ANALYSIS OF RANDOM FOREST CLASSIFICATION FOR MAPPING OF SPATIAL DISTRIBUTION COFFEE PLANTS ON AGROFORESTRY (CASE STUDY: PART OF MOUNT PUNTANG AREA, PASIRMULYA, BANJARAN DISTRICT, WEST JAVA)

• Main Author: Tridawati, Anggun
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/50701
• Institution: Institut Teknologi Bandung

Indonesia is the fourth largest coffee producer in the world after Brazil, Colombia, and Vietnam. Statistically, the area of coffee plantations in Indonesia is 1.2 million hectares with a production of 500 kg/ha. In Indonesia, information regarding the distribution of coffee plantations and the effectiveness of their products are limited. Therefore, it is important to obtain an assessment of the spatial distribution of coffee plantation. Remote sensing is one method that can provide information about land cover mapping at a relatively low cost. Random forest is a widely used algorithm for the classification of remotely sensed data. This study aims to map coffee plants in agroforestry system derived from random forest classification model generated obtained multisensory, multitemporal, and multiresolution remote sensing data from Geoeye-1, Sentinel-2, and DEMNAS. The data used in this study are 1) pansharp GeoEye-1, used for the extraction of entropy, correlation, mean, standard deviation, and contrast texture, 2) multitemporal Sentinel-2, used for NDVI vegetation index, brightness, greenness, and wetness, 3) digital elevation model (DEM), used for extraction of elevation, aspect, and slope, and 4) aerial photograph, used as a reference in making of training points and testing points. These 29 input variables then evaluated to determine the most important variable and to map coffee plants. The results of this study indicate that the random forest algorithm using optimal parameters (ntree: 1000, mtry: all variables, and minimum node size: 6) can be used for mapping coffee plants providing total value, kappa statistics, producer accuracy, and user accuracy, respectively, 79.333%, 0.774, 92,000%, and 90.790%. In addition, random forest classification using the 12 most important variables also provided producer accuracy and user accuracy of 79.333%, 0.774, 91.333%, and 84.570%, respectively.