Ground beetle occurrence and moor degradation: modelling a bioindication system by automated decision-tree induction and fuzzy logic

Abstract

Abstract Mostly due to anthropogenic influences, moors in Central Europe are undergoing rapid succession. We tried to indicate five degradation stages of a peculiar type of moors, so-called kettle-hole moors, which are located in the region of terminal moraines left by the Pleistocene glaciation up until 10,000 years B.P. Using a database of more than 12,000 individuals of carabid beetles belonging to more than 100 species sampled in 25 sites of various stages of degradation, we tried to construct a bioindication system by the use of machine learning techniques. Model-tree induction yielded a classifier consisting of three decision-trees which was moderately successful in classifying moors into the correct degradation stages by using information on only nine species, thus reducing the biotic information to a necessary minimum and creating an extremely parsimonious model. A translation of the decision-trees into fuzzy rule-based models increased bioindicatory efficiency: only 1 of the 10 unseen cases used for validation deviated more than one class from the correct degradation stage. At the moment, this model reflects a static picture of kettle-hole moor degradation, confirming what can be said after macroscopic examination. Future studies must show whether the more subtle dynamic aspects of moor degradation can also be indicated by carabid beetle occurrence.