Land-cover Mapping from Sentinel Time-Series Imagery on the Google Earth Engine: A Case Study for Hanoi

摘要

Over the past decade, satellite image processing is an overwhelming bulk of work. Recently, with rapid development in information technology, Google released Google Earth Engine (GEE), which is a powerful cloud computing platform, to help to improve the performance of geospatial big data archives and processing. In this study, we deployed a machine learning model to evaluate the capability of time series Sentinel imagery (Sentinel 2 A/B and Sentinel 1A) in landcover mapping for Hanoi in 2019. First, we evaluated several traditional machine learning models, as a result, XGBoost classifier stands out as the best model with 86% overall accuracy (OA). As Hanoi is a frequent cloud-covered area, the combination of optical data and radar data helps to improve the quality of the landcover map in 2019. The use of GEE has made it easier and faster through the provided JavaScript API when ensuring high accuracy.