GISdevelopment.net --> Application --> Miscellaneous
Oil Spill Detection and Monitoring from Satellite Image
S. B. Mansor, H. Assilzadeh, H.M. Ibrahim, A. R. Mohamd
Spatial and Numerical Modeling Laboratory
Institute of Advanced Technology, University Putra Malaysia,
43400 Serdang, Selangor, Malaysia
Email : firstname.lastname@example.org, email@example.com
The very nature of marine oil spills massive quantities covering vast areas open ocean and/or coastline necessitates the use of satellite remote sensing to supplement other aerial observations. Remote sensing instrumentation is constrained to work in areas of the electromagnetic spectrum, which transmit sufficient quantities of radiation such as the microwave, thermal, near infrared, visible and near ultraviolet. Within these windows, there is usually further attenuation of radiation by cloud cover and inclement weather. In general, satellites using electro-optical detectors have problems seeing through cloud cover and at night. There is also problems with the majority of optical remote sensing techniques used for oil spill detection which is the high number of false alarms; i.e. oil slick look-a-likes, phenomena which give rise to signals which appear to be oil but are not (O’Neil, 1983; Goodman, 1989 and 1992; Schnell, 1992; Thornton, 1992;). Visible satellite systems are susceptible to false alarms due to sun glint, wind sheen, bottom features, cloud shadows, and biogenic material such as surface weeds and sunken kelp beds (Goodman, 1988 and 1989; Fingas et al., 1990 and 1992).
There are certain times when visual techniques and optical satellite image are unsuitable for mapping of oil spill; it is in these cases where radar remote sensing is required. These situations include spills covering vast areas of the marine environment, and when the oil cannot be seen or discriminated against the background. The discrimination of oil in these circumstances presents several unique problems. The remotely sensed data collected in these situations often provide complex signatures, which must be deciphered in order to locate the spilled oil. Environmental conditions such as precipitation, fog, and the amounts of daylight present also may pose problems especially in optical images.
In this study in order to minimize these problems and reduce the response time and facilitate the decision-making process in the event of oil spill, application of radar image as an operational tool has been suggested. The aim of this study is to construct a developed early warning system for the potential and accidental oil spills in water environment. New techniques such as radar image processing have been used in order to make the proper distinctions between different thickness of oil spill; The results of this research shows that with radar information, the signature of oil can be used to detect minute concentrations of hydrocarbon (oil spill) on the sea and it can distinguish between different types of its thickness.
Materials and Methods
Figure 1 shows the algorithm of processing Synthetic Aperture Radar (SAR) image for detection and classification of oil spill. A SAR data was selected for analysis, taken on 10 Oct. 1997 over the Straits of Malacca, covering around 110 x 150 km of Peninsula Malaysia from the states of Johor up to Malacca. The weather condition was extracted from ground stations over the study area. The maximum speed of seawater was reported around 0.25 m/s and the maximum for wind was 1.4 m/s.
Figure 1 Oil spill detection and classification algorithm for SAR image
Picture Analysis, Correction, and Enhancement (PACE) are a group of application programs in PCI image analysis software, which providing extensive digital image processing functions. These programs have been used to process radar data for detecting oil spill in water body.
Pre-processing of radar image include Antenna pattern correction (APC), Radiometric correction, and Geometric correction; have been applied to prepare correct data from SAR image.
Antenna Pattern Correction (APC) performs a radiometric balancing on synthetic aperture radar data to compensate for non-uniform illumination in the range direction due to the antenna pattern. Then the image has been geocoded by GCPWork interface in PCI image analysis software using image to map geocoding method. In this method the uncorrected image was selected and suitable georeference system has been defined for the image, then by choosing “Collect GCPs” command on PCI’s GCPWork, the “GCP Selection and Editing” panel loaded to select ground control points (GCP) by using topographic maps of the study area.
Post processing of radar image for detection of oil spill is including of image enhancement, texture analysis, dark slick detection, feature extraction, scaling and filtering. First the calibrated radar brightness image was generated from a Radarsat SAR image through SARBETA function in PACE interface. SARBETA generates a radar brightness channel from the input scaled radar channel using the gain offset and scaling. Then a set of texture was calculated for all pixels on the image through the texture analysis function in PACE interface. The measurements were based on second-order statistics computed from the gray level co-occurrence matrices. The textures measured after scaling have been used as input channels for classification algorithm.
To extract detailed information about oil spill, scaling radar standardized the texture values. Image Gray Level Scaling (SCAL) program in PACE, performs a linear or nonlinear mapping of the image gray levels to a desired output range. This program is typically used to scale data from "high" resolution (32 and 16-bit) channels to "low" resolution (16 and 8-bit) channels. These channels are then applied for oil spill image classification based on texture analysis results.
The image was then preceded to classify oil spill using supervised maximum likelihood algorithm. Speckles appearing on SAR images are natural phenomenon generated by the coherent processing of radar echoes (Lee 1986). The presence of speckle not only reduces the interpreter's ability to resolve fine detail, but also makes automatic segmentation of such images difficult. The gamma map filter is primarily used on radar data to remove high frequency noise. Finally to create automatic detection engine for any oil spill radar images PCI’s Modeler programming interface has been used to provide an interactive methodology for the development of all described oil spill image analysis and processing flows.
To design such a model oil spill image processing modules were placed on the PCI’s Modeler canvas, and then connected the modules via pipes to create a process flow (Figure 2). Parameters for the spill detection modules were set in advance and the model was launched. During the execution of the model, graphical cues followed the data flow through the process.
Figure 2 Automatic detection of oil spill using Visual Modeler was designed to detect any spilled oil in a short time.
Results and Discussions
Figure 3 shows the results of pre-processing after APC, radiometric and geometric correction. Figure 4 shows the image gray level scaling and quantization, which applied on, calibrated radar brightness image to perform a linear mapping of image gray levels to a desired output range. This program is typically used to scale data from "high" resolution (32 and 16-bit) channels to "low" resolution (16 and 8-bit) channels.
Figure 3 Geometric Correction based on GCPWorks Main panel.
Figure 4 Scaling performs a linear mapping of image gray levels to a desired output range.
Gamma MAP Filter is primarily used on radar data to remove high frequency noise (speckle), while preserving high frequency features (edges). Figure 5 shows that the filter has smoothed the image, without removing edges or sharp features in the images.
Figure 5 Enhancing radar image using FSPEC -- SAR Speckle Filters command in Xpace
Figure 6 and 7 show the results of texture analysis for oil spill image. For oil spill detection Homogeneity and Angular Second Moment were found as two effective texture analyses for oil spill detection and classification. The texture measures used by this program were based on R.M. Haralick et al. 1973 & 1979 and R. W. Conners et al.1980 who were demonstrated the application of textural features for image classification. These functions were applied on 32-bit image acquired from brightens value analysis. To extract detailed information about oil spill, Scaling standardized the texture values of radar image.
Figure 6 Effects of the homogeneity function of texture analysis on the image.
Figure 7 Effects of the Contrast function of texture analysis on the image.
Figure 8 and 9 described the best results for automatic detection of oil spill in SAR image based on SCAL program, gamma image filtering and texture analysis program. The results where then saved on a new channel. In this images the edge of spilled oil is clearly presented and the spill look-likes are also eliminated.
Figure 8 Classified oil spill image by visual modeler in two regions; oil spills area in red and polluted area in black color.
Figure 9 Grey values of original image (channel 1) and attribute values after Homogeneity (Channel 2) and Angular Second Moment (Channel 3) texture analysis. Channel 2 specified the all polluted area and channel 3 rectified only very high-polluted area (spilled area).
Maximum Likelihood was found as the best algorithm for oil spill classification. Figure 10 shows the composite image classification results using texture analysis and scaled grey level image. Classification analysis results where given the important information about oil spill thickness and area of spillage. This figure presents the oil spill in three different classes according to thickness of the spilled oil.
Figure 10 SAR image classification and attribute DN values of the image pixels showing three different classes for oil spill.
Remotely sensed data are used for detecting oil spill to support the contingency plan at a specific location in Straits of Malacca. Several methods of image processing are applied for this task such as Gamma distribution analysis, texture analysis, image composite analysis and image classification. After pre processing (radiometric correction, APC, geometric correction, pixel size conversion) the image was applied to detect oil spill and its characteristics. This paper was focused on automatic dark slick detection and classification as early warning system for oil spill contingency planning.