Predictors of pre-European deforestation on Pacific islands: a re-analysis using modern multivariate non-parametric statistical methods

The many different islands of the Pacific provide a natural experiment that is ideal for studying predictors affecting plant cover and habitat-type. Previous research into pre-European deforestation and forest replacement across the Pacific islands detected multiple significant environmental and cultural variables. Here we re-analyse data for 67 islands using modern multivariate non-parametric statistical methods. For this analysis, neither parametric assumptions nor transformations are needed. According to our results there are congruities, but also differences from previous work. Although our analysis controls the familywise error rate, we found more relevant variables. Regarding deforestation, rainfall is the most important variable, but tephra and absolute latitude are also highly significant. Rainfall and tephra are negatively correlated with deforestation, whereas absolute latitude and deforestation have a positive correlation. Regarding forest replacement, area, dust and tephra are highly significant. These three variables are negatively correlated with the extent of forest replacement. In summary, we confirm the strong influence of environmental predictors.