Coexistence mechanisms dependent on environmental variation not only explain the competitive coexistence in species rich communities (e.g. plants), but also address the spatial and temporal dynamics of biological invasions. Although well developed in theory, those mechanisms lack empirical tests. We designed and conducted a field study to test the hypothesis that population fluctuations driven by spatial environmental variation at smaller scales give rise to the spatial storage effect at larger scales to enhance the coexistence between invasive and native species. We chose the invasive desert winter annual plant Brassica tournefortii and its two competing native species Chaenactis stevioides and Plantago ovata as focal species. We set up experimental plots under a spatial hierarchy: from the smallest scale of neighborhoods to the intermediate scale of subhabitats (open vs. under creosote shrub) and to the largest scale of habitats (sand flat vs. semi-active dune). We compared the above ground biomass of target individuals subject to neighbor-removal versus control treatment in order to measure the individuals’ response to the environment (E) and to the competition (C). Using these measurements we calculated the covariance between each species’ E and C (EC covariance) at both low and normal density state, which are essential variables for quantifying the storage effect.
Results/Conclusions
We showed that guided by an appropriate statistical model, it was possible to quantify the strength of the spatial storage effect operating between multiple species across multiple scales. It was necessary to have the empirical knowledge of the sampling errors of E in order to obtain accurate estimates of the EC covariance. The sampling errors were traditionally measured by having two neighborhoods assigned for E measurement in an experimental cluster in addition to the neighborhood assigned for C measurement. A higher spatial resolution can be achieved by separating clusters used for measuring EC covariance and those for measuring sampling errors of E and thus reducing the number of neighborhoods in each cluster from three to two.