Graduation Year


Document Type




Degree Name

Doctor of Philosophy (Ph.D.)

Degree Granting Department

Marine Science

Major Professor

Chuanmin Hu, Ph.D.

Committee Member

Kendra L. Daly, Ph.D.

Committee Member

Mark E. Luther, Ph.D.

Committee Member

Ian R. MacDonald, Ph.D.

Committee Member

David A. Palandro, Ph.D.


multiband, oil detection, oil emulsion, oil spill assessment, sun glint, thickness classification


Assessment of oil spills in the ocean using passive remote sensing (i.e., reflected sun light) faces two challenges: detect oil presence/absence and quantify oil volume. While the optical properties of oil allow it to be differentiated from the surrounding marine environment, sun glint can facilitate oil presence/absence detection because the oil-water spatial contrast is enhanced due to wave dampening. However, sun glint also modulates the magnitude and shape of the spectral reflectance of surface oil. In addition to this difficulty, the most critical challenge is how to quantify oil volume (or thickness) through remote sensing. To date, such quantifications have mainly been based on laboratory hyperspectral measurements over known oil volume for both oil emulsions and non-emulsions. Application of such laboratory-based methods to the real ocean environment faces two significant problems: 1) the observing conditions can be dramatically different (e.g., presence sun glint), and 2) lack of remote sensors with sufficient spectral bands and spatial resolution to apply the laboratory-based methods or to address the heterogeneity of oil slicks.

The objectives of this research are to understand oil slick reflectance spectra in the marine environment, delineate oil footprint, and develop practical methods to classify oil emulsions from non-emulsions and classify oil thickness, thus providing useful tools for oil spill assessment and for decision-making during an oil spill accident. Specifically, the objectives are to: 1) understand the various spatial and spectral oil-water contrasts in optical remote sensing imagery under different observing conditions; 2) develop algorithms and schemes to detect oil slicks, classify oil type (oil emulsion versus non-emulsion), and estimate oil thicknesses using multiband optical remote sensing imagery; and 3) apply the algorithms and schemes in the assessment of oil spill accidents. The Gulf of Mexico (GoM) is selected as the focus of this research because the continental slope of the GoM is recognized as a major hydrocarbon province with widely distributed natural hydrocarbon seeps and where two of the largest marine oil spills occurred (the Ixtoc-I oil spill in 1979 and Deepwater Horizon oil spill in 2010).

The several approaches used to address these objectives include: 1) a literature search; 2) controlled tank measurements to understand oil-water spatial and spectral contrasts under various observing conditions; 3) a multi-sensor analysis to examine the spatial and spectral characteristics of oil slicks; 4) a step-wise classification scheme to classify oil type and oil thickness; and, 5) the application of the developed methods to several oil spill events through case studies.

Firstly, a thorough review of previous laboratory-developed reflectance—thickness relationships of both crude oil and oil emulsion is performed and compared to reflectance spectra collected by several satellite and airborne sensors (MERIS, MODIS, MISR, Landsat, AVIRIS) from the Deepwater Horizon oil spill (Chapter 2). Interpretation of the oil-water spatial and spectral contrasts under different observing conditions suggests that besides oil thickness, several other factors also affect oil–water spatial and spectral contrasts. These include sun glint strength, oil emulsification state, optical properties of surrounding water, and spatial and spectral resolutions of remote sensing imagery.

The impact of sun glint strength on oil slick detection is further investigated in Chapter 3, where concurrent (1-2 hours) image pairs collected by MODIS/Terra, MODIS/Aqua, and VIIRS over the same oil slicks from natural seeps are used to quantify the sun glint threshold, under which thin oil films cannot be observed. The threshold is determined to be 10-5–10-6 sr-1 for MODIS Terra and MODIS Aqua, and 10-6–10-7 sr-1 for VIIRS.

The impact of pixel resolution on spill detection is evaluated by studying oil slick morphology and size distributions for different oil thickness classes derived by the USGS using fine spatial resolution (~7.6 m) hyperspectral AVIRIS imagery collected over the Deepwater Horizon oil spill in the GoM (Chapter 4). Oil slicks are found to be elongated in shape for all thickness classes (≤50 μm but thicker than sheen, 50—200 μm, 200—1000 μm, and >1000 μm). They are found to be highly heterogeneous as well, where most of the medium-resolution (30-m) pixels would be mixtures of different thickness classes of oil, or mixtures of oil and oil-free water. According to the AVIRIS derived results, to detect oil thicker than sheen with oil fractional pixel coverage >50% for at least half of the oil containing pixels, a 30-m or higher spatial resolution sensor would be needed. This suggests that most satellite remote sensing must consider mixed pixels when conducting analysis of spatial and spectral contrasts.

Based on the above understandings of oil-water spatial and spectral contrasts under different sun glint conditions, a stepwise classification scheme is proposed to extract oil features, classify oil types (oil emulsion versus non-emulsion), and classify oil thicknesses of each type under no glint condition and under various sun glint conditions in multiband optical imagery (Chapter 5). After oil feature extraction, reflectance in the Near Infrared and ShortWave Infrared (SWIR) bands is used to classify oil type, where elevated reflectance indicates oil emulsions. For oil emulsions, a histogram matching technique is used to compare the multiband measurements with hyperspectral AVIRIS measurements to classify oil thickness under various sun glint conditions. For the non-emulsion oil, a ratio between SWIR and blue bands is used to classify oil thickness. Furthermore, the spectral bands deemed necessary to apply the step-wise classification scheme and to discriminate false-positives are determined to be 480, 560, 670, 860, and 1600 nm.

The methods developed above are applied to several oil spill events as case studies (Chapter 6, 7 and 8). The Ixtoc-I oil spill footprint (over its > 9-month spill period) has been mapped with Landsat Multispectral Scanner and Coastal Zone Color Scanner (Chapter 6). The satellite-derived oil trajectory patterns agree well with physical modeling and field observations in the past. Another case study focuses on the ongoing oil spill in the MC-20 site in the northern GoM, where the spill is assessed systematically using medium- to high-resolution (10-30 m) optical remote sensing imagery between 2004 and 2016 (Chapter 7). These data allow for the determination of oil slick presence frequency and average spill size; further, the cumulative oil footprint are derived with daily discharge rate estimated. Finally, a multi-sensor day-and-night approach, along with numerical modeling is used to track an oil tanker collision event in the East China Sea, where the unique value of VIIRS night time data is demonstrated (Chapter 8).

In summary, this dissertation provides a better understanding of oil-water spatial and spectral contrasts in multi-band optical remote sensing imagery, from which a step-wise classification scheme is developed to extract oil slick features, classify oil emulsion from non-emulsion, and estimate oil thicknesses in each type. The methods are then used in several case studies to assess oil spills. Although further research is still required to refine the methods and to provide direct field validation, the findings here expand our current knowledge in remote sensing of oil spills using multiband optical imagery. In particular, when compared with the remote sensing capacity during the DeepWater Horizon oil spill (where satellite remote sensing could only provide maps of oil presence/absence), the findings here suggest that much better data products can be derived from existing satellite platforms, to not only show oil presence/absence, but to also classify oil type and thickness, in future spills, for improved response and assessment.