Human alterations of the Earth have had, and continue to have, a tremendous impact on ecosystems, with land transformations and climate change considered major drivers of worldwide biodiversity loss (Mooney & Hobbs, 2000; Vitousek et al., 1997). While land transformations have an obvious impact on species distributions through the loss of habitat, the effects associated with climate change may also cause shifts in species' ranges, which increases the risk of invasive species (Bertelsmeier et al., 2015). Global trade shuffles species composition by introducing new taxa into now disturbed ecosystems (Hulme, 2009; Levine & D'Antonio, 2003; Marco & Santini, 2015; Meyerson & Mooney, 2007; Seebens et al., 2017). Climate change-driven range expansion of invasive species (Walther et al., 2009) may exacerbate impacts on local ecosystems and tremendously increase the costs associated with conservation, agriculture, forestry, and fisheries (reviewed in Vilà et al., 2010).
Many ants are particularly problematic invaders that can rapidly spread into an area and negatively impact native species, agriculture, and human health (Lessard et al., 2009; Lowe et al., 2000). Native to central South America, the red imported fire ant (RIFA), Solenopsis invicta (Buren), is a notorious invasive species that has become established in many parts of the world, including Central and North America, the Caribbean, East Asia, Australia, and New Zealand (Chen et al., 2020; Sung et al., 2018; Sutherst & Maywald, 2005). RIFA was introduced in the United States through the Port of Mobile, Alabama, in the 1930s (Buren et al., 1974). From there, colonies expanded rapidly throughout the southern United States (Callcott & Collins, 1996; Sung et al., 2018). In 2000, the US Department of Agriculture estimated that RIFA was expanding by approximately 193 km (120 miles) per year (Kemp et al., 2000). At present, RIFA occurs throughout the southern United States, from Florida in the east through Texas and more sporadically in the Southwest where it has been reported from New Mexico, Arizona, and California. The northern limits of its introduced North American range extend into Missouri in the Midwest and southern Virginia in the east (
In its introduced range, RIFA favors disturbed sites including public lands, businesses, and recreational areas such as lawns, gardens, urban and suburban parks, as well as roadsides and agricultural fields and pastures (Gutrich et al., 2007; King & Tschinkel, 2008; Porter & Tschinkel, 1993; Rosson, 2004; Tschinkel, 2006). Numerous negative impacts of RIFA on native biota have been the subject of many studies and are relatively well documented (see review in Tschinkel, 2006). To name a few examples, imported fire ants alter soil chemistry (Lafleur et al., 2005; Lockaby & Adams, 1985), inflict damage to crop seeds (Morrison et al., 1997), decrease seed dispersal (Ness, 2004), negatively affect native ants (Hook & Porter, 1990; Kaspari, 2000), reduce foraging in small mammals (Orrock & Danielson, 2004), and depredate bird nests (Stake & Cimprich, 2003). The painful sting of this aggressive species is known for its negative effects on wildlife, livestock, and humans (Langkilde, 2009; Vinson, 2013; Wang et al., 2019). In agricultural systems, S. invicta has been shown to impact farm equipment through wear on blades (qualitatively described in Vinson, 2013) and to reduce crop yields by as much as 33% in RIFA-infested soybean plots (Adams et al., 1983). Consequently, in the United States, RIFA is a major ecological and socioeconomic problem causing an estimated US $6 billion economic impact annually (Drees & Lard, 2006).
While some human activities are heavily affected by the presence of RIFA, others such as trade and transportation are also major driving forces behind its spread (King & Tschinkel, 2008). In addition to human-assisted dispersal, temperature (Morrill, 1977; Porter, 1988), precipitation (Thorvilson et al., 1992; Wojcik, 1983), and vegetation type (Rosson, 2004) also influence the areas suitable for RIFA establishment and continued persistence. Temperature and precipitation have often been reported as the main constraints limiting RIFA spread (Chen et al., 2020; Morrison et al., 2005). It is generally assumed that the drier conditions in the west and the colder temperatures in the north have slowed its spread (Vinson, 1997), although urbanization may facilitate the spread of RIFA in colder climates where ants can overwinter in electrical equipment and insulation (Thorvilson et al., 1992). Due to RIFA's high reproductive and adaptive capacity (Ross & Shoemaker, 2008), global climate change, and current urbanization trends, its range is predicted to continue to expand (Morrison et al., 2005).
Managing invasive species that have become established is extremely difficult (Hoffmann et al., 2011). Key factors in combating further spread of invasive species include early detection, quarantines, mapping, and prediction of areas of future spread. These factors are not mutually exclusive, as, for example, detection can be used in mapping and in crafting policy decisions such as establishing quarantines. Early detection and subsequent limit on spread are among the most important mitigation tools in slowing the spread of invasive species (Lodge et al., 2006). The Federal Imported Fire Ant Quarantine was established in Virginia in 2009 and included the independent cites of Chesapeake, Hampton, Newport News, Norfolk, Poquoson, Portsmouth, Suffolk, Virginia Beach, and Williamsburg and the counties of James City and York. In 2019, the independent cities of Emporia and Franklin and the counties of Brunswick, Isle of Wight, Greensville, Mecklenburg, and Southampton were added to the quarantine zone (Miller & Allen, 2019). The quarantine restricts the movement of soil, plant products, and equipment used for moving soil and other items that are capable of harboring ants and requires the inspection of such items before transportation (APHIS, 2019).
Mapping and modeling species distributions is another important tool for tracking and managing the spread of invasive species and predicting the impacts of climate change on their distributions (Miller, 2010). Previous models have shown that the range of RIFA is likely to continue to expand (Chen et al., 2020; Korzukhin et al., 2001; Morrison et al., 2005; Sung et al., 2018; Sutherst & Maywald, 2005). These models, however, have not been validated after publication, so we do not know how well they reflect the current expansion of RIFA in Virginia and elsewhere.
In this study, we investigated the current spread of S. invicta in Virginia, situated along the northernmost edge of RIFA's introduced North American range, and qualitatively compared it with an earlier predictive model (Morrison et al., 2005). In addition, we constructed a species distribution model (SDM) and explored the effects of climate on RIFA range expansion in North America. We complemented existing opportunistic occurrence data, collected by the Virginia Department of Agriculture and Consumer Services (VDACS), with our own systematic presence/absence surveys. Our efforts allowed us to examine the effects of climate conditions on recent RIFA expansion trends. These data support the importance of mapping and modeling species distributions for the management of invasive species and can provide further insight into the effects of global climate change on RIFA spread.
MATERIALS AND METHODS Current range of the RIFA in Virginia Roadside surveysDuring the summers of 2020 and 2021, we conducted 175 visual surveys along public roadways in southern Virginia. The survey methods were adapted from the United States Department of Agriculture “Imported Fire Ant Program Manual” (APHIS, 2019).
We selected 23 roadways that extended approximately 40 km perpendicular to the current Imported Fire Ant Quarantine boundaries. Roadways were selected using ArcGIS Online (ESRI, 2020) and were situated approximately 16 km (10 miles) apart from each other. During each survey, we stopped every 8 km (5 miles) along each roadway (median of 6 stops per roadway in 2020; total of 125 stops) and looked for evidence of RIFA presence. Due to the orientation of some roadways, some stops overlapped, and these were sampled only once. All sampled roadways are presented in Figure 1. In 2021, we limited the roadway surveys to within 16 km (10 miles) of known RIFA localities. During data validation in QGIS (QGIS.org, 2022), the coordinates of a single road sampling point were found to be located in North Carolina by approximately 100 m. This point was assigned to the neighboring Halifax County, Virginia, because we assumed this to be due to an inaccurate GPS measurement.
Enlarge this image.FIGURE 1. Map of 2020–2021 Virginia roadside surveys. Red imported fire ant quarantined counties (as of 2019) are shaded in gray, roadside survey transects are represented by black lines, and roadside survey locations are represented by open circles. All locations were sampled in 2020. Locations sampled in 2021 are indicated with an additional, wider, concentric circle.
The survey region is one of Virginia's most heavily populated areas. Land cover primarily consists of urban and suburban sprawl, cropland, and other disturbed areas intermixed with wetlands and remaining tracts of mixed second-growth woodlands. Surveys were conducted during daylight hours when air temperatures were between 18 and 32°C. Before surveying each stop, we examined its suitability using Google Maps. A location was deemed unsuitable if there was construction, if the location had no safe turnoff, or if it was located on private property. If a site was deemed unsuitable, we choose a new site within 1 km of the previous location. Exact stopping locations were determined at the time of the survey and were adjusted for parking and safety of the surveyor (MM). Roadways were grouped by convenience and randomly selected in advance via a random number generator. At each stop, we walked 0.4 km (0.25 miles) along each side of the road and visually surveyed for mounds, small piles of exposed soil, or other soil disturbances (APHIS, 2019). The characteristic RIFA mounds are readily distinguished from those of other local species by the absence of visible entrances. Upon encountering a mound, we disturbed the top layer of the soil with a stick or a shovel and confirmed the identity of the ants rushing out of the nest based on morphological characteristics (Collins & Scheffrahn, 2001).
Reports from the VDACSIn addition to roadside surveys, we used multiyear infestation reports from the VDACS covering the period 2016–2020. These reports focused on areas outside of the Imported Fire Ant Quarantine zone. VDACS provided coordinates with 0.402-km (0.25 miles) accuracy to obscure sensitive information. This approach resulted in a few points, located along the North Carolina border, falling outside of Virginia's borders (by 10–200 m). These points were assigned to the neighboring Halifax County, Virginia.
Other data sourcesBetween January 2020 and October 2021, we used information from citizen science organizations and other institutions to gather additional RIFA occurrence data. Citizen science data were provided by participating individuals from the Virginia Master Naturalist program or were harvested from research grade iNaturalist records. iNaturalist is a community science platform for identifying, mapping, and sharing biodiversity observations worldwide. Additionally, we surveyed county Cooperative Extension offices, and while no specific locations were recorded through this approach, we were able to gain a better understanding of RIFA's spread in the area. Lastly, presence/absence data were also gathered from an ongoing ant diversity project conducted by the Virginia Museum of Natural History (KI). This dataset contained a data point from Giles County from 1990 which was deemed to be based on a “hitchhiker” and was excluded from subsequent analyses.
Qualitative comparison with a previous predictive modelWe compared the current S. invicta range in Virginia with an earlier predictive model (Morrison et al., 2005) by overlaying our occurrence data with the model's maps and visually comparing the two distributions. We used geographic landmarks to align the two maps as accurately as possible and utilized the same 0.5° × 0.5° cell grid (approx. 55 × 55 km) used in the Morrison et al. (2005) model. We then qualitatively compared the differences between the model's predictions and our data by identifying the areas where RIFA was predicted to be present at different points in time. We annotated our occurrence observations with the time that Morrison et al. (2005) predicted RIFA to be present at these locations. Because the original data and model are no longer available (Morrison, personal communication), a more quantitative comparison of current observations and model predictions was not possible.
Prediction of habitable range with current and future climate predictionsTo create our SDM, we harvested worldwide S. invicta presence data over all available time points from the Global Biodiversity Information Facility (GBIF), a free open access source of global biodiversity data (
We tested if the distribution of the presence/pseudo-absence data could be explained by environmental variables. We used the bioclimate variable dataset from WorldClim, which offers free high spatial resolution data of global weather and climate for past observations and predicted future conditions (Fick & Hijmans, 2017). The dataset includes 19 bioclimate variables. Based on Sung et al. (2018), we selected a priori five variables most likely associated with S. invicta spread: annual mean temperature (Bio1), maximum temperature of the warmest month (Bio5), minimum temperature of the coldest month (Bio6), annual precipitation (Bio12), and precipitation of the driest month (Bio14). Based on previous work, Sung et al. (2018) argue that fire ants are vulnerable to minimum and maximum temperatures in winter and summer, respectively (Jemal & Hugh-Jones, 1993; Peterson & Nakazawa, 2007), and that they are somewhat sensitive to annual precipitation (Ward, 2009; cited in Sung et al., 2018).
All temperatures were given in 0.1°C and all precipitation in millimeter. To get more meaningful odds ratio estimates, we divided the temperatures by 10 in the model (changing the scale from 0.1 to 1°C), annual precipitation by 1000 (changing the scale to meters), and precipitation of the driest month by 100 (changing the scale to decimeters). We used historical (current) data at a spatial resolution of 30 s or approximately 1 km2. We used future climate data CMIP5 at a resolution of 2.5 min for 2050 and 2070 to make predictions of RIFA spread based on expected environmental variables and our model trained on current data. The data we used were future IPPC5 climate projections from global climate models (GCMs) for the stabilization scenario representative concentration pathways (RCP 4.5).
The methods and code for our SDM follow Zurell et al. (2018), with a few revisions to adapt to our data using R version 4.3.0 (R Core Team, 2020). The SDM was constructed to model the potential distribution of RIFA under historical and future climate conditions. For model fitting, we used a binomial generalized linear model (GLM) available through the mecofun package (Zurell, 2020) and trained the model on all US data available through GBIF. We entered each of the climatic variables as a linear predictor and a quadratic predictor and set the two-sided significance level to α < 0.05. We removed the only nonsignificant quadratic term (maximum temperature of the warmest month) from the final model. After training the model, we extracted the odds ratios by using the exponential function on estimated coefficients. These odds ratios indicate the relative increase in the odds of occurrence for one unit increase in the predictor. We then qualitatively evaluated its accuracy by projecting the distribution of RIFA under current climate conditions. We also assessed the model fit more formally by splitting the occurrence and pseudo-absence dataset into a training and a test set to calculate the area under the curve (AUC). Similar to Sung et al. (2018), our AUCs were in the range between 0.939 and 0.949, indicating very good prediction of presence and absence. Once the SDM was fitted and assessed, we projected future spread over geographic space.
RESULTS Current range ofIn the period 2020–2021, we recorded a total of 158 confirmed RIFA occurrences. From the 2020 roadway surveys, we found RIFA present at 9 of the 125 sites surveyed, 8 of these sites occurred at the site closest to the starting point (within 8 km of the Imported Fire Ant Quarantine zone). From the 2021 roadway surveys, we found RIFA present at 6 of the 50 sites surveyed, 5 of these sites occurred at the site closest to the starting point. The number of observations reported from VDACS increased every year since 2016: 6 reports in 2016, 17 reports in 2017, 65 reports in 2018, 71 reports in 2019, and 113 reports in 2020.
Prior to 2020, no RIFA observations were recorded in Virginia through iNaturalist. Between January 2020 and October 2021, there were 25 observations reported through iNaturalist. A total of seven observations were reported through the Virginia Museum of Natural History dataset, all of which before 2020. Additional five observations in the period 2020–2021 were collected through other sources. We also recorded a total of 605 absence points from VDACS and the roadside surveys.
Based on these data, 113 of the total 158 RIFA occurrences were reported outside of the quarantine boundaries in 2020 and 2021. Most of these observations were in Halifax (36.3%), Lunenburg (32.7%), and Sussex (15.9%) counties, with fewer observations in Charlotte (11.5%), Dinwiddie (2.7%), and Chesterfield (0.9%) counties. Prior to 2020, VDACS reported a few RIFA observations in the City of Richmond area (Figure 2); however, these observations appear isolated and were not confirmed during our surveys.
Enlarge this image.FIGURE 2. Current spread of Solenopsis invicta (Buren) in Virginia. Occurrence records are color-coded based on predictions by Morrison et al. (2005) and show that fire ants have spread faster than anticipated: current range (yellow), 2040–2049 range (orange), and 2080–2089 range (red). Isolated records from the City of Roanoke, and Giles and Montgomery counties (see text) are not shown.
Both westward and northward expansions in Virginia are greater than those expected 15 years ago (Morrison et al., 2005; Figure 2). For example, by 2021, the range of RIFA in Virginia had expanded to about 275 km (five cells in Morrison et al., 2005) further than that predicted, including areas in the range predicted for 2080–2089. Along the North Carolina border, expansion has increased by approximately 128 km (two to three cells) past the range expected by Morrison et al. (2005). Northward expansion is approximately 55 km (one cell) past the expected range from the Morrison et al. (2005) model (Figure 2).
Our SDM is based on GBIF occurrence data and generated pseudo-absence data (Figure 3). Our results indicate that four of the five bioclimatic variables we selected a priori for the model building (i.e., annual mean temperature, minimum temperature, average precipitation, and minimum precipitation) are significant predictors of RIFA presence. All terms, except for maximum temperature of the warmest month, had a significant quadratic term, whereas maximum temperature was a significant linear predictor of RIFA presence (Table 1).
Enlarge this image.FIGURE 3. Model predictions of current Solenopsis invicta (Buren) relative probability of occurrence in the United States on a continuous scale from 0 to 1. Dark green represents high-risk areas, yellow represents mid-risk areas, orange represents low-risk areas, and gray represents no risk areas. Semitransparent black crosses represent red imported fire ant occurrence data from Global Biodiversity Information Facility. Note that the model relies on pseudo-absence data distributed across the United States (not shown).
TABLE 1 Red imported fire ant species distribution model coefficients with associated 95% CIs and probability values.
| Predictor variable | Odds ratio (95% CI) | p |
| Annual mean temperature (in °C) | 36.1356 (23.2824–56.8141) | <0.001* |
| Squared annual mean temperature (in °C2) | 0.9073 (0.8960–0.9184) | <0.001* |
| Maximum temperature of the warmest month (in °C) | 0.8733 (0.8200–0.9294) | <0.001* |
| Minimum temperature of the coldest month (in °C) | 1.3422 (1.2648–1.4237) | <0.001* |
| Squared minimum temperature of the coldest month (in °C2) | 1.0095 (1.0050–1.0143) | <0.001* |
| Annual precipitation (in m) | 1.1943 (1.0249–1.3911) | 0.023* |
| Squared annual precipitation (in m2) | 0.9728 (0.9668–0.9789) | <0.001* |
| Precipitation of the driest month (in dm) | 3.9276 (3.2540–4.7512) | <0.001* |
| Squared precipitation of the driest month (in dm2) | 0.9249 (0.9114–0.9386) | <0.001* |
Note: The asterisks denote significant values at p < 0.05.
Because of the presence of both significant linear and quadratic terms, the partial response curves provided a better representation of how these variables are related to the probability of RIFA occurrence compared to the SDM coefficients (Table 1; Figure 4). The odds ratio for the linear term of annual mean temperature suggests an increase in the odds of occurrence by almost 36.1-fold for each increase in degrees Celsius, which is then curtailed by a decrease in odds through the quadratic term for annual mean temperature, where each increase in degrees Celsius squared reduces the odds 1.102-fold (Table 1). Together, this means that if the annual mean temperature is below 10°C or above 25°C, our model predicts that RIFA cannot persist in an area (Figure 4A). The quadratic term for maximum temperature during the warmest month was nonsignificant and was removed from the final model. For each degree Celsius increase in maximum temperature, there was a 1.145-fold decrease in the odds (Table 1; Figure 4B). Minimum temperature of the coldest month below −10°C also restricts RIFA range (Figure 4C), and with every degree Celsius increase, the odds of occurrence increase approximately 1.342-fold (Table 1). While the quadratic term is significant, the magnitude is so small that it does not seem to affect noticeably the probability of ants in the relevant ranges (Table 1; Figure 4C). Similar to mean temperature, annual precipitation and precipitation of the driest month show a significant quadratic term. While an increase of 1 m in annual precipitation leads to a 1.194-fold increase in odds of ant presence, this is countered by a 1.028-fold decrease for 1-m2 increase of annual precipitation, or a 3.928-fold increase in odds for each decimeter precipitation of the driest month countered by a 0.925-fold decrease for each decimeter squared, respectively (Table 1). For the data range considered herein, this means that annual precipitation is limiting above approximately 1500 mm (Figure 4D), whereas precipitation during the driest month is limiting toward both the high- and low-end values with an optimum around 90 mm (Figure 4E).
Enlarge this image.FIGURE 4. Partial response curves showing red imported fire ant relative occurrence probability.
Based on model parameters, we made predictions regarding the spread of S. invicta under current and future conditions (Figure 5). The prediction of current spread across the continental United States is reflective of the training data on a larger scale (Figure 3), except west of the Mississippi River where our model predicts suitable conditions, but there is a gap in empirical occurrence data (Figure 5A). The habitable area of RIFA in the southeastern United States is likely to increase by 2050 and 2070, especially through the lower Midwest. Westward expansion through Texas and parts of the Southwest is less likely (Figure 5B,C). In Virginia, high-risk invasion areas are mostly located along the border with North Carolina, along Virginia's Coastal Plain, and the Delmarva Peninsula (Figure 5D). Lower suitability areas are currently situated through much of central and western Virginia. While most of the Piedmont region will be at risk of invasion by 2050 (Figure 5E), by 2070, RIFA might find suitable climatic conditions even as far west as the City of Roanoke (Figure 5F). Most of the western, mountainous part of the state is predicted to be unsuitable for RIFA establishment under current climatic conditions (Figure 5D), and the Blue Ridge Mountains may remain unsuitable (Figure 5E,F).
Enlarge this image.FIGURE 5. Relative probability of occurrence of Solenopsis invicta (Buren) spread in (A–C) the United States and (D–F) Virginia on a continuous scale from 0 to 1 (dark green represents high-risk areas, yellow represents mid-risk areas, orange represents low-risk areas, and gray represents no risk areas): (A, D) current possible range, (B, E) future possible range for 2050, and (C, F) future possible range for 2070. All predictions are based on WorldClim data.
We provide new data on the distribution of the invasive S. invicta in Virginia, along one of the current expansion fronts of this species in the United States. We qualitatively compared this distribution with a frequently cited model (Morrison et al., 2005) and demonstrated an unexpected, rapid spread of RIFA beyond previous predictions. Furthermore, we fitted a new SDM on a continental scale to GBIF fire ant occurrence data and confirmed a set of bioclimatic variables most strongly predictive of RIFA presence. We predict that the rapid RIFA spread may continue into suitable habitat and that, in a future under a warming climate, the current boundaries may be further permeated. Overall, these data demonstrate the necessity of sound SDMs to inform appropriate management decision and prevention efforts.
Between 2016 and 2020, RIFA observations have expanded westward into Virginia's Halifax County. In 2020–2021, expansion continued further west into Pittsylvania County while northward expansion continued into Charlotte, Lunenburg, Dinwiddie, Sussex, and Chesterfield counties. While most RIFA observations were located just outside the 2019 quarantine boundaries, isolated observations have been made as far north as the City of Richmond. In summary, new RIFA occurrences were found in seven counties (Charlotte, Chesterfield, Dinwiddie, Halifax, Lunenburg, Pittsylvania, and Sussex) beyond the quarantine zone. Comparing these results with predictions from Morrison et al. (2005), by 2021, expansion had increased by five cells (each cell corresponds to approx. 55 km), including areas beyond the predicted 2040–2049 range. Our results show that both westward and northward RIFA expansions in Virginia are greater than those expected 15 years ago, including expansion into areas predicted to be occupied by S. invicta only by 2080–2089 (Morrison et al., 2005). As a result, wildlife, land managers, farmers, and the general public in these areas are likely to be affected much earlier than expected.
We used a newly fitted SDM to estimate the potential distribution of RIFA under current conditions and future climate projections. Comparing the model's results with occurrence data shows that (1) qualitatively the RIFA range in Virginia is well predicted along the North Carolina border, (2) RIFA is not occupying the currently suitable habitat in its entirety, and (3) the suitable habitat for RIFA is expected to increase under future climate change scenarios, especially if the current trajectory of climate change does not abate. Based on these results, RIFA is likely to continue to expand its range northward along the Atlantic coast and further north into the Midwest (Figure 5).
Our model differs from many of the earlier RIFA colony growth and CLIMEX models in that it uses a logistic regression on presence/pseudo-absence data with climatic variables as predictors, instead of modeling ant ecophysiology (Killion & Grant, 1995; Korzukhin et al., 2001; Morrison et al., 2005; Pimm & Bartell, 1980; Stoker et al., 1994; Sutherst & Maywald, 2005). A previous study, using a slightly different approach for fitting presence/pseudo-absence data, found that such models performed well in predicting the spread of invasive ant species in New Zealand (Ward, 2006). However, our model also shows the limits of SDMs and their predictions based on presence and pseudo-absence data. A potential reason why there are no RIFA records from the Delmarva Peninsula, even though it is predicted to be highly suitable for S. invicta establishment, could be that the Chesapeake Bay acts as a barrier to dispersal (Figure 5). Such geographical barriers would not be captured in a model solely relying on environmental predictors. A spatially explicit model, taking into account dispersal patterns, connectivity and mode of spread, may better predict future range expansions. In this respect, previous modeling attempts that link fire ant life history and ecology to climatic variables could be expanded to include the human component (i.e., transportation of ants to new places; Killion & Grant, 1995; Korzukhin et al., 2001; Morrison et al., 2005; Pimm & Bartell, 1980; Stoker et al., 1994).
S. invicta occurrence is likely to vary from year to year based on changes in temperature (Morrill, 1977), moisture (although Sutherst & Maywald, 2005 suggest that moisture is not a limiting factor), and land management practices (Summerlin et al., 1976). For example, an unusually cold winter or unusually dry summer could lower ant survival (James et al., 2002; Morrill, 1977) and temporarily alter a species' range. Land management practices such as repeated mowing and treating RIFA mounds with insecticides may cause ants to relocate though recolonization, a relatively quick process (Summerlin et al., 1976). In this regard, the recent rapid expansion we found could reflect current weather patterns and management practices rather than climatic conditions. However, it is important to note that our SDM was trained on US data (excluding our prospective Virginia data) and was able to predict the current range we observed better than a previous model (Morrison et al., 2005). Our model is in line with the models by Korzukhin et al. (2001) and Sutherst and Maywald (2005), although winter kill was predicted to be high and ecological suitability to be low in areas where our data indicate fire ants now to be firmly present.
The discrepancy between our data and the predictions made by the model of Morrison et al. (2005) demonstrates that their model either (1) underestimated the rate of change in global climate, (2) did not adequately capture the climatic factors affecting RIFA spread, or (3) the introduced RIFA populations have been freed of earlier climatic constraints through rapid adaptation (Kosmala et al., 2018; Phillips et al., 2006; Shine et al., 2011). We cannot conclusively select among these options because the model from Morrison et al. (2005) was not available for examination. However, their use of an ecophysiological approach suggests points 1 or 3, whereas the saltatory nature of the expansion with stasis followed by a rapid expansion (Callcott & Collins, 1996; see our video from those data at
Based on our results, we should continue to monitor and put in place measures to reduce the chance of RIFA establishment, especially along its invasion front in Virginia. In the future, we might expect RIFA to spread into Kentucky, and even Indiana (Figure 5B,C). On the other hand, it is unlikely that West Virginia will be colonized (Figure 5B,C), presumably because of the Appalachian Mountains and the associated colder temperatures. Continued monitoring is needed for early detection of newly invaded areas, which is in turn critical to managing the spread of invasive species, including RIFA. Comparison of our empirical data with the model of Morrison et al. (2005) shows that validating SDMs is as important as creating them and that models predicting RIFA expansion should be continually validated and, if necessary, updated as the expansion continues. It is therefore important that model results are open and can be readily used in a more formal comparison with real world data than what we have done here (e.g., using AUC; Ward, 2006).
There is no reason to assume that the negative impacts of RIFA (e.g., Hook & Porter, 1990; Kaspari, 2000; Morrison et al., 1997; Ness, 2004; Orrock & Danielson, 2004; Stake & Cimprich, 2003) will be different in the newly invaded areas. Accurate models can be an important tool in the management of invasive species and the mitigation of their negative impacts. For example, VDACS' efforts focus on limiting the spread of RIFA beyond the quarantine boundaries and on preventing new infestations across Virginia, goals that are likely in common with other regional Departments of Agriculture. Importantly, early detection is key for reducing the spread and for eradicating newly established colonies (Lodge et al., 2006), and proactive surveillance such as delimiting and roadside surveys, as we have done here, are rare in invasive ant monitoring. Most early detection, at least in Virginia, relies on public reporting. Model accuracy could help optimize the management of RIFA by guiding surveying efforts (Yemshanov et al., 2017). This in turn could help slow down the expansion. Ultimately, the management of invasive species is difficult, and complete eradications of established, invasive species are extremely rare (Hoffmann et al., 2011).
In conclusion, our study provides an insight into the range expansion of S. invicta at the border of its suitable US habitat and confirms some of the climatic factors associated with its current and future spread. More broadly, these data provide good examples of the effects of changing climate conditions on invasive species expansion and the necessity of frequent model validation of the prediction and therefore management of invasive species' spread and establishment.
AUTHOR CONTRIBUTIONSAll the authors contributed to the study conception and design. Data collection was performed by Morgan Malone and Kaloyan Ivanov. Data analysis was performed by Morgan Malone and Roger Schürch. The first draft of the manuscript was written by Morgan Malone, and all the authors commented on previous versions of the manuscript. All the authors read and approved the final manuscript.
ACKNOWLEDGMENTSWe are grateful to Dr. Sean Malone for his expertise, resources, and continuous support in the field and the Tidewater AREC staff, especially Austin Branch and the rest of the Entomology Team for their support in the field. We thank everyone who helped in gathering fire ant observation data including Eric Day, Tina Macintyre, Cynthia Gregg, the VDACS Plant Protection Inspectors (Noland Hudson, Cindy Hubbard, Joshua Martin, Jessica Driver, Mohamed Abdalla, and Brenda Johnson-Asnicar), the Virginia Museum of Natural History, and the Virginia Master Naturalists. We are also grateful to Dr. Margaret Jane Couvillon for comments on an earlier draft of this manuscript. Lastly, we would like to thank two anonymous reviewers for Ecosphere for their suggestions to improve the manuscript.
FUNDING INFORMATIONThis work was supported by the USDA National Institute of Food and Agriculture, Hatch project VA-160129 to Roger Schürch.
CONFLICT OF INTEREST STATEMENTThe authors declare no conflicts of interest.
DATA AVAILABILITY STATEMENTThe data collected for this project as well as de-identified data provided by the Virginia Department of Agriculture and Consumer Services (VDACS) and the data from the Virginia Museum of Natural History are archived at Virginia Tech's public data repository (Malone et al., 2023;
© 2023. This work is published under http://creativecommons.org/licenses/by/4.0/ (the “License”). Notwithstanding the ProQuest Terms and Conditions, you may use this content in accordance with the terms of the License.
Details
1 Department of Entomology, Virginia Polytechnic Institute and State University, Blacksburg, Virginia, USA
2 Department of Recent Invertebrates, Virginia Museum of Natural History, Martinsville, Virginia, USA
3 Tidewater Agricultural Research and Extension Center, Virginia Polytechnic Institute and State University, Suffolk, Virginia, USA