We conducted this study on six trails located in the Central Andes Mountains, in Mendoza, Argentina (Fig. 1). These trails lie mostly in protected areas, and are the main entrance routes to the local Provincial Parks for tourists and climbers. We surveyed three trails in Cordón del Plata Provincial Park (1755 km2, 69° 26' W, 32° 58' S): Lomas Blancas, Piedra Grande and Morro Negro; and three trails in Aconcagua Provincial Park (657 km2, 69° 26' W, 32° 58' S) and surrounding areas: Quebrada de Vacas, Quebrada de Horcones, and Quebrada de Vargas. These trails are informal, not professionally designed, and are used by both hikers and domestic livestock. Many sections of the trails had multiple secondary trails generating impacts beyond the main trail (Barros et al. 2013, 2020). We covered an elevation gradient between 2400 and 3570 m a.s.l. (See Supplementary Material, Table S1 for further details). These areas constitute an internationally popular tourism and recreation destination due to the stunning landscapes with peaks over 5000 m, including mount Plata (5968 m a.s.l.) and Aconcagua (6962 m a.s.l.) (Barros et al. 2013). For example, in the 2018–2019 season (November to April), ca. 9000 people visited Aconcagua Park and 6000 Cordón del Plata Park (Secretaría de Ambiente y Ordenamiento Territorial, 2020). These areas also support livestock (mainly mules, horses, and cows) for human subsistence and transportation of mountain equipment.
The protected areas were created with the aim of conserving glaciers, rivers, Andean ecosystems, and archeological sites (Barros et al. 2013). The region has a great diversity of microclimates, determined by a complex topography, which generates different vegetation physiognomies (Méndez 2004; Morello et al. 2012). The climate is cold and dry, with precipitation concentrated mainly in the winter, between May and August (Morello et al. 2012). The average annual precipitation in Cordón del Plata Park is 398 mm (1979–2015), whereas Aconcagua Park is drier, with an annual precipitation of 100 mm (2003–2013) (Barros and Pickering 2014; Trombotto et al. 2020). The soils are, in general, little developed and exhibit substantial spatial heterogeneity in depth and granulometric composition (Méndez, 2004; Méndez et al., 2006). The vegetation consists of scrubland communities (including Adesmia pinifolia, Nassauvia axillaris, and Berberis empetrifolia) shrubby steppes (Adesmia subterranea and Azorella monantha, among others) and herbaceous steppes (Acaena pinnatifida and Phacelia secunda, among others). Between 3800 and 4200 m a.s.l. the vegetation cover is sparse and dominated by slow growing perennial herbs (e.g. Chaetanthera pulvinata, Nassauvia pinnigera and Nototriche transandina in Aconcagua, and Colobanthus subulatus, Nassauvia cumingii, and Senecio crithmoides in Cordón del Plata). With more than 500 native plant species identified, the Cordón del Plata Park has greater plant diversity than Aconcagua, where over 120 vascular plant species have been recorded (Méndez 2004, 2007; Méndez et al. 2006).
We carried out the field surveys in the summer season (January-March) of 2018 and 2019. Selected sites were intensively used for mountaineering activities and had a broad elevational range, with an average difference in elevation of 650 m a.s.l. between the start and the end of the trail. The sampling followed the T-trail survey protocol developed by the Mountain Invasion Research Network (MIREN) (Liedtke et al. 2020). We surveyed twenty transects along each trail, located approximately every ca. 35 m of elevation starting at the trailhead, avoiding areas with secondary trails. Each transect consisted of three 2 m x 10 m plots arranged in a T-shape, for a total of 120 transects (360 plots) (Fig. 2). In each plot, we identified and estimated the cover of all vascular plant species, both native and non-native. To estimate the level of livestock activity (cows, horses and mules), we estimated dung density on each plot (Ender et al. 2017). We collected plant specimens that could not be identified in the field and subsequently identified them with herbarium specimens and taxonomic keys in the Ruiz Leal Herbarium of the Argentine Institute for Dryland Research (IADIZA, CONICET Science and Technology Center, Mendoza). We classified species according to their origin and life forms using the database from the Darwinion Botanical Institute (Instituto de Botánica Darwinion 2018). We recorded the trail track, elevation and transect location with a GPS device and later processed them with QGIS and R software to determine the distance of each transect to the start of the trail.
We conducted all analyses with R version 3.6.1 (R Core Team 2019). To evaluate non-native plant success along mountain trails spanning the elevation gradient, we considered three response variables for non-native plants: occurrence (presence/absence of any non-native species), richness (number of non-native species present), and cover (the summed cover of all non-native species per plot). We applied Zero-altered Generalized Linear Mixed Models (two-part models) using the glmmTMB function (Brooks et al. 2017). In these models, presence-absence data are modeled with a binomial probability distribution with a logit link function (models the probability that a zero value is observed), while the non-zero observations are modeled with a truncated Poisson model for species richness with a log link function and a Beta distribution with a logit link function for cover (Zuur et al. 2009; Damgaard and Irvine 2019). Two-part models are appropriate for studies of non-native species invasions, as they usually have an overabundance of zeros in sites lacking non-natives (Damgaard and Irvine 2019). In our survey, 32% of the plots lacked non-natives (Supplementary Material, Fig. S1). In addition, these models allowed us to answer two questions in the same statistical structure: what determines the occurrence of non-native species and, in plots where non-natives were present, what determines their richness and cover. We eliminated non-significant predictors and compared the fit of the two models with a likelihood ratio test (LRT).
To assess the effect of trails on non-native plants success we included six explanatory variables in the models, 1) distance of each plot from the edge of the trail (plot number within the T-transect: plot 1 - adjacent to trail edge, plot 2 - intermediate plot, plot 3 - plot furthest from the trail edge); 2) the distance from the trailhead, considering both distances as proxy measures of a gradient of disturbance and propagule pressure; 3) dung density as a measure of livestock use intensity; 4) elevation (m a.s.l.). Finally, we consider two variables related to the composition of the resident community: 5) the coordinates of each plot resulting from a non-metric multidimensional scaling (NMDS1 and NMDS2), and 6) shrub cover (sum of all shrub covers per plot). We performed the non-metric multidimensional analysis considering the proportional cover of native species using the metaMDS function of the package vegan (Oksanen et al. 2019) in R (k = 2, trymax = 20, distance = Bray-Curtis). We checked that these variables were not too strongly correlated with each other based on Pearson's correlation coefficient, using the ggpairs function of the GGally package (Schloerke et al. 2020). Based on the criteria proposed by Dormann et al. (2013), we considered them to be too correlated if they had a Pearson's coefficient greater than I0.7I, which was the case for none of these six variables.
We also considered as additional variables precipitation (Supplementary Material, Fig. S2) and mean temperature (Supplementary Material, Fig. S3) data based on the period 1979–2013 obtained from the CHELSA global climate database (Karger et al. 2016, 2017) and downscaled to 1.2 arc second resolution, with a tile size of roughly 90 m2 at this latitude, using Geographically Weighed Regressions based on elevation, slope, northness, eastness, distance from the ocean, and potential solar radiation derived from the SRTM high-resolution Digital Elevation Model (NASA Shuttle Radar Topography Mission. SRTM 2013) following the procedure outlined in (Lenoir et al. 2017; Lembrechts et al. 2019). We assessed the correlation of these variables with the rest of the predictor variables and also relate these variables to the ordination obtained from the multidimensional scaling analysis (NMDS) using the envfit function of the vegan package. Temperature was highly and negatively correlated with elevation (corr. -0.76) and precipitation was positively correlated with the NMDS1 axis (corr. 0.75). We therefore included elevation instead of temperature in the models, as it resulted in a lower AIC and as the elevational gradient may represent other unmeasured covariates that may influence establishment and propagation of non-native plants (e.g. atmospheric pressure, vapour pressure, solar radiation (Körner 2007)). Similarly, we decided to include the NMDS1 and NMDS2 axes in the models instead of precipitation as it improved the AIC and these values would be representative of the community composition and the climatic gradient among which they change (Supplementary Material, Table S6 and S7).
We scaled the metric predictors (distance from trailhead, dung density, elevation, NMDS1, NMDS2, shrub cover, temperature, and precipitation) to a mean of zero and a standard deviation of one to make regression coefficients directly comparable. We included trail identity as a random factor in the models. To plot the partial effects of the fixed-effect predictor variables we use the Effect package (Fox and Weisberg 2018).