Effects of data prevalence on species distribution modelling using a genetic takagi-sugeno fuzzy system

Abstract

Uncertainties originating from observation data and modelling approaches can affect model accuracy and thus impact on the applicability and reliability of a model. This paper aims to assess the effects of data prevalence (i.e., proportion of presence in the entire data set) on species distribution modelling and habitat preference evaluation using a 0-order genetic Takagi-Sugeno fuzzy model. The effects were evaluated based on the model accuracy and habitat preference curves (HPCs). In order to avoid the data uncertainty, virtual species data were generated using hypothetical HPCs under different assumptions on the interaction between habitat variables and habitat preference of a virtual fish. In total, thirteen data sets under three different interaction scenarios were generated. The model accuracy of resulting models was different according to the data prevalence, whereas different trends between data sets under different interaction scenarios were observed. Although the HPC shapes were similar across data sets, the HPCs were different according to the data prevalence, of which a higher prevalence can result in a uniform HPC. This study demonstrates possible influences of data prevalence on the species distribution modelling. Further study is needed for a better solution to cope with the prevalence-related problems in ecological modelling.