In 2013, Evans et al. reported 28 invasive termite species worldwide. Distribution data for these species were extracted from reliable sources including Global Biodiversity Information Facility (GBIF 2024), Sistema de Informação sobre a Biodiversidade Brasileira (SiBBr 2024) and the University of Florida Termite Collection (UFTC) curated by Scheffrahn (2019). Of the 28 species, six had less than 30 occurrences after curation (removing duplicates, excluding coordinates with low accuracy and coordinates from countries where the species had not been reported in the literature). Detailed information on the remaining 22 species, including their respective families, feeding groups, nesting types, breeding systems, target damage, native and invaded ranges and spread methods are given in Table 1 (see Suppl. material 2 for occurrences of the 22 initial invasive species and GBIF DOIs). From this subset, we narrowed our focus on the 10 most invasive termites, discarding those with limited spread capacity (e.g. those inhabiting rotten wood) or those lacking recent reports. Six of them belong to the Kalotermitidae family (Cryptotermes brevis (Walker, 1853), Cr. domesticus (Haviland, 1898), Cr. dudleyi (Banks, 1918), Cr. havilandi (Sjöstedt, 1900), Incisitermes immigrans (Snyder, 1922) and I. minor (Hagen, 1858)), three to the Rhinotermitidae family (Coptotermes formosanus (Shiraki, 1909), Co. gestroi (Wasmann, 1896) and Reticulitermes flavipes (Kollar, 1837)) and one from the Termitidae family (Nasutitermes corniger (Motschulsky, 1855)). All produce secondary reproductives and consume wood (Evans et al. 2013). All have a life stage in which they nest in a single piece of wood in which the colony is founded. The Kalotermitidae nest within their food source for the entire life of the colony (i.e. single-piece nesting termite; Abe (1987)); the Rhinotermitidae establish their incipient colony inside a piece of wood or at the wood-soil interface before settling in the soil (Ferraz and Cancello 2004; de Lima et al. 2006); for Nasutitermes corniger, the construction of the final carton nest, whether on a tree or on the ground, may follow the establishment of an incipient colony in a piece of wood (Thorne and Haverty 2000). These life-history traits make them particularly good invaders (Evans et al. 2013).

The curation process of our ten species yielded an average of 313 occurrences per species (Table 1). Three sets of 1000 randomly selected pseudo-absence points were generated for each species using Biomod2 version 4.2-4 package (Thuiller et al. 2009) in the software R version 4.2.0 (R Core Team 2022). These pseudo-absence points were chosen with equal weighting for both presence and absence, following the methodology described by Barbet-Massin et al. (2012).

Explanatory varisables were standardised to the same resolution (0.25°, the resampling was done using the nearest-neighbour algorithm), dimensions (nrow = 600, ncol = 1440), extent (xmin = -180°, xmax = 180°, ymin = -60°, ymax = 90°) and format (WGS84 EPSG:4326) using the software QGIS (QGIS 2023). For each species, a raster stack was built with explanatory variables using the package raster_3.6-14 (Hijmans et al. 2015).

The entire modelling and evaluation process was conducted in R4.2.0 using the Biomod2 4.2-4 package. We performed modelling analyses by integrating bioclimatic, land-use, connectivity and elevation variables into our models. Additionally, we ran models exclusively using bioclimatic variables to assess outcomes and discern any divergences. The same algorithms (FDA, RF, MAXNET, GAM, GLM, GBM) as described earlier were used for all the modelling. For model training, only 75% of the randomly-selected occurrences were utilised, while the remaining 25% were kept for model evaluation. The performance of the models was assessed using two metrics: True Skill Statistic (TSS, Allouche et al. (2006)) and Area Under a Receiver Operating Characteristic (ROC or AUC, Hanley and McNeil (1982)). To validate the models, a 5+1-fold cross-validation was conducted for each species. This resulted in a total of 108 models (six algorithms multiplied by 5+1 cross-validations multiplied by three pseudo-absence samplings) fitted for each species, considering both current and future shared socioeconomic pathways. The importance (i.e. the contribution to the model) of each variable (climatic, land-cover, connectivity and elevation) for each species was evaluated using the variable importance command from Biomod2.

Ensemble modelling, which combines individual forecasts into a consensus projection (Araújo and New 2007), has become a popular technique in species distribution modelling (Hao et al. 2019). However, it may not be always the best approach if all default settings of Biomod2 are chosen, as it can lead to underperforming models compared to well-tuned individual models (Valavi et al. 2022). Therefore, except for MAXNET, FDA and GBM, the default settings of Biomod2 were not used. For RF, downscaled performance is improved by adding the sampsize option and ntrees was set to 1000 as recommended by Valavi et al. (2022). GAM performed best using the GAM_mgcv algorithm, the binomial family (logit) and the REML method (Pedersen et al. 2019). GLM utilised the binomial family (logit) (Hastie et al. 2001). To ensure that only high-performing individual models were included in the ensemble modelling, a threshold of 0.75 was set for the TSS metric. The weighted average method was employed to create the consensus model, as it provides more robust predictions alongside the mean compared to other methods (Marmion et al. 2009). Finally, the same metrics (TSS and ROC) used for the individual models were used to assess the quality of the final consensus model. For ROC, models can be considered poor for values in the range 0.5–0.7, fair in the range 0.7–0.9 and excellent when the value ranges between 0.9 and 1 (Swets 1988). For TSS, models can be considered poor for values ranging from 0.2 to 0.5, useful when ranging from 0.6 to 0.8 and can be considered good to excellent when ranging from 0.8 to 1 (Coetzee et al. 2009).

To facilitate the visual comparison between present and future scenarios, the range size function from Biomod2 was used instead of relying on multiple maps. This function computes the number of pixels that are lost, stable or gained, along with their relative proportions, when comparing two species distribution models. To perform this analysis, the current and future ensemble models were transformed into binary predictions (absence: 0 or presence: 1) by applying an optimised threshold derived from TSS (Thuiller et al. 2009). This approach allowed us to generate a single map that delineates the regions where each species is contracting, stabilising, expanding or absent. Additionally, using Biomod2, the net increase (percentage of pixels predicted to be gained compared to the number of pixels currently occupied) and decrease (percentage of pixels currently occupied and predicted to be lost) were calculated for each species. This methodology provides a clear representation of the spatial dynamics for each species and quantifies the net changes in terms of percentage for each species, offering valuable insights into the projected shifts in their distributions.

To delineate potential high-risk invasion areas, the same methodology as outlined in the previous section was used. However, “lost” and “absent” pixels were converted to 0, while “stable” and “increase” pixels were converted to 1. This adjustment enabled the summation of values for each species across all pixels, providing an estimate of the potential number of species in each pixel under the selected scenario.

The evaluation process supports the robustness and accuracy of the models in predicting species distributions. For ROC, all the individual models could be considered excellent (0.900–1), ranging from 0.968 to 1, while all the ensemble models were close to perfect, ranging from 0.986 to 1 with an excellent average of 0.996. For TSS, all the individual models could be considered good to excellent (0.800–1), ranging from 0.819 to 0.999, while all the ensemble models were good to excellent ranging from 0.879 to 0.994 with an excellent average of 0.958 (Suppl. material 1: S4).

The analysis of variable importance revealed that bioclimatic variables were overall the most important predictors, followed by our four connectivity variables. In contrast, the significance of elevation and land-cover variables (excluding urban land) appears comparatively lower than other variables (Fig. 1, Table 2). The combined use of bioclimatic, land-use, connectivity and elevation variables as predictors refine our projections by reducing by an average of 313% the areas suitable to the establishment of all termite species under study (Table 3).

Figure 1.

Number of times each variable was identified as the most important (First), second most important (Second) and third most important (Third) for each species.

Amongst the top three most important variables, a bioclimatic variable ranked first for six species and urban land for four species. Ranking second in importance, bioclimatic factors held for six species, whereas urban land held for two species. Accessibility to Cities (ATC) and elevation (ELE) held this position for one species. Lastly, the third-rank variable category encompassed bioclimatic factors for four species, C4 perennial crops for two species, ATC for two species, leisure vessels (LVE) and elevation for one species (Fig. 1). Our results underscore the prevailing role of bioclimatic variables and the substantial influence of trade, transport and demography factors (i.e. the connectivity variables) in shaping the distribution patterns of invasive termite species. Concerning the land-cover variables (excluding urban land), only one variable, C4 perennial crops, managed to reach the top three positions. In addition to several bioclimatic and most land-cover variables, the population variable (POP) failed to secure a position in the top three average contributions. Amongst the bioclimatic variables, those linked to temperature held a significant presence in the top three, occurring seven times more frequently as variables tied to precipitation (Fig. 1, Table 2, Suppl. material 1: S5 for the response curves).

The majority of the ten invasive species demonstrate significant potential for occupying a wide range of habitats for the current climate conditions and socioeconomic development. Although no overarching trends apply to all our species, some preferences can be observed with our models. For instance, species such as Cryptotermes brevis, Cryptotermes domesticus, Incisitermes immigrans, Incisitermes minor, Coptotermes formosanus, Coptotermes gestroi and Reticulitermes flavipes all show preference for large and well-connected urban areas, while Cryptotermes dudleyi, Cryptotermes havilandi and Nasutitermes corniger appear to be slightly more restricted to environments resembling their native habitats (Table 3, Suppl. material 1: S6 for all the maps). Below, we will describe the results of the models for each species, focusing solely on their statistical performance without considering their ecological context or the actual conditions of their habitats. For a deeper analysis incorporating these factors, please refer to the Discussion section.

In the short term (2021–2040) and under the SSP2-4.5 scenario, four species are projected to have a significant (> 20%) expanded range: I. minor by 35%, R. flavipes by 38%, Co. gestroi by 39% and Co. formosanus by 68%. In contrast, three species are expected to experience a significant decline in their habitat range: N. corniger by 22%, Cr. domesticus by 23% and Cr. dudleyi by 37%. Three species, Cr. brevis, Cr. havilandi and I. immigrans, are forecast to maintain their current distribution with minimal variations during this period and scenario (Fig. 2, Suppl. material 1: S7).

Figure 2.

Potential projected range shift (left) and high-risk invasion map (right) for selected periods and socioeconomic-shared pathways.

When considering the long term (2041–2060) under the same SSP2-4.5 scenario, the trend remains consistent, except that one more species, Cr. havilandi, is expected to experience a significant range expansion, by 47% instead of the previous 16% observed in the short term (Fig. 2, Suppl. material 1: S7). For most species, the change in range becomes even more pronounced during this later period.

Shifting to a more pessimistic scenario (“fossil-fuelled scenario”, SSP5-8.5) reveals a broader impact, with six species significantly increasing their habitat range. Co. gestroi is expected to expand by 28%, R. flavipes by 47%, Cr. domesticus and I. minor by 81%, Cr. havilandi by 105% and Co. formosanus by 175%. Conversely, Cr. dudleyi and N. corniger could experience a significant reduction in range, by 29% and 35%, respectively, under this scenario; while Cr. brevis could see a slight decrease of 12% (Fig. 2, Suppl. material 1: S7).

Overall, the average range shift is consistently positive for each scenario and increases over time as the combined effects of climate change and socioeconomic developments intensify. Specifically, in the short term (2021–2040 SSP2-4.5), the average range shift is low at 10%. In the long term (2041–2060) under the same scenario, this increases to 19%. In a scenario characterised by higher fossil fuel reliance, the range shift reaches 46% (Suppl. material 1: S7). These percentages represent net changes in terms of pixels and do not offer insights into the specific regions that may be affected in the future. The right side of Fig. 2 addresses precisely this limitation by showing the number of species potentially suitable in each pixel of the map for each scenario. Upon closer examination, regardless of the scenario, it appears clear that major cities and thriving economic hubs are the most at risk, especially in North America, South America, western Africa, Europe and the Asia Pacific Region. In these areas, the number of potential species for the worst scenario is between 5 and 10. When focusing on individual countries, several seem to be especially at high risk: the US, Brazil, Nigeria, China, Indonesia, Japan and Australia. These countries could host a substantial number of cities with a high number of invasive species. Nevertheless, all continents have significant areas suitable for more than four species at a time (Fig. 2).

To look at the range shift for each species separately, a comprehensive set of 30 distinct maps providing a visual depiction of the changes for each species, timeframe and shared socioeconomic pathway has been developed (Suppl. material 1: S6). Overall, the trend shows that, under more pessimistic scenarios and over time, species tend to shift towards higher latitudes, both northwards and southwards. Most species are, however, expected to experience a slightly reduced range in lower latitudes (Table 3, Suppl. material 1: S6). For instance, a striking comparison can be made amongst three contrasting species, Incisitermes minor, Reticulitermes flavipes and Nasutitermes corniger. I. minor, originating from the south-western US and northern Mexico, has spread likely through the transportation of infested furniture to eastern US, Canada, China, various islands in the Pacific Ocean and Japan. In a pessimistic scenario (SSP5-8.5 2041–2060), I. minor is expected to extend its range across the US, Europe, Japan, China and Australia, particularly in densely urbanised regions (Fig. 3, Table 3). In contrast, according to our models, both R. flavipes and N. corniger show a lower dependency on urban environments, especially N. corniger. R. flavipes is native to eastern US and northern Bahamas and has been introduced to Canada, Europe, South America and Easter Island. Under similar scenarios, its range potentially expands significantly across the US, Europe, southern Australia, southern Africa and southern South America. However, at the same time, its suitable range could decrease in southern US and in southern parts of China (Fig. 3, Table 3). On the other hand, N. corniger, originating from Neotropical Regions (Central and South America, West Indies), has established itself in only a few areas, in Florida and New Guinea. According to our models, N. corniger is projected to experience significant declines in its potential suitable tropical range, particularly in Africa and Brazil (Fig. 3, Table 3).

Figure 3.

Potential projected range shift for Incisitermes minor, Reticulitermes flavipes and Nasutitermes corniger between potential current suitability and the period of 2041–2060 under the SSP5-8.5 scenario.

Our study validates the significance of bioclimatic conditions as fundamental variables to understand termite distribution patterns (Eggleton 2000); amongst these factors, temperature emerges as the most influential determinant. We also note the minor influence of precipitation variables compared to temperature; a finding consistent with Guerreiro et al. (2014). It is worth noting, however, that invasive termites tend to establish nests within controlled-temperature human structures (Su and Scheffrahn 1998), highlighting the importance of considering not only bioclimatic factors, but also variables reflecting termite movements through these human structures (“connectivity” in this study). Our analysis highlights the significant role of the connectivity variables, particularly urban cover and Accessibility to Cities (ATC) and, to some extent, leisure vessels (LVE), in explaining the distribution of invasive termites. This echoes previous research showing the substantial predictive power of socioeconomic variables, ranking second only to bioclimatic and habitat variables (Bellard et al. 2016). The significance of human activity in facilitating the establishment of invasive termite populations is notable, given their primary reliance on wood, a ubiquitous commodity found in every household, boat and city worldwide. Regions with higher human activity should be predisposed to facilitating establishment of invasive termite populations. However, our results suggest that the human population (POP) layer contributes relatively little to the predictive capacity of our distribution models. This could be attributed to its high correlation with urban cover (0.53), suggesting potential redundancy with greater efficiency. Notably, areas such as seaports, airports and industrial zones, despite exhibiting lower population densities, exert a more pronounced influence on invasion risk compared to population density itself (Bellard et al. 2016).

Elevation (ELE) was found to play a minor role in predicting the distribution of invasive termites. While Guerreiro et al. (2014) reported a significant contribution of elevation to their model for Cr. brevis, our modelling revealed that the distribution of this species is primarily shaped by a combination of bioclimatic (bio11) and connectivity (urban and ATC) factors, with elevation contributing by less than 2%. If Guerreiro et al. (2014) used four variables (three bioclimatic factors and elevation) per species, our approach, utilising a comprehensive set of eight variables per species, likely attributed greater significance to other variables, reducing the prominence of elevation.

A previous study had integrated land-cover variables to project the spatial distribution of two invasive termite species (Tonini et al. 2014), but the extent of their contribution to the models was not reported. Here, excluding urban cover, the impact of land-cover variables – encompassing the cultivation of crops, including C3 perennial and C4 annual varieties, as well as forested, deforested and pasture areas – showed limited influence on the models. These results are expected given that agricultural crops are subject to significant disruption from human activities (e.g. pesticide applications). However, forthcoming environmental regulations and enhanced human health measures could bolster the survival rates of invasive termite species within these ecosystems (Haifig et al. 2008).

Most of the ten invasive termites we studied show the ability to occupy a wide range of habitats, especially urban areas, confirming the global threat posed by invasive termites. Contrary to previous descriptions (Buczkowski and Bertelsmeier 2017), our models portray a more confined distribution, but with a heightened focus on urban and well-connected areas. This result is in line with the biology of invasive termites, often establishing themselves initially in urbanised localities before spreading to natural environments.

We also noted important differences for Cr. brevis, with a narrower suitable range compared to previous reports (Guerreiro et al. 2014; Buczkowski and Bertelsmeier 2017). This species primary inhabits in buildings outside its endemic range (Scheffrahn et al. 2009) and places of similar endemic climate such as Morocco (Najjari et al. 2023). Our modelling accentuated a preference for cities and economically developed regions, especially in the Americas and Africa. Interestingly, Asia and Europe, aside from major cities like Tokyo, Jakarta, Manila, Shanghai and the Guangdong-Hong Kong-Macao Greater Bay Area, showed less suitability, likely due to the presence of other Cryptotermes species in the Asia Pacific habitat (Scheffrahn et al. 2009; Guerreiro et al. 2014). Similarly, our models for Cr. domesticus, Cr. dudleyi, Cr. havilandi and I. immigrans identified limited distribution along the coastlines of tropical countries in Africa and South America, as well as suitability in most of Southeast Asia, especially Indonesia. The suitability is particularly high in major cities of these regions like São Paulo, Rio de Janeiro, Jakarta, Greater Bay Area or Lagos. This contrasts with the broader suitability suggested by Buczkowski and Bertelsmeier (2017) for these four species across tropical regions. These variations might be attributed to our comprehensive integration of land-cover, elevation and connectivity variables: elevation was a significant variable for Cr. domesticus, while C4 perennial crops was for Cr. dudleyi, leisure vessels (LVE) for Cr. havilandi, and both Accessibility to cities and urban land for I. immigrans. A similar divergence emerged for I. minor in higher latitudes, highlighting the strong influence of large urbanised and connected areas (e.g. most large US cities, southern California, São Paulo, London, Belgium, Madrid, Greater Bay Area and main cities of eastern Australia).

Our models for Co. formosanus and Co. gestroi also showed a distribution heavily associated with urban areas (see also Li et al. (2013) and Tonini et al. (2014)) both in their native and introduced regions. The use of connectivity variables for both species likely contribute to these results, accounting for 50% and 44% to the models, respectively. On the other hand, we identified suitability for Co. gestroi in European and American cities, such as Paris, London, Madrid or New York, although such suitability is highly improbable due to the species’ restriction to tropical regions, necessitating favourable conditions, such as adequate humidity and high temperatures (Li et al. 2013). Nonetheless, our models indicate high suitability for Co. gestroi in more likely regions, such as large economic areas in tropical and subtropical regions like western Africa, Southeast Asia or south-eastern China. Conversely, Coptotermes formosanus, with a warm temperate to subtropical distribution (Cao and Su 2016), presented heightened risk in large cities of south-eastern US and Asia, as well as South America. As the effects of climate change reshape ecosystems globally, disparities between the two species are likely to become more apparent, especially considering instances like Co. formosanus potentially expanding its range into the Korean Peninsula due to increasing temperatures (Lee et al. 2021).

Regarding Reticulitermes flavipes, our models suggest a threat to most cities in temperate and subtropical regions, particularly in its native range in the US, but also in Europe and eastern Asia, with a notable focus on urban areas (the urban layers contribute 36% to the projections). Our results partially disagree with those of Buczkowski and Bertelsmeier (2017) who suggested potential suitability in higher latitudes (e.g. Iceland, Norway, Alaska or Patagonia) and the Tropics (Indonesia, Ecuador, Colombia). Such locations appear unlikely given the current distribution of the species (Evans et al. 2013). Modelling also shows that the distribution of R. flavipes does not extend to cold regions, such as the Alps or the Carpathians in Europe, a limitation attributed to the elevation layer.

Finally, our results suggest that N. corniger is highly adapted to tropical regions. Our results agree with those of Buczkowski and Bertelsmeier (2017), except that the latter also designate parts of southern Argentina, Chile, Morocco, the Arabian Peninsula and Australia as potentially suitable. This seems unlikely, as this species is strictly restricted to tropical regions, its actual distribution ranging from southern Mexico to southern Brazil and northern Argentina, including most of the West Indies (Scheffrahn et al. 2005).

Facon et al. (2006) proposed three scenarios to understand the relationship between invasive species and their new environments, shedding light on the role of human activities in biological invasions. The first scenario suggests that invasion is limited by the small population size of the invasive propagule (to establish a viable population), but changes in migration patterns, possibly caused by human activity, can trigger the invasion. The second scenario suggests that invasion can proceed if the introduced species finds a suitable match with the environment; this adequacy can be facilitated by changes in the biotic or abiotic environment, changes often influenced by human activities. The third scenario highlights genetic changes in the invader as a factor initiating invasions, including reduced genetic variance, inappropriate range of adaptive variation in the original species and maladaptation due to excessive migration. Invasive termites fit perfectly within this framework. Migration change (e.g. by wood exchanges through furniture or private vessels) and human activities in urban areas offer many opportunities to establish and invade new territories. Evans et al. (2013) identified three characteristics common to all 28 invasive termite species that increase their likelihood of successful propagule. First, invasive termites are all wood feeders (Table 1).

Secondly, nesting in wood is particularly advantageous for invasions given the ubiquitous presence of wood in households worldwide (e.g. Grace et al. 2009). While some species live, eat and nest exclusively in wood (called single-piece nester, here the Kalotermitidae species), others (here the Rhinotermitidae species) begin their colonies within wood or at the soil-wood interface before establishing their colonies underground or in another piece of wood (Ferraz and Cancello 2004; de Lima et al. 2006). The latter are called intermediate-piece nesters and will forage outside their colony after this single-piece stage (Abe 1987). Nasutitermes corniger, though typically nesting on trees, poles, walls or directly on the ground (Thorne 1980), can also establish an incipient colony inside small wood pieces (Thorne and Haverty 2000; Scheffrahn et al. 2014). All ten invasive species can, therefore, use wood as a hidden mean of transport for at least part of their lifecycle.

All ten invasive species share a third characteristic: the ability to produce secondary reproductives, typically through neoteny of nymphs (nymphoid reproductives), workers or pseudergates (ergatoid reproductive) or through the retention of alates (adultoid reproductive) (Myles 1999). In the case of N. corniger, ergatoids are produced, rendering any piece of wood with foraging workers a viable propagule at any time of year (Thorne and Noirot 1982). These ergatoids, combined with the fact that they are polygynous, probably allowed N. corniger to invade New Guinea 100 years ago via sugar trade shipping from the West Indies, travelling more than 15,000 km (Roisin and Pasteels 1986; Scheffrahn 2013).

Consequently, all ten of our highly-invasive species are capable of nesting in wood, whether in furniture or in boats (Scheffrahn 2023; Chouvenc, personal communication, January 2024), rendering invasions extremely likely, be it from one household to another or from a marina to nearby coastal residences, as exemplified by N. corniger or Coptotermes (Scheffrahn and Crowe 2011; Scheffrahn et al. 2014; Hochmair et al. 2023; Scheffrahn 2023). Flying imagoes are attracted by light poles and illuminated houses near marinas, serving as an initial entry point into the lands (Scheffrahn et al. 2014; Scheffrahn, personal communication, January 2024). However, once arriving in a new environment, invasive species may face new challenges in establishing themselves if the environment differs or is already saturated by other species. To succeed under such conditions, the species may undergo substantial adaptations (Facon et al. 2006). For example, native populations of Reticulitermes flavipes consist of colonies headed by monogamous pairs of primary reproductives; in contrast, introduced populations in France exhibit a breeding structure where hundreds of related neotenics reproduce while colonies lose intraspecific aggression and get a propensity to fuse (Perdereau et al. 2013; Perdereau et al. 2015). These changes in colonies composition and organisation confer advantages in terms of resource exploitation and competition, thus facilitating ecological dominance.

Some termite species have biological and/or behavioural characteristics that can help them invade and survive in new territories. For instance, Reticulitermes species are naturally well-adapted to low temperatures and can move the nest to deep underground during the winter (Cabrera and Kamble 2001; Takata et al. 2023). This behavioural trait can prove decisive in extending their range northwards (Cabrera and Kamble 2001; Takata et al. 2023). Additionally, climate change may relax constraints on naturalisation and, hence, increase the hybridisation risk amongst invasive and native populations (e.g, Co. gestroi and Co. formosanus in Florida), further influencing invasion dynamics (Chouvenc et al. 2015; Fournier and Aron 2021). Elevated temperatures also enhance termite foraging activity (Kasseney et al. 2011; Zanne et al. 2022). For example, Co. gestroi and Co. formosanus have a significantly higher wood consumption rate between 22–35 °C than between 10–15 °C (Patel et al. 2019). Finally, biological invasions could also be facilitated by the bridgehead effect (Lombaert et al. 2010). Introduced populations are often the source of other invasions (Yang et al. 2012; Sherpa et al. 2019; Blumenfeld and Vargo 2020). These secondary invasions, therefore, originate from populations that have already succeeded in overcoming all the barriers – geography, survival, reproduction, dispersal, environment – that punctuate from introduction to propagation invasion processes. For instance, native to eastern Asia, Co. formosanus first established populations in Hawaii before invading the US mainland (Blumenfeld et al. 2021).

Urbanisation is an inevitable phenomenon as projected by the United Nations Department of Economic and Social Affairs (UN DESA 2020): up to two-thirds of the global population is expected to reside in urban areas by 2050. Moreover, the number of megacities (with over 10 million inhabitants) is projected to rise from 33 in 2018 to potentially 43 in 2030. Urbanisation and megacities, notably through the urban heat-island effect (Szulkin et al. 2020), will therefore constitute invasion hotspots that may serve as bridgeheads in surrounding natural habitats as temperatures increase due to climate change. Our study demonstrates that most species could thrive in a changing climate and an increasingly urbanised world, particularly under a fossil-fuelled future (SSP5-8.5). This finding aligns with other studies on different and shared species and using different variables, collectively suggesting a consensus that climate change will generally increase the distribution of most invasive or pest termites (e.g. Tonini et al. (2014); Buczkowski and Bertelsmeier (2017); Lee et al. (2021)). While some species might experience localised contractions in their ranges, they are anticipated to make gains in other regions, particularly within densely urbanised areas. This trend is evidenced by our incorporation of connectivity variables, which reveals the potential for more substantial and costly damage. For instance, our modelling shows that Cr. brevis could experience a reduced tropical range, yet extend its distribution into urbanised higher latitude regions. Thus, a contraction in range does not necessarily translate into a reduction in damage costs, especially for termites that predominantly target human structures, such as species from the Kalotermitidae and Rhinotermitidae families. Conversely, other species like Co. formosanus could mainly expand their range, with close to zero reduction compared to their potential current suitable habitat. The Formosan subterranean termite could potentially thrive in higher latitudes in a fossil-fuelled world, placing most American, European and eastern Asian cities at risk. On the other hand, N. corniger will lose significant suitability in a fossil-fuelled world within Brazil and central Africa, probably due to the combined effects of increased deforestation and less favourable bioclimatic conditions.

Overall, most of these ten invasive termites will thrive in a changing climate and a heavily transformed world marked by escalating urbanisation, particularly under a fossil-fuel-dependent trajectory. Even if our models do not consider the full force and speed of future connectivity, the undeniable expansion potential of the ten termite species and, therefore, the damage concomitant with invasions, underscores the urgency of addressing climate change, urbanisation and growing connectivity. These factors will be crucial in contributing to the spread of invasive termites, posing a significant threat not only to the economies of invaded regions, but also, to some extent, to biodiversity and ecosystem functioning.

The authors have declared that no competing interests exist.

No ethical statement was reported.

This research was funded by the French Community of Belgium through a grant from the Fonds de la Recherche Scientifique – Fonds National pour la Recherche Scientifique (FRS-FNRS, no. J.0110.17 to DF; FNRS/FRIA funds to ED) and the Université libre de Bruxelles (Fonds Defay to DF).

Edouard Duquesne: Data curation, Formal analysis, Methodology, Software, Visualisation, Writing – original draft.

Denis Fournier: Conceptualisation, Funding acquisition, Supervision, Visualisation, Writing – original draft, Writing – review & editing.

Edouard Duquesne https://orcid.org/0000-0003-2105-9434

Denis Fournier https://orcid.org/0000-0003-4094-0390

All of the data that support the findings of this study are available in the main text or Supplementary Information.