Estimating resistance surfaces using gradient forest and allelic frequencies.

Area of Science:

• Ecology
• Genetics
• Conservation Biology
• Machine Learning

Background:

• Landscape connectivity is crucial for biodiversity conservation, especially with increasing fragmentation.
• Traditional methods often rely on genetic distance and landscape distance, with inherent statistical assumptions.
• Existing models face limitations in handling complex environmental interactions and data.

Purpose of the Study:

• To introduce and evaluate resistance Gradient Forest (resGF), a novel machine learning approach for generating resistance surfaces.
• To compare resGF performance against established methods using genetic simulations and real-world datasets.
• To demonstrate resGF's potential for improving landscape connectivity analysis and conservation planning.

Main Methods:

• Adapted the gradient forest algorithm, an extension of random forest, to create resistance surfaces (resGF).
• Compared resGF with maximum likelihood population effects (MLPE), random forest-based least-cost transect analysis, and species distribution models.
• Utilized genetic simulations and two published datasets for performance evaluation.

Main Results:

• resGF accurately identified the true surface influencing genetic diversity in univariate scenarios.
• In multivariate scenarios, resGF performed comparably to other random forest methods and outperformed MLPE-based approaches.
• The method effectively handles multiple environmental predictors without traditional linear model assumptions.

Conclusions:

• resGF offers a robust alternative to conventional methods for modeling landscape connectivity.
• This machine learning approach enhances the understanding of genetic diversity patterns and landscape influences.
• resGF has significant potential to inform and improve long-term biodiversity conservation strategies.