University of Twente Student Theses


Dark spot detection for characterization of oil spills using polsar remote sensing

Kulshrestha, Anurag (2018) Dark spot detection for characterization of oil spills using polsar remote sensing.

[img] PDF
Abstract:Oil spills have been a cause of concern for environmental agencies, governments and coastal habitats.Accurate and fast knowledge about the location and characteristics of oil spills is extremely useful for oil spillcontainment and clean-up operations. In this research, the potential of polarimetric SAR data in detecting andcharacterizing oil spills is studied. The study is conducted using quad-polarized UAVSAR data, hybridpolarized RISAT-1 data and dual-polarized TerraSAR-X data of an experimental oil spill exercise (NORSE-2015) in North Sea, Norway conducted on 10th June, 2015. The environmental conditions during the oil spillexercise were rough with wind speed being consistently above 10 m/s. In this exercise, four different type ofoils were spilled into the sea: a simulated plant oil (PO), and three emulsions of mineral oil with 40% oil(E40), 60% oil (E60), and 80% oil (E80) respectively with water making up the remaining volume in eachemulsion. Distinguishing between different type of oil slicks and from similar look-alikes is difficult due tothe similarity of their radar backscatter signals. To overcome this challenge, polarimetric SAR data is used toderive the polarimetric parameters which relate to the physical properties of the scatterers on the sea surface.Some of these features are used to detect oil spills using Expectation Maximization of Gaussian MixtureModels. The parameters of the algorithm are optimized using the UAVSAR dataset and hence the method istested on all available datasets. The method is found to show better performance for RISAT-1 dataset ascompared to TerraSAR-X dataset. The strongest factor for incorrect results is found to be the high windwave caused due to high wind speeds. The shadows created by the high surface gravity waves acts as lookalikes. The slick areas extracted using this process helps in realizing that the stretching of the slicks is in thedirection of wind and its extent is also proportional to the amount of oil in the slicks. The extracted slickareas are then used to compare the polarimetric features on the basis of their potential to separate oil slicksfrom water and from other oil slicks. The determinant of the covariance matrix is found to be the mosteffective feature for oil-water and oil-oil class separabilities. Therefore, a covariance matrix based Wishart-maximum likelihood classifier (W-MLC) is chosen for oil spill classification. The results of this classificationare found to be much better than Gaussian based MLC, with no misclassification in the near range and a gainin overall classification accuracy of 16-34%. The oil probability output from this classification is then used tomodel oil spills as Gaussian probability surface models. Some of the probability surfaces models are found tocorrectly estimate the orientation and areal extent of the spill. Moreover, this method is also able toprobabilistically separate E40 from E60 from each other by correctly estimating relative peak probability forE60 as compared to E40. This method gave higher probability for PO, thereby indicating low mixing of POwith the sea water. It is concluded that probabilistic surface modelling is useful in oil spill categorization andtherefore, can be optimised to include other ancillary information to further improve the quality of oil spillclassification.
Item Type:Essay (Master)
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Programme:Geoinformation Science and Earth Observation MSc (75014)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page