This study aims to evaluate the performance of machine learning methods for mapping bamboo grove using multi¬temporal Landsat-8 OLI imagery. Representative six machine learning methods such as Artificial Neural Network (ANN), Support Vector Machine (SVM), k-Nearest Neighbor (kNN), Random Forest (RF), C5.0 and eXtreme Gradient Boosting (XGB) were employed to generate models for predicting land use / land cover (LULC) classes including bamboo grove. 500 points of ground reference data were used to calibrate the model parameters and to validate their classification performance based on the repeated 10 fold cross validation method with Cohen's kappa. The result showed that the highest performance was achieved by RF (0.818), closely followed by XGB (0.814), C5.0 (0.803), SVM (0.799), ANN (0.756), and finally kNN (0.739). Variable importance metrics suggested that the NIR and SWIR portions were the key to map bamboo groves through LULC classification.
抄録全体を表示