Abstract: Application of the Unsupervised Semi-variogram Textural Classifier (USTC) for the Detection of Natural Oil Seeps Using RADARSAT-1 Data Obtained Offshore the Amazon River Mouth, Brazil

Miranda, F. P. and C. M. Bentz - Petrobras/Cenpes; C.H. Beisl - Gorceix Foundation; J.A. Lorenzetti and C.ES. Araújo - INPE; C.L. Silva Jr. - Oceansat

The remote detection of natural oil seeps is invaluable for efficient planning of expensive exploration activities in deepwater frontier areas. This paper describes a pilot project offshore the Amazon River mouth. The region is cloud-covered all year round, which poses difficulties for the sole use of remote sensing systems in the visible and infrared portion of the electromagnetic spectrum. Spaceborn radars are therefore the ideal choice, since they are only weakly affected by weather conditions, and are able to detect natural and man-made oil slicks.

Oil slicks are normally identified as dark areas of low radar backscatter. However, such a phenomenon is not unique to the presence of oil. False targets include low wind zones or shadow zones behind obstacles, natural surfactants, local upwelling, shallow marine vegetation, and areas of heavy rainfall. Ambiguities in oil slick detection using radar systems can be tentatively resolved with the aid of ancillary information. RADARSAT-1 images in the ScanSAR Narrow (SN1) beam mode were used in this investigation; data acquisition took place on 25 March, 18 April, 12 May, 29 June, and 23 July, 1997. Ancillary information was derived from existing tectonic maps and from Advanced Very High Resolution Radiometer (AVHRR) temperature maps.

The unsupervised semi-variogram textural classifier (USTC) was selected as the classification algorithm to enhance ocean surface features using RADARSAT-1 data. This is a deterministic classifier that combines textural and radiometric information. Textural information is described by the shape and value of the circular semi-variogram function; it is also described by the Digital Number (DN) variance in a circular neighborhood defined by a very large lag distance. The radiometric information is conveyed by the despeckled DN value. The chosen speckle reduction algorithm was the median filter. Unsupervised classification was carried out using the isodata clustering algorithm. After unsupervised classification, the results obtained from the clustering program were merged using interactive class aggregation (an aggregate is a grouping of one or more classes considered to be of thematic significance).

Results of USTC classification highlighted areas of rough texture mostly associated with internal waves. Areas of smooth texture were also discriminated from an intermediate background and were interpreted as low wind zones, natural surfactants or oil slicks. Thin, elongated patches of smooth image texture were considered natural oil seeps whenever adjacent to normal (listric) and strike-slip faults.

AAPG Search and Discovery Article #90933©1998 ABGP/AAPG International Conference and Exhibition, Rio de Janeiro, Brazil