Improving the Accuracy of Land Cover Classification Based on NDVI Time Series Library

Authors: Yanlin Yue*, Key Laboratory of Geographic Information Science, Ministry of Education, East China Normal University, Shunbao Liao, College of Ecology and Environment, Institute of Disaster Prevention, Guangxing Gi, Key Laboratory of Geographic Information Science, Ministry of Education, East China Normal University, Jincai Zhao, Key Laboratory of Geographic Information Science, Ministry of Education, East China Normal University, Zhizhu Lai, Key Laboratory of Geographic Information Science, Ministry of Education, East China Normal University
Topics: Remote Sensing
Keywords: land cover classification; accuracy improvement; NDVI time series library; time series similarity measure.
Session Type: Paper
Day: 4/6/2019
Start / End Time: 5:00 PM / 6:40 PM
Room: Balcony B, Marriott, Mezzanine Level
Presentation File: No File Uploaded


For improving the accuracy of land cover classification, this study proposed a novel approach. In our study, the existing land cover product (MODIS MCD12Q1), was first divided into two parts: high precision area and low precision area. Then, NDVI time series library was constructed and the land cover information of the study region was extracted based on this library. Time series similarity measure was utilized to extract land cover information, and the results demonstrated that minimum distance method (MD) performed better than spectral angle mapper method (SAM). Meanwhile, we proved that there existed a close relevance between NDVI time series and land cover classification. Above the procedures and analyses, a new land cover product (AMCD_2013) was produced. At last, we evaluated the classification accuracy of the original land cover product (MCD12Q1) and the improved land cover product (AMCD_2013). The comparison results of comparative evaluation (accuracy assessment based on existing product) clearly noticed that the overall accuracy and Kappa coefficient of AMCD_2013 were improved 10.29% (from 72.76% to 83.05%) and 0.25 (from 0.40 to 0.65) respectively, compared with MCD12Q1. In addition, the sample evaluation (accuracy assessment based on field survey samples) results demonstrated that the overall accuracy of AMCD_2013 (81.72%) was 17.20% higher than that of MCD12Q1 (64.52%). Above results proved the feasibility of the method that improving the accuracy of land cover classification through ameliorating the low precision area of the existing land cover product based on NDVI time series library.

Abstract Information

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

To access contact information login