How to map large scale mangroves with Google Earth Engine

Authors: Ying Lu*, Department of Geography, University at Buffalo, Le Wang, Department of Geography, University at Buffalo
Topics: Remote Sensing
Keywords: Global mangrove, Google Earth Engine, deep learning, data sampling
Session Type: Paper
Day: 4/5/2019
Start / End Time: 9:55 AM / 11:35 AM
Room: Roosevelt 6, Marriott, Exhibition Level
Presentation File: No File Uploaded

Due to the deforest of mangroves caused by climate change and human activity, large scale mangrove maps are under requirement. As data processing is tough for large scale mangrove mapping, regional scale studies dominated historically.
Fortunately, the publication of Google Earth Engine(GEE) offers a good platform to produce large scale maps.The objective of this study is to map the distribution of mangrove forests in large scale with Google Earth Engine (GEE) using Landsat imagery. Data pre-processing, data sampling and classification with deep learning were three major steps in this study. Initially, interested areas were divided into several climate regions. Regions unlikely to be mangroves were masked out in advance, such as water bodies and inland areas. Then, a data sampling strategy was applied to derive a reliable sample dataset for training and testing. Mangrove maps from other literature were synthesized. As mangroves located in inter-tidal areas, water and vegetation indexes were analyzed to drive the most suitable sample data. At last, deep learning classification models were trained for every climate region separately. The large scale mangrove map was a combination of the classification results from each climate region. Due to the widely distributed mangroves, China was selected as a study area. The overall mangrove area in China indicated a decrease trend. This result will serve as a good reference for the sustainable development and environment protection of the world. What’s more, with further improvement, the method is possible to be applied in global scale.

Abstract Information

This abstract is already part of a session. View the session here.

To access contact information login