The use of Landsat-8 and Sentinel-2 imageries in detecting and mapping rubber trees

摘要

Information on rubber tree (Hevea brasiliensis) areas and stages of rubber tree growth is needed in making decisions to maximise land use and for efficient farm management. The use of conventional methods in collecting this information requires a long time, high costs, and constraints to access certain areas. Therefore, this study was conducted to evaluate Landsat-8 OLI and Sentinel-2 images in detecting and mapping the rubber tree area. This study presents a pixel-based supervised classification approach to obtain an accurate map of land cover and rubber tree growth stage distribution using resampled 10 m spatial resolution of Sentinel-2 and pansharpened 15 m Landsat-8 OLI. Seven land cover classes (bare soil, water, mature rubber, immature rubber, oil palm, forest, and built-up area) were classified using support vector machine (SVM), artificial neural network (ANN) and spectral angle mapper (SAM). The results showed that the highest classification accuracy was obtained using SVM, 87.22% for Sentinel-2 and 85.74% for Landsat-8. Next, the classification accuracies of ANN were almost similar with 86.17% and 82.39% for Sentinel-2 and Landsat-8, respectively. SAM has produced less than 60% of acceptable accuracy for both datasets. The performance of the aforementioned classifiers was statistically tested using a McNemar test. The test showed that the p-value between SVM and ANN was not significant and thus, ANN and SVM produced similar accuracies and outperformed SAM for both cases. In this study, the best output produced via SVM from Sentinel-2 was selected to produce the thematic map due to the spatial accuracy advantage of Sentinel-2 compared to Landsat-8. The calculated areas of immature and mature rubber from the thematic map were 7.79 km(2) and 10.93 km(2), respectively, which then used to estimate the number of tappers needed for the management of rubber. It is concluded that the Sentinel-2 Multispectral Instrument (MSI) data can be recommended to be used in rubber cultivation area assessment.