Discovering the most important factors for communities of soil microarthropods using machine learning

Abstract

In agricultural soil a set of anthropogenic events shapes the ecosystem processes and populations. The risk of impact from anthropogenic sources on the soil environment is almost exclusively assessed for chemicals, although in agriculture other factors like crop and tillage have large impact too. Thus, the farming system as a whole should be evaluated and ranked according to its environmental benefits and impacts. Our starting point is the availability of data sets describing the agricultural events and the soil biological parameters. Using that datasets and machine learning methods for inducing regression and model trees, we produced empirically based models useful for predicting the soil quality in terms of quantities describing the soil microarthropod community from agricultural measures. However, inducing models for predicting soil quality is not our only goal. What we are also interested is to discover additional knowledge on a higher level and identify the most important factors for population densities of springtails and mites and their biodiversity. We do that by preferring smaller and simpler models to bigger and more complex models, while trying to minimize the performance loss of the models at the same time. Using that approach we identify that microarthropod communities are most sensitive to crops and tillage.