The Qinghai–Tibet Plateau accounts for about 26.8% of China’s land area, covering the region of 26°00′12″–39°46′50″N and 73°18′52″–104°46′59″E ( Figure 1D ). Topographically, the QTP is high in the northwest and low in the southeast, with an average altitude of over 4000 m, a total area of about 2.5 million square kilometers, and complex and diverse landform types. There are significant regional differences in climate on the QTP, with temperatures showing a gradual decrease from south to northwest, and precipitation a distribution of more precipitation in the southeast and less in the northwest. The QTP is sensitive to climate change and is an ideal place for researching how alpine vegetation systems respond to global changes.

The following methods/data sources were used to collect data on the occurrence of L. virgaurea and L. sagitta: (1) From 2019 to 2021, a field survey was conducted using GPS to record elevation, longitude, and elevation data; (2) the Global Biodiversity Information Facility (www.gbif.org/); (3) the National Specimen Information Infrastructure (http://www.nsii.org.cn/2017/); and (4) the China National Knowledge Infrastructure (http://www.cnki.net/). We collected a total of 465 data on the occurrence of L. virgaurea on the QTP, and 128 on the occurrence of L. sagitta. These distribution data of L. virgaurea and L. sagitta on the QTP were then filtered using ArcGIS 10.2 to ensure that each 10×10 km grid cell contained only one record, which ultimately gave us 276 occurrences of L. virgaurea and 118 occurrences of L. sagitta for subsequent analysis ( Figure 1 ).

We selected 27 environmental variables for the model to predict the probability distributions of L. virgaurea and L. sagitta on the QTP ( Table 1 ), including 19 biological variables, three topographical variables, three soil variables, and two human activity variables. More specifically, to depict the current climatic situation, 19 biological variables from the WorldClim database, version 2.0 (https://www.worldclim.org/), were used with a spatial resolution of 30 arc-seconds, while Geospatial Data Cloud (https://www.gscloud.cn/) provided the three topographical variables, the National Tibetan Plateau Data Center (https://data.tpdc.ac.cn/home) provided the three soil variables, and the Human Footprint Index and actual livestock carrying capacity (https://data.tpdc.ac.cn/home) were the two human activity variables (Duan and Luo, 2021; Liu, 2021).

Future climatic variables were projected for two future periods (the 2050s and 2070s) using the BCC-CSM2-MR global climate model (GCM) under the conditions of four Shared Socioeconomic Pathways (SSPs) in CMIP6 (phase 6 of the Coupled Model Intercomparison Project): SSP126 (low GHG emissions: CO₂ emissions cut to net zero around 2075); SSP245 (intermediate GHG emissions: CO₂ emissions around current levels until 2050, then falling but not reaching net zero by 2100); SSP370 (high GHG emissions: CO₂ emissions double by 2100); and SSP585 (very high GHG emissions: CO₂ emissions triple by 2075). BCC-CSM2-MR was chosen because, based on nine widely used GCMs (BCC-CSM2-MR, CNRM-CM6-1, CNRM-ESM2-1, CanESM5, GFDL-ESM4, IPSL-CM6A-LR, MIROC-ES2L, MIROC6, and MRI-ESM2-0), researchers found that, when compared to its previous-generation (CMIP5) version, BCC-CSM2-MR greatly improved the simulation of temperature and precipitation changes in China, and more so than the other eight tested models (Xin et al., 2018).

Many variables exhibit spatial collinearity, which can lead a model to suffer from overfitting, which will ultimately impact the prediction outcomes (Hu and Liu, 2014). In order to filter the environmental variables, we applied Pearson correlation analysis to the 27 environmental variables. We then retained those variables with |r|<0.8, according to the Pearson correlation principle (Fang et al., 2021). This meant that, ultimately, eight climatic variables, three topographical variables, three soil variables, and two human activity variables were selected to participate in the model ( Table 2 ). Finally, we used the ArcGIS resampling tool to resample the resolution of the 16 selected environmental variable layers to 1 km for model analysis.

To predict the possible habitat distributions of L. virgaurea and L. sagitta on the QTP under climate change, the MaxEnt model, which is a model based on the theory of maximum entropy, was selected. More specifically, we used version 3.4.4 of MaxEnt to model the distribution points of L. virgaurea and L. sagitta on the QTP, as well as 13 environmental variables, and randomly selected 75% of the species distribution points as the training data and 25% as the test data. We then repeated the procedure 15 times using the cross-validation approach, wherein the maximum number of iterations was 1000, the output format was logistic, and an ASCII file was the final output (Guo et al., 2016).

The area under the ROC (receiver operator characteristic) curve value was used to measure model performance, which is a threshold-independent approach used to identify presence from absence in species occurrence (Katz and Zellmer, 2018). A model’s accuracy can be judged as excellent if the area under curve (AUC) value is between 0.9 and 1, good if AUC is between 0.8 and 0.9, fair if AUC is between 0.7 and 0.8, poor if AUC is between 0.6 and 0.7, and failed if AUC is between 0.5 and 0.6 (Phillips et al., 2006). The true skill statistic (TSS) has a value ranging from −1 to 1. The higher the TSS value, the greater the consistency between observed and predicted values, and the better the model effect; the lower the TSS value, the worse the consistency and the worse the model prediction impact (Fang et al., 2021). On this basis, the model’s performance was assessed using a combination of AUC and TSS.

The key ecological constraints restricting the ranges of L. virgaurea and L. sagitta on the QTP were identified by calculating the contribution of each environmental variable to the expected results in MaxEnt using the knife cut technique (Li et al., 2021). The ASCII files were imported into ArcGIS10.2 and transformed into raster data before being classified into four groups using the spatial analysis tool “reclassify”—namely, unsuitable habitat, low habitat suitability, medium habitat suitability, and high habitat suitability. The distributions and extents of appropriate habitats for L. virgaurea and L. sagitta on the QTP were determined.

To clarify the relationship between the predicted suitable habitat index and environmental variables of L. virgaurea and L. sagitta, this study used regression analysis. ArcGIS was used to extract the environmental data and the suitable habitat index of the distribution points. SPSS23 was used for regression analysis of key environmental variables whose cumulative contribution rate to the extracted data and the model was greater than 90%.

The average AUC and TSS values of model runs for L. virgaurea and L. sagitta were evaluated in the Maxent model ( Table 3 ). The results show that the prediction of the potential distribution areas of L. virgaurea and L. sagitta under current and future climatic conditions were highly accurate and performed extremely well. Under current environmental conditions, the mean AUC and TSS values of the L. virgaurea and L. sagitta model runs were (0.934, 0.922) and (0.943, 0.909), respectively. Furthermore, the average AUC and TSS values of the L. virgaurea and L. sagitta model runs were greater than 0.88, indicating that they were more accurate and performed better

We obtained the contribution rate of each variable in the MaxEnt model and analyzed the environmental variables that had a greater impact on the prediction result. The cumulative contribution rate of the four types of variables was calculated ( Table 2 ). Climatic variables accounted for 10.6% and 37.1% of the result for L. virgaurea and L. sagitta, respectively. Meanwhile, topographical variables accounted for 12.8% and 9.8%, soil variables accounted for 20.1% and 8.4%, and human activity variables accounted for 56.5% and 44.7%, respectively. The proportions for the human activity variables were larger, demonstrating that they had a greater impact on the model prediction results. In summary, human activity variables had the greatest impact on the potential distribution of L. virgaurea and L. sagitta, followed by climatic variables, while topography variables had the least influence.

The key factors, which had a cumulative contribution rate of more than 90% to the MaxEnt model, were used for regression analysis along with the suitable habitat index of distribution points ( Table 4 ). HFP and elevation had substantial (P<0.01) influences on the distribution of both L. virgaurea and L. sagitta, whilst S_TN had no effect. Bio15 had significant (P<0.01) impacts on the distribution of L. virgaurea and L. sagitta, respectively, whereas Bio13 had significant (P<0.01) effects on the distribution of L. sagitta, and Bio15, SLOP and ASPE had no significant influence on the distribution of L. sagitta. This indicates that human activity is the primary factor influencing the spread of L. virgaurea and L. sagitta on the QTP and that the distribution area and range of these two poisonous plants has expanded as the HFP intensity has increased.

For the present day, the results show that the suitable invasive areas for L. virgaurea and L. sagitta are mainly distributed in the eastern and southwestern part of the QTP, southwestern Gansu, eastern Qinghai, western Sichuan, southeastern Tibet, and northwestern Yunnan, with a total area of 8.92×10⁵km² and 10.65×10⁵km² ( Figures 2A, B ), respectively. Under the current climatic scenario, the potential distribution areas of L. virgaurea and L. sagitta account for 12.8%, 7.8%, 4.3%, 16.6%, 9.2%, and 4% of the total area of the QTP, respectively, and are mostly concentrated in Qinghai, Gansu, Sichuan, and Tibet ( Figure 3 ).

The total potential distribution area of L. virgaurea was projected to increase substantially over time under the four future emission scenarios. Specifically, the area accounted for 34.2%, 35%, 35.1% and 34.8% of the total area of the QTP in the 2050s under SSP126, SSP245, SSP370 and SSP585, respectively; and 32.2%, 31.8%, 32% and 32.5% in the 2070s ( Figure 4 ). The area under the SSP126 scenario increased the most from the 2050s to the 2070s, whereas it decreased under the other three scenarios. For the 2070s under SSP126, the overall area expanded the largest when compared to the existing distribution area, reaching 12.96×10⁵km² ( Figure 2A ). From the 2050s to the 2070s, the area of high habitat suitability increased and then decreased, with the largest extent in the 2050s under SSP370, at 2.89×10⁵km² ( Figure 2A ), while the area of medium habitat suitability increased and then decreased. In comparison to the existing distribution area, the prospective distribution region of L. virgaurea on the QTP was projected to expand owing to the effects of future climate change, with the amount of appropriate habitat peaking in the 2050s, mostly concentrated on the QTP. These locations are expected to be more favorable for the growth and reproduction of L. virgaurea in the face of climate change.

The total potential distribution area of L. sagitta was projected to grow greatly over time under the four future emission scenarios. Specifically, the area accounted for 32.2%, 31.8%, 32% and 32.5% of the total area of the QTP in the 2050s under SSP126, SSP245, SSP370 and SSP585, respectively; and 31%, 32.9%, 32.8% and 32.1% in the 2070s ( Figure 5 ). The area under SSP245 increased the most from the 2050s to the 2070s, followed by those under SSP370, SSP126 and SSP585, which all produced a decrease in the distribution area. For the 2070s under SSP370, the total area expanded the greatest when compared to the existing distribution area, reaching 12.01×10⁵km² ( Figure 2B ). From the 2050s to the 2070s, the area of high habitat suitability increased gradually, with a maximum area of 2.53×10⁵km² for the 2070s under SSP370, and the amount of medium habitat suitability decreased and subsequently increased under SSP245, SSP126 and SSP370 ( Figure 2B ). Under the effects of future climate change, the prospective distribution region of L. sagitta on the QTP is expected to expand in comparison to the existing distribution area. Under future climatic conditions, L. sagitta will likely be primarily distributed in the eastern and southeastern parts of the QTP, and the distribution areas of high habitat suitability will primarily be in southeastern Qinghai, indicating that these locations are more suitable for the growth of L. sagitta, and their climate is the primary reason for their expected expansion.

In this study, the climate factors and topographical factors of four emission scenarios (SSP126, SSP245, SSP370, and SSP585) under future climate change were adopted, and the MaxEnt model was used to simulate and predict the potential distribution areas and distribution ranges of L. virgaurea and L. sagitta on the QTP against the background of future climate change from the 2050s to 2070s. The results showed that, under the four emission scenarios, the distribution areas and distribution ranges of different habitats of L. virgaurea and L. sagitta increased and decreased with the change in future climate. This demonstrates that future climate change will have an important impact on the potential distribution range and suitable habitat of these two plant species. Among all the climatic factors, Bio15 and Bio13 were the most important, indicating that the growth and distribution of L. virgaurea and L. sagitta are strongly related with precipitation and temperature. Against the future climate background, with the increase in emissions intensity, the precipitation and temperature of the QTP will increase, and the change in temperature will be obviously different at different elevations. The overall finding was that the temperature increase in the high-elevation areas of the northern QTP will be greater than that in the low-elevation areas of the southeastern QTP; and meanwhile, the most significant increase in precipitation is expected in the northwest region (Luo et al., 2022). Plant growth is determined by water and temperature, and when drought and annual accumulated temperature change, plant growth will be severely limited (Ganjurjav et al., 2016). The growth and distribution of plants has an important relationship with soil nutrients. The soil TN content of high-density L. virgaurea and L. sagitta communities is higher. In the future, with the increase in temperature, the soil biological activities related to soil N transformation will be promoted, which will accelerate the soil N cycle, increase the content of soil available N, and further promote the growth and distribution of L. virgaurea and L. sagitta (Mei and Ma, 2021). The MaxEnt simulation predicted the geographical distribution of L. virgaurea and L. sagitta on the QTP in the context of future climate change, and this information can be used to monitor and provide advanced warnings regarding the occurrence of these poisonous plants, as well as serve as a reference for the formulation of prevention and control measures in key areas.

In summary, this study predicted the current and future distribution of L. virgaurea and L. sagitta through the MaxEnt model. Although there are a few studies that have reported the distribution of poisonous plants and predicted the potential distribution of poisonous plants at local scales in some regions under climate change (Huang et al., 2020), there is, however, a lack of reports on the widespread distribution of these two poisonous plants on the QTP, especially considering the joint effects of climate change and human disturbance. Our study provides a basis for understanding environmental factors and the reasons for the widespread distribution of two poisonous plants, as well as serving as a reference for managing the spread of poisonous plants in the future. However, we only considered the role of abiotic factors in predicting the spread of poisonous plants. It is also very important to study the role of biological factors, such as plant intraspecific and interspecific competition, in predicting the spread of poisonous plants in the future.