| Abstract: |
| Oil palm plantations are rapidly expanding in the tropics, which leads to deforestation and other associated damages to biodiversity and ecosystem services. Forest researchers and practitioners in developing nations are in need of a low-cost, accessible and user-friendly tool for detecting the establishment of industrial oil palm plantations. Google Earth Engine (GEE) is a cloud computing platform which hosts publicly available satellite images and allows for land cover classification using inbuilt algorithms. These algorithms conduct pixel-based classification via supervised learning. We demonstrate the use of GEE for the detection of industrial oil palm plantations in Tripa, Aceh, Indonesia. We performed land cover classification using different spectral bands (RGB, NIR, SWIR, TIR, all bands) from our Landsat 8 image to distinguish the following land cover classes: immature oil palm, mature oil palm, non-forest non-oil palm, forest, water, and clouds. The overall accuracy and Kappa coefficient were the highest using all bands for land cover classification, followed by RGB, SWIR, TIR, and NIR. Classification and Regression Trees (CART) and Random Forests (RFT) algorithms produced classified land cover maps which had higher overall accuracies and Kappa coefficients than the Minimum Distance (MD) algorithm. Object-based classification and using a combination of radar- and optic-based imagery are some ways in which oil palm detection can be improved within GEE. Despite its limitations, GEE does have the potential to be developed further into an accessible and low-cost tool for independent bodies to detect and monitor the expansion of oil palm plantations in the tropics. |
|
