Urban areas are expanding rapidly, with the majority of the global and US population inhabiting them. Urban forests are critically important for providing ecosystem services to the growing urban populace, but their health is threatened by invasive insects. Insect density and damage are highly variable in different sites across urban landscapes, such that trees in some sites experience outbreaks and are severely damaged while others are relatively unaffected. To protect urban forests against damage from invasive insects and support future delivery of ecosystem services, we must first understand the factors that affect insect density and damage to their hosts across urban landscapes. This study explores how a variety of environmental factors that vary across urban habitats influence density of invasive insects. Specifically, we evaluate how vegetational complexity, distance to buildings, impervious surface, canopy temperature, host availability, and density of co-occurring herbivores impact three invasive pests of elm trees: the elm leaf beetle (Xanthogaleruca luteola), the elm flea weevil (Orchestes steppensis), and the elm leafminer (Fenusa ulmi). Except for building distance, all environmental factors were associated with density of at least one pest species. Furthermore, insect responses to these factors were species-specific, with direction and strength of associations influenced by insect life history. These findings can be used to inform future urban pest management and tree care efforts, making urban forests more resilient in an era where globalization and climate change make them particularly vulnerable to attack. Keywords: urban forest, invasive species, impervious surface, temperature, species interactions.

Insect Density

At each sampling period, we measured insect density on four branches of each tree, one branch in each cardinal direction (N, S, E, and W). The sampling unit was a 30 cm terminal branch (Dahlsten et al., 1993; Rodrigo et al., 2019), and we assumed equal leaf area per branch. All sampled branches were in the lower canopy up to 3 meters from the ground, and branches that could not be reached from the ground were accessed using a ladder. Sampled branches were haphazardly chosen from a distance where insects were not distinguishable to avoid sampling bias.

On each tree branch, we counted individuals of each observable insect stage: beetle eggs, larvae, and adults (the beetle pupates in cryptic locations such as under bark or in the soil, and thus pupae were not counted); weevil leaf mines and adults; and the number of leaves with leafminer mines. Individual leafminer mines were not counted because adult females lay multiple eggs per leaf, and it is common for mines to merge and become indistinguishable from one another as larvae develop. Thus, it was not possible to count the number of individual mines for this species. Leafminer adults were not counted because this stage had disappeared for the season by the start of the first sampling period. The total number of leaves on each branch was also recorded. In addition to serving as the response variable for our environmental hypotheses, insect density of each species was also used as predictor variables for the co-occurring herbivore hypothesis.

Tree 0 indicates the end of the dataset.

Urban Site Factors

Host Availability (AllElm_Density)

We measured host availability digitally by counting the number of elm trees within a 100 meter buffer around each tree using QGIS version 3.10.12 (QGIS Development Team, 2022) and a dataset of publicly managed trees provided by municipal forestry departments. We chose a 100 meter radius because significant changes in insect density are detectable for multiple insect species at this spatial scale (Sperry et al., 2001). Although Siberian elm is a preferred host of the insects in this system, other species of elm may also serve as hosts and were thus included in this data set. Following digital assessment, we verified all counts in situ to capture any visible privately owned trees and verify that trees in the dataset were still alive and present in the field. Despite efforts to avoid spatial autocorrelation, four trees had 100 meter buffers that overlapped with the buffer of another tree (that is, two locations where two trees had overlapping buffers). Because the maximum overlap was <14% of the buffer area, we retained these trees in our analyses.

Vegetational Complexity (SCI_0_500)

We measured the structural complexity of the vegetation in a 10 x 10 meter area around each tree following Shrewsbury & Raupp (2000, 2006). Specifically, we sectioned off a 10 x 10 meter area around each study tree and divided this area into one hundred  1² meter plots. In each of these plots, we recorded five vegetation categories: ground cover (e.g., mulch or turf grass), herbaceous plants (e.g., garden annuals/perennials, tall native grasses), shrubs (e.g., hydrangea, boxwood, barberry), understory trees (e.g., juniper, plum, crabapple, small Siberian Elm), or overstory trees (those with mature canopy including ash, pine, and other elm). One point was awarded for each vegetation type present, resulting in 0-5 points awarded in each plot. To quantify complexity of the vegetation in a continuous way, points were summed for all one hundred plots. Thus, each tree received a vegetational complexity score between 0 and 500. 

Building Distance (Building Distance_m)

To assess the local availability of structures for insect overwintering, we measured the distance of each sampled tree to the nearest building in meters as in Speight et al (1998). This was performed digitally using QGIS version 3.10.12 (QGIS Development Team, 2022) and the ESRI Standard Basemap, which displays built structures.

Impervious Surface (ImperviousSurface_20m)

Impervious surface data were obtained through the USGS Multi-Resolution Land Characteristics Consortium (Dewitz & US Geological Survey, 2021) on a 30 x 30 meter scale and processed using QGIS version 3.10.12 (QGIS Development Team, 2022). We used the zonal statistics tool to calculate the percentage of impervious surface within a 20 meter buffer surrounding each sampled tree, which is more predictive of herbivorous insect density than impervious surface at larger spatial scales (Just et al., 2019). Although impervious surface data were not available at a smaller spatial scale, the zonal statistics tool allowed us to obtain an estimate of impervious surface within 20 meters of each tree using 30 x 30 meter data by computing an average impervious surface value based on weighted averages of the extent to which each 30 x 30 meter pixel overlapped with the 20 meter buffer around a tree.

Canopy Temperature (MeanTemp_Night)

Canopy temperature at each tree was measured every 1.5 hours via the iButton Thermochron (model DS1921G-F5). Temperature logging began at 7:30AM MST on June 12 and ended at 7:30AM MST on August 25 for a total of 1,185 data points per logger. We placed each logger in a compostable container to prevent contact with direct sunlight and attached them with a zip tie to branches approximately 2-3 meters from the ground. We placed temperature loggers on the east side of the tree wherever possible or on the west side of the tree if a stable eastern location was not available. Despite efforts to minimize contact with direct sunlight, several loggers recorded artificially inflated temperatures. This made mean and maximum temperatures impractical for analysis. We used mean nighttime temperature in the following analyses (7:30PM-7:30AM MST, n=666 measurements per logger) because the urban heat island effect is less variable, occurs more frequently, and is more intense in urban canopies at night compared to the day (Du et al., 2021; Sun et al., 2019).