Predicting non-native seaweeds global distributions: The importance of tuning individual algorithms in ensembles to obtain biologically meaningful results

Modelling non-native marine species distributions is still a challenging activity. This study aims to predict the global distribution of five widespread introduced seaweed by focusing on two mains aspects ensemble modeling process: (1) Does enforcement less complex models (in terms number predictors) help in obtaining better predictions? (2) What are implications tuning configuration individual algorithms ecological realism? Regarding first aspect, datasets with different predictors were created. second four and three configurations tested. Models evaluated using common evaluation metrics (AUC, TSS, Boyce index TSS-derived sensitivity) realism. Finally, stepwise procedure for model selection was applied build ensembles. trained large predictor dataset generally performed than reduced dataset, but some exceptions. configurations, Random Forest (RF) Generalized Boosting (GBM) scored highest metric values average, even though, RF response curves most unrealistic non-smooth GBM showed overfitting species. Linear (GLM) MAXENT, despite their lower scores, fitted smoother (especially at intermediate complexity levels). Reliable biologically meaningful predictions achieved. Inspecting include final ensembles its have been demonstrated be crucial this purpose. Additionally, we highlight importance combining quantitative (based multiple metrics) qualitative realism) methods selecting optimal configurations.