Focus area

Models developed in this study utilize a global coverage. However, we selected a focus area to illustrate the ability of human footprint-inclusive SDMs to inform management decisions at regional scale. The focus area encompasses four countries—Great Britain, France, Belgium and The Netherlands—including the British Channel and the southern part of the North Sea. The region has a long history of trade and travel, and includes important commercial ports. These intensive activities across national borders have led to the introduction of over 6,000 non-native animal, plant and other species to this area, both from other European regions and further afield [21]. The focus area affected the selection of species for modelling and the interpretation of regional spatial patterns of invasion.

Major predictors of invasive species’ global distribution

The AUC of models ranged between 0.90 and 0.99 (average 0.97±0.02), and sensitivity between 0.73 and 1.0 (average 0.91±0.05), thereby suggesting a relatively high performance. The analysis of model outputs grouped by four major habitat-taxon groups revealed certain generalities in the response of terrestrial plant and animal, freshwater and marine invasive species to global environmental and human gradients (see extended results from SDM in S7 and S8 Tables). As expected, temperature related variables explained the largest amount of the potential distribution of the invasive species investigated (average for terrestrial animals = 53.6%, terrestrial plants = 62.5%, freshwater = 45.4%, marine = 26.2%; Fig 1). Aquatic inland organisms, terrestrial plants and terrestrial animals showed a similar response to mean annual temperature (Fig 2A), peaking around 10°C; but different suitability optima at increasing minimum temperatures (Fig 2B). The response of marine invaders to minimum water temperature was variable although it generally peaked at 15°C (Fig 3A).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Permutation importance of environmental and socio-economic predictors in species distribution models.

Variables are ordered by their overall mean importance. T = temperature, PP = precipitation, HII = Human Influence Index, MHI = Marine Human Influence. Bars represent the standard deviation of the mean value. Insert pie-charts summarize the influence of major groups of variables on the distribution of the four taxon-habitat groups. Temperature related variables were most important in explaining invasive species distribution, followed by the human footprint.

https://doi.org/10.1371/journal.pone.0125801.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Response curves showing the relationship between spacial suitability scores extracted with MaxEnt and seven of the most important drivers of their global distribution.

Lines represent the combined response of terrestrial animals (red), terrestrial plants (green) and freshwater organisms (blue) evaluated in this study. Pointed lines represent 95% confidence intervals around the mean. Spatial suitability for invasive species generally showed a unimodal response to temperature related variables, increased with Human Influence and population density, and closeness to transport networks (roads and ports).

https://doi.org/10.1371/journal.pone.0125801.g002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Change in spatial suitability for marine organisms along the most important drivers of their global distribution.

Pointed lines represent 95% confidence intervals around the mean. Like in Fig 2, the spatial suitability for marine invaders was highest at intermediate temperatures and increased with Human Influence.

https://doi.org/10.1371/journal.pone.0125801.g003

The permutation importance of the human footprint was highly variable, with values between 1.2 and 87.3% (average for terrestrial animals = 21.4%, terrestrial plants = 23.0%, freshwater = 25.8%, marine = 21.1%, Fig 1). Proximity to ports, roads, the Human Influence Index (HII) and cholorphyll-a—used as proxy for oceanic eutrophication—were particularly important factors (Fig 1). Invasive species showed a positive response to increasing human influence values, with HII>25 leading to habitat suitability scores over 50% for most species (Fig 2C). This factor reached highest permutation importance in models developed for terrestrial animals, particularly for insects such as the Mediterranean fruit-fly (Ceratitis capitata) and the Argentine ant (Linepithema humile). In the marine environment, most species also showed increasing occurrence potential with cumulative Marine Human Impact (MHI) values (Fig 3B), although this variable achieved lower permutation importance than expected (<1%). Suitability for marine invaders peaked at chlorophyll-a concentrations around 20–30 mg/m³ (Fig 3C). An important driver of the distribution of freshwater invaders was proximity to commercial ports (Fig 1C and Fig 2D), which achieved >40% permutation importance in the models calibrated for the walking catfish (Clarias batrachus) and American shrimp (Gammarus fasciatus) (S3 Table). Proximity to roads was a particularly important driver for plant species distribution. In fact, the distribution of two plants, the Kudzu (Pueraria lobata montana) and Kahili ginger (Hedychium gardnerianum) was largely explained by road proximity (permutation importance>30%), showing the maximum probability of invasion within 2 km from roads (as in Fig 2E). Population density and land-use were the least significant predictors of invasive species distribution included in the models. Nevertheless, population density>2500 habitants/km² led to invasion suitability >50% (Fig 2F), whereas urban and agricultural areas were the land-uses under highest risk of invasion (Fig 2G).

Mapping invasive species distribution at global and regional scale

The ‘heat map’ presented in Fig 4 synthesizes information regarding the risk associated with a diverse set of aquatic and terrestrial organisms displaying a wide range of origins, pathways of invasion and habitat preferences. In the focus area (Fig 4B), the ‘heat map’ highlighted ports and urban areas around the British Channel and southern part of the North Sea as potential hotspots of invasive species. Occurrence potential decreased towards the NW in Great Britain, and towards the SE in France, Belgium and The Netherlands. The surroundings of commercial ports like the Thames, Southampton, Rotterdam, Antwerp and Boulogne-Sur-Mer showed the highest number of potential future invaders, with up to 36 different invasive species showing values above the selected likelihood threshold. Major urban areas like London, Liverpool, Paris, Amsterdam, Utrecht, Ghent or Brussels can also be important gateways of invasion according to our future ‘heat map’. In the marine environment, the combined suitability for invaders was higher towards the north-east of the North Sea, along the coast of Norway (Fig 4B).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4.

Heat map showing cumulative probability of presence of 72 invasive species at (A) the global scale, and (B) across the focus area in Great Britain, France, Belgium and The Netherlands. High cumulative risk scores can be found around ports and urban areas around the British Channel and southern part of the North Sea.

https://doi.org/10.1371/journal.pone.0125801.g004

Major predictors of invasive species’ global distribution

Confirming observations made in previous studies that implemented climatic and non-climatic factors [mostly land cover, e.g. 29; but see 20 for other human activities], climate-related variables in our models were the most important in determining the global scale distribution of 65 out of 72 invasive species analysed. Amongst climate factors, temperature-related variables explained about half of the distribution of terrestrial and freshwater organisms, which is in accordance with previous studies [20, 27, 47].

Araujo et al. [48] revealed that differences in the global distribution of species are largely driven by their cold tolerance level, with upper temperature tolerance levels being relatively similar across species. Likewise, our models showed that maximum temperature optima are centred around 25°C for all continental organisms, and that optimum minimum temperatures are generally lowest for freshwater organisms, medium for terrestrial plants, and highest for terrestrial animals. Water temperature was also a primary driver of biological invasions in the marine environment, with species spanning a wide range of suitable water minimum temperatures, from 5 to 25°C. This partially confirms the observation of Zerebecki & Sorte [49] that marine invaders tend to occupy habitats with broader temperature ranges and higher maximum temperatures than natives.

Despite the dominant role of temperature in SDM, the human footprint explained about a quarter of the global distribution of invasive species, with only minor differences between terrestrial and aquatic habitats (Fig 1). Although results were very variable between species, models revealed common patterns across the four habitat-taxon groups investigated. Firstly, all groups—but particularly invasive plants—featured distributions that were influenced by the location of transport routes. This confirms previous authors who argued that transport networks promote the dispersal of non-native species by altering habitats, stressing native species, and providing movement corridors [12, 16, 19, 50]. For instance, roads have been shown to promote hunting, fishing, passive harassment of animals, and landscape modifications [50]. Transport routes therefore enhance immigration rates of new species and the spread of non-native species that have already arrived.

All of the modelled species showed a consistent positive response to the Human Influence Index and the Marine Human Impacts, which suggests that the probability of an invasion grows with the intensity of habitat use by humans [20, 51]. For a varied range of organisms, the probability of invasion exceeded 50% at Human Influence values above 25, which at the global scale correspond to the most industrialized regions of Europe, North America, South East Asia and South America (S2 Fig). Moreover, regions with Human Influence values above 25 largely coincide with global hotspots of invasive species richness recently reported by Leprieur et al. [51] and Béllard et al. [29]. This is presumably because the human activities responsible for the introduction of non-native species such as horticulture, pet trade, hunting or fishing, are more common in densely populated areas. Amongst species modelled in this study, the distribution of insects was particularly strongly affected by the Human Influence Index. Although invasive insects can fly over short distances and are also carried by the wind and animals, their main long-distance dispersal is human-assisted, as contaminants of imported plants, soil, flowers or fruit [52]. Despite the positive response of marine invaders to the degree of human influence, Marine Human Impacts registered low permutation importance, even though this factor summarized the effect of multiple human activities that have direct or indirect impacts on marine ecosystems (e.g. shipping, fishing, pollution). In contrast, primary productivity (chlorophyll-a)—used in this study as a proxy for eutrophication although it can also indicate water upwelling—was an important factor in explaining the distribution of marine invasions. These observations, together with the high permutation importance showed by temperature and nutrients such as nitrate, support a relevant role of the match between physicochemical characteristics of the donor and receiving waters for marine invasions, as suggested by Seebens et al. [53].

Proximity to ports was identified as an important predictor not only for aquatic species that can be transported in ballast water or attached to the ship’s hull, but also for some terrestrial plants and animals. The importance of ports as gateways for non-native species is well known, with shipping intensity and distance between origin and recipient ports having been identified as the most important factors determining invasion risk [53]. Invasive species are transported as commodities and deliberately released or escape from captivity, or can be involuntarily transported as contaminants or stowaways [12, 54]. A further source for the importance of port proximity may lay in the fact that port closeness also reflects coastline proximity. Coastal landscapes are being transformed as a consequence of the increasing demand for infrastructures to sustain residential, commercial and tourist activities. Thus, intertidal and shallow marine habitats are largely being replaced by a variety of artificial substrata (e.g. breakwaters, seawalls, jetties) that are very susceptible to invasion [55]. In addition, invasive species are usually tolerant to harsh environmental conditions and may benefit from reduced native biodiversity—thus biotic resistance—in inter-tidal areas. Finally, population density and land-use were often removed from models during variable selection in favour of other socio-economic indicators, which suggest that their effects were already accounted for by factors such as the Human Influence Index, road and port proximity.

Mapping invasive species distribution at regional scale

Spatial patterns of invasion in the focus area suggest there is room for expansion of current as well as future invaders, with vast areas highly suitable for up to 36 different species, which poses a tremendous challenge in terms of prevention and management. Spatially, the SE of England and coastal areas of Belgium and The Netherlands are under the highest risk of multiple invasions, with risk progressively decreasing outwards, i.e. north and westwards in Great Britain, and south and eastwards on the continent. We can therefore consider the north-eastern part of the focus area (and in particular urban areas adjacent to major ports like London, Oostende, Zeebrudge, Rotterdam and Antwerp) as a hot spot of invasion. The inclusion of human footprint proxies into SDMs was translated into highest cumulative risk scores in close relation to the location of commercial ports, transport networks, population density and the intense use of landscapes for industry, urbanization or recreation purposes (i.e. high propagule pressure), which altogether amplifies the potential for invasion.

In the marine environment, the suitability for invaders was highest around the ports of the British Channel and southern part of the North Sea. That said, in general the focus area was predicted as fairly unsuitable for most of the potential (not yet present) marine species. This is likely due to the fact that these species mostly originate from the Pacific Ocean and Mediterranean Sea, both of which are characterized by rather different climatic conditions from the focus area. This observation agrees with a recent study about the risk of marine invasions caused by global shipping [53]. In spite of their substantial maritime traffic volume, no European port was included in the rank of the 20 ports with highest invasion risk, a fact that authors attributed to the lack of match in environmental conditions between donor and host ports. This supports our prediction that the risk of new marine invasions in the focus area is relatively low, at least from some of the world’s worst invaders investigated here.

We are aware that the analyses performed in this study suffer from multiple limitations. First, results from this study refer to the species investigated and further studies with a more representative set of invaders are needed to generalize our interpretations. In terms of predictors, although the factors evaluated in this study provide valuable information on the effect of the human footprint on invasive species, the inclusion of other predictors more directly related to propagule pressure and species dispersal (e.g. aquaculture, horticulture, shipping frequency) may further improve the predictive capacity of models. In addition, variables related to habitat conditions at regional to local scale, such as vegetation state, water and soil chemistry or habitat structure, might be helpful in refining the areas under greatest threat. In terms of modelling, a number of factors may affect the accuracy of SDMs, some of which are related to the unrealistic ecological assumptions behind models (e.g. species in equilibrium with environment, no dispersal limitations, lack of biotic interaction) and extensively described elsewhere (e.g. [35, 37]). Sample size, bias and clustering of species’ occurrences may affect the modelled distribution of a species [23], and were thus treated with careful consideration. Bias towards highly accessible areas due to uneven sampling effort is a common and pervasive problem in species occurrence data, and may partly account for the importance of human footprint variables in our models. The extent to which sampling bias and propagule pressure has contributed to the patterns described in this study is difficult if not impossible to discern. We therefore advocate a cautious evaluation of data quality and testing different modelling options to optimize final predictions.

Overall, our study provides a comprehensive overview of the relative importance of environmental factors and the human footprint to explain the global scale occurrence of invasive species. Analyses suggest that the occurrence of some of Europe’s most damaging invaders is primarily limited by temperature tolerance. Nonetheless, the human footprint explained on average a quarter of invasive species distribution and was directly linked to the vectors and pathways of distribution for invasive species. Amongst human related factors, transport networks were the most influential over the distribution of invasive plants; terrestrial animals were favoured by the degradation of natural ecosystems; port proximity determined the presence of freshwater invaders; and eutrophication was closely related to marine invasions. These findings are of primary importance since they confirm that the relationship between invasive species and the human footprint–widely investigated at local to regional levels—can be scaled up to the global level. Spatial predictions further confirmed the capacity of the human footprint to correct suitability scores by promoting highest risk values in areas where propagule pressure can be presumed high (i.e. close to transport networks and densely populated areas). These results, consistent across a wide range of species and habitats, suggest that climate is important, but not enough to anticipate future invasions. Consequently, human related information—currently available in the form of easily accessible maps and databases—should be routinely implemented into predictive frameworks to inform upon geographically targeted policies to prevent and manage invasions. Otherwise we might be seriously underestimating the species and areas with the highest risk of invasion. Yet, current linkages between SDM modelling and species management are currently weak and require further progress [56].