Application of machine learning techniques to the analysis of soil ecological data bases: relationships between habitat features and Collembolan community characteristics

We applied novel modelling techniques (neural networks, tree-based models) to relate total abundance and species number of Collembola as well as abund ances of dominant species to habitat characteristics and compared their pre dictive power with simple statistical models (multiple regression, linear r egression, land-use-specific means). The data used consisted of soil biolog ical, chemical and physical measurements in soil cores taken at 396 points distributed over a 50 x 50 m sampling grid in an agricultural landscape in southern Germany. Neural networks appeared to be most efficient in reflecti ng the nonlinearities of the habitat-Collembola relationships. The underlyi ng functional relations, however, are hidden within the network connections and cannot be analyzed easily. Model trees - next in predictive power to n eural networks - are much more transparent and give an explicit picture of the functional relationships. Both modelling approaches perform significant ly better than traditional statistical models and decrease the mean absolut e error between prediction and observation by about 16-38%. Total carbon co ntent and measurements highly correlated with it (e.g. total nitrogen conte nt, microbial biomass and respiration) were the most important factors infl uencing the Collembolan community, This is in broad agreement with existing knowledge. Apparent limitations to predicting Collembolan abundance and sp ecies number by habitat quality alone are discussed. (C) 2000 Elsevier Scie nce Ltd. All rights reserved.