Integration of statistical modelling and multi-source remote sensing data for automatic and robust oil spill detection

Lead Research Organisation: Newcastle University
Department Name: Sch of Computer Science

Abstract

An Oil Spill is a form of marine pollution which can have a serious impact on the environment as well as to commercial interests. Timely information on the location, extent and movement of an oil slick can be used to effectively manage the response. For oil spill response, satellite imagery provides information that can be used to support various missions, including assessing the initial (and potential future) extent and impact of a spill, planning response operations and monitoring the effectiveness of the response as a whole. In addition, the free, full and open access to ESA Sentinel-1A/1B Synthetic Aperture Radar sensor data and Sentinel-2A/2B optical sensor data has been revolutionising the way remote sensing is used in several sectors. Due to their very short revisit times, high spatial resolution, and good spectral resolution these sensors make it possible for us to detect oil spills in an automatic way.
A commonly used sensor for oil spill detection is the Synthetic Aperture Radar (SAR). SAR is an active microwave radar and is particularly prevalent due to its all-weather and all-day imaging capabilities. SAR images have been shown to provide a good source for identifying oil slicks, as they show up clearly as Dark Formations in the image. Unfortunately, oil spills are not the only phenomena that may produce a Dark Formation in a SAR image. Phenomena such as low wind conditions, algae blooms and natural films all produce similar Dark Formations. This makes deciding whether a Dark Formation has been produced by an oil spill or another phenomenon a complex task. Much of the recent literature focuses on using sophisticated classification methods to improve the classification accuracy of oil spill detection from SAR imagery. In addition to this, it has been suggested the use of multi-source data, such as the combination of optical and SAR imagery, to improve oil spill detection systems.
This project aims to combine spatial-temporal models under a functional data analysis framework with remote sensing techniques to develop an automatic and robust oil spill detection system with ESA's Sentinel-1/2 data. Specific objectives are as follows:
* To develop spatial-temporal models under functional data analysis framework to integrate data from different sources and combine models with remote sensing techniques to develop classification/clustering tools.
* To accurately discriminate between oil-spills and look-alike phenomena. This is believed to be the biggest challenge in detecting oil spills from Synthetic Aperture Radar imagery. Optical imagery will be used to assist in this process.
* To automatically analyse oil spills three types of information should be extracted. Firstly, the location, extent and surface area of the spills. Secondly, the response resource of the spills. Finally, the evolution of the spills.

Publications

10 25 50

Studentship Projects

Project Reference Relationship Related To Start End Student Name
EP/R512588/1 01/10/2017 30/09/2021
1948804 Studentship EP/R512588/1 01/10/2017 30/09/2021 Julian George Austin