AI Predicts Habitats for Wide-Ranging Carnivores in Diverse Regions

Jim Crocker
16th May, 2024

AI Predicts Habitats for Wide-Ranging Carnivores in Diverse Regions

Random forest derived resource selection function depicting predicted continent-wide mountain lion habitat suitability in North America from data collected from 476 mountain lions between 2002–2020

Image adapted from: O’Malley et al. / CC BY (Source)

Key Findings

  • The study from the University of Montana focused on predicting mountain lion habitat selection across North America
  • Random forest models outperformed traditional logistic regression models in predicting mountain lion habitat use
  • Random forest models revealed complex, nonlinear relationships in habitat selection that logistic regression models missed
Resource selection functions (RSFs) are essential tools for ecologists and wildlife managers to understand how animals choose their habitats. These functions use statistical models to identify which environmental features are selected or avoided by animals. Traditionally, RSFs have been performed using logistic regression, a type of generalized linear model. However, logistic regression has limitations, particularly in handling complex interactions and nonlinear relationships. Recently, researchers at the University of Montana have explored the use of machine-learning methods, specifically random forest (RF), to improve the predictive performance of RSFs[1]. The study from the University of Montana investigates the seasonal resource selection patterns of mule deer (Odocoileus hemionus) using both logistic regression and random forest models. Their findings show that RF models can detect nonlinear relationships and complex interactions that are challenging to capture with standard logistic regression models. This improved predictive skill provided new insights into the habitat preferences of mule deer, demonstrating significant differences in predicted habitat suitability when RF models were applied. This study builds on earlier research that highlights the importance of accurately identifying critical resources for animal populations. For example, one study on woodland caribou emphasized the need to consider both resource selection and predation risk when evaluating habitat quality[2]. The researchers found that while resource selection positively correlated with survival, incorporating predation risk greatly improved the model's accuracy. This indicates that a comprehensive approach, considering multiple factors, is crucial for understanding habitat selection and its impact on animal survival[2]. Another relevant study reviewed the use of random-effects models in resource selection to account for individual variation and unbalanced sample designs[3]. By including random effects, researchers can better interpret the data and improve model fit, addressing common limitations in traditional RSF models. This approach allows for more accurate estimation of both population-level and individual-level responses, which is essential for making informed conservation decisions[3]. The University of Montana study also aligns with previous research advocating for the use of machine-learning methods in ecological studies. For instance, a study on mule deer demonstrated that RF models could uncover complex interactions and nonlinear relationships that logistic regression might miss[4]. This is particularly important for predicting habitat suitability across different geographic areas, as RF models can provide more reliable and nuanced predictions. In summary, the University of Montana's research underscores the advantages of using machine-learning methods like random forest for predicting habitat selection. By addressing the limitations of traditional logistic regression models, RF models offer improved predictive performance and deeper insights into animal habitat preferences. This study, along with earlier research on resource selection and model selection techniques, highlights the evolving landscape of ecological modeling and the importance of adopting advanced methods to enhance our understanding of wildlife habitat selection.

EnvironmentWildlifeEcology

References

Main Study

1) Machine learning allows for large-scale habitat prediction of a wide-ranging carnivore across diverse ecoregions

Published 15th May, 2024

https://doi.org/10.1007/s10980-024-01903-2

Related Studies

2) Linking habitat selection and predation risk to spatial variation in survival.

https://doi.org/10.1111/1365-2656.12144

3) Application of random effects to the study of resource selection by animals.

Journal: The Journal of animal ecology, Issue: Vol 75, Issue 4, Jul 2006

4) A machine-learning approach for extending classical wildlife resource selection analyses.

https://doi.org/10.1002/ece3.3936


Related Articles

An unhandled error has occurred. Reload 🗙