Modelling can predict where invasive species will appear, and where needs safeguarding. A new study shows where the data comes from matters. After just 100 invasive plant observations in the US, models using only US data better predict its spread than those incorporating global range information.
Nicholas Young and colleagues examined how models can use data from a plant’s range to predict where it is most likely to invade in the USA. They found that the best models vary depending on the invasion stage. The best models for the earliest stage, under 50 observations, used US and global data.
Yet when there were more than 100 observations in the USA, something odd happened. The traditional advice to build distribution models using a plant’s entire global range (native + invaded) actually performed worse in most cases. The authors suggest that data quality may be an issue.
Global predictors generally are coarser (grain or cell size), less accurate locally, and have a limited quantity available while US predictors may be available at finer resolutions, are often more accurate, and may have a greater variety and diversity of predictors available.
Another factor can be niche shifts. When plants invade new regions, they can often find new opportunities. In the case of Japanese stiltgrass, Young & colleagues found almost no overlap between the US and global niches. These differences can make local observations more valuable than global data.
The team applied their method to 13 different species from Tree of Heaven, invading in 1841, to Air Potato, invading in 1965. They found their 100 observation tipping point worked across vines, forbs, shrubs, trees and grasses, suggesting it works as a general approach to modelling.
Our findings show that once an invader accumulates 100 occurrences after spatially filtering to at least 5 km in the invaded US range, models developed using occurrences from the global range are significantly worse than other model strategies and that models should be developed using invaded range occurrences from this point forward in the invasion.
The study provides guidance for workers with early invasion detection and rapid response efforts. Improving modelling will help develop a a more accurate and efficient approach to identifying high-risk areas, enabling better allocation of limited monitoring and control resources.
Young, N. E., Williams, D. A., Shadwell, K. S., Pearse, I. S., & Jarnevich, C. S. (2025). How to model a new invader? US-invaded range models outperform global or combined range models after 100 occurrences. Ecological Applications, 35(2), e70010. https://doi.org/pbqd
Cross-posted to Bluesky & Mastodon.
Image: Lysimachia nummularia / Canva.
Related