Detection of Sea Surface Temperature Fronts from SAR Images

Li Zhao aSchool of Marine Sciences, Nanjing University of Information Science and Technology, Nanjing, Jiangsu, China
dFisheries and Oceans Canada, Bedford Institute of Oceanography, Dartmouth, Nova Scotia, Canada

Search for other papers by Li Zhao in
Current site
Google Scholar
PubMed
Close
,
Tao Xie bLaboratory for Regional Oceanography and Numerical Modeling, Qingdao National Laboratory for Marine Science and Technology, Qingdao, Shandong, China
cSchool of Remote Sensing and Geomatics Engineering, Nanjing University of Information Science and Technology, Nanjing, Jiangsu, China

Search for other papers by Tao Xie in
Current site
Google Scholar
PubMed
Close
,
William Perrie dFisheries and Oceans Canada, Bedford Institute of Oceanography, Dartmouth, Nova Scotia, Canada

Search for other papers by William Perrie in
Current site
Google Scholar
PubMed
Close
,
Ming Ma eBeijing Institute of Applied Meteorology, Beijing, China

Search for other papers by Ming Ma in
Current site
Google Scholar
PubMed
Close
,
Jingsong Yang fState Key Laboratory of Satellite Ocean Environment Dynamics, Second Institute of Oceanography, Ministry of Natural Resources, Hangzhou, Zhejiang, China

Search for other papers by Jingsong Yang in
Current site
Google Scholar
PubMed
Close
,
Chengzu Bai eBeijing Institute of Applied Meteorology, Beijing, China

Search for other papers by Chengzu Bai in
Current site
Google Scholar
PubMed
Close
, and
Rick Danielson dFisheries and Oceans Canada, Bedford Institute of Oceanography, Dartmouth, Nova Scotia, Canada

Search for other papers by Rick Danielson in
Current site
Google Scholar
PubMed
Close
Free access

Abstract

Sea surface temperature (SST) fronts are important for fisheries and marine ecology, as well as upper-ocean dynamics, weather forecasting, and climate monitoring. In this paper, we propose a new approach to detect SST fronts from RADARSAT-2 ScanSAR images, based on the correlation of SAR-derived wind speeds using the gray level cooccurrence matrix (GLCM) approach. Due to the large differences between the correlation of wind speeds for SST fronts compared to other areas, SST fronts can be detected by the threshold method. To eliminate small-scale features (or noise), the 30 km scale is used as the length threshold for the detection of the SST fronts. The proposed method is effective when wind speeds are between 3 and 13 m s−1. The overall accuracy of our method is about 93.6%, which is sufficient for operational applications.

© 2022 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).

Corresponding author: Tao Xie, xietao@nuist.edu.cn

Abstract

Sea surface temperature (SST) fronts are important for fisheries and marine ecology, as well as upper-ocean dynamics, weather forecasting, and climate monitoring. In this paper, we propose a new approach to detect SST fronts from RADARSAT-2 ScanSAR images, based on the correlation of SAR-derived wind speeds using the gray level cooccurrence matrix (GLCM) approach. Due to the large differences between the correlation of wind speeds for SST fronts compared to other areas, SST fronts can be detected by the threshold method. To eliminate small-scale features (or noise), the 30 km scale is used as the length threshold for the detection of the SST fronts. The proposed method is effective when wind speeds are between 3 and 13 m s−1. The overall accuracy of our method is about 93.6%, which is sufficient for operational applications.

© 2022 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).

Corresponding author: Tao Xie, xietao@nuist.edu.cn

1. Introduction

Sea surface temperature (SST) fronts are typical mesoscale oceanic processes, which can be defined as narrow boundaries separating waters with different temperatures; these play an important role in fisheries and marine ecology (Mauzole et al. 2020). In particular, SST fronts interact with the atmosphere in the exchange of heat and moisture. Therefore, SST fronts are crucial geophysical parameters for upper-ocean dynamics, weather forecasting and climate monitoring.

Recently, satellite remote sensing has provided a powerful tool to carry out continuous observations of SST fronts over large areas. SST data derived from optical sensors such as Moderate Resolution Imaging Spectroradiometer (MODIS) or Advanced Very High Resolution Radiometer (AVHRR) have been widely used for SST front studies since the 1970s (Legeckis 1978). Several studies have obtained good results in detecting SST fronts from these data. In these works, the automatic detection of SST fronts is done by a gradient algorithm, a histogram algorithm, or a combination of these methodologies (Shimada et al. 2005; Belkin and O’Reilly 2009; Miller 2009; Kirches et al. 2016). With the development of artificial intelligence, deep learning methods such as convolutional neural networks (CNNs) have been introduced to identify SST fronts and promising results have been achieved (Sun et al. 2018). However, it is well known that detection of SST fronts from optical images is difficult in cloudy days.

Spaceborne synthetic aperture radar (SAR) can work in all-weather conditions, during day and night, and can penetrate clouds. The echo signal received by SAR is sensitive to the roughness of the sea surface, which is induced by wind and which can be retrieved from SAR images using geophysical model functions (GMFs) (Fang et al. 2017, 2018, 2019). Moreover, it has been reported that SST fronts are related to wind variations (Johannessen et al. 1996; Beal et al. 1997; Chelton et al. 2004; Song et al. 2006; Kim et al. 2017). The relationship between SST gradients and SAR-derived wind variations, at high spatial resolution, provides a potential methodology for detection of SST fronts (Xie et al. 2010; Kuang et al. 2012).

In this paper, a high-resolution method is developed for SST frontal detection based on RADARSAT-2 ScanSAR images. In our approach, wind speeds are derived from SAR images and thereafter used to detect SST fronts, rather than by the application of the normalized radar cross section (NRCS) images. One reason that we choose wind speed images for identification of SST fronts is the dependence of the NRCS on incidence angles. Due to the large variation in incidence angles, ranging from about 20° to 40° for ScanSAR images, the NRCS can be greatly influenced by incidence angles and generally decreases along the range direction. Section 2 describes the study area and data. Our method is illustrated in section 3. Results and discussion are given in section 4. Finally, conclusions are presented in section 5.

2. Study area and data

a. Study area

The Gulf Stream is the most powerful warm current in the global ocean, transporting warm tropical water to mid- and high latitudes and profoundly affecting Earth’s climate system. It is associated with strong sea surface temperature gradients. Therefore, the Gulf Stream region is an ideal location for the study of SST fronts. Figure 1 illustrates our study area and the location of 30 RADARSAT-2 SAR images. Details regarding these SAR images are described in section 2b.

Fig. 1.
Fig. 1.

Location of the study area and 30 RADARSAT-2 SAR images with SST fronts.

Citation: Journal of Atmospheric and Oceanic Technology 39, 12; 10.1175/JTECH-D-22-0009.1

b. RADARSAT-2 SAR images

This study is based on 30 RADARSAT-2 SAR images, in ScanSAR Narrow (SCN) mode, with 300 km swath width and 25 m pixel spacing in both azimuth and range directions. The range of incidence angles is about 19°–40°. This mode is very suitable for observing SST fronts and other mesoscale features due to its large swath width and high resolution. Only VV polarization images are used to detect SST fronts.

c. Wind direction data

External wind direction data are used to retrieve sea surface wind speeds from the SAR images. In this study, daily and gridded QuikSCAT wind directions are used. The resolution of these data is 25 km × 25 km. The time discrepancy between SAR images and QuikSCAT data is less than half an hour. For trade winds over the Gulf Stream, it is reasonable to assume that wind directions change very slowly.

d. SST data

To validate our method, blended MODIS and AVHRR SST data were used. Some of these were averaged over several days to minimize cloud contamination. The resolution of the SST data is 1 km. These raw SST data were acquired from the U.S. Navy Research Laboratory (NRL) and processed at the Meteorology and Oceanography (MetOc) Centre in Halifax. Since the Gulf Stream can change location by up to 30 km day−1, the correspondences between SAR and SST data may not be exact due to the time difference.

3. Methods

a. SAR image preprocessing

In the first step of our algorithm for SST fronts detection, SAR image preprocessing is conducted; this includes absolute radiometric calibration and pixel resampling for VV polarization, as well as incidence angle calculation. Absolute radiometric calibration is performed to convert the digital numbers found in the SAR products to backscattering values, by applying a constant offset- and range-dependent gain to the SAR imagery in VV polarization. For detected products, the NRCS can be obtained by the following formula:
σ0=DN2+BA,
where σ0 is the backscattering values in linear units, DN is digital number, and A and B are the gain value and offset contained in the sigma-naught (σ0) lookup table (LUT), respectively. To reduce the speckle effects on wind speed retrieval, we make a 40 × 40 pixel boxcar average on σ0. As a result, the resampled pixel spacing is 1 km. The radar incidence angle is also needed in the wind speed retrieval methodology and can be calculated by
θ=sin1(A/C)×180/π,
where θ is the radar incidence angle in degrees, and C is the gain value that can be derived from the beta-naught (β0) LUT.

b. Retrieval of wind speed

The second step is to carry out wind speed retrieval from the preprocessed NRCSs in VV polarization, based on empirical GMFs. The CMOD series, e.g., CMOD4, CMOD5, CMOD5.N are the most widely used GMFs for wind speed retrieval from SAR images (Vachon and Dobson 1996; Hersbach et al. 2007; Hersbach 2010). The CMOD functions relate the NRCS of the sea surface to the radar incidence angle, wind speed at 10 m height, and wind direction, relative to the antenna look direction. Therefore, one can retrieve wind speed from SAR imagery if the radar incidence angle, wind direction and NRCS in VV polarization are known. After SAR image processing, the only unknown parameter is wind direction. The wind direction from QuikSCAT is interpolated to processed SAR image grids. We use CMOD5.N to estimate wind speed at 1 km resolution, because previous studies have demonstrated that it works best with RADARSAT-2 SAR data (Zhang et al. 2011).

c. Extraction of sea surface texture features

The third step is extraction of texture features from wind speed images. The extraction of sea surface texture features is based on the gray level cooccurrence matrix (GLCM). The GLCM, denoted S(i, j, d, ϑ), gives the cooccurrence probability between gray levels i and j, at a particular distance d and direction ϑ, and is defined as (Clausi 2002)
S(i,j,d,ϑ)=P(i,j,d,ϑ)i,j=1KP(i,j,d,ϑ)S(i,j,d,ϑ),
where P represents the number of occurrences of gray levels i and j within this particular window, given a certain (d, ϑ) pair, and K is the quantized number of gray levels. To account for possible rotation of the SST fronts, the GLCM is averaged over four directions (0°, 45°, 90°, 135°).
Based on the GLCM values, the following texture features can be calculated (Zakhvatkina et al. 2017):
Homogeneity=i,j=1KS(i,j,d,ϑ)1+(ij)2,
Energy=i,j=1K[S(i,j,d,ϑ)]2,
Correlation=i,j=1K(iμx)(jμy)S(i,j,d,ϑ)σxσy,
Contrast=i,j=1K(ij)2S(i,j,d,ϑ),
where μx and μy are mean values of rows and columns, and σx and σy are the standard deviations of rows and columns.

By analyzing the texture features of wind speed images calculated using (4)(7), we can obtain the optimal texture features for detection of SST fronts. Because there is large difference between the correlations of wind speed for SST fronts compared to other areas, SST fronts can be detected by the correlations image. However, it is difficult to find the features of SST fronts in the homogeneity, energy, or contrast images. Therefore, only the correlations image for wind speed is used to detect SST fronts in this work. More details are given in section 4c.

Several factors are important during the calculation of the correlations image, such as the size of the sliding window, the cooccurrence distance, the moving step of the window, and the quantized number of gray levels. We set the moving step to 1 to keep the resolution of the texture features the same as that of the wind speed. The remaining parameters have been chosen by trial and error, in order to optimize the overall results. For our study, the size of the sliding window is 7 × 7; the cooccurrence distance is 1; and the quantized number of gray levels is 32.

d. Detection of SST fronts

Because of the large difference between the correlation of wind speed for SST fronts compared to other areas, SST fronts can be detected from the correlations image using the threshold method. For a given threshold T, the pixel located at any given grid position (i, j) of the correlations image, is identified as a SST front if
Correlation(i,j)T.
In analysis of the SAR observations, we find that T = 0.8 is sufficient to identify SST fronts because of the very strong correlation that they have to wind speed.

A previous study has shown that mesoscale SST fronts in the Gulf Stream should be continuous linear features which are at least 5 km in length (Sikora and Ufermann 2004). This is a basic condition for SST fronts. From an analysis of 41 SAR images and collocated, cotemporal SST images, Kuang et al. (2012) suggested that a length scale in excess of 30 km is required for actual retrieval of SST fronts from SAR images. In this study, the 30 km length scale is also used as the threshold for detection of SST fronts and to eliminate small-scale features (or noise). After removing small-scale features, it is then possible to detect and identify SST fronts.

4. Results and discussion

a. NRCS calculation

As an example to demonstrate our method, four SAR images S1–S4 are investigated. The NRCS images with 1 km resolution in VV polarization for SAR images S1–S4 are shown in Fig. 2, after completion of the preprocessing step. Due to differences (or gradients) between SST fronts and other areas, the outlines of the SST fronts are clearly evident in all SAR images. However, the differences are quite small, especially in SAR images S3 and S4. Other oceanic phenomenon can possibly be confused with these SST fronts. Moreover, the incidence angle has a significant effect on NRCS values due to the large range in its variation, from about 20° to 40°. In general, the NRCS in VV polarization decreases with increasing incidence angle and this can be clearly seen in SAR images S2, S3, and S4. Therefore, some features of the SST fronts might not be displayed due to the influence of incidence angle. As a consequence, it is quite difficult to detect SST fronts from NRCS images directly.

Fig. 2.
Fig. 2.

NRCS values in VV polarization for the SAR images at (top left) 1016:29 UTC 15 Sep 2008 (S1), (top right) 2206:58 UTC 25 Oct 2008 (S2), (bottom left) 2210:58 UTC 28 Apr 2009(S3), and (bottom right) 2211:34 UTC 22 May 2009(S4). RADARSAT-2 data and products from MacDonald, Dettwiler, and Associates Ltd., all rights reserved.

Citation: Journal of Atmospheric and Oceanic Technology 39, 12; 10.1175/JTECH-D-22-0009.1

b. Wind speed retrieval

Using the QuikSCAT wind directions, the wind speeds at 1 km resolution were retrieved from VV-polarized NRCSs using CMOD5.N GMF, as shown in Fig. 3. The mean wind speeds for SAR images S1, S2, S3, and S4 are 3.70, 5.97, 7.01, and 7.52 m s−1, respectively. The SST fronts can be clearly seen by the associated larger local gradients in wind speeds. Compared to Fig. 2, the SST front features are much more clearly displayed in the wind speed images of Fig. 3 as a result of the elimination of the incident angle effects.

Fig. 3.
Fig. 3.

Wind speed retrieved from VV-polarized NRCS.

Citation: Journal of Atmospheric and Oceanic Technology 39, 12; 10.1175/JTECH-D-22-0009.1

c. Texture feature extraction

We focus on SAR image S2. To optimally choose texture features for detection of SST fronts, the texture features of the corresponding wind speed image are calculated using (4)(7), and displayed in Fig. 4. All texture features have been normalized to the range (0, 1). The correlation values of wind speed for SST fronts are very strong, exceeding 0.8, while the correlation values for other areas are relatively very small. In comparison, it is quite difficult to identify SST fronts in the images for the homogeneity, energy, and contrast values. Hence, the preferred method for detection of SST fronts is by making use of the correlation field for wind speeds.

Fig. 4.
Fig. 4.

Texture features revealed from the values for homogeneity, energy, correlation, and contrast, for wind speed derived from SAR image S2, using Eqs. (4)(7).

Citation: Journal of Atmospheric and Oceanic Technology 39, 12; 10.1175/JTECH-D-22-0009.1

Therefore, the correlation values for wind speed are calculated for SAR images S1–S4, as displayed in Fig. 5. As expected, quite high correlation values are obtained for wind speeds for SST fronts. Under these conditions, small-scale features also have strong correlation values for wind speed. However, they can be treated as noise because our focus is to detect SST fronts from SAR images. They can be removed by using the length threshold.

Fig. 5.
Fig. 5.

Correlation values for wind speed corresponding to SAR images S1–S4.

Citation: Journal of Atmospheric and Oceanic Technology 39, 12; 10.1175/JTECH-D-22-0009.1

The correlation values for NRCSs in VV polarization mode have also been calculated and are shown in Fig. 6. Compared with Fig. 5, the correlations for SST fronts are weaker. Moreover, some features of the SST fronts are lost as a result of the strong incidence angle effect, especially in SAR images S2, S3, and S4. This proves that NRCS is not suitable for SST fronts detection.

Fig. 6.
Fig. 6.

Correlation values for the NRCSs in VV polarization of the SAR images.

Citation: Journal of Atmospheric and Oceanic Technology 39, 12; 10.1175/JTECH-D-22-0009.1

d. SST fronts detection

Identification of the detected SST fronts (black lines) overlaid on SST images are shown in Fig. 7, after applying our method to the correlations of wind speeds shown in Fig. 5. There are some time differences between collocated SAR images and SST data. There is therefore some inconsistency between SST fronts in the SAR images and blended SST images because of the shifting tendency of the Gulf Stream. In Fig. 7 (S4), the SST front indicated by the red box cannot be seen from SST map. This might be caused by the inconsistency mentioned above, related to changes in the Gulf Stream with time.

Fig. 7.
Fig. 7.

SST fronts detected from SAR images, overlapped on blended SST data: (top left) 3-day-averaged SST data (S1), (top right) 15 h earlier (S2), (bottom left) 4 h later (S3), and (bottom right) 4 h earlier (S4).

Citation: Journal of Atmospheric and Oceanic Technology 39, 12; 10.1175/JTECH-D-22-0009.1

To further validate our proposed method, a total of 30 SAR images were tested. The results were compared with the SST fronts extracted from SST images from MODIS or AVHRR observations, through visual interpretation. Considering the inconsistency caused by the time differences, for every pixel of the SST fronts in the SST images, our criteria is that if we can find a corresponding pixel of the SST fronts within a two-pixel distance in the SAR images, then we say the SST fronts are detected in the SAR images. The overall accuracy is 93.6%. Due to biases in the SAR data, such as may result from high winds, breaking waves, or heavy rain, some SST fronts cannot be detected. However, our results suggest that 91.2% of the SAR-detected fronts are SST fronts. Therefore, the performance of our method is sufficient for further applications. Overall, there is no exact correspondence between SST images and SAR images. However, it is hard to remove all the “noise” related to other complex oceanic features, because their spatial scales can exceed 30 km. The detection accuracy of our method is slightly lower than that (∼95%) of Kuang et al. (2012). Comparing our detection results for S2 and S3 with results from their method, we find that they are able to detect more fronts whereas the locations of fronts detected by our method are more accurate. Our method is effective when wind speeds are between 3 and 13 m s−1, which is similar to the range 3–12 m s−1 suggested by Kuang et al. However, our method is easier to implement and can potentially be transitioned to operational applications because the additional calculations related to wind stress and multistage spatial filtering are not needed. An additional advantage is that errors that might be introduced by converting wind speed to wind stress are avoided, especially in low wind speed conditions.

5. Conclusions

In this study, a high-resolution method is proposed to detect SST fronts from SAR images, based on 1-km-scale texture features, i.e., the correlation values for SAR-derived wind speeds. SST fronts can be detected from correlations images using the threshold method because of large differences in the correlation values for wind speed for SST fronts, compared to surrounding areas in the SAR image. In this methodology, for observations of SST fronts in the Gulf Stream region, a scale of 30 km should be used as the length threshold, to eliminate small-scale features or noise. Compared to previous studies, the performance of our method is comparable, but easier to implement.

In future work, how to discriminate SST fronts from other competing oceanic processes should be considered. We suggest that deep learning methodologies might provide a pathway forward, as this approach has achieved good success in classifications of targets in SAR images.

Acknowledgments.

This work was partly supported by the National Key R&D Program of China (2021YFC2803302), the Natural Science Foundation of Jiangsu Province (Grants YJGL-YF-2020-16 and JSZRHYKJ202114), Postgraduate Research and Practice Innovation Program of Jiangsu Province (KYCX20_0930), and the China Scholarship Council for 2 year’s study at the Bedford Institute of Oceanography (202008320523). Additional support was also provided by Canada’s Surface Water and Ocean Topography (SWOT) programs and MAXSS, the ESA initiative for Maximum Atmospheric Extreme Satellite Synergy.

Data availability statement.

The RADARSAT-2 SAR imagery are provided by MacDonald, Dettwiler, and Associates Ltd., the QuikSCAT data are from Remote Sensing Systems (https://remss.com/), and SST data are from the U.S. Navy Research Laboratory (https://www.nrl.navy.mil/) and processed at the Meteorology and Oceanography (MetOc) Centre in Halifax.

REFERENCES

  • Beal, R. C., V. N. Kudryavtsev, D. R. Thompson, S. A. Grodsky, D. G. Tilley, V. A. Dulov, and H. C. Graber, 1997: The influence of the marine atmospheric boundary layer on ERS 1 synthetic aperture radar imagery of the Gulf Stream. J. Geophys. Res., 102, 57995814, https://doi.org/10.1029/96JC03109.

    • Search Google Scholar
    • Export Citation
  • Belkin, I. M., and J. E. O’Reilly, 2009: An algorithm for oceanic front detection in chlorophyll and SST satellite imagery. J. Mar. Syst., 78, 319326, https://doi.org/10.1016/j.jmarsys.2008.11.018.

    • Search Google Scholar
    • Export Citation
  • Chelton, D. B., M. G. Schlax, M. H. Freilich, and R. F. Milliff, 2004: Satellite measurements reveal persistent small-scale features in ocean winds. Science, 303, 978983, https://doi.org/10.1126/science.1091901.

    • Search Google Scholar
    • Export Citation
  • Clausi, D. A., 2002: An analysis of co-occurrence texture statistics as a function of grey level quantization. Can. J. Remote Sens., 28, 4562, https://doi.org/10.5589/m02-004.

    • Search Google Scholar
    • Export Citation
  • Fang, H., T. Xie, W. Perrie, L. Zhao, J. Yang, and Y. He, 2017: Ocean wind and current retrievals based on satellite SAR measurements in conjunction with buoy and HF radar data. Remote Sens., 9, 1321, https://doi.org/10.3390/rs9121321.

    • Search Google Scholar
    • Export Citation
  • Fang, H., T. Xie, W. Perrie, G. Zhang, J. Yang, and Y. He, 2018: Comparison of C-band quad-polarization synthetic aperture radar wind retrieval models. Remote Sens., 10, 1448, https://doi.org/10.3390/rs10091448.

    • Search Google Scholar
    • Export Citation
  • Fang, H., W. Perrie, G. Zhang, T. Xie, S. Khurshid, K. Warner, J. Yang, and Y. He, 2019: Ocean surface wind speed retrieval using simulated RADARSAT constellation mission compact polarimetry SAR data. Remote Sens., 11, 1876, https://doi.org/10.3390/rs11161876.

    • Search Google Scholar
    • Export Citation
  • Hersbach, H., 2010: Comparison of C-band scatterometer CMOD5.N equivalent neutral winds with ECMWF. J. Atmos. Oceanic Technol., 27, 721736, https://doi.org/10.1175/2009JTECHO698.1.

    • Search Google Scholar
    • Export Citation
  • Hersbach, H., A. Stoffelen, and S. de Haan, 2007: An improved C‐band scatterometer ocean geophysical model function: CMOD5. J. Geophys. Res., 112, C03006, https://doi.org/10.1029/2006JC003743.

    • Search Google Scholar
    • Export Citation
  • Johannessen, J. A., R. A. Shuchman, G. Digranes, D. R. Lyzenga, C. Wackerman, O. M. Johannessen, and P. W. Vachon, 1996: Coastal ocean fronts and eddies imaged with ERS 1 synthetic aperture radar. J. Geophys. Res., 101, 66516667, https://doi.org/10.1029/95JC02962.

    • Search Google Scholar
    • Export Citation
  • Kim, T.-S., K.-A. Park, X. Li, A. A. Mouche, B. Chapron, and M. Lee, 2017: Observation of wind direction change on the sea surface temperature front using high-resolution full polarimetric SAR data. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens., 10, 25992607, https://doi.org/10.1109/JSTARS.2017.2660858.

    • Search Google Scholar
    • Export Citation
  • Kirches, G., M. Paperin, H. Klein, C. Brockmann, and K. Stelzer, 2016: GRADHIST—A method for detection and analysis of oceanic fronts from remote sensing data. Remote Sens. Environ., 181, 264280, https://doi.org/10.1016/j.rse.2016.04.009.

    • Search Google Scholar
    • Export Citation
  • Kuang, H.-I., W. Perrie, T. Xie, B. Zhang, and W. Chen, 2012: Retrievals of sea surface temperature fronts from SAR imagery. Geophys. Res. Lett., 39, L10607, https://doi.org/10.1029/2012GL051288.

    • Search Google Scholar
    • Export Citation
  • Legeckis, R., 1978: A survey of worldwide sea surface temperature fronts detected by environmental satellites. J. Geophys. Res., 83, 45014522, https://doi.org/10.1029/JC083iC09p04501.

    • Search Google Scholar
    • Export Citation
  • Mauzole, Y. L., H. S. Torres, and L.-L. Fu, 2020: Patterns and dynamics of SST fronts in the California Current System. J. Geophys. Res. Oceans, 125, e2019JC015499, https://doi.org/10.1029/2019JC015499.

    • Search Google Scholar
    • Export Citation
  • Miller, P., 2009: Composite front maps for improved visibility of dynamic sea-surface features on cloudy SeaWiFS and AVHRR data. J. Mar. Syst., 78, 327336, https://doi.org/10.1016/j.jmarsys.2008.11.019.

    • Search Google Scholar
    • Export Citation
  • Shimada, T., F. Sakaida, H. Kawamura, and T. Okumura, 2005: Application of an edge detection method to satellite images for distinguishing sea surface temperature fronts near the Japanese coast. Remote Sens. Environ., 98, 2134, https://doi.org/10.1016/j.rse.2005.05.018.

    • Search Google Scholar
    • Export Citation
  • Sikora, T. D., and S. Ufermann, 2004: Marine atmospheric boundary layer cellular convection and longitudinal roll vortices. NOAA Synthetic Aperture Radar Marine User’s Manual, 321–330, www.sarusersmanual.com.

    • Search Google Scholar
    • Export Citation
  • Song, Q., P. Cornillon, and T. Hara, 2006: Surface wind response to oceanic fronts. J. Geophys. Res., 111, C12006, https://doi.org/10.1029/2006JC003680.

    • Search Google Scholar
    • Export Citation
  • Sun, X., C. Wang, J. Dong, E. Lima, and Y. Yang, 2018: A multiscale deep framework for ocean fronts detection and fine-grained location. IEEE Geosci. Remote Sens. Lett., 16, 178182, https://doi.org/10.1109/LGRS.2018.2869647.

    • Search Google Scholar
    • Export Citation
  • Vachon, P. W., and F. W. Dobson, 1996: Validation of wind vector retrieval from ERS-1 SAR images over the ocean. Global Atmos. Ocean Syst., 5, 177187.

    • Search Google Scholar
    • Export Citation
  • Xie, T., W. Perrie, and W. Chen, 2010: Gulf Stream thermal fronts detected by synthetic aperture radar. Geophys. Res. Lett., 37, L06601, https://doi.org/10.1029/2009GL041972.

    • Search Google Scholar
    • Export Citation
  • Zakhvatkina, N., A. Korosov, S. Muckenhuber, S. Sandven, and M. Babiker, 2017: Operational algorithm for ice–water classification on dual-polarized RADARSAT-2 images. Cryosphere, 11, 3346, https://doi.org/10.5194/tc-11-33-2017.

    • Search Google Scholar
    • Export Citation
  • Zhang, B., W. Perrie, and Y. He, 2011: Wind speed retrieval from RADARSAT‐2 quad‐polarization images using a new polarization ratio model. J. Geophys. Res., 116, C08008, https://doi.org/10.1029/2010JC006522.

    • Search Google Scholar
    • Export Citation
Save
  • Beal, R. C., V. N. Kudryavtsev, D. R. Thompson, S. A. Grodsky, D. G. Tilley, V. A. Dulov, and H. C. Graber, 1997: The influence of the marine atmospheric boundary layer on ERS 1 synthetic aperture radar imagery of the Gulf Stream. J. Geophys. Res., 102, 57995814, https://doi.org/10.1029/96JC03109.

    • Search Google Scholar
    • Export Citation
  • Belkin, I. M., and J. E. O’Reilly, 2009: An algorithm for oceanic front detection in chlorophyll and SST satellite imagery. J. Mar. Syst., 78, 319326, https://doi.org/10.1016/j.jmarsys.2008.11.018.

    • Search Google Scholar
    • Export Citation
  • Chelton, D. B., M. G. Schlax, M. H. Freilich, and R. F. Milliff, 2004: Satellite measurements reveal persistent small-scale features in ocean winds. Science, 303, 978983, https://doi.org/10.1126/science.1091901.

    • Search Google Scholar
    • Export Citation
  • Clausi, D. A., 2002: An analysis of co-occurrence texture statistics as a function of grey level quantization. Can. J. Remote Sens., 28, 4562, https://doi.org/10.5589/m02-004.

    • Search Google Scholar
    • Export Citation
  • Fang, H., T. Xie, W. Perrie, L. Zhao, J. Yang, and Y. He, 2017: Ocean wind and current retrievals based on satellite SAR measurements in conjunction with buoy and HF radar data. Remote Sens., 9, 1321, https://doi.org/10.3390/rs9121321.

    • Search Google Scholar
    • Export Citation
  • Fang, H., T. Xie, W. Perrie, G. Zhang, J. Yang, and Y. He, 2018: Comparison of C-band quad-polarization synthetic aperture radar wind retrieval models. Remote Sens., 10, 1448, https://doi.org/10.3390/rs10091448.

    • Search Google Scholar
    • Export Citation
  • Fang, H., W. Perrie, G. Zhang, T. Xie, S. Khurshid, K. Warner, J. Yang, and Y. He, 2019: Ocean surface wind speed retrieval using simulated RADARSAT constellation mission compact polarimetry SAR data. Remote Sens., 11, 1876, https://doi.org/10.3390/rs11161876.

    • Search Google Scholar
    • Export Citation
  • Hersbach, H., 2010: Comparison of C-band scatterometer CMOD5.N equivalent neutral winds with ECMWF. J. Atmos. Oceanic Technol., 27, 721736, https://doi.org/10.1175/2009JTECHO698.1.

    • Search Google Scholar
    • Export Citation
  • Hersbach, H., A. Stoffelen, and S. de Haan, 2007: An improved C‐band scatterometer ocean geophysical model function: CMOD5. J. Geophys. Res., 112, C03006, https://doi.org/10.1029/2006JC003743.

    • Search Google Scholar
    • Export Citation
  • Johannessen, J. A., R. A. Shuchman, G. Digranes, D. R. Lyzenga, C. Wackerman, O. M. Johannessen, and P. W. Vachon, 1996: Coastal ocean fronts and eddies imaged with ERS 1 synthetic aperture radar. J. Geophys. Res., 101, 66516667, https://doi.org/10.1029/95JC02962.

    • Search Google Scholar
    • Export Citation
  • Kim, T.-S., K.-A. Park, X. Li, A. A. Mouche, B. Chapron, and M. Lee, 2017: Observation of wind direction change on the sea surface temperature front using high-resolution full polarimetric SAR data. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens., 10, 25992607, https://doi.org/10.1109/JSTARS.2017.2660858.

    • Search Google Scholar
    • Export Citation
  • Kirches, G., M. Paperin, H. Klein, C. Brockmann, and K. Stelzer, 2016: GRADHIST—A method for detection and analysis of oceanic fronts from remote sensing data. Remote Sens. Environ., 181, 264280, https://doi.org/10.1016/j.rse.2016.04.009.

    • Search Google Scholar
    • Export Citation
  • Kuang, H.-I., W. Perrie, T. Xie, B. Zhang, and W. Chen, 2012: Retrievals of sea surface temperature fronts from SAR imagery. Geophys. Res. Lett., 39, L10607, https://doi.org/10.1029/2012GL051288.

    • Search Google Scholar
    • Export Citation
  • Legeckis, R., 1978: A survey of worldwide sea surface temperature fronts detected by environmental satellites. J. Geophys. Res., 83, 45014522, https://doi.org/10.1029/JC083iC09p04501.

    • Search Google Scholar
    • Export Citation
  • Mauzole, Y. L., H. S. Torres, and L.-L. Fu, 2020: Patterns and dynamics of SST fronts in the California Current System. J. Geophys. Res. Oceans, 125, e2019JC015499, https://doi.org/10.1029/2019JC015499.

    • Search Google Scholar
    • Export Citation
  • Miller, P., 2009: Composite front maps for improved visibility of dynamic sea-surface features on cloudy SeaWiFS and AVHRR data. J. Mar. Syst., 78, 327336, https://doi.org/10.1016/j.jmarsys.2008.11.019.

    • Search Google Scholar
    • Export Citation
  • Shimada, T., F. Sakaida, H. Kawamura, and T. Okumura, 2005: Application of an edge detection method to satellite images for distinguishing sea surface temperature fronts near the Japanese coast. Remote Sens. Environ., 98, 2134, https://doi.org/10.1016/j.rse.2005.05.018.

    • Search Google Scholar
    • Export Citation
  • Sikora, T. D., and S. Ufermann, 2004: Marine atmospheric boundary layer cellular convection and longitudinal roll vortices. NOAA Synthetic Aperture Radar Marine User’s Manual, 321–330, www.sarusersmanual.com.

    • Search Google Scholar
    • Export Citation
  • Song, Q., P. Cornillon, and T. Hara, 2006: Surface wind response to oceanic fronts. J. Geophys. Res., 111, C12006, https://doi.org/10.1029/2006JC003680.

    • Search Google Scholar
    • Export Citation
  • Sun, X., C. Wang, J. Dong, E. Lima, and Y. Yang, 2018: A multiscale deep framework for ocean fronts detection and fine-grained location. IEEE Geosci. Remote Sens. Lett., 16, 178182, https://doi.org/10.1109/LGRS.2018.2869647.

    • Search Google Scholar
    • Export Citation
  • Vachon, P. W., and F. W. Dobson, 1996: Validation of wind vector retrieval from ERS-1 SAR images over the ocean. Global Atmos. Ocean Syst., 5, 177187.

    • Search Google Scholar
    • Export Citation
  • Xie, T., W. Perrie, and W. Chen, 2010: Gulf Stream thermal fronts detected by synthetic aperture radar. Geophys. Res. Lett., 37, L06601, https://doi.org/10.1029/2009GL041972.

    • Search Google Scholar
    • Export Citation
  • Zakhvatkina, N., A. Korosov, S. Muckenhuber, S. Sandven, and M. Babiker, 2017: Operational algorithm for ice–water classification on dual-polarized RADARSAT-2 images. Cryosphere, 11, 3346, https://doi.org/10.5194/tc-11-33-2017.

    • Search Google Scholar
    • Export Citation
  • Zhang, B., W. Perrie, and Y. He, 2011: Wind speed retrieval from RADARSAT‐2 quad‐polarization images using a new polarization ratio model. J. Geophys. Res., 116, C08008, https://doi.org/10.1029/2010JC006522.

    • Search Google Scholar
    • Export Citation
  • Fig. 1.

    Location of the study area and 30 RADARSAT-2 SAR images with SST fronts.

  • Fig. 2.

    NRCS values in VV polarization for the SAR images at (top left) 1016:29 UTC 15 Sep 2008 (S1), (top right) 2206:58 UTC 25 Oct 2008 (S2), (bottom left) 2210:58 UTC 28 Apr 2009(S3), and (bottom right) 2211:34 UTC 22 May 2009(S4). RADARSAT-2 data and products from MacDonald, Dettwiler, and Associates Ltd., all rights reserved.

  • Fig. 3.

    Wind speed retrieved from VV-polarized NRCS.

  • Fig. 4.

    Texture features revealed from the values for homogeneity, energy, correlation, and contrast, for wind speed derived from SAR image S2, using Eqs. (4)(7).

  • Fig. 5.

    Correlation values for wind speed corresponding to SAR images S1–S4.

  • Fig. 6.

    Correlation values for the NRCSs in VV polarization of the SAR images.

  • Fig. 7.

    SST fronts detected from SAR images, overlapped on blended SST data: (top left) 3-day-averaged SST data (S1), (top right) 15 h earlier (S2), (bottom left) 4 h later (S3), and (bottom right) 4 h earlier (S4).

All Time Past Year Past 30 Days
Abstract Views 775 229 0
Full Text Views 298 180 44
PDF Downloads 318 199 48