Neighborhood predictors of plant performance

We developed models of inter-individual interference to predict the fecundity of individuals in populations of the annual plant species, Arabidopsis thaliana. An individual plant is modeled as having a neighborhood which is a circular area of fixed radius with the plant at its center. Other plants which share the circle with the focal plant are termed neighbors of the focal plant. We developed an index of neighborhood interference which is the independent variable in a non-linear regression model that predicts individual plant fecundity. We present methods of exploratory data analysis that are useful in determining a best neighborhood radius, defined as that radius which minimizes residual sum of squares, and in deciding on the functional form of the interference index. In developing the interference index for Arabidopsis, we focus on aspects of the spatial distribution of neighbors: their number, distance and angular dispersion.We found that a best (or optimal) neighborhood radius can be resolved, which provides the best predictor of plant performance. Fecundity predictors based on adult neighbors were noticeably better than those based on neighbors at the seedling stage. Rosettes of Arabidopsis may change location during development (they fall over) and the new "fallen" positions do provide some improvement in the predictor. Taking into account distance to neighbors within the neighborhood provided only negligible improvement in the model. Finally, the incorporation of angular dispersion in the crowding index produced a considerably better fit. The fecundity predictor that included number of neighbors and angular dispersion in the crowding index explained about 70% of the variation in individual seed set.