Spotted lanternfly (SLF), Lycorma delicatula (White) (Hemiptera: Fulgoridae), is a non-native planthopper that recently established in the Northeastern United States. Little is known about the spatial dynamics of its invasion and key drivers associated with its regional spread. Here, using field survey data from a total of 241,366 survey locations from 2014–2019 in the eastern USA, we quantified rates of SLF spread and modeled factors associated with the risk of SLF invasion. During the study period, SLF invasion appears to be associated with both short- and long-distance dispersal. On average, the number of newly invaded counties per year increased since initial discovery, with 0–14 long-distance dispersal events per year and median jump distances ranging from 55 to 92 km/year throughout the study period. Radial rates of spread, based on two of the three analysis methods applied, varied from 38.6 to 46.2 km/year. A Cox proportional hazards model suggested that risk of SLF invasion increased with a proxy for human-aided dispersal, human population per county. We anticipate that SLF will continue to spread via both long- and short-distance dispersals, especially via human activities. Efforts to manage SLF populations potentially could target human-mediated movement of SLF to reduce rates of spread.

Biological invasion, Cox proportional hazards, spatiotemporal, invasive species, radial spread

Though most non-native pests fail to establish after arrival, those that successfully found reproducing populations can subsequently spread via a coupling of population growth with dispersal. The dispersal of many invading species is characterized by the simultaneous occurrence of local diffusion and occasional long-distance dispersal (Hastings et. al. 2005; Liebhold and Tobin 2008). Information on what factors drive spread of a non-native pest can guide management to contain its populations and reduce their impacts to ecosystems and economic costs (Sharov and Liebhold 1998; Liebhold and Kean 2019). Understanding the factors that drive spread is particularly important for newly established species, for which dispersal behaviors and population growth characteristics are often unknown.

Spotted lanternfly, Lycorma delicatula (White) (Hemiptera: Fulgoridae), is a non-native planthopper recently established in the United States. The species is native to southeast Asia, but recently invaded the USA in Berks County, Pennsylvania in 2014 (Barringer et al. 2015). Spotted lanternfly (SLF) is univoltine and lays egg masses on a variety of surfaces, including tree bark, stone, motor vehicles, and trains (Urban 2019). In addition to indiscriminate egg deposition, SLF also has a wide breadth of host use. This pest feeds on over 70 species of herbaceous and woody plants belonging to over 20 families, though it prefers tree of heaven (Ailanthus altissima), especially as a late instar (Dara et al. 2015; Parra et al. 2017). Notably, SLF feeds on apple (Malus spp.) and grape (Vitis spp.), both important agricultural plants in the Northeastern USA. Feeding on grape has reportedly resulted in lower fruit quality, less fruit production, and elevated mortality, though minimal impacts to fruit tree health have been reported (Urban 2019). The most conspicuous impact of SLF in forests is the accumulation of honeydew in the understory, which results in sooty mold growth that limits photosynthesis and growth of understory plants (Ding et al. 2006; Parra et al. 2017). There is also evidence that aggregation of SLF can cause weeping wounds on trees, resulting in crown dieback (Dara et al. 2015). While detrimental impacts on tree of heaven might be beneficial due to its status as an invasive plant, SLF is considered a serious pest due to its negative impacts on agricultural crops and native trees.

Despite regulations by the state of Pennsylvania that prohibit movement of any SLF living stage (e.g. egg masses, nymphs, adults) or material potentially harboring the pest (e.g. firewood, nursery stock, etc.) outside of a quarantine area, SLF has spread from Pennsylvania to seven surrounding states as of 2019 (Fig. 1). Because SLF was detected in the USA recently, there is little information on how this species spreads or what drives its invasion. Though the body of knowledge on this insect is growing, many aspects of SLF spread, especially the role of environmental drivers, are unknown. Elucidating how this pest spreads can inform future management and survey efforts.

Figure 1.

County-level distributions of spotted lanternfly (SLF) in the eastern USA. Distribution of SLF detections and establishments by year based on USDA Animal and Plant Health Inspection Service and Pennsylvania Department of Agriculture visual survey data. Counties with hash marks had SLF detections that failed to establish. We define a county as invaded when the county experiences at least two consecutive years of SLF detection, and define year invaded as the first of those two consecutive years. Counties with white color were not surveyed.

The ranges of introduced species are influenced by a multitude of anthropogenic factors and habitat features. For SLF, climatic niche models indicate that half of the USA, including most of the New England, Mid-Atlantic, and Pacific Coast states, is at risk of invasion (Wakie et al. 2020). While these climatic niche models provide valuable information on where SLF can potentially establish, analyses of spread can provide insight into how quickly SLF will arrive and what habitat and/or anthropogenic factors affect the dynamics of SLF spread. SLF has undergone several long-distance dispersal events that likely resulted from human-mediated transportation (Eddy 2018; Scheid 2020). Tree of heaven is also more abundant in urban areas, and thus human activities may increase both propagule pressure and habitat suitability. However, the rate of spread, including the frequency and distance of long-distance dispersal events, and drivers of spread have not been quantified. Therefore, we investigated how anthropogenic and habitat factors are related to SLF spread.

We analyzed the known geographical distribution of SLF (2014–2020) in the USA to quantify its rates of spread and identify factors that influence its invasion risk. Our goals were to: 1) describe the patterns of SLF spread following the initial detection in 2014, and 2) identify key drivers that are associated with SLF spread. For our second goal, we used known occurrences of SLF in conjunction with habitat and anthropogenic variables to determine the most important factors driving county-level invasion risk across the study area, defined below. We hypothesized that anthropogenic factors are important drivers of SLF spread, given the ability of this insect to lay inconspicuous eggs on a variety of materials, including motor vehicles and trains (Urban 2019).

The SLF distribution data analyzed in this study were derived from visual surveys conducted from 2014–2019 by the US Department of Agriculture, Animal and Plant Health Inspection Service (APHIS) and the Pennsylvania Department of Agriculture (PDA). We also used SLF county-level presence data for 2020 from the New York State Integrated Pest Management Program (Cornell 2021). Survey data include geospatial coordinates for survey locations as well as the number of SLF observed (if any). A total of 241,366 survey locations were obtained for this study (Suppl. material 1). Given the irregularity of survey locations and potential biases (e.g. surveys at expected SLF locations) and to render data at an equivalent scale as the 2020 presence data, we converted counts to county-level presence/absence records and used county as our unit of analysis.

The survey data contained many points that we identified as failed establishments in which SLF were observed in a county in a given year but were absent in surveys of the same county in subsequent years. These detections were likely either populations that failed to establish or regulatory incidents, such as dead SLF adults found in transported materials, and thus we did not treat them as invasions. Hereafter, we refer to detections as establishments plus failed establishments and establishments as only populations that persisted for more than one survey year in consecutive years within a county. Moreover, we categorized each invaded county in year n as contiguous or non-noncontiguous based on the presence or absence, respectively, of an invaded neighboring county in year n-1.

Described below are methods we used to 1) determine aspects of spread dynamics, such as jump distances and spread events into contiguous vs. non-contiguous counties, 2) compare three methods of estimating spread rates, and 3) fit a Cox proportional hazards model estimating time-to-invasion as a function of variables representing spatial proximity to existing SLF populations (henceforth referred to as spatial proximity), habitat suitability, and anthropogenic influences. Our study area was defined as the area of the eastern USA invaded in 2019 plus a buffer distance of 355 km, equal to the maximum observed jump distance (see “Characterization of spread events” in Methods). This study area was used for all subsequent analyses. Counties, which are the level at which quarantines and other management decisions are set, served as the unit of analysis for all analyses. All analyses were conducted using R version 4.0.2 (R Core Team 2020).

Spread of invasive species is often characterized by both short- and long-distance dispersal. In many systems, short-distance dispersal is caused by the natural movement of organisms (e.g. flight behavior) while long-distance dispersal is caused by accidental human movement (Hastings et al. 2005). Even small amounts of long-distance dispersal can result in greatly elevated rates of spread (Shigesada et al. 1995). So far in the SLF invasion, movement appears to consist of both short- and long-distance dispersal. Little is known about natural dispersal in this species. Our results suggest, however, that risk of long-distance movement increases with human population density, likely reflecting the propensity of SLF to become associated with objects transported by humans, such as when SLFs oviposit onto train cars and motor vehicles (Urban 2019).

A higher number of new establishments occurred in contiguous than in non-contiguous counties, but several long-distance jumps were observed and the frequency of jumps appears to be increasing (Table 1). Human-mediated long-distance dispersal events are responsible for spread outside of the center of invasion, allowing for invasion of a larger geographic area than would be possible via insect movement alone. For example, the established population in northern Virginia (Frederick County) is believed to have originated from shipments from a stone yard in Pennsylvania (Eddy 2018). As SLF spreads, there may be increases in both long- and short-distance movement due to increases in numbers of source populations or increases in population size. Indeed, there have been additional long-distance dispersals beyond our study period (2014–2019), such as the discovery of SLF in Switzerland County, IN in July 2021 and the identification of SLF in a bug collection at the Kansas State Fair in early September 2021 (Edwards 2021; Indiana DNR 2021). A Mann Whitney U test showed no significant difference between jump distance distributions in detected but non-established vs. established populations, indicating that jump dispersal events are not necessarily more likely to persist if they are closer to the point of establishment (i.e. have closer spatial proximity). Ranges of jump distances were visually similar in range for both detected and established populations (Fig. 3), signifying that established jumps went at least as far as jumps that failed to establish. Shigesada et al. (1995) demonstrated that such long-distance dispersal events typically result in faster rates of spread as well as accelerating patterns of radial spread. SLF spread rates could increase in this way, and we observed the largest increases in radial spread in the last two years of the study period (2018 and 2019), potentially indicating accelerating spread.

Our estimates of spread rate varied between methods, with the effective range radius method estimating the highest spread rate. The large differences observed between these methods may reflect the discontinuous nature of SLF spread. Measurement of the radial rate of spread of invading organisms was originally envisioned for continuous range expansion (e.g. Skellam 1951) and may not fully capture discontinuous spread such as observed here, which is also reflected in the low variance explained by distance regression spread estimation (r² = 0.098) (Fig. 4B). The effective range radius approach may provide a more representative measure of spread in this situation as it accounts for both the frequent long-distance dispersals and subsequent spread into the counties between contiguous and non-contiguous counties. For example, a long-distance dispersal event established a SLF population in northern Virginia in 2018, and in 2019, SLF spread to several counties between the eastern Pennsylvania invasion area and the new area in northern Virginia (Fig. 1). The effective range radius approach accounts for the cumulative invaded area as these counties are occupied in subsequent years, whereas boundary displacement does not include those counties in estimates of radial spread. That is, counties closer to the previously invaded area following a long-distance jump are enclosed by the convex hull polygon and do not influence future boundary displacements as they become invaded. However, the effective range radius approach alone may overestimate spread as entire county areas are summed as invaded, while only portions of each county are actually invaded.

Therefore, based on the findings presented here, we estimate the radial spread rate at around 40 km/year based on the average of the two more reliable methods (i.e. effective range radius and boundary displacement). If SLF were allowed to spread without any intervention, spread might be much higher given considerable management efforts are currently targeted to suppress SLF populations and limit movement. For example, active management programs conducted by USDA APHIS include egg scraping, sanitation (i.e. host tree removal) around SLF detections, and insecticide application to tree of heaven designated as trap trees (USDA-APHIS 2018). Insecticide applications were used primarily to kill SLF landing on trap trees and, in later applications, to determine efficacy of insecticides for use on fruit and residential trees (Urban, Calvin, and Hills-Stevenson 2021). Additionally, the State of Pennsylvania’s quarantine on movement of goods out of the invaded area is implemented to limit spread of SLF. It is also important to note SLF is in the early stages of invasion, and the spread rate may increase as this pest continues to colonize new locations in the USA.

Results of the Cox proportional hazards model suggested that anthropogenic factors, specifically human density, are stronger drivers of SLF spread than forested area or availability of host trees. The role of humans in facilitating spread of invading organisms is a common phenomenon. Known international and domestic pathways of human-mediated spread of tree pests include transportation of pests on live plants (Liebhold 2012) and wood products (e.g. packing materials or movement of firewood) (Yemshanov et al. 2012), though pests can also be transported on non-host materials, such as on stone imports as with SLF. Domestic pathways of human-mediated spread include movement of firewood, transportation via vehicles (e.g. trains, motor vehicles), and “hitchhiking” on travel gear (e.g. hiking gear) and/or pets. Given that SLF lays eggs indiscriminately, human-mediated spread is not limited to host materials. Humans could facilitate the spread of this pest via travel (e.g. automobiles, trains) and movement of both host and non-host materials from an invaded area. Gilbert et al. (2004) came to similar conclusions in their analyses of the horse chestnut leafminer Cameraria ohridella Deschka & Dimic (Lepidoptera, Gracillariidae), finding that geographical variation in human population density explained most of the variation in historical spread of this species. Similarly, in an analysis of 79 damaging forest pests, Liebhold et al. (2013) found human population density associated with both spatial proximity and number of invasive forest pests per county across the USA. However, with all such analyses of historical spread, there is always some possibility that statistical associations may be caused in part by more intensive surveying and reporting in more populated areas.

The invasion of tree of heaven in the eastern USA more than 200 years prior to the arrival of SLF may have facilitated the insect’s initial establishment, causing an “invasional meltdown” (Feret 1985; Simberloff and Von Holle 1999) in which invasion of one species facilitates the invasion of another. Tree of heaven is the preferred host for SLF and SLF fitness (survival and fecundity) is maximized when feeding on tree of heaven, but this pest can survive and reproduce without access to tree of heaven (Uyi et al. 2020). In addition, SLF’s ability to feed on a wide breadth of plant species (more than 70 species) increases the likelihood of the insect encountering a suitable host following dispersal (Dara et al. 2015). The final Cox proportional hazards model did not include a significant effect of tree of heaven abundance on SLF spread, and therefore we found no evidence that this tree species has influenced SLF spread. Surveys for tree of heaven were conducted by many different people including volunteers and residents and thus the resulting data are likely not reflective of true distribution of tree of heaven, despite verification by EDDMapS reviewers. However, given the association of tree of heaven with urban and disturbed environments, it is probable that more accurate tree of heaven distribution data may correlate highly with human population, which emerged as a significant predictor of SLF spread. In addition, as SLF invasion progresses, additional relationships to host trees or other environmental variables may become apparent or the importance of such variables may vary geographically.

Spatial proximity will remain an important predictor in the future spread of this pest, rendering estimation of SLF populations an important step in assessing spread. Current challenges in estimating SLF populations are primarily lack of long-term, systematic population assessment data and difficulties detecting small populations. The SLF-specific dispersal kernel we estimated here provided the best estimates of spatial proximity based on available distribution data but it was limited by the coarse spatial scale of county-level data and the limited temporal replication. We anticipate that as more data are collected on SLF populations, the estimated dispersal kernel could be refined and thus enhance model predictions.

There are a few limitations involved in our study. First, the data used in these analyses consisted of visual surveys that were located based on perceived risk of SLF establishment. These data were not collected in a systematic fashion, and thus there is potential for sampling bias and imperfect detection, e.g. overlooking of individuals. Though work is underway on developing traps to efficiently survey for SLF (Francese et al. 2020), a sensitive SLF-specific trapping system has not yet been widely implemented. The lack of a pest-specific trap increases risk of missed detections in visual assessments, especially for low population densities. Missed detections occur in many invasive species surveillance programs and can lead to underestimation of the extent of species ranges as well as biased estimates of invasion speeds (Guillera-Arroita 2016; Mang et al. 2016). Given these potential biases, we used counties as the unit of analysis, and the estimates of spread rate as well as drivers of local spread at a finer resolution may be different. We also assumed counties with only a single year of SLF detection indicated populations that failed to establish and thus were not detected in future surveys. Failure to establish could be the result of stochastic dynamics or Allee effects, both of which can drive low-density, newly invaded populations to extinction (Liebhold and Tobin 2008). For example, Liebhold and Bascompte (2003) concluded that low density gypsy moth Lymantria dispar (L.) populations are likely to reach extinction without intervention, and in their analysis, most of the populations that did go extinct without treatment did so within a year of detection. Where management efforts are in place, failure to establish could also be the result of local eradication efforts. However, there is also a possibility that low-density populations did indeed persist, but due to difficulties in detecting this pest without specific lures or traps, small populations went undetected.

Focusing efforts on assessing populations and on estimating spatial proximity is important in describing and predicting spread of non-native pests. Our findings suggest that SLF has spread from 2014–2020 primarily through local diffusion with less frequent but consistent long-distance dispersal from previously established populations with influence from human populations. Based on the results presented here, we anticipate that SLF will continue to spread in the USA, though management and eradication efforts may effectively reduce population densities, reproductive potential, and ultimately rate of spread. Additional monitoring efforts to prevent and detect long-distance dispersals may prove useful, especially regarding transports of materials from areas with existing SLF populations.

We thank staff of USDA APHIS and Pennsylvania Department Agriculture for providing survey data and for providing feedback on this manuscript. The spotted lanternfly visual survey data used in or part of this publication was made possible, in part, by APHIS. This publication may not necessarily express the views or opinions of the APHIS. This research was partially supported by National Science Foundation Macrosystems Biology grant 1638702 to S.F. and A.L., the USDA McIntire-Stennis program and USDA Forest Service grant 21-CR-11330145-065 to S.F., and grant EVA4.0, No. CZ.02.1.01/0.0/0.0/16_019/0000803 financed by Czech Operational Programme “Science, Research, and Education” to A.M.L.