Structural Image Classification for Oil Spill Contamination Mapping in West Siberia using Quickbird data

This work presents the final results of the OSCaR pilot study (Oil Spill Contamination mapping in Russia) and shows a synergetic concept for an object based and multitemporal mapping and classification system for terrestrial oil spill pollution using a test area in West-Siberia. An object oriented image classification system is created to map contaminated soils and vegetation in high resolution data and changes in the oil exploration well infrastructure. Due to the limited spectral resolution of Quickbird data context information and image object structure are used as additional features building a structural object knowledge base for the area. The distance to industrial land use and infrastructure objects is utilized to classify crude oil contaminated surfaces. Additionally the potential of Landsat data for dating of oil spill events using change indicators is tested with multi temporal Landsat data from 1987, 1995 and 2001. OSCaR defined 3 different subprojects: 1. high resolution mapping of crude oil contaminated surfaces, 2. mapping of industrial infrastructure change, 3. dating of oil spill events using multitemporal Landsat data. Validation of the contamination mapping results is done with field data from Russian experts provided by the Yugra State University in Khanty-Mansiyskiy. The developed image structure classification system showed good results for the severely polluted areas with good overall classification accuracy but did also reveal the need for the direct mapping of hydrocarbon substances. Oil spill event dating with Landsat TM/ETM+ data was very much limited by the comparably low spatial resolution of Landsat TM 5 data, small scale character of oil spilled surfaces and limited information about oil spill dates.