Frontiers in Ecology and Evolution (Nov 2022)

Modeling the distribution of invasive species (Ambrosia spp.) using regression kriging and Maxent

  • Ki Hwan Cho,
  • Jeong-Soo Park,
  • Ji Hyung Kim,
  • Yong Sung Kwon,
  • Do-Hun Lee

DOI
https://doi.org/10.3389/fevo.2022.1036816
Journal volume & issue
Vol. 10

Abstract

Read online

Invasion by non-native species due to human activities is a major threat to biodiversity. The niche hypothesis for invasive species that rapidly disperse and disturb ecosystems is easily discarded owing to eradication activities or unsaturated dispersal. Here, we used spatial and non-spatial models to model the distribution of two invasive plant species (Ambrosia artemisiifolia and Ambrosia trifida), which are widely distributed, but are also being actively eradicated. Regression kriging (RK) and Maxent were used to predict the spatial distribution of the two plant species having eradication targets for decades in South Korea. In total, 1,478 presence/absence data points in the Seoul metropolitan area (∼11,000 km2 in northeastern South Korea) were used. For regression kriging, the presence/absence data were first fitted with environmental covariates using a generalized linear model (GLM), and then the residuals of the GLM were modeled using ordinary kriging. The residuals of GLM showed significant spatial autocorrelation. The spatial autocorrelation was modeled using kriging. Regression kriging, which considers the spatial structure of data, yielded area under the receiver operating curve values of 0.785 and 0.775 for A. artemisiifolia and A. trifida, respectively; however, the values of Maxent, a non-spatial model, were 0.619 and 0.622, respectively. Thus, regression kriging was advantageous as it considers the spatial autocorrelation of the data. However, species distribution modeling encounters difficulties when the current species distribution does not reflect optimal habitat conditions (the niche habitat preferences) or when colonization is disturbed by artificial interference (e.g., removal activity). This greatly reduces the predictive power of the model if the model is based solely on the niche hypotheses that do not reflect reality. Managers can take advantage of regression modeling when modeling species distributions under conditions unfavorable to the niche hypothesis.

Keywords