Constructing ecological interaction networks by correlation analysis: hints from community sampling

摘要

A set of methodology for constructing ecological interaction networks by correlation analysis of community sampling data was presented in this study. Nearly 30 data sets at different levels of taxa for different sampling seasons and locations were used to construct networks and find network properties. I defined the network constructed by Pearson linear correlation is the linear network, and the network constructed by quasi-linear correlation measure (e.g., Spearman correlation) is the quasi-linear network. Two taxa with statistically significant linear or quasi-linear correlation are determined to interact. The quasi-linear network is more general than linear network. The results reveled that correlation distributions of Pearson linear correlation and partial linear correlation constructed networks are unimodal functions and most of them are short-head (mostly negative correlations) and long-tailed (mostly positive correlations). Spearman correlation distributions are either long-head and short-tailed unimodal functions or monotonically increasing functions. It was found that both mean partial linear correlation and mean Pearson linear correlation were approximately 0. The proportion of positive (partial) linear correlations declined significantly with the increase in taxa. The mean (partial) linear correlation declined significantly with the increase of taxa. More than 90% of network interactions are positive interactions. The average connectance was 9.8% (9.3%) for (partial) linear correlation constructed network. The parameter λ in power low distribution (L(x)=x-λ) increased as the decline of taxon level (from functional group to species) for the partial linear correlation constructed network. λ is in average 0.8 to 0.9. The number of (positive) interactions increased with the number of taxa for both linear and partial linear correlations constructed networks. The addition of a taxon would result in an increase of 0.4 (0.3) interactions (positive interactions) in the partial linear correlation constructed network. And the addition of a taxon would result in an increase of 3 interactions (positive interactions) in the linear correlation constructed network. For partial linear correlation constructed network, the network connectance decreased as the number of taxa. The constant connectance hypothesis did not hold for our networks. It was found that network structure changed with season and location. The same taxon in the network would connect to different taxa as the change of season and location. A higher level of species aggregation may used to find a more stable network structure. Positive interactions were considered to be caused mainly by mutualism, predation/parasitism, etc. the number and portion of positive interactions may be the most important indices for community stability and functionality. Mutualism is the most significant trophic relationship, seconded by predation/parasitism, and competition is the worst for community stability.