A Quantitative Comparison to Water Column Correction Techniques for Benthic Mapping Using High Spatial Resolution Data

Mapping of substrate cover types such as Coral Reefs, Algae, Seagrass, etc. in coastal regions has been performed in wide areas using satellite-based observation since mid 1980's. Although the high spatial resolution data such as IKONOS images has overcome some spatial limitations, still low signal-to-noise ratio due to destructive atmospheric effects and intervening water column has led to a problematic image processing task. In this study, two pan-sharpened IKONOS data sets of Eastern part of Kish Island, located in the south of Iran in the Persian Gulf, acquired on 5 November 2001 and 23 June 2004, have been tested. Following registration using a 1:1000 scale topographic map of the region produced by Kish Free Organization in 2003, atmospheric impacts have been removed applying pseudo invariant features (PIFs). Two distinct methods, namely, Bierwirth and Lyzenga (Depth Invariant Indices (DII)) have been utilized for elimination of water column effects. Efficiency of these methods has been evaluated, comparing between the accuracy of generated thematic maps. Prior to classification, the bottom reflectance images (Bierwith based outcomes) were divided into two zones according to DOP index within red band. The other processing stages were identical during the procedure. Results of this research have shown that the use of Bierwirth model performed well against Lyzenga method in both data sets. In other words, Kappa index of Bierwirth was 0.67, 0.74 rather than the 0.64, 0.71 of the Lyzenga, respectively. However, these contrasts were more remarkable within live coral classes, when analyzing their user's and producer's accuracy. However both methods were almost poor in deep areas regarding the assignment of 10-12 m depth range to deep water category. This trend was more pronounced in the case of Lyzenga method. The use of hyper spectral data sets may lead to a better performance of the Lyzenga method.