Skip to main content


Title: Use Of Pattern Recognition To Identify The Source Of An Oil Spill On An Inland Water
Author: W Whalley
Author: P Robotham
Author: M O'Connor
Author: Environment Agency
Document Type: Monograph
Annotation: Environment Agency Project ID:EAPRJOUT_355, Representation ID: 89, Object ID: 1699
This Technical Report presents, the findings~ of National. R and D Project El-050 aThe Application of Neural Networks to Pattern Recognition and Comparison - A Feasibility to neural networks and made no Studya. Although the projectas title specificallyreferred mention oil ;spills, its principal aim was clearly to determine the feasibility of using -pattern recognition teclmiques to identify the. source. of an oil spill using chromatographic / spectrographic data;- It was agreed at an early stage that the scope .of the study be expanded to include statistical methods of pattern recognition, since it was found that- far more.research work has been done in this field using these meth0ds.tha.n using neural networks. A comprehensive literature search was carried out to -determine the state-of-the-art and .the extent to which currently available software. might meet the Agencyas needs. Several key researchers and consultants world-wide were contacted to seek their opinions and advice. In addition,: a desk-top study of the basic nature of-the pattern recognition task posed by this spec and problem was carried out. The conclusions drawn from these initial studies were that: l much work has been done using statistical methods of pattern recognition to identify the source of a crude oil spill at sea, and that a proven computer package (EUROCRUDE) is now available for this purpose; l there are presently .no computer.packages based on neural networks that would be suitable for use by.Agency staff.for routine chemical fingerprinting of oil samples; l the problem of identifjkg the source of a refined oil spill on an inland water is sufEiciently difberent from that posed by a crude oil spill at sea as to require a different solution; and l the current state-of-the-art makes it feasible to develop a source identificatiorrpackage for oil spills on inland waters using. computer-based pattern recognition techniques. A preliminary investigation, based on data supplied by the Agency, was carried out to test the viability of a-potential solution tothe source identification ,problem The proposed solution involves a two stage process: firstly, to determine the type of retied oil responsrble for the spill by comparing the Gngerprints of the field samples with those of a range of standard reiined oils; and secondly, to identify the source of the spill by .comparing specific features of the f%rgerprints~ of the field samples with the corresponding features in the fingerprints of samples taken from potential sources. The viability of the first stage was not inadoubt; since there was ample evidence in the literature to show that it can be done: However, the proposed second stage involved a novel approach, so a study was carried out. to test its viability using data:fiom a specific case study. This study was based on a visual pattern recognition exercise in which 20 voluteers attempted to match-patterns in localised sections of the GC traces of. field samples: with-: those in the ~corresponding sections of samples derived from .-several potential sources. The high degree of .successachieved by the. participants in identifying. the true source clearly indicated that an algorithm based on the matching of localised features could provide the basis for an effective computer-based source identification system Recommendations have been made for a phased programme of research to. develop a comprehensive source identification-system that would improve the reliability.and defensibility of evidence present in court. Keywords: river, pollution, liability, oil spills, chromatography, fingerprinting, pattern recognition, neural networks, chemometrics. R and D Technical Report E72 vii 1.
Publisher: Environment Agency
Subject Keywords: Oil spill; Pattern recognition; Inland water
Extent: 97
Total file downloads: 11

Download PDF    Display PDF in separate tab