Comparison of Image Analysis Tools in Mapping Water-Bottom Sediment Distributions from a Modern Carbonate Platform Using Multi-Spectral Landsat Data
Analysis and interpretation of multi-spectral satellite data is a valuable approach for mapping water-bottom sediment types in modern carbonate depositional settings. The current investigation presents Cocos (Keeling) Islands, an isolated carbonate atoll in the Eastern Indian Ocean, as a case study to evaluate the viability and efficiency of three image analysis tools used in various geologic applications. The tools ER Mapper, Stratimagic, and Axiovision, which employ different statistical algorithms, were used to analyze variations in multi-spectral satellite data to produce water-bottom sediment distribution maps.
ER Mapper is a standard software package specifically designed for displaying and manipulating raster data sets. Stratimagic is pattern recognition software that, in its conventional application, uses a rigorous self-organizing neural network to classify 3-D seismic waveform data sampled in time. In this study, the different satellite spectral bands represent a sample value in a “pseudo waveform”, or attribute vector, that serves as input to Stratimagic’s neural network engine. Axiovision is a standard petrographic software application selected for its ease and utility in separating image pixels based on color in a supervised classification. The utility of these applications is that they function by binning individual pixels into sets of distinct clusters such that the differences in spectral characteristics between clusters are maximized and differences within each cluster are minimized. The resulting spectral cluster maps provided the basis for subsequent geologic interpretations of surficial sediment types that are calibrated to published sediment data.
Results indicate that all three applications are capable of producing geologically reasonable sediment distribution maps for Cocos Atoll. The strengths and weaknesses of each tool, however, are quite distinct. ER Mapper allows for quick, detailed unsupervised classifications, but determining the optimal initial statistical parameters can be challenging. Stratimagic, with its theoretically robust algorithms, also produces detailed classifications, but requires pre-classification preparation of the dataset. Axiovision was found to be the most simple and interactive approach for sediment mapping, but because it lacks geo-spatial functionality, interpretations must be subsequently geo-rectified to preserve correct spatial position and scale.
AAPG Search and Discovery Article #90090©2009 AAPG Annual Convention and Exhibition, Denver, Colorado, June 7-10, 2009