Research On The Classification Of China's Vegetation Type Of Formation Based On The Community Detecting Algorithms

The classification of Chinese vegetation has been a key focus of biologists.In this research,with respect to 5 temperal zones including tropic zone,tropic and subtropic zone,subtropic zone,temperate combined with cold temperate zone and alphine,subalphine,paramos and permanent freezing zone,1079 formations located in China's land are investigated by constructing coexistence network with defining their connection as the cooccurance in the 2-dimension hydrothermal grid.With the networks based on temperal zones built,further analysis on the structure and function of networks is executed and to detect clusters in the coexistence networks,integrated network clustering is implemented with 4 different community detection algorithms which includes community structure detection based on edge betweenness,community structure detection based on short random walks,community structure detection based on greedy optimization of modularity and community structure detection based on the leading eigenvector of the community matrix.Besides,a community detection algorithm based on percolation theory is put forward and its effectiveness on clustering is tested with above 5 different coexistence networks.The main results are listed as following:(1)The five coexistence networks based on varied temperal zones are well-connected with few isolated nodes appearing in the structure.Both the weighted degree and connection weights follow a power-law distribution,which indicates a small portion of nodes have a large portion of the connection while the other nodes of networks have a small portion of connection demonstrating the scale-free characteristic.(2)With the clustering concerning maximizing modularity,community structure detection based on greedy optimization of modularity and community structure detection based on the leading eigenvector of the community matrix demonstrated the highest classifying accuracy of 67.24%while community structure detection based on edge betweenness shows the highest accuracy of 80.24%when clustering based on specific number of classes for the coexistence network of tropic zone.For the other 4 networks,community structure detection based on edge betweenness illustrates the consistent best behavior with the accuracy being 77.98%,71.00%,79.17%and 75.08%respectively when clustering based on maximizing modularity and community structure detection based on the leading eigenvector of the community matrix all shows the highest accuracy of 87.42%,87.35%,95.06%and 91.49%respectively when clustering based on specific number of classes.(3)A algorithm based on percolation theory is put forward.For the networks of tropic zones and temperate combined with cold temperate zone,the accuracies of the algorithm based on percolation are 79.77%and 91.83%,which is almost equal to the best behavior of aboved 4 community detection.For the other 3 networks,the accuracy reaches 81.70%,78.68%and 82.88%,which is apparently lower than the best behavior of 4 community detection,but close to the second best in accuracy.In the research,Chinese vegetation classification is studied with network techniques and the effectiveness of using temperature and precipitation to classify Chinese vegetation is tested,which offered theory reference for plant classification and biology research including biodiversity recognization and vegetation restoration.