Optimizing k-Nearest Neighbors for Mapping Vegetation Cover of Land Desertification Areas using Landsat 8 image

Authors: Guangxing Wang*, Southern Illinois University at Carbondale, Hua Sun, Research Center of Forestry Remote Sensing & Information Engineering, Central South University of Forestry and Technology, Changsha 410004, Hunan, China, Qing Wang, Department of Geography and Environmental Resources, Southern Illinois University, Carbondale IL USA, Hui Lin, Research Center of Forestry Remote Sensing & Information Engineering, Central South University of Forestry and Technology, Changsha 410004, Hunan, China, Jiping Li, Research Center of Forestry Remote Sensing & Information Engineering, Central South University of Forestry and Technology, Changsha 410004, Hunan, China, Siqi Zeng, Research Center of Forestry Remote Sensing & Information Engineering, Central South University of Forestry and Technology, Changsha 410004, Hunan, China
Topics: Remote Sensing, Arid Regions, China
Keywords: Land desertification, Optimized k-Nearest Neighbors, Landsat image, Vegetation cover mapping, Duolun County, Kangbao County
Session Type: Poster
Day: 4/13/2018
Start / End Time: 1:20 PM / 3:00 PM
Room: Napoleon Foyer/Common St. Corridor, Sheraton, 3rd Floor
Presentation File: No File Uploaded


Land degradation and desertification in arid and semi-arid areas is of great concern. Accurately mapping vegetation cover in the areas is critical, but challenging because of being sparsely vegetated, rarely populated and difficult for collecting field observations. Traditional methods can’t accurately predict vegetation cover in the areas. Nonparametric k-nearest neighbors (kNN) algorithm widely used in estimation of forest parameters is a good alternative because of its characteristics and flexibility in which k most similar neighbors are selected based on an image space instead of a geographic space. However, using a global constant k value in kNN limits its ability of increasing prediction accuracy due to spatial variability of vegetation cover in the areas. In this study, an improved method that optimizes spatially determining the number of nearest plots (k value) for kNN, denoted with Opt_kNN, was proposed to map vegetation cover in both Duolun County and Kangbao County located in Inner Mongolia and Hebei Province of China, respectively, using Landsat 8 images and sample plot data. The Opt_kNN was compared with several widely used methods to improve the mapping for the study areas. The results showed that: 1) Most of the red and near infrared bands relevant vegetation indices had significant contributions to improving the mapping accuracy; 2) Opt_kNN led to higher prediction accuracy than all other methods; 3) Opt_kNN generated spatially variable and optimized k values. This implied that the improved method is very promising to increase the accuracy of mapping vegetation cover in arid and semi-arid areas.

Abstract Information

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

To access contact information login