Urbanization, globalization, and climate change are all important contributing factors to increasing rates of biological invasion. Understanding how these factors influence the introduction and spread dynamics of novel invasive species remains a critical concern for maintaining forest ecosystem health and function. The spotted lanternfly (Lycorma delicatula; SLF) is a plant hopper native to China, India, and Vietnam. The insect was first found in southeastern Pennsylvania in 2014 and has the potential to become a widespread economic pest, with particular threats to agricultural crops including grapes and fruit trees. The SLF has shown a strong host preference for tree of heaven (Ailanthus altissima) for both feeding and egg deposition, which is common throughout the eastern and western United States. In a collaboration among the state of PA, the U.S. Forest Service, and the University of Delaware, we are modeling the spread of SLF based on the state-wide distribution of tree of heaven, railroads, and roads. We are using computer vision and deep learning neural networks to map the tree of heaven, and are building individual-based models that incorporate stage-structured population model estimates of survival and reproduction probabilities of SLF to simulate and predict the spread of SLF over the next 10 years. These models and resulting predicted occupancy maps for SLF can help pinpoint areas of high risk, and improve targeted efforts to slow the spread of SLF throughout Pennsylvania and beyond.
Results/Conclusions
We trained neural network models using over 10,000 images of tree of heaven scraped from the internet, and can identify (with > 95 % accuracy) this species within Google Street View imagery along roads throughout our study area. We developed a neural network model that also identifies tree of heaven from above the canopy, that we trained using drone, aerial, and high-resolution satellite imagery. Preliminary results from individual-based model predicted distributions of SLF that incorporate the estimated tree of heaven distribution will be presented along with methodology for ranking areas based on risk and probability of SLF invasion.