Detailed and automated classification of land use/land cover using machine learning algorithms in Google Earth Engine
Xia Pan, Zhenyi Wang, Yong Gao, Xiaohong Dang, Yanlong Han
Abstract
All the supervised classification methods need sufficient and efficient samples, which are commonly labeled by visual inspection. In this study, to resolve the issues of insufficient training samples and time-consuming, a novel method for detailed and automated LULC classification by LC_Type1 of MCD12Q1 IGBP schemes in the GEE cloud platform was proposed based on the RF and CART classifiers. The results present that the validation overall accuracy of the RF classifier is higher than the CART, 87.24% in Australia, and 85.18% in the USA, respectively. The automated classification results of the RF classifier are more concentrated than CART, which the RF classifier is more suitable for this automated method. Moreover, the proposed method can accomplish accurate, detailed, and automated LULC classification based on the GEE which is making satellite imagery computing an efficient, flexible, and fast process. The workflow provides a reliable method for detailed, automated, and remotely LULC classification.