Forest change mapping using remote sensing and Google Earth Engine
Permanent link to Research Commons versionhttps://hdl.handle.net/10289/16299
With advancements in data sources, software, and image analysis techniques, remote sensing has become an efficient method for forest classification. However, access to this technology has been limited for developing countries due to the high cost of high resolution images and analysis software. A potential solution is that NASA and the European Space Agency provide free access to mid-low resolution satellite images. In addition, Google Earth Engine (GEE), a free cloud-based geospatial analysis platform, has allowed researchers from developing countries to conduct research without relying on costly remote sensing software. This study evaluates the suitability of the freely available images and the GEE platform for different forest management applications in Sri Lanka. The research focused on three different scenarios in vegetation mapping: identifying home gardens, monitoring invasive pines, and forest cover mapping in a tropical montane region.Home garden is an agroforestry class seen in tropical countries often overshadowed by global land cover classifications. This study used a random forest classification algorithm to classify the home garden, utilizing terrain data and Sentinel-2A images as the dataset. The results confirmed that the red band of Sentinel-2 and textural metrics derived from grey-level co-occurrence matrix analysis are effective in identifying home gardens from other forestry classes.Monitoring the spread of invasive pines is important for forest management in Sri Lanka. This study used Landsat satellite images for the years 2000 and 2021 to detect the exotic Pinus caribaea species invasion of native habitats. By using an image difference technique, this research identified that the extent of these invasive pines had an overall decline over the past 21 years. Further, results showed that broadleaved forests and grasslands located within 100m of the pine plantations’ borders were susceptible to invasion. In tropical regions, analysing forests can be difficult due to cloud cover. This research also focused on methods to enhance classification accuracy using multiple data sources: Sentinel-1, Sentinel-2, PALSAR-2, and elevation. Six different methods were tested. It was found that combining radar data with optical and topographical data improved vegetation classification accuracy. In addition, feature importance analysis using the random forest algorithm indicated that radar data plays a significant role in tropical land use classification.The three case studies used in this research demonstrate that GEE and the freely available mid-low resolution satellite images make the application of remote sensing in Sri Lanka a viable solution for the monitoring and mapping of landcover.
The University of Waikato
All items in Research Commons are provided for private study and research purposes and are protected by copyright with all rights reserved unless otherwise indicated.
- Higher Degree Theses