We used South African National Parks (SANParks) as a model system as it has 752 alien plant species recorded across 19 national parks (Spear et al. 2011, Foxcroft et al. 2017). The SANParks estate covers an area of about 39,000 km² and spans eight of the nine biomes in South Africa. Using the full list of alien species recorded in SANParks by Spear et al. (2011), we extracted a subset of those alien plants we considered to be transformer species.

Transformer species were defined as the “subset of invasive plant species that change the character, condition, form, or nature of ecosystems over a substantial area relative to the extent of that ecosystem” (Richardson et al. 2000, McGeoch et al. 2006). From the overall SANParks species list we extracted those species that we considered transformers using information from four key publications, (i) Henderson (2001), declared alien weeds and invasive plants, (ii) Nel et al. (2004), a classification of invasive alien plant species in South Africa, (iii) van Wilgen et al. (2008a), a biome scale assessment of the impact of invasive alien plants, and (iv) van Wilgen et al. (2008b) prioritising species and catchments for guiding invasive alien plant management in South Africa). We also considered national legislation that regulates the management of alien and invasive plants in South Africa (CARA; Conservation of Agricultural Resources; Act 43 of 1983, as amended 2001). The criteria for designating alien plant species as transformers in our assessment were, (i) species recorded as ‘transformers’ or ‘potential transformers’ in Henderson (2001), (ii) species classified in Henderson (2001) as ‘special effect weeds’ AND also listed in two or more of the other publications (but in CARA only if listed as a Category 1 prohibited species) and (iii) four species in the SANParks list were considered transformer species based on the authors’ personal observations in parks and supported by expert opinion. The latter resulted in the inclusion of the following species: Austrocylindropuntia cylindrica, Aristolochia littoralis, Bryophyllum delagoense and Pontederia cordata. Pontederia cordata was also recorded in Henderson (2001) as a ‘special effects weed’ and in the CARA regulations as a Category 3 invader.

This selection process resulted in a list of 139 alien plants regarded as transformer species (see Suppl. material 1: Table S1 for species list and data). By using a post hoc approach we aimed to elucidate those pathways that should be considered future management priorities for these species, and also the range and frequency of types of impacts likely to be experienced.

The potential pathways of introduction and impacts that alien plant species may have in national parks in South Africa were determined by the authors as a group, using literature (for example Mack et al. 2000, Morse et al. 2004, Randall et al. 2008, Vilà et al. 2010, Wells et al. 1986). As the classification was based on information not only from South Africa but also from other parts of the world, where data on impacts and pathways associated with species on our list are available, and it has not been proven that they actually occur in the parks under study, we term the pathways and impact categories ‘potential’ (see Rumlerová et al. 2016). The rationalisation of categories resulted in a list of eight locally-relevant pathways (nationally and within parks) of introduction and dispersal: rivers; ornamental plants; roads, paths, trails, tracks; contaminated construction material, equipment, soils; agriculture; clothing; food or produce; and dispersal by animals (Table 1) and 13 impacts: fire properties; geomorphology; hydrological regimes; nutrient/mineral dynamics; light; pH, salinity, alkalinity; physical structure; facilitation; alteration of successional process; competition; hybridisation; poison, allelopathy; and disruption of ecological interactions (Table 2). This was done to prevent the inclusion of pathways and impacts not relevant to the local context. While there are many recognised pathways by which alien species are introduced, for example 32 categories listed in Hulme et al. (2008), the eight included here were deemed practical for our purposes and for management application in a protected area on a local scale. While some pathways seem counterintuitive for protected areas, all eight were deemed relevant. For example, ornamental plants are often cultivated in tourist camps and staff accommodation, or can be found at former farm houses/abandoned structures now part of a protected area. Similarly, agriculture is largely relevant for former agricultural land now incorporated into protected areas, but is also relevant where species are introduced directly adjacent to protected areas.

For each species the likelihood of being introduced by each of the eight pathways, and of having negative impacts in each of the 13 impact categories, was assessed using three primary local resources (Wells et al. 1986, Henderson 2001, Bromilow 2010), supplemented by international literature, (ISSG 2015 – Global Invasive Species Database) where the findings were deemed locally applicable by our expert judgement. We acknowledge that a species that has a diverse range of potential impacts does not necessarily equate to having the most severe impact (see Blackburn et al. 2014, Rumlerová et al. 2016). We were instead interested in quantifying the range of impacts that a species may exert on a system. This would indicate the different protected areas’ objectives that may be compromised and thus the threats requiring prioritisation.

Three options were used to describe whether a species has the potential to result in an impact described by each of the 13 categories: (i) Yes – the species has been documented to impact in this way or there is other evidence, including authors’ specialist judgement, that the species will do so. (ii) No – the species does not impact in this manner or the impact is very unlikely and has never been documented for this species. (iii) Unknown – there is too little information to make a confident decision as to whether the species may impact in this manner, but this is not implausible given the biology of its taxonomic group. To be conservative, unknown records were treated as ‘No’ records for some analyses (detailed below). For pathways of entry, all pathways for each species could confidently be scored as ‘Yes’ or ‘No’ (i.e. no species/pathway combination was scored as ‘Unknown’). In addition to the impact and pathway data we also recorded family, life-form, park invaded (Spear et al. 2011) and biome invaded (van Wilgen et al. 2008a, b).

Species were divided among authors and scored for pathways and impacts. Thereafter, a subset of species was randomly selected by category to check for consistency within, and between, categories and authors. Categories where inconsistencies were identified were systematically verified by the group for all species individually, specifically comparing entries within and between categories. The data were also checked by grouping species based on their similarity (Jaccard index) of impacts, particularly the number of impacts shared. Species that appeared to be outliers were then further examined to ensure data consistency.

Distribution of species across life-forms, families, parks, biomes, pathways and impact categories

Species were counted across life-forms, families, pathways and impact categories, to determine the status of transformer species in SANParks. For this analysis, the afore-mentioned data were transformed to binary as follows: Yes – 1, No – 0, Unknown – 0.

To determine the importance of each variable we tested for significant differences between the numbers of species counted within each category. The data were expanded into unique combinations across each category, resulting in a total of 32,718 records. The variables for impacts were maintained as Yes–No–Unknown, from which combinations including Unknown records were then excluded from the analyses. Analyses were run in R 3.0.2 (R Development Core Team 2010), using the base stats library and the chi-squared contingency table and goodness-of-fit tests.

Relationship between impact and biome, park and pathway type

A Generalised Linear Model with quasi-poisson error distribution was used to examine the relationships between the count of numbers of impacts per species, with the number of pathways by which it can invade the biomes and parks in which it occurs. The analysis was performed on all 139 species, using the glm function in R to determine the relationship of the number of impact types with the number of biomes, parks and pathways per species.

Similarity in species clusters by pathways and impacts

We assessed the occurrence of groups of species with similar pathways of introduction or similar impacts to identify groups for which particular management strategies might be effective. A statistical test for non-independence of columns, using a Spearman’s rank order correlation matrix was performed in R. The variables were weighted as follows for impacts and pathways: Yes – 1, No – 0, Unknown – 0.

Spearman rank correlations were conducted between all variables to exclude strongly correlated variables (r_s > 0.60). None of the pathway variables were highly correlated (See Suppl. material 2: Table S2: Spearman rank order correlation matrix of variables for pathways) nor were any of the impact variables (See Suppl. material 2: Table S3: Spearman rank order correlation matrix of variables for impacts).

A binary species by impact matrix, and species by pathway matrix, was constructed and the Jaccard’s index calculated in Estimate S 7.51 (Colwell 2013), and used to represent the similarity of impact and pathway types between species. Cladograms were then constructed in Primer (Clarke and Gorley 2006) using group averaging. The groupings of species were examined, noting their shared impacts and pathways, mean number of impacts and pathways, and taxonomic representation in the groups.

The transformer plant species present in parks represent 43 families, with the three most represented families being Fabaceae (20% of all the taxa assessed), Myrtaceae (9%) and Cactaceae (8%), and all other families contributing 5% or less. There were significant differences among life forms of transformer species (χ² = 118.7626, df = 8, P < 0.001; Table 3), with trees (37.4%) or tree/shrubs (17.2%) over-represented and the six other life-forms less represented (See Suppl. material 2: Table S4).

There were significant differences in the number of transformer species per biome (χ² = 155.7173, df = 7, P < 0.001; Table 3), with the fynbos (78% of the taxa assessed), then forest (48%) and savanna (45%), having the highest number of taxa (See Suppl. material 2: Table S5). The succulent karoo (6%) and arid savanna (9%) have the fewest transformer species recorded. In agreement with this, there were significant differences between the number of species recorded per park (χ² = 372.3872, df = 18, P < 0.001), the pattern thereof largely similar to the biomes. Table Mountain National Park (hereafter Table Mountain), which is fynbos dominated, including 60% of the transformers, followed by Garden Route National Park (Garden Route), which is forest and fynbos dominated including 45%, and Kruger National Park (Kruger), a savanna protected area, including 45% of the species (See Suppl. material 2: Table S6: Total count and percentage of species per biome). Golden Gate Highlands National Park (Golden Gate), Agulhas National Park (Agulhas), Addo Elephant National Park (Addo) and Camdeboo National Park (Camdeboo) have moderate levels of transformers, whereas the remaining 12 parks have significantly lower transformer species richness.

We found a significant difference in the numbers of species within each pathway category (χ² = 193.6135, df = 7, P < 0.001; Table 3). Ornamental species (78%), rivers (63%) and animal dispersion (48%) may be considered the most important pathways of introduction and spread, with numbers of species, and thus likelihood of invasion through these pathways being higher (Figure 1). Roads, paths, trails and tracks introduce less than a half (40%) of the species assessed, while the highest of the next four pathways, agriculture, contaminated materials, clothing and food is responsible for introducing less than 29% (Figure 1) (See Suppl. material 2: Table S6: Total count and percentage of species per pathway and life form).

For impacts, there is a significant difference in the numbers of species within each impact category (χ² = 346.9231, df = 12, P < 0.0001; Table 3). Nearly all 139 species are capable of direct competition with native species (Figure 2; See Suppl. material 2: Table S7: Total count and percentage of species per impact type and life form). The next most frequent types of impacts are changes to physical structure, light and then hydrological regimes. Trees and shrubs are represented in all impact categories (Figure 2).

There was no relationship between the number of impact types per species and the number of biomes (P = 0.331) or parks in which the species occurred (P = 0.131) (Table 4). The only significant relationship showed that species with more impacts are likely to be introduced by more pathways (P < 0.0001; Table 4).

The pathway cluster analysis separated the species into three main groups and four sub-groups (Figure 3; See Suppl. material 2: Figure S1 for detailed species names). The first group (group a; Figure 3) consists of 45 species that are introduced by a mean of 3.8 pathways per species, predominantly roads, paths, trails and tracks (91% of the species that have this as a pathway fall in this group only), ornamentals (82%) and rivers (78%). The second group (group b; Figure 3) is a large group of 77 species that are introduced by a mean number of 2.3 pathways, which for most species includes introduction as ornamentals (94%) and via rivers (79%). For the most part, examination of clusters at the finest scale did not reveal readily interpretable patterns. Only five out of the 13 Acacia species comprised a single cluster (group h; Figure 3), which, falling in sub-group (d) indicates their ability to disperse via four pathways.

In the cluster analysis of impact categories, three main groups were observed (Figure 4; See Suppl. material 2: Figure S2 for detailed species names). The first group of 12 species (group a; Figure 4) had fewer impacts (mean of 1.63) with the majority of species impacting via competition (67%). The second group contains 97 species with a mean number of 5.63 impacts, representing all 13 impacts. Competition (98%) and physical structure (93%) were the most important. The third group (group c; Figure 4) includes 30 species, which are characterised by a mean number of 3.67 impacts per species. All these species include competition (100%) and 93.3 % of the species impact through poison or allelopathy.

In contrast to the cladogram for pathways, there were four instances where related species clustered together based on the similarities of their impacts. All four Opuntia and two Cylindropuntia species (group h; Figure 4) were clustered, as were all six Eucalyptus species (group I; Figure 4), all seven Pinus species (group j; Figure 4) and all 13 Acacia species (group k; Figure 4). The cacti include competition and physical structure as the most important impacts. The Eucalyptus, Pinus and Acacia species include both competition and physical structure as key impacts, but also fell into the only sub-group in which fire was important.

The two most important pathways of invasion identified for transformer species into national parks included use as ornamental species and rivers. An additional two pathways appear to play a role as vectors, although to a lesser extent, including dispersal by animals and along roads, paths, tracks and trails. The results from the analyses all point to the high likelihood that many of the species currently in SANParks (~80%) were introduced for ornamentation. This can be illustrated in two parks, Kruger and Table Mountain. Kruger has a long history of plant introductions and the control of ornamental plants was first recommended in 1935 (Joubert 1986). However, by 2003 more than 250 ornamental plant species were recorded in Kruger (Foxcroft et al. 2008), including 35% of the species in our list. Work by Spear et al. (2013) showed human population density surrounding a park to be a significant driver of invasion into a park. Similarly, areas with high levels of natural vegetation along the boundary of Kruger proved to be a filter to plant invasions into the park (Foxcroft et al. 2011). The use of ornamental plants at the urban-protected area interface has been shown to increase the threat to urban protected areas such as Table Mountain (Alston and Richardson 2006). Many ornamentals appear to have few other introduction pathways, suggesting that if these species were removed from ornamental use at least some species would potentially be prevented from invading in future. Ornamental species potentially remain one of the easier pathways to manage within protected areas using, for example, policy guidelines (e.g. in Kruger, Foxcroft et al. 2015) and incentive schemes to replace alien species with native species occurring within parks and potentially those in close proximity. However, propagule pressure from outside the park is harder to control. For many of the ornamental species, rivers also form important invasion pathways, necessitating working with the nursery and landscaping industry and promoting initiatives to plant indigenous alternatives outside parks and increasing surveillance in riparian areas.

Rivers have been widely acknowledged as key dispersal vectors for invasion (Richardson et al. 2007, Esler et al. 2008, Naiman and Décamps 1997, van Wilgen et al. 2007, Jarošík et al. 2011) and more than 60% of the species in our list can disperse along rivers. Propagules transported by water flow can be widely dispersed during floods, and riparian zones and rivers banks provide highly suitable habitat (Alpert et al. 2000). Surveillance activities along rivers should be flagged as a priority area to detect new species and changes in distribution (van Wilgen et al. 2007, Forsyth et al. 2012). Trees and tree-shrubs, which are likely to be more conspicuous and easier to detect, comprise only just over half of the list, suggesting that increased effort needs to be made to detect less visible species.

Although animals are widely considered to be major dispersers of invasive plants (e.g. Vavra et al. 2007, Guerrero and Tye 2011, Kueffer et al. 2009; Oatley 1984, Gosper et al. 2005), we found only half of the species in SANParks may disperse in this manner. This is surprising as most parks have native vertebrates that could utilise alien plants. There are notable examples, however, where animals form an important dispersal mechanism for species introduced via other pathways into a park. For example, in Kruger Opuntia stricta was introduced as an ornamental plant but due to baboons and elephants utilising the fruit it became widely invasive (Foxcroft and Rejmánek 2007).

In contrast with work done in a number of studies (e.g. Pauchard and Alaback 2004, Stohlgren et al. 2013, Lonsdale and Lane 1994, Lonsdale 1999, Tyser and Worley 1992, Gelbard and Belnap 2003), we found roads and tracks to be surprisingly less frequently listed (40% of the species). As most parks have large road networks, tracks and pathways, whether for tourism or management purposes, the comparatively lower importance of this pathway is fortunate. However, there are important examples of this pathway such as alien species found along hiking and cycling trails in Table Mountain (Bouchard et al. 2015), as well as the fact that Table Mountain is an urban park within the city of Cape Town, the rapid spread of Parthenium hysterophorus along roadsides leading there and in Kruger (Foxcroft et al. 2009) and Pennisetum setaceum in Camdeboo (Masubelele et al. 2009), which is also partly an urban park. For management purposes, sections of path can be delineated for increased surveillance and fortuitously, populations confined to roadsides, should be comparatively easier to control than other pathways.

Assessing the transformer species richness per park and biome provides some insights into the potential invasibility of an area. For example, Kruger includes about 350 alien plant species, which is about 100 alien plant species more than in Table Mountain (~240) (Spear et al. 2013). However, less than 20% of the species in Kruger are transformer plants and more than a third of the species in Table Mountain and Garden Route are transformers. Moreover, the high endemism in the Fynbos biome (Rebelo et al. 2011) and high levels of habitat loss highlights that Table Mountain should be a priority for alien species management. Garden Route, containing both fynbos and forest, should likewise be considered a high management priority. Conversely, in the arid regions, parks such as Kalahari Gemsbok, Richtersveld and Augrabies Falls National Parks are less likely to become invaded by a large suite of alien plant species, of which most are likely to be restricted to rivers and drainage lines. This does not, however, indicate immunity from other invasions. Ornamental species, for example from the Cactaceae, which are introduced and nurtured in gardens could escape once established (Novoa et al. 2015). Implementing policy to prohibit the use of ornamental species in the parks therefore provides an opportunity for ongoing prevention and thereby further minimising the already low diversity of invasions in these arid parks. Species such as Prosopis spp., which are river dispersed, are highly likely to remain problematic in arid areas and the importance of the impact on hydrology, especially ground water (Dzikiti et al. 2017) highlights that this species should remain a priority.

Additional support for prioritising pathways may be gained from associations or shared traits of species that clustered together, while for some groups it is clear that prioritising one or even a few pathways will not be enough to curb spread and integrated approaches will be required. For example, all Acacia species share four of the eight pathways, with five of the 13 species sharing exactly the same pathways. These clusters together with the large body of work on Acacia (Richardson et al. 2011, Wilson et al. 2011) allow broader generalisations to be inferred for this group. Similarly, six of the seven Pinus spp. were clustered and based on current knowledge (e.g. Richardson 2006), potential pathways for other Pinus spp. may be similar.

That nearly all transformer species compete directly with native species is not an entirely unexpected result. More importantly, however, a large proportion (~70%) of the species showed the potential to impact in at least four additional ways. This most frequently included impacts such as altering hydrological regimes, changing light properties of invaded habitats, changing the physical structure of invaded areas, fire properties and succession. At a higher level in our categorisation these impacts were included as community structure, community composition and ecosystem level processes. These combined impacts can lead to cascading effects which are less easy, if at all possible, to reengineer (Meiners and Pickett 2013). Legacy effects can persist even after clearing has taken place (Larious and Sudding 2013) and can influence the ability of a system to recover following control efforts and whether additional interventions are required.

Four of the most represented naturalised genera globally were recorded in our list (Pyšek et al. 2017), and include some of the most frequently listed impacts. The Opuntia and Cylindropuntia spp., Eucalyptus spp., Pinus spp. and Acacia spp. each formed clusters of similar impacts. Excluding direct competition, physical structure was listed as the most important impact for the Cactaceae. Due to the dominance of trees and tree-shrubs in the transformer group, these species made up about half of the direct competition category and 40% of the species that can potentially change physical structure. These include the Eucalyptus, Pinus and Acacia species, but for these species specifically, impacts also included light, hydrology and fire. For example, species in the Fabaceae can significantly increase biomass and intensity of fires (van Wilgen and Richardson 1985), compounding long-term soil erosion (Scott et al. 1998) and other ecosystem level impacts on biogeochemistry (Yelenik et al. 2004). Therefore these species, in particular, are important and should be prioritised. In addition, groups of similar species may be advantageous in that similar management actions may be possible across the species.

By assessing each species against the eight pathway and 13 impact categories, we aimed to determine a relative risk profile for each species that could assist in determining the threat that the species posed to a protected area. The significant relationship between pathways and impacts indicates that the more pathways a species can use to disperse, the higher the likelihood that the species will become problematic.

For protected areas in our study, a species introduced by multiple pathways can be expected to be distributed over a larger area and should be given a higher priority. For example, when spreading along rivers, riparian vegetation may be displaced, causing substantial changes to the geomorphology, vegetation and community structure and composition (Hejda et al. 2009), species communities and river bank collapse, while simultaneous spread across the landscape more broadly (e.g. grass or shrublands to alien tree dominated systems) can alter ecosystem processes (Raizada et al. 2008, Martin et al. 2009), fire regimes (e.g. Table Mountain, Forsyth and van Wilgen 2008, and Andropogon gayanus in Kakadu National Park in northern Australia, Rossiter et al. 2003), hydrology and nutrient cycling/biogeochemistry (e.g. Carbon-Nitrogen-water-leaf litter interactions, Ehrenfeld 2003).