CROSS DETECTION BASED ON SYNTHETIC APERTURE RADAR (SAR) DATA AND NUMERICAL WAVE MODEL (WAM)

Xiao-Ming LI(1)(2), Susanne Lehner(1) , Ming-Xia HE(2) Johannes Schulz-Stellenfleth(1)

(1)German Aerospace Centre (DLR), Oberpfaffenhofen, 82234 Wessling, Germany, Email: [email protected] (2) Remote Sensing Institute, Ocean University of China, Qingdao, 266003, China

ABSTRACT that is characterized by a sea and one or more systems is called a mixed sea or confused sea. If The present paper is about the detection of cross their directions differ, the is called a cross based on ERS-2/SAR wave mode data and comparison sea. A cross sea case which occurred in the South-East to WAM model. A case of cross seas observed by Pacific was captured clearly by consecutive ERS-2 ERS-2/SAR was analyzed for its generation and SAR wave mode images acquired on August 10, 2000 development by wave mode data together with the at 17:30 UTC. One of the consecutive imagettes is WAM model two-dimensional (2-D) spectra. The sea shown in Fig. 1. The case is studied using 2-D spectra, surface elevation is estimated from these wave mode including the WAM model [2] spectra, cross spectra of data with cross sea features using a CMOD type tilt SAR complex data [3] and the non-linear retrieved algorithm. An effective method to remove the speckle PARSA spectra [4]. By a simple travel kinematic noise from SAR wave mode images is introduced, too, model, generation and development of cross seas is in order to get better results of sea surface elevation. demonstrated.

1. Introduction

It is well known that Synthetic Aperture Radar (SAR) provides directional ocean wave and surface wind information on a continuous and global scale [1]. Due to the high resolution of SAR data, it is possible to analyze the structure of the ocean wave field, such as wave groups and individual wave behavior. In the framework of the project WAVEALTAS, ESA provided a two-year’s ERS-2/SAR wave mode raw data set, which was reprocessed to single-look- Figure 1. ERS-2/SAR wave mode data imaging a cross complex SAR images at DLR using the BSAR sea, acquired on August 10, 2000 at 17:30 UTC processor [1]. The processed data set contains Traditionally, SAR ocean wave measurements are typically between 1300 and 1500 images of 10km by carried out in the spectral domain to estimate the two- 5km size daily. dimensional spectrum. One approach is realized in the 1.1 SAR ocean wave spectra algorithm MPI (Max-Planck Institute) scheme [5]. Later by Two types of ocean waves usually characterize the sea making using of the SAR complex image, the cross surface, namely wind sea and swell. The first refers to spectra method was derived to retrieve ocean wave waves influenced by the local wind, the latter to waves propagation directions without ambiguity [3]. This is that have propagated out of the generating area and are the standard algorithm for ASAR wave mode data [6]. thus no longer affected by the local wind. A sea state However, all these applications do not make use of full

______Proc. ‘Envisat Symposium 2007’, Montreux, Switzerland 23–27 April 2007 (ESA SP-636, July 2007) information on the two-dimensional sea surface 2. Data Set Description elevation field provided by SAR. For this study ERS-2 SAR images and ERA-40 model 1.2 Sea Surface Elevation data are used. Sea surface elevation is estimated from SAR wave 2.1 ERS-2 SAR wave mode data mode imagettes, e.g. as done by the LISE algorithm ERS SAR wave mode data is acquired over the ocean developed by DLR [7]. The objective of this paper is every 200 km along the satellite track with the to introduce a technique to estimate the sea surface coverage of 5km x 10km, when image mode data is elevation field from SAR wave mode by a CMOD not requested. type tilt algorithm ([email protected]). It is A two-year wave mode dataset from ERS-2 SAR well known that the geophysical model function acquired during 1999 and 2000, which has been CMOD describes the relationship between wind speed, reprocessed to single-look-complex data at DLR [1]. wind direction, antenna look direction and incidence 2.2 Model data angle [8]. The incidence angle chosen for SAR wave mode images is set around 23◦. This does not take After the success of ERA-15, European Centre for account into the tilt caused by the sea surface Medium-Range Weather Forecasts (ECMWF) is elevation. Although in steep wave situations, the performing their second reanalysis called ERA-40 [10], which covers 45 years, from 1957 until 2002. change of tilt is significant. Given the wind speed U10 and wind direction and by using CMOD, thus the tilt Starting 1991, data obtained from the angle of the individual ocean wave is estimated. This altimeters on board of ERS-1 and ERS-2 are is the main idea of the approach introduced in the assimilated into ERA-40 wave data. paper. The Numerical wave model used for the research is 1.3 Speckle noise removing in SAR images the well-known WAM [2]. It is the so called the third generation model (cycle 4), in which the wave SAR images are generated by coherent processing of spectrum is computed by integration of the energy the scattered signals and they are highly susceptible to balance equation. The model resolution is chosen as 1◦ specking effects [9]. The presence of speckle in SAR by 1◦ and forced by the ERA-40 high resolution wind image reduces the ability of information extraction, field (1◦ by 1◦). especially when the ratio of signal to noise is low over the ocean. It is observed that the energy due to the 3. Observation of Cross Seas imaged ocean wave field is concentrated in narrow angular sectors of the image spectrum. Thus it is The case of cross seas imaged by SAR occurred in the possible to choose a proper threshold to remove the southeastern Pacific on Aug.10, 2000. An impressive speckle noise in the image by a Fourier Fast pattern of crossing ocean wave systems is observed on Transform (FFT) type of filter. at least 8 consecutive images, i.e. on a distance of more than 1000 km. The most distinct peaks can be The paper is structured as follows: in section 2, the observed on the image which is situated at 23.06S and data set used in this research is introduced. Section 3 is 111.6W degrees as shown in Fig. 1. Three consecutive about the generation and development of the cross seas imagettes through the cross sea area are shown in Fig. case. The technique of removing speckle noise and 2. The hindcast WAM model spectra (upper row), estimation of sea surface elevation on SAR wave observed cross spectra (middle row with wave mode data is introduced in the fourth section. traveling direction ambiguity removed by using the imaginary part of the cross spectrum) and retrieved PARSA spectra (lower row) are shown as well. It is clear from these contour plots that the cross sea As expected the PARSA inversion algorithm turns the contains two distinct swell systems travelling to peak of swell Sne towards the azimuth direction. In a northeast and northwest. These are denoted as Sne and next step the dissipation scheme of the WAM model

Snw hereby respectively. The two swell systems are will be compared to the inverted SAR spectra. most dominant in the last imagette, which is shown in

Fig. 2. The PARSA retrieved Sne swell system peak 4. Estimation of Sea Surface Elevation wave length is about 400m, which agrees well with the 4.1 Technique of speckle noise removing observed cross spectral peak, but the WAM model Radar signals returned from the scatterers within the hindcast result shows only about 300m. SAR resolution cell are added up coherently during the Gonzalez et al., [11] introduced a simple kinematical process, which makes images susceptible to speckle. wave model which has shown that swell generally Basically the speckle is signal-dependent and acts like obeys linear wave theory of propagation and does not multiplicative noise. Therefore, in the SAR image seem to be affected by propagating through zones of analysis, the multiplicative noise model is used [12]. steady wind. With the simple kinematical model and Iσ = I *S (1) swell group speed calculated from the WAM model In which, Iσ is SAR intensity, I is the cross section result. It is estimated that Swell S was generated by a ne information and S is speckle noise. storm about 4000km away near to the Antarctic In the SAR one-dimension directional spectrum, the Continent and after travelled 96 hours it arrived at the energy is concentrated in very small angular sectors. observation point. Swell S was generated by an anti- ne This property gives the possibility to remove the clockwise low pressure about 2200km away and speckle and extract the useful information by setting a travelled about 54 hours to the observation point. In proper threshold. Before doing the (FFT), the Fig. 4 swell travelling and cross sea generation is logarithm is calculated. Thus the multiplicative shown on a simplified map. The black dotted line speckle noise in SAR image is transformed into shows the swell great circle route from the generation additive noise i.e. area (green triangle) to the observation at the imagette log(Iσ) = log(I)+log(S) (2) location (Black Square, 23.06S and 111.6W). The After the FFT, the speckle is still additive, i.e. yellow lines indicate swell traveling distance during F(log(Iσ)) = F( log(I))+ F(log(S)) (3) one day. Using the proper thresholds, the speckle can be The imaginary part of the SAR cross spectrum shows removed and the intensity image can be retrieved by a strong peak of the swell system S as shown in nw inverse FFT. This method is denoted as LOG-FFT Fig.3. This peak is underestimated in the WAM method. Fig. 6 shows the speckle reduced result of the model. This is due to the fact that the input wind field second imagette shown in Fig.2. from ERA-40 at the generation area of the swell was The filtered result shows that the speckle noise is weak. This can be concluded from the comparison to reduced significantly. Two different swell systems can the Quikscat wind field in Fig. 5. The maximum wind be observed clearly. speed measured by Quikscat is about 30m/s larger than 4.2 Technique of sea surface elevation estimation ERA-40 model result about 20m/s. As mentioned above, the elevation of sea surface will For the S swell system, it can be observed that the ne cause radar cross section changes. Different tilt of the energy contained on the system is largest in the sea surface generates different local incidence angles imagette (the rightmost one Fig. 2) closest to the in every pixel of the SAR image. The procedure to generation area and decreases northward along the estimate the sea surface elevation is demonstrated in orbit due to the swell dissipation. the following. sim First, the normalized radar cross section (NRCS) σ0 individual ocean waves properties (e.g. crest height of the imagette is simulated by using the collocated and length).

ERA-40 wind field data (wind speed U10 and wind direction ψ), the antenna look direction φ and the Acknowledgment incidence angle θ is 23◦ in the CMOD5 function. The The ERS-2/SAR wave mode raw data were kindly calibration constant applied for SAR wave mode supplied by ESA in the framework of AO images is taken to be -44.96dB [13]. WAVEATLAS. We thank for the ECMWF make the Second, from the intensity value of pixels in imagettes, ERA-40 data set available freely. obs the observed σ0 measured by SAR is computed by using the calibration constant. The cost function J is Reference defined as Eq.4 1. S. Lehner, J. Schulz-Stellenfleth, J.B. Schättler, obs sim 2 H. Breit, J. Horstmann. (2000). Wind and Wave J()=( (U100 ,,, )()−+ U100 ,,, θϕψσαθϕψσα ) (4) Measurments Using Complex ERS-2 Wave Mode J(α) optimizes the angle α, the difference between data, IEEE TGRS, Vol.38, No. 5, Pp. 2246-2257. local incidence angle due to tilt and 23◦. The minimum 2. WAMDI Group. (1988). the WAM model a third of the cost function corresponds to the best fit of local generation ocean wave prediction model, Journal incidence angle. of Physics , 18, pp. 1775-1810. In a further step, the slope in every pixel from the local 3. G. Engen and H. Johnson. (2000). SAR ocean incidence angle is estimated. The sea surface elevation wave inversion using image cross spectra, IEEE of SAR wave mode imagettes is obtained by TGARS, Vol.33, pp. 329-360. integrating the slope in every pixel along the range 4. J. Schulz-Stellenfleth, S. Lehner, D. Hoja. (2005). direction. A parametric scheme for the retrieval of two- Fig. 6 shows the intensity value of pixels in the dimensional ocean wave spectra from synthetic selected purple rectangular area of the filtered aperture radar look cross spectra, J. Geophys. imagette shown in Fig. 5. Fig. 7 is the corresponding Res., Vol. 110 sea surface elevation result using the CMOD tilt 5. K. Hasselmann and S. Hasselmann. (1991). On algorithm. In a next step, this first result will be the nonlinear mapping of an ocean wave spectrum validated against buoy data and model results. into a synthetic aperture radar image spectrum, J.

Geophys. Res., vol.96, pp.10713-10729 Summary 6. Envisat ASAR Level 2 products Algorithms: A case of cross seas captured clearly by SAR is http://envisat.esa.int/dataproducts/asar/CNTR2-7- analyzed for swells generation and dissipation based 1.htm on SAR wave mode data and WAM model. The 7. J. Schulz-Stellenfleth and S. Lehner. (2004). comparison among WAM model spectra, cross spectra Measurement of 2-D Sea Surface Elevation Fields and PARSA non-linear inverted spectra is using Complex Synthetic Aperture Radar Data, demonstrated. SAR measurements of sea state could IEEE TGARS, Vol. 42, No 6, pp 1149-1160. be used to validate the quality of the wind field driving 8. Stoffelen, A.C.M and D.L.T. Anderson. (1997). the WAM model. A CMOD type technique for sea Scatterometer Data Interpretation: Derivation of surface elevation estimation is introduced. A LOG- the transfer function CMOD4, J. Geophys. Res., FFT speckle noise reduction method used on ERS-2 vol.102, pp.5676-5780. SAR wave mode data is applied. It is effective for 9. Jong-Sen Lee. (1981). Speckle analysis and noise removal and extraction of information in smoothing of Synthetic Aperture Radar images, Computer Graphics and Image Processing, 17, 12. Alpers, W. and K. Hasselmann. (1982). Spectral pp.24-32 signal to clutter and thermal noise properties of 10. http://www.ecmwf.int/research/era/ ocean wave imaging synthetic aperture radars, Int. 11. F.I. Gonzalez, B. Holt, and F.G. Tilley. (1987). J. Rem. Sens., Vol. 3, pp. 423-446 The age and source of ocean swell observed in 13. J. Horstmann., S. Lehner, H. Schiller. (2003). hurricane Josephine, Johns Hopkins Tech. Digest, Global wind speed retrieval from SAR, IEEE 8, pp. 94-99 TGARS, Vol. 10, pp. 2277-228

WAM spectra

Cross-spectra Module

PARSA-spectra

Figure 2. SAR Imagettes, WAM Model spectra (m4), cross spectra (m2), PARSA retrieved spectra (m4)

Figure 5. Filtered result by LOG-FFT method of third imagette shown in Fig.2 Figure 3. Cross spectra imaginary part of rightmost imagette shown in Fig. 2

Figure 4. Sketch map of cross sea system generation Figure 6. Intensity value of pixels in Fig. 5 purple rectangular area

Figure 7. Corresponding Sea surface elevation

Figure. 5 Wind fields of ERA-40 and Quikscat at

generation area of Swell Snw