Linear mixture modelling applied to ikonos data for mangrove mapping

Abstract

Mixed pixels problem in remotely sensed satellite data often results in poor classification accuracy. However, high spatial resolution sensors such as IKONOS and QuickBird, are expected to classify mangrove species more accurately than conventional coarse spatial resolution satellite images. Nevertheless, conventional classification technique such as maximum likelihood could not improve the classification accuracy when such high-resolution images are applied. In order to improve the classification accuracy Linear Mixture Model (LMM) was applied in this study to classify the mangrove forest at Sungai Belungkor, Johor, Malaysia. High spatial resolution satellite imagery (IKONOS) was used to map three different mangrove species in the area. The application of LMM to mangrove classification involved image preprocessing, endmember selection, inversion of LMM and finally the accuracy assessment. Accuracy assessment was carried out between the fraction of pixels estimated from LMM and the fraction measured in the field. The results of accuracy assessment gave a correlation coefficient value of ~0.8 for endmember bakau minyak and ~0.6 for bakau kurap and “others” type of mangrove species. An error image was also created to compare the best fitting spectrum produced by the inversion of LMM to the original observed spectrum where the maximum RMS error was only 5%. The accuracy of LMM was also assessed based on a generalized area based confusion matrix. Areas that were correctly classified according to reference data and classified data were 63 % and 73%, for bakau minyak, 62 % and 60% for bakau kurap and 58% and 52% for “others” type of mangrove respectively. These accuracies obtained from LMM were higher in comparison to the classification results derived from maximum likelihood with the inclusion of texture information or minimum distance to mean classifier