Biological records (citizen science observations of species) provide information about species occurrences over large spatial scales, but the data are spatially patchy with lots of data from some locations and no data from others (spatially biased data). Species distribution models (SDMs) can “fill in” unsampled locations by predicting which species are present, but only if models can make good predictions despite being trained with spatially biased data. We tested the effect of spatial sampling bias, sample size, and the choice of species distribution modeling method on predictions of species distribution models when applied at the spatial scale of Ireland. We present two lines of evidence, a simulation study and a study in which we manipulated biological records of birds in Ireland to introduce spatial biases similar to those found in data for other taxonomic groups. Based on those results, we produced species distribution models for Ireland that fill in knowledge gaps for millipedes (Diplopoda), a taxonomic group for which sampling in Ireland is spatially patchy.
Results/Conclusions
1) Spatial bias at an intensity comparable to that found in average Irish biological records datasets did not hurt predictive performance of SDMs compared to when models were trained with minimally biased data.
2) Increasing the total amount of data is more important than increasing the spatial evenness of data when the goal is to predict species occurrence using species distribution modeling.
3) Species distribution models for selected millipede species showed adequate performance, but sample size may still limit our ability to model many millipede species in Ireland. For the purpose of predictive species distribution modeling, increasing the total number of millipede records is likely more important than increasing the spatial evenness of recording.