Authors: Heng Wan*, Virginia Polytechnic Institute & State University
Topics: Remote Sensing, Geographic Information Science and Systems, Land Use and Land Cover Change
Keywords: Image classification, random forest, land cover change
Session Type: Poster
Start / End Time: 5:20 PM / 7:00 PM
Room: Napoleon Foyer/Common St. Corridor, Sheraton, 3rd Floor
Presentation File: No File Uploaded
Abstract: In the US, National Land-cover database (NLCD) has been widely used as the baseline land-cover data for a variety of research and applications. Currently, there is a 5-year time lag between the image-capture date and the product-release date. During this time lag, some areas may have experienced substantial land cover changes, especially in urban/suburban settings. There are also land change modelling applications that require higher temporal frequency (e.g., annual) input data. This study aims at finding an effective method to update the NLCD with a high thematic accuracy. Using 2006/2011 NLCD as baseline product, we designed an image analytical approach to derive annual land cover products by integrating Landsat 7, Landsat 8, and Sentinel 2 multi-temporal imagery. Several machine learning algorithms, Random Forest, neural network (NN), and support vector machine (SVM) are applied as the classification algorithm, and different sampling methods for obtaining the training data are tested to find the best combination for image classification. Albermarle-Pamlico Estuarine System (APES) is selected as the research area due to the rich reference data available for accuracy assessment.