A screening system to predict wildfire risk of invasive plants

Abstract

Globally, invasive plant-fueled wildfires have tremendous environmental, economical, and societal impacts, and the frequencies of wildfires and plant invasions are on an upward trend globally. Identifying which plant species tend to increase the frequency or severity of wildfire is important to help manage their impacts. We developed a screening system to identify introduced plant species that are likely to increase wildfire risk, using the Hawaiian Islands to test the system and illustrate how the system can be applied to inform management decisions. Expert-based fire risk scores derived from field experiences with 49 invasive species in Hawai′i were used to train a machine learning model that predicts expert fire risk scores from among 21 plant traits obtained from literature and databases. The model revealed that just four variables can identify species categorized as higher fire risk by experts with 90% accuracy, while low risk species were identified with 79% accuracy. We then used the predictive model to screen > 140 recently naturalized plants in Hawai′i to illustrate how the screening tool can be applied. The screening tool identified a managebly small set of species (6% of naturalizations in the last ~ 10 years) that are likely to pose a high fire risk and can be targeted for eradication or containment to reduce future wildfire risks. Because the screening system uses general plant traits that are likely relevant to fire risk in drylands around the world, it can likely be applied with minimal modification to other regions where invasive plants pose potential fire risks.