Peat Depth Prediction System Using Long-Term MODIS Data And Random Forest Algorithm: A Case Study in Pulang Pisau, Kalimantan

Abstract

Peatlands have an important role as global climate regulators because they store global amounts of carbon which, if degraded, will result in increased concentrations of greenhouse gases in the atmosphere. Peatland mapping using satellite imagery is considered effective for classifying a land cover area. Previous studies concluded that satellite imagery can be used to classify a peat area and a non-peat area. In this study, we use satellite imagery with a mounted MODIS sensor from 2015-2019 and calculate the index from MODIS bands. The Machine Learning (ML) method was used for generating a peat depth in Pulang Pisau, Kalimantan. Random Forest (RF), Support Vector Machine (SVM), Support Vector Regressor (SVR), Gradient Boosting (GB), and Ada Boost (AB) models were used to generate a peat depth map. The best performance was achieved by RF Classifier with accuracy 0.93 and RF Regressor with ${R}^{2}=0.88$