Evaluating species distribution model predictions through time against paleozoological records

Area of Science:

• Paleoecology
• Ecological Modeling
• Climate Change Biology

Background:

• Species distribution models (SDMs) are crucial for predicting species' responses to climate change.
• Current SDM evaluation primarily focuses on model fit to present-day distributions.
• Long-term performance assessment of SDMs under environmental change remains under-explored.

Purpose of the Study:

• To independently assess the predictive accuracy of SDMs through time using paleozoological records.
• To compare conventional model evaluation metrics with novel 'paleometrics' for historical distribution hindcasting.
• To investigate the influence of input data and modeling algorithms on SDM performance.

Main Methods:

• Utilized Holocene presence/absence records of the hartebeest (Alcelaphus buselaphus).
• Applied various SDM algorithms to occurrence datasets (point data, current range maps, 'natural' range maps).
• Compared conventional metrics (TSS_c, AUC_c) with paleometrics (TSS_p, AUC_p, Boyce_p, F2-score_p, Sorensen_p).

Main Results:

• Found a weak correlation between conventional metrics and paleometrics.
• Identified that models excelling at current distributions may not reliably hindcast historical ones.
• Demonstrated that input data choice and algorithm significantly impact SDM predictions and performance.

Conclusions:

• Advocate for paleometric assessment of SDM performance, especially metrics focusing on presence prediction (F2-score_p, Sorensen_p).
• Emphasize the potential unreliability of absence data in paleozoological records.
• Recommend integrating paleo-records to improve understanding of environmental constraints on species distributions over time.