vw. 
HYPERSPECTRAL PARTIAL UNMIXING TECHNIQUE FOR OIL SPILL 
TARGET IDENTIFICATION 
Foudan Salem,? Menas Kafatos ^ 
? Research Scientist, fsalem@scs.gemu.edu 
p Director/Dean, mkafatos(@compton.smu.edu 
Center for Earth Observing and Space Research 
4400 University Dr. (MSN5C3) 
SCS / George Mason University, Fairfax, VA, 22030, USA 
KEY WORDS: Hyper-spectral, Pollution, Imagery, Remote sensing, Sea, Monitoring, Detection. 
ABSTRACT 
In this study, advanced techniques for oil spill detection and oil spill type identification using hyperspectral AVIRIS data are 
presented. Spectrally complex areas are highly dimensional and are consequently difficult to fully unravel. In our previous study, 
the Spectral Linear Unmixing (SLU) technique showed many limitations for operational applications, since signatures for all the 
target materials in the scene must be spectrally identified. Our new methodology emphasizes the ability to distinguish oil slicks from 
the background using the Partial Unmixing (PU) technique. Both the data reduction and the pixels projection methods are used for 
distinguishing thick, slick oil from dispersed oil; moderate and thin oil sheens; polluted water; and tarry oil. It was developed in part 
to partially un-mix the oil target pixels from the background mixed pixels. This method improve on the SLU technique because it 
dose not require prior knowledge of the background material spectral signatures. Our analysis applies to oil spill targets with the 
assumption that all pixels are pure and they are not mixed with background materials. In the specific case of the Santa Barbara 
coastal zone event (March, 2002), the changes in the oil slick occurred from the north (oil spill source) to the south due to the high 
sea waves and strong current effects. 
Our study is focusing on target identification for oil slick. We show that oil spill on sea water can be clearly identified. 
1. INTRODUCTION 
The classification of oil is extremely complicated due to the 
variance in the optical properties of different oil spill types. 
Sea waves currently lack positive discrimination and cause poor 
contrast and mixing of many oil spill types. There is difficulty 
in optically identifying thick oil slicks spectra from streaks and 
oiled water.. Also, it is difficult some times to optically identify 
oil slick spectral signatures for oil spills on the scene. A 
complete spectral mixing of a complicated AVIRIS scene may 
not always be possible or even desired. High-quality data of 
spectrally complex areas are very high dimensional and 
difficult to fully separate. There is a need for a more selective 
method to increase the ability to identify regions of interest for 
the desired regions. Therefore, using more advanced 
techniques such as the Partial Unmixing (PU) is very efficient 
for increasing the reliability of the analysis. 
The improved signal to noise AVIRIS data complemented by 
new data reduction and processing techniques permits 
unambiguous oil identification and spectral unmixing of 
subpixel targets; subtle spectral differences enhanced in the 
data include oil types and polluted water discrimination. This 
allows the detailed detection of smaller oil spill areas. The 
techniques developed so far classify oil spills and verify their 
effectiveness experimentally, which in turn will make it 
possible to model water-leaving radiances from different types 
of oil slicks. Analysis methods focus on classifying each pixel 
into a single class by identifying the main material in the pixel 
(Richard, et al. 2002). 
Our new methodology emphasizes the ability to distinguish oil 
slicks from the background using the Partial Unmixing (PU) 
technique. It was developed in part to partially un-mix the oil 
target pixels from the background mixed pixels. Our model 
focuses on distinguishing the abundance of targets under 
investigation from background features. The PU techniques are 
used to identify oil spill targets in the presence of a complex 
background and when there is no ground truth information. 
In practice, with multispectral techniques, one method alone is 
not conclusive in all oil spill detection (Goodman and Fingas, 
1988). Often, oil has no specific characteristics that distinguish 
it from the background. Taylor (1992) studied oil spectra in the 
laboratory and field and observed flat spectra with no usable 
features distinguishing it from the background (Taylor, 1992). 
Techniques that separate specific spectral regions did not 
increase detection capability. 
1.1 Case Study: Santa Barbara 
Santa Barbara County is home to the most intensive offshore oil 
development on the West Coast. For decades, Santa Barbara 
County has been sensitive to offshore oil drilling. In 1969, 
California's biggest oil spill fouled the Santa Barbara Channel 
with about four million gallons of crude oil. Moreover, at least 
10,000 gallons of oil have been spilled from an undersea pipe 
near Santa Barbara. As oil production continues offshore, 
tourism has grown dramatically, increasing the threat of oil- 
related injuries. 
1.2 AVIRIS Data Set 
1329