Within five riparian field sites ranging from alder woods to abandoned meadows we systematically sampled (Table 1, Fig. 1) the environmental conditions and the vegetation. Each site comprises patches of Impatiens glandulifera and heterogeneous micro-habitats regarding tree cover and soil moisture. All sites are located in the region around Bayreuth, Germany at an elevation between 345 m a.s.l. and 426 m a.s.l.. Within each site, plots were arranged on a grid of 20 m × 20 m. Every grid intersection point was used as position for a plot of 2 m × 2 m, independently of environmental condition, I. glandulifera or native vegetation cover (Fig. 1). This systematic design allowed us a representative sampling over the whole gradient of environmental conditions and I. glandulifera cover and regression analysis instead of comparison between invaded and uninvaded plots only. Across the five sites 114 plots of 2 m × 2 m were established in total (Table 1). In spring two plots were not accessible because of a very high soil water content and one plot was hit by a fallen tree, decreasing the number of replicates to 111 in spring.

Figure 1.

Plot design used for this study. Plots were arranged on a grid laid over the study site. Positions for light and soil moisture measurements within each plot are marked.

The herb layer vegetation was surveyed in summer while I. glandulifera was flowering (2016-07-12/08-17), and in spring while the spring geophytes were flowering (2017-04-20/05-04). All vascular plant species were determined using standard literature (Schmeil et al. 2011, Eggenberg and Möhl 2013, Jäger et al. 2013, Aas 2017, Jäger 2017), and their cover was estimated according to extended Braun-Blanquet scale (Braun-Blanquet 1964, Reichelt and Wilmanns 1973). Additionally the maximum height of the vegetation, and in summer 2017 the cover of I. glandulifera was measured as described in Suppl. materials 1, 2. The environmental variables volumetric soil water content, light (relative photosynthetically active radiation), tree layer species composition and cover, and Ellenberg indicator values were gathered once in each plot in summer. The volumetric soil water content (named soil water content hereafter) was measured on 2017-05-23/05-28 in the uppermost soil layer with a SM-150 sensor (Delta-T Devices). Recordings were taken at four positions per plot and their median was calculated (Fig. 1). Relative photosynthetically active radiation (PAR, Parent and Messier 1996, Gendron et al. 1998) was measured with Quantum sensors (LICOR) when trees were fully foliate (2016-08-03/10-03). In the plot four single point records were taken just above the herb layer (Fig. 1). Simultaneously a reference value was taken at a totally unshaded site nearby using a second Quantum sensor. This sensor was connected to a BayEOS logger (BayCEER, University of Bayreuth) taking records every 30 s and saving them as means over 5 minutes. Relative PAR was calculated with the median of the point records within the plots divided by the particular logged reference matching in time. All light measurements were taken under a homogeneous overcast sky with the sun invisible and no rain, and always between 11:00 and 17:00. The tree layer was characterized estimating the cover of each tree species separately (2017-08-17/09-07) according to extended Braun-Blanquet scale (Braun-Blanquet 1964, Reichelt and Wilmanns 1973). Mean Ellenberg indicator values for light L, soil moisture F, soil nutrients N, and soil reaction R were calculated per plot based on the summer vegetation. Based on additional information included in the F-value, the percentage of plants preferring either periodically wet soils or constantly wet soils was calculated. Hereafter, these parameters are named index of periodically wet soil and index of constantly wet soil.

All statistical analyses were done with the software package R 3.5.2 (R Core Team 2018). To find the polynomial model best describing the dependence of cover of I. glandulifera on light and soil water content a multiple regression analysis was performed. To identify environmental variables affecting the cover of I. glandulifera, we performed an automated model selection (Bartoń 2018) separately for summer and spring vegetation. First of all a global model was built with the cover of I. glandulifera as response variable and 13 predictor variables that were expected to affect the cover of I. glandulifera: relative PAR (squared because of hump-shaped relationship), soil water content (squared), number of tree species, cover of these tree species occurring at least in 20 plots (Alnus glutinosa, Salix fragilis, Acer pseudoplatanus, Fraxinus excelsior, Betula pendula), Ellenberg values N, R, and indices for periodically or constantly wet soils. Ellenberg values L, and F were excluded because they correlated with relative PAR and soil water content, (Pearson correlation coefficient 0.549 and 0.544 respectively). All variables were standardized to zero mean and unit variance (VEGAN, (Oksanen et al. 2018)). For analysis of spring vegetation, the cover of I. glandulifera was log-transformed to counter heteroscedasticity of the model. The study site was considered as a random factor (NLME, (Pinheiro et al. 2018)). Next a set of models with combinations of all parameters was generated from the global model and the models were weighted by their AICc (MuMIn, (Bartoń 2018)). Models with ΔAICc > 2 were used to calculate the relative importance of each variable as the sum of AICc weights of all models including the variable.

Using the variables resulting from the model selection, we performed a piecewise structural equation model (piecewiseSEM, (Lefcheck 2016)) to test the effects of the environmental variables on I. glandulifera, and how in turn I. glandulifera affects the resident vegetation (all species except I. glandulifera). This also allowed us to infer whether the resident vegetation is more affected by I. glandulifera or by the environment based on the regression coefficients of the SEM. The resident vegetation was represented by species number, total cover (sum of the cover of all resident species in a plot) and the cover of those herbaceous species occurring in more than 20 plots. The construction of the initial models is visualized in Suppl. material 3, Fig. 1. The SEMs were fitted separately for summer and spring vegetation, and within the models the study site was considered as a random factor. For each path in the piecewise structural equation model, a standardized regression coefficient (β) and its significance were calculated as well as conditional R²-values for all response variables.

To analyze plant community composition in summer, or respectively spring, we performed a Detrended Correspondence Analysis of the cover of the resident plant species with downweighting of rare species (DCA, package VEGAN (Oksanen et al. 2018)). DCA was confirmed to be appropriate because the DCA-axis gradient length was more than four times the standard deviation. Cover of I. glandulifera, as well as environmental parameters, were post-hoc fitted into the DCA result. Additionally, a Constrained Correspondence Analysis (CCA, VEGAN) was performed with the same data constraining the resident community with I. glandulifera cover. With an ANOVA-like permutation test (VEGAN) significance of the constraints was tested.

With the summer dataset of the year 2016, we analyzed whether the impact of I. glandulifera on the resident vegetation differed between micro-habitat groups. The groups were created by dividing the dataset according to the median of light (23.9 % PAR) and soil water content (51.5 %). Subsequently, they are named moist–bright (n = 30), wet–bright (n = 28), moist–dark (n = 27) and wet–dark (n = 29). For each of this groups separately and for the complete dataset impact of I. glandulifera on various variables representing the resident vegetation was analyzed: Impact on species number, Shannon-index and total plant cover was tested with linear models. Some parameters in the wet–dark group were log-transformed to counter heteroscedasticity of the models. Impact on cover of Filipendula ulmaria, Phalaris arundinacea and Urtica dioica was tested with a quantile regression (R package QUANTREG (Koenker 2018)) because data were not homogenous in variance hence linear regression was not the appropriate test (Cade and Noon 2003). We took the 0.50, 0.75, 0.85 and 0.95 quantiles emphasizing the upper quantiles because after visual inspection of the data we expected I. glandulifera to especially restrict maximum cover of other plants. For each quantile regression, standard errors and p-values were calculated by bootstrap analysis. Impact of I. glandulifera on community composition was tested with a DCA (with downweighting of rare species) and with CCA (VEGAN).

I. glandulifera occurred in about 80 % of all plots in summer (87 of 114) and in spring (91 of 111, Fig. 2). Especially in spring the cover of I. glandulifera was often very low and rarely above 25 %. In summer I. glandulifera reached more than 50 % cover in 28 plots. By summer 2017 the cover of I. glandulifera changed largely in few plots (Suppl. material 1). In summer, I. glandulifera plants were higher than the resident vegetation if they reached more than 20 % cover, while in spring I. glandulifera was always lower than the resident vegetation (Suppl. material 2). The resident vegetation consisted of in total 128 plant species in summer 2016 and 109 in spring 2017 ranging from 2 to 20 species per plot. None of these plant species was an endangered one. Besides I. glandulifera further alien species were recorded: Lamium argentatum occurred in 14 plots, sometimes reaching more than 75 % cover. Fallopia japonica, Lysimachia punctata, Bidens frondosa and Epilobium ciliatum each occurred in only one plot with always less than 5 % cover. Most frequent native species were typical ones of tall herbaceous vegetation at riparian sites (Fig. 4A, B). Especially Urtica dioica, Filipendula ulmaria, and Phalaris arundinacea in summer, and additionally the geophytes Ranunculus ficaria and Anemone nemorosa in spring could reach cover of more than 50 %. In spring further early flowering species such as Corydalis cava, Caltha palustris, Polygonum bistorta, Cardamine amara and Alliaria petiolata occurred.

Figure 2.

Distribution of cover of Impatiens glandulifera in summer 2016 and spring 2017. Shown categories correspond to the Braun-Blanquet scale.

Light (relative PAR) and soil water content spanned nearly the whole gradient from 0– 100 %. However, Ellenberg values that correlated with light and soil moisture showed rather smaller gradients (L-value for light 4–7.5, F-value for soil moisture 5.5–9.3) indicating that there were medium light conditions and no sites with dry soils. I. glandulifera occurred over the whole range of light and soil water content measured in this study, but in summer it reached high cover mainly at 50–70 % light and 30–40 % soil water content (Fig. 3). The relationship between I. glandulifera and light and soil water content was hump-shaped being a typical species reaction on a long environmental gradient (light: linear model, f(x) = x + x² + x³, F_(3,110) = 7.221, R² = 0.142, p < 0.001). Notably soil water content on its own had only low explanatory power (linear model, f(x) = x + x³, F_(2,111) = 4.88, R² = 0.064, p = 0.009) but in combination with light the R² increased to 0.208 (Fig. 3A, B).

Figure 3.

Cover of Impatiens glandulifera in summer 2016 in dependence of light and volumetric soil water content. A Cover of I. glandulifera is represented by point size and color as given in the legend. Grouping the plots into the micro-habitats moist–bright, wet–bright, moist–dark and wet–dark is based on the medians of light and soil water content. B Fitted function of the same data shown in 3d-space. f(cover) = light + light² + light³ + water content + water content³. Linear model, R² = 0.208, F_(5,108) = 6.928, p < 0.001, n = 114.

The piecewise SEM revealed that in summer 39 % of the variation in the cover of I. glandulifera was explained by the environmental variables identified as important by the model selection (R² = 0.39, Fig. 4A, Suppl. material 4: Table S1). The reaction of I. glandulifera to light was unimodal hence the cover was highest at moderate light (β = -0.294). The cover of I. glandulifera was enhanced by a high Ellenberg value N for nutrients and by periodically wet soils (Fig. 4A). In contrast it was reduced by constantly wet soil and cover of the specific tree species Acer pseudoplatanus, Fraxinus excelsior, and Alnus glutinosa. In turn, I. glandulifera had no impact on the number of plant species but on plant cover. It strongly reduced the cover of U. dioica (β = -0.387), slightly that of F. ulmaria (not significant, p = 0.073) and the total cover of the resident vegetation. Besides the effect of I. glandulifera the parameters representing the resident vegetation were mainly directly affected by the environmental variables. For example, the number of plant species increased with the number of tree species and strongly decreased with increasing Ellenberg value N. Cover of U. dioica was determined by Ellenberg value N, index of constant wet soil and by cover of A. glutinosa similarly to I. glandulifera.

Figure 4.

Results of the piecewise structural equation modeling for summer (A) and spring (B). Arrows show significant correlations between the environmental variables shown to be important by the model selection (Suppl. material 4, Table 1), cover of Impatiens glandulifera and resident vegetation parameters. Included resident species are the most frequent ones: Ranunculus ficaria (Ran.ficaria), Urtica dioica, Filipendula ulmaria (Fil. ulmaria), Galium aparine, Anemone nemorosa (Ane. nemorosa), Phalaris arundinacea (P. arundinacea), Galeopsis tetrahit (Galeo.tetrahit), Stellaria nemorum (Ste. nemorum) and Impatiens noli-tangere (I. noli-tangere). Arrows show significant correlations, red arrows negative ones, black arrows positive ones. The thicker the arrows, the higher are the standardized regression coefficients (β), which are stated next to the arrows. R² values for the component models are given within the boxes of all response variables. Variables that are directly connected to I. glandulifera are highlighted by gray colored boxes. For the spring model cover of I. glandulifera was log-transformed. n = 114 plots for summer, n = 111 for spring.

The piecewise SEM on spring vegetation showed that 30 % of the variation of the I. glandulifera cover was explained by the environmental variables identified as important by the model selection (Fig. 4B, Suppl. material 4: Table 1). As in summer, constantly wet soils and cover of the tree species A. glutinosa and Betula pendula reduced the cover of I. glandulifera, and reaction to light was unimodal (Fig. 4B). In contrast to summer, periodically wet soils were not found to be important (model selection, Suppl. material 4: Table 1), and the increase of I. glandulifera cover with increasing Ellenberg value N was not significant (SEM, Fig. 4B). Also I. glandulifera did not affect U. dioica and total plant cover, and the cover of Galeopsis tetrahit was even slightly increased. However, the cover of R. ficaria and A. nemorosa were reduced by I. glandulifera. The resident vegetation was mainly directly affected by the environmental variables and by interactions between the resident species. For example, the cover of R. ficaria and G. aparine were positively correlated and the cover of U. dioica and R. ficaria reduced the species number.

In summer I. glandulifera had no impact on plant community composition: The cover of I. glandulifera did not correlate with the axes of a DCA of the resident community (p = 0.222, Fig. 5) and was not able to constrain resident community in a CCA (p = 0.116, without figure). In contrast the resident plant community in spring was strongly affected by the cover of I. glandulifera of the previous summer 2016 (p < 0.001, DCA; p = 0.052, CCA; Fig. 5), and slightly by current cover in spring 2017 (p = 0.048, DCA; p = 0.551, CCA). In summer and in spring the resident plant community was also shaped by most of those environmental variables important for the cover of I. glandulifera (Fig. 5).

Figure 5.

Ordination (DCA) of the resident plant community in summer 2016 and spring 2017. Cover of I. glandulifera in summer 2016 (Imp 16) and in spring 2017 (Imp 17) and important environmental variables (Suppl. material 4, Table 1) as well as volumetric soil water content (soil.water) were post-hoc fitted onto the DCA. Significant variables are shown as arrows. Plots are indicated as crosses, species as open circles. Most abundant species are labeled. Not significant environmental variables and Eigenvalues of DCA axes are given in boxes below the plots. n = 114 plots for summer, n = 111 for spring. For abbreviations of the species names see Suppl. material 5: Table S2.

With the summer dataset four micro-habitat groups were created reflecting different conditions of light and soil water content (Fig. 3). These groups differed in their plant community composition (DCA p = 0.008, CCA p = 0.001). In each group the cover of I. glandulifera ranged from 0 to at least 80 % but its mean differed between groups, being highest in the wet–bright group (Table 2). The impact of I. glandulifera on plant cover was different between micro-habitat groups (Table 2, Suppl. material 6: Fig. S2). The cover of I. glandulifera reduced the total plant cover in all micro-habitat except for the wet–dark group. The cover of U. dioica was reduced in the moist–bright and wet–dark groups, as well as the cover of F. ulmaria in the two bright micro-habitats. These were exactly those micro-habitats where the highest average cover of these species was observed (Table 2, Suppl. material 6: Fig. S2). In contrast, the cover of Phalaris arundinacea was not negatively affected at all, but its cover slightly increased with the cover of I. glandulifera under low light conditions. I. glandulifera had no impact on plant species number and Shannon-index in any micro-habitat group. In contrast plant species composition was changed under bright conditions especially with high soil water content. Within the wet–bright micro-habitat for example Calystegia sepium, Glechoma hederacea, and Polygonum bistorta tended to occur at high cover of I. glandulifera (CCA). In contrast Carex acutiformis, Geranium palustre and also some species of very wet plots as Galium elongatum and Equisetum fluviatile occurred at low cover of I. glandulifera (Suppl. material 7: Fig. 3). Considering all parameters representing the native vegetation I. glandulifera had the lowest impact in the wet–dark micro-habitat where also the cover of I. glandulifera was lowest.

Within our study sites, I. glandulifera occurred over a broad range of environmental conditions but it was unevenly distributed within the sites forming invaded and uninvaded patches. Its cover correlated with environmental variables. A positive effect of nutrients and moderate light as well as low importance of soil water content (measured at one point in time), is consistent with literature (Andrews et al. 2005, Čuda et al. 2014). However, we showed that soil water content in combination with light was a good predictor for the cover of I. glandulifera, with the cover being highest at moderate light and moderate soil water content. Ellenberg values indicated moreover that constantly high soil moisture had a negative effect on the cover but in summer periodically wet soils were favorable. A high N-supply is also more important in summer than for early establishment in spring. Considering a larger spatial scale, such a patchy occurrence can enable the co-existence of species that would outcompete each other within one patch (Amarasekare 2003). For example, in our study I. glandulifera and U. dioica coexisted within one study site forming a patchy mosaic.

We found that I. glandulifera reduced the cover of the resident vegetation but it had no impact on species composition in summer or on plant species alpha-diversity at all. Thus the resident plant species seem to be able to coexist within I. glandulifera stands, albeit reaching only lower cover. Changes in I. glandulifera cover from year-to-year as they are reported in literature (Kasperek 2004) and which were also observed in our study, should then enable the resident plants to recover when I. glandulifera declines leading to co-existence at a larger time-scale (Stouffer et al. 2018).

I. glandulifera especially reduced the cover of the most dominant native species. Species were most affected in those micro-habitats where their average cover was highest and in each season those species with the highest cover were the most affected ones. These were Urtica dioica and Filipendula ulmaria in summer, and Ranunculus ficaria and Anemone nemorosa in spring. We suggest that this is due to competition for space and resources strengthening at high cover. Still, it cannot be ruled out that also less frequent species with lower cover are affected by I. glandulifera. Rare occurrence and thus small sample size of a species as well as huge variability result in low statistical power and may lead to an underestimation of the effect of the invader (Davidson and Hewitt 2014).

Similar to other studies comparing plots with and plots without I. glandulifera, we are not able to show a causal impact of I. glandulifera on native vegetation but only correlations (Hejda and Pyšek 2006). However, in our study the link to environmental conditions can help to disentangle negative correlation because of different spatial niches from negative correlation because of suppression within one spatial niche. A. nemorosa and I. glandulifera could be an example for different spatial niches, because A. nemorosa was enhanced by a high cover of Alnus glutinosa which in contrast reduced the cover of I. glandulifera. U. dioica however, seems causally suppressed by I. glandulifera. The cover of both species correlated negatively, and according to the SEM they were favored by the same environmental conditions. Experimental studies confirm that U. dioica is negatively affected by I. glandulifera and that this effect is larger than vice versa (Tickner et al. 2001, Gruntman et al. 2014, Bieberich et al. 2018).

The habitat depending impact of I. glandulifera indicates that the impact gets stronger with increasing cover of I. glandulifera. This is also indicated by Cockel et al. (2014). In our study the wet–dark micro-habitat with the lowest cover of I. glandulifera, was the least affected. The plant species composition was most affected in the wet–bright micro-habitat which had also a strong gradient and highest average cover of I. glandulifera. Species that tended to occur only in plots without I. glandulifera generally occurred less frequently (for example Equisetum fluviatile) while those that tended to occur at high cover of I. glandulifera (for example Glechoma hederacea) were very common ones.

Micro-habitat specific interactions between native species and an invader can also be due to micro-habitat specific performance of the plant species. If two C-strategists compete for resources, which should be the case with our dominant species, the magnitude of competition is highest under most favorable as well as under most unfavorable environmental conditions (stress-gradient hypothesis, Maestre et al. 2009). In the strongly competitive situation inferiority of the natives in fitness leads to suppression by the invader (MacDougall et al. 2009). We suggest that this can explain the micro-habitat depending impact of I. glandulifera on U. dioica and F. ulmaria. Both natives were most reduced by I. glandulifera when they grew under environmental conditions that were, according to the SEM, most favorable for them (U. dioica in the moist–bright group and F. ulmaria in bright micro-habitats). U. dioica was additionally negatively affected by I. glandulifera in the wet–dark micro-habitat which was shown by the SEM to be unfavorable for U. dioica.

Plant species composition in summer was not affected by I. glandulifera but in spring it was, despite the fact that I. glandulifera plants were smaller than the resident vegetation in spring. The reason could be a seasonally varying allelopathic effect of I. glandulifera because it is known, that in spring I. glandulifera has a higher content of the allelopathic compound 2-MNQ compared to summer (Ruckli et al. 2014a). In a previous experimental study we showed that I. glandulifera has a species-specific allelopathic and competitive impact on native plants especially in the seedlings- and juvenile-stage (Bieberich et al. 2018). Furthermore, cover of I. glandulifera from the previous summer 2016 affected species composition in spring while it did not affect the resident community in summer 2016 itself. Allelopathic legacy effects (Grove et al. 2012) may explain this: 2-MNQ could persist in the soil (Ruckli et al. 2014b) and affect early growing plants even before I. glandulifera germinates.

Negative impact on biodiversity and ecosystem functions, processes and services are the criteria to grade an alien species as invasive (Ehrenfeld 2010, Hulme et al. 2013, Barney et al. 2013). German and European Union nature conservation authorities emphasize the impact on diversity and threat to other species (European Union 2014, Nehring et al. 2015). Taking this study and all available ones into account, the impact of I. glandulifera on plant species diversity can be rated to be relatively low (Hejda and Pyšek 2006, Hejda et al. 2009, Diekmann et al. 2016, Čuda et al. 2017) even if some studies showed stronger effects (Hulme and Bremner 2006, Cockel et al. 2014, Kiełtyk and Delimat 2019). Indeed, we found a negative impact on the dominant natives, U. dioica and F. ulmaria, but they are in general very common and widespread in Central Europe and thus not expected to be threatened (Schreiber 1958, Weber 1995). However, suppression of abundant dominant plant species could lead to changes in ecosystem processes as they account for functions such as primary production and nutrient cycles (Grime 1998).

The micro-habitat and season dependent impact of I. glandulifera requires that its invasion risk has to be assessed separately for different habitats. We found the lowest impact in the wet–dark micro-habitat which corresponds to alder swamp-forests. The impact was highest at bright conditions, as abandoned meadows, but especially under high soil moisture as found in marshes or open patches of swamp-forests. Special attention should be given to habitats with rare or specialized communities or with distinct spring communities. For nature conservation this is a great opportunity to develop more targeted management strategies of I. glandulifera and invasive species in general with vigorous efforts only in selected habitats.