Detection and classification of oil spills in MODIS satellite imagery

Abstract

Using satellite imagery to achieve an early and accurate identification of oil spills will contribute towards the reduction of their impact on the marine ecosystem. Satellite imagery provided by the synthetic aperture radar (SAR) sensors are widely used for this task over the multi-temporal and multi-band visible near infra-red (VNIR) sensors. This is due to the SAR imaging capabilities through clouds, dust storms, soot and at night times, which limit the capability of VNIR sensors. However, gaps in knowledge exist regarding whether satellite ocean-colour sensors are capable of identifying unreported oil spills as true positives and whether they are able to discriminate them from lookalikes with the least uncertainty, particularly in arid land regions characterised with nearly cloud-free conditions. It was therefore, the goal of this research to develop reliable and robust methodology for data processing and interpretation of oil spills observed by VNIR sensors.

The Moderate Resolution Imaging Spectroradiometer (MODIS) is a VNIR-type sensor that was selected for this project for a number of reasons: it is characterised with adequate multi-spectral features (36 spectral bands 0.405-14.385 ?m) spread over three spatial resolutions (250, 500 and 1000 m); and its data is freely distributed in near-realtime. MODIS bio-geophysical products processed in this study such as sea surface temperature (SST4 and SST) and chlorophyll-a (Chlor-a) have also proven their usefulness in providing complementary data.

As a result of this investigation, two methods were proposed: The spectral contrast shift (SCS) and the surface algal bloom index (SABI).

The SCS identifies oil spills and classifies their thickness by using MODIS extreme (maximum and minimum) top-of-atmosphere radiance (TOA) values in the 250 m/pixel resolution bands: the red (?1=645 nm) and the NIR (?2 =859 nm) measured over a relatively small area selected to encompass part of an unknown class and part of the surrounding pure sea water. The method has produced consistent and highly sensitive results independent of sun-glint illuminations. Oil spills have SCS values lying within the range 0.02-0.04±0.002 varying by 0.01 corresponding to different thicknesses of oil. The SCS succeeded also in classifying surface floating blooms having SCS values greater than or equal to 0.20.

The SABI is a four-band relationship, which according to MODIS 500 m/pixel resolution, is made up of the difference between the TOA radiance responses in the NIR and the red bands (aggregated from the 250 m resolution group) to the sum of the TOA radiance responses in the blue (?3=469 nm) and green (?4=555 nm) bands. The SABI aims to discriminate biological floating species that may appear as an oil spill look-alike without the need to perform complex corrections for atmosphere and sun-glint effects. The SABI succeeded in classifying 95% of surface blooms that had values greater than or equal to a baseline value of -0.10. Oil spills, however, always appear at values lower than the surface bloom baseline value.

More information

Identifiers

Catalogue record

Export record