Biodiversity Monitoring Based On Hyperspectral Images Of Plant Canopies In Chenshan

Biodiversity is the basis of sustainable development of human society.The detailed monitoring and research on it is the inevitable requirement of the United Nations Convention on Biological Diversity,and as well as scientific guarantee to deal with the rapid loss of biodiversity.As a supplement to the classical method of biodiversity surveying,the survey method based on hyperspectral image provides a prospect for greatly improving the efficiency and precision of biodiversity monitoring.Existing studies have proved that there is a close relationship between hyperspectral images of plants in simple indoor scenes and biodiversity.This provides a theoretical basis for monitoring biodiversity based on UAV-based canopy hyperspectral images.However,it is still unknown whether this relationship can be detected in the complex outdoor light environment,especially in the plant community with high canopy species diversity.This shortage,to some extent,limited the capacity of hyperspectral images based biodiversity monitoring and related studies.In this study,using UAV-based hyperspectral imaging system to collect the canopy hyperspectral images of 121 tree species.Based on ground survey data,the relationships between canopy spectral diversity and species diversity and phylogenetic diversity are studied under high plant species diversity environment,and the possibility of monitoring species diversity through species identification method is explored.The main conclusions are as follows:(1)there is a significant positive relationship between canopy spectral diversity and species diversity(R~2=0.68,P<0.01),species diversity increased with the increasing spectral diversity;(2)there is no significant correlation between canopy spectral diversity and phylogenetic diversity(R~2=0.02,P=0.08),and canopy spectral distance has a very low interpretation for phylogenetic distance(R~2=0.05,P<0.01);(3)canopy hyperspectral images combined with multi-scale 3D deep convolutional neural network can achieve an overall classification accuracy of 96%.This indicates that even in the plant communities with high species diversity,high precision species diversity monitoring can be carried out by using canopy hyperspectral images.These results confirm that there is still a strong relationship between spectral diversity and species diversity in the outdoor complex light environment,and provide a scientific basis for the further development of species diversity monitoring based on hyperspectral images.At the same time,it proves the feasibility and superiority of including species identification into biodiversity monitoring,which is an important step for the ultimate realization of subtropical evergreen broad-leaved forest biodiversity monitoring based on hyperspectral images.