Population distribution models: species distributions are better modeled using biologically relevant data partitions

Area of Science:

• Ecology
• Biogeography
• Computational Biology

Background:

• Predicting species distributions using models often results in high omission errors, especially for widespread species.
• Pooling subspecies or races in analyses can mask important spatial variations within a species' distribution.
• The oldfield mouse (Peromyscus polionotus) serves as a model for studying widespread species distribution challenges.

Purpose of the Study:

• To compare the performance of different maximum entropy models for predicting the geographic distribution of the oldfield mouse.
• To evaluate whether subdividing data by subspecies or geographic quadrants improves model accuracy compared to a single, pooled model.
• To identify optimal strategies for modeling the distribution of widespread species.

Main Methods:

• Developed a presence-only maximum entropy model using all available presence locations for Peromyscus polionotus.
• Created two composite maximum entropy models: one dividing the distribution into four geographic quadrants, and another by fifteen subspecies.
• Contrasted model performance using Area Under the ROC Curve (AUC) and omission curves.

Main Results:

• All models achieved high Area Under the ROC Curve (AUC) values, indicating good overall performance.
• The composite model using fifteen subspecies demonstrated a significantly better representation of the known distribution compared to quadrant-based or whole-species models.
• Geographic quadrant models and the whole-species model performed less effectively than the subspecies-based composite model.

Conclusions:

• Area Under the ROC Curve (AUC) values alone are insufficient for evaluating differences in model predictability; omission curves should also be used.
• Partitioning widespread species data into biologically relevant units, such as subspecies, substantially enhances distribution model performance.
• This approach offers a practical and informative method for improving species distribution modeling across various applications.