Improved prediction of Canada lynx distribution through regional model transferability and data efficiency

Area of Science:

• Ecology
• Conservation Biology
• Spatial Modeling

Background:

• Species distribution models (SDMs) are crucial for conservation but face challenges in transferability across geographic regions.
• Understanding regional variations in animal-environment relationships is key to improving SDM transferability.
• Canada lynx (Lynx canadensis) habitat modeling in the northwestern US requires robust and transferable SDMs.

Purpose of the Study:

• To assess how regional differences in Canada lynx-environment relationships affect SDM transferability.
• To evaluate the impact of GPS data sample size on SDM performance and transferability.
• To generate reliable spatial predictions of Canada lynx distribution in the northwestern US.

Main Methods:

• Used Canada lynx GPS data from Washington, Montana, and Wyoming populations (1996-2015).
• Assessed regional variation using Principal Components Analysis (PCA).
• Developed ensemble SDMs for individual populations and a combined model, testing transferability with independent data and varying GPS sample sizes.

Main Results:

• Regional variations in lynx-environment relationships were observed via PCA, influencing individual population model transferability.
• A single SDM incorporating data from all populations performed as well as or better than single-population models.
• SDM performance plateaued around 30% of the GPS data, indicating sample size redundancy.

Conclusions:

• Combining data from multiple populations enhances SDM transferability and performance for Canada lynx.
• A reduced GPS data sample size (around 30%) is sufficient for robust SDM predictions.
• Precipitation and temperature are key predictors for Canada lynx distribution models in the southern part of their range.