Species distribution modelling
The present breeding distribution of L. c. superciliosus was modelled based on presence-only data and seven non-correlated bioclimatic variables using the MaxEnt algorithm (Figure 3). Minimum training presence omission rates for test localities were 1.9% and the model was evaluated to estimate suitable habitat without problematic overfitting. The present prediction of suitable areas (Figure 3a) reflected the taxon’s known breeding range according to Lefranc & Worfolk (1997).
Projection of the model to three climatic models of the LGM (Figure 3b-d) inferred that the coastal area of the currently submerged ECS continental shelf, and an inner continental range corresponding to the present CRECS, were predicted to be suitable breeding habitat forL. c. superciliosus . Suitability for these regions was estimated to be relatively lower than in the Japanese archipelago, within which suitable breeding habitat receded southwards but largely remained during the LGM. Most of the Ryukyu islands were not predicted to be suitable for the species during the LGM, based on our model.