Edge correction and regression models for quantifying single-tree influence on understory vegetation
Doktorsavhandling, 2004

The understory is the layer of vegetation in the forest situated under the canopies of the trees. Some species of understory vegetation benefit from the surrounding trees, e.g by the provided nutrients, while others are restricted, e.g. by the limited amount of light. In this thesis, statistical methods have been developed that allow the quantification of the effect of the trees on the vegetation. This is of importance for understanding ecological dynamics as well as for decisions regarding biodiversity and forest management. The effect of the trees was calculated with an index called influence potential on a quadrat (IPQ), which uses the size of the trees and their spatial distribution. The abundance of the understory was assessed by the proportion of ground covered. The Finnish Forest Research Institute provided the data consisting of observations on trees and vegetation from more than 3000 plots distributed over Finland. In Paper I, an edge correction for IPQ was developed using tools from spatial point processes. The correction eliminates the bias that originates when trees outside the plot are ignored in the calculations. Paper II suggests a logistic model for studying the absence and presence of a plant species, but conditioned on a sufficient statistic for the large-scale effects. These effects are present throughout the data and are due to factors such as climate and latitude. By conditioning, the effects are accounted for but do not require to be estimated. Paper III develops regression models for proportions assuming that the errors are Beta distributed. It uses an alternative parameterization that is more flexible and allows estimation methods not possible with the standard form. Paper IV is an exploratory study of the relationship between IPQ and the abundance of 12 species of understory vegetation.

statistical modeling


logistic regression

spatial point process

edge effects

beta distribution

influence potential



Sharon Kühlmann-Berenzon

Göteborgs universitet

Chalmers, Institutionen för matematisk statistik





Doktorsavhandlingar vid Chalmers tekniska högskola. Ny serie: 2078