Using multivariate cross correlations, Granger causality and graphical models to quantify spatiotemporal synchronization and causality between pest populations

Background: This work combines multivariate time series analysis and graph theory to detect synchronization and causality among certain ecological variables and to represent significant correlations via network projections. Four different statistical tools (cross-correlations, partial cross-correlations, Granger causality and partial Granger causality) utilized to quantify correlation strength and causality among biological entities. These indices correspond to different ways to estimate the relationships between different variables and to construct ecological networks using the variables as nodes and the indices as edges. Specifically, correlations and Granger causality indices introduce rules that define the associations (links) between the ecological variables (nodes). This approach is used for the first time to analyze time series of moth populations as well as temperature and relative humidity in order to detect spatiotemporal synchronization over an agricultural study area and to illustrate significant correlations and causality interactions via graphical models. Results: The networks resulting from the different approaches are trimmed and show how the network configurations are affected by each construction technique. The Granger statistical rules provide a simple test to determine whether one series (population) is caused by another series (i.e. environmental variable or other population) even when they are not correlated. In most cases, the statistical analysis and the related graphical models, revealed intra-specific links, a fact that may be linked to similarities in pest population life cycles and synchronizations. Graph theoretic landscape projections reveal that significant associations in the populations are not subject to landscape characteristics. Populations may be linked over great distances through physical features such as rivers and not only at adjacent locations in which significant interactions are more likely to appear. In some cases, incidental connections, with no ecological explanation, were also observed; however, this was expected because some of the statistical methods used to define non trivial associations show connections that cannot be interpreted phenomenologically. Conclusions: Incorporating multivariate causal interactions in a probabilistic sense comes closer to reality than doing per se binary theoretic constructs because the former conceptually incorporate the dynamics of all kinds of ecological variables within the network. The advantage of Granger rules over correlations is that Granger rules have dynamic features and provide an easy way to examine the dynamic causal relations of multiple time-series variables. The constructed networks may provide an intuitive, advantageous representation of multiple populations' associations that can be realized within an agro-ecosystem. These relationships may be due to life cycle synchronizations, exposure to a shared climate or even more complicated ecological interactions such as moving behavior, dispersal patterns and host allocation. Moreover, they are useful for drawing inferences regarding pest population dynamics and their spatial management. Extending these models by including more variables should allow the exploration of intra and interspecies relationships in larger ecological systems, and the identification of specific population traits that might constrain their structures in larger areas.