A crucial asset in the management of invasive species is the open-access sharing of data on the range of invaders and the progression of their spread. Such data should be current, comprehensive, consistent and standardised, to support reproducible and comparable forecasting efforts amongst multiple researchers and managers. Here, we present the lydemapr R package containing spatiotemporal data and mapping functions to visualise the current spread of the spotted lanternfly (Lycorma delicatula, White 1841) in the Western Hemisphere. The spotted lanternfly is a forest and agricultural pest in the eastern Mid-Atlantic Region of the U.S., where it was first discovered in 2014. As of 2023, it has been found in 14 states according to State and Federal Departments of Agriculture. However, the lack of easily accessible, fine-scale data on its spread hampers research and management efforts. We obtained multiple memoranda-of-understanding from several agencies and citizen-science projects, gaining access to their internal data on spotted lanternfly point observations. We then cleaned, harmonised, anonymised and combined the individual data sources into a single comprehensive dataset. The resulting dataset contains spatial data gridded at the 1 km² resolution, with yearly information on the presence/absence of spotted lanternflies, establishment status and population density across 658,390 observations. The lydemapr package will aid researchers, managers and the public in their understanding, modelling and managing of the spread of this invasive pest.

Biological invasions, crop pest, data science, forecasting, Lycorma delicatula, management, open access data, reproducibility, spread modelling

Due to the globalisation of trade and the homogenisation of urban and suburban habitats, the accidental introduction and establishment of invasive species is ever more likely (Hulme 2009). When establishment goes undetected and eradication becomes less viable, the goal should be to mitigate the negative effects generated by invasive species (Diagne et al. 2020; Fantle-Lepczyk et al. 2022; Leroy et al. 2022). In doing so, one of the main challenges is tracking the spread of established invasive alien species so that control measures to slow spread, reduce impact and conserve biodiversity can be effectively enacted (Robertson et al. 2020). High quality data on past and present spread of invasives are key to model invasive spread accurately enough to provide robust forecasts on which to base management decisions.

A multitude of modelling techniques to forecast spread is available to researchers (Fisher 1937; Skellam 1951; Higgins and Richardson 1996; Kot et al. 1996; Neubert et al. 2000; Travis and Dytham 2002; Clark et al. 2003; Jongejans et al. 2011; Rodrigues and Johnstone 2014; Hudgins et al. 2017). Despite different assumptions and approaches to the modelling itself, fitting and validating models rely on longitudinal, spatially-explicit data on the occurrence or density of the spreading invasive species. Different models need to be built upon the same standardised data for comparisons between models to reflect genuine differences in model assumptions (e.g. Norberg et al. 2019). Comparing models with standardised data highlights which biological aspects of spread coded in each model are crucial to manage (Sakai et al. 2001). In addition, building models on the same data provides a more solid ground to combine them into ensemble models, which offer a higher degree of reliability compared to a single model (Araújo and New 2007). However, there are three hurdles that must be overcome before such standardised data for modelling be made available.

The first hurdle that must be overcome when developing a standardised dataset on invasive spread is to develop relationships with the agencies, institutions and citizen-science projects collecting data on the invasive of interest. For pests with negative impact on agricultural activity or forest habitats, local agencies, state departments and research institutions associated with the species first discovery are likely to operate data collection. If the pest is spreading across geopolitical boundaries, multiple organisations with different jurisdictions and areas of operation are likely to collect field data. In addition, easy-to-identify pests are likely to attract public attention and involvement, fostering the collection of citizen-science data (Dickinson et al. 2010; Catlin-Groves 2012; Sullivan et al. 2014; Kobori et al. 2016; Johnson et al. 2020; Norman-Burgdolf and Rieske 2021; Santaoja 2022). Obtaining access to the data often requires directly contacting the maintainer of the dataset in the relevant institution and obtaining memoranda-of-understanding to use the data once shared. Each agency will follow unique data sharing agreements, which need to be discussed in-depth at this stage.

Once the data are obtained, the heterogeneity of the data collection protocols adopted by different agencies requires several additional steps to harmonise the survey results before they can be combined into a single dataset (Kelling et al. 2009). This second hurdle is often the most time-consuming and requires a high degree of eco-informatic skill in data handling and management (Michener and Jones 2012). Non-standardised data collection demands an in-depth understanding of the collection protocols used in order to match the information collected across different surveys (Hampton et al. 2013). For this reason, harmonisation often demands an active collaboration with the agencies that collected the data, to ensure the data are interpreted correctly, especially when surveys lack metadata (Jones et al. 2019).

The third hurdle is essential, yet not often acknowledged: data anonymisation. Calls to make scientific knowledge more accessible and transparent have pushed ecological data to be published alongside many scientific papers (Reichman et al. 2011). This process is paramount to improve collaboration and repeatability of scientific studies, although some limitations need to occur to ensure sharing open access data is done safely (Lindenmayer and Scheele 2017; Lunghi et al. 2019). One such limitation concerns data at high spatial resolution, the publication of which could infringe upon individual privacy and personal interests (Zipper et al. 2019). Due to this, invasive spread data need to be carefully and fully anonymised to ensure stakeholders are protected and served. This is especially true when knowledge on the infested state of a property could cause its value to decrease or the value of the goods produced to be affected (Zhang and Boyle 2010; Kovacs et al. 2011). Anonymisation practices include the removal of personal information, as well as data handling that reduces the spatial resolution to an optimal compromise between conveying relevant information and safeguarding privacy.

The spotted lanternfly (Lycorma delicatula, White 1845; often referred to as SLF in literature) was first discovered in the United States in Berks County, Pennsylvania, in 2014 (Barringer et al. 2015; Dara et al. 2015) and, by 2023, spread to 14 states across the Northeastern, South-Atlantic and Midwestern United States (Urban et al. 2021; NYIPM 2023). This phloem-feeding planthopper is native to China and was likely introduced accidentally via a shipment of landscaping materials. The spotted lanternfly is known to feed on over 100 species of plants (Barringer and Ciafré 2020; Murman et al. 2020; Huron and Helmus 2022) and poses a major economic burden on viticulture as it feeds on grapevines reducing total yield and plant vigour (Urban 2020). There is a high risk of spotted lanternfly impacting the global wine market by spreading to areas like California and Europe (Huron et al. 2022).

State agencies and the United States Department of Agriculture (USDA) have collected large amounts of data on spotted lanternfly spread through field surveys. In addition, given the species is easily recognised and hard to misidentify, an extensive campaign to educate the public has promoted the collection of citizen-science data. Data are collected through individual use of well-established applications such as iNaturalist, which allow for users to record geo-referenced observations of wildlife sightings, as well as through the use of applications developed ad hoc by State Departments of Agriculture to collect data on the spotted lanternfly. Given the variety of sources and the refinement of protocols for data collection, the data on this species are heavily heterogeneous. Currently, any research team analysing the spread of the pest has to invest a significant amount of time processing the data before using them in model construction and validation (Wakie et al. 2020; Cook et al. 2021; Huron et al. 2022; Jones et al. 2022; Ramirez et al. 2023).

Here, we describe the R package lydemapr (Lycorma delicatula mapping in R), containing an up-to-date, fully anonymised and regularly refined, longitudinal, spatially-explicit dataset of spotted lanternfly records throughout the United States since its first discovery. The dataset includes information derived from field surveys and citizen-science observations and reports observed presence/absence of this invasive species in surveyed areas, as well as the presence of established populations and estimates of population density. In addition, the package contains tools to visualise the data by mapping them and to obtain summary tables of the dataset. The goal of this package is to provide a baseline for future modelling efforts to forecast the spread of the spotted lanternfly and to foster more effective collaboration between agencies and researchers. The lydemapr package was fully developed in R (R Core Team 2021) and is available as an online repository at https://github.com/ieco-lab/lydemapr.

The dataset contained in the package represents an anonymised and condensed comprehensive record of data collected by several federal agencies, state agencies and citizen-science projects on the presence, establishment and population density of the spotted lanternfly in the United States (Fig. 1). Sources include the Departments of Agriculture for the States of Pennsylvania, Delaware, Indiana, Maryland and New Jersey; the New York State Department of Agriculture and Markets; the Virginia Department of Agriculture and Consumer Services; the Virginia Polytechnic Institute and State University; the United States Department of Agriculture; and public reporting from iNaturalist. The field data were collected through a variety of methods, including surveys aiming to estimate establishment status and spotted lanternfly population density, control actions to manage population through egg mass destruction and trapping of nymphs and adults and citizen-science observations collected through self-reporting or direct involvement through research projects. Self-reporting tools include two separate platforms developed by the Pennsylvania Department of Agriculture (PDA) in association with Penn State University (PSU) and the New Jersey Department of Agriculture (NJDA). In addition, we included data collected through an independent citizen-science projects of limited duration run by the Virginia Polytechnic Institute and State University and the Virginia Cooperative Extension.

Figure 1.

Conceptual graph describing the process leading to the distribution of the R package lydemapr. Data are collected by individual sources through multiple surveying processes. The datasets compiled this way are gathered from the sources and individually processed, then combined into a single comprehensive dataset. This is anonymised through both a censoring step and a spatial transformation to reduce spatial resolution. For the spatial transformation, latitude and longitude of individual survey points are rounded to the centroids of a 1-km² resolution grid. The aggregated and anonymised dataset is distributed through the package, together with functions to visualise the spread of the invasion through time.

At the date of this publication, the aggregated and anonymised dataset contained 658,390 individual observations pertaining to 61,715 point-locations throughout the United States collected between 2014 and 2021. These 61,715 point-locations represent centroids of a 1 km² grid at which the geospatial data were aggregated for anonymisation. The exact latitude and longitude of each survey contained in the geospatial data collected by the sources were rounded to the coordinates of the centroids. This approach, while removing the ability to derive property-level information from the dataset, allowed us to distribute survey-level information the data users can summarise as it best fits their needs. All variables containing traceable information regarding personal names, business names, contact information and comments were also removed from the dataset. The choice of 1 km² was agreed upon by all data contributing agencies to represent a compromise that provides high-resolution spatial data to enable precise spatial forecasting modelling while preserving privacy of the distributed data.

The individual observations recorded in the dataset derive from surveys and individual reporting conducted in 25 states across 8 years. The data points organised by year and state are summarised in Table 1. The distribution of data points by state is greatly skewed towards highly-impacted states. While Pennsylvania and the neighbouring states of Delaware, Maryland, New Jersey, New York and Virginia account for over 95% of data points (630,688 out of 658,390), other states in the western part of the country only account for a handful of surveys, mostly as a result of anecdotal reporting. Across time, it is easy to appreciate how surveying effort has increased, likely due to both the spread of lanternfly and to a higher investment of resources.

About 40% of the total data points were obtained through citizen-science projects; the well-established PDA and NJDA public reporting tools provided over 250,000 individual data points since 2019, while iNaturalist added just over 10,000 points. While management and surveying efforts led by state and federal agencies often focus on the leading edge of the invasion, where control actions are more effective, public reporting provides a constant and consistent source of data at the core. This helps the monitoring of these areas to be consistent and protracted in time, without subtracting important resources and work hours from managing the edge. In addition, iNaturalist provides constant, yet scattered, observations in areas where the surveying effort is not focused, as they are far from the invasion range. Those observations can then be confirmed by specialists during spatially-targeted surveys. The reliability of individually-reported records might vary with the experience and knowledge of the reporter. For this reason, in the dataset, records collected through citizen-science efforts are clearly distinct from records collected through expert-led surveys through the use of different categories under the variable “collection_method”. This allows users of the data to only focus on records deriving from management and control actions, if necessary.

The lydemapr package can be installed in two different ways. The public repository allows the user to install the package directly from GitHub, by executing the following command in a local R or RStudio instance: devtools::install_github(“ieco-lab/lydemapr”, build_vignette = TRUE). This requires the package devtools (Wickham et al. 2022) and its dependencies to be installed locally. Alternatively, the package can be obtained by cloning the repository from the GitHub page https://github.com/ieco-lab/lydemapr. The package can then be installed locally by opening the file lydemapr.Rproj in RStudio and clicking “Install package” in the Build tab (or by executing the command devtools::install()). Once the package is installed, the user has access to the complete dataset, which can be loaded by typing lydemapr::lyde in the R console. In addition, the package contains a rarefied and summarised version of the same dataset at a lower spatial resolution (10 km²), which can be accessed by typing lydemapr::lyde_10k instead. All information concerning package installation and data access is also available at the front page of the GitHub repository.

The R package structure allows us to update the dataset regularly as more data become available and if funding is obtained to support this initiative. In addition, a live GitHub repository grants us the ability to add functionalities and to improve the visualisation and summary tools included.

If the user is only interested in accessing the data without using the R package or is unfamiliar with R, all datasets contained in lydemapr are available for download through Zenodo (DOI: 10.5281/zenodo.7976229), where the user can download the data (in .csv format) and Metadata associated with it.

For a summary overview of the data, the function lyde_summary() provides a breakdown of the dataset, showing the number of data points collected each year in each state where data have been collected (Table 1). The package contains two customisable functions that can be used to visualise the data spatially. The function map_spread() provides an up-to-date map displaying the progression of the established invasion range through time, in addition to the locations of surveys which did not detect established populations (Fig. 2). Function arguments allow the user to select the spatial resolution at which the data should be mapped (choosing between 1 and 10 km²) and the spatial extent of the figure produced. A second function included in the package, map_yearly() maps the findings of the survey efforts in terms of the species' population density. The visualisation is broken down by the year the surveys were conducted (Fig. 3). Through this visual depiction, it is possible to observe where survey efforts have been focusing on each year, as the invasion front progressed.

Figure 2.

Map produced through the package function map_spread(). The map shows the year of first discovery of established populations of the spotted lanternfly (coloured points) in 1-km² grid cells across the eastern United States, as well as the location of negative survey records for the establishment of the species (grey crosses).

Figure 3.

Map produced through the package function map_yearly(), showing the population density of spotted lanternfly assessed yearly in 10-km² grid cells across the eastern United States (red tiles).

The dataset we provide on the spread of the spotted lanternfly, a high-impact forest and grapevine pest, will be useful in a variety of current and future efforts. Several models have been developed to forecast the future spread and establishment potential of spotted lanternfly in the United States and globally (Jung et al. 2017; Wakie et al. 2020; Huron et al. 2022; Jones et al. 2022; Lewkiewicz et al. 2022; Maino et al. 2022). Statistical forecasting models (e.g. Wakie et al. 2020; Huron et al. 2022; Jones et al. 2022) heavily rely on high resolution spatial data to derive future predictions. Leveraging this big data-set will allow new models to be developed and current ones to be refined and improved. On the other hand, mechanistic mathematical models (Lewkiewicz et al. 2022; Maino et al. 2022), despite building their predictions through a bottom-up approach that involves a deeper understanding of the species’ own biology and ecology, require spatial data for validation and model tuning. To ensure future models can be compared and combined through ensemble procedures, these models should be based on the same historic and present spread data of spotted lanternfly, reaffirming the importance of a unified and readily available dataset.

From a management standpoint, a comprehensive data-set can provide additional information on population trends through time in specific areas, allowing for the expansion of current studies (Cook et al. 2021), as well as offering insight on the efficacy of control actions over time. In addition, our openly-accessible and comprehensive dataset has broad applications in education, to promote citizen-science initiatives in under-surveyed areas, but also to provide an opportunity for data science projects for students. As the issues related to the spread of invasive species are often issues students experience first-hand, working on this dataset can represent an engaging learning opportunity.

There were two unexpected challenges to creating the lydemapr dataset. One of the main challenges we encountered was the heterogeneity in the data collection methods. This challenge greatly inflated the time, effort and eco-informatic data-coding skills required to aggregate the data. The heterogeneity was greater in the first few years (until about 2019), when more and more agencies were becoming involved, but the coordination between them was low. To solve conflicts encounters when harmonising the data, which occurred, in particular, when combining different methods to score population density of spotted lanternfly, we contacted directly the maintainers of the individual datasets for insight. An additional challenge we faced was reaching a compromise between safeguarding the privacy of stakeholders while providing a high-resolution dataset to allow accurate forecasting and management planning. Protecting individual interests while allowing data to be shared openly is a topic of current relevance (Zipper et al. 2019). The resolution of 1 km² used in our dataset was reached after thorough discussions with the agencies involved, to ensure no breach of privacy occurred. Paramount to overcome both challenges was a tight collaboration with the agencies. We contacted data maintainers soon after a new agency was becoming involved in data collection, to start developing a relationship of trust and cooperation. This created an open line of communication with the agencies collecting the data from the field and curating the individual datasets and produced a feedback loop that we believe strengthens the quality and reliability of our dataset.

SDB and MRH conceived the paper, gathered the data, produced the comprehensive dataset and wrote the code for the package. LB, PK, JL and GRP provided survey data and helped harmonise it across sources. All authors contributed with the writing of the manuscript.

The package, containing the open access data, is stored as a public repository at https://github.com/ieco-lab/lydemapr. Additionally, versions of the 1 km² and 10 km² datasets are stored on Zenodo DOI: 10.5281/zenodo.7976229.

We would like to thank Eric Day for providing data on a citizen-science project run by the Virginia Polytechnic Institute and State University and the Virginia Cooperative Extension. We thank Jocelyn Behm, Stefani Cannon, Anna Carlson, Jason Gleditsch, Stephanie Lewkiewicz, Sam Owens, Payton Phillips and Timothy Swartz for their insightful comments on early drafts. This work was funded by the United States Department of Agriculture Animal and Plant Health Inspection Service Plant Protection and Quarantine under agreements AP19PPQS&T00C251, AP20PPQS&T00C136, AP20PPQS&T00C118, AP22PPQS&T00C146 and AP22PPQS&T00C097; the United States Department of Agriculture National Institute of Food and Agriculture Specialty Crop Research Initiative Coordinated Agricultural Project Award 2019-51181-30014; the Pennsylvania Department of Agriculture under agreements 44176768, 44187342, C9400000036, C94000833 and C940000835; and the California Department of Food and Agriculture under agreement A20-0850-000-SA.