Classification of oil palm in Indonesia using formosat-2 satellite image

Indonesia is the biggest exporter crude palm oil (CPO) in the world since 2006. Total export of Indonesian's CPO and its derivatives in 2007 was about 11 million tons or equal
to US$ 6.2 billion. It is a valuable sector that supports Indonesian economics, but it also causes environmental and social impacts. Deforestation is a sensitive issue related to oil palm plantation expansion. Sustainable oil palm development is needed to reduce environmental impacts and to meet economics purpose. Through the utilization of remote
sensing (RS) technology, this study has tried to support sustainable oil palm development.
Cimulang oil palm plantation that lies in district of Bogor, West Java Province, Indonesia was chosen as study area. High spatial resolution FORMOSAT-2 satellite image that has 4 multispectral bands (8 m spatial resolution) and 1 panchromatic band (2 m spatial resolution) was used in this study. The objectives of this study are to classify
growing stages of oil palms using only multispectral bands and to classify growing stages of oil palms using multispectral bands plus texture information of FORMOSAT-2 data, to test the accuracy of both classification results, and to support sustainable palm development by providing more often updated oil palm land use map. Texture extraction
through image matching by correlation and maximum likelihood supervised classification method has been applied in this study. The result shows that overall accuracy for
multispectral image classification is 66.4%. Triangular oil palms planting pattern that has space 9 m apart between trees can be seen visually in 2 m panchromatic image of
FORMOSAT-2 data and it also can be extracted automatically by texture analysis through image matching by correlation. This texture information then added to multispectral bands
for classification. The overall accuracy result of multispectral bands with texture information is 76.8%. Image classification accuracy has improved (10.4 %) if the
classification process employed not only multispectral bands but also added with the
texture information.