The models predict widespread environmental suitability for Ae. aegypti in both the eastern and western United States (Fig. 1a), with the highest degree of suitability across the south and lower suitability along the northern margins of the range. Counties classified as suitable for Ae. albopictus were primarily in the eastern part of the country, though areas of suitability were indicated also along the west coast. The lower predicted suitability for Ae. albopictus in western counties may be partly due to the outsized influence of presence observations from eastern counties used for model fitting (Fig. 1b). The number (percentage) of counties classified as environmentally suitable, based on models with 90 or 99% sensitivity, ranged from 1,443 (46%) to 2,209 (71%) for Ae. aegypti and from 1,726 (55%) to 2,329 (75%) for Ae. albopictus.

Based on models with 99% sensitivity, a total of 2,146 counties (69%) were predicted to be suitable for both species, and 2,392 counties (77%) were predicted to be suitable for at least one species. Some counties, primarily in the Northeast, upper Midwest and Pacific Northwest were predicted to be suitable for Ae. albopictus but not Ae. aegypti. Conversely, several counties, primarily in the Southwest and California were predicted to be suitable for Ae. aegypti but not Ae. albopictus. In total, 719 counties (23%), mainly in the upper Midwest, northern Great Plains and Mountain West, were predicted as not suitable for either species.

The predictive model for Ae. aegypti had an AUC of 0.84. Among the variables considered, the best model for Ae. aegypti was based on two variables: winter (December–February) cumulative GDDs >10°C and maximum temperature during the warmest month (Bio5). Winter cumulative GDDs >10°C was the more important variable (permutation importance = 94.3%) but the maximum temperature during the warmest month also contributed significantly (permutation importance = 5.7%). The response curve indicated that any county in which winter GDDs were more than zero had moderate–to-high suitability (Fig. 2a). In a practical sense, this result can be interpreted as meaning that a county is predicted to be suitable for Ae. aegypti if at least 1 d, on average, during the 90-d December–February period has an average temperature exceeding 10°C. Suitability increases between 0 and 700 GDD, and then levels off for higher values, which are not common (areas with 700 or more GDDs in winter are mainly in south Texas and Florida). Suitability increased with increasing maximum temperatures from 15 to 36°C during the warmest month (i.e., mid-summer) but decreased as temperatures increased thereafter (Fig. 2b).

The AUC for the Ae. albopictus model was 0.85. Two predictors contributed to the distribution of Ae. albopictus, winter (December–February) cumulative GDDs >10°C (52.7%) and precipitation in the driest month (BIO14; 47.3%). The response curve for winter GDDs for Ae. albopictus was similar to the response curve described above for Ae. aegypti: any county in which average winter GDDs were more than zero had high suitability (probability > 0.50) (Fig. 2c). Suitability increased nearly linearly as average precipitation in the driest month increased beyond 10 mm (Fig. 2d).

Using county-level presence records for Ae. aegypti and Ae. albopictus from 1960 to 2016, we developed environmental suitability maps for these two species across the contiguous United States as a function of climatic variables. Cumulative GDDs during the winter months, an indicator of overall warmth, were the most important predictive variable for both species and was positively associated with suitability. For Ae. aegypti, summer temperatures also had an overall positive influence on the likelihood of a county being classified as suitable, but contributed substantially less to the model compared with winter GDDs. For Ae. albopictus, total rainfall during the driest month was also a significant predictor of suitability with a permutation importance only ~5% less than GDDs. The fact that GDDs are the most important predictor in both models is a noteworthy result, because the response curves for GDDs indicate that the probability of presence for both species strongly increases if at least 1 d, on average, during the 90-d December–February period has an average temperature exceeding 10°C. Based on our models, we estimate that up to 77% of counties in the contiguous United States are suitable under current climate conditions for at least one of these medically important vector species to survive and reproduce at least during the warm part of the year.

A recent survey of mosquito records from 1995 through 2016 revealed that Ae. aegypti and Ae. albopictus have been reported from 220 and 1,368 counties, respectively (Hahn et al. 2017). However, recognizing that vector surveillance was not conducted in all counties or with equal intensity across counties where surveillance occurred, and that surveillance methods often differ among counties, the reported records likely underestimate the actual number of counties where either species currently exists (Hahn et al. 2017). Indeed, historical presence records dating back to 1960 for Ae. aegypti and the introduction event in 1985 for Ae. albopictus (Fig. 1 and Table S1 [online only]) show a broader geographic distribution than what is revealed by more contemporary collection records from 1995 onwards (Hahn et al. 2017). Although this change for Ae. aegypti may have resulted in large part from improved sanitation and decreased residential water storage over the last 50 years, the older records underscore the need for vigilance as they clearly indicate climatic suitability for the mosquito. Competition with other, more recently introduced mosquito species that exploit containers as larval development sites, including satyrization from Ae. albopictus (Bargielowski et al. 2013, 2015), could also impact local populations of Ae. aegypti (O’Meara et al. 1995, Juliano et al. 2004, Lounibos et al. 2010) and thus may be a confounder in climate-only based models.

The maps presented here identify counties that may be suitable for either species but where recent collection records are lacking. Specifically, our models indicate that, depending on the threshold used to classify counties as suitable (99 or 90% sensitivity), from 1995 through 2016 Ae. aegypti was reported from between 10 and 15% of counties classified as suitable for Ae. aegypti, whereas Ae. albopictus was reported between 59 and 79% of counties classified as suitable for Ae. albopictus. Thus, intensified surveillance efforts for these species launched in response to the ongoing Zika outbreak are likely to produce many more county records. As illustrated in Fig. 1, increasing model sensitivity results in more counties classified as suitable from which there are no mosquito records. We anticipate that Ae. aegypti and Ae. albopictus will be found more commonly in counties classified as suitable based on the lower 90% sensitivity threshold compared with the higher 99% threshold. Counties predicted suitable with 90% sensitivity should therefore be a top priority for expanded mosquito surveillance efforts. However, because eggs or immatures of Ae. aegypti and Ae. albopictus can be transported over long distances in tires and other sundry water-holding containers it also should be noted that they can suddenly appear and most likely also proliferate during the warmest part of the year anywhere within the geographic areas delineated by the 99% sensitivity model. Based on these considerations we chose to present multiple models with sensitivities ranging from 90 to 99%. Comparing the reported species distributions from Hahn et al. (2017) to our suitability models, records of Ae. aegypti are particularly sparse in the eastern United States north of Florida (Fig. 3a). By comparison, counties predicted to be suitable for Ae. albopictus, but where recent records are lacking primarily in the Great Lakes region, the Northeast, parts of the Midwest, and the central and southern Great Plains (Fig. 3b).

Despite a small number of presence records, our Ae. aegypti model obtained an AUC of 0.84. Our finding that winter GDDs >10°C (December–February) are predicted to largely define the range for this mosquito in the United States is consistent with the global range of Ae. aegypti being bounded by the winter 10°C iso-cline (Christophers 1960). More specifically, the model’s response curve indicated that suitability increased with increasing GDDs up to ~700 GDDs. This finding is consistent with previous studies based on observations showing that Ae. aegypti presence/suitability is positively associated with average cold-season temperatures in North America (Lozano-Fuentes et al. 2012, Eisen et al. 2014). For example, Eisen et al. (2014) showed that cities with year-round suitability for Ae. aegypti have average temperatures during the coldest month that are >20°C, such as Key West, Florida. In addition to GDDs, the finding that environmental suitability for Ae. aegypti is also positively associated with temperatures during the warmest month is also consistent with the observation-based findings of Eisen et al. (2014). The shape of the model’s response curve, showing highest suitability between 30 and 36°C, is likely due to the combination of optimal conditions for immature development – development rates for larvae and pupae increase up to about 33–36°C but decrease at higher water temperatures (Focks et al. 1993, Eisen et al. 2014) – and survival rates at all life stages, which typically peak between 20 and 30°C but can still be high at temperatures >30°C (Focks et al. 1993, Brady et al. 2013, Eisen et al. 2014).

Our prediction map for suitability of Ae. aegypti is generally consistent with other recent prediction maps (Capinha et al. 2014, Khormi and Kumar 2014, Campbell et al. 2015, Kraemer et al. 2015a, Monaghan et al. 2016a, b). Overall, our model predicted more extensive suitability in the western United States compared with these previous efforts, a result that may be related to the inclusion of very recent records following the introduction and spread of the mosquito in California. Although Ae. aegypti occurs most commonly in high-density urban settings, we chose not to employ human population data as a model predictor because even rural counties typically have population centers able to support Ae. aegypti if the climate is suitable.

Our model indicates higher suitability for Ae. albopictus in much of the eastern tier of the United States compared with the west, which is consistent with the presence records (Fig. 1b) and other published studies (Benedict et al. 2007, Medley 2010, Fischer et al. 2011, Ogden et al. 2014, Campbell et al. 2015, Kraemer et al. 2015a, Proestos et al. 2015). Suitability was based almost equally on winter GDDs and precipitation during the driest month of the year. Although availability of water is required for both species, the strong eastern bias in presence points for Ae. albopictus compared with Ae. aegypti may help to explain why precipitation was significant in predicting the suitable range for the former, but not the latter species.

Studies report that Ae. albopictus can survive in temperatures ranging from −5 to 40.6°C (Gao et al. 1984, Smith et al. 1988). We excluded minimum temperature of the coldest month (BIO6) and mean temperature of the coldest quarter (BIO11) because both variables are strongly correlated with winter GDDs (correlation coefficients >0.7 and 0.8, respectively). Our model indicates that winter GDDs are associated with the presence of Ae. albopictus, a result that is consistent with previous studies (Hawley 1988, Brady et al. 2014). Our models indicate that the probability of presence for Ae. albopictus is greater than for Ae. aegypti for “colder” counties with winter GDDs between 0 and 175 (Fig. 2c vs Fig. 2a). This may be related to the fact that Ae. albopictus eggs have the ability to diapause (Hawley 1988), whereas Ae. aegypti eggs do not (Christophers 1960); diapause may enhance survival in areas with cooler winters (Thomas et al. 2012). Our model also identified precipitation in the driest month as another key limiting factor for suitability of Ae. albopictus, a result that is supported by others (Alto and Juliano 2001, Rochlin et al. 2013, Cunze et al. 2016). In the absence of human-mediated water sources, precipitation in dry seasons is necessary for egg deposition and increases the availability of suitable breeding habitats (Medlock et al. 2015). A greater reliance of Ae. albopictus, as compared with Ae. aegypti, on water sources filled exclusively by precipitation rather than in part by human action (Faraji and Unlu 2016) may partially explain why it is not widespread in the largely arid and semi-arid western United States. Additionally, humid air masses are generally associated with precipitation (e.g., Higgins et al. 1997), and thus dry-season rainfall may be a proxy of the lower limits of humidity which largely impact the egg and adults stages of Ae. albopictus (Waldock et al. 2013).

A limitation of both models is that they are trained using macroclimatic (county-level) data. Previous studies indicate that Aedes (Stegomyia) mosquitoes have the ability to inhabit suitable microclimates when macroclimatic conditions may be otherwise unsuitable (Hayden et al. 2010, Lima et al. 2016). This could apply either to cooler and moister microhabitats in very hot and dry areas in the southwestern United States or especially warm habitats, such as urban heat islands, in more northern parts of the country. This characteristic suggests that our models may underestimate suitable ‘pockets’ for Aedes (Stegomyia) presence within counties deemed unsuitable by our models; this effect may be particularly plausible near the geographic margins of suitability where macroclimatic conditions are just beyond the suitable range.

An additional limitation is that presence records used to inform our models did not differentiate between introduced or transient populations and established populations because no such information is available across the United States. Our model was not overly sensitive to geographic outliers, but nonetheless, areas on the margins of where the mosquitoes can survive year-round without repeated introductions cannot be separated from those in which the climate was classifiied as suitable for year-round survival. Despite these limitations, the map is useful for highlighting areas that are climatically suitable at least for summer survival of mosquitoes (Monaghan et al. 2016), if introduced, and highlight areas where additional surveillance may be indicated, particularly in high-risk settings for introductions such as tire piles (Yee 2008).

We caution that our models provide information on where mosquitoes could survive and reproduce, but they do not provide estimates of mosquito abundance within suitable counties. Moreover, they should not be interpreted as predictive of risk for human exposure to primarily human-borne chikungunya, dengue and Zika viruses, as this requires the establishment of local human-mosquito transmission cycles and other factors in addition to the simple presence of a mosquito vector in unknown numbers. Nonetheless, our predictive maps of climate suitability may aid in prioritizing mosquito surveillance to areas where Ae. aegypti or Ae. albopictus have not been reported, but where climatic conditions are suitable for establishment. Counties predicted suitable with 90% sensitivity should be a top priority for Aedes (Stegomyia) vector surveillance, as they are more likely than counties classified as suitable at 99% sensitivity to yield mosquitoes. If human mediated water sources are available anywhere in the Ae. aegypti range, it is possible to also encounter Ae. albopictus because winter temperatures are high enough to support mosquito populations and the precipitation deficit may be overcome by alternative water sources. Finally, as additional surveillance data become available, these models should be refined to improve accuracy of predictions.