IntroductionMesoscale wind fields over the oceansurface are of great interest for studyingprocesses in the boundary layer and forobtaining a better understanding ofcoastal processes. Spatial coveringwind fields are frequently derived fromspace-borne scatterometers (SCAT).The SCAT and the Synthetic ApertureRadar (SAR) aboard the Europeanremote-sensing satellites ERS-1 andERS-2 operate both at 5.3 GHz

(C-band) at moderate incidence anglesbetween 18° and 59°. For these inci-dence angles, the backscatter signal ofthe rough ocean surface is primarilycaused by the short surface gravitywaves, which are in resonance with theincident radiation of the sensor. For incidence angles of ERS SARs,between 20° and 26°, the range ofscattering wavelengths extends from8.2 to 6.5 cm. These waves are strongly influenced by the local wind

field and allow the backscatter to be ameasure of the wind.

For several applications, especially incoastal regions the spatial resolution ofthe ERS SCAT (45 × 45 km, with aswath width of 500 km) is insufficient.The ERS SAR’s high resolution of about25 × 25 m and swath width of 100 kmoffer a unique opportunity to derivemesoscale wind fields over the oceansurface. A lot of effort has been

ocean windfields 8

Ocean wind fields and their variability derived from SAR

J. Horstmann1, Wolfgang Koch1, S. Lehner2 & W. Rosenthal1

1GKSS Forschungszentrum GeesthachtMax-Planck-Str. D-21502 Geesthacht, Germany

phone: +49-4152-87.1567 – fax: +49-4152-87.1565 – e-mail: Horstmann@gkss.de

2Deutsche Forschungsanstalt für Luft- und Raumfahrt e.V., DFDD-82230 Wessling, Germany

phone: +49-8153-28.2828 – fax: +49-8153-28.1445 – e-mail: Susanne.Lehner@dlr.de

The Synthetic Aperture Radar (SAR) aboard the European Space Agency’s remote-sensing satellites (ERS-1 and ERS-2)acquires images that can be used to derive wind fields over the ocean surface. The SAR measures the backscatter, which is ameasure of the roughness of the ocean surface. The roughness is strongly influenced by the local wind field so that the radarbackscatter can be used to measure the wind. For the derivation of the wind field from SAR images the empirical C-bandmodels are applied, which were originally developed for the wind scatterometer (SCAT), that operates at the same wavelengthas the SAR. Scenes at different geographical locations and under different meteorological and oceanographical conditions wereselected to test the applicability of the ERS SAR for retrieving wind fields over the ocean. Especially in coastal regions, the SAR-derived mesoscale wind fields give a lot of additional information. As an example, an ERS SAR-retrieved wind field is comparedto ground truth measurements and to the results of a mesoscale atmospheric model.

Vachon P & FW Dobson, 1996:Validation of wind vector retireval fromERS-1 SAR images over the ocean,Global Atmos. Ocean Sys. 5, 177-187.

Vandemark D, FC Jackson, EJ Walsh &B Chapron, 1994: Airborne radar meas-urements of ocean wave spectra andwind speed during the ERS-1 SARwave experiment, Atmosphere-Ocean32, 143-178.

Vandemark D, JB Edson & B Chapron,1997: Altimeter estimation of sea-sur-face wind stress for light to moderate

winds, J. Atm. Oceanic Tech. 14, 716-722.

Wackerman C, C Rufenach, RShuchman, J Johannessen & KDavidson, 1996: Wind vector retrievalusing ERS-1 synthetic aperture radarimagery, IEEE Trans. Geosc. RemoteSensing 34, 1343-1352.

Witter DL & DB Chelton, 1991: AGeosat altimeter wind-speed algorithmand a method for altimeter wind-speedalgorithm development. JGR 96, 8853-8860.

Acknowledgements We thank the many involved in data collection and processing for the GrandBanks and Hi-Res experiments.Atmospheric Environment Service,Canada, provided the wind field datafrom the Grand Banks experiment. Thisstudy was supported in part by NASA’sPhysical Oceanography program andthe Office of Naval Research SpaceRemote Sensing program.

invested in the derivation of wind speedfrom SAR images, [see e.g. Alpers et al.1994, Chapron et al. 1994, Scoon et al.1996, and Horstmann 1997].Furthermore, the derivation of winddirection in addition to the wind speedwas derived from SAR images, [see e.g.Wakerman et al. 1996, Vachon et al.1996, Korsbakken et al. 1996, andLehner et al. 1998]. In the first part ofthis article the method to derive windspeeds and wind directions from ERSSAR images is introduced. In the second part an example at the Germancoast of the Baltic Sea is given andcompared to a mesoscale numericalmodel. Finally, the ERS SAR-derivedwind fields are applied to investigate thespatial variability of the wind field.

Wind from SARFor derivation of the wind speed fromERS-1 and 2 SAR images the empiricalC-band models CMOD4 [Stoffelen &Anderson 1997] or CMOD IFR2 [Quilfen& Bentamy 1994] can be used. Bothmodels were originally developed forthe SCAT aboard ERS-1 and 2. Thesemodels require input of the normalisedradar cross section (NRCS), the incidence angle of the radar beam, andthe wind direction. The NRCS and theincidence angle can be derived in astraightforward manner from the ERSSAR data. Due to the fact that smallchanges in NRCS result in largechanges in wind speed, the SARimages have to be calibrated as accurately as possible.

Details about the calibration are given inLaur et al. 1997. The wind direction canbe derived from ERS SAR images ifwind streaks are present. Figure 1shows an ERS-1 SAR image with clear-ly imaged wind streaks from which thewind direction is derived with an 180°ambiguity. In approximately 65% of theinvestigated SAR images, wind streaksare present from which the wind direc-tion can be extracted. Shadowing of thewind field for instance due to thecoastal topography removes the 180°ambiguity of wind direction. For deriva-tion of the wind direction a 10 × 10 kmsub-image is analysed with a two-dimensional fast Fourier transform (FFT).The results of one sub-image are plot-ted in Figure 2. The grey levels repre-sent energies of the power spectrumwhich are plotted for wavelengthsbetween 500 and 1500 m. The resultingwind direction (solid line) is perpendicu-lar to the dashed line depicting thedirection of peak power of wind streaks.

Comparison with a numerical modelFor the study of coastal processes,high-resolution wave and current models are used which require detailedwind information to calculate hydro-graphic parameters. Therefore, it isimportant to test the capability of ERS

ers-sar9

SAR image of the island Rügen at the Baltic coast of Germany, taken on 12 August1991, at 21:07 UTC from ERS-1. The solid black lines in the right part of the imagegive the orientation of the visible wind streaks from which the wind direction wascomputed for 10 × 10 km thick lines and 5 × 5 km thin lines. The 180° ambiguitycould be removed due to the shadowing of the coast.

Power spectrum of an area in the openwater north of Rügen. Wavelengthsbetween 500 and 1500 m are taken forthe estimation of wind direction. Themain energy is along an orientation of26° versus north; the resulting mainwind direction is orientated perpendi-cular.

①

②

SAR to reproduce spatial variations ofthe wind field. Due to the complextopographic structure, the island Rügennear the German coastline at the BalticSea was chosen as a test site. TheERS-1 SAR image of this area is shownin Figure 1. For validation, themesoscale atmospheric ‘GeesthachtSimulation’ model GESIMA, developedat GKSS Research Centre is used. It isa three-dimensional non-hydrostaticmesoscale model of atmospheric circu-lation [details are provided in Kapitzaand Eppel 1992]. For this application, itwas set up with 8 layers between 0 and1500 m and a horizontal resolution of 1 × 1 km. The roughness length overthe water surface is given according toCharnock’s relation and over land fromland use charts. For the upper boundary condition at 1500 m height,stationary wind speeds of 16 m/s coming from 306° were prescribed, andthe model was run assuming neutralatmospheric stratification. The NRCSand incidence angles from the ERS-1SAR image from 12 August 1991, at21:07 UTC were computed with a hori-zontal resolution of 1 × 1 km. For the wind derivation a constant winddirection of 290° was assumed. Figure3 shows the results of GESIMA and ofthe ERS SAR image using the CMOD4.The SAR image shows much finer detailin wind structure and a higher spatialvariability than the modelled wind field.This is caused by the differencebetween the snapshot of a highly

turbulent wind field from ERS SAR andthe simulation with a mesoscale modelassuming a stationary situation. On theocean surface north of Rügen, the SARimage shows wind rolls from which thewind direction was derived. The GESI-MA model cannot reproduce these fea-tures. This shows that the ERS SARdata give a lot of additional informationon the turbulent structure of the atmos-phere. East of the island of Rügen, thewind shadowing shows the same orderof magnitude. The wind speed dropsdown to about 6 m/s behind the island,but picks up again toward the openocean. The geometrical location of thewind shadowing in the ERS SAR andGESIMA data is slightly different, show-ing again that the results of GESIMAyield a much smoother solution than thedistribution of wind speed in nature.Since the model has been run for sta-tionary conditions, a full coincidencewith the ERS SAR derived wind fieldcannot be expected.

Spatial wind variabilityThe wind fields from ERS SAR give aunique opportunity to investigate thevariation of the wind field in space. Inthe conventional method using analysisof time series from a point measure-ment, Taylor’s hypothesis is assumed.Taylor considers the turbulence to befrozen as it advects past the measuringsensor. Therefore, the wind speed canbe used to translate turbulence meas-urements as a function of time to their

corresponding measurements in space.The ERS SAR-derived wind fields are asnapshot of the surface wind field of anarea of up to 100 × 100 km. Figure 4shows an example of a SAR imagefrom 1 December 1992 northwest ofthe Shetland Islands. There are range-travelling waves with approximately 300m wavelengths visible in the completeERS SAR image. The superimposedsolid lines represent the wind directionscomputed from the imaged windstreaks. Due to shadowing off the coastof the Shetlands the 180° ambiguitycould be removed. Using the meanwind direction, the wind speed wascomputed at a resolution of 100 × 100 m. This resulted in a mean windspeed of 14 m/s. The wind-speedspectrum was computed from the windspeeds with a two-dimensional FFT. Toget an idea of the distribution of thespectral amplitude in wave number, thewind-speed spectrum was integratedover all directions and plotted in a double logarithmic plot (Fig. 5). Thewave number and the correspondingwavelength are plotted versus the spec-tral density of the wind speed. Betweenwavelengths of 56 km and 2 km, thespectral density decreases. For lowerwavelengths the spectral density beginsto increase. This tendency wasobserved in several investigated SARimages. The slope of the spectrabetween 100 km and 2 km wavelengthsvaried between -1.1 and -1.6 while theslope below 2 km is approx. 1. Thechange of the spectral density in Figure 4 at wavelengths of approx. 300 m is due to the sea-state. Spectraldensity decreases with increasingwavelengths are a commonly observedphenomenon in the troposphere [Gage& Nastrom 1986] and at lower scales inthe boundary layer [Kaimal et al. 1972].Speckle effects dominate for resolutionsbelow 2 × 2 km. Due to the distribution,the slope for wavelengths below ~2 kmis approx. 1 and dominates the windvariability.

ConclusionApplication of the C-band modelsCMOD4 and CMOD IFR2 to ERS SARimages result in valuable information onmesoscale wind fields over the ocean. Ifwind streaks appear in ERS SAR

ocean windfields 10

Wind speeds computed by the Geesthacht simulation model of the atmosphere(GESIMA) on the left side and by the C-band model (CMOD4) from ERS-1 SARdata on the right side. The solid lines represent isotachs in m/s.

③

images, wind direction can be retrievedwith an 180° ambiguity which can beremoved if wind shadowing is visible.Comparisons of SAR-derived windspeeds to results of the mesoscalemodel GESIMA showed finer details ofwind speeds in the ERS SAR data,resulting in richer information in com-parison to the model results. The spatialwind variation over the ocean surfacewith ERS SAR images showed a dis-tinct reduction of spectral density withan increasing wave number and a slopeaccording to k-5/3.

ReferencesAlpers WR & B Brümmer, Atmosphericboundary layer rolls observed by thesynthetic aperture radar aboard theERS-1 satellite, JGR 99, 12613-12621,1994.

Chapron B, T Elfouhaily & V Kerbaol, A SAR speckle wind algorithm, in Proc.2nd ERS-1 Workshop, pp. 5-40,Ifremer, Brest, France, 1994.

Gage KS & GD Nastrom, Theoreticalinterpretation of atmospheric wavenum-ber spectra of wind and temperatureobserved by commercial aircraft duringGASP, J.Atmos.Sci. 43:7, 729-740,April 1996.

Horstmann J, Untersuchung zurWindgeschwindigkeitsbestimmung ausdem Radar mit synthetischer Apertur anBord der ERS-1/2-Satelliten, externalGKSS report, GKSS 97/E/55, ISSN0344-9629, 1997.

Kaimal JC, JC Wyngaard, Y Izumi & ORCote, Spectral characteristics of surfacelayer turbulence, Q.J. Roy.Meteor. Soc.98, 563-589, 1972.

Kapitza H & DP Eppel, The non-hydro-static mesoscale model GESIMA, I -Dynamical equations and tests, Contrib.Atmos.Phys. 65 (2), 129-146, May1992.

Korsbakken E, Quantitative wind fieldretrievals from ERS SAR images, YGTreport, ESA ESTEC, Printed by TIDCRepro., 1996.

Laur H, P Bally, P Meadows, JSanchez, B Schaettler & E Lopinto,Derivation of the backscattering coeffi-cient σ 0 in ESA ERS SAR PRI products,Tech. Note ES-TN-RS-PM-HL09, issue2, rev.4, 41 p., ESA, Frascati, Italy, May1997.

Lehner S, J Horstmann, W Koch & WRosenthal, Mesoscale wind measure-ments using recalibrated ERS SARimages, JGR, in press.

Quilfen Y & A Bentamy, Calibration/Validation of ERS-1 scatterometer preci-sion products, Proc. IGARSS 1994,

ers-sar11

ERS-1 SAR image of the west coast of the Shetland Islands, taken on 1 Dec. 1992,at 10:28 UTC. The highlighted area was taken to investigate the wind variability. Thesuperimposed solid lines represent the wind direction as computed from the wind streaks.

The spectral density versus wavenumber is plotted in log-log coordi-nates. The spectral density was computed from the wind-speed spectrum. The corresponding wave-lengths are plotted under the wavenumber axis.

④

⑤

IntroductionIt has been well established thatSynthetic Aperture Radar (SAR) imagesover the ocean can provide wind infor-mation over small spatial scales. SARimage data can be used quantitativelyas an imaging scatterometer if the SARis calibrated and a suitable windretrieval model is available [Vachon &Dobson 1996; Wackerman et al. 1996;Fetterer et al. 1998]. Canada’s RadarsatSAR (C-band HH polarisation) with itsScanSAR wide mode characterised bya swath width of 500 km at a spatialresolution of 100 m, can provide bothsynoptic-scale and small-scale views ofthe imprint of mesoscale meteorologicalprocesses and features on the ocean’ssurface.

Atmospheric mesoscale cyclonic vortices that develop in the regions

between sea ice (or cold land) and theocean during cold air outbreaks are ofincreasing interest [Twitchell et al. 1989;Heinemann & Claud 1997; Mitnik et al.1996]. Often referred to as polar lows,intensive mesoscale cyclones areformed poleward of major jet streams incold air masses or frontal zones; theircloud mass is mainly convective in origin [Businger & Reed 1989]. Withsmall spatial scales (< 500 km) andshort life times (24 to 48 hours), it maybe difficult for numerical weather predic-tion models to fully simulate their initia-tion, temporal evolution, and spatialstructure. Since polar mesoscalecyclones usually occur in data sparseregions, satellite images have been animportant tool for the detection andcharacterisation of these phenomena[Heinemann & Claud 1997; Mitnik et al.1996]. These high-latitude outbreaks

also have a role in understanding andquantifying the intense fluxes of heatand moisture from sea to air thataccompany oceanic deep convection[Marshall et al. 1998].

In this paper, we consider two such fea-tures observed in (thus far) uncalibratedRadarsat ScanSAR images. We thenconsider the more generic issue of windretrieval from Radarsat SAR images andpresent wind speed validation resultsfrom single-beam mode observationstaken over ships and buoys.

Labrador Sea mesoscale cyclonesFigure 1 shows a Radarsat ScanSARwide image of an intense mesoscalecyclone (MC) over the Labrador Sea.The image was acquired for routine icesurveillance [Bradley et al. l998] on 29December 1997. On this date, the sea-

windfield structure 12

Wind field structure and speed from Radarsat SAR images

P.W. Vachon1, I. Chunchuzov1* & F.W. Dobson2

1Natural Resources Canada, Canada Centre for Remote Sensing588 Booth Str., Ottawa, K1A 0Y7 Canada

2 Canada Dept. of Fisheries & Oceans, Bedford Institute of OceanographyP.O.Box 1006, Dartmouth, N.S. B2Y 4A2 Canada

* Under contract to CCRS from Noetix Research Inc.. 265 Carling Ave., Suite 403. Ottawa, Ont. K1S 2E1 Canada.

Two examples of Radarsat ScanSAR wide images of mesoscale cyclones in the Labrador Sea are presented, along with windretrieval validation results from both the ERS SARs and the Radarsat SAR. The results show that SAR images may be used todeduce information about both the spatial structure and the wind speed of mesoscale phenomena near the ocean’s surface.

Scoon A, IS Robinson, & PJ Meadows,Demonstration of an improved calibra-tion scheme for ERS-1 SAR imageryusing a scatterometer wind model,Int.J. Remote Sens. 17 (2), 413-418,1996.

Stoffelen A & D Anderson, Scattero-meter data interpretation: Estimationand validation of the transfer functionCMOD4, JGR 102, 5767-5780, 1997.

Vachon P & FW Dobson, Validation ofwind vector retrieval from ERS-1 SARimages over the ocean, The GlobalAtm. & Ocean Syst., 5, 177-187, 1996.

Wakerman CC, CL Rufenach, RSchuchman, J.A. Johannessen & K.Davidson, Wind vector retrieval usingERS-1 synthetic aperture radar imagery,JGR 34, 1343-1352, 1996.

AcknowledgementsThis work was supported by theBundesministerium für Bildung undForschung (BMBF) under contract No. 03F0165C. The ERS SAR imageswere kindly provided by ESA as part ofthe SARPAK project, under grantAO02.D113.