Beyond streetlights: predicting first establishment locations of non-native tree pests with generative models

Summary

Non-native tree pests threaten food security, biodiversity, and climate resilience. As border inspections are imperfect, some species become established. Predicting where these species first establish is key to avoid impact. Since many first detections are made by volunteers and accidental, there might be a bias in the reports toward areas where people look. The thesis introduces a generative modeling approach to correct for this bias, revealing that while cities and warmer climates are key factors, the importance of cities is often overestimated. The study also finds that pests from warmer regions tend to establish in Europe's warmer areas, likely driven by trade, host availability and climate suitability. These findings help refine surveillance strategies and improve early detection. The research emphasizes the need to account for sampling bias and offers a new approach to modeling presence-only data commonly used in species distribution modelling.