Introduction
The application of remote sensing techniques to environmental disaster prediction, detection and monitoring is a field which is rapidly developing. These techniques are particularly useful when applied to incidents which occur in remote regions of the world, not readily accessible from the ground or the air. The variety of sensors currently in orbit make it possible to observe a particular area on the Earth's surface at regular intervals thus giving us the ability to monitor change. A timely and appropriate application for this new technology would be to monitor the extensive network of gas and oil pipelines in the permafrost regions of Siberia. Pipelines in this region are particularly vulnerable because of the difficulty in maintaining the infrastructure currently in place. This problem is compounded by the extreme climatic conditions whereby the extremes of temperature and the freeze/thaw cycle in the permafrost combine causing great mechanical stresses in the pipes. Interpretation of visible, near infra-red, thermal and radar imagery from Earth observation satellites allow changes in vegetation health and bare ground caused by pollution stress to be monitored, and also recognition of conditions which might lead to pipeline ruptures.
      This article describes a case study of the 1994 oil spill in the Russian Republic of Komi carried out by the Marine Environment Group of the UK's National Remote Sensing Centre (NRSC) Ltd. The study serves as an example of how remote sensing can be used as an element of a system to provide advance warning of a potential disaster. The data used for the study were collected by the synthetic aperture radar (SAR) sensor onboard the European Space Agency's ERS-1 satellite. The images were used to observe the spill site, and to pinpoint and monitor the progress of the spill through change detection. It is likely that if a continuous watch were being kept on the pipeline network, by a combination of both radar and optical sensors, this particular incident might have been detected earlier and the consequences minimized.

Background
The spill took place along a section of pipeline some 50 km north of the town of Usinsk, near the Arctic Circle. A series of small leaks occurred throughout the summer of 1994, leading up to a large rupture in September. A large earthen dam was constructed to prevent the leaked oil spreading away from the pipeline. Consequently, a large reservoir of oil built up behind this dam. At the end of September, after a period of heavy rain, the dam collapsed, allowing the stored oil to spill out onto the tundra, and into the nearby water system. The contaminated waterways included the Kolva and the Usa rivers, which feed into the ecologically sensitive Pechora River.
      Estimates of the amount of oil involved varied wildly at the time, but the currently accepted figure is 100,000 tons. By comparison, the Exxon Valdez accident involved about 13,000 tons.
      Initial cleanup operations were hampered by the extreme cold and short days of the arctic winter. However, these same conditions have also had the beneficial effect of slowing the advance of the oil by making it more viscous and locking it into the river ice. When the spring thaw comes, it is likely that the bulk of this will be released and flow towards the Barents Sea by way of the Pechora River. Oil contamination here could have severe implications for the fishing and agricultural activites in these area.

Remote Sensing Observations of the Spill
When news of the spill broke in the Western press, NRSC began to carry out a study of the spill site. Conditions of heavy cloud cover and short periods of daylight made it impossible to use purely optical images, such as those produced by the American Landsat satellite and the French SPOT system. However, radar images are not affected by these conditions, and images from spaceborne radar systems have already been used extensively to study oil slicks at sea.
      Four SAR images of the spill area were analyzed with the following dates in 1994: August 20, September 29, October 16 and November 2. Figure 4 shows a map generated from three of the images, showing the area around Usinsk, and the location of the spill site. These dates were chosen to cover the period from the time when the spill was contained by the dam to a date after the dam had broken. The dam actually collapsed on September 27, a few days before the date reported in the Western press, and was actually located some 50 km north of the town. The scarcity of accurate information on both the time and location of events surrounding the spill was a heavy limitation on the amount of useful work that could be done with the resources to hand. For instance, the October 16 image turned out not to extend far enough north to cover the spill site, and so could not be used in the analysis. Further spills are now known to have occurred later in November, but unfortunately, imagery for these is not currently available.
      A useful feature of radar imagery is that man-made and predominantly metal objects tend to show up as very bright features. An initial inspection of the area around Usinsk allowed the town itself to be identified along with the Usa and Kolva rivers and some other artificial features such as pipelines, bridges and airfields. Although some of the pipeline is buried, it was possible to trace its course by looking for surface features associated with the pipe, such as clearings through forested areas. Other variation in the images was seen due to different terrain types - forest, marshland, water bodies, etc., all present different appearances in the radar image. The location of the feature detected by NRSC in the imagery, shown in Figures 1-4, was confirmed as being at the site of the spill by representatives of the Greenpeace field team which had been in the area to monitor the cleanup.
      The main image processing technique employed to identify and characterize the oil residue from the spill was change detection through color compositing of a time sequence of images. This is an established technique for detecting areas where the ground cover has changed in character over time; it involves firstly geocorrecting the individual images to a common map projection, so that they can be overlaid on top of each other, and registered exactly. A color composite of the time sequence is then produced, with images from three dates superimposed in red, green and blue. Where little or no change has occurred, one may expect the result to be an addition of roughly equal amounts of the three primary colors, producing either white, black and shades of grey, or very pale colors. Where much change has occurred, one or two colors will predominate, resulting in the highlighting of change areas in strong hues.
      Figures 1-3 show an enlargement of the spill area indicated in Figure 4, located by a combination of the method described above, and through ground information from Greenpeace. Figures 1-3 show the development of the spill over the period from August to November. It is also possible to estimate the area of ground covered by oil, although at present, this can only be an approximate figure, based on visual interpretation of the image. Figure 1 shows the oil pooling up around the pipeline, but presumably contained by the dam. It appears to cover an area about 200 m on either side of the pipe. In Figure 2, the dam has burst. The oil now covers a much larger expanse of ground, stretching almost 1 km to the south of the pipeline, and with an area of the order of 30-40 hectares. In Figure 3, a further month has elapsed, and the oil is still visible, but the extent of the oil affected area has changed. This is attributed to the dispersion of some of the oil into nearby water systems, removal by cleanup and possibly changes in the local environment. By November, the winter freeze had set in, with heavy snow fall, causing changes in the way the ground reflects the radar beam and hence the way it appears in the image. This could also explain why the pipeline is not as apparent in the November image as in the other images.
      Further opportunities for remote sensing observations of this site may arise in the spring when weather conditions have improved. Longer daylight hours and reduced cloud cover allow the resumed use of high resolution optical sensors.

Conclusions
Despite the lack of available information about the site, this short project has demonstrated the utility of radar data in locating and mapping areas of oil spilled on land. Active radar instruments such as the one on ERS-1 are particularly suited for this role in the former Soviet Union area, due to their ability to penetrate cloud and darkness, and their capability to emphasise artificial objects and areas of oil. If remote sensing can provide timely information on an oil spill under the restricted conditions of this pilot study, it can be expected to work in a fully supported operational mode. One can expect better results in conditions where high resolution optical data can be used in conjunction with radar data. The goal of such a monitoring system would not be to detect oil explicitly, but to identify the appearance of suspicious changes on the ground close to the pipelines, which would be targeted for further investigation by air or ground survey. In view of the remoteness of many of the world's oil fields, and limitations on resources available for monitoring and cleanup operations, it is evident that a remote sensing system could provide a valuable and cost effective contribution to the environmental health of these areas.

Credits:
Thanks go to Mike Brimson for speedy processing of the SAR images, Dr. Adrian Huntley of NRSC Applications Group for help in SAR interpretation, and Christian Ripley for help in preparing the article. All the image processing for this project was carried out on a Sun SPARC 2 workstation using ER-Mapper Version 4.0 image processing software. ERS-1 images kindly provided by European Space Agency. In addition, thanks to Greenpeace for use of ground photographs. For further information, contact Greenpeace Communications, 5 Bakers Row, London, EC1R 3DB, UK, Tel: +44 (0)171 833 0600.

About the Authors:
Paul Allen joined the National Remote Sensing Centre Ltd in 1994, and currently works on applications of satellite and airborne remote sensing to oceans and coastal zones. He may be reached by phone, +44 1252 3541464, fax, +44 1252 375016, or email, [email protected]
Simon Wilson is on temporary attachment to NRSC Ltd., working on the use of remote sensing and GIS technology to investigate the environmental impact of oil spills on fragile ecosystems.

Back