Ecological multivariate systems offer a suitable data set on which to apply recent advances in information theory and causality detection. These systems are driven by the interplay of various environmental factors: meteorological and hydrological forcing, which are often correlated with each other at different time lags; and biological factors, primary producers and decomposers with both autonomous and coupled dynamics. Here, using conditional spectral Granger causality, we quantify directional causalities in a complex atmosphere-plant-soil system involving the carbon cycle. Granger causality is a statistical approach, originating in econometrics, used to identify the presence of linear causal interactions between time series of data, based on prediction theory. We first test to see if there was a significant difference in the causal structure among two treatments where carbon allocation to roots was interrupted by girdling. We then expanded the analysis, introducing radiation and soil moisture. The results showed a complex pattern of multilevel interactions, with some of these interactions depending upon the number of variables in the system. However, no significant differences emerged in the causal structure of above and below ground carbon cycle among the two treatments.