Modelling natural forest expansion on a landscape level by multinomial logistic regression

Abstract

Natural forest expansion is one of the most relevant landscape changes in many temperate countries. Although large areas are involved, relatively few studies have been carried out with the objective of unravelling the specific impact of the individual factors characterising the sites prone to such a process. The aim of this article is to present a research tool for assessing the factors characterising farmland sites prone to natural conversion from crop growing and pasture to forests and other wooded land (OWL), and for predicting the probability of such a land-use change. The methodological approach is based on multinomial logistic regression. As a case study, the approach was applied to land-use classification repeated on the same sites in a large area of central Italy on two successive occasions, spanning two decades, from the beginning of the 1980s up to 2002. Of all the factors assessed, landscape attributes were identified as a sufficient subset for quantitative prediction of change from farmland to OWL or to forest. The tested modelling approach is explicitly empirical and planning-oriented. From a quantitative point of view, the precision of the models may be only indicative for assessing land-use change probability for single observations, while it is appropriate for predicting mean probabilities at a landscape mapping level, where it is possible to sample a number of sites. At this level, the approach is a useful tool for simulating future landscape scenarios related to natural forest expansion.