You must have a javascript-enabled browser and javacript and stylesheets must be enabled to use some of the functions on this site.
 
   workshops     

 

On C-Band SAR Based Oil Slick Detection in the Baltic Sea

Markku SimilĂ€(1) , Istcvan Heiler(1) , Juha Karvonen(1) , and Kimmo Kahma(1)

(1) Finnish Institute of Marine Research, PO Box 2, 00561 Helsinki, Finland

Abstract

We are developing an operational algorithm utilizing satellite-borne C-band SAR instruments (Radarsat-1, Envisat, Radarsat-2) for oil slick detection in the Baltic Sea area. In this paper we will report on four different aspects related to this problem: 1) At which wind speed the sea surface is totally roughened with visible waves, 2) How often the wind conditions in the Baltic Sea allow a reliable oil spill detection, 3) Our experience on the usability of three backscattering models in the oil spill detection problem, and 4) How the radar incidence angle affects on the oil detection statistics ?.

On the 9th September in 2002 we were able to record the change of totally calm water surface (wind speed 1.2 m/s) to become gradually covered with larger and larger wind induced rough surface patches with a digital video camera. In this case (water temperature $20^o C$) the sea surface was covered with wind roughened patches with 50 % coverage at wind speed 2.8 m/s and with 100 % coverage at wind speed 3.4 m/s. This implies that a threshold for a reliable oil spill detection (capillary waves dampened by natural or oil film) in the Baltic Sea is about 4.0 m/s (conservative choice).

The oil can reliably be detected by C-band SAR sensor only if the wind is not too low or too high. If the wind speed is high enough, waves induced by strong wind break an oil spill and drag it in the ocean sub-surface where it can not be detected. Oil spill observations have not been reported in the Baltic Sea if the wind speed has exceeded 8 m/s. To assess how often the SAR can detect oil spills with a great confidence, the wind statistics of KalbÄdagrund, a lighthouse in the Gulf of Finland with 30 km distance to coast line, during the years 1991-1999 were studied. Monthly wind statistics will be shown. E.g., in December the wind speed is between 4 m/s to 8 m/s only 34 % of time, and between 4 m/s to 10 m/s 51 % of time. From the oil detection point of view, the best month is June with percentages 55 % and 71 %, respectively.

To incorporate wind information with the oil slick detection procedure, three backscattering models were studied: the integral equation model (IEM) [1], CMOD4 [2] and CMOD5 [3] models. Comparisons of the model results to our SAR data suggest that of these models CMOD5 best captures the SAR backscattering behaviour for Radarsat-1 SAR data over the Baltic Sea, in most cases the dymamic ranges of the CMOD5 data are similar to those of the corresponding SAR images, unlike for IEM and CMOD4. However, the model and SAR histogram shapes do not always correspond to each other very well. Partly this due to uncertainties in the used forcing wind fields (the weather forecast HIRLAM model, 22 km grid size, temporal resolution 1 h). Validation of the HIRLAM model results show that even 2 m/s wind speed error may occur at very low wind speeds (less than 4 m/s), the rms being about 1-2 m/s. Our experiences indicate that it is impossible to utilize the models directly in thresholding the SAR data. Instead, the modelled wind field can be used as an aid in oil slick detection. In many cases the HIRLAM model is capable of locating areas of low backscattering, and we can then compare the model values with the local SAR backscattering and if both SAR and model values are low, the dark area is very probably caused by wind.

Our data set consists of 54 RADARSAT-1 ScanSAR scenes. Human operators have identified 70 potential oil spills from these images and given confidence level values for these. So far we have computed some basic statistics for these spills. It was observed that mean difference between the backscattering levels of oil slick and surrounding open water area decreased drastically from about 4 dB (at incidence angle of $20^o$) to about 1.5 dB (at incidence angle of $40^o$). Almost all low confidence oil slick observations occurred at incidence angles over $30^o$. All computed oil slick statistics, including various shape measures as a function of incidence angle will be presented.

[1] K.S. Chen, A.K. Fung, D.E. Weissman, A backscattering model for ocean surface, IEEE Trans. Geosci. Rem. Sens., GE-30, pp 811-817, 1992.

[2] A. Stoffelen, D. Anderson, Scatterometer data interpretation: Estimation and validation of the transfer function CMOD4, Journal of Geophysical Research, Volume 102, Issue C3, pp. 5767-5780, 1997.

[3] H. Hersbach, CMOD5, An Improved Geophysical Model Function for ERS C-Band Scatterometry, Report of the Europena Centre Medium-Range Weather Forecasts (ECMWF), 2003.

 

Full paper

 

  Higher level                 Last modified: 07.10.03