A comparative test of the predictive power of neighbourhood models in natural populations of Lasallia pustulata

Three different neighbourhood models were tested to predict individual perf ormance in 50 natural populations of the saxicolous lichen Lasallia pustula ta (L.) Merat. Mean distance to neighbours was clearly the best predictor, accounting for most of the variation in 70% of the populations. In contrast a model based on the number of neighbours within a circle of fixed radius usually had the lowest predictive power. Polygon areas generated by Dirichl et tessellation had a predictive power slightly less than the nearest neigh bour approach. The predictive power of all three neighbourhood models was s ignificantly positively correlated, and the polygon and the nearest neighbo ur model was strongly so. The differences in predictive power are interpret ed as reflecting the degree of realism included in the models. The nearest neighbour approach uses actual distances to neighbours, a fairly direct mea sure of degree of interference in crowded populations. Tessellation models use these distances to generate semi-empirical areas of influence. In contr ast the circle model circumscribes a neighbourhood in an arbitrary and abst ract manner, and only secondarily take into account the number of organisms within that area. Considering the comparative merits of the models, it is a paradox perhaps that the most frequently used model in previous studies h as been the circle model.