University of Twente Student Theses


Seagrass identification and mapping using remote sensing and Google Earth Engine

Anand, Ankita (2020) Seagrass identification and mapping using remote sensing and Google Earth Engine.

Full text not available from this repository.

Full Text Status:Access to this publication is restricted
Abstract:Seagrass, the aquatic flowering plants is the most valuable habitat of marine ecosystems. They contribute to the global carbon cycle by storing large volumes of carbon thereby controlling Earth’s temperature. Seagrass species across the world are under a major threat due to anthropogenic impact and climate change, calling for solutions ensuring their conservation and global-scale mapping for regular monitoring. Image processing of composite tiles of satellite images covering large areas has subsequently led to “Big Data” problems. Cloud computing has bridged the gap by addressing data management issues; and GEE is its recent advancement offering geospatial analysis. This study utilizes the cloud computing power, Google earth engine, remote sensing and machine learning methods to propose a framework for mapping seagrass beds using Landsat-8 satellite data. The approach is to exploit GEE in-built machine learning algorithms such as Classification and Regression Trees (CART), Random Forests (RF) and Support Vector Machine (SVM) to analyse a pixel-based and object-based image analysis (OBIA) methods for mapping and monitoring seagrass habitat. The study shows, a kernel-based, SVM outperformed all the classifiers producing maximum accuracy of 92.50%, whereas RF produced the accuracy of 90.00% and CART gave 85.00%. Also, object-based image classification is performed by replacing the pixel grid with the group of meaningful regions defining natural boundaries called superpixels. Hence, GEE in-built superpixel segmentation algorithm, Simple Non-Iterative Clustering (SNIC) is implemented for OBIA and SVM classifier was used to perform classification. The overall accuracy obtained by pixel-based method using SVM is 92.50% and kappa coefficient is 0.81 while overall accuracy achieved by implementing superpixel clustering for OBIA is 95.00% and kappa coefficient is 0.87. This increase in accuracy of 2.50% signifies the improvement in the standard method of image analysis, i.e. pixel-based methods. The use of the superpixel segmentation technique was found effective for OBIA. This study is the contribution to Indian marine biodiversity for sustainable conservation and management of seagrass.
Item Type:Essay (Master)
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Programme:Geoinformation Science and Earth Observation MSc (75014)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page