South Africa is one of the seventeen most biodiverse countries globally (e.g., Myers et al. 2000; Algotsson 2009; Paknia et al. 2015), otherwise known as megadiverse countries. Although it is one of the countries with a relatively low number of established alien invasive species (Seebens et al. 2017), those that have invaded successfully have had significant impacts on terrestrial ecosystems, their biodiversity, and the services they deliver (Van Wilgen et al. 2008).

Of the 2170 species of spiders recorded from South Africa so far (Dippenaar-Schoeman et al. 2015), 74 species are cosmopolitan or introduced (Dippenaar et al. 2011; Foord et al. 2011; Haddad et al. 2013; Dippenaar-Schoeman et al. 2015). Some of these species have a naturally broad distribution extending beyond the Afrotropical Region, while others have clearly been introduced from other continents.

Badumna longinqua (L. Koch, 1867) (Araneae: Desidae) is indigenous to eastern and southern Australia and is one of several Australian web-building spiders that has benefited favorably from the human impacts of urbanization (Main 2001), which has seen its abundance increase significantly in synanthropic environments. This has created source populations that have facilitated its human-mediated dispersal to other continents and islands.

The first country that B. longinqua invaded was New Zealand, where it was first recorded in 1896 from French Pass at the northern end of the South Island (Simon 1899). It would have been introduced accidentally from Australia and is likely to have established in New Zealand during the European colonial period (1830–1880). It is found throughout New Zealand in all habitats except native forest, and is especially common around human habitations (Forster 1970; Paquin et al. 2010). Since then, B. longinqua has been recorded from the western U.S.A. (Gertsch 1937) and Hawaii (Roth & Nishida 1997), Mexico (Leech 1972), Argentina (Pompozzi et al. 2013), Brazil (Simó et al. 2015), Uruguay (Capocasale & Pereira 2003; Simó et al. 2011), Japan (Okumura et al. 2009), Germany (Kielhorn & Rödel 2011) and South Africa (Haddad et al. 2013), with early specimen records from the U.S.A. dating back to 1918 (Leech 1971). As such, this species provides an excellent example among arachnids of the effects of expansion in global trade during the past century in facilitating accidental species introductions (Seebens et al. 2017).

Badumna longinqua can be easily distinguished from other South African cribellate web-building spiders, particularly amaurobiids (Almeida-Silva 2013) and phyxelidids (Griswold 1990), by the distinctive abdominal markings (Figs. 1a, 2a, c) and unique structure of their cribellate webs, usually comprising a two-dimensional sheet-like capture web with zig-zagging threads accompanied by a retreat to one side (Figs. 1b–d,  S1a (JoA-S-20-069.s1 Supplementary Figure S1.pdf), c, d, f). Among desids, Badumna can be recognized (e.g., Gray 1983; Paquin et al. 2010) by the distinct transverse ridge at the posterior edge of the epigynal atrium (Fig. 2c), the S-shaped embolus of the male palp (Fig. 2d) and the comparatively stout median apophysis (Fig. 2e), with B. longinqua being distinguished from congeners by details of the form of the epigynal ridges, the male palpal structure, and structure of the tarsal organs and trichobothrial bases (e.g., Gray 1983).

Figure 1.

Habitus and web photographs of Badumna longinqua in South Africa: (a) Female with retreat in a leaf, Lorraine, Port Elizabeth. (b) Web constructed in a fence post, Kamma Park, Port Elizabeth. (c) Web amongst cycads in a garden, Thornhill. (d) Webs constructed on the roof of a house, Lorraine, Port Elizabeth.

Figure 2.

Habitus and genitalic structure of Badumna longinqua, female (a, b) and male (c–e) from Port Elizabeth. (a), (c) Dorsal habitus. (b) Ventral epigyne. (d) Palp, ventral view. (e) Same, retrolateral view.

The current paper presents the first comprehensive review of the South African records of B. longinqua, which was recently reported for the first time from the Grassland Biome (Haddad et al. 2013). Further, we used various predictive ecological (bioclimatic) niche models and analytical methods to determine potentially suitable areas for B. longinqua in South Africa and the factors that explain its current distribution. We also determine its possible future range expansion and potential distribution based on hypothetical future climate scenarios.

METHODS

Distribution data.—Distribution data on the occurrence of B. longinqua was gathered from South Africa based on specimens collected in the field by the first author and a few donated specimens, which were identified using illustrations from several taxonomic publications (Forster 1970; Gray 1983; Paquin et al. 2010). Following confirmation of the species' identity, photographic records were sourced from several public resources, including the South African National Survey of Arachnida Virtual Museum (online at  http://www.arc.agric.za:8080/), the iSpot website (online at  http://www.ispotnature.org/communities/southern-africa), and through circulation of e-mails to colleagues at various academic, conservation and public institutions nationwide. Although the majority of records relate to field-collected specimens (20 point localities; see Appendix 1), photographic records were sourced from another five localities (Appendix 2; see  Supplemental Figure S1 (JoA-S-20-069.s1 Supplementary Figure S1.pdf), online at  https://doi.org/10.1636/JoA-S-20-069.s1), which were suspected to belong to B. longinqua through comparisons with photographs of the spiders and their webs taken by the author (Fig. 1), whose corresponding specimens were sampled. This gives a total of 23 point localities from which B. longinqua has been recorded in South Africa. The whole of Australia was used for the native range of B. longinqua, and a total of 202 native records were sourced from GBIF (28 July 2020).

Distribution modeling.—An initial model describing the distribution of B. longinqua in its native range was developed using all 19 bioclimatic variables in WorldClim (available online at  http://www.worldclim.org/). This initial model was refined by including a subset of five environmental variables that together explained more than 90% of the native distribution (Table 1). Environmental variables (global climate data) were at 15 × 15′ arc minutes resolution. The political borders of the species' native range (Australia) were used as a mask (Phillips 2008) to avoid over-predicting and generating low specificity. The models were then projected onto the same environmental variables for South Africa. This was done to determine whether the same conditions (environmental variables, and their ranges) characteristic of B. longinqua's native range exist in South Africa. Colinearity of the 19 bioclimatic variables were accounted for through a step-wise procedure of calculating the variance inflation factor (VIF) for all the variables, removing the variable with the highest VIF, and repeating the procedure until all the variables had a VIF < 10.

Table 1.

Model performance of four modeling algorithms, namely BRT, GLM, MAXENT and RFT, using AUC, COR, TSS and Deviance statistics to assess accuracy. Mean performance of models were evaluated using K-fold cross-validation and Bootstrapping. Abbreviations: AUC – area under curve; BRT – boosted regression trees; COR – point-biserial correlation; GLM – Generalized Linear Models; MAXENT – Maximum Entropy Species Distribution Modeling; RFT – random forest trees; TSS – true skill statistic.

Model evaluations were done through cross-validation and bootstrapping. The K-fold cross-validation consisted of splitting the data into five equal parts and then fitting the model five times. Each time, one part is used as the test data and the other four parts as training data. Model evaluation by Bootstrapping involves sampling with replacement until the database is the same size as the original database. Observations not selected are used for the evaluation. Models were evaluated for accuracy (i.e., goodness of fit) by using the area under the receiver operator curve (AUC) values, with an AUC value of 1.0 indicating a near perfect prediction and a value of 0.5 a prediction that is no better than random (Fielding & Bell 1997). Models with an AUC of >0.9 can be considered very good, with AUC of 0.7–0.9 as good, and <0.7 as uninformative (Swets 1988; Baldwin 2009). In addition to AUC, true skill statistic (TSS), point-biserial correlation (COR), and Deviance were used to evaluate models (Naimi & Araújo 2016).

Four modeling methods were used, namely Generalized Linear Models (GLM), Maximum Entropy Species Distribution Modeling (MAXENT) (Phillips et al. 2006; Phillips & Dudík 2008), random forest trees (RFT) and boosted regression trees (BRT). Model predictions were done based on ensemble output that was weighted by the Deviance for each model (Naimi & Araújo 2016).

Species response curves were modeled using the methods proposed by Elith et al. (2005). The relative importance of variables was assessed by correlating the predicted values to values predicted when a variable is randomly permuted. Permutations of important variables will lead to lower correlation, and correlation of 1 would therefore give an indication of variable importance (Thuiller et al. 2009).

RESULTS

Invasion history.—Badumna longinqua specimens were first collected in South Africa in January 2010 in Lovemore Park, Port Elizabeth (now Gqeberha) in the Eastern Cape, and initially misidentified by the first author as a species of Amaurobiidae. Subsequently, specimens were collected in the vicinity of Bloemfontein (Free State Province) and elsewhere in Port Elizabeth (Appendix 1). It was only during a collection event in Bloemfontein in 2011 that visiting researcher Dr Charles Griswold (California Academy of Sciences, San Francisco) recognized specimens of this species as belonging to Badumna, which was the first confirmation of the occurrence of this genus on the African continent.

During the following years, the species was collected at several additional sites in the Eastern and Western Cape Provinces (Appendix 1). Circulation of photos of B. longinqua in September 2015 led to several photographic records of spiders and webs being provided by members of the public (Appendix 2;  Supplemental Figure S1 (JoA-S-20-069.s1 Supplementary Figure S1.pdf)), which confirmed quite a wide distribution of the species along the southern coast of South Africa. Although B. longinqua was only formally recorded in 2010, it is likely that it was introduced to South Africa several decades ago, evidenced by its wide distribution in the southern parts of the country—particularly its almost ubiquitous occurrence in synanthropic habitats in this area—and by the high population densities observed in many cities and towns in this area, e.g., more than 50 individuals per square meter on flat surfaces (See  supplemental Figure S1a (JoA-S-20-069.s1 Supplementary Figure S1.pdf)).

Microhabitat preferences.—To date, almost all records are associated with synanthropic habitats, notably fences, gardens, buildings (walls, gutters and roofs), and horticultural nurseries (Figs. 1b–d;  Supplemental Figure S1a (JoA-S-20-069.s1 Supplementary Figure S1.pdf),  c (JoA-S-20-069.s1 Supplementary Figure S1.pdf),  d (JoA-S-20-069.s1 Supplementary Figure S1.pdf),  f (JoA-S-20-069.s1 Supplementary Figure S1.pdf)), a pattern consistent with other parts of the world where it has been introduced (e.g., Simó et al. 2011), as well as in its native Australia (Main 2001). Webs are also regularly constructed on a variety of automobiles, which may have contributed to its dispersal. At only one locality (Oyster Bay) has B. longinqua been collected from farm outbuildings and farming implements. In South Africa, the species has not yet been encountered in undisturbed natural habitats. Although regularly collected from Eucalyptus bark in South America and its native Australia, the species has not been collected from plantations or isolated Eucalyptus trees in synanthropic environments in South Africa.

Current distribution and potential range of Badumna longinqua in South Africa.—Model performance averaged at 0.96 for AUC, 0.78 for COR, 0.86 for TSS and 0.36 for deviance (Table 1). Random forest models performed the best, with an AUC of 0.98, COR of 0.81, TSS (0.88) and deviance (0.28). Both cross-validation and bootstrapping suggest that models based on training data performed equally well in predicting test data (Table 1).

Based on projections from Australian records (Fig. 3a), suitable habitat is mainly restricted to the southern coast of South Africa, extending to the east, with the entire interior of the country (including central Free State records) indicated as extremely unsuitable for B. longinqua (Fig. 3b). The southwestern parts of the country were considered to be of very low suitability. Only a single record has been collected in Cape Town and the surrounding areas (Appendix 1), despite information on this species being passed on to a number of invasion biologists and the amateur arachnologists working in this part of South Africa.

Figure 3.

(a) Predicted native distribution of Badumna longinqua in Australia. (b) Projected introduced distribution in South Africa. Circles represent actual observations of Badumna longinqua. Key indicates habitat suitability on a scale of 0 (least suitable) to 1 (optimum).

The records and modeled native distribution range of B. longinqua in Australia (Fig. 3a) suggest that it prefers consistently humid, warm climates, with warm summers (Kottek et al. 2006). Projected distributions of B. longinqua point to suitable climatic conditions along the southern coast of South Africa in particular, but also along the east coast (Fig. 3b), and were mainly associated with the Cfb climate category (Fig. 4) of the Köppen-Geiger climate classification (Beck et al. 2018).

Figure 4.

Koppen-Geiger climate classification (a) for present day (1980–2016) and (b) into the future (2071–2100). Circles represent actual observations of Badumna longinqua in South Africa. Key indicates global climate zones according to the classification of Kottek et al. (2006). PE, Port Elizabeth/Gqeberha; EL, East London; BG, Bedford Gap.

Four bioclimatic variables were consistently the most important for all modeling approaches and evaluation criteria: Bio15 – Precipitation seasonality; Bio18 – Precipitation of the warmest quarter; Bio19 – precipitation of the driest quarter; and Bio2 – Mean diurnal range in temperature (see  Supplemental Figure S2 (JoA-S-20-069.s2 Supplementary Figure S2.pdf), online at  https://doi.org/10.1636/JoA-S-20-069.s2). The species seems to be particularly sensitive to increases in seasonality of precipitation ( Supplemental Figure S3 (JoA-S-20-069.s3 Supplementary Figure S3.pdf), online at  https://doi.org/10.1636/JoA-S-20-069.s3), while increases in precipitation in both the warmest quarter and the driest quarter increased the probability of observing B. longinqua. Increased ranges in diurnal temperature also decreased the likelihood of occurrence in an area ( Supplemental Figure S3 (JoA-S-20-069.s3 Supplementary Figure S3.pdf)).

DISCUSSION

In California, Burger et al. (2001) found that the proportion of non-native species in spider assemblages declined with distance inland from the coast. In South Africa, B. longinqua is more common and widespread in its coastal range, notably at altitudes less than 300 m. The milder climatic conditions along the southern coast, where rainfall occurs all year round, temperatures are generally warm and summers are warm but not extremely hot, contrast greatly with those in central South Africa, characterized by distinctive wet-dry seasons, very low winter rainfall, hot summer days frequently above 35°C, and cold winters with regular subzero minimum temperatures.

However, suitable climatic conditions are found at higher elevations to the east of its current range, in the eastern parts of the Eastern Cape and parts of KwaZulu-Natal (Fig. 4a). Their absence here so far could be related to low sampling intensity in this part of the country, but might also be linked to the larger diurnal range in temperature and the rainfall seasonality of these regions. So, although these regions experience warm summers and rainfall tends to be year-round, the climate is more variable. The ‘Bedford gap’ (Lawes 1990), a semi-desert intrusion between Port Elizabeth and East London that divides the eastern and south-eastern parts of the Cape Floristic Region, might act as a further barrier to dispersal. This gap has played an important role in the biogeography of a range of taxa (e.g., Kulenkampff et al. 2019; Taylor et al. 2020). Predicted future climate (Beck et al. 2018) suggests that this gap will become larger (Fig. 4b), further limiting the dispersal of this species. Although garden irrigation is a common practice in central South Africa, particularly during summer, this does not seem to have facilitated the spread of B. longinqua in this area.

Considering its currently known distribution in South Africa, the most plausible point of introduction and subsequent spread of B. longinqua was through Port Elizabeth harbor, which has approximately 7% of the shipping market share amongst southern African ports (Fraser & Notteboom 2012). It is likely that populations spread to central South Africa through transportation of imported or transported goods (particularly plants, flowerpots and other gardening infrastructure) from the southern coastal parts of its range, particularly Port Elizabeth.

In considering these factors, it is clear that the two populations (i.e., southern vs central South Africa) have displayed contrasting levels of establishment, according to the classification system of Van Wilgen et al. (2014). The southern population has passed through Phase 2 (incursion), Phase 3 (spread), and can now be considered as a Phase 4 species (dominant), as, through competitive exclusion, B. longinqua has displaced most web-building spiders typically found in synanthropic habitats in southern South Africa (Haddad, pers. obs.), e.g., Theridion purcelli O. Pickard-Cambridge, 1904 and Latrodectus geometricus C.L. Koch, 1841 (Theridiidae), Neoscona spp. (Araneidae), and the cosmopolitan Uloborus plumipes Lucas, 1846 (Uloboridae). This aspect of competitive exclusion needs to be more thoroughly investigated and quantified in the future to determine which species are most susceptible to displacement by B. longinqua.

In contrast, the populations from central South Africa may have been introduced to the Bloemfontein area on multiple occasions (Phase 2) but have failed to establish and spread beyond the horticultural nurseries where they were collected, suggesting that unsuitable climatic conditions could cause repeated annual mortality of populations during the very cold winters. It is plausible that the Port Elizabeth population of B. longinqua serves as a bridgehead invasive population (see Guillemaud et al. 2011), from which introductions to other areas such as Bloemfontein have occurred through ground transport. Considering that Bloemfontein populations have to date been exclusively collected from nurseries, it is likely that new individuals are introduced annually with the transportation of ornamental pots and plants from southern South Africa.

The lack of collection data of B. longinqua prior to 2010 could be explained by two factors: (1) the southern parts of the country have historically been poorly sampled for spiders, and as a result, collecting records for these degree-square grids are generally at a much lower level here than many other parts of the country (Foord et al. 2011, 2020); (2) although there is a spider collection housed in the Albany Museum in Grahams-town, approximately 150 km east of Port Elizabeth, the lack of an arachnological curator and researcher there for several decades has contributed greatly to the undersampling of spiders, particularly in the coastal parts of the Eastern Cape. This has resulted in a paucity of identified spider material that could potentially contribute taxonomic data to ecological studies, including the current one. It is quite plausible that there are some B. longinqua specimens amongst this unidentified material that could vastly improve our knowledge of the invasion history of the species in South Africa. However, predicted future climatic conditions in South Africa may probably not be favorable for this species' continued spread in the country.