Detection and Prediction of Peatland Cover Changes Using Support Vector Machine and Markov Chain Model

Abstract

Detection and prediction of peatland cover changes needs to be done in the rapid rate of deforestation in Indonesia. This work applied Support Vector Machine (SVM) and Markov Chain Model on multitemporal satellite data. The study area is located in the Rokan Hilir district, Riau Province. SVM classification technique used to extract information from satellite data for the years 2000, 2004, 2006, 2009 and 2013. The Markov Chain Model was used to predict future peatland cover. The SVM classification result showed that the Kappa accuracy of peatland cover classification is more than 0.92. The non vegetation areas increased to 307% and the sparse vegetation areas increased to 22% between 2000 and 2013, while dense vegetation areas decreased to 61%. Prediction of future land cover by the Markov Chain Model showed that the use of multitemporal satellite data with 3 years interval provides accurate result for predicting peatland cover changes.