The Influence of Environmental Heterogeneity on Fertilization-Driven Patterns of Distribution and Yield in Medicinal Plants

Abstract

Medicinal plant production is essential for global health, yet how temperature, precipitation, and soil properties influence yield responses to fertilization remains poorly understood. Here, we conducted a meta-analysis of 668 observations from 79 studies, focusing on a wide range of plant species, to evaluate how nutrient inputs and environmental factors interact to shape medicinal plant productivity. We found that latitude, MAT, and MAP jointly determine global medicinal plant yield patterns. Yield increased with latitude and MAT but declined under prolonged fertilization and higher MAP. Optimal cultivation regions were identified between 15° and 35° absolute latitude, where temperature and precipitation conditions were most favorable. Compared with the arid environment of desertic climates, other regions, especially those with higher MAP in tropical areas, show a higher increase in yield. Our findings highlight that shifting precipitation-temperature regimes under climate change will affect fertilization outcomes on medicinal plant yield, emphasizing the need for spatiotemporally adaptive nutrient management strategies to ensure the sustainable yield of medicinal plants.

1. Introduction

With the continuous advancements in medicine, the focus of healthcare has gradually expanded beyond disease treatment to include prevention, post-treatment care, and rehabilitation [1]. Compared to synthetic pharmaceuticals, medicinal plants are widely valued for their naturally derived bioactive compounds and generally lower risk of adverse effects [2,3]. In response to rising global market demand, the cultivation area of medicinal plants is expanding worldwide—with the 2024 global herbal medicine market valued at USD 180.4 billion (6.3% CAGR 2025–2031) and China’s cultivation area growing 5.87% annually [4]. However, the challenges associated with continuous cropping are becoming more prominent, leading to yield declines, deterioration in medicinal quality, and disruption of soil ecosystems [5].

Agronomic practices and management strategies are crucial in sustaining the yield of medicinal plants [6,7]. Farmers often resort to intensive chemical fertilization and pesticide application to maximize yields [8]. While these inputs can enhance short-term productivity, they may also weaken the medicinal properties of plants and introduce risks such as pesticide residues and heavy metal accumulation [9]. In contrast, organic fertilizers have been shown to improve soil physicochemical properties, mitigate acidification, and enhance nutrient availability [10,11]. Similarly, the promoting effect of fertilization on plant growth is influenced by different environmental factors in different regions [12]. Temperature, precipitation, soil properties, and spatial factors such as latitude, which are referred to as spatiotemporal heterogeneity factors, often have an impact on the effectiveness of fertilization. Therefore, we need to find appropriate nutrient management strategies. Environmental factors such as temperature, humidity, and soil characteristics further influence soil microbial colonization and the bioavailability of active compounds [13,14], making fertilization outcomes highly variable. Therefore, understanding how spatiotemporal heterogeneity shapes fertilization efficiency is critical for optimizing medicinal plant cultivation.

Climate change is expected to disrupt natural and agricultural ecosystems substantially [15]. The production of medicinal plants is susceptible to climate fluctuations [16,17], as shifts in temperature and precipitation can alter species distribution, chemical composition, and overall medicinal quality. Currently, a significant proportion of medicinal plants in global trade originates from wild collection [18]. However, future climate scenarios predict range shifts or contractions for many species, impacting their availability and sustainable harvest [19]. These dynamics affect the effectiveness of fertilization strategies, as the nutrient requirements of medicinal plants must be matched with soil nutrient availability [20]. Regions with naturally high fertility may require less supplementation, whereas nutrient-deficient areas demand precise fertilization interventions [21]. A systematic understanding of spatiotemporal heterogeneity in soil nutrient dynamics is crucial for achieving these objectives.

In this study, through meta-analysis, we obtained the effects of fertilization on different types of medicinal plants in different regions. Then, we extracted the influences of annual precipitation, annual average temperature, spatial factors, environmental factors, and soil physical and chemical properties on the medicinal plants during the fertilization process. Furthermore, using these climatic factors and the Maxent model, we simulated the potential global regions suitable for the growth of medicinal plants. For this region, we used deep learning model to simulate the response of global medicinal plant growth to fertilizers based on the obtained data (Figure S1). MaxEnt models species—environment niches for spatial—agronomic insights, while deep learning captures complex, non—linear interactions in agronomic data—both filling gaps in standard meta—analysis tools. Through the above process, we aim to address the following key questions: (1) How does fertilization influence the yield of medicinal plants across global agroecosystems? (2) To what extent do spatial and climatic factors (e.g., latitude, temperature, precipitation) regulate the effectiveness of fertilization? (3) Can optimal cultivation zones for medicinal plants be identified based on climatic suitability and fertilization responsiveness?

2. Materials and Methods

2.1. Data Collection

Peer-reviewed journal articles published between 1 January 1990 and 1 June 2024 were collected using the Institute for Scientific Information Web of Science (http://apps.webofknowledge.com (accessed on 29 August 2024)). The keywords in the search were “medicine” or “medicinal plant” AND “fertilizer” or “fertilization”. To assess the effect of fertilization on medicinal plants’ yield, publications had to meet the following criteria: (1) they should be controlled (i.e., fertilized vs. non-fertilized treatments); (2) the article should be written in English; (3) the study should have been conducted under open-field conditions; (4) yield or yield-related indices for medicinal plants are reported; and (5) the study duration of the experiment is longer than one year/growing season. Based on these criteria, 79 articles (Table S1) were obtained.

We extracted the primary information on the field of medicinal plants for each study. To extract the results presented in the figures, we used the digitizing software GetData Graph Digitizer 2.25. Measurements from different ecosystem types, climatic types, and nutrient addition levels within a single study were treated as independent observations. We used the hierarchical effect model based on random effects and acknowledged that this method has certain limitations. Subsequently, we conducted more detailed subgrouping to ensure the stability of the results. When a study included multiple sampling dates, we only included data from the peak growing season in agroecosystems’ mature medicinal plants stage. Furthermore, the changed factors of spatiotemporal heterogeneity were used to analyze the response of soil productivity to fertilization, including climatic factors (mean annual temperature (MAT) and mean annual precipitation (MAP)), management duration (plant and fertilization duration (Time), experiment start year (Year) and the types of fertilizers used (biological organic fertilizer (BF), chemical fertilizer (CF), organic fertilizer (OF))). Further, soil properties in experiments (pH) were also collected and recorded from articles and supplement materials. The classification of different factors is as follows. The division of absolute latitude includes 0–30°as the low-latitude region; 30–60°as the mid-latitude region. The temperature zone division: 0–10 °C as the low-temperature zone; 10–20 °C as the medium-temperature zone; 20–30 °C as the high-temperature zone. The precipitation zone division: 0–200 mm as the arid zone; 200–400 mm as the semi-arid zone; 200–400 mm as the semi-humid zone; 400–800 mm as the humid zone. The fertilization cycle division: t-1: 0–12 months; t-2: 11–24 months; t-3: 24–36 months; t-4: 36–48 months. All relevant data have been uploaded to the public database Figshare to ensure accessibility. The data can be accessed via the following link: https://figshare.com/s/a4b37e6b7d818c0648b0 (accessed on 10 July 2025).

2.2. Calculation of the Individual Response Ratio

We used Stata SE 16.0 for the meta-analysis and used the natural log of the ratio as a metric to compare the effect sizes of different response variables between fertilized and non-fertilized treatments [22]. The lnRR (response ratio) of plant produce index was calculated as the natural log of the ratio between the mean of fertilized (X_t) and non-fertilized (X_c) treatment groups as follows:

$$ln{RR}_{\left(i\right)}=\ln \left({\displaystyle \frac{X_t}{X_c}}\right)=\ln X_t-\ln X_c$$

(1)

$$lnRR={\displaystyle \frac{{\sum }_{i=1}^n{lnRR}_{\left(i\right)}}{n}}$$

(2)

The variance of the effect size (v) of X was computed as follows:

$$v_{\left(i\right)}={\displaystyle \frac{{SD}_t^2}{n_t{X}_t^2}}+{\displaystyle \frac{{SD}_c^2}{{n_{c}X}_c^2}}$$

(3)

$$v={\displaystyle \frac{{\sum }_{i=1}^n{v}_{\left(i\right)}}{n}}$$

(4)

where i means the number index of plant productivity in each article. SD_c and SD_t are the standard deviations of the non-fertilized control and fertilized treatment groups, and n_c and n_t are the sample sizes of the control and treatment groups, respectively. Most included studies reported standard errors (SEs), which were transformed into standard deviations (SDs) according to the following equation:

$$SD=SE\sqrt{n}$$

where n is the sample size; in several studies in which neither SD nor SE was given, we assigned an SD using the package ‘metagear’ in R software(version 4.4.1). In simple terms, it assumes that in a certain type of study, the standard deviation of the effect size is roughly in a certain proportional relationship with the mean. Thus, the missing standard deviation can be estimated using the known relationship between the mean and standard deviation of the group.

2.3. Calculation of the Overall Response Ratios

We chose the multilevel-effects model to calculate the weighted response ratio (RR₊₊) and 95% confidence interval (CI) using the rma. mv function in the ‘metafor’ package [23]. If the 95% CI values did not overlap zero, the effect of fertilizer addition was considered significant (p < 0.05). To test the responses that differed among different groups, we calculated the between-group variability (Q_M) of the variables affecting the effect of fertilization management on the medicinal yield. When the Q_M values were significant (p < 0.05), the responses among treatments were considered meaningful. Publication bias associated with the overall response ratios was tested using the Rosenthal fail-safe number. If the fail-safe was higher than the threshold value (N > 5k + 10; k was the number of case studies), we assumed no publication bias (Table S2).

2.4. MaxEnt Model on Global Potential Location Distribution of Medicine Material

Weather data were downloaded from WorldClim 2.1 (http://www.worldclim.org/ (accessed on 11 November 2024)), and all of them included 19 climatic variables (Bio1–Bio19, Table S3), with a spatial resolution of 10′. Species distribution modeling was performed using MaxEnt software (version 3.4.1). Model settings included a 25% test set and a 75% training set. The number of replicates was set to 10 using the subsample replicated run type. The maximum number of iterations was set to 5000, with a 10% training threshold applied. The importance of climate variables was assessed using the Jackknife method [24].

The ArcGIS (10.8.1) software was used to process the Maxent simulation data, and the average value of 10 simulations was imported into it and converted into Raster data. The raster value represented the survival probability of medicinal plants, and the probability p was reclassified. The predicted distribution division of medicinal plants was divided into five categories, namely, unsuitable areas, low suitability areas, general suitable areas, medium suitability areas, and higher suitability areas. ArcGIS calculated the number of pixels in different suitable areas, and the proportion of pixels in the total pixels was calculated. The area of each suitable area was calculated according to the global land area. The contribution rate of environmental factors and the Jackknife test were used to evaluate the importance of each environmental factor. The receiver operating characteristic (ROC) curve assesses the accuracy of model prediction results (Figure S2). The area under the ROC curve is called the area under curve (AUC), ranging from 0.5 to 1.0. A higher variable importance value suggests a stronger contribution of the corresponding environmental factor to the prediction model, thereby improving the model’s predictive accuracy [25].

2.5. XGBoost and Deep Learning Models on Global Response of Medicine Yield to Fertilization Prediction

Based on the growth data of medicinal plants in different geographical regions, combined with climate and soil properties, this study constructed an XGBoost model to evaluate the relative importance of various environmental factors on medicinal plants’ yield [26]. The research data included the following variables: (a) geographic information: longitude, latitude; (b) climate variables (4 in total): MAP, MAT, temperature annual range, and mixed temperature—humidity weighted metric; (c) soil properties (6 in total): total organic carbon, pH, available water capacity, cation exchange capacity, electrical conductivity of the extract and soil bulk density; (d) Growth indicators of medicine material: the response ratio of yield and yield were used as a measure of the growth status of medicinal plants. The model was run in R (4.3.3) using package ‘xgboost’, using XGBoost’s own feature importance analysis method to calculate the importance of each environmental variable based on the gain value. All variables were sorted from high to low by gain value, and feature importance bar charts were drawn to identify the factors that have the most significant impact on the growth of medicinal plants. In addition, we used deep learning to build a global prediction model based on the top 5 factors in terms of importance. In briefly, in ArcGIS, the longitude and latitude data of the test points were collected and matched with MODIS soil physicochemical data and Worldclim climate data (MAT; MAP; Cation—Exchange Capacity (CEC); Mixed Temperature—Humidity Weighted Metric (MTWM); Available Water Capacity (AWC); Temperature annual range (TAR); Bulk density; Total Organic Carbon (TOC); Electrical Conductivity of the Extract (ECE)), and the yield response ratio and the fitting curve of relevant data were simulated in MATLAB 2018a, establishing a mathematical relationship between the yield response ratio and the aforementioned soil and climate variables (Figure S3). Then the response ratio of medicine at the predicted points was deduced by using the fitted curve [27].

2.6. Statistical Analysis

The distribution of the collected experimental data and the response ratio of medicine yield to fertilization were visualized using the R packages ‘ggmap’ to illustrate the global spatial distribution of the materials in R (4.3.3). The relationship between response ratio of medicine yield to fertilization and environmental factors was calculated based on Spearman correlation according to R packages ‘ggpubr’ and ‘dplyr’. Environmental factors include latitude, MAT, MAP, soil pH, Time, and Year. A p-value < 0.05 was considered statistically significant.

3. Results

3.1. Distribution of Collected Data

The global distribution map of medicinal plant experiments highlights regional research activity. It shows that Southeast Asia, Europe, and the Middle East are the primary centers of medicinal plant research. Conversely, relatively little research has focused on medicinal plant production in the Southern Hemisphere, and no studies have been conducted in North America or Oceania (Figure 1).

3.2. Effect of Spatiotemporal Heterogeneity on Medicine Yield Under Fertilization Management

On a global scale, compared to no fertilization, fertilizer application significantly increased medicinal plant yield (+28.73%). However, subgroup analyses based on climatic factors (MAT and MAP), temporal factors (fertilization duration), soil types, and fertilizer types revealed varying responses to fertilization. In terms of fertilization duration, short-term fertilization significantly increased the yield response (t-1: +28.53%; t-2: +26.95%), while long-term fertilization had no significant effect (Figure 2). In terms of climatic factors, fertilization in arid environments had more pronounced positive effects than in moist environments. Fertilization in arid and semi-arid areas positively affected medicinal plant yield (Arid: +36.15%; Semi-arid: +43.43%, respectively). The improvement effect of fertilization in low latitudes on medicinal plant yield was higher than that in high latitudes. Moreover, in terms of soil types, the improvement effect in clay soil is higher than other types of soil.

3.3. Correlation Between Medicine Yield and Environmental Factors

Spearman correlation analysis showed that the latitude positively affected the yield of medicinal plants, but it was not significant (R = 0.047, p = 0.34; Figure 3a). Climate factors, such as MAT, improved the response of yield of medicinal plants to fertilization (R = 0.081, p = 0.041; Figure 3b), and MAP had a negative relationship with the response of yield of medicinal plants to fertilization (R = −0.15, p < 0.001; Figure 3c). Soil pH did not affect the response of yield of medicinal plants to fertilization. In terms of time factors, fertilization duration and starting year both caused a decline in response of yield of medicinal plants to fertilization.

3.4. Potentially Suitable Medicinal Plant Habitats

For the combined influence of climate, topography, and soil factors, the total area of suitable medicinal plants habitat was found to be 4972.58 million ha, accounting for around 33.37% of the total terrestrial area. The spatial distribution of medicinal plants’ habitat suitability shows a gradually decreasing trend from the tropic of Cancer and Capricorn to the poles (Figure 4) and has significant aggregation characteristics. Highly suitable habitats are mainly distributed in humid coastal areas, including Central and South China, the Korean Peninsula, Central Asia, and the Mediterranean, which account for 2.92% of the total land area. Moderately suitable habitats account for 4.21% of the total land area and are mainly distributed in Central North America and South Africa. Poorly suitable habitats account for 66.63% of the total land area and are primarily distributed in high-latitude areas, including Nordic and Mongolian regions. The unsuitable habitats occupy the most significant area (66.63% of the total land area) and are mainly located in northwestern Asia, the northern region of North America, and the equatorial region. The average test AUC of this repeat run is 0.91, and the standard deviation is 0.04, showing high confidence.

3.5. Global Prediction of Medicine Yield to Fertilization Management According to Environmental Heterogeneity

SHapley Additive exPlanations (SHAP) values were used to assess the effects of multiple features on the model output (Figure 5a). These values help unravel the degree and direction of influence that different features exert, with the color gradient further denoting how the magnitude of feature values relates to their impact, thus providing a comprehensive and interpretable view of feature—model output interactions. High-impact features, such as MAP and CEC, were identified as key contributors to model decision-making, whereas low-impact features played relatively minor roles. MAP and CEC exhibited higher SHAP values compared to other variables (SHAP > 0.075), indicating their greater importance in influencing the model outcomes (Figure S4).

The global distribution of predicted values for the natural logarithm of the response ratio of medicinal plant yield, with color gradients representing varying predicted values across different latitude and longitude regions, is visualized as this result. (Figure 5b). Based on the deep learning model predictions, fertilization management for medicinal plant production near the Tropic of Cancer and the Tropic of Capricorn showed the most beneficial results. Interestingly, regions in North America demonstrated strong positive effects on medicinal plants to fertilization (predicted lnRR > 0.5). In contrast, the response ratio in North Africa indicated predominantly negative effects of fertilization on the yield of medicinal plants.

4. Discussion

Here, we comprehensively analyzed the role of spatial and time factors in controlling medicinal plants’ yield globally. For the fertilization response of medicinal plants’ yield, a tenuous positive trend was observed across absolute latitudes, though this pattern lacked statistical significance. The MAT has a significant influence on the yield. However, MAP and time factors (fertilization duration and start year) were negatively correlated with yield. XGBoost model indicated that climatic factors (e.g., MAT) were more critical than time factors (e.g., fertilizer duration) and soil properties (e.g., soil pH) in controlling medicinal plants’ yield. Moreover, the MaxEnt module indicated that the habitats of suitable medicinal plants and efficient fertilization areas overlap highly, focusing on areas near the Tropic of Cancer and the Capricorn. Currently, our research mainly focuses on the northern latitudes, and there is relatively little data analysis for other regions. This might affect our judgment. In the future, appropriately expanding the collection scope will be more appreciated in future studies.

Our results showed that fertilizer addition improved medicinal plant yield by 28.73%. Notably, we observed a weak positive trend, where the apparent benefit of fertilization to medicinal plant yield showed a tendency to increase with absolute latitude. Latitude is associated with variations in climate, day length, and temperature, which can potentially influence plant physiological responses and growth patterns. For instance, the MAT and latitude can affect the nitrogen-to-phosphorus ratio of plant leaves [27]. However, the current data do not provide sufficient evidence for a significant correlation. Generally, as latitude increases (moving closer to the poles), temperatures tend to decrease, and seasonal variations in day length become more pronounced [28]. Medicinal plants in higher latitudes often experience cooler temperatures and shorter growing seasons, which can limit their natural growth potential [29]. Under such conditions, adequate fertilization provides essential nutrients that can promote more efficient growth and yield accumulation during the limited growing period [30]. Moreover, lower temperatures in these regions may reduce the decomposition rate of organic matter, potentially leading to lower natural soil fertility [31]. Fertilizer supplementation can thus compensate for nutrient deficiencies, ensuring improved plant productivity [32].

Precipitation and temperature are crucial climatic factors that significantly impact the yield of medicinal plants [33]. Both factors directly affect plant growth, development, and the production of bioactive compounds, which are essential for the medicinal value of these plants [29]. Our findings indicate a weak positive correlation between MAT and medicinal plant yield, suggesting that MAT may exert a subtle positive influence on it. Excessive rainfall can lead to waterlogged soils, which reduce oxygen availability to plant roots, inhibit nutrient uptake, and promote root rot, ultimately limiting plant growth [34]. Even with fertilization, water saturation in the soil can impede nutrient absorption, reducing the overall effectiveness of fertilization efforts [35,36]. Applying additional irrigation to replenish water in areas with low precipitation will be more beneficial for the fertilization effect.

Biodiversity and habitat suitability typically exhibit a latitudinal gradient. Subtropical regions near the Tropic of Cancer experience relatively stable temperatures, abundant sunlight, and sufficient precipitation, creating mild ecological conditions and high productivity that are conducive to plant growth [37]. However, as one moves toward the poles, temperatures decrease, seasonality intensifies, and the growing season shortens, with extreme climatic conditions restricting the survival of most plants. Our research confirms this latitudinal gradient, showing a gradual decline in biomass from the Tropic of Cancer and the Tropic of Capricorn toward the poles. We found that coastal areas of southern China have a warm, humid climate, favorable for the growth of a wide variety of medicinal plants. In contrast, the Mongolian Plateau and northern Europe experience cold and arid conditions, with medicinal plants distributed sparsely. Extreme climate and unique soil conditions in these high-latitude regions further limit the growth and survival of medicinal plants [38,39]. Winters are extremely cold, leading to a short growing season, and soils are often nutrient-poor or highly acidic. Additionally, light and temperature conditions restrict photosynthesis and energy accumulation, preventing many medicinal plants from completing their life cycles. For example, medicinal plants on the Mongolian steppes are mostly cold- and drought-tolerant, while northern Europe, with its extensive permafrost coverage, supports only a limited variety of species. Furthermore, equatorial regions present challenges such as continuous high temperatures and humidity, poor soil drainage, and high pest and disease pressures, which many medicinal plants cannot tolerate [17,40].

The study indicated that the fertilization management effect of medicinal plant production was most significant near the Tropic of Cancer and the Tropic of Capricorn. The unique climatic conditions of this region, such as sufficient light and suitable temperatures, interact with fertilization practices to create a favorable environment for the growth of medicinal plants, allowing fertilization to maximize its catalytic role. The response ratio of medicinal plants to fertilization in North Africa was generally low. This outcome may be related to the region’s arid climate, low soil fertility, and limited water availability, which collectively constrain the effectiveness of fertilization. Under such environmental conditions, nutrient uptake by plants may be limited, thereby reducing the potential benefits of added fertilizers. Furthermore, in water-scarce environments, there is a greater risk of soil salinization associated with fertilizer application, which could further inhibit plant growth. Based on secondary data, the model-predicted results we obtained provide a preliminary reference for global medicinal plant fertilization management and may assist in guiding regions to explore more scientific fertilization strategies aimed at improving medicinal plant yield and quality.

5. Conclusions

In this study, we observed that spatiotemporal heterogeneity is an essential factor affecting the production of medicinal plants to fertilization variability, in which latitude, MAT, and MAP have a more significant impact on the response of medicinal plants’ yield to fertilization. The maxent model was established to find suitable growing areas of medicinal plants globally, and highly suitable habitats near the Tropic of Capricorn and the Tropic of Cancer were found. In addition, the response ratio of medicinal plants under fertilization management in this suitable area was predicted, and it was found that it also had a good positive effect near the Tropic of Cancer. Our research can help researchers and growers formulate precise fertilization schemes according to the spatial and temporal conditions, such as climate and soil in different regions at different times, improve fertilizer utilization, yield, and quality, and promote industrial development.