Detection and Dispersion of Oil Spills from Satellite Optical Images in a Coastal Bay

Abstract

The spatial distribution of an oil spill and its temporal dispersion within a coastal bay were investigated using high-resolution optical images. A neural network (NN) method was applied to Landsat and DubaiSat-2 images to detect the oil spill. We conducted field observations to measure spectral characteristics of the oil spill and the oil-free sea surface. We were able to detect and eliminate pixels corresponding to ship and ship shadows on the satellite image, resulting insuccessful oil spill detection. A new recursive NN method using a near-infrared band was developed to classify oil types into thick or film-like oil and to estimate their areal extents. To understand potential causes of the temporal evolution of the oil spill, we performed numerical modeling with atmospheric and oceanic inputs. Overall, trajectories of oil particles controlled by tidal currents showed good agreement with the detection results from satellite data. Slight discrepancies occurred between satellite data and results from the model simulation using only tidal currents, particularly in the southeastward dispersion or in the spreading of film-like oils into the northern inner channels. This was attributed to the effect of wind-driven Ekman drift. This study suggests that tidal currents played an important role in the temporal dispersion of oil in the bay during initial phases, when wind conditions were relatively weak, and that the Ekman