Canadian Forest Service Publications
Potential distribution of emerald ash borer: what can we learn from ecological niche models using Maxent and GARP? 2012. Sobek-Swant, S.; Kuza, D.A.; Cuddington,K.; Lyons, D.B. Forest Ecology and Management. 281:23-31.
Available from: Great Lakes Forestry Centre
Catalog ID: 33964
Available from the Journal's Web site. †
† This site may require a fee.
We use two ecological niche modeling methods, Maxent and GARP, to model the potential distribution of emerald ash borer (Agrilus planipennis, EAB) in its invaded (Canada, USA) and native range (eastern Asia). For each algorithm (Maxent and GARP), we constructed three different models based on native or invaded data or a combination thereof. All GARP models yielded higher area under the curve (AUC) values and had therefore higher discriminatory power than the corresponding Maxent models, and the area predicted as suitable by GARP in North America was generally larger and included most infested sites even at the known range edges. In Asia, habitat suitability predicted by both algorithms was low and models trained with invaded coordinates did not transfer well to the native range. We found that none of the Maxent models provided a prediction precise enough for reliable risk assessment and the development of management plans, but a GARP model trained in the native range performed well when validated with data from the invaded range. Based on this GARP model, EAB may be able to extend its North American range further south, north and west covering roughly half (49%) of the natural range of the most common affected ash species (Fraxinus americana, F. nigra, F. quadrangulata, and F. pennsylvanica). While our results demonstrate that native data may be useful for risk assessment of invasive species, a validation of early predictions based on these data is only possible with a time lag, which is lacking for most species. Due to uncertainties associated with ecological niche models for invaders, native data may not be sufficient at all times for long-term risk assessment. With regard to the latter, we recommend frequent re-evaluation of models based on more current monitoring. Combining ecological niche models with more mechanistic approaches based on experimental data may reduce uncertainty and improve risk assessment.