Mapping long-distance jump dispersal risk by leveraging transport infrastructure data and sampling bias correction: a case study with the invasive spotted lanternfly (Lycorma delicatula)

Abstract

Long-distance jump dispersal accelerates invasive species spread. Analysis of past jumps can identify invasion pathways and point to locations at risk of future colonization, but spatial bias in survey records obscures patterns. To inform surveillance for the spreading invasive grape pest Lycorma delicatula (Hemiptera: Fulgoridae), we gathered 82,890 agency records of established invasive populations from the United States. We used a clustering algorithm to identify 207 jump dispersal sites, and we developed a bias-aware comparison dataset by filtering records to identify proactive search effort at locations not yet known to hold L. delicatula. We measured points' proximity to a co-produced list of 22 property types encompassing categories with high vehicle traffic, railroads, and related contexts that may transport L. delicatula. Bootstrapping analysis showed that proactive surveys were concentrated near hypothesized influential property types, confirming substantial sampling bias. With this bias accounted for, historical jump dispersal sites were closer than expected to 15 property types, particularly distribution centers, truck rentals, flea markets, and importers. We also compared the lengths of road and rail around these points, finding that jump sites were strongly associated with elevated rail density, whereas models showed limited support for roads as jump sites. Finally, we used the high-risk features to project risk of future jumps across the eastern U.S. using ensemble random forests. Predicted high-risk areas were northeast Illinois, southeast Michigan, and central Ohio. This framework is generalizable, allowing various pathways to be assessed and jump dispersal risk to be projected based on spatially biased surveillance data.