Anais XII Simpósio Brasileiro de Sensoriamento Remoto, Goiânia, Brasil, 16-21 abril 2005, INPE, p. 4489-4491.
Change detection target by semivariogram textural classification added to practical
application of wind speed retrieval using RADARSAT-1/SAR imagery
Luciana de Souza Soler
Coordenadoria de Pesquisa e Pós-Graduação em Engenharia Civil ,Universidade Federal do
Rio de Janeiro(COPPE/UFRJ), Rio de Janeiro –RJ, Brasil
lsoler@coc.ufrj.br
Abstract. Coastal surveillance activities have increased in recent years, so accurate information as target
detection and wind speed are required to oil spill contingency plans. The main goal is to test practical
applications to SAR wind speed retrieval algorithms into a developed controlling method for oil rigs positions. A
semivariogram textural classifier (STC) was applied in RADARSAT-1/SAR images of ScanSAR Narrow and
Extended Low 1 beam modes. The comparison between the measured position of fixed and floating targets and
the data provided by official organizations was done to validate the methodology. Limitations regarding wind
vector field obtained by atmospheric modeling can be connected to poor resolution. Thus, SAR wind speed
retrieval algorithms have been applied to the images in order to get a more appropriate resolution to coastal
applications.
Keywords: target detection, SAR, wind speed, detecção de alvos, SAR, velocidade do vento.
1. Introduction
Coastal surveillance and environmental monitoring has motivated the development of
automatized feature extraction tools using remote sensing data. Activities of intrinsic risk as
oil exploitation and transportation demand appropriate contingency plans for emergency
situations as well as illegal oil spills. Hence, there is a need of controlling the fixed, floating
and moving targets position in exploitation areas, so that one can correctly connect them to
eventual oil spills.
Target detection by Synthetic Aperture Radar (SAR) has been extensively studied in
recent years, and several techniques can be found in literature (Li and Zenio, 1996; Perlovsky
et al., 1997; Howard et al., 1999). Among these techniques, the use of contextual information
added to radiometric values of each pixel provides a more complete scenario for digital
classification algorithms. Considering this, a contextual classifier based on semivariogram
function was applied to a set of ten multitemporal SAR images in order to detect fixed,
floating and moving target on the sea (Soler and Landau, 2003).
Although efficient, the contextual targets classification method has presented limitations
regarding to incident angles of radar systems and high wind speed occurrence. Some images
classification resulted in a poor discrimination of targets which was connected to lack of
accuracy in geometrical correction. However, this can be due to high wind speed occurrence
which was not revealed by the wind field modeled data used considering its poor spatial
resolution. Thus, SAR wind speed retrieval algorithms have been applied to the images in
order to get a more appropriate resolution to coastal applications.
2. Study Area
Located along the coast of Rio de Janeiro and Espírito Santo States, Campos Basin region
keeps 86.9% of the total Brazilian oil and gas reserves (ANP, 2002). Exploitation activities in
Campos are very significant in deep waters (depths greater than 1000 m). The most important
oceanographic feature present in the area is the predominantly southward flux of Brazil
Current (BC). However, BC may take different flow directions along Campos Basin due to
the submarine elevation of Abrolhos and the shelf break offshore of Capes São Tomé and
Frio. These discontinuities can cause flow instability and origin meanders and eddies with
4489
Anais XII Simpósio Brasileiro de Sensoriamento Remoto, Goiânia, Brasil, 16-21 abril 2005, INPE, p. 4489-4491.
diameters between 50 and 200 km that may last for weeks. These features in SAR images can
be misclassified as oil slicks, so ancillary data as wind at ocean surface and atmospheric
conditions must be considered. Beyond this, the constant passage of cold fronts during
wintertime causes a significant change in wind regime at ocean surface along the coast (Stech
and Lorenzzetti, 1992). However, these changes can not be detect by wind scatterometer data
and buoys sampling due to their coarse spatial resolution, so wind field retrieval by SAR
may diminish these limitations (Korsbakken et al. 1998, Kim and Moon 2002).
3. Methodology
In the first phase of the methodology, a set of ten RADARSAT-1/SAR images was used in
the oil rigs detection study. The images specifications, the classification results and the wind
speed observed in modeled data are presented in Table 1. The ScanSAR Narrow A and B
beam modes used (SNAR and SNB respectively) present image swaths of 300 km and 50 m
of nominal resolution.
Table 1. Images specifications with the numbers of fixed, floating and moving targets
identified in each image and the wind speed interval observed in the modeled data.
Date and beam
mode
25/03/2002
SNAR
13/04/2002
SNB
18/04/2002
SNAR
12/05/2002
SNAR
19/05/2002
SNB
01/08/2002
SNAR
12/10/2002
SNAR
Orbit
Fixed targets
Floating
targets
Moving targets
Wind speed
interval (m/s)
Descending
40
10
58
3<|v|<8
Ascending
41
10
62
3<|v|<8
Descending
40
10
30
2<|v|<6
Descending
32
10
37
5<|v|<7
Descending
42
10
60
6<|v|<8
Ascending
34
07
54
6<|v|<9.5
Descending
28
07
29
8<|v|<13
After the pre-processing steps of geometrical corrections and speckle filtering, the
Semivariogram Textural Classifier (STC) was applied to images. Detailed information about
the classification procedure can be seen in Miranda and Carr, (1994) Soler and Landau
(2003).
In the second phase of the methodology a comparative analysis will be done between
wind speed obtained by atmospheric modeling (which values are show in Table 1) and the
one estimated by the same SAR images used to oil rigs distinguishing. Wind speed retrieval
algorithm CMOD4 (Stoffelen ans Anderson, 1997) has been applied to SAR images.
The CMOD4 algorithm was initially developed to ERS-1 Scaterometer wind retrieval, but
significant results has been reached when used with the same end in SAR images
(Wackerman et al. 1996, Korsbakken et al. 1998, Vachon et al. 2000). Radar scatterometer as
SeaWinds/QuikSCAT-1 provides global wind vector field at spatial resolution of 50 km with
an accuracy of 2.0 m/s in speed, 20° in direction and a directional ambiguity of 180°. On the
other hand, by using SAR images to wind vector field estimation, one can reach a spatial
resolution of 10 km, an accuracy of 1.5 m/s in speed and 33° in direction. Although SAR
wind vector field estimation also results in a directional ambiguity of 180°, it can be solved by
using external data as the data used in the first phase of the present study or scatterometer
data.
4490
Anais XII Simpósio Brasileiro de Sensoriamento Remoto, Goiânia, Brasil, 16-21 abril 2005, INPE, p. 4489-4491.
In the third phase of the study, it is intended to apply the above detailed methodology in
Mexican Gulf where SAR images and in situ measurements of wind intensity and direction
will be used. Thus, not only will it be possible to estimate wind speed, but also wind
direction. It is intended to compare CMOD4 algorithm to CMOD_IFR2 (Kim and Moon,
2002) and to used new computational techniques as neural networks to obtain the wind speed
as an improvement of actual algorithms used.
4. Previous and Expected Results
The occurrence of wind speed higher than 8 m/s was a determinant aspect to images from
August 1st and October 12th present a lower rate of detection in both: fixed and floating
targets. It is expected to confirm this range limitation by using a great number of images in
Mexican Gulf.
The classification result of image from May 12th showed a noticeable diminishing of
fixed targets. When compared to the other ScanSAR Narrow images, the expressive RMS
error in geometrical correction (about of 50%) may be associated to the poor discrimination of
features. However, a detailed comparison between wind vector fields must be done in order to
evaluate possible errors in atmospheric modeling. It is expected this comparison reveals the
efficiency of wind vector field estimated by SAR images.
References
Agência Nacional do Petróleo (ANP). Anuário Estatístico Brasileiro do Petróleo e Gás Natural. Rio de
Janeiro: ANP, 2002. Disponível na biblioteca digital URLib:
<http://www.anp.gov.br/petro/dados_estatisticos.asp>. Acesso em: 30 jun. 2002.
Howard, D; Roberts, S.C.; Brankin, R. Target detection in SAR imagery by genetic programming. Advances in
Engineering Software, v. 30, p. 303-311, 1999.
Kim, D.; Moon, W.M. Estimation of sea surface wind vector using RADARSAT data. Remote Sensing of
Environment, v. 80, p. 55– 64, 2002.
Korsbakken, E.; Johannessen, J. A.; Johannessen, O. M. Coastal Wind Field Retrievals from ERS SAR Images.
Journal of Geophysical Research, v. 103, n. C4, p. 7857-7874, 1998.
Li, J.A.; Zelnio, E.G. Target detection with synthetic aperture radar. IEEE Trans. on Aerospace and
Electronic Systems, v. 32, p. 613-627, 1996.
Miranda, F.P.; Carr, J.R.. Application of the Semivariogram Textural Classifier (STC) for vegetation
discrimination using SIR-B data of the Guiana Shield, northwestern Brazil. Remote Sensing Reviews, v. 10, p.
155-168, 1994.
Perlovsky, L.I.; Schoendorf, W.H.; Burdick, B.J.; Tye, D.M. Model-based neural network for target detection in
SAR images, IEEE Trans. on Image Processing, v. 6, p. 203-216, 1997.
Stech, J.L.; Lorenzzetti, J.A. The response of the South Brazil Bight to the passage of wintertime cold fronts.
Journal of Geophysical Research, v. 97, n. C6, p. 9507-9520, 1992.
Soler, L.S.; Landau, L. Change detection target in RADARSAT-1/SAR imagery by semivariogram textural
classification. In: 30th International Symposium on Remote Sensing of Environment, 2003, Honolulu.
Proceedings. [CDRom].
Stoffelen, A.; Anderson, D.L.T. Scatterometer data interpretation: Estimation and validation of the transfer
function CMOD4. Journal of Geophysical Research, v. 102, n. C3, p. 5767-5780, 1997.
Vachon, P. W., Dobson, F. W. Wind retrieval from RADARSAT SAR images: selection of a suitable C-band
HH polarization wind retrieval model. Canadian Journal of Remote Sensing, v. 26, n. 4, p. 306–313, 2000.
Wackerman, C. C., Rufenach, C. L., Shuchman, R. A., Johannessen, J. A.,Davidson, K. L. Wind vector retrieval
using ERS-1 synthetic aperture radar imagery. IEEE Transactions on Geoscience and Remote Sensing, v. 34,
p. 1343– 1352, 1996.
4491