Abstract: To explore an efficient method for mapping young mangrove forest exactly in local level,Unsupervised classification and object-oriented nearest neighbor classification were test on WorldView-3 remote sensing image in Maoweihai bay in Guangxi,south China,where a plenty of young mangrove forests grow.The results indicated that the overall accuracies of unsupervised and object-oriented classification were 95.8% and 96.2% respectively,and the kappa indexes were 0.906 8 and 0.913 7 respectively,that meant two simple methods could be used to accurately map the distribution of young mangrove forest.But the former output represented only the crown coverage of young trees and did not include the bare bead between the trees,and there was a significant salt and pepper effect on the map,and the latter output was the extent of young tree distribution for it included not only the extent of young trees crown but also the beach near by the trees,therefore,the object-oriented classification was more suitable for extracting the extend information of young mangrove forest than pixel-based method.Young mangrove forests have small crowns,high resolution remote sensing image must be used to map their extent,1.0 m or small resolution of images were recommended,0.3 m resolution of image was preferable.On the other hand,the images acquired in low tide period were needed.

Key words: young mangrove forest, remote sensing, information extraction, Maoweihai Bay of Guangxi