Modelling freshwater species distributions can be challenging as biases due to obscure or difficult to observe species may lead to underestimating the true habitat suitability. In this paper, authors present a solution to this problem involving optimizing models based on multiple objectives rather than just one.
Using data from the GBIF-mediated Limnodata Neerlandica dataset the authors evaluated the performance of their multi-objective optimization (MOO) approach—called the non-dominated sorting genetic algorithm II—against regular single objective models of 11 pollution-sensitive macroinvertebrate species (including Cloeon dipterum shown above).
Their results showed that the MOO approach is two-to-four times more efficient at identifying large range distributions, while only requiring four per cent longer runtimes per training. To support decision-making, the authors propose a closer collaboration between model developers and freshwater managers to set environmetal standard limits on a more objective basis.