Introduction

Urbanisation is a major driver of global environmental change, with urban infrastructure and human populations encroaching on and modifying natural ecosystems (Faulkner 2004; Coleman 2017; Cresswell and Murphy 2017). Urbanisation leads to the degradation and removal of natural spaces for plants and animals and the introduction of invasive of species which create new biological interactions and alter the physical environment (McDonnell et al. 1993; Kinzig and Grove 2001; Hamer and McDonnell 2008; Štajerová et al. 2017). Wetlands are an ecosystem type particularly threatened by urbanisation. Wetland ecosystems are the most rapidly declining ecosystems world-wide (Balmford et al. 2002; Sieben et al. 2016) and even international treaties such as the Ramsar Convention and identification of World Heritage areas, that recognise their importance for the conservation of biodiversity, have failed to halt this decline, which is anticipated to accelerate further with the additional impacts of climate change (Nicholls et al. 1999; Burgin et al. 2016). In Australia, natural wetlands have been lost from much of the continent, with areas of particular concern including the south-eastern and south-western regions of the continent, where increasing urbanisation means wetlands are lost or highly modified to accommodate concentrated anthropogenic land uses (Burgin et al. 2016). On the Swan Coastal plain in south-west Western Australia, 70% of wetlands have been lost by activities that facilitate urban development, including drainage, clearing of native vegetation and road construction (Lemly et al. 2000).

Management and protection of remaining wetlands is essential for conservation in urban areas because of their importance to the maintenance of urban biodiversity (Burgin et al. 2016). Wetland habitats are especially vulnerable to invasive plants as up to a quarter of the most invasive plant species worldwide are wetland species, which then benefit from accumulation of materials, such as sediment and nutrients, in wetland habitats (Zedler and Kercher 2004). Invasive plants are a threat to ecosystems globally and can dominate disturbed or sensitive vegetation communities, reducing plant diversity and structural heterogeneity and impacting ecosystem function (Schweiger et al. 2018) and retention of biodiversity (Stein et al. 2014). Invasive wetland weeds are particularly prone to forming monocultures (Zedler and Kercher 2004) and these monocultures can reduce habitat quality for fauna by reducing structural complexity and resource availability, impacting predation pressure, thermal quality, food availability and ultimately the overall suitability of the habitat (Huey 1991; Webb et al. 2005; Lee et al. 2006). Invasion of wetlands by weeds typically reduces mammal, bird, fish and insect diversity and abundance, and in particular impacts on species that are habitat specialists (Zedler and Kercher 2004).

Large reptiles are the top predators of many Australian ecosystems, with diverse effects on ecosystem structure and function (Pianka 1986; Kuch et al. 2005; Read and Scoleri 2015; Doody et al. 2021). Tiger snakes (Notechis scutatus) are large, venomous elapid snakes of southern Australia that are typically affiliated with wetland habitats. These wetlands support populations of frogs (the primary prey of tiger snakes) and waterbodies that are used by tiger snakes to evade predators (Aubret 2004; Lettoof et al. 2020). Tiger snakes can be a conspicuous top predator in anthropogenic landscapes where their presence and abundance is sensitive to an intact food web and habitat availability (Sergio et al. 2008); thus they are considered bioindicators of wetland health (Lettoof et al. 2022). Western tiger snakes (N. s. occidentalis) persist in a subset of urban wetlands within the urban sprawl of Perth city (Lettoof et al. 2022).

Some of the urban wetlands inhabited by western tiger snakes are dominated by introduced kikuyu grass (Cenchrus clandestinus), a globally recognised invasive species, particularly in wetlands (Bird et al. 2013; Boon and Tesfamichael 2017). The aggressively competitive nature of many introduced grass species and their tolerance of above-ground disturbance and sedimentation makes them particularly successful invasive species (Zedler and Kercher 2004; Ferdinands et al. 2005; Sieben et al. 2016). Abundant kikuyu lowers native plant species richness and cover (Gaertner et al. 2011) and has potentially similar impacts on fauna. For example, in wetlands in northern Australia, birds avoid areas dominated by invasive para grass (Urochloa mutica) in favour of more heterogenous natural habitat (Ferdinands et al. 2005). However, not all fauna is negatively affected by the invasion of exotic grasses (Douglas et al. 2006). For some reptiles, habitat structure is more important than plant species composition (Garden et al. 2007) and populations may persist in highly modified habitats if structural components are retained within the landscape. For example, the presence of structurally complex habitats created by the invasive woody shrub lantana (Lantana camara) supports large numbers of a rare species of shadeskink (Saproscincus rosei; Virkki et al. 2012). Invasive grasses such as kikuyu may also provide suitable habitat for the frogs on which tiger snakes prey (Zavaleta et al. 2001; Maerz et al. 2005) and prey availability is an important driver of snake abundance and habitat use (McCauley et al. 2006; Battles et al. 2013; Zipkin et al. 2020).

A broadly inhospitable matrix of urban area restricts tiger snakes on the Swan Coastal Plain to the wetlands they inhabit (Lettoof et al. 2021b), with little opportunity for individuals to move between habitat patches. Consequently, the quality of these wetlands and availability of prey are factors crucial for maintenance of viable populations of this top predator. An over-abundance of kikuyu appears to have substantially changed the vegetation composition and habitat structure at many of these wetlands. The flow-on effects for prey abundance, predation risk and thermal quality for tiger snakes has not been determined so here we investigate the impacts of kikuyu invasion on these variables for wetlands, by comparing a range of biotic and abiotic metrics in kikuyu-dominated wetlands against wetlands that are dominated by native vegetation communities. We quantify vegetation structure, prey availability (frog abundance), predation risk (for juveniles) and thermal quality of available microhabitats to better understand how kikuyu might impact wetland fauna and assist land managers to better manage degraded urban wetlands to maximise wetland ecosystem function and faunal biodiversity.

Methods

Study sites

We examined western tiger snake habitats at two wetlands dominated by kikuyu and two wetlands dominated by native vegetation, all within 50 km of the Perth CBD, in south-west Western Australia (Fig. 1). Western tiger snake populations have been previously documented as occurring at these wetlands (Lettoof et al. 2021a). At Herdsman Lake (HL; 31.92°S, 115.80°E) and Kogolup Lake (KL; 32.12°S, 115.83°E) kikuyu was the dominant ground cover vegetation while at Loch McNess in Yanchep National Park (Y; 31.54°S, 115.68°E) and Black Swan Lake (BS; 32.47°S, 115.77°E) native vegetation predominated. Although the total area of the wetlands varied greatly the area of potential tiger snake habitat at these wetlands was similar for each site, ranging from 49.1 ha (HL) to 59 ha (Y; Cornelis et al. 2023). We were limited to four wetland sites due to the requirement for sites that (1) had a distinctly kikuyu or native species dominated vegetation, (2) a substantial tiger snake population, (3) were accessible for our study and (4) occurred on the Swan Coastal Plain within the urban sprawl of the Perth metropolitan area. Within these four wetland sites, we identified a priori three broad riparian habitat categories occurring within 50 m of the waterbody; open grassland (abundant ground cover of either uncut kikuyu grass or native grasses with no canopy cover or mid-story vegetation), sedge (mid-story level dense native sedges or uncut kikuyu grass growing against bullrush bordering waterbodies), and woodland habitats (canopy cover and understory vegetation, occasionally with mid-story vegetation; Fig. S1) and use the terms ‘kikuyu’ or ‘native’ to distinguish the broad wetland communities (i.e. two kikuyu dominated wetlands and two native-vegetation dominated wetlands) from the finer-scale riparian habitat categories (open, sedge and woodland) within these sites.

Fig. 1
figure 1

Wetlands that were surveyed in the Swan Coastal Plain. BS black swan lake, HL herdsman lake, KL kogolup lake, Y yanchep national park. Red font indicates wetlands classed as dominated by native vegetation, blue font indicates wetlands dominated by kikuyu grass. Satellite images were obtained from Google Earth in 2022. Scale bar = 500 m

Full size image

Habitat structure

To quantify habitat structure, we recorded vegetation features along two 100 m transects in each habitat type at each wetland. Every 10 m along the transect we recorded canopy cover using a spherical densiometer (Forest Densiometers, Model A), light availability with a digital lux meter (DR.meter LX101BS, when the reading in ambient unshaded light was ca. 720,000 lx), the number of trees with stems > 150 mm circumference at breast height (1.4 m) within a 2 m radius, the number of woody plants with stems < 150 mm circumference at breast height within a 2 m radius, the height of the tallest understorey plant, an estimate of the percent ground cover within a 2 m radius and the vegetation density. Vegetation density was quantified by inserting a 50 cm × 50 cm board inside the vegetation (2 m from the observer), taking a photograph, and then using ImageJ 1.x (Schneider et al. 2012) to calculate the surface area of the board not obscured by vegetation. If the vegetation was too dense to insert the board we recorded vegetation density as 100% (Fox et al. 1996; Colman et al. 2014, 2015).

All statistical analyses were conducted in R studio version 1.4.1103 (RStudio Team 2020). We used principal component analysis (PCA) to examine associations between the three a priori riparian habitat types (open, sedge and woodland) and vegetation communities (kikuyu or native vegetation) based on the measured habitat characteristics. The FactoMineR (Lê et al. 2008) and factoextra (Kassambara and Mundt 2020) packages were used to generate principal components (PCs); PCs with eigenvalues > 1 were considered useful for inference (Roznik and Reichling 2021). An analysis of deviance was then used to compare the PCs of the riparian habitat types and to compare sites with kikuyu to those with native vegetation, with Tukey’s post-hoc pairwise comparisons (Roznik and Reichling 2021) achieved with the emmeans package (Lenth et al. 2018).

Prey availability

We measured the abundance of prey (frogs) in the different vegetation communities by conducting standardised 10 min active visual searches in the open, sedge and woodland riparian habitat types at the four wetlands. The search commenced in the centre of each habitat type and observers walked a steady pace while counting individual frogs detected with a spotlight. All transects were at least 100m apart, and surveyed once per night in a randomised order; Yanchep and Black Swan (native) were surveyed for five nights each, and Kogolup Lake and Herdsman Lake (kikuyu) were surveyed on four nights each, for a total of 54 visual searches. All transects were selected based on known presence of tiger snakes (JC & DL, pers. obvs.). Frogs were identified to species. Only a single wetland was surveyed each night to mitigate any temporal effects caused by time of night. Surveys were conducted in October and November (2019), which is peak calling season for the dominant prey species Litoria moorei, Litoria adelaidensis and Limnodynastes dorsalis (Hoskin et al. 2015; Lettoof et al. 2020). All surveys began after 19:00 on evenings when ambient temperature at this time was > 15 °C and were completed before 22:30.

Generalised linear mixed-effects models (GLMMs; lme4 package, Bates et al. 2014) with Tukey’s post-hoc pairwise comparisons were used to assess the differences in frog abundance among vegetation communities and habitats. We used a two-step process for assessing differences: Firstly, frog abundance was broadly compared between vegetation communities and habitat types; fitting a GLMM (Poisson error structure) with total frog count of the two most commonly observed species of frog present at all sites (slender tree frog and motorbike frog) per transect per night as the response variable and vegetation community (kikuyu or native vegetation) and habitat type (open, sedge or woodland) as fixed effects. Site and transect were included as random effects to account for the repeated surveys at each site on different days, after Crawley (2007). We also ran these models with just slender tree frog and just motorbike frog count as the response variable, respectively. The second approach was to compare the abundance of slender tree frog and motorbike frog with habitat structure variables (e.g. canopy and ground cover, light, tree density, understory cover etc.) to determine which specific structural components of a habitat type were influencing frog abundance. We fitted a GLMM (Poisson error structure) with frog count data as the response variable, the habitat structure variables as predictor variables, and site as a random effect. Predictor variables were first scaled and centred to improve model fitting and correlated variables were removed to reduce multicollinearity, retaining variables with variance inflation factors < 5 and pairwise correlations < 0.7, identified using the usdm and psych packages (Naimi 2015; Revelle 2015; Lettoof et al. 2020). For correlated variables the least ecologically relevant predictors were removed, after Dormann et al. (2013). The global model included canopy cover, vegetation density, vegetation height and woody vegetation as predictor variables. Model fit was then assessed with pseudo-R2 values calculated with the glmmADMB package (Skaug et al. 2013). Following this, model selection was performed using the dredge function from the MuMIn package (Barton 2015) and all sub models were ranked according to Akaike’s information criterion, with a sample size correction (AICc). All models with ΔAICc < 2 were considered useful for inference.

Predation risk

We assessed the differences in predation risk for young tiger snakes among vegetation communities and habitats using clay models. We moulded 120 soft clay models, representing juvenile tiger snakes because snakes are most vulnerable to predation when they are small (Lima and Dill 1990; Webb et al. 2005), and 120 spherical control balls from nontoxic Plastiplay™ brown modelling clay (Fig. S2A). Total body length of the artificial snakes was ca. 25 cm, width 1 cm and they were moulded into a sinusoidal shape. Control balls were 4 cm in diameter and used to determine if predators discriminated foreign objects from artificial snakes (Nordberg and Schwarzkopf 2019). The artificial snakes and balls were deployed at the four study sites in each of the three different habitat types. In each habitat type, ten artificial snakes were placed in exposed areas to simulate basking sites, and ten were hidden under vegetation to simulate refuge sites. Artificial snakes were separated by 5 m, with each model accompanied by a control ball at a distance of 25 cm. All clay models were checked every day for a 5 day period and marks left by predators were counted, scored by intensity (depth) of attack and the predators were identified as birds, rodents or cats, based on the shape of the marks left by the predators (Webb and Whiting 2005) or classified as unknown if the clay models were missing. Small shallow marks, often left by house mice (Mus musculus), were scored as exploratory rather than predatory as they did not represent a realistic predatory attempt on a juvenile tiger snake; these were therefore omitted from analyses (Fig. S2B). Experiments were conducted in autumn and repeated in spring for a total of 10 days of sampling at each site.

We used a Student’s t-test to compare the difference in total predation rate between snake models and control balls. We used the same two-step process used to analyse frog abundance data for assessing differences in predation counts both among vegetation communities and habitat types and for habitat structure variables. The response variable for these GLMMs was whether a model had experienced a predation attack (binomial yes/no). The global model included vegetation density, ground cover, light availability and vegetation height as predictor variables.

Thermal quality of vegetation

To determine if kikuyu grass or native vegetation differ in their thermal quality we monitored the temperatures within potential tiger snake shelter sites comprised of these either of these two types of vegetation with operative temperature models constructed using 30 cm long sections of copper pipe (32 mm diameter, wall 1 mm thick) filled with water and painted to approximate the reflectivity of a tiger snake (Seebacher and Shine 2004; Lutterschmidt and Reinert 2012). In the centre of each model we suspended a small temperature logger (Thermochron iButton) which recorded temperature at 10min intervals. The temperature logger was waterproofed (Plasti Dip) and wrapped in plastic film. Temperatures of the operative models were calibrated against tiger snake carcasses over a range of thermal and radiative conditions (Fig. S3). We deployed models in the field and calculated their thermal properties via a custom-written VB program (Visual Basic V6; Cooper and Withers 2004).

Three operative temperature models were deployed at each wetland, one in each of the three habitat types, placed within what we refer to as a shelter site (deep within the vegetation where we have observed tiger snakes retreating or emerging; pers obs.). Data were recorded at 10 min intervals for 7 days every month for 12 months. We calculated the mean, maximum and minimum operative environmental temperature (Te) of the sheltered microhabitats. Then we calculated the thermal quality (de) of tiger snake shelter sites as the absolute value of how much Te in shelter sites deviated from the thermal set-point range (Tset) for tiger snakes (Blouin‐Demers and Weatherhead 2002), determined by Ladyman and Bradshaw (2003) for snakes from Herdsman Lake in a thermal gradient as 24.0 ± 0.9 °C to 30.9 ± 0.3 °C. The resulting data were then grouped into four seasons and we compared the differences in thermal quality between sites with kikuyu grass and native vegetation and the habitat types within sites. To compare seasonal habitat temperature characteristics between vegetation types, we used a GLMM (Gaussian error structure) with the temperature characteristic as the response variable, vegetation type, habitat type and season as fixed predictor variables and site as a random effect to account for repeated measurements at each site. Tukey’s post-hoc pairwise comparisons were used to examine differences between variables.

Results

Habitat structure

Three PCs were retained (eigenvalue > 1) from the PCA analysis of habitat structure, collectively accounting for 73.7% of the total variance. PC1 explained 35.5% of the variance and was associated with canopy cover (Canopy), number of trees (Trees), woody vegetation (Woody) and low light availability (Lux; Table S1). PC2 explained 23.8% of the variance and was associated with ground cover (Ground), vegetation density (Density) and canopy cover. PC3 explained 14.4% of the variation and was associated with understory height. Collectively, the structure of kikuyu grass habitats was more homogenous than native vegetation habitats (Fig. 2) and differed significantly for PC2 (F5, 238 = 495, P < 0.001). The structure of our a priori riparian habitat types were significantly different for PC1 (F5, 234 ≤ 149, P < 0.001), although no pairwise differences were detected for kikuyu or native vegetation types (P ≥ 0.526). For PC3, open vegetation types were statistically similar (P = 0.856) for native and kikuyu dominated sites, but sedge and woodland riparian habitats differed for sites dominated by kikuyu compared to native vegetation (P < 0.001).

Fig. 2
figure 2

PCA ordination of habitat characteristics in western tiger snake (Notechis scutatus occidentalis) habitats from four different wetlands, dominated by either invasive vegetation (Herdsman Lake and Kogolop Lake) or native vegetation (Black Swan Lake and Yanchep National Park) in the Perth region, Western Australia. A The ordination of the habitat characteristics of the kikuyu grass and native vegetation plant communities, B ordination of the different habitat types within the wetlands KO kikuyu open, KS kikuyu sedge, KW kikuyu woodland, NO native open, NS native sedge, NW native woodland and. PC1 explains 35.5% and PC2 21% of the total variance. Arrows denote the influence of specified habitat characteristics and ellipses denote 80% of the spread of the groups

Full size image

Prey availability

Two species of frog were frequently observed during surveys at all four study sites: slender tree frog (Litoria adelaidensis) and motorbike frog (Litoria moorei). Only two individuals of the rattling froglet (Crinia glauerti) were detected at a single site (Yanchep) and the moaning frog (Heleioporus eyrie) was not detected at Herdsman Lake (Table 1). Consequently, these two species were excluded from all subsequent analyses and we only investigated the abundance of slender tree frog and motorbike frog.

Table 1 Total number of frogs seen during visual surveys at four wetlands in Perth, Western Australia. Wetlands include HL herdsman lake, KL kogolup lake, BS black swan lake and Y yanchep national park. Frog species are the slender tree frog (Litoria adelaidensis; STF) and motorbike frog; MBF)
Full size table

Total frog abundance (slender tree frog and motorbike frog combined) was not significantly influenced by vegetation community (kikuyu grass or native vegetation) or habitat type (open, sedge and woodland; χ2 2 ≤ 3.340, P ≥ 0.188); however, the interaction between vegetation community and habitat type did significantly influence total abundance of frogs (χ2 2 ≤ 26.55, P < 0.001), with a greater abundance of frogs in the native open habitat compared to the other native habitat types and the open kikuyu habitat; the open kikuyu habitat also had a significantly lower abundance of frogs than the kikuyu woodland habitat (post-hoc comparison P ≤ 0.003; Fig. 3). Further, there were significant differences in the abundance of frogs among habitat types within vegetation communities. Native open habitats had significantly more frogs than the native sedge and (NS) and native woodland habitats (NW; post-hoc comparisons P < 0.001) in sites dominated by native vegetation. In At sites dominated by kikuyu grass communities, total frog abundance was significantly lower in the open habitat compared to the kikuyu woodland (KW; post-hoc comparisons P = 0.001).

Fig. 3
figure 3

Model predicted mean (± SE) frog abundance in wetlands of Perth, Australia, dominated by either kikuyu grass or native vegetation. A Shows patterns in total frogs (all species summed), B shows patterns for slender tree frogs (Litoria adelaidensis) and C shows patterns for motorbike frogs (Litoria moorei). * denotes a significant difference between habitat types within vegetation communities

Full size image

We observed larger numbers of slender tree frogs in native vegetation compared to kikuyu-dominated vegetation, although this was not significantly different (χ21 = 2.893, P = 0.088). However, habitat type and the interaction between vegetation community and habitat type significantly influenced abundance of the slender tree frog (χ2 2 ≤ 8.489, P < 0.040) with the native woodland having significantly lower abundance of slender tree frog than the native sedge and open habitat types (post-hoc comparison P ≤ 0.003). Motorbike frog abundance was not influenced by vegetation community or the interaction between vegetation community and habitat type (: χ2 2 ≤ 2.736, P ≥ 0.254); however habitat type did significantly influence motorbike frog abundance (χ22 = 7.940, P = 0.018) with woodland habitats having a greater abundance of motorbike frogs than open habitats (post-hoc comparison P ≤ 0.026). The abundance of both species was generally the same, for equivalent habitat types, for kikuyu and native vegetation dominated sites (post-hoc comparisons; P ≥ 0.072), except the slender tree frog was significantly more abundant in open habitats in sites dominated by native vegetation. Canopy cover, vegetation density and ground cover (%) were important predictors of both species; slender tree frog abundance was negatively associated with canopy cover and ground cover, and positively associated with vegetation density (Table 2), whereas motorbike frog abundance was positively associated with canopy cover, ground cover and understorey height and as was negatively associated with vegetation density (Table 2).

Table 2 Habitat characteristics influencing the abundance of slender tree (STF; Litoria adelaidensis) and motorbike frogs (MBF; Litoria moori). Results from generalised linear mixed-effects model indicating the best predictor variables for submodels with ΔAICc. Results from generalised linear mixed-effects model indicating the best predictor variables for submodels with ΔAICc < 2 and the null model identifying the strongest predictors of abundance based on ΔAICc and weight in four wetlands in the Perth region, Western Australia. The estimate and standard error (estimate) and P-value are reported for each predictor variable. The conditional R2 (R2GLMM(c)), degrees of freedom (df), log-likelihood (logLik), Akaike’s Information Criterion (AICc), difference in AICc between models (ΔAICc), and weight are reported for each model. Response indicates whether frog abundance had a positive ( +) or negative (-) response to the habitat characteristics. Significant p-values are in bold
Full size table

Predation risk

Of the 1200 observations of clay models, 57 predation attempts were recorded for artificial snakes and 17 for control balls. Artificial snakes were significantly more likely to be predated than control balls (t958 = 3.98, P < 0.001). Most predation marks could be identified as being a consequence of either birds (54%) or rodents (40%; Table 3). Due to the infrequency of predation attempts, we could not investigate differences in the frequency of different predator interactions with the models. All predation data were therefore pooled into a single response variable, ‘predation’.

Table 3 Predation attempts on artificial snakes in vegetation communities that are either dominated by either introduced kikuyu grass or native plant species. Predation attempts were identified as being made by a bird, rodent, cat, or unknown species, and were quantified in three habitat types (classified as open, sedge, and woodland)
Full size table

There was no significant difference in the probability of predation on artificial snakes between vegetation communities (χ2 1 = 1.27, P = 0.257). The probability of predation was significantly different among habitat types (χ2 2 = 6.86, P = 0.032) although post-hoc comparisons of pairwise habitats showed no significant differences (P ≥ 0.270; Fig. 4). The structural variables of light availability and occurrence of woody vegetation were the best predictors of predation probability for artificial snakes, as these variables occurred in most of the five top models (Table 4). Predation probability increased with light availability and decreased with woody vegetation count.

Fig. 4
figure 4

Model predicted mean (± SE) probability of predation for clay model snakes in wetland vegetation communities dominated by either introduced kikuyu grass (Cenchrus clandestinus) or native vegetation in Perth, Australia. Predation attempts were quantified in three habitat types (open, sedge and woodland)

Full size image
Table 4 Habitat characteristics influencing the predation risk for clay models of juvenile snakes. Submodels and the null model identifying the strongest predictor variables of clay model snake predation probability based on ΔAICc. Submodels and the null model identifying the strongest predictor variables of clay model snake predation probability based on ΔAICc < 2 and weight in four wetlands in the Perth region, Western Australia. The estimate and standard error (estimate) and P-value are reported for each predictor variable. The conditional R2 (R2GLMM(c)), degrees of freedom (df), log-likelihood (logLik), Akaike’s Information Criterion (AICc), difference in AICc between models (ΔAICc) and weight are reported for each model. Response indicates whether predation probability had a positive ( +) or negative (−) response to the habitat characteristics. Significant p values are in bold
Full size table

Thermal quality of vegetation

Calibration of the operative temperature models indicated they accurately represented the temperatures an actual tiger snake would attain (see description in Online Resource 1 and Fig. S3). Most of the thermal variables were highly correlated with demax (R ≥ 0.90), except for Te, max (R = 0.68) and de0 (proportion of time de was equal to 0; R = 0.43), so we only analysed vegetation and riparian habitat effects for these three variables. The Te,max differed seasonally (F3 = 413, P =  < 0.001), being highest in summer and lowest in winter (Fig. 4). The Te,max in habitat types did not differ within each season (post-hoc comparisons P ≥ 0.190), except for at sites dominated by native vegetation, when open habitat had significantly higher Te,max than sedge habitat during summer (post-hoc comparisons P < 0.001; Fig. 5). The dominant vegetation type (kikuyu or native) had no influence on demax (F1 = 0.05, P = 0.837) but there was a significant difference in demax among habitat types (F2 = 31.8, P < 0.001) with the highest demaxs occurring in open habitats and the lowest demaxs in the woodland habitats (Fig. 5). Season also significantly influenced demax (F3 = 275.53, P < 0.001) which was lowest during summer, highest during winter and similar in autumn and spring (Fig. 5), with no interaction (F6 = 0.21, P = 0.972; Fig. 5). Neither the dominant vegetation type (F1 = 0.04, P = 0.857) or riparian habitat type (F2 = 1.69, P = 0.184) influenced the mean proportion of time that the thermal quality of the vegetation fell within Tset (de0). However there was a significant interaction between vegetation community, habitat type and season (F6 = 3.04, P = 0.005); within a season, there were no differences between equivalent habitat types in native or kikuyu vegetation (post-hoc comparisons P ≥ 0.240) but there were some differences between habitat types within the same vegetation community e.g. the highest mean de0 occurred in kikuyu-domianted woodland during summer and differed from open and sedge habitats in the same season (post-hoc comparisons P ≤ 0.001; Fig. 5).

Fig. 5
figure 5

Temperature parameters (A maximum operative environmental temperature; Te, B maximum deviation of environmental temperature, demax and C proportion of time the environmental temperature in sheltered microhabitats fell within the body temperature set-point range; de0) for tiger snakes (Notechis scutatus occidentalis) at four wetlands in Perth, Western Australia recorded during four seasons. Parameters were measured in sheltered microhabitats in three habitat types at sites where the vegetation is dominated by either introduced kikuyu grass or native vegetation

Full size image

Discussion

We investigated a suite of habitat factors likely to influence populations of western tiger snakes in urban wetlands where native vegetation communities have been largely replaced by a single invasive species, kikuyu grass. Our results indicate that habitat compositional heterogeneity in wetlands with kikuyu grass has been reduced and that prey availability may be reduced in areas that have become dominated by kikuyu. However, despite the reduction in bare ground and increase in vegetation density at sites invaded by kikuyu, the transition from native vegetation to invasive grass appears to have had little effect on the predation risk for juvenile snakes or the thermal properties of shelter sites. Together, these findings indicate that the current extent of invasion of kikuyu grass in wetlands of the Swan Coastal Plain offers an overall similar habitat quality for tiger snakes to natural vegetation and is unlikely to substantially impact population persistence in these urban wetlands.

Habitat structure

Diverse habitat structure is a primary driver of biodiversity worldwide (Culbert et al. 2013; Boeye et al. 2014; von Takach et al. 2020). Our results indicated that habitat structure is similar between kikuyu-dominated and native-dominated vegetation, but is different between habitat types (open/sedge/woodland), supporting our a priori determination of habitat categories for the riparian vegetation within these wetlands. Increased vegetation density, ground cover and overall homogenisation drove the observed differences in the structure of kikuyu dominated vegetation communities compared to native vegetation equivalents. Kikuyu grows as a dense matrix of stems, which facilitates its colonisation and reduces inter-plant distance compared to native grasses, resulting in less bare ground and reduced plant diversity, transforming native vegetation communities into a monoculture (Gonzalez 2009; Litt and Steidl 2011; Lindsay and Cunningham 2012; Bradshaw et al. 2013; Abom et al. 2015).

The domination of an environment by a monoculture of a single invasive grass can have adverse effects on the ecosystem. For example, the invasion of gamba (Andropogon gayanus) and para (Urochloa mutica) grasses in northern Australia results in higher fuel loads and higher intensity fires than in uninvaded habitats, due to higher biomass and lower moisture content of the invasive species (Rossiter et al. 2003; Douglas and O’Connor 2004; von Takach et al. 2022). The most common impact of habitat structural homogeneity from invasive grasses, however, is the loss of native plant (Stephens et al. 2008), invertebrate (Douglas and O’Connor 2003) and vertebrate (Ferdinands et al. 2005; Cook and Grice 2013; Stanton et al. 2018) biodiversity—largely attributed to a loss of suitable habitat usually provided by structurally complex vegetation communities. Similar to these studies, we found that the invasion of kikuyu grass changed the habitat structure of riparian wetland vegetation. While we quantified the impact of this change on two common frog species, we encourage further research into how this invasion may affect other wetland biota in these urban wetlands.

Prey availability

The primary difference we found between kikuyu and native-dominated vegetation was the native open habitats had a greater abundance of slender tree frog and motorbike frog (combined), than the kikuyu open equivalent. In general, current levels of kikuyu in the wetlands we investigated made no difference to the total abundance of frogs, despite frogs being amongst the most sensitive vertebrate taxa to anthropogenic habitat modification (Nowakowski et al. 2017). However, the abundance of slender tree frog and motorbike frog among habitat types differed within and between wetlands dominated by kikuyu compared to native vegetation. Slender tree frog had higher abundance in the native open habitats, with the lowest abundance found in open kikuyu habitats, consistent with the species having a negative response to anthropogenic disturbance (Liu et al. 2021). The motorbike frog however, is one of the most tolerant of Australian frogs to habitat modification; indeed it often benefits from anthropogenically altered environments (Liu et al. 2021). It is therefore not surprising that motorbike frogs are tolerant of kikuyu grass invasion. Their negative association with vegetation density explains their highest abundance in the woodland habitats and is presumably a trait that enhances their overall success in highly human-modified environments. Compared to slender tree frogs, motorbike frogs also have a larger body size, broader geographical distribution and larger clutch size, all traits associated with tolerance of modified habitats (Nowakowski et al. 2017; Liu et al. 2021). As motorbike frogs contribute a substantial proportion of tiger snake diet in these wetlands (Lettoof et al. 2020), their presence may explain why tiger snakes have persisted in these wetlands. However, the reduction in abundance of smaller (34–47 mm) slender tree frogs likely reduces prey availability for young tiger snakes in kikuyu-dominated habitats and may be of concern for the recruitment of juvenile snakes into populations.

We observed the lowest abundance of all frogs in the open kikuyu habitat, although it is possible that the dense structure created by kikuyu grass made it difficult to detect all individuals (Heard et al. 2008; Vences et al. 2008). This habitat type is a monoculture of dense grass, and structurally very different from the open native vegetation equivalent which is comprised of several structurally diverse plant species and interspersed with bare ground. Plant structure and composition can have substantial impacts on amphibian ecology as vegetation influences species interactions including predation and foraging opportunities, reproduction, disease and abiotic conditions such as temperature (Burrow and Maerz 2022). Invasive grass monocultures usually result in a loss of abundance and diversity of native birds (Catling 2005; Skórka et al. 2010), reptiles (Valentine et al. 2007; Hacking et al. 2014), frogs (Grant and Samways. 2016; DeVore and Maerz 2014; Falaschi et al. 2020) and rodents (Sammon and Wilkins. 2005) globally. The abundance of tiger snakes two dominant prey species was highest in native open, native sedge and both woodland habitats, where the greater structural complexity and woody vegetation is likely favourable (Landsman and Bowman 2017), potentially reducing predation risk (DeVore and Maerz 2014; Norbury and van Overmeire 2019). As large expanses of riparian wetland vegetation in Perth can be kikuyu grass monocultures (Hill et al. 1996; Department of Environment and Conservation 2012), the persistence of these frogs and their tiger snake predators could be facilitated by the planting and maintenance of riparian grasses (Schoenoplectus spp.), sedges (Gahnia decomposita and/or Lepidosperma longitudinale) and native woodland trees (Banksia, Melaleuca and Eucalyptus spp) to supplement and supress the dense kikuyu vegetation (e.g. Burrow and Maerz 2022).

Predation risk

Predation attempts on juvenile snake clay models occurred at low frequency, with birds and rodents the predominant predators. The low frequency of predation attempts (4.75% of 1200) on artificial juvenile tiger snakes is comparable to the predation rates (7.44%) observed for artificial juvenile broad headed snakes (Hoplocephalus bungaroides; Webb and Whiting 2006) and uropeltid snakes (5%; Cyriac and Kodandaramaiah 2019). Using clay models as a proxy for live animal interactions can have limitations (Bateman et al. 2017), yet we consider our predation attempts to be an accurate representation of at least bird and rodent predation at our study sites, as predation attempts were significantly higher on snake models compared to control balls. The depredated model snakes had strike marks primarily on the head and tail and occasionally had been flipped by an avian predator and pecked on the ventral surface, typical strategies for predators of snakes (Webb and Whiting 2006). It is likely that predation of juvenile tiger snakes at these sites is in reality a rare occurrence and indeed the observed predation rate could be an overestimate as live juvenile snakes spend limited time basking (Webb and Whiting 2005) and artificial snakes do not respond with anti-predator behaviours.

Environmental habitat structure influences the predation risk for reptiles by determining the protective qualities of refuge sites or structural features predators use to hunt (Hawlena et al. 2010; Martin and Murray 2011; Steidl et al. 2013). The considerable habitat structural changes caused by invasive grasses in wetlands can modify trophic interactions sufficiently to impact on overall prey survivorship (DeVore and Maerz 2014). The dominant predators at our study sites are largely visual hunters, thus the artificial snakes were probably more conspicuous in habitats that have reduced structural complexity and increased light availability (Daly et al. 2008; Sato et al. 2014). This is consistent with other studies that found greater predator diversity along open habitat edges (Anderson and Burgin 2008). Birds specifically use elevated perches along habitat edges to forage in open environments (Hansen et al. 2019) and can show a preference for hunting in more homogenous habitats (Hawlena et al. 2010). Our results suggest that kikuyu dominated vegetation is not influencing predation risk for juvenile tiger snakes but riparian vegetation with woody stemmed vegetation and structural complexity provides safer habitats for young tiger snakes, highlighting the value of maintaining sedge and woodland habitats in these urban wetlands.

Thermal quality of vegetation

Kikuyu grass dominated wetland habitats had similar thermal properties to those dominated by natural vegetation. Similarly, Abom et al. (2015) found that the thermal quality within grader grass (Themeda quadrivalvis) did not differ from within native vegetation, offering thermally equivalent shelter sites for a number of skink and snake species. As the thermal environments within these vegetation communities are so similar, despite a change in habitat composition, the way tiger snakes thermoregulate and exploit the thermoregulatory opportunities in these environments has remained unaffected (Cornelis 2021). Not all vertebrate species are disadvantaged by the invasion of exotic grasses (Malo et al. 2012; Lindenmayer et al. 2017); the thermal quality and dense structure provided by kikuyu grass may be particularly suitable for snakes as they can easily move through the vegetation and thermoregulate (McDonald and Luck 2013; Abom et al. 2015). The only notable differences were that riparian woodland habitats provided the most stable thermal environments with the least extreme temperature fluctuations, which is unsurprising given the effects of canopy cover in providing shade and reducing radiation to the sky (Breshears et al. 1998; Yates et al. 2000). In summer in kikuyu dominated communities, riparian woodlands provided the highest proportion of optimal tiger snake temperature compared to other habitats (Table S2). The most challenging temperatures for tiger snakes occur in summer and winter. In Perth during summer, Ta regularly exceeds 40 °C and winter nights can drop below 0 °C (Bureau of Meteorology 2022). Snakes are at considerable risk of exceeding their voluntary thermal maxima (35.5 °C) in summer or critical thermal minima (2 °C) during winter (Shine 1977; Lillywhite 1980; Shine and Mason 2004). In some cases, habitat types in either vegetation community approached these temperatures but the thermal stability, with generally lower maximum and higher minimum temperatures, offered by the woodland habitat types may offer tiger snakes the most valuable thermoregulatory and hibernation opportunities (Table S2).

Management and conservation implications

Seventy percent of urban wetland habitat on the Swan Coastal Plain between Wedge Island and Mandurah Western Australia has been lost to urban development and agriculture (Halse 1989). Their high biodiversity value means that appropriate conservation and management of remaining wetlands is of upmost importance. However, of the remaining 30% of wetlands on the Swan Coastal Plain, only 17% are vegetated with predominantly native vegetation (Hill et al. 1996). Weeds have been identified as a key threat to Perth metropolitan wetlands (Conservation and Parks Commission 2017) and as such weed management and rehabilitation plans have been implemented by many local governments and/or the Department of Biodiversity, Conservation and Attractions. Complete eradication of kikuyu grass and revegetation with native species can be extremely difficult (Department of Environment and Conservation 2012) and may not be successful in some situations. Our work suggests that total eradication of kikuyu grass is not necessary to maintain western tiger snake populations in wetlands of the Swan Coastal Plain and indeed large-scale removal of kikuyu may leave snakes and their prey without shelter sites. We suggest that an alternative to eradication that may improve habitat quality for the snakes and their major prey items would be to increase structural habitat complexity, particularly in areas of homogenous open kikuyu habitat. This could be achieved by removing smaller plots of kikuyu grass and planting more structurally complex native species, with scattered canopy cover. A suitably large area of kikuyu grass would need to be removed to allow for the establishment of these slower growing native species as perennial grasses such are kikuyu can quickly smother native plants (Department of Environment and Conservation 2012), but buffer areas of kikuyu should be left to provide habitat while the native vegetation becomes established. Though this method may result in positive outcomes for local western tiger snake and frog populations, more research is needed to determine how kikuyu dominated vegetation is used by other fauna, and the potential impacts of restoration on other flora and fauna species within the wetland ecosystem.

In Perth, tiger snakes are only known to persist in seven of the potential hundreds of available wetlands (Hill et al. 1996; Lettoof et al. 2021a). Although the exact habitat requirements necessary for supporting tiger snake populations are unknown, we suspect adequate vegetation structure and area plays a vital role. Over-grown kikuyu grass provides the necessary structural features in wetlands where native riparian vegetation no longer exists, providing suitable prey, anti-predator and thermal functions. Future research efforts should focus on assessing differences in habitat characteristics of wetlands with and without tiger snakes to determine the factors that allow tiger snakes to persist in urban wetlands, how tiger snakes utilise the habitats we described in this study (e.g. space-use, movement and habitat selection preferences), and the population size and stability of remnant populations.