Premise

Researchers often use ecological niche models to predict where species might establish and persist under future or novel climate conditions. However, these predictive methods assume species have stable niches across time and space. Furthermore, ignoring the time of occurrence data can obscure important information about species reproduction and ultimately fitness. Here, we assess and compare ecological niche models generated from full-year averages to seasonal models 

Methods

In this study, we generate full-year and monthly ecological niche models for Capsella bursa-pastoris in Europe and North America to see if we can detect changes in the seasonal niche of the species after long-distance dispersal. 

Key Results

We find full-year ecological niche models have low transferability across continents and there are continental differences in the climate conditions that influence the distribution of C. bursa-pastoris. Monthly models have greater predictive accuracy than full-year models in cooler seasons but no monthly models are able to predict North American summer occurrences very well.

Conclusions

The relative predictive ability of European monthly models compared to North American monthly models suggests a change in the seasonal timing between the native range to the non-native range. These results highlight the utility of ecological niche models at finer temporal scales in predicting species distributions and unmasking subtle patterns of evolution.

Here, we present the mean Partial Area Under the Curve (pAUC) value for comparisons of European monthly models on North American occurrences. We evaluated the performance of models using a pAUC analysis to focus on the most informative metrics of predictive ability for our study. We chose an admissible omission error rate of 0.15 for each model and 300 bootstrapped iterations. We evaluated the European models’ predictive ability on North American occurrences of C. bursa-pastoris to test for a seasonal niche shift. Each month of North American occurrences was evaluated against all twelve models resulting in 144 evaluations of the European model's predictive ability on North American occurrence records.