Estimation and Mapping Forest Attributes Using "k Nearest Neighbor" Method on IRS-P6 LISS III Satellite Image Data

 
Roya Abedi*, Amir Eslam Bonyad
 
Department of Forestry, Faculty of Natural Resources, University of Guilan, IRAN
* Corresponding author: abedi.roya@yahoo.com
 

Abstract. In this study, we explored the utility of k Nearest Neighbor (kNN) algorithm to integrate IRS-P6 LISS III satellite imagery data and ground inventory data for application in forest attributes (DBH, trees height, volume, basal area, density and forest cover type) estimation and mapping. The ground inventory data was based on a systematic-random sampling grid and the numbers of sampling plots were 408 circular plots in a plantation in Guilan province, north of Iran. We concluded that kNN method was useful tool for mapping at a fine accuracy between 80% and 93.94%. Values of k between 5 and 8 seemed appropriate. The best distance metrics were found Euclidean, Fuzzy and Mahalanobis. Results showed that kNN was accurate enough for practical applicability for mapping forest areas. 

Key words: Forest attributes, IRS, k Nearest Neighbor (kNN). 
 
Ecologia Balkanica, 2015, vol. 7, Issue 1, pp.93-102

Article № eb.15119, ICID: 1164568 [Full text - PDF]pdf

Back to content