowing behind coasts are visible, from which the wind direction can be computed. Due to the much higher resolution (25 X 25 m² compared to 45 X 45 km² of the SCAT), the SAR images open the opportunity to derive mesoscale wind fields, which has not been possible before.

The object of investigation is to derive mesoscale wind fields over the ocean surface from the normalized radar cross section (NRCS) of ERS SAR images and compare these results to model and ground truth data (Lehner et al., 1996). Wind speeds from ERS SAR images have been derived e.g. by Alpers and Brümmer, 1994, Chapron et al., 1994, Rosenthal et al., 1995, Johannessen et al., 1996 and Scoon et al., 1996. In these articles either recalibration is not considered or extensive comparisons to models and ground truth is not given.

For moderate incidence angles between 20^o and 60^o the backscatter from the rough ocean surface is primarily explained by resonant Bragg scattering (Wright, 1966). The backscatter signal is caused by the water wave component which is in resonance with the incidence radiation. The resonant water wave number k_w is related to the electromagnetic wave number k_el of the radar according to

k_w = 2k_el sin alpha,

where alpha is the incidence angle of the radar beam. In case of ERS SAR with incidence angles between 20^o and 26^o the range of scattering wavelengths extends from 8.2 cm to 6.5 cm. Therefore, the NRCS can be used to evaluate the wind speed, which strongly influences the small scale roughness.

For derivation of wind speeds from ERS SAR images the empiric C-band model CMOD4 developed for ERS SCAT (Stoffelen and Anderson, 1993) is used. As input for computation the incidence angle, wind direction and the NRCS is needed. Therefore the SAR images have to be accurately calibrated. Unfortunately it became evident that the ERS SAR data are not properly calibrated (Meadows and Wright, 1994). In case of high wind speeds or in the near range of the SAR image the NRCS can be up to 5 dB lower than expected. This phenomenon is assigned to the receiver gain settings. It causes a too high input power into the analog to digital convertor (ADC), when large areas of relatively high backscatter are imaged. This leads to ADC-saturation (Meadows and Wright, 1994). Due to the reduction of the receiver gain settings of the SAR aboard ERS-2, the power loss of these images is less serve, but still present. This is especially the case for strong wind conditions and in the near range of the SAR images. In consequence, ERS SAR images used for computation of the wind speeds have to be recalibrated.

Power loss effect on wind measurements

The first calibration procedure relating the amplitudes A of the SAR.PRI data to the NRCS is given in Laur, 1992.

NRCS = A² * K^-1 * sin alpha * (sin alpha_ref)^-1

Where K is a calibration constant for the reference angle alpha_ref = 23^o. The factor sin alpha / sin alpha_ref takes the variation with incidence angle into account. The improved calibration procedure considers in addition the power loss due to ADC-saturation and changes in replica power (an internal SAR calibration), into account. The complete calibration formula is given in Laur et al., 1996

NRCS = A² * K^-1 * sin alpha * (sin alpha_ref)^-1 * IRP * RRP^-1 * power loss,

where IRP denotes the replica power of the image and RRP abbreviates the replica power of the reference image. The so far mentioned corrections are in the range of 1 dB while the power loss correction can be higher than 6 dB and is thus crucial. For scenes processed before appropiate dates a correction for the latest antenna pattern has to be applied (Laur et al., 1996).

Using the CMOD4 and the power loss correction tables from Laur et al., 1996 the relative error due to the power loss is estimated as a function of wind direction and wind speed at an incidence angle of 23^o for ERS-1 (figure 1) and for ERS-2 (figure 2), respectively. From these figures we can conclude that at least for ERS-1 images the power loss correction has to be always applied.

Fig. 1: Estimated relative reduction of wind speed in % for ERS-1 SAR due to the power loss for an incidence angle of 23^o. The wind direction is given clockwise from flight direction, e.g. 90^o corresponds to upwind.

Fig. 2: The same as in figure 1 but for ERS-2 SAR.

Errors of CMOD4 derived wind speeds

To compute the wind speed from ERS SAR images, the wind direction is needed as input into the CMOD4. In figure 3 the dependency of the NRCS on wind direction is plotted as computed with the CMOD4. The relative error in wind speed due to an uncertainty of +/- 10^o in wind direction is plotted in figure 4. The error is maximal for wind directions 215^o and 140^o (clockwise from flight direction).

Fig. 3: Dependency of NRCS on wind direction for wind speeds between 2 m/s and 26 m/s as computed with CMOD4. The wind direction is given clockwise from flight direction, e.g. 90^o corresponds to upwind.

Fig. 4: Relative error of wind speed in % due to an uncertainity in wind direction of +/- 10^o for an incidence angle of 23^o. The wind direction is given clockwise from flight direction, e.g. 90^o is corresponds to upwind.

Another source of error is the approximation used in the recalibration of the ERS SAR.PRI images. For large homogeneous surfaces like the open ocean this works well, but fails at sharp boundaries, such as coast lines and in near range at high wind conditions. In this case the land has a relative low NRCS compared to the water surface. This leads to a reduction of the NRCS over the water due to a sliding window which is used for estimation of the power loss. Especially for ERS-1 SAR this leads to unrealistically low values of wind speeds near the coast.

Comparison to ground truth

We derived the wind speeds from 20 recalibrated ERS-1/2 SAR images at different geographical locations and compared them to 10 minutes mean wind speed data sets from the `Royal Netherlands Meteorological Institute' (KNMI) (Koek, 1995). Figure 5 shows scatter plots of ground truth measurements versus the values from the SAR for 20 scenes considered. Out of the 20 SAR images 14 showed distinct wind streaks. In figure 5 (left side) a comparison of wind direction from SAR images to KNMI ground truth measurements is given. The measurements compare well, with a bias of 1.53^o and a correlation of 0.99. Figure 5 (right side) shows the scatter plot of wind speed with a bias of 0.6 m/s and a correlation of 0.78.

Fig. 5: Scatter plots of wind direction from SAR versus wind direction from KNMI (left side) and of wind speed from SAR versus wind speed from KNMI (right side).

Comparison to the mesoscale wind model GESIMA

Since no mesoscale, area covering measuring method exists, we used the mesoscale atmospheric model GESIMA to estimate the wind field around the island Rügen in the Baltic Sea. GESIMA is a three dimensional non-hydrostatic model with terrain following coordinates. A detailed description is given by Kapitza and Eppel, 1992. For our comparison GESIMA was run for a neutral atmosphere and a horizontal resolution of 1 km². The lower boundary condition for the friction velocity over the sea were given by Charnocks relation. The bottom stress over land was taken into account from land use charts. The results are compared to the wind speeds derived from the power loss corrected ERS-1 SAR image of the island Rügen (12. August 1991 at 21:07 UTC).

In figure 6 the isotaches are plotted as computed from GESIMA and in figure 7 from ERS-1 SAR with a fixed wind direction via CMOD4, respectively. On the ocean surface north of Rügen the SAR image shows wind rolls from which the the wind direction is derived. These features can not be reproduced by the GESIMA model. Relative changes in wind speed due to shadowing effects, e.g. east of Rügen, are clearly visible in both figures. The wind shadowing shows the same order of magnitude. The wind speed drops down to about 6 m/s behind the island before picking up again towards the open ocean. The geometrical location of the wind shadowing in the SAR and GESIMA are slightly different, showing that the results of GESIMA yield a much smoother solution than the distribution of wind speed in real nature. The SAR image shows much finer detail in wind structure and a higher variability. This is due to the difference between the snap shot of ERS-1 SAR of a highly turbulent wind field and the mesoscale model simulation assuming a stationary situation.

Fig. 6: Contour lines represent the wind speeds in m/s as computed by the GESIMA model arround the island Rügen in the Baltic Sea.

Fig. 7: Contour lines represent wind speeds as computed rom the ERS-1 SAR image via CMOD4 arround the island Rügen in the Baltic Sea.

Conclusions

Application of the CMOD4 algorithm to power loss corrected ERS-1/2 SAR images gives valuable information of mesoscale wind fields over the ocean. The measurements compare well to ground truth. When wind streaks are present in the SAR image, wind direction is measured to an accuracy of 5 degrees. Comparisons to the mesoscale model GESIMA show much finer details of wind speed in the SAR data, giving a lot of additional information to the model results. Power loss correction of SAR images to calibrated NRCS remains a problem especially for ERS-1 images near coast lines. For wind measurements the use of ERS-2 images is generally recommended.

References

Alpers, W.R. and B. Brümmer, Atmospheric Boundary Layer Rolls Observed by the Synthetic Aperture Radar Aboard the ERS-1 Satellite, J. Geophys. Res., Vol. 99, pp. 12 613-12 621, 1994.

Chapron, B., T. Elfouhaily and V. Kerbaol, A SAR Speckle Wind Algorithm, Proceedings of the Second ERS-1 Workshop, IFREMER Brest, BP 70, 29280 Plouzane, France, 1994.

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

Kapitza, H. and D.P. Eppel, The Non-Hydrostatic Mesoscale Model GESIMA. Part I: Dynamical Equations and Tests, Beitr. Phys. Atmosph., Vol. 65, No. 2, pp. 129-146, May 1992.

Koek, F.B., Monthly Bulletin North Sea, Royal Netherlands Meteorological Institute, KNMI, ISSN 0168-8332, Vol. 12, 1995.

Laur, H., ERS-1 SAR Calibration: Derivation of Backscattering Coefficient sigma naught in ERS-1.SAR.PRI Products, ESA, ESRIN, Issue 1, Rev. 0, October 1992.

Laur, H., P. Bally, P. Meadows, J. Sanchez, B. Schaettler and E. Lopinto, Derivation of the Backscattering Coefficient sigma naught in ESA ERS-1/2 SAR.PRI Data Products, Issue 2, Rev.1, 1996.

Lehner ,S., J. Horstmann, W. Koch and W. Rosenthal, Mesoscale Wind Measurements Using Recalibrated ERS SAR Images, submitted to J. Geophys. Res., December 1996.

Meadows, P.J. and P.A. Wright, ERS-1 SAR Analogue to Digital Convertor Saturation, Proceedings of the CEOS SAR Calibration Workshop, Ann Arbor Michigan, pp. 24-37, September 1994.

Rosenthal, W., S. Lehner, J. Horstmann and W. Koch, Wind Measurements Using ERS-1 SAR, Proc. of the Second ERS Applications Workshop, London, UK, pp. 355-358, December 1995.

Scoon, A., I.S. Robinson and P.J. Meadows, Demonstration of an Improved Calibration Scheme for ERS-1 SAR Imagery using a Scatterometer Wind Model, Int. J. Remote Sensing, Vol. 17, No. 2, pp. 413-418, 1996.

Stoffelen, Ad and D. Anderson, Characterisation of ERS-1 Scatterometer Measurements and Wind Retrieval, Proc. Second ERS-1 Symposium- Space at the Service of our Environment, Hamburg, Germany, ESA SP-361, pp. 997-1001, October, 1993.

Wright, J.W., Backscattering from Capillary Waves with Application to Sea Clutter, IEEE Trans. on Antennas and Propagation, Vol. AP-14, No. 6, pp. 749-754, November 1966.

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