Rapid upwards spread of non-native plants in mountains across continents

Species distributions are rapidly shifting globally, with mountain ecosystems particularly responsive due to steep environmental gradients over short distances. Historically, high elevations have had relatively few invasions by non-native plants, likely due to lower human disturbance and harsh climatic conditions. However, climate warming and increasing human activity at higher elevations are changing this situation; mountain roads already facilitate non-native establishment at higher elevations. Prior regional studies report mixed trends in non-native upslope expansions, and consistent, global time-series assessments have been lacking. The authors posit that apparent stronger upslope shifts for species starting at low elevations may arise partly from statistical artifacts: regression toward the mean and geometric constraints along bounded elevational gradients limit observable shift directions and magnitudes. Therefore, null models are necessary to distinguish true biological signals from expectations under chance. The study asks: (1) has non-native species richness increased in mountains in the last decade? and (2) are upper elevation limits moving upslope, particularly faster than expected at lower elevations?

Previous work shows non-native plant introductions typically begin at low elevations, spreading upward over time; older introductions reach higher elevations and broader elevational ranges. Studies of native species often find a negative relationship between initial upper range limits and subsequent shifts, attributed to climate responses, water balance changes, growing season length, tolerance breadths, and microsite buffering. Regional assessments of non-native plants have found no average expansion over a decade in parts of Europe or upward expansions over longer periods in Europe, Hawaii, and California. Given geometric constraints and regression toward the mean, elevation-dependent shift patterns can emerge without biological change, underscoring the need for null-model evaluation before ecological interpretation.

Design and data collection: Using the MIREN standardized protocol, elevational gradients along mountain roads and adjacent seminatural vegetation were surveyed in 11 regions across five continents (from 68° N to 37° S), including New South Wales and Victoria (Australia), Switzerland, Central and South Chile, Montana and Oregon (USA), Hawaii (USA), Tenerife (Spain), Kashmir (India), and Norway. In each region, typically three roads (two in Central Chile, four in Hawaii, five in Victoria) open to vehicle traffic were selected, spanning from the lowest point with meaningful elevation change to the road’s highest point. Each road was evenly stratified into 20 sampling transects (about 60 per region; total 651 transects). At each transect, three 2 × 50 m plots were arranged in a T-shape: one along the roadside (0 m) and two perpendicular at 50 m and 100 m from the road (when accessible). This yielded 651 roadside, 481 at 50 m, and 440 at 100 m plots. Surveys were conducted at 5-year intervals during peak growing seasons starting 2007/2008. Species data: Presence and abundance (ordinal classes) of non-native vascular plants (per World Flora Online taxonomy) were recorded. Analyses aggregated plot data to transect-level presence–absence. Species not identified to species level and singletons (recorded only once within a region) were excluded. The dataset comprised nearly 15,000 observations for 616 non-native species (reduced to 480 after excluding singletons) across regions. Analytical approach: Species richness per region and year (n=27 region-year observations) was modeled via linear mixed-effects models (Gaussian errors) with year since first survey (0, 5, 10) as a fixed effect and region as a random intercept; likelihood ratio tests compared models to intercept-only equivalents. Percentage change in richness was modeled similarly with the intercept fixed to zero. Range limits: For species present in both first and last surveys (5 or 10 years apart), upper elevation limits were defined as the 90th percentile of observed elevations. Range shifts were computed as differences between last and first surveys’ 90th percentiles. Elevations were centered and scaled within regions for cross-region summaries. Intercept-only LMMs (region as random intercept) tested overall changes. Region-specific linear models (unstandardized elevation) estimated mean shifts, weighting observations by species’ total frequency of occurrence across first and last surveys to reduce influence of rare, outlier records. Sensitivity filters assessed robustness by retaining only species with >5 and >10 occurrences across all years (filters 1,2) and >10 per region per year (filter 3). Elevation dependence and null models: To test elevation-dependent shifts, models regressed upper-limit shift on initial upper-limit elevation (t1), weighting by species frequency, within and across regions. Mean shifts accounting for elevation were evaluated at each region’s median elevational gradient. A null-model framework randomized observed shift magnitudes across the gradient 10,000 times per region by assigning new initial elevations (with replacement) from surveyed elevations, truncating to respect geometric bounds (preventing impossible shifts beyond minimum/maximum elevations). For each iteration, a weighted linear regression of shift versus initial elevation was fitted; fitted values across 10,000 iterations provided 95% confidence intervals for expected relationships under randomness. Deviations were deemed significant where observed fitted relationships fell outside these CIs. Two alternative nulls varied the vector of initial elevations: (1) 200 equally spaced elevations (narrower CIs) and (2) only observed species’ initial elevations (wider CIs). All analyses were conducted in R 4.0.3 using lme4; spatial visualization used sf and ggmap.

• Species richness: Across regions, non-native plant richness increased significantly by 1.56 ± 0.57% per year (~16% per decade; χ² = 6.822, P = 0.009). Absolute richness showed a near-significant increase of 0.46 ± 0.23 species per year (χ² = 3.190, P = 0.074). Seven of 11 regions (64%) had net richness increases over the study period; four regions (Oregon, Central Chile, Victoria, Kashmir) showed small net losses, with non-monotonic changes in some time series.
• Upper elevational limits: Mean shifts between first and last surveys were upslope in all regions except Montana and Hawaii (USA). Significant upslope mean shifts were detected in four regions (two additional near-significant). Patterns persisted under stricter frequency filters; regions with very low occurrences could lose significance or be excluded due to small sample sizes.
• Elevation dependence: Across regions, shifts were more upslope at lower elevations, decreasing linearly with elevation and sometimes becoming slightly downslope at high elevations (LMM: F1,579.89 = 54.82, P < 0.001). After accounting for elevation by evaluating at the gradient median, average upper limit shifts were significantly upslope in seven regions. The overall standardized cross-region shift (0.07 ± 0.04) was not significant (t = 1.77, d.f. = 9.43, P = 0.109), reflecting regional heterogeneity.
• Null-model tests: Observed negative relationships between initial elevation and shift exceeded null expectations (greater upslope than expected by chance at low/mid elevations) in at least seven regions (both Australian regions, South Chile, Tenerife, Hawaii, Kashmir, Montana). An alternative null model added Oregon to this list. Some regions showed greater-than-expected downslope shifts at high elevations (e.g., Tenerife, Montana), depending on the null specification.
• Drivers and context: No relationship was found between regional average yearly temperature increase (2000–2016) and mean range limit shifts (F1,9 = 0.447, P > 0.521), and regions with 10-year series did not show larger net shifts than 5-year series, suggesting short-term warming was not the primary driver. Detection and magnitude estimates were influenced by transect spacing; regions with larger distances between plots (e.g., Hawaii, Kashmir, Tenerife, Central Chile) could underdetect small shifts or inflate small changes. Overall evidence indicates rapid upward expansion of non-native species’ upper elevational limits in 10 of 11 regions within 5–10 years, especially at lower/mid elevations and along roads.

The study demonstrates a rapid, global trend of non-native plants expanding upslope along mountain road corridors. While elevation-dependent negative relationships between initial positions and shifts can arise from statistical and geometric constraints, null-model comparisons show that in most regions the observed upslope advances at low/mid elevations exceed chance expectations, indicating true biological spread. Likely mechanisms include ongoing dispersal from low-elevation introduction hubs, higher propagule pressure and disturbance near human infrastructure, and longer growing seasons at lower elevations facilitating spread. Many species at low elevations are recent introductions and not yet in equilibrium with climatic limits, producing nested elevational distributions and continuing upward expansion. Downslope shifts at high elevations were less common beyond null expectations and may reflect management removals, episodic harsh conditions, or stochastic dynamics affecting rarer high-elevation occurrences; weighting by species frequency reduced stochastic influence. Over the study window, climate warming did not explain regional differences in shift magnitudes, but longer-term warming is expected to further open higher elevations to invasion. The analysis underscores the importance of applying null models in interpreting range shift patterns along bounded gradients to avoid attributing artifact-driven patterns to ecological mechanisms.

Non-native plant species are rapidly expanding their upper elevational limits in mountains worldwide, with significant upslope movements detected in 10 of 11 regions within just 5–10 years. Roads act as conduits and suitable habitats, likely accelerating spread relative to undisturbed areas. Given increasing anthropogenic pressures and future climate warming, exposure of high-elevation ecosystems to biological invasions will likely intensify. Management implications include the need for continued monitoring, early detection along road networks, and proactive biosecurity in mountain regions. Future research should identify traits linked to rapid spread and high impact, evaluate impacts on native communities, and determine ecosystem features that heighten invasion risk. Methodologically, integrating null-model frameworks into analyses of elevational range dynamics will be crucial for robust inference as time-series data accumulate.

• Observational constraints along bounded elevational gradients can bias apparent elevation-dependent shifts; addressed via null models but still an inherent limitation.
• Heterogeneous transect spacing among regions (and along single roads) can underdetect small shifts or overestimate small changes; some regions (e.g., Hawaii) had large between-transect distances and many species recorded on single roads.
• Sampling focused on roadsides and nearby seminatural plots; results may overrepresent corridor-driven spread and not fully capture invasion into undisturbed high-elevation habitats.
• Short time horizon (5–10 years) limits detection of longer-term climate-driven responses; climate anomalies during survey years may influence richness and occurrence.
• Low occurrence of some species/regions reduced power; applying frequency filters led to exclusion of regions with few species and loss of significance in some cases.
• Use of the 90th percentile for upper limits reduces outlier influence but may miss extreme occurrences; presence–absence aggregation across plots may obscure fine-scale abundance dynamics.