Evaluation of the Reliability of Classifiers for the Mapping of Mangrove Forest Using Landsat TM Images

Abstract

In Indonesia land use management of many areas has undergone significant changes due to a variety of human activities. The historical land use management at EMRP (Ex Mega Rice Project) area, Central Kalimantan, has been expanded for crops production in 1996/1997 and usage is divided into many classes. Mangrove as a potential carbon sink, having the largest amount of biomass found with in the Indo-Pacific area, was focused on as potential rehabilitated land to promote biomass restoration. As a result it has become a main topic of research. Several Lands at images taken in 1997, 2006 and 2010 were used and classified to monitor mangrove distribution using Maximum Likelihood Classification (MLC), Spectral Angle Map per (SAM) and Support Vector Machine (SVM) methods, then comparisons were made. The land cover type includes, forest, mangrove forest, water, cloud, and other land use (bare land, settlement, bush shrub swamp, dry land agriculture, savanna, paddy field, and bush shrub). Result shows that SVM based on statistical and vector approach appears to be a superior method, with a high accuracy of 0.95±0.003. The smallest standard deviation of classification accuracy also showed that SVM is relatively more stable in mangrove forest mapping.