An efficient approach to answering ecological and biogeographical questions, species distribution modelling depend on reliable occurrence data and suitable model algorithms. In models of terrestrial species, geographical bias can be ameliorated through filtering, however, less is known about the uncertainties affecting models of marine species, which are also influenced by a scarcity of predictor variables.
In a study of the common seagrass Cymodocea nodosa, researchers used GBIF-mediated occurrences to explore how a number of factors may affect the performance of models. First, researchers limited the number of predictors to only half of the available variables, which led to an expected and significant reduction in performance.
To alleviate effects of observed sampling bias, the authors constructed a weighted filter taking into account general marine sampling efforts, national GBIF-publishing activity and level of country development. Surprisingly, however, the filters did not improve the performance of any model.
The authors finally verified the models by comfirming hindcasted predictions of so-called glacial refugia with regions where the highest genetic diversity of C. nodosa is found today.