Research on the synergistic prediction of the suitable distribution and chemical components of Panax Notoginseng under the background of climate warming

Abstract

Panax notoginseng is a well-known research species in China. The issue of continuous crop barriers has led to a reduction in suitable habitats in Wenshan. In the context of global warming, it is far from adequate to merely predict the suitability distribution of P. notoginseng under future climate, as it remains unclear whether the Highly suitable habitats in the results are also the high content areas. We used machine learning for the first time to predict the changes in chemical components under future climate scenarios and found that it was more scientific than the Biomod2 based on environmental variables. The results show that TreeBagger is an effective tool to predict the chemical composition of P. notoginseng under future climate change. Bio7 was an important environmental variable affecting the distribution of P. notoginseng, and Bio15 was the environmental variable that had the greatest impact on the quality of P. notoginseng. Comprehensive evaluation indicated that Pre11 might be a key factor affecting the distribution and quality of P. notoginseng. In the future, under climate change, most of the Highly suitable habitats for P. notoginseng in Wenshan area will be downgraded to moderately suitable and lowly suitable areas. Therefore, it is necessary to focus on protecting its ecosystem and plan the cultivation of P. notoginseng in Yuxi and its surrounding areas, which can effectively relieve the cultivation pressure in Wenshan area. This study can provide new ideas and methods for the future research on the species distribution of medicinal plants.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12870-025-07169-3.

Introduction

Global climate change has led to significant variability in temperature and precipitation patterns, along with an increased frequency of extreme climate events, making it one of the most pressing concerns in ecological research [1]. These climatic changes are profoundly affecting species distributions, with medicinal plants being particularly vulnerable as variations in temperature and precipitation differentially influence both their geographic range and biochemical composition. Studies have demonstrated that future climate change will inevitably alter the extent of suitable cultivation areas, and such shifts have already substantially hindered the development of sustainable utilization strategies [2]– [3]. Certain high-value medicinal plants face increasing endangerment due to the combined pressures of economic demand, climate change, and human activities [4]. Commercial cultivation has emerged as a crucial conservation strategy for these species. Among them, P. notoginseng (Burk.) F.H. Chen (commonly called P. notoginseng), derived from the dried root of the plant, represents one of China’s most valuable bulk-traded medicinal herbs. With a cultivation history spanning over 600 years, it is revered as “the king of ginseng” and considered “priceless gold” in traditional medicine. This important herb is typically marketed in raw form or processed into traditional Chinese medicine formulations, maintaining a dominant position in regional markets [5]– [6]. The species possesses the efficacy of activating blood circulation and removing blood stasis, subduing swelling and fixing pain, nourishing and strengthening the body, and resisting tumors [7]. It’s favored by consumers for both its healthcare and therapeutic roles, making it one of the economic lifelines of Yunnan Province and a major advantage for industrial development and rural revitalization. P. notoginseng mainly grows in mountainous areas at an altitude of 1,200-2,000 m [8]. and Wenshan City, Yunnan Province, the hometown of P. notoginseng, is the traditional origin of P. notoginseng. In recent years, with the rapid development of the P. notoginseng industry and the shortage of land resources caused by factors such as continuous cropping barriers during cultivation, P. notoginseng cultivation has spread to other counties and cities in Yunnan Province, with Wenshan as the center of its traditional production area. Currently, the P. notoginseng plantation industry in Honghe Prefecture, Yuxi City, Kunming City, and Qujing City has formed a certain scale [9]– [10]. Currently, P. notoginseng is mainly cultivated through human intervention, but it is confronted with a serious issue of repeated cultivation. Generally speaking, a piece of land that has been planted with P. notoginseng requires consecutive cultivation with crops such as corn for more than ten years before it can be replanted [11]. P. notoginseng prefers a warm climate in winter and a cool one in summer, does not tolerate extreme cold and heat, and favours a semi-shaded and moist ecological environment. Since P. notoginseng has extremely strict requirements for the growing environment and there is no effective method to overcome the obstacle of continuous cropping, resulting in the continuous reduction of the P. notoginseng cultivation area and the decline in yield, it is urgent to expand the P. notoginseng cultivation area and enhance the quality and yield of P. notoginseng. However, the introduction and cultivation of medicinal plants must adhere to the principle of consistency between the environment and medicinal use, and the selection of areas similar to the origin environment for the introduction and cultivation of medicinal plants is the key to ensuring the quality of medicinal plants and meeting the market demand [12]. Currently, only the distribution of P. notoginseng in the current climate has been predicted in conjunction with chemical composition in known studies, but there is a gap in research on how the chemical composition will change in the context of future climate. Therefore, it is important to predict the distribution of P. notoginseng under future climate change along with the chemical composition content.

Species distribution modeling is an approach often utilized for biodiversity conservation. It mainly encompasses Generalized Boosted Models, Generalized Additive Models, Generalized Linear Models, Artificial Neural Network, Classification Tree Analysis, one Rectilinear Envelope Similar to BIOCLIM, Flexible Discriminant Analysis, Regression Splines, Random Forest, Multivariate Adaptive Maximum Entropy Models, and the like [13]. Among them, the MaxEnt model is relatively extensively used because it can quantify the influences of climatic factors on species distribution [14]. Since it is a single model, the simulation results are prone to differ from those of other models. Biomod2, which integrates more than ten commonly utilized single algorithms for SDM, generates better results than a single model. It is worth noting that the prediction of the fitness zones of P. notoginseng in the current study was mainly based on the MaxEnt model [15]– [16]. Due to the limitation of a single model, it is necessary to use a more comprehensive Biomod2 model to predict the fitness zone of P. notoginseng.

However, there is a shortage in the research related to the chemical content of medicinal plants under future climate change, leading to the results of the species distribution modeling being only able to indicate their suitability for growth, and not simultaneously represent high chemical content in Highly suitable habitats. The simultaneous prediction of species distribution together with its chemical content under future climate would further enhance the final simulation results. Bagging in integrated learning algorithms is a classical algorithm, and random forest (RF) is one of the most representative approaches due to its outstanding performance and generalization capacity [17]. It has been applied in current research for the yield prediction of wheat under future climate [18–21]. TreeBagger, a classification and regression tool developed in Matlab, pertains to the Random Forest technique. It assesses the model results based on the root-mean-square error and is much easier to apply in prediction. Combining TreeBagger’s ability to predict species content in future climates with the predictions of the Biomod2 model could further optimize studies of suitable habitat for P. notoginseng.

The scientific hypothesis of this study is that future climate change will have differential impacts on highly suitable and high quality habitats for P. notoginseng. The novelty was the successful prediction of P. notoginseng content under future climate scenarios in combination with machine learning, which in combination with Biomod2 yielded areas that are both highly suitable and high quality habitats, and Bridging the gap between previous studies. Therefore, the objective of this study was to predict the saponin content of P. notoginseng under future climate conditions via machine learning, and to assess comprehensively the suitable habitats of P. notoginseng under the future climate through spatial analysis in combination with the outcomes of the Biomod2 model, so that the Highly suitable habitats are in line with the regions of high content, the moderately suitable areas are in line with the regions of medium content, and the low suitable areas are in line with the regions of low content, and to clarify the influences on the distribution of P. notoginseng. Meanwhile, the environmental variables affecting the distribution and quality of P. notoginseng were disclosed. The results of this study can relieve to a certain extent the pressure of plantation planning due to the continuous cropping barrier of P. notoginseng, and offer a more comprehensive scientific foundation for the introduction and cultivation of related medicinal plants in the future.

Materials and methods

Collection of species distribution data and environmental variables

Species distribution points were obtained from the Chinese Virtual HeRb1arium (CVH: http://www.cvh.ac.cn/), the Global Biodiversity Information Facility (GBIF: 10.15468/dl.6vde6u), and the National Specimen Information Infrastructure (NSII: http://www.nsii.org.cn/). Preliminary data regarding the distribution of P. notoginseng were obtained from 3332 entries. When the distribution points are excessively close to one another or the data are duplicated, manual screening of the distribution points might cause overfitting of the model results. Therefore, we employed the ‘spThin’ package in R software to conduct 50 iterations of spatial thinning on the species distribution records, applying a 5 km buffer distance to ensure the scientific validity of the point data [22]. 2883 distribution points were utilized for the screening of distribution points. The 2883 distribution point records are the final outputs of the screening process. This includes 45 field trip records, climatic conditions at the sampling sites were recorded in detail through Arcgis software and were categorized into five regions for subsequent analysis (Table S1). Future climate data were acquired from four shared socioeconomic pathway (SSP126, SSP245, SSP370, and SSP585) scenarios of the global circulation model BBC-CSM2-MR downloaded from WorldClim [23]. The main focus was on the distribution of suitable habitat for P. notoginseng over the next 80 years, 2021–2040 s, 2041–2060 s, 2061–2080 s, and 2081–2100 s. After reviewing the relevant literature, it was found that the contribution of climatic variables far exceeded that of soil variables [15]– [16, 24, 25]. Therefore, a total of 32 environmental variables related to the species’ distribution were selected, including 19 climatic factors (Bio1-19) 12 precipitation factors (Pre1-12) and elevation, at a resolution of 30 s (1km × 1 km), raster data in tif format, with the coordinate system set to WGS 1984, and finally, ArcGIS Pro was used to standardize the coordinate system, number of rows, and number of columns for all raster data.

In addition, 45 P. notoginseng samples (No. YAAS20240723001) collected from field investigations were identified by Dr. Huang Hengyu from Yunnan University of Traditional Chinese Medicine. All field studies and the acquisition of plant materials were conducted in accordance with local and national regulations, and no additional permits or licenses were required. These samples are preserved in the Herbarium of the Institute of Medicinal Plants, Yunnan Academy of Agricultural Sciences.

MaxEnt modeling and filtering environment variables

MaxEnt has become one of the models widely used for medicinal plant distribution prediction in recent years because of its high accuracy. Before simulating the prediction of the potential distribution of P. notoginseng under the current scenario through the distribution point locations and environmental variables described above, the model output was formatted as a Cloglog transformation [14]. In order to allow sufficient time for the model to converge, 70% of the distribution point data were used as a training model and 30% of the data were used for model evaluation with 5000 iterations, and model predictions were averaged after 30 repetitions [26]. The modeling was continued after excluding environmental variables contributing 0 based on the prediction results until there were no variables contributing 0. Environmental variables are correlated with each other, and when the correlation is too strong it is easy to lead to overfitting of the prediction model, so it is necessary to screen the environmental variables before constructing the final prediction model [27]. When the correlation coefficients between environmental variables are too high, they can have an impact on the model results. However, relying on correlation coefficients alone may exclude ecologically relevant variables, so variance inflation factor (VIF) analysis was used to more fully assess multicollinearity, a step accomplished through R software (Table S2).

Biomod2 modeling

The GBM, GLM, CTA, ANN, RSE, FDA, MARS, RF, GAM, XGBOOST, and MAXENT models were constructed in R software using the Biomod2 (4.2-4) program package, respectively. Prior to model building, MaxEnt was optimized using the Wallace (2.1.2) program package, and after model building the feature combinations (FC) and regularized multipliers (RM) were used as evaluation metrics.RM was taken in the range of [0.5, 3.5] with an interval of 0.5, and the FCs were L, H, LQH, and LQHP, and a total of 36 models were built for the Wallcae test [28]. Artificial models had strong performance when delta. AICc < 2 and close to 0 for different FCs were evaluated [29]. The parameters of other single models are default values, so the optimized parameters are RM = 0.5, FC = LQHP, and the single model is completed and further formed into an ensemble model EnsembleModel after evaluated by two metrics, RCO and TSS (Table S3). ROC is one of the measures of model performance, and AUC is a sensitivity map to specificity, which is currently a judgmental assessment index with a high degree of discriminatory power, and when ROC > 0.9, it suggests that the model’s performance is stable and excellent [26]. Related studies have shown that using only AUC as a predictor for spatial distribution modeling is inadequate. Hence, TSS was introduced as a supplementary measure as it inherits all the advantages of the Kappa statistic and has a significant correlation with ROC. TSS in this study mainly makes use of distribution points and pseudo-distribution points to calculate the prediction success of the model. The closer the TSS is to 1, the higher the prediction accuracy of the model. And when the TSS is closer to −1, it implies that the model is closer to a random model with pseudo-distribution points, and the worse its accuracy is [14]. Subsequently, 70% of the distribution points were employed for model training and 30% for model validation. After constructing the model and randomly generating 1,000 pseudo-distribution points, the run was repeated 10 times (Fig. S1). Additionally, the integrated model built by the Biomod2 model in this study consists of four single models, namely ANN, RF, GBM, and GAM (Fig. S2).

Fieldwork and sample collection for P. notoginseng

Five areas with concentrated distribution points were selected for sample collection of P. noteginseng, namely Kunming, Yuxi, Honghe, Wenshan, and Qujing, with a total of 45 sampling points, and more than 30 samples of P. noteginseng were collected from each locality, and species identification was carried out by Prof. Hengyu Huang of Yunnan University of Traditional Chinese Medicine (Fig. S3). Finally, the P. notoginseng samples were dried to constant weight using an electrothermal constant temperature dryer (Shanghai Yiheng Scientific Instrument Co., Ltd.), with the temperature set at 50℃. The samples were then dried to constant weight using a portable high-speed grinder. The samples were pulverized using a portable high-speed grinder and passed through a 100-hole sieve, and the final sample powder obtained was stored in a sealed bag for subsequent chemical analysis.

High performance liquid chromatography (HPLC) analysis

Plotting of saponin control standard curve

Weighed the ginsenoside R1 10.0 mg, Rg1 20.0 mg, Rb1 20.0 mg and Rd 10.0 mg in 10 mL volumetric flasks, add methanol to dissolve and dilute to the scale, and then shake well to make a control solution of P. notoginsengoside R1 1.0 mg/mL, Ginsenoside Rg1 2.0 mg/mL, Ginsenoside Rb1 2.0 mg/mL, Ginsenoside Rd1 1.0 mg/mL, Ginsenoside Rb1 2.0 mg/mL, Ginsenoside Rd 1.0 mg/mL. Rb1 2.0 mg/mL, ginsenoside Rd 1.0 mg/mL. Precisely measure 0.1, 0.2, 0.4, 0.8, 1.0, 1.5 and 3.0 mL of control solution of P. notoginseng ginseng ginsenoside Rg1, saponin R1, ginsenoside Rd and ginsenoside Rb1 in a 5 mL volumetric flask with methanol to the scale, and then formulate into a series of control solution. Pipette 1 µL of the control solution into the ultra-high performance liquid chromatograph, record the peak area of the control at different concentration gradients, and calculate the standard curve equation between the peak area and the concentration of the control.

Saponin content test

The sample powder was accurately weighed 40.0 ± 0.2 mg in a clean test tube, 2.0 mL of methanol was added, and the extraction was sealed and ultrasonicated for 40 min. After the extraction, the loss of methanol was replenished, and the extract was filtered by 0.22 μm organic microporous filter membrane to obtain the sample solution. 1 µL of the sample solution was aspirated and injected into an ultra-high performance liquid chromatograph (UPLC) with the mobile phase of ultrapure water (A)-acetonitrile (B) at a flow rate of 0.35 mL/min, the column temperature was 45 ℃, and the detection wavelength was 203 nm. The sample solution was extracted with the mobile phase of ultrapure water (A)-acetonitrile (B) at a flow rate of 0.35 mL/min, the column temperature was 45 ℃, and the detection wavelength was 203 nm. 56%~20% B; 13 ~ 14 min, 20% B for gradient elution. The saponins in the chromatograms of the samples were fingered and the peak areas were recorded with reference to the retention time of the control, and the corresponding saponin contents were calculated according to the standard curve of the control [24] (Table S4).

TreeBagger modeling to predict saponin under future climates

TreeBagger is a bagged classification tree, and Y is usually a set of class labels, so this study focuses on the regression problem, and Y will be used as a numerical vector to correspond to “regression” and “method”, respectively. After determining the saponin content of P. notoginseng using HPLC, a TreeBagger regression model was constructed by combining the data set of the current climate of the field site with the corresponding saponins, and then predicting the corresponding saponin content based on the data set of environmental variables of the future climate. The model performance was mainly evaluated by RMSE, the smaller RMSE, the better the model performance [30]. R² was used to demonstrate the fitting ability and stability of the model, the closer it is to 1, the better it represents its ability. In addition, the sub-steps are operated by MATLAB14.0 software to complete.

Visualization of saponin content of P. notoginseng under current and future climates

In order to combine the saponin contents of P. notoginseng with its suitable habitats under different climatic conditions, the saponins R1, Rb, Rd and Rg1 were visualized using kriging interpolation. Before using kriging interpolation, the data set was first exploratively interpolated in ArcGIS Pro using the “Geostatistical Analysis Tool”, and analyzed using ordinary kriging, simple kriging, empirical Bayesian kriging, and general kriging, respectively. The empirical Bayesian kriging interpolation method was finally used for the subsequent studies.

Ranking of suitable habitats

The use of natural discontinuities (Jenks) method was used to classify suitable habitats for P. notoginseng as it is often used to classify suitable habitats for medicinal plants [14]. Thus, the suitable habitats of P. notoginseng were classified into four categories: unsuitable habitats (0-0.15), Low-suitable habitats (0.15–0.45), Medium-suitable habitats (0.45–0.75) and High-suitable habitats (0.75-1). Due to the different saponin contents of P. notoginseng in different climates, it was not possible to classify them manually, so the four saponins R1, Rb, Rd and Rg1 were classified into four categories based on the percentage of their content values: Low (0–45%), Medium (45%−55%), High (55%−70%) and Very High (70%−100%), respectively.

Saponin content of P. notoginseng superimposed on suitable habitats

The quantitative data of P. notoginseng were overlaid with the suitable habitats through the “Overlay Analysis” in the tool, which corresponded to the four zones of saponin content in descending order: low (0–45%), medium (45%−55%), high (55%−70%) and very high (70%−100%). 100%). In this way, the suitable habitat for P. notoginseng can be obtained not only for the growth and planting of P. notoginseng, but also for its high content region at the same time.

Correlation analysis

The values of the environmental variables corresponding to the P. notoginseng field sites were extracted and correlated with their saponin content using Spearman’s correlation analysis to analyze the key factors affecting their quality.

Center of gravity transfer

The center of mass of suitable habitats for P. notoginseng based on climatic variables and suitable habitats incorporating content distribution were calculated using the “Spatial analyst” tool of ArcGIS pro. In the known studies, most of the areas in Yunnan belong to the suitable growing area of P. notoginseng, so in order to determine the variation of the most suitable growing area, this study only focused on the highly suitable area for center of mass transfer.

Results and discussion

Distribution of P. notoginseng species

According to the 2,883 distribution point records, P. notoginseng was mainly concentrated in the eastern part of Yunnan, with a small portion in southeastern Sichuan, southwestern Chongqing, southwestern Guizhou and western Guangxi. Overall, the distribution of P. notoginseng was mainly spreading from eastern Yunnan to the surrounding areas (Fig. 1).

Fig. 1.

Record of geographical distribution of P. notoginseng

Environmental variables affecting P. notoginseng distribution

Prior to establishing the integrated model, 11 species distribution models were set up to predict the suitable habitats of P. notoginseng. The results of Figure S4 reveal that among the six different integration methods, the environmental variable exerting the greatest influence on the distribution of most P. notoginseng was Bio7, and the order of the magnitude of the contribution rate was once ranked as Bio7 > Pre11 > Ele > Bio3 > Pre4 > Bio15. The main environmental variables affecting the distribution of P. notoginseng were Bio7 and Pre11, and the remaining four environmental variables had a negligible influence as their contribution rate was less than 10%. The main environmental variables influencing the distribution are Bio7 and Pre11, and the remaining four environmental variables do not have a significant effect because their contribution is less than 10%.

Potential suitability distribution of P. notoginseng under current climate conditions

Model simulations showed that most of the areas suitable for P. notoginseng growth were concentrated in Yunnan, with a small number concentrated in Tibet and southeastern Sichuan, and suitable habitats in Guizhou and Hunan were mainly located at the border with Yunnan. In addition to the northwest and northeast parts of Yunnan that are suitable for the growth of P. notoginseng, there are some moderately suitable areas and lowly suitable areas in the central part of Yunnan in the northwest and southwest directions, and Highly suitable habitats in other areas, indicating that P. notoginseng can adapt to most of the climatic types in Yunnan (Fig. 2).

Fig. 2.

Potential suitability distribution of P. notoginseng under current climate scenarios

Potential suitability distribution of P. notoginseng under future climates

Under the influence of future climate change, the SSP126 scenario does not show significant changes at 30 s, 50 s and 70 s, and at 90 s, the Highly suitable habitats in western Yunnan will degrade to moderately suitable and lowly suitable areas. The SSP245 scenario shows that the Highly suitable habitats at 30 s will increase in the north and south, and the Highly suitable habitats at 50 s and 70 s will decrease over time, and at 90 s the distribution will reverting to the distribution of the current scenario. In the SSP370 scenario, the socio-economic development begins to emphasize the economy, CO₂ emissions are increased, and the ecological environment begins to be damaged, and it can be clearly seen that the degradation of the highly suitable area of P. notoginseng in Yunnan is getting worse and worse with the increase of time. In the SSP585 scenario, along with the deepening of ecological damage, the suitable habitat of P. notoginseng in the 70 s and 90 s will be dominated by moderately suitable zones, with the highly suitable zones less than 50% under the current climate scenario, and the highly suitable zones in the neighboring regions of Yunnan will also be degraded to moderately suitable zones and lowly suitable zones (Fig. 3). The above results only represent its suitability for P. notoginseng growth, but the content in future climate scenarios is not clear, therefore, the saponin content of P. notoginseng in future climate scenarios will be predicted in the next study.

Fig. 3.

Potential suitability distribution of P. notoginseng under different future climate scenarios

Performance evaluation of treebagger

Highly suitable zone is not equal to high quality zone, it may be suitable for plant growth, but it does not necessarily mean that it has high quality P. notoginseng. Therefore, to further determine the P. notoginseng under future climate that is suitable for P. notoginseng growth and is a high quality zone, the four saponin contents of P. notoginseng under future climate were predicted using the TreeBagger model, and the results showed that most of the original values and the The results show that most of the original and predicted values fit well, the RMSE values of Fig. S4A, B and C are less than 2, which proves that the model predicts the effect of small error, the RMSE value of Fig. S4D is 2.176, which may be caused by too large a discrepancy between the data of Rg1, based on the reliable range of the four saponins content of the R² results show that the TreeBagger model is highly stable (Fig. S5). Evaluating the suitability of P. notoginseng in conjunction with quantitative data in future climates will further optimize the model and make the results more accurate.

Machine learning predicts the distribution of saponin content of P. notoginseng under future climate scenarios

Fieldwork sampling of five regions in Yunnan Province, including Wenshan, Kunming, Honghe, Qujing, and Yuxi, based on surveys and literature, and determination of their saponin contents revealed that P. notoginseng from WS was of the highest quality and had the highest contents of the saponins R1, Rg1, Rb1, and Rd, and that the four saponin contents of the five origins were ranked in descending order as WS > YX > KM > HH > QJ (Fig. S6). The four saponin contents of P. notoginseng under different climate scenarios were visualized as the distribution in Yunnan after using empirical Bayesian kriging interpolation. In the current climate scenario, the regions with higher saponin content were found to be Yuxi and Wenshan, respectively (Fig. 4). In Fig. S7-Fig. S10, the high saponin content of P. notoginseng under the future climate scenario is mainly concentrated in WS, and the low content is mainly distributed in KM, HH, and QJ and some neighboring areas. After spatial superimposition, it can be further analyzed as a potentially suitable distribution of P. notoginseng in highly suitable and high quality zones.

Fig. 4.

Distribution of different saponin contents of P. notoginseng under current climate scenarios. (A) R1. (B) Rg1. (C) Rb1. (D) Rd

Potential suitability distribution of P. notoginseng under different climate scenarios after spatial superposition

In contrast to the prediction model based on environmental variables, in the final results, the suitable habitat for P. notoginseng was mainly concentrated in Kunming, Yuxi, Honghe, Wenshan, and Qujing, and the majority of its area was highly suitable (Fig. 5). In the future SSP126 scenario, there was little change in the suitable habitat for P. notoginseng, and at 90 s, a part of the highly suitable area would be reduced in Wenshan and the southern part of the Red River. In the future SSP245 scenario, the Highly suitable habitats in Kunming and northern Qujing would decrease at 50 s, 70 s and 90 s, and the reduction in Highly suitable habitats in Wenshan and southern Honghe would be more obvious at 70s. In the future SSP370 scenario, at 70 s and 90 s, a large number of moderately adapted zones would emerge in the central region connecting Yuxi, Kunming, Qujing, and Honghe over time, suggesting that this part of the region is no longer suitable for growing high-quality and highly adapted P. notoginseng, with the overall situation in Kunming being the most severe. In the SSP585 scenario, which is accompanied by continuous and large emissions of fossil fuels and CO₂, Highly suitable habitats for P. notoginseng will be reduced by at least 40% in the 70 s and 90 s, with Highly suitable habitats concentrated in Yuxi and Honghe. It is worth noting that in the 70 s, Wenshan will not be the highest-quality area for P. notoginseng cultivation (Fig. 6). Overall, under the influence of future climate, the suitable planting area of P. notoginseng will remain relatively stable, the low suitability area and medium suitability area will show an increasing trend, and the high suitability area will be degraded to medium suitability area and high suitability area as time goes on under different climate scenarios. Among them, the area of highly suitable area will decrease by 46.04% and 43.99% respectively in 2070 s and 2090 s in the SSP585 scenario (Table 1).

Fig. 5.

Combined distribution of P. notoginseng suitability and saponin content in the current scenario after spatial superposition

Fig. 6.

Integrated distribution of P. notoginseng suitability and saponin content under spatially superimposed future climate scenarios

Table 1.

Area of P. notoginseng suitable habitats under different future climates

Scene	Period	Hight suitable area	Change	Medium suitable area	Change	Low suitable area	Change	No suitable area
Current	/	10.23	/	0.97	/	0.17	/	0.00
SSP126	2030s	9.69	−5.28%	1.40	44.33%	0.27	58.82%	0.01
	2050s	9.61	−6.06%	1.45	49.48%	0.29	70.59%	0.02
	2070s	9.53	−6.84%	1.50	54.64%	0.33	94.12%	0.01
	2090s	8.60	−15.93%	2.11	117.53%	0.58	241.18%	0.08
SSP245	2030s	9.96	−2.64%	1.17	20.62%	0.23	35.29%	0.01
	2050s	9.12	−10.85%	1.76	81.44%	0.41	141.18%	0.08
	2070s	8.86	−13.39%	2.11	117.53%	0.34	100.00%	0.05
	2090s	8.79	−14.08%	2.02	108.25%	0.55	223.53%	0.02
SSP370	2030s	9.35	−8.60%	1.66	71.13%	0.32	88.24%	0.04
	2050s	9.94	−2.83%	1.24	27.84%	0.18	5.88%	0.01
	2070s	8.32	−18.67%	2.80	188.66%	0.24	41.18%	0.01
	2090s	7.81	−23.66%	3.22	231.96%	0.33	94.12%	0.01
SSP585	2030s	9.08	−11.24%	1.74	79.38%	0.48	182.35%	0.07
	2050s	9.28	−9.29%	1.66	71.13%	0.38	123.53%	0.04
	2070s	5.52	−46.04%	4.71	385.57%	0.98	476.47%	0.15
	2090s	5.73	−43.99%	5.01	416.49%	0.54	217.65%	0.03

Note: The area unit is × 10⁴ km²

Dynamic distribution of suitable habitat for P. notoginseng under future climate impacts

Figure 7 shows that different climate scenarios and time periods have diverse impacts on the highly suitable area of P. notoginseng, and at the same time, it can reflect the response of the ecological environment to the effects of anthropogenic activities. Under the SSP126 scenario, the highly suitable area for P. notoginseng due to human activities mainly decreases in the southern part of Wenshan and the southeastern part of the Honghe River, and a small amount of highly suitable area decreases in Kunming and northern Qujing in the 2050 s, 2070 s, and 2090 s, with the highly suitable area in northern Kunming increasing over time, which is the most pronounced in the 2030 s and 2090s. The two scenarios, SSP245 and SSP370 scenarios, are similar in future climate change, with highly suitable zones mainly increasing in the northern region and decreasing in the southern region in 2030 s and 2050 s, and decreasing in the northwestern and southern regions in 2070 s and 2090s. The SSP585 scenario has more highly suitable zones decreasing in 2030 s compared with the other scenarios, and a small amount of highly suitable zones decreases in most areas of Wenshan in 2070s. In 2070 s, the most severe reduction in Highly suitable habitats was observed in most areas of Wenshan, and in 2090 s, the most severe reduction in Highly suitable habitats was seen in the north of the central part of Qujing and Kunming.

Fig. 7.

Dynamics of the combined distribution of suitability and saponin content in Highly suitable habitats for P. notoginseng under spatially superimposed future climate scenarios

Center-of-mass transfer of P. notoginseng in future climates

The comparison of the center-of-mass transfer of P. notoginseng based on environmental variables and the superimposed center-of-mass transfer of P. notoginseng revealed that the center-of-mass transfer based on environmental variables mainly took place in Chuxiong, Yuxi and Puer, while the superimposed center-of-mass transfer mainly occurred in Kunming, Honghe and Wenshan (Fig. 8A). This indicates that the results are more precise with the addition of quantitative data, as Kunming, Honghe and Wenshan are the main cultivation sites of P. notoginseng, and the results obtained with the assistance of content data are closer to reality. In Fig. 8B, under different climatic influences in the future, the highly suitable area of P. notoginseng generally demonstrated a trend of migration from south to north, and it is notable that the center of mass of the superimposed highly suitable area of P. notoginseng was mainly shifted to the Honghe region in the future (Fig. 8C). The results of both center of gravity shifts suggested that Wenshan, which is a highly suitable area and a high content area under the current climate, will be degraded to a moderately suitable area or a low suitable area under the influence of the future climate, which proves that the ecological environment in Wenshan needs to be protected in the future.

Fig. 8.

Changes in center-of-mass transfer in P. notoginseng. (A) Scope of change. (B) Center-of-mass transfer for Biomod2 modeling based on climatic factors. (C) Center of mass transfer after spatial superposition

The effect of environmental variables on distribution and quality of P. notoginseng

Correlation analysis indicated that Rg1 was significantly negatively correlated with Bio7, suggesting that the higher the Bio7, the lower the content of Rg1. All the contents of the four saponins were significantly and positively correlated with Bio15, signifying that Bio15 is a key environmental variable influencing the quality of P. notoginseng. Notably, the high and low contents of Rb1 and Rg1 were mainly impacted by moisture (Fig. S11). Generally, the mass of P. notoginseng is more influenced by moisture rather than temperature. It is evident that among the different integrated models, Bio7 and Pre11 are the key environmental variables influencing the distribution of P. notoginseng. In the EMmean model employed in this study, the contribution rate of Bio7 exceeds 20%, while that of Pre11 surpasses 15% (Fig. S12). Considering that both suitable habitats and quality are equally important for P. notoginseng, Pre11 emerges as the most influential environmental variable for its distribution.

Discussion

Dominant factors affecting the distribution and quality of P. notoginseng pairs

Among the environmental variables that affect the distribution of the suitability of P. notoginseng, Bio15 has a marginal influence, but it is a crucial variable influencing the level of saponin content of the four P. notoginseng saponins. The altitude range of 1400–1800 m is the optimum range for the cultivation of P. notoginseng. Beyond the suitability range, the increase in Ele will lead to an increase in solar radiation and a higher possibility of P. notoginseng black spot disease. Additionally, even though there was no correlation with any of the environmental variables in this study, altitude had a significant effect on the growth of P. notoginseng because environmental factors such as temperature, light, and rainfall were highly related to higher altitude [31]. Overall, the environmental variables that had a strong influence on the distribution of the suitability of P. notoginseng, in descending order, were Bio7 > Pre11 > Ele > Bio3 > Pre4 > Bio15, and those that had a strong impact on the quality of P. notoginseng were Bio15 > Pre4 > Pre11 > Bio7 > Bio3 > Ele. Previous studies have shown that seasonal precipitation is an important environmental variable influencing the distribution of P. notoginseng. From June to November is the hotter season in Yunnan, during which the precipitation requirement will be higher as this is the period when dry matter and nutrients accumulate more rapidly [32]. Therefor, Bio15 has a very strong effect on P. notoginseng content. Excessively high temperatures during this period can accelerate the evaporation of soil moisture, resulting in the failure of P. notoginseng seedling planting. Therefore, adequate water supplementation is necessary to ensure root development. When considered together, Pre11 is perhaps the most influential environmental variable on P. notoginseng, which affects both its distribution and quality. Existing research suggests that P. notoginseng prefers warmer environments and does not thrive when temperatures or precipitation (drought or flooding, etc.) are unsuitable [25]. During periods of heavy rainfall or high solar radiation, artificial intervention may be an option to control moisture and heat to assist P. notoginseng growth.

Transformation of suitable habitats for P. notoginseng under future climate change

Under the influence of diverse future climate scenarios, the suitable habitats of P. notoginseng will undergo alterations, among which the majority of the Highly suitable habitats in the regions of Wenshan, Kunming, and Qujing will be downgraded to moderately suitable areas and low suitable areas. It is notable that the degradation in the 2070 s and 2090 s under the SSP585 scenario is the most severe, indicating that with the deterioration of the ecological environment, the increase in caRb1on dioxide emissions, and the reduction of land resource occupancy, the Highly suitable habitats in these three regions are significantly affected. Wenshan, as a GI production area of P. notoginseng, has the highest saponin content. In the final results of this study, Highly suitable habitats are both Highly suitable habitats for P. notoginseng and areas with high saponin content, while the degradation to moderately suitable areas and low suitable areas represents a decrease in their saponin content as well. Among all the projections of suitable habitats for P. notoginseng under future climate change, the Yuxi region undergoes the least change, maintaining over 90% of Highly suitable habitats. Along with the growth in population size, the greater economic demands required for social development, and the environmental damage caused by human activities, the probability of occurrence of the two scenarios SSP370 and SSP585 is much higher than that of the SSP126 and SSP245 scenarios. Therefore, the three regions dominated by Wenshan must focus on protecting the ecological environment in their future development and develop neighboring regions, such as Yuxi and Honghe, as a means to alleviate the planting pressure on regions such as Wenshan and to safeguard the characteristics of local industries. In the future center of mass transfer, the distribution of P. notoginseng based on environmental variables mainly occurs in the central area connected by Chuxiong, Yuxi, and Puer, and when content change is taken into account, its distribution mainly occurs in the central area connected by Kunming, Honghe, and Wenshan. The results show that combining the species distribution model with the content change in the future is a favorable solution for the study of the suitable habitat of P. notoginseng. It is appropriate to conduct a study on the suitable habitat of P. notoginseng as it conforms to the layout of the Three-Year Action Plan for Understory Chinese Materia Medica Cultivation in Yunnan Province (2021–2023) issued by the government (http://lcj.yn.gov.cn/html/2022/fazhanguihua_0107/65067.html).

Limitations and future prospects

In this study, we used machine learning to predict the content distribution of P. notoginseng under future climate change, and then combined with the results of Biomod2 modeling to further evaluate its potential habitat, which has high quality planting areas while ensuring suitable growth, and this method can provide a new scientific basis and technical means for the cultivation of related medicinal plants in the future. However, since the determination of saponin content included only a few common growing regions, the final results are inevitably limited. In subsequent studies, quantitative data from neighboring regions should be continuously supplemented to make up for the shortcomings in the current study.

Conclusion

In this study, we investigated environmental changes and mitigation of P. notoginseng succession disorder through species distribution modeling, machine learning and chemical analysis, and for the first time, machine learning was used to predict the saponin content of P. notoginseng under different future climate scenarios, successfully predicting potential habitats of P. notoginseng that are both high fitness and high quality zones under future climate scenarios. Under the future climate change, the ecosystem of Wenshan region needs to be protected as a priority, and Yuxi and Honghe can be the candidate areas for the introduction of P. notoginseng cultivation in the future, which can effectively alleviate the planting pressure in Wenshan region. In previous studies, the highly suitable zones und the most influential environmental variables on P. notoginseng distribution and quality were Bio7 and Bio15, respectively, and the comprehensive assessment found that Pre11 is likely to be the key factor influencing P. notoginseng distribution and qualityer different future climate scenarios were not representative of high quality zones, which is effectively remedied by our results. They provide a scientific basis for the introduction of P. notoginseng in the future and are of great significance for the environmental protection of related areas. However, due to the lack of abundant collection sites and the instability of the data, the follow-up study needs to focus on collecting samples from various regions in Yunnan Province to further validate the results of the study.

Authors’ contributions

Peiyuan Li: Writing-original draft, Writing-review and editing, Conceptualization, and Formal analysis, Conceptualization, Writing-review and editing; Zhitian Zuo: Visualization, and Supervision Investigation and Resources; Yuanzhong Wang: Resources, Investigation, Supervision and Funding acquisition.

Funding

This work was financially supported by the Key Project of the Joint Special Project on Agricultural Basic Research in Yunnan Province (Grant ID: 202501BD070001-017).

Data availability

The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics and consent to participate declarations

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.