Low current and future spotted lanternfly suitability in New York wine-growing regions tempers vineyard risk

invasive species, spotted lanternfly, tree-of-heaven, risk assessment, climate change, 11 species distribution modeling, remote sensing, Lycorma delicatula 12 13 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted May 31, 2025. ; https://doi.org/10.1101/2025.05.28.656630doi: bioRxiv preprint 2

14 Invasive species disrupt natural and agricultural ecosystems, leading to cascading shifts and 15 economic losses. The spotted lanternfly (Lycorma delicatula) has spread rapidly across the 16 northeastern United States, facilitated by its ecological flexibility, human activities, and the 17 widespread presence of its favored host, tree-of-heaven (Ailanthus altissima). This polyphagous 18 insect threatens New York vineyards, but the level of risk is uncertain. Using ~22,000 tree-of-19 heaven and ~10,000 SLF observations, species distribution models were developed for New 20 York State using Random Forests. Models incorporate ~1km scale resource availability, remote 21 sensing, human influence as well as geographic and climatic variables to better reflect factors 22 that define niche breadth. A tree-of-heaven suitability index was used as a predictor of spotted 23 lanternfly suitability. The species distribution models for tree-of-heaven and spotted lanternfly 24 had high accuracy (98%). To quantify vineyard risk, tree-of-heaven and spotted lanternfly 25 suitability were integrated with a distance weighted measure of vineyard proximity. Habitat 26 suitability and vineyard risk were modeled under both current (2011-2040) and worsening future 27 (2041-2070) climate scenarios. For tree-of-heaven the statewide increase in suitable habitat is 28 modest and high suitability in the wine-growing regions actually decrease from 1.4% to 1.3%. 29 High suitability nearly doubles between current and future conditions for spotted lanternfly, but 30 the area at high risk is still less than 1.5% of the wine-growing regions. Though the future risk 31 SLF poses to vineyards in the Finger Lakes and Hudson Valley is projected to increase in area 32 and intensity, the risk burden under worsening climate conditions in the near-future is less than 33 expected. These results highlight the need for fine scale management strategies and species-34 specific estimates in response to climate change and resource availability across the landscape. 35 36 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted May 31, 2025. ; https://doi.org/10.1101/2025.05.28.656630doi: bioRxiv preprint 3

37 Climate change has exacerbated the spread of invasive species around the world (Finch et al., 38 2021). Invasive species cause cascading environmental harm, including shifts in community 39 biodiversity and economic loss (Xu et al 2006, Oliveira et al 2012, Boivin et al 2016). The 40 spotted lanternfly (Lycorma delicatula (White) [Hemiptera:Fulgoridae]), a planthopper native to 41 China, is highly polyphagous, feeding on over 100 plant species across 33 families (Uyi et al., 42 2021). Adults lay eggs on materials that are often moved through human activities, increasing the 43 rate of population spread (Elsensohn et al., 2024, Strömbom et al. 2024). Spotted lanternfly also 44 successfully persists under a wide range of climatic conditions (Keena et al. 2023). High 45 phenotypic plasticity may be as important as genetic diversity in maintaining fitness in novel 46 habitats and under changing climatic conditions (Ghalambor et al., 2007). Species with extensive 47 geographic distributions often demonstrate high phenotypic plasticity (Bennett et al., 2019; 48 Valladares et al., 2014), improving success in novel regions of recent invasion and under novel 49 conditions due to climate change (Thompson et al., 2020). 50 51 The spotted lanternfly can survive and develop on a single plant host species, but “diet mixing” 52 by feeding on multiple host species enhances fitness and egg production (Laveaga et al., 2023). 53 Adults demonstrate preference for tree-of-heaven (Ailanthus altissima (Mill.) Swingle), a 54 widespread invasive deciduous tree that was first introduced to the U.S. in 1784 (Soler and 55 Izquierdo, 2024). Due to its broad diet breadth, spotted lanternfly presents a high economic risk 56 to many crops, but especially cultivated grapes (Vitis labrusca, Vitis vinifera and Vitis vinifera 57 interspecific hybrids) (Leach and Leach 2020, Murman et al., 2020 and Huron et al., 2022). 58 Spotted lanternfly exposed to a mixed diet of grapevine and tree-of-heaven resulted in greater 59 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted May 31, 2025. ; https://doi.org/10.1101/2025.05.28.656630doi: bioRxiv preprint 4 nymphal development, egg production, and body mass compared to those fed on either species 60 alone (Laveaga et al., 2023). Since its 2014 introduction in Pennsylvania, the spotted lanternfly 61 has caused over $300 million in damage to U.S. viticulture (Harper et al., 2019). 62 63 Excessive spotted lanternfly feeding reduces grapevine ability to store carbohydrates and 64 nitrogen in the root system in the autumn, compromising plant health and likely reducing winter 65 survival (Harner et al., 2022; Lavely et al., 2022). New York state is the third largest producer of 66 grapes in the country with 1,400 vineyards and generating $6.65 billion in economic activity in 67 the state (Dunham & Associates, 2022). Grapes grown in New York are used for wine, juice and 68 table grapes with vineyards distributed across four major production zones. New York State is 69 characterized by microclimatic variation with minimum annual extreme temperatures as low as -70 35oC, making it an important region to help define the cold hardiness limits of spotted lanternfly 71 (Gómez-Marco and Hoddle 2022, Turbelin et al. 2024). The widespread distribution of tree-of-72 heaven along with spotted lanternfly’s phenotypic plasticity, polyphagy, propensity for 73 hitchhiking and potential for economic damage make this an ideal system to examine habitat 74 suitability and develop landscape-informed risk models in the context of global change. 75 76 Correlative species distribution models have been widely utilized to estimate landscape habitat 77 suitability and climate change impact on both native and invasive species and have proven useful 78 in guiding policymaking (Gillson et al., 2013). These models establish statistical relationships 79 between a species' current geographic distribution and relevant climatic parameters, enabling 80 researchers to predict potential spread which can be used for regional risk assessment (Pacifici et 81 al., 2015; Evans et al., 2015). Unfortunately, many species distribution models do not incorporate 82 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted May 31, 2025. ; https://doi.org/10.1101/2025.05.28.656630doi: bioRxiv preprint 5 the broad diversity of variables beyond climatic ones, that we know to be important in defining a 83 species niche (Mod et al., 2016). 84 85 High-resolution remote sensing techniques can identify ecological conditions that promote the 86 establishment and spread of invasive species, allowing vulnerable areas to be identified (Wang 87 and Hu, 2021). Remote sensing imagery reflects real-time data on land use, vegetation health, 88 and climatic conditions and can improve the accuracy of ecological models (Zhang et al., 2017). 89 Combining remote sensing data with climate change projections is especially relevant for spotted 90 lanternfly, whereby remote sensing can help identify the distribution of tree-of-heaven, 91 improving site-specific early detection and informing targeted management strategies (Koch, 92 2021, Hao et al., 2024). 93 94 In anticipation of the risk to New Yorks grape growing industry, the New York State Department 95 of Agriculture and Markets initiated a statewide spotted lanternfly monitoring effort in 2020 that 96 has since collected over 23,000 observations of both spotted lanternfly and tree-of-heaven. 97 Observations of tree-of-heaven from the Global Biodiversity Information Facility (GBIF) 98 provide an important additional dataset of distribution across New York State. Using these 99 observational data to quantify the risk of spotted lanternfly to New York vineyards represents an 100 especially timely effort since in 2024 spotted lanternfly were found in the Finger Lakes region of 101 New York. The Finger Lakes are one of the primary grape-growing regions in the state, located 102 more than 200 miles away from the site of initial introduction. We aimed to produce a tree-of-103 heaven index as an explanatory variable in modeling the spotted lanternfly habitat suitability. 104 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted May 31, 2025. ; https://doi.org/10.1101/2025.05.28.656630doi: bioRxiv preprint 6 Using tree-of-heaven suitability, spotted lanternfly suitability and vineyard density across the 105 state we quantify the risk spotted lanternfly poses to vineyards. Our specific goals were to: 106 1. Quantify tree-of-heaven habitat suitability using a large observational dataset, climatic, 107 geographic and remote sensing variables 108 2. Assess the value of including a tree-of-heaven suitability index as an explanatory variable in 109 defining spotted lanternfly habitat suitability in addition to other environmental variables and 110 the intensity of human activity. 111 3. Quantify the potential risk of spotted lanternfly suitability for New York vineyards under 112 both current and future predictive modeling. 113

114 The focal region for this research is the state of New York, United States. Most of the grapes 115 grown in the state are used for grape juice production (68%), but New York is also the third 116 largest wine producing state (Dunham & Associates, 2022). Both wine and juice grapes are 117 primarily grown in four regions across the state: Lake Erie, the Finger Lakes, the Hudson Valley 118 and on Long Island. The New York Wine and Grape Foundation shared 6,609 vineyard locations 119 across the state (Figure 1). The climatic and geographic remote sensing variables used for both 120 tree-of-heaven and spotted lanternfly suitability ranged in resolution from 1-km to 30-m (Tables 121 S1, S2). 122 Climatic Variables 123 To model tree-of-heaven and spotted lanternfly habitat suitability, quarterly and monthly mean 124 climatic variables were downloaded from CHELSA CMIP6 ISIMIP3 version 2.1(Karger et al., 125 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted May 31, 2025. ; https://doi.org/10.1101/2025.05.28.656630doi: bioRxiv preprint 7 2021). Focusing primarily on GFDL-ESM4 datasets, we assessed 14 climatic variables under 126 both current (2011-2040; SSP 370, RCP 7.0) and future (2041-2070; SSP 585, RCP 8.5) 127 conditions (Table S1). 128 129 Humidity is a critical environmental variable that is frequently ignored when defining habitat 130 suitability for ectotherms (Brown et al. 2023). Monthly near-surface relative humidity from the 131 National Center for Atmospheric Research (NCAR) CMIP5 CCSM4 model was included as an 132 additional climatic variable (See Supplementary Methods: Climatic variables). We used ArcGIS 133 Pro to extract the near-surface humidity estimates for years corresponding to our current (2011-134 2040, RCP 6.0) and future (2041-2070, RCP 8.5) climatic scenarios. 135 136 Geographic Variables 137 All geographic variables used in habitat suitability models are detailed in Table S2. Mean 30-arc 138 sec Digital Elevation Model via U.S. Geological Survey GMTED 2010 files were used to obtain 139 the topographic position and elevation of New York State. The soil properties dataset consisted 140 of three raster datasets describing the following soil properties: available water capacity, field 141 capacity, and soil porosity (Boiko et al., 2021). 142 143 Vegetation productivity is effectively measured using satellite-based remote sensing by the 144 Normalized Difference Vegetation Index (NDVI) (Leitão and Santos, 2019) which also 145 characterizes differential nitrogen and chlorophyll levels, as well as crop vigor and biomass 146 (Farooque et al., 2023). Mean values of NDVI between August 1, 2023, and November 1, 2023 147 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted May 31, 2025. ; https://doi.org/10.1101/2025.05.28.656630doi: bioRxiv preprint 8 from Sentinel-2 Harmonized Multispectral Instrument (MSI) were used to characterize New 148 Yorks’ autumn season (see Supplementary Methods: Geographic variables). 149 150 Tree-of-Heaven is a highly shade-intolerant species that thrives in disturbed, high-light 151 environments, including forest edges, canopy gaps, roadsides, and open fields (Fryer, 2010). 152 Dense, mature forests with canopy cover exceeding 50% inhibit TOH establishment and 153 persistence due to insufficient light availability (Fryer, 2010; Isler et al., 2023). The Landsat 154 dataset (2010; version 4) was reclassified to include only 30m pixels with forest canopy cover at 155 or below a 40% threshold across the state (Figure S1). The development of this dataset, with 156 edge habitat defined by limited tree canopy-cover, was used to identify pixels with optimal 157 conditions for tree-of-heaven. 158 159 To account for the importance of human-assisted transportation corridors in the spread of Spotted 160 Lanternfly (Elsensohn et al., 2024), we included the human influence index dataset from NASA 161 Socioeconomic Data and Applications Center (SEDAC), the Last of the Wild Project, 2018 162 Release version 3 (Venter et al., 2018) (see Supplementary Methods: Geographic variables). 163 164 Modeling 165 This study was implemented using R version 4.4.1 (R Core Team, 2024) and ArcGIS Pro version 166 3.3 (Esri, 2024). We employed a wide range of Comprehensive R Archive Network (CRAN) 167 packages; such as SDMtune (Vignali et al., 2020), flexsdm (Velazco et al., 2022), and blockCV 168 (Valavi et al., 2019). Random Forest classification was used to improve risk assessment (Chen et 169 al., 2021). Default settings in all distribution models were used unless stated otherwise. 170 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted May 31, 2025. ; https://doi.org/10.1101/2025.05.28.656630doi: bioRxiv preprint 9 171 Georeferenced observations of tree-of-heaven were obtained from both the New York State 172 Department of Agriculture and Markets (NYSDAM, N= 11,217) and GBIF (N= 10,926), 173 resulting in 22,143 observations. Only verified observations from NYSDAM were used for 174 spotted lanternfly, resulting in 10,143 observations. True absence data for both tree-of-heaven 175 and spotted lanternfly were unavailable, therefore pseudo-absences were generated by randomly 176 sampling an equal number of species observations (see Supplementary Methods: Data pre-177 processing). 178 179 A K-fold cross-validation approach with spatial blocks was implemented to minimize spatial 180 autocorrelation and ensure an independent dataset for validation (Soley-Guardia et al., 2024) 181 (See Supplementary Methods: Model optimization for more information). This modeling 182 approach is most often recommended for spatially structured data, regardless of the degree of 183 clustering, spatial autocorrelation, or species abundance class (Mushagalusa et al., 2024). This 184 resulted in a balanced presence-absence dataset of 36,073 tree-of-heaven and 35,312 spotted 185 lanternfly for our current model and 34,669 tree-of-heaven and 35,034 spotted lanternfly for our 186 future model. Model performance was evaluated by generating a confusion matrix; particularly 187 looking at accuracy, sensitivity, specificity, positive prediction value, and negative prediction 188 value. To assess model accuracy, we compared the root mean square error between both the 189 training-testing (RMSE) and validation (CV-RMSE) datasets (Nayak et al., 2022). 190 191 To produce the tree-of-heaven index, a weight function was implemented to compute importance 192 scores for each explanatory variable. The Mean Decrease Gini index was extracted and 193 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted May 31, 2025. ; https://doi.org/10.1101/2025.05.28.656630doi: bioRxiv preprint 10 normalized so that the weights sum to 1. The final raster output emphasizes the most informative 194 variables while minimizing the impact of those with lower predictive value (Liu and Zhao, 195 2017). All 19 explanatory variables were then reclassified to differentiate between suitable and 196 unsuitable conditions (see Supplementary Methods: Tree-of-heaven index). 197 198 To develop an additive risk index for vineyards across New York State, first a spatial analysis 199 was conducted in ArcGIS Pro, using tree-of-heaven and spotted lanternfly habitat suitability 200 raster datasets relative to the 6,609 vineyard locations across the state (Figure 1). A 10-km buffer 201 zone around each vineyard location was implemented to define the habitat proximal to vineyard 202 polygons. The ‘Zonal Statistics as Table’ function was used to compute the 90th percentile for 203 both tree-of-heaven and spotted lanternfly habitat suitability within each buffered vineyard area. 204 This statistical measure was selected to represent extreme habitat suitability conditions rather 205 than average conditions, ensuring that areas of high risk were effectively identified. This 206 approach was applied to the spatial extent of each buffered vineyard rather than statewide to 207 avoid assigning risk values to areas that are not ecologically suitable for vineyards, thereby 208 reducing the likelihood of generating inaccurate spatiotemporal patterns. An additive risk index 209 (MacKenzie, 2015) was calculated between the tree-of-heaven 90th percentile and spotted 210 lanternfly 90th percentile (see Supplementary Methods: Vineyard risk). 211 212 The Inverse Distance Weighting (IDW) function was applied to account for additional risk based 213 on vineyard proximity to the nearest neighboring vineyard. The vineyard point layer served as 214 the input dataset (N=6,609), with the newly calculated additive risk index field used as the Z-215 score parameter. The Z-score parameter serves as the input for estimating spatial risk patterns at 216 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted May 31, 2025. ; https://doi.org/10.1101/2025.05.28.656630doi: bioRxiv preprint 11 unsampled locations based on vineyard proximity. The power parameter (p) was set to three to 217 ensure that high risk regions were well delineated, while distant points had little to no influence 218 (Esri, 2024). Finally, the IDW raster output was masked based on the spatial extent of New York 219 State showcasing a final vineyard risk map. The vineyard risk map showcases a gradient 220 representation of vineyard risk across New York State, categorizing regions into various risk 221 zones based on tree-of-heaven and spotted lanternfly interactions. 222 223 224 225 Figure 1: New York State with its wine regions highlighted in grey: (A) Lake Erie, (B) Finger 226 Lakes, (C) Hudson Valley, and (D) Long Island. The map illustrates 6,609 vineyard locations. 227 228 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted May 31, 2025. ; https://doi.org/10.1101/2025.05.28.656630doi: bioRxiv preprint 12

229 Tree-of-heaven habitat suitability 230 Our tree-of-heaven model under current (2011-2040; SSP 370, RCP 7.0) climatic conditions 231 demonstrated high accuracy (98.76%), sensitivity (99.76%), specificity (83.65%), positive 232 prediction value (98.92%), and negative prediction value (95.95%). The cross-validated RMSE 233 was 0.1040. The climatic and geographic variables driving regional differentiation for tree-of-234 heaven included 19 explanatory variables. Our tree-of-heaven model under future (2041-2070; 235 SSP 585, RCP 8.5) climatic conditions demonstrated high accuracy (98.98%), sensitivity 236 (99.85%), specificity (79.50%), positive prediction value (99.09%), and negative prediction 237 value (95.92%). The cross-validated RMSE was 0.094. The importance scores for 19 explanatory 238 variables identified by the tree-of-heaven suitability model were input as a weight function for 239 the tree-of-heaven index (see Supplementary Results: Tree-of-heaven index, Figure S2). 240 241 According to our findings, Long Island and Hudson Valley were identified as moderate-to-very 242 high in suitability, whereas the Finger Lakes and Lake Erie regions were identified as low-to-243 moderate suitability (Figure 2A). Lake Erie has the least suitable habitat for tree of heaven of the 244 wine-growing regions with just 2.2% currently categorized as moderate and 0.07% as high 245 (Table S3). In this region tree-of-heaven suitability remains stable. As for the Finger Lakes and 246 Hudson Valley regions, both remain essentially stable between current and future predictions of 247 tree-of-heave suitability. In the Long Island region however suitability for tree-of-heaven is 248 expected to decline in the future (Figure S3; Table S3). 249 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted May 31, 2025. ; https://doi.org/10.1101/2025.05.28.656630doi: bioRxiv preprint 13 250 251 Figure 2: Statewide habitat suitability for tree-of-heaven in New York State in A) under current 252 conditions (2011-2040) with model Accuracy = 98.76%, Sensitivity = 99.76%, Specificity = 253 83.65%, and in B) under future conditions (2041-2070) with a warming climate (Accuracy = 254 98.98%, Sensitivity = 99.85%, Specificity = 79.50%), based on 19 explanatory variables. White 255 represents unsuitable habitat, light green is low likelihood of occurrence (0.1-25%), green 256 represents moderate likelihood of occurrence (26-75%), and dark green represents very high 257 likelihood (>75%). 258 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted May 31, 2025. ; https://doi.org/10.1101/2025.05.28.656630doi: bioRxiv preprint 14 The most influential explanatory variables, according to their respective weights, for modeling 259 tree-of-heaven between both current and future climatic scenarios were mean daily air 260 temperature of the wettest quarter, growing degree days heat sum above 0°C, net primary 261 production, mean daily minimum air temperature of the coldest month, and frost change 262 frequency (Figure 3). 263 Our current model suggests that tree-of-heaven thrives in cooler temperatures during the wettest 264 quarter, which in New York fall between June and August. However, under a worsening climate 265 scenario, the influence of this variables on suitability is expected to decline though the range of 266 suitability increases (Figure 3, Tables S4, S5). Future climate projections (SSP 585, RCP 8.5) 267 indicate that temperatures during the wettest quarter will rise in New York overall (Table S4), 268 though predictions in the grape-growing regions are variable (Table S6). Our current model 269 indicates that habitat suitability starts declining when mean daily air temperatures of the coldest 270 month are -8.15°C or warmer (Figure S4), though this temperature rises in the future. When 271 comparing our computed weights under both current and future climate scenarios, the 272 significance of this variable becomes more influential in the future (Figure 3). 273 274 Growing degree days above 0°C shows high suitability until about 3750°C, after which 275 suitability declines (Figure S4). Under future climatic conditions, suitability starts to rapidly 276 decline at around 4000°C. This variable remains one of the most important variables for 277 modeling tree-of-heaven under both current and future climate scenarios. Net primary production 278 refers to the amount of energy a plant stores as biomass over time. Under current and future 279 climatic conditions, suitability for tree-of-heaven decreases with higher net primary productivity 280 (Figure S4). 281 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted May 31, 2025. ; https://doi.org/10.1101/2025.05.28.656630doi: bioRxiv preprint 15 282 283 Figure 3. A bar plot illustrating the comparison between explanatory variables for modeling 284 habitat suitability under both the current (SSP 370, RCP 7.0) and future (SSP 585, RCP 8,5) 285 climatic scenarios, based on their relative weights, in A) for all 19 variables used in the tree-of-286 heaven model and in B) for all 15 variables used in the spotted lanternfly habitat suitability 287 model. Weights were computed by normalizing the Mean Decrease Gini index from our Random 288 Forest model. Black bars represent the current climatic scenario and grey represents the future 289 climatic scenario. 290 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted May 31, 2025. ; https://doi.org/10.1101/2025.05.28.656630doi: bioRxiv preprint 16 Frost change frequency refers to the number of instances in a given period during which the daily 291 minimum temperature or daily maximum temperature fluctuates around the freezing point of 292 0°C. Under current climatic conditions, frost-change frequency across New York varies by 293 location, with averages ranging from 40 to 180 frost days per year (Seggos, 2021). Under future 294 climate change scenarios, New York State is projected to experience a significant decrease in 295 frost days due to rising temperatures, however the frost change frequency is much more 296 regionally variable (Tables S4, S6, Figure S5). 297 298 Spotted lanternfly habitat suitability 299 The climatic and geographic variables driving regional differentiation for spotted lanternfly 300 included 15 explanatory variables. The spotted lanternfly model under current (2011-2040; SSP 301 370, RCP 7.0) climatic conditions demonstrated high (99.19%), sensitivity (99.77%), specificity 302 (84.82%), positive prediction value (99.38%), and negative prediction value (93.87%). The 303 RMSE was 0.08 and the CV-RMSE was 0.08 indicating high accuracy. The Hudson Valley was 304 identified as very high in suitability (Figure 4). The Finger Lakes and Long Island regions were 305 identified to be moderate in suitability, whereas Lake Erie was low in suitability. Our spotted 306 lanternfly model under future (2041-2070; SSP 585, RCP 8.5) climatic conditions also 307 demonstrated high accuracy (99.09%), sensitivity (99.67%), specificity (83.07%), positive 308 prediction value (99.37%), and negative prediction value (91.01%). As for model accuracy, the 309 RMSE was 0.08 and the CV-RMSE was 0.08. There is very little change expected statewide in 310 overall habitat suitability for spotted lanternfly under worsening conditions in the near-future, 311 with some regions of the state showing increased suitability while others decline (Figure S6). 312 313 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted May 31, 2025. ; https://doi.org/10.1101/2025.05.28.656630doi: bioRxiv preprint 17 314 315 316 Figure 4: Habitat suitability for the spotted lanternfly across New York State A) under current 317 conditions (Accuracy = 99.19%, Sensitivity = 99.77%, Specificity = 84.82%) and in B) future 318 conditions based on a warming climate (Accuracy = 99.09%, Sensitivity = 99.69%, Specificity = 319 83.07%), based on 15 explanatory variables. White represents unsuitable habitat, light brown is 320 low likelihood of occurrence (0.1-25%), brown represents moderate likelihood of occurrence 321 (26-75%), and dark brown represents very high likelihood (>75%). 322 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted May 31, 2025. ; https://doi.org/10.1101/2025.05.28.656630doi: bioRxiv preprint 18 Under current conditions, Lake Erie has ~31% of the region categorized as low suitability for 323 spotted lanternfly with less than 1% moderate (Table S3). Though the future sees some slight 324 increase in suitability, moderate and high suitability area still account for less than 1% of the 325 region. Similarly, in the Finger lakes ~33% is categorized as low suitability, 1% as moderate, and 326 0.10% as high. In the future, high suitability doubles, but is still less than 1%. In the Hudson 327 Valley region, currently ~39% is categorized as low suitability, with ~2% as moderate and 0.63% 328 as high. In the future, areas categorized as low will increase in suitability, with ~37% of the area 329 as low, ~3% as moderate, and 0.86% as high. In the Long Island region, ~65% is currently 330 categorized as low suitability and 0.8% as moderate. In the future, suitability decreases in area, 331 with ~42% of the area as low and 0.13% as moderate (Figure 4; Table S3). 332 333 The most influential explanatory variables, according to their respective weights, for modeling 334 the spotted lanternfly between both current and future climatic scenarios were near-surface 335 relative humidity, mean daily mean air temperature of the wettest quarter, human influence index 336 (Figure 3B). 337 338 Relative humidity is the number one variable in predicting spotted lanternfly habitat suitability 339 under current conditions and the second most important variable in the near-future. The mean 340 daily mean air temperature of the wettest quarter typically occurs between June and August in 341 New York State and ranges from -2.1 to 25.6 oC (Table S4). Spotted lanternfly are highly 342 suitable in regions that experience between -3°C and 20°C during the wettest quarter (Figure S7, 343 Table S5). 344 345 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted May 31, 2025. ; https://doi.org/10.1101/2025.05.28.656630doi: bioRxiv preprint 19 The variables that changed the most in predicting distribution between the current and future 346 conditions included: tree-of-heaven index, mean daily mean air temperatures of the driest 347 quarter, isothermality, and frost change frequency (Figure 3B). The highest suitability for spotted 348 lanternfly suitability is characterized by areas with low to moderate human influence (Figure 349 S7). Under current conditions, regions where all environmental factors agree and support tree-of-350 heaven growth, are important for predicting spotted lanternfly suitability (Figures 2, 4 and S2). 351 However, in the future, the importance of tree-of-heaven as a predictor declines (Figure 3B). 352 353 Isothermality tells us how much the temperature fluctuates during a 24-hour period relative to 354 how much it varies over the whole year. Under current conditions, isothermality levels above 0.2 355 (20%) are associated with higher spotted lanternfly suitability (Table S3, Figure S7). In the 356 future, predicted probability is even higher, though isothermality levels above that begin to 357 decrease habitat suitability (Figure S7). Climate change predictions for New York indicate that 358 while average temperatures are expected to rise, the changes in isothermality are projected to be 359 minimal (Table S4). This suggests that the relationship between daily and annual temperature 360 variations will remain relatively stable, even as overall temperatures increase. 361 362 Spotted lanternfly suitability is positively correlated with higher number of events of frost 363 change frequency —approximately 50 to 100 events during current conditions and 100 to ~150 364 events during future conditions (Table S5, Figure S7). 365 366 367 368 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted May 31, 2025. ; https://doi.org/10.1101/2025.05.28.656630doi: bioRxiv preprint 20 Vineyard Risk Assessment 369 370 The risk to the Western New York grape growing region in Lake Eris is the lowest with ~76% of 371 currently categorized at low risk and ~19% at low-to-moderate (Table S7). In the future risk 372 stays essentially the same in this region with ~77% of the area at low risk, ~17% at low-to-373 moderate. As for the Finger Lakes region, currently ~58% of this important grape growing region 374 is categorized at low risk, with ~21% at low-to-moderate, ~18.8% at moderate, and ~3% at 375 moderate-to-high. In the future, areas categorized as low and moderate will increase slightly in 376 risk, with ~5% at moderate-to-high, and ~2% at high (Figure 5). The Hudson Valley region has 377 the highest risk burden with ~20% of the region categorized as moderate-to-high risk and ~2.3% 378 at high. In the future, areas already with moderate and high risk will increase with ~21% at 379 moderate-to-high, ~2% at high, and ~2.4% at very high. In the Long Island region has currently 380 ~41% of the region classified as low-to-moderate risk and ~27% at moderate. In the future, risk 381 decreases further, with only ~9% at low-to-moderate (Figure 5; Table S7). 382 383 In addition to the area at risk is the total change in risk burden between current and future 384 conditions. Regionally, vineyards in the Finger Lakes and the Hudson Valley are projected to 385 experience the largest increase in risk, at nearly 20% in some areas (Figure S8). On the other 386 hand, the change in risk experience regionally by vineyards in the Lake Eri region is only 387 expected to increase 2%, while the Long Island region is projected to see a decline of in overall 388 vineyard risk under worsening near-future climate conditions (Figure S8). 389 390 391 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted May 31, 2025. ; https://doi.org/10.1101/2025.05.28.656630doi: bioRxiv preprint 21 392 393 Figure 5: Vineyard risk across New York State, categorizing regions into different risk zones 394 based on tree-of-heaven and spotted lanternfly interactions and vineyard proximity. Green 395 represents very low risk (≤3%), yellow represents moderate suitability (≤28%), and red 396 represents very high suitability (≤64%). In A) the current risk is shown (2011-2040), while B) 397 shows future risk under a warming climate scenario (2041-2070). 398 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted May 31, 2025. ; https://doi.org/10.1101/2025.05.28.656630doi: bioRxiv preprint 22

399 The invasive spotted lanternfly is both an economic threat and useful system for modeling 400 invasive species dynamics and how climate change may drive shifts in distribution. Here, we 401 combined tree-of-heaven index in a spotted lanternfly species distribution model and 402 incorporated climate projection modeling to quantify the current and future risk spotted 403 lanternfly poses to the New York wine and grape industry. 404 405 This study demonstrates the value of integrating extensive observational datasets and satellite 406 imagery with traditional climatic variables to assess invasive species distribution and their 407 associated risks to agriculture. Although the spotted lanternfly might benefit from warmer 408 temperatures in the future, our findings suggests that the effects of climate change on the 409 interaction between tree-of-heaven and spotted lanternfly result in a far more limited range 410 expansion than expected. 411 412 New York State will experience an increase in precipitation, warmer temperatures, and longer 413 growing seasons, thereby contributing to an environment that favors both grape development and 414 pest survival (Karger et al., 2023). The wine regions are projected to experience increases in both 415 temperature and precipitation during cold periods; however, temperature and precipitation 416 seasonality are expected to decline. Warmer temperatures and increased precipitation during 417 colder periods may expand the suitable habitat range for both tree-of-heaven and spotted 418 lanternfly, facilitating their spread into new areas, but the expansion in suitable habitat in New 419 York state appears minimal. 420 421 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted May 31, 2025. ; https://doi.org/10.1101/2025.05.28.656630doi: bioRxiv preprint 23 Under both current and future climatic scenarios, growing degree days emerged as one of the 422 most influential variables for modeling tree-of-heaven habitat suitability. The state's average 423 temperatures have already increased by approximately 3°F since 1970, and projections suggest 424 an additional rise of up to 3°F by 2080, particularly in northern regions (Seggos, 2021). This 425 warming trend is expected to result in milder winters with fewer days of frost and reduced snow 426 cover. Studies have shown that tree-of-heaven is susceptible to frost damage, particularly in 427 juvenile stages, which restricts its spread to higher elevations (Clark and Wang, 2020). The 428 association between lower net primary productivity and lower forest edge with higher tree-of-429 heaven suitability are likely driven by tree-of-heaven often thriving at the forest edge and in 430 disturbed landscapes, while contiguous forests are likely both more productive and less suitable. 431 432 Though frost-change-frequency increases in importance for predicting suitable habitat for tree-433 of-heaven under future climate conditions, there is essentially no increase in suitable habitat 434 across New York for tree-of-heaven. This demonstrates the importance of multiple variables in 435 defining habitat suitability and that future climate scenarios will see shifting conditions in way 436 that may mute the potential for habitat expansion. Given that tree-of-heaven is widely distributed 437 to the south of New York and considering climate change models predict a warming trend for 438 New York, it is likely that the temperature variables that identify upper thresholds from the 439 model reflect this warming trend rather than an actual shift in upper temperature thresholds. 440 441 Near-surface relative humidity was the most influential variable for modeling the spotted 442 lanternfly under both the current and future climatic conditions. Currently, the optimal range for 443 spotted lanternfly reproduction and growth is between 69% and 80% humidity increasing slightly 444 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted May 31, 2025. ; https://doi.org/10.1101/2025.05.28.656630doi: bioRxiv preprint 24 in the future matching the expected increase in humidity across New York. Higher relative 445 humidity levels can reduce desiccation stress, thereby improving survival rates, especially during 446 vulnerable life stages such as egg and nymphal development. Humidity plays a huge part in 447 spotted lanternfly development and reproduction, especially in summer (Keena et al., 2024). 448 Studies have indicated that in laboratory settings, egg hatch success rates are significantly 449 influenced by humidity levels, with higher relative humidity correlating with increased 450 hatchability (Liu, 2022). 451 452 Modeling the distribution of the spotted lanternfly under both the current and future climatic 453 conditions, the mean air temperature of the wettest quarter (between June and August) seems to 454 play a key role in defining habitat suitability across New York State. In both the Finger Lakes 455 and Hudson Valley, this variable is expected to increase, whereas Lake Erie and Long Erie will 456 decrease. This suggests that elevated temperatures during this period can accelerate spotted 457 lanternfly development, enhance survival rates, and potentially increase reproductive success. In 458 addition, the mean air temperature of the driest quarter is expected to increase in the future and 459 becomes an important factor in spotted lanternfly establishment in the future. Similarly, Wakie et 460 al. (2019) identified the mean temperature of the driest quarter to be most influential when 461 modeling spotted lanternfly habitat suitability. 462 463 Both the tree-of-heaven and human influence indices remain highly ranked in modeling spotted 464 lanternfly suitability under current climatic conditions. This finding underscores the dual reliance 465 of the species on both biological and anthropogenic factors for its establishment and spread. 466 Tree-of-heaven serves as a primary host plant, providing essential resources for feeding and 467 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted May 31, 2025. ; https://doi.org/10.1101/2025.05.28.656630doi: bioRxiv preprint 25 reproduction, while human influence—such as trade, transportation networks, and 468 urbanization—facilitates long-distance dispersal to neighboring regions (Murman et al., 2020; 469 Strömbom et al., 2024). However, in the future the tree-of-heaven index becomes less influential 470 while the human influence index stays highly influential. This difference potentially highlights a 471 tradeoff between favorable climatic conditions and preferred host establishment. Climatic factors 472 become increasingly important in facilitating habitat suitability for the spotted lanternfly 473 compared to available resources needed for establishment. 474 475 The increasing importance of frost-change-frequency in defining habitat suitability for spotted 476 lanternfly in the future was unexpected. Freeze-tolerant ectotherms overwintering above ground, 477 as opposed to those overwintering below ground, often experience less physiologically stressful 478 conditions (Irwin and Lee, 2003). Spotted lanternfly eggs are freeze-tolerant ectotherms; 479 however, they are susceptible to damage from repeated freeze-thaw cycles, which can 480 compromise their viability. A decrease in freeze-thaw cycles may therefore enhance egg survival 481 rates, leading to higher population densities in the spring, but the two regions that experience 482 decreased frost-change-frequency show opposing directions of future suitability (increase in the 483 Hudson Valley and decrease in Long Island). 484 485 The findings in this study emphasize the importance of taking a sub-regional modeling approach 486 to risk assessment. The sub-regional variation that our models identify at a 30-m scale enables 487 policymakers to strategically deploy limited resources to mitigate the potential economic impact 488 of spotted lanternfly on the New York grape industry, both in the immediate time frame as well 489 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted May 31, 2025. ; https://doi.org/10.1101/2025.05.28.656630doi: bioRxiv preprint 26 as in the future. It also gives growers a sense of agency in assessing their own level of risk and 490 engaging in monitoring efforts of their vineyards. 491 492 While the application of insecticides has been shown to effectively reduce spotted lanternfly 493 populations, concerns regarding environmental impacts and effects on non-target species persist 494 (Leach et al., 2019). Risk maps generated from these models can help direct monitoring efforts to 495 vineyards in high-risk zones, focusing pest control in areas with the greatest potential impact, 496 thereby reducing un-targeted pesticide use (Koch et al., 2019). Efforts to remove tree-of-heaven 497 to curb the spread of spotted lanternfly have been successful, though and resource labor intensive 498 (Young, Bell and Morrison 2020). Notably, while high populations of the spotted lanternfly were 499 initially recorded in Pennsylvania, it is anticipated that these populations may be stabilizing as 500 natural predators adapt to feeding on them and resource availability shifts (Johnson et al., 2025). 501 Additionally, regionalized risk assessments enable more efficient economic decision-making, 502 allowing stakeholders to balance short-term intervention costs with the longer-term benefits of 503 preserving grapevine health and productivity, reducing vineyard losses, and ensuring the 504 viability of local viticulture industries (Chapman et al., 2019). By forecasting likely hotspots 505 under a high emissions SSP 585, RCP 8.5 (2041-2070) climate change scenario, growers can 506 prioritize alternative management practices in regions projected to face higher risks (Harper et 507 al., 2019). Climate change forecasting extends the utility of this model under predicted future 508 climate scenarios in the state of New York. The dynamic interaction between the potential and 509 realized niche of invasive species highlights the necessity for adaptive management strategies 510 and ongoing monitoring efforts. 511 512 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted May 31, 2025. ; https://doi.org/10.1101/2025.05.28.656630doi: bioRxiv preprint 27 Though our models provide valuable insights, several limitations should be acknowledged. 513 Under-sampling of both tree-of-heaven and spotted lanternfly in in Northern New York State and 514 oversampling in New York City may have influenced specific distribution predictions, however 515 strong model predictive power and low RMSE temper these concerns somewhat. Future research 516 should concentrate on enhancing sampling methodologies, incorporating finer (e.g., 10-m) high-517 resolution climate data, and investigating the impact of biotic factors, like competition and 518 predation, on species distributions. Another important avenue for exploration involves a more in-519 depth analysis of vineyard-specific risks, since vineyard-specific management strategies may 520 further shift vulnerability. 521 522 The overarching goal of this study was to inform stakeholders and growers about the current 523 habitat suitability and geographical range of spotted lanternfly and its primary invasive host, 524 tree-of-heaven. We also aimed to identify the probable patterns of future niche expansion for 525 both species under high emissions RCP 8.5 climate change scenario. We successfully created 526 high-performing models by integrating climatic and geographic variables to predict potential 527 future species habitat suitability and assess the risk of spotted lanternfly invasion and 528 establishment for New York State vineyards. Our results indicate that the Hudson Valley and the 529 southwestern region of the Finger Lakes are at highest risk of spotted lanternfly establishment in 530 the future and are therefore most likely to experience economic damage due to the spotted 531 lanternfly though habitat suitability are far more restricted in area than expected under a 532 worsening climate scenario in the near-future. This is because niche breadth is defined by many 533 climatic variables which are expected to shift in often opposing directions resulting in less 534 expansion despite warming temperatures. 535 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted May 31, 2025. ; https://doi.org/10.1101/2025.05.28.656630doi: bioRxiv preprint 28 536 It is also important to clarify that risk does not guarantee invasion and that our models highlight 537 potential hotspots where spotted lanternfly is more likely to spread. In these regions, proactive 538 management strategies, like targeted monitoring and early intervention, are essential. Providing 539 grape industry stakeholders with spotted lanternfly invasion risk assessments in the short-term 540 can assist in the strategic allocation of resources and provide more accurate information for 541 proactive management. 542 543 As climate change alters our natural environment, implementing proactive pest management 544 strategies becomes essential. The models presented in this study provide valuable insights that 545 can assist stakeholders, policymakers, and growers in making well-informed decisions regarding 546 resource allocation, pest control, and monitoring approaches. The integration of climate 547 scenarios (SSP 370, RCP 7.0 and SSP 585, RCP 8.5) enhances the assessment of long-term risks 548 by highlighting key variables driving spotted lanternfly niche expansion and contraction. 549 Increased temperatures and precipitation patterns in both scenarios do facilitate habitat suitability 550 for the spotted lanternfly. However, extreme temperatures in the Hudson Valley and Long Island 551 could potentially put stress on populations, reducing habitat suitability. Collectively, these 552 variables suggest that climate change has the potential to contribute to both the expansion and 553 contraction of the spotted lanternfly's niche, emphasizing the need for targeted, sub-region-554 specific management strategies. From both environmental and economic perspectives, this model 555 can be used as a tool to quantify risk in both the immediate timescale as well as the future has the 556 potential to improve the sustainability of the U.S. grape production. 557 558 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted May 31, 2025. ; https://doi.org/10.1101/2025.05.28.656630doi: bioRxiv preprint 29

559 We would like to thank the New York State Department of Agriculture and Markets, especially 560 Michael Formicelli, and USDA APHIS Plant Protection and Quarantine (PPQ) for providing 561 essential datasets for our models. We would like to thank members of the NYSIPM including 562 Brian Eshenaur and Daniel Olmstead, Cornell Cooperative Extension including Hans Walter-563 Peterson, Jennifer Russo, Terry Bates and Jeremy Schuster, and Cornell AgriTech including 564 Gregory Loeb and Stephen Hesler, who provided feedback on interpretation of these findings. 565 Additionally, this work was funded by NASA Acres (80NSSC23M0034) and the New York Wine 566 and Grape Foundation (68141). We are extremely grateful to all collaborators that have 567 facilitated this study. 568 569

570 Bennett, S., C.M. Duarte, N. Marbà, T. Wernberg. (2019) Integrating within-species variation in 571 thermal physiology into climate change. Philosophical Transactions of The Royal Society B, 572 374:1778. https://doi.org/10.1098/rstb.2018.0550 573 Boiko, O., S. Kagone, G. Senay. (2021) Soil properties dataset in the United States. Dataset. 574 USGS. https://doi.org/10.5066/P9TI3IS8 575 Boivin, N.L., M.A. Zeder, D.O. Fuller, M.D. Petraglia. (2016) Ecological consequences of 576 human niche construction: Examining long-term anthropogenic shaping of global species 577 distributions. Proceedings of the National Academy of Sciences, 113:6388-6396. https://doi-578 org.proxy.library.cornell.edu/10.1073/pnas.1525200113 579 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted May 31, 2025. ; https://doi.org/10.1101/2025.05.28.656630doi: bioRxiv preprint 30 Brown, J.J., M. Pascual, M.C. Wimberly, L.R. Johnson, C.C. Murdock. (2023) Humidity – The 580 overlooked variable in the thermal biology of mosquito-borne disease. Ecology letters, 26:1029–581 1049. 582 Chapman, D., O.L. Pescott, H.E. Roy, R. Tanner. (2019) Improving species distribution models 583 for invasive non-native species with biologically informed pseudo-absence selection. Journal of 584 Biogeography, 46:1029-1040. https://doi.org/10.1111/jbi.13555. 585 Chen, J., L. Liu, J. Pei, M. Deng. (2021) An ensemble risk assessment model for urban rainstorm 586 disasters based on Random Forest and Deep Belief Nets: A case study of Nanjing, China. Natural 587 Hazards, 107:2671-2692. https://doi.org/10.1007/s11069-021-04630-y 588 Clark, J., Y . Wang, P.V . August. (2014) Assessing current and projected suitable habitats for tree-589 of-heaven along the Appalachian Trail. Philosophical Transactions of the Royal Society B: 590 Biological Sciences, 369:20130192. https://doi.org/10.1098/rstb.2013.0192. 591 Du, Z., Y . Wu, Z. Chen, et al. (2021) Global Phylogeography and Invasion History of the Spotted 592 Lanternfly Revealed by Mitochondrial Phylogenomics. Evolutionary Applications, 14:915-930. 593 https://doi.org/10.1111/eva.13170 594 Dunham, J. (2022) New York economic impact study of the wine industry. WineAmerica. 595 Retrieved 8/10/2024 from https://wineamerica.org/economic-impact-study/new-york-wine-596 industry/ 597 Elsensohn, J.E., S. Wolford, A. Tabb, T. Leskey. (2024) Experimental evidence supports the 598 ability of spotted lanternfly to hitchhike on vehicle exteriors as a mechanism for anthropogenic 599 dispersal. Royal Society Open Science, 11:240493 https://doi.org/10.1098/rsos.240493 600 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted May 31, 2025. ; https://doi.org/10.1101/2025.05.28.656630doi: bioRxiv preprint 31 Esri (2024) ArcGIS Pro Version 3.3 [Software]. Environmental Systems Research Institute, Inc. 601 Evans, T.G., S.E. Diamond, M.W. Kelly. (2015) Mechanistic species distribution modelling as a 602 link between physiology and conservation. Conservation Physiology, 3:cov056. 603 https://doi.org/10.1093/conphys/cov056 604 Farooque, A., H. Afzaal, R. Benlamri, S. Al-Naemi et al. (2023) Red-green-blue to Normalized 605 Difference Vegetation Index translation: A robust and inexpensive approach for vegetation 606 monitoring using machine vision and generative adversarial networks. Precision Agriculture, 607 2:1097-1115. https://doi.org/10.1007/s11119-023-10001-3 608 Finch, D.M., J.L. Butler, J.B. Runyon, C.J. Fettig, et al. (2021) Effects of climate change on 609 invasive species. In: Poland, T.M., T. Patel-Weynand, D.M. Finch, C.F. Miniat, D.C. Hayes, V .M. 610 Lopez (eds) Invasive Species in Forests and Rangelands of the United States. Springer, Cham. 611 Fryer, J.L. (2010) Ailanthus altissima. In: Fire Effects Information System. U.S. Department of 612 Agriculture, Forest Service, Rocky Mountain Research Station, Fire Sciences Laboratory. 613 https://www.fs.usda.gov/database/feis/plants/tree/ailalt/all.html 614 Ghalambor, C.K., J.K. McKay, S.P. Carroll, D.N. Reznick. (2007) Adaptive versus non-adaptive 615 phenotypic plasticity and the potential for contemporary adaptation in new environments. 616 Functional Ecology, 21:394-407. https://doi.org/10.1111/j.1365-2435.2007.01283.x 617 Gillson, L., T.P. Dawson, S. Jack, M.A. McGeoch. (2013) Accommodating climate change 618 contingencies in conservation strategy. Trends in Ecology & Evolution, 28:135-142 619 https://doi.org/10.1016/j.tree.2012.10.008 620 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted May 31, 2025. ; https://doi.org/10.1101/2025.05.28.656630doi: bioRxiv preprint 32 Gómez-Marco, F. and M.S. Hoddle. (2022) Effects of freezing Lycorma delicatula egg masses 621 on nymph emergence and parasitization by Anastatus orientalis. Frontiers in Insect Science, 622 2:937129. 623 Hao, Z., L. Lin, C.J. Post, E.A. Mikhailova. (2024) Monitoring the spatial-temporal distribution 624 of invasive plant in urban water using deep learning and remote sensing technology. Ecological 625 Indicators, 162:1122061. https://doi.org/10.1016/j.ecolind.2024.112061 626 Harfoot, M.B., A. Johnston, A. Balmford, N.D. Burgess et al. (2021) Using the IUCN Red List to 627 map threats to terrestrial vertebrates at global scale. Nature Ecology and Evolution, 5:1510-1519. 628 https://doi.org/10.1038/s41559-021-01542-9 629 Harper, J.K., W. Stone, T.W. Kelsey, L.F. Kime. (2019) Potential economic impact of the spotted 630 lanternfly on agriculture and forestry in Pennsylvania. Center for Economic and Community 631 Development. Retrieved 8/10/2024 from https://aese.psu.edu/research/centers/cecd 632 /publications/economic-impact/spotted-lanternfly-2019.pdf 633 Harner, A.D., H.L. Leach, L. Briggs, M. Centinari. (2022) Prolonged phloem feeding by the 634 spotted lanternfly, an invasive planthopper, alters resource allocation and inhibits gas exchange 635 in grapevines. Plant Direct, 6:e452. https://doi.org/10.1002/pld3.452 636 Huron, N.A., J.E. Behm, M.R. Helmus. (2022) Paninvasion severity assessment of a U.S. grape 637 pest to disrupt the global wine market. Communications Biology, 5:655. 638 https://doi.org/10.1038/s42003-022-03580-w 639 Irwin, J.T. and R.E. Lee. (2003) Cold winter microenvironments conserve energy and improve 640 overwintering survival and potential fecundity of the goldenrod gall fly, Eurosta solidaginis. 641 Oikos, 100:71-78. https://doi.org/10.1034/j.1600-0706.2003.11738.x 642 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted May 31, 2025. ; https://doi.org/10.1101/2025.05.28.656630doi: bioRxiv preprint 33 Isler, J., H. Bugmann, M. Conedera, T. Thrippleton. (2023) Long-term dynamics of tree-of-643 heaven (Ailanthus altissima) in central European forests. European Journal of Forest Research, 644 142:1149-1166. https://doi.org/10.1007/s10342-023-01582-9 645 Johnson, A.E, S Hermann, K. Hoover (2025) Predation of spotted lanternfly (Lycorma 646 delicatula) by generalist arthropod predators of North America. Arthropod-Plant 647 Interactions.19:31 Pennsylvania State University. https://doi.org/10.1007/s11829-025-10138-0 648 Karger, D.N., O. Conrad, J. Böhner, T. Kawohl et al. (2021) Climatologies at high resolution for 649 the Earth’s land surface areas. EnviDat. https://doi.org/10.16904/envidat.228.v2.1 650 Karger, D.N., M.P. Nobis, S. Norman, C.H. Graham, N.E. Zimmermann. (2023) CHELSA-651 TraCE21k – high-resolution (1 km) downscaled transient temperature and precipitation data 652 since the last glacial maximum. Climate of the Past, 19:439-456. https://doi.org/10.5194/cp-19-653 439-2023 654 Keena, M.A., G. Hamilton, D. Kreitman. (2023) The potential climate range of spotted lanternfly 655 may be broader than previously predicted. Frontiers in Insect Species, 3:1092189. 656 https://doi.org/10.3389/finsc.2023.1092189 657 Keena, M.A. (2024) Effects of temperature on the survival of spotted lanternfly active life stages 658 when held without food. Agricultural and Forest Entomology, 26:366-372. 659 https://doi.org/10.1111/afe.12619 660 Koch, F.H. (2021) Considerations regarding species distribution models for forest insects. 661 Agricultural and Forest Entomology, 23:393-399. https://doi.org/10.1111/afe.12458 662 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted May 31, 2025. ; https://doi.org/10.1101/2025.05.28.656630doi: bioRxiv preprint 34 Laveaga, E., K. Hoover, F.E. Acevedo. (2023) Life history traits of spotted lanternfly 663 (Hemiptera: Fulgoridae) when feeding on grapevines and tree of heaven. Frontiers in Insect 664 Science, 3:1091332. https://doi.org/10.3389/finsc.2023.1091332 665 Lavely, E., L. Iavorivska, O. Uyi, D.M. Eissenstat et al. (2022) Impacts of short-term feeding by 666 spotted lanternfly (Lycorma delicatula) on ecophysiology of young hardwood trees in a common 667 garden. Frontiers in Insect Science, 2:1080124. https://doi.org/10.3389/finsc.2022.1080124 668 Leach, H., D.J. Biddinger, G. Krawczyk, E. Smyers, J.M. Urban. (2019) Evaluation of 669 insecticides for control of the spotted lanternfly, Lycorma delicatula (Hemiptera: Fulgoridae), a 670 new pest of fruit in the Northeastern US. Crop Production, 124:104833. 671 https://doi.org/10.1016/j.cropro.2019.05.027 672 Leach, H. and A. Leach. (2020) Seasonal phenology and activity of spotted lanternfly (Lycorma 673 delicatula) in Eastern US vineyards. Journal of Pest Science, 93:1215-1224. 674 https://doi.org/10.1007/s10340-020-01233-7 675 Leitão, P.J. and M.J. Santos. (2019) Improving models of species ecological niches: A remote 676 sensing overview. Frontiers in Ecology and Evolution, 7:9. 677 https://doi.org/10.3389/fevo.2019.00009 678 Liu, H. (2022) Oviposition selection in spotted lanternfly: Impact of habitat and substrate on egg 679 mass size and hatchability. Frontiers in Insect Science. 2:932433. 680 https://doi.org/10.3389/finsc.2022.932433 681 Liu, Y . and H. Zhao. (2017) Variable importance-weighted random forests. Quantitative Biology, 682 5:338-351. https://doi.org/10.1007/s40084-017-0121-6 683 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted May 31, 2025. ; https://doi.org/10.1101/2025.05.28.656630doi: bioRxiv preprint 35 MacKenzie, C.A. (2015) Summarizing risk using risk measures and risk indices. Risk Analysis, 684 34:2143-2162. http://dx.doi.org/10.1111/risa.12220. 685 Mod, H.K., D. Scherrer, M. Luoto and A. Guisan. (2016). What we use is not what we know: 686 environmental predictors in plant distribution models. Journal of Vegetation Science, 27:1308-687 1322. https://doi.org/10.1111/jvs.12444 688 Mushagalusa, C.A., A.B. Fandohan, R. Glèlè Kakaï. (2024) Random Forest and spatial cross-689 validation performance in predicting species abundance distributions. Environmental Systems 690 Research, 13:23. https://doi.org/10.1186/s40068-024-00352-9 691 Murman, K., G.P. Setliff, C.V . Pugh, M.J. Toolan et al. (2020) Distribution, survival, and 692 development of spotted lanternfly on host plants found in North America. Environmental 693 Entomology, 49:1270-1281. https://doi.org/1093/ee/nvaa126. 694 National Center for Atmospheric Research (2017) WCRP CMIP5: National Center for 695 Atmospheric Research (NCAR) CCSM4 model output collection. Center for Environmental Data 696 Analysis. http://catalogue.ceda.ac.uk/uuid/3982248dd440444ebee68e2ae64a0333 697 Nayak, H.S., J.V . Silva, C.M. Parihar, T.J. Krupnik et al. (2022) Interpretable machine learning 698

to explain on-farm yield variability of high productivity wheat in northwest India. Field 699 Crops Research, 287:108640. https://doi.org/10.1016/j.fcr.2022.108640 700 Oliveira, C.M., A.M. Auad, S.M. Mendes, M.R. Frizzas. (2012) Economic impact of exotic 701 insect pests in Brazilian agriculture. Journal of Applied Entomology. 137:1-15. https://doi-702 org.proxy.library.cornell.edu/10.1111/jen.12018 703 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted May 31, 2025. ; https://doi.org/10.1101/2025.05.28.656630doi: bioRxiv preprint 36 Pacifici, M., W.B. Foden, P. Visconti, J.E.M. Watson et al. (2015) Assessing species vulnerability 704 to climate change. Nature Climate Change, 5:215-224. https://doi.org/10.1038/nclimate2448 705 R Core Team (2024) R: A language and environment for statistical computing. R Foundation for 706 Statistical Computing, Vienna, Austria. Version 4.4.1. https://www.R-project.org/ 707 Seggos, B. (2021) Observed and Projected Climate Change in New York State: An Overview. 708 Department of Environmental Conservation. Accessed March 2025 at 709 https://extapps.dec.ny.giv/docs/administration_pdf/ccnys2021.pdf 710 Soler, J. and J. Izquierdo. (2024) The invasive Ailanthus altissima: A biology, ecology, and 711 control review. Plants, 13:931. https://doi.org/10.3390/plants1307093 712 Soley-Guardia, M.; Alvarado-Serrano, D.F.; Anderson, R.P . (2024) Top ten hazards to avoid 713 when modeling species distributions: a didactic guide of assumptions, problems, and 714 recommendations. Ecography, 4: e06852. https://dio.org/10.1111/ecog.06852 715 Strömbom, D., A. Stands, J.M. Graham, A. Crocker et al. (2024) Modeling human activity-716 related spread of the spotted lanternfly (Lycorma delicatula) in the US. PLOS ONE, 19: 717 e0307754. https://doi.org/10.1371/journal.pone.0307754 718 Thompson, L., S.D. Powers, A. Appolon, P. Hafker et al. (2020) Climate-related geographical 719 variation in performance traits across the invasion front of a widespread non-native insect. 720 Journal of Biogeography, 48:405-414. doi: 10.1111/jbi.14005 721 Turbelin, A. J., B. J. Sinclair, J. Rost, and A. D. Roe (2024) Cold tolerance strategy and lower 722 temperature thresholds of Lycorma delicatula egg masses. Preprint: Research Square. Updated 723 June 19, 2024. https://doi.org/10.21203/rs.3.rs-4601482/v1 724 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted May 31, 2025. ; https://doi.org/10.1101/2025.05.28.656630doi: bioRxiv preprint 37 USDA-NRCS (2024) Web soil survey. United States Department of Agriculture, Natural 725 Resources Conservation Service. Accessed September 1, 2024. 726 https://www.nrcs.usda.gov/resources/data-and-reports/web-soil-survey 727 US Forest Service (2025) Forests of New York, 2023 FIA Annual Snapshot. U.S. Department of 728 Agriculture, Forest Service. https://doi.org/10.2737/FS-RU-595 729 Uyi, O., J.A. Keller, E. Swackhamer, K. Hoover. (2021) Performance and host association of 730 spotted lanternfly (Lycorma delicatula) among common woody ornamentals. Scientific Reports, 731 11:15774. https://doi.org/10.1038/s41598-021-95376-x 732 Valavi, R., J. Elith, J.J. Lahoz-Monfort, G. Guillera-Arroita, D. Warton. (2019) blockCV: An R 733 package for generating spatially or environmentally separated folds for K‐fold cross‐validation 734 of species distribution models. Methods in Ecology and Evolution, 10:225-232. 735 https://doi.org/10.1111/2041-210X.13107 736 Valladares, F., S. Matesanz, F. Guilhaumon, M.B. Araújo et al. (2014) The effects of phenotypic 737 plasticity and local adaptation on forecasts of species range shifts under climate change. Ecology 738 Letters, 17:1351-1364. https://doi.org/10.1111/ele.12348 739 Velazco, S.J.E., M.B. Rose, A.F. Alves de Andrade, I. Minoli, J. Franklin. (2022) flexsdm: An R 740 package for supporting a comprehensive and flexible species distribution modelling workflow. 741

in Ecology and Evolution, 13:1661-1669. https://doi.org/10.1111/2041-210X.13874 742 Venette, R.C., D.J. Kriticos, R.D. Magarey, F.H. Koch et al. (2010) Pest risk maps for invasive 743 alien species: A roadmap for improvement. BioScience, 60:349–362. https://doi-744 org.proxy.library.cornell.edu/10.1525/bio.2010.60.5.5 745 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted May 31, 2025. ; https://doi.org/10.1101/2025.05.28.656630doi: bioRxiv preprint 38 Venter, O., E.W. Sanderson, A. Magrach, J.R. Allan et al. (2018) Last of the wild project, version 746 3 (LWP-3): 2009 Human Footprint, 2018 Release. Palisades, New York: NASA Socioeconomic 747 Data and Applications Center (SEDAC). https://doi.org/10.7927/H46T0JQ4. 748 Vignali, S., A.G. Barras, R. Arlettaz, V. Braunisch. (2020) SDMtune: An R package to tune and 749 evaluate species distribution models. Ecology and Evolution, 10:11488-11506. 750 https://doi.org/10.1002/ece3.6786 751 Walker, G., M. Robertson, M. Gaertner, L. Gallien, D.M. Richardson. (2017) The potential range 752 of Ailanthus altissima (tree of heaven) in South Africa: The roles of climate, land use, and 753 disturbance. Biological Invasions, 19:1-16. https://doi.org/10.1007/s10530-017-1597-8 754 Wang, M. and C. Hu. (2021) Satellite remote sensing of pelagic sargassum macroalgae: The 755 power of high resolution and deep learning. Remote Sensing of Environment, 264:112631. 756 https://doi.org/10.1016/j.rse.2021.112631 757 Xu, H., H. Ding, M. Li, S. Qiang et al. (2006) The distribution and economic losses of alien 758 species invasion to China. Biological Invasions. 8:1495-1500. https://doi.org/ 10.1007/s10530-759 005-5841-2 760 Young, CC., J.C. Bell and L.W. Morrison. (2020) Long-term treatment leads to reduction of tree-761 of-heaven (Ailanthus altissima) populations in Buffalo National River. Invasive Plant Science 762 and Management 13:276-281. doi: 10.1017/inp.2020.28 763 Zhang, Y ., X. Xiao, X. Wu, S. Zhou, G. Zhang, Y. Qin, J. Dong. (2017) A global moderate 764 resolution dataset of gross primary production of vegetation for 2000-2016. Scientific Data, 765 4:170165. https://doi.org/10.1038/sdata.2017.165 766 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted May 31, 2025. ; https://doi.org/10.1101/2025.05.28.656630doi: bioRxiv preprint