A Method to Identify Dacrydium pierrei Hickel Using Unmanned Aerial Vehicle Multi-source Remote Sensing Data in a Chinese Tropical Rainforest

• Published in: Journal of the Indian Society of Remote Sensing Vol. 50; no. 1; pp. 25 - 35
• Main Authors: Peng, Xi, Liu, Haodong, Chen, Yongfu, Chen, Qiao, Wang, Juan, Li, Huayu, Zhao, Anjiu
• Format: Journal Article
• Language: English
• Published: New Delhi Springer India 2022
• Subjects: Earth and Environmental Science; Earth Sciences; Remote Sensing/Photogrammetry; Research Article; Unmanned aerial vehicles (UAV); Chinese tropical rainforest; Tree species identification; Feature selection; Random forest; Multi-source remote sensing data
• ISSN: 0255-660X; 0974-3006
• DOI: 10.1007/s12524-021-01453-z

Identifying special species in tropical forests is an important topic in forest resource management, and the use of a single type of remote sensing data for identification of species has limited accuracy. To analyze the ability of various unmanned aerial vehicle (UAV) remote sensing data for identifying target species, this study used three types of UAV remote sensing data (light detection and ranging (LiDAR), red, green, blue (RGB), and multispectral) to identify Dacrydium pierrei Hickel ( D. pierrei ) in Chinese tropical forests. The study compared the effects of using various combinations of UAV remote sensing data on the accuracy of D. pierrei identification and identified the optimal combination. (1) Random forest feature selection improved the accuracy of identification of D. pierrei by UAV multiple source remote sensing data: The producer accuracy (PA) was increased up to by 4.62%. (2) The following eight features were most useful for identifying D. pierrei : four features from multispectral images (DR_Standard, RE_Standard, DR_Mean, and B_Brightne), two features from RGB images (B_Standard and B_Mean), and two features from LiDAR images (INT_kurtosis and INT_aad). (3) Combining remote sensing data by integrating up to three types of data sources improved the accuracy of D. pierrei identification. When using a single type of remote sensing data, multispectral data gave the highest identification accuracy. When combining two types of remote sensing data, RGB and multispectral data achieved the best overall effect, and the highest overall identification accuracy, of more than 90%, was obtained by combining three types of remote sensing data.