Plantation Tree Species Classification With Multi-source High Resolution Remotely Sensed Data

The global availability of high spatial resolution images makes mapping tree species distribution possible for better management of forest resources.Previous research mainly focused on mapping single tree species,but information about the spatial distribution of all kinds of trees,especially plantations,is often required.This research selected Wangyedian Forest Farm as study area,five kinds of features extracted from multi-temporal ZiYuan-3 multispectral and stereo images were considered:(1)pixel-based spectral features such as spectral signatures and vegetation indices;(2)spatial-based features such as textural images and image segmentation;(3)temporal features such as growing and deciduous seasons;(4)height–based variables that can reflect the difference of forest stand structures;and(5)topographic factors such as slope,aspect and elevation.Through comparative analysis of different datasets(leaf-off,leaf-on,and their combination),different combination of variables(V1-spectral band;V2-V1+vegetation indices,textures,segmented shapes indices,topographic factors;V3-V2+height features)and different classifiers(maximum likelihood classifier(MLC),artificial neural network(ANN),knearest neighbor(kNN),decision tree(DT),random forest(RF),and support vector machine(SVM))in improving land cover(all forest and non-forest classes),forest(all tree species classes),and tree species mapping performance.The results show that:(1)Use of multiple source data—spectral bands,vegetation indices,textures,and topographic factors—considerably improved land-cover and forest classification accuracy compared to spectral bands alone,which the highest overall accuracy of 84.5% for land cover classes was from the SVM,and of 89.2% for forest classes was from the MLC.The combination of multiple source data also improved land cover classification by 3.7-15.5% and forest classification by 1.0-12.7% compared to spectral image alone.(2)The combination of leaf-on and leaf-off seasonal images further improved classification accuracies by 7.8-15.0% for land cover classes and by 6.0-11.8% for forest classes compared to single season spectral image.(3)MLC provided better land-cover and forest classification accuracies than machine learning algorithms when spectral data alone were used;but some machine learning approaches such as RF and SVM provided better performance than MLC when multiple data sources were used;further addition of canopy height features into multiple source data had no or limited effects in improving land-cover or forest classification,but indeed improved some tree species such as birch and Mongolia scotch pine classification accuracies.(4)Considering tree species classification,Chinese pine,Mongolia scotch pine,red pine,aspen and elm,and other broadleaf trees have classification accuracies of over 92%,and larch and birch have relatively low accuracies of 87.3% and 84.5%.However,these high classification accuracies are from different data sources and classification algorithms,and no one classification algorithm provided the best accuracy for all tree species classes.This research aims to identify suitable variables and algorithms for classifications of land cover,forest,and tree species.This research implies the same data source and classification algorithm cannot provide the best classification results for different land cover classes.It is necessary to develop a comprehensive classification procedure using expert-based approach or hierarchical-based classification approach that can employ specific data variables and algorithm for each tree species class.