Proceedings of the 2nd Philippine Geomatics Symposium (PhilGEOS) 2013: Geomatics for a Resilient Agriculture and Forestry November 28-29, 2013 |University of the Philippines, Diliman, Quezon City | http://dge.upd.edu.ph/philgeos2013 ________________________________________________________________________________________________________________________ MAPPING THE STARCH-RICH SAGO PALMS THROUGH MAXIMUM LIKELIHOOD CLASSIFICATION OF MULTI-SOURCE DATA Jojene R. Santillan Research Laboratory for Applied Geodesy and Space Technology, Training Center for Applied Geodesy and Photogrammetry, University of the Philippines, Diliman, Quezon City, Metro Manila, 1101, Philippines, E-mail: santillan.jr2@gmail.com, jrsantillan@up.edu.ph , ABSTRACT An approach to map Sago palms using multi-source datasets is presented in this paper. The approach consisted of applying the Maximum Likelihood Classifier (MLC) to a stack of calibrated and co-registered ALOS AVNIR-2 image, NDVI, Envisat ASAR image, and an ASTER GDEM of the study area located in Agusan del Sur, Mindanao, Philippines. The MLC was applied to eight combinations of the ALOS AVNIR-2 reflectance bands, NDVI, Envisat ASAR and ASTER GDEM to derive classification maps. Results indicate that more than 90% overall classification accuracy and more than 90% Producer’s Accuracy for Sago Palm can be both achieved if MLC is applied to any of the 8 combinations. However, the User’s Accuracy for Sago palm is only 77% when using the ALOS AVNIR2 image alone as input which indicates overestimation of Sago palm classification due to high commission errors. Better User’s Accuracy of 91.37% was achieved when MLC was applied to a combination of ALOS AVNIR 2, NDVI, Envisat ASAR and ASTER GDEM. With this significant increase in User’s accuracy, the approach of applying MLC to a combination of these multi-source datasets shows potential in mapping Sago palms in other areas in the Philippines. Keywords: Sago palm, ALOS-AVNIR2, Envisat ASAR, ASTER GDEM, Philippines INTRODUCTION Background The Sago palm (Metroxylon sagu), as shown in Figure 1, has a trunk which contains starch. It is reported to be the highest starch producer at 25 tons per hectare per year (Bujang, 2008). It is now grown commercially in Malaysia, Indonesia and Papua New Guinea for production of starch and/or conversion to animal food or fuel ethanol (McClatchey et al., 2006). Because of the Sago palm’s significant economic benefits, there has been keen interest by the Philippine government for its mass propagation in order to develop and sustain a large-scale sago starch industry. For this to be realized, mapping the location of existing Sago palms is necessary in order to determine current supply as well as to characterize its habitat. Once these characteristics have been identified, it is then possible to locate other areas that have the same habitat characteristics for Sago palms to grow. Large clusters of Sago palms have been reported to exist in marshlands and other wetlands of Mindanao in southern Philippines. Confirming the locations of these clusters is not only difficult due to in-accessability; it is also expensive especially when done using conventional field mapping techniques. The use of remote sensing data and techniques is considered to be the best alternative. Figure 1. The sago palm in clusters. Shown in (a) are mature Sago palms, while younger ones are shown in (b). The Sago palms are similar in structure to coconuts and oil palms. (a.) (b.)