Fig. 1 Geographic distribution of provenance (a) and species occurrence records (b) for introduced plants included in this study. Propagules were obtained from 435 provenances through field collection, exchange with other botanical gardens, and purchase, all coordinated by the Xishuangbanna Tropical Botanical Garden, Chinese Academy of Sciences (XTBG). This study includes 1,232 species, with 3,152,039 occurrence records obtained from the Global Biodiversity Information Facility (GBIF, https://www.gbif.org/) and cleaned by removing outliers. The number of occurrence records per species had a median of 47, a mean of 2,767, and ranged from 1 to 500,332. The location of XTBG (21°41′ N, 101°25′ E) is marked by a black diamond. Red circles indicate provenance in (a) and species occurrence records in (b).

Fig. 2 Climatic differences between Xishuangbanna Tropical Botanical Garden, Chinese Academy of Sciences (XTBG) and the distribution of plants in the wild based on the proportion of the climatic distribution of each species that was more extreme than the climate at XTBG, illustrated with one of the study species: Ficus virens. Red long-dashed vertical lines indicate the climatic parameter values at XTBG; blue lines indicate the climate at the provenance; green short-dashed vertical lines indicate the median climate across occurrence records of the species; and gray shaded curves represent the density of the species climatic distribution constructed using kernel density estimation, with climatic data for occurrence records extracted by matching GBIF distribution records with WorldClim 2.1. This metric was derived by calculating the cumulative proportion of species occurrence records falling on the side more extreme than the XTBG threshold. Specifically, for minimum temperature of coldest month (mtcm), we calculated the proportion of occurrence records with temperatures warmer than that at XTBG; for maximum temperature of warmest month (mtwm), we calculated the proportion of occurrence records with temperatures colder than that at XTBG. For mean annual temperature (mat), the metric depended on the relative position between the species median and XTBG: When the species median was lower, we calculated the proportion of occurrence records with temperatures lower than that at XTBG; otherwise, we calculated the proportion of occurrence records with temperature higher than that at XTBG. For precipitation variables—annual precipitation (map), precipitation of driest month (pdm), and precipitation of wettest month (pwm)—we consistently calculated the proportion of occurrence records with precipitation lower than that at XTBG. This approach emphasizes climatic properties of the whole species distribution, as opposed to propagule provenance under the assumption of local adaptation, and may be useful when species’ climatic niches are largely determined by the tolerance of generalist genotypes.

Fig. 3 Comparison of empirical support for mixed-effects Cox proportional hazards models assessing effects of climatic difference metrics and growth form on survival of introduced plants. The x-axis indicates candidate model numbers (1-63) grouped by the number of climatic variables included (decreasing from 6 to 1 from left to right, indicated by alternating gray and white bands); the y-axis indicates Akaike information criterion differences (ΔAIC, relative to the best-supported model). Models incorporated three climatic difference metrics: difference from climate at provenance (blue), difference from median climate across occurrence records (orange), and proportion of extreme climat (red); and two random-effect structures: uncorrelated species random effects (circles) and phylogenetically correlated random effects (triangles). The gray dashed line indicates the ΔAIC = 3 threshold. Panel (b) provides a detailed view of the upper range (ΔAIC ≤ 10) in panel (a), highlighting models with highest support (ΔAIC ≤ 3). All models included growth form as a fixed effect.

Fig. 4 Comparison of mixed-effects Cox proportional hazards models for introduced plant survival in relation to climatic differences and growth form, and corresponding effect estimates. (a) Combinations of climatic variables included in each model. The ordinate shows six climatic difference variables: difference in mean annual temperature (Δmat), difference in minimum temperature of coldest month (Δmtcm), difference in maximum temperature of warmest month (Δmtwm), difference in annual precipitation (Δmap), difference in precipitation of driest month (Δpdm), difference in precipitation of wettest month (Δpwm). The abscissa shows all non-empty combinations of these variables. Gray dots indicate variable inclusion; vertical dotted lines mark models with the highest empirical support (ΔAIC ≤ 3), with variables in these models highlighted by large black dots. The numbers and alternating gray/white bands at the top serve as visual separators for model groups. All models included growth form as an explanatory variable in the analysis. (b) Effect estimates from the best-supported models (ΔAIC ≤ 3). Point estimates (red for temperature, blue down for precipitation, black diamonds for growth form) and 95% confidence intervals are shown for nine top-ranked models (Model IDs 130-152). Coefficients are log (hazard ratio); statistical significance at P = 0.05 is indicated when 95% confidence intervals do not overlap 0 (gray horizontal line). For climatic variables, regression coefficients > 0 indicate increased mortality risk per unit increase in climatic difference; for growth form, regression coefficients < 0 indicate lower mortality risk for herbal than for woody plants.

Fig. 5 Effects of varying climatic values on predicted introduced plant survival, according to one of the models with highest empirical support (ΔAIC = 0; Model 13 in Fig. 3, Model 139 in Fig. 4). The curves relate survival time (abscissa) to probability of plant survival at XTBG (ordinate). In each panel, the curve labeled “XTBG” describes the predicted survival when the median climate of the species distribution is identical to the climate of XTBG (i.e., the baseline prediction when climatic difference equals zero); other curves correspond to survival conditions at four representative values: the maximum observed difference, and three equally spaced intermediate levels between zero and the maximum. Survival curves are based on the Nelson-Altschuler-Aalen estimator of the hazards. Panels (a-b) show the effect of differences in median mean annual temperature (ΔmatM); panels (c-d) show the effect of differences in median minimum temperature of coldest month (ΔmtcmM); while panels (e-f) show the effect of differences in median precipitation of wettest month (ΔpwmM).

Fig. 6 Comparison of empirical support for mixed-effects Cox proportional hazards models relating introduced plant survival to soil differences and growth form. Empirical support for 60 models was measured as differences in the Akaike information criterion (ΔAIC). Models quantified soil differences between XTBG and species’ wild distributions in two ways: difference from soil at provenances (labeled “Provenance soil”) and difference from the median soil across occurrence records of a species (labeled “Median soil across occurrence records”). Additionally, models accounted for evolutionary relationships among plantings using uncorrelated species random effects (labeled “species random effect”) or phylogenetically correlated species random effects (labeled “phylogenetic random effect”). The abscissa indicates candidate model numbers (1-15), and the ordinate indicates ΔAIC values. Blue circles represent provenance soil with uncorrelated species random effects (s), and red circles represent the median soil across occurrence records with uncorrelated species random effects (s); blue triangles represent provenance soil with phylogenetic random effect, and red triangles represent the median soil across occurrence records with phylogenetic random effect. Gray dashed lines in both panel (a) and panel (b) indicate ΔAIC = 3, serving as the threshold to identify models with highest empirical support (ΔAIC ≤ 3). Panel (b) Magnified view the lower part of the ordinate in panel (a) (0-10), highlighting the models with highest empirical support (ΔAIC ≤ 3). Growth form was included as an explanatory variable in all models.

Fig. 7 Comparison of mixed-effects Cox proportional hazards models for introduced plant survival in relation to soil differences and growth form, with effect estimates. (a) Combinations of soil variables included in each model. The ordinate (y-axis) displays four soil difference variables: difference in topsoil sand content (ΔT_SAND), difference in topsoil organic carbon content (ΔT_OC), difference in topsoil gravel content (ΔT_GRAVEL), and difference in subsoil bulk density (ΔS_BULK_DENSITY); the abscissa (x-axis) shows 15 candidate models (numbered 1-15) representing all possible combinations of these variables. Dots indicate variable inclusion; large black dots mark variables present in models with the highest empirical support (ΔAIC ≤ 3, indicated by vertical dotted lines). Numbers at the top of panel (a) denote the number of soil variables included in each model (1-4). All models included growth form as an explanatory variable. (b) Effect estimates from the highest empirical support models (ΔAIC ≤ 3). The abscissa (x-axis) indicates model numbers (14, 8, 15, 11), and the ordinate (y-axis) represents the natural logarithm of the hazard ratio (log hazard ratio). Point estimates and 95% confidence intervals (vertical lines) show regression coefficients. Statistical significance (P ≤ 0.05) is indicated when confidence intervals do not overlap zero (gray horizontal line). regression coefficients > 0 for soil variables indicate increased mortality risk per unit increase in soil difference; for growth form, regression coefficients > 0 indicate higher mortality risk for herbal compared to woody plants.

Fig. 8 Predicted effects of soil variable differences on introduced plant survival curves based on the model with the highest empirical support (Model 14 in Fig. 6). The panels display predicted survival probability (ordinate) over time since survival time (abscissa). The left column presents predictions for woody plants and the right column for herbal plants; the upper row (a-b) shows the effect of topsoil organic carbon content difference (ΔT_OCL), and the lower row (c-d) shows the effect of topsoil gravel volume percentage difference (ΔT_GRAVELL). In each panel, the curve labeled “XTBG” represents the reference scenario with zero soil difference from XTBG; the other four curves correspond to four equally-spaced gradient levels of that soil variable across the range from 0 to the maximum observed difference (including the maximum). Survival curves were derived from Nelson-Aalen cumulative hazard estimates.