Machine learning reveals host, microclimate, and landscape drivers of spotted lanternfly occurrence in Maryland

Abstract

Abstract Spotted lanternfly ( Lycorma delicatula, SLF) is an invasive planthopper with substantial ecological and economic impacts across the United States. In Maryland, SLF threatens urban forests, vineyards, and agricultural systems, prompting an urgent need for spatially targeted surveillance and population reduction strategies. To support these efforts, we developed an explainable machine learning framework using presence-only occurrence records from the Global Biodiversity Information Facility to map SLF hotspots and identify environmental and landscape-level drivers of spread. Our approach enables intervention planning at Census Block Groups (CBGs), with direct relevance to pest management and regulatory decision-making. Our results show that reported occurrences increased sharply in recent years, with recorded hotspots concentrated in Baltimore City, Baltimore County, Anne Arundel County, and Howard County. Across CBGs, Tree-of-heaven ( Ailanthus altissima ), the primary host, was the strongest predictor of presence (mean |SHAP| contribution of 27.9%; Spearman ρ ≈ +0.33), followed by local temperature and specific land-cover fractions. Although statewide occurrence peaks in warm months, CBG-level modeling indicated higher counts in relatively cooler microhabitats (ρ ≈ −0.25 between SLF presence and local temperature), consistent with fine-scale refugia that mitigate heat/moisture stress and align life-stage timing with thermal optima. Land-cover fractions provided additional explanatory signal within the model: Developed Low-Intensity CBGs showed a negative association, Pasture/Hay were positively associated; increases in sumac ( Rhus typhina : secondary host plant) and human population also contributed to presence. Finally, we identified priority CBGs in Maryland for targeted interventions including host-tree removal near transport corridors and along urban–agricultural edges. This interpretable framework supports sustainable regulatory goals by defining measurable nature-based indicators (e.g., host-tree reduction) and is readily transferable to other jurisdictions for data-driven early detection, control, and optimized resource allocation.