Skip to main content
U.S. flag

An official website of the United States government

Comparison of four modeling tools for the prediction of potential distribution for non-indigenous weeds in the United States

September 19, 2017

This study compares four models for predicting the potential distribution of non-indigenous weed species in the conterminous U.S. The comparison focused on evaluating modeling tools and protocols as currently used for weed risk assessment or for predicting the potential distribution of invasive weeds. We used six weed species (three highly invasive and three less invasive non-indigenous species) that have been established in the U.S. for more than 75 years. The experiment involved providing non-U. S. location data to users familiar with one of the four evaluated techniques, who then developed predictive models that were applied to the United States without knowing the identity of the species or its U.S. distribution. We compared a simple GIS climate matching technique known as Proto3, a simple climate matching tool CLIMEX Match Climates, the correlative model MaxEnt, and a process model known as the Thornley Transport Resistance (TTR) model. Two experienced users ran each modeling tool except TTR, which had one user. Models were trained with global species distribution data excluding any U.S. data, and then were evaluated using the current known U.S. distribution. The influence of weed species identity and modeling tool on prevalence and sensitivity effects was compared using a generalized linear mixed model. Each modeling tool itself had a low statistical significance, while weed species alone accounted for 69.1 and 48.5% of the variance for prevalence and sensitivity, respectively. These results suggest that simple modeling tools might perform as well as complex ones in the case of predicting potential distribution for a weed not yet present in the United States. Considerations of model accuracy should also be balanced with those of reproducibility and ease of use. More important than the choice of modeling tool is the construction of robust protocols and testing both new and experienced users under blind test conditions that approximate operational conditions.

Citation Information

Publication Year 2018
Title Comparison of four modeling tools for the prediction of potential distribution for non-indigenous weeds in the United States
DOI 10.1007/s10530-017-1567-1
Authors Roger Magarey, Leslie Newton, Seung C. Hong, Yu Takeuchi, Dave Christie, Catherine S. Jarnevich, Lisa Kohl, Martin Damus, Steven I. Higgins, Leah Miller, Karen Castro, Amanda M. West, John Hastings, Gericke Cook, John Kartesz, Anthony Koop
Publication Type Article
Publication Subtype Journal Article
Series Title Biological Invasions
Series Number
Index ID 70191912
Record Source USGS Publications Warehouse
USGS Organization Fort Collins Science Center