Premium
Evaluation of the R package ‘ resistancega ’: A promising approach towards the accurate optimization of landscape resistance surfaces
Author(s) -
Winiarski Kristopher Jonathan,
Peterman William E.,
McGarigal Kevin
Publication year - 2020
Publication title -
molecular ecology resources
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 2.96
H-Index - 136
eISSN - 1755-0998
pISSN - 1755-098X
DOI - 10.1111/1755-0998.13217
Subject(s) - univariate , resistance (ecology) , sample size determination , biology , variance (accounting) , multivariate statistics , statistics , resistance distance , sample (material) , autocorrelation , biological system , selection (genetic algorithm) , fitness landscape , mathematics , ecology , computer science , machine learning , physics , graph , accounting , line graph , graph power , business , population , demography , sociology , thermodynamics , discrete mathematics
Understanding how landscape features affect gene flow has implications for numerous fields including molecular and evolutionary ecology. Despite this, modelling landscape resistance surfaces has remained a significant challenge. The R package resistancega was developed to provide a framework for optimizing landscape resistance surfaces. In this study, we assessed ResistanceGA's ability to recover the true resistance surface under a variety of scenarios, including when the underlying surface: (a) had different levels of spatial autocorrelation and (b) was transformed into a resistance surface using different functional transformations. These scenarios were evaluated with regard to varying sample size and varying levels of variance in the measure of genetic distance. We also assessed the ability of ResistanceGA to identify the true resistance surface among alternative correlated surfaces. In univariate simulations, correlation between the true and optimized resistance surfaces remained high with increased variance in genetic distance, but only when sample size was moderate to high (≥50). Model selection error was also driven by sample size with low type I error when simulations had moderate to high sample sizes, even with moderate to high variance in genetic distance and correlated alternative surfaces. ResistanceGA also performed well in multivariate simulationsbut had more difficulty identifying the true data generating surfaces when genetic data were simulated using an agent‐based approach (especially with individual‐based genetic data). Overall, our simulations highlight the ability of ResistanceGA to accurately optimize resistance surfaces but also underscore challenges in optimizing landscape resistance surfaces, especially with highly stochastic individual‐based data.