Mapping oil spills on sea water using spectral mixture analysis of hyperspectral image data Javier Plaza * , Rosa Pérez, Antonio Plaza, Pablo Martínez and David Valencia Neural Networks and Signal Processing Group (GRNPS), Computer Science Department, University of Extremadura, Avda. de la Universidad S/N, 10071 Cáceres, SPAIN. ABSTRACT During the last years, several terrestrial ecosystems have suffered from large spill oil events threatening coastal habitats and species. Some recent examples include the 2002 Prestige tanker oil spill in Galicia, Northern Spain, as well as repeated oil spill leaks evidenced in the Santa Barbara coastline in California, and the Patuxent river (Chesapeake watershed) in Maryland. Both spaceborne and airborne hyperspectral sensors allow detailed identification of materials, and very accurate (sub-pixel) estimates of their fractional abundance covers. In the event of an oil spill, the information produced by remotely sensed hyperspectral instruments can be used to design an effective environmental oil spill protection and response plan, which could help to reduce the environmental consequences of the spill and cleanup efforts, as well as to protect human life. In this paper, we discuss a novel automated hyperspectral target detection technique for determining the level of oil contamination of polluted areas in the shoreline. The method is based on the simultaneous use of spatial and spectral information by extended mathematical morphology operations. Both simulated and real hyperspectral data, collected over polluted areas, are used in this work to illustrate the effectiveness of the proposed approach. Keywords: Standoff detection, Oil spill detection and mapping, Hyperspectral imagery, Morphological analysis. 1. INTRODUCTION Large spills of oil and related petroleum products in the marine environment can have serious biological and economic impacts 1 . Public and media scrutiny is usually intense following a spill, with demands that the location and extent of the oil spill be identified. Hyperspectral remote sensing is playing an increasingly important role in oil spill response efforts 2,3,4 . Recent improvements in sensor technology, space power, computers, pattern recognition algorithms, and communication systems suggest that efficient standoff detection and identification systems are feasible 5 . These systems involve passive and active methods for sensing of chemical and biological materials when the sensor is physically separated from the site of interest. Nearly any chemical or biological element can be a pollutant, meaning that in large enough quantities it has the potential for causing ecological damage. In the event of an oil spill on sea water, fast and accurate determination of hazard areas is needed, particularly if monitoring large quantities of oil spilled. During the last years, several terrestrial ecosystems have suffered from large spill oil events threatening coastal habitats and species. Some recent examples include the 2002 Prestige tanker oil spill in Galicia, Northern Spain 6 (see Fig. 1), as well as repeated oil spill leaks evidenced in the Santa Barbara coastline in California, and the Patuxent river (Chesapeake watershed) in Maryland. Chemical measurements of man’s influences on coastal environments, such as those mentioned above, fall into three general classes: 1) the assessment of the major chemical constituents of sea water including salt, dissolved oxygen, major nutrients (nitrogen, phosphate, silicate) and carbon; 2) the quantification of trace elements, principally metals, in the water, the bottom, and the sea life; and 3) the measurement of pollutant hydrocarbons including synthetic organics and petroleum hydrocarbons. The result of many such measurements is to determine where the nutrients and pollutants are in the ecosystem. In order to discover whether an area is polluted or otherwise influenced, its chemical and biological characteristics must be compared with some area that is as similar as possible but seems not to have been affected by man. These comparative sites are called reference or control stations 1 . Generally, it is sensible to have as many reference sites as possible so that a range of normal values is available for comparison. * E-mail: aplaza@unex.es; Phone: +34 927257195; Fax: +34 927257203. Chemical and Biological Standoff Detection III, edited by James O. Jensen, Jean-Marc Thériault, Proc. of SPIE Vol. 5995, 599509, (2005) · 0277-786X/05/$15 · doi: 10.1117/12.631149 Proc. of SPIE Vol. 5995 599509-1