Changes in the ecosystem functions along the elevation gradients.
Since most prior studies have largely focused on the influence of PTE transport on biodiversity in polluted areas with a single natural habitat30,31, the correlation between ecological functions and PTE transport has never been fully analyzed at multiple spatial scales (i.e., across climate zones). However, in the natural habitats of the Qilian Mountains, most soil and plant-mediated functions varied linearly along the elevation gradients. For the natural ecosystem functions mediated by plants and soil, climate and topography jointly predicted the changes in habitat and ecosystem functions (Fig. 3). On average, these factors explained 65% (± 23% (s.d.), ranging from 20% to 99%; Supplementary materials Table 4) of the changes in the ecosystem functions for the natural habitats over all the elevation gradients. Most soil- and plant-mediated functions peaked at lower, and middle elevations (2370-3051 m a.s.l), where the highest MAP (353.91~445.93 mm year-1), and relatively low MAT (4.42~6.02℃) were recorded (Fig. 4). In contrast, pH, electrical conductivity (EC ), and total salt content declined with elevation, whereas alkaline N peaked at lower-middle elevations (Fig. 4). For the natural ecosystem functions, the effect of topography (i.e., slope and aspect) was less important than that of climatic heterogeneity (MAT, win10, and ET0) in predicting ecosystem functions (Fig. 3). For the plant-mediated ecosystem functions, the absolute effect strength values were, on average, higher for climate change (i.e., ET0, MAT, and win10), whereas the soil-mediated ecosystem functions were strongly explained by both topography (i.e., slope and elevation) and climate change (i.e., MAT, win10, and ET0) (linear mixed effect model, P < 0.05).
———————-
Place Fig 3 here
———————-
The majority of the ecosystem functions (explanatory variable = 92% ± 5% (SD), P < 0.005) were altered by the presence of PTEs, with the interaction of topographic heterogeneity and climate change (Fig. 5a). In most cases, the presence of PTEs varied with both the climate and topography of the study sites in different climate zones (Supplementary materials Table 2). The two-way interaction models (PTEs-topography and PTEs-climate) for different elevations explained most of the 21 ecosystem functional variables (explanatory variable = 39% ± 20% (s.d.), P < 0.005, explanatory variable = 47% ± 26% (s.d.), P < 0.005) (Supplementary materials Table 2). In the three-way interaction models (climate-PTEs-topography), both climatic and topographic variables were input, causing the explained variation in ecosystem functions to increase to 53% ± 26% (s.d.) and indicating that the interactive model supported the empirical data more strongly (Supplementary materials Table 2). The climate zones in which the strongest effects of PTE intensity on ecosystem functions were observed varied for different response variables (Fig. 4). For example, the C/N content of the plants in the PTE-polluted habitats decreased as the elevation increased, which occurred especially in the mine areas of the warm and relatively humid low-elevation mountain grasslands (2370-2500 m a.s.l) (residual effect = 0.21, R2 = 0.57, P < 0.005). The NDVI of the PTE-polluted habitats in the arid low-elevation mountain desert steppe (1000-1500 m a.s.l) changed significantly, as explained by the local climate (residual effect = -0.21, R2 = 0.90, P < 0.005; Supplementary materials Tables 2, 3). The large degree of support for the climate-PTEs-topography interaction models across ecosystem functions was robust to different PTEs criteria (Supplementary Note 1, Supplementary materials Tables 5, 6 and Supplementary Materials Figs. 8, 9) and allowed us to consider potentially confounding environmental variables that systematically change with elevation (Supplementary Note 1 and Supplementary materials Tables 5, 6). As plant species and ecosystem functions may respond more strongly to certain factors related to the ecological risk induced by PTE transport, we tested whether the uncertainty of ecological risk was affecting the prediction of the response variables. Our results showed that most of the models equally supported the prediction of the response variables with or without considering the uncertainty of ecological risks. Furthermore, we also tested three models incorporating only a subset of the original data, and the results supported the complete three-way interaction models (climate-PTEs-topography) in most cases (explanatory variable = 63% ± 23% (s.d.), P < 0.005; Supplementary Materials Figs. 2-4 and Supplementary Materials Tables 6, 7, 8).
———————-
Place Fig 4 here
———————-