Land-Cover Mapping of Agricultural Areas Using Machine Learning in Google Earth Engine

摘要

Land-cover mapping is critically needed in land-use planning and policy making. Compared to other techniques, Google Earth Engine (GEE) offers a free cloud of satellite information and high computation capabilities. In this context, this article examines machine learning with GEE for land-cover mapping. For this purpose, a five-phase procedure is applied: (1) imagery selection and pre-processing, (2) selection of the classes and training samples, (3) classification process, (4) post-classification, and (5) validation. The study region is located in the San Salvador basin (Uruguay), which is under agricultural intensification. As a result, the 1990 land-cover map of the San Salvador basin is produced. The new map shows good agreements with past agriculture census and reveals the transformation of grassland to cropland in the period 1990-2018.