Spatiotemporal Patterns and Driving Mechanisms of Heavy Metal Accumulation in China’s Farmland Soils Based on Meta-Analysis and Machine Learning

Abstract

To elucidate the long-term spatiotemporal patterns and key drivers factors, this study employed a meta-analysis of data from soil containing Potentially Toxic Elements (Cd, As, Cr, Hg, and Pb) in Chinese farmland soils from 2003 to 2025. The geoaccumulation index, the potential ecological risk index, and standard deviation ellipses were used to assess the spatiotemporal evolution of heavy metal accumulation and ecological risk, while the Random forest–SHapley Additive exPlanations (RF-SHAP) method was employed to identify driving mechanisms. At the national scale, Cd and Hg are significantly enriched relative to the background values, whereas As, Cr, and Pb remained at relatively low levels, with enrichment ranked as Cd > Hg > Pb > Cr > As. Cd and Hg indicated mild pollution, but the Sichuan Basin emerged as a hotspot, where Cd reached moderate pollution and showed strong ecological risk, and Hg also exhibited high ecological risk. Over the past two decades, the contamination center shifted from coastal to southwestern inland regions, with an expanded and more dispersed distribution. Since 2017, Cd and Hg pollution levels have stabilized, suggesting that the aggravating trend has been preliminarily curbed. Industrial waste and wastewater discharge, irrigation and fertilization were identified as the primary anthropogenic factors of soil heavy metal accumulation, while climatic factors (temperature, precipitation, and solar radiation) and soil physicochemical properties (pH, clay content, and organic matter) played fundamental roles in spatial distribution and accumulation. Our findings call for targeted predictive research and policies to manage heavy metal risks and preserve farmland sustainability in a changing climate.

1. Introduction

With the acceleration of industrialization, urbanization, and agricultural modernization in China, heavy metal contamination of agricultural soils has become a prominent issue, resulting in widespread environmental and health concerns [1,2]. Globally, approximately 14% to 17% of the arable land is affected by heavy metal pollution [3]. In China, contamination levels in agriculture soils frequently exceed the national standard, with As, Cd, Cr, Hg, and Pb identified as the major pollutants [4]. Elevated concentrations of these elements not only disrupt soil physicochemical properties and reduce agricultural productivity but also bioaccumulate through crop uptake and transfer along the food chain, ultimately posing serious human health risks such as carcinogenesis and teratogenesis [5,6,7]. Therefore, a comprehensive understanding of the spatiotemporal trends and driving factors of soil environmental quality is crucial for scientifically assessing pollution risks, safeguarding food security and public health, and formulating effective remediation and management strategies.

Owing to the challenges associated with large-scale sampling and data acquisition, most studies have focused on megacities, heavily polluted regions, or river basins [8,9,10], revealing the characteristics and driving factors of heavy metal pollution in localized areas. Although some scholars have investigated the spatial distribution and driving factors of soil pollution at the national scale, Cd and Hg have been identified as the metals posing the most significant risks to farmland soils of China, with pollution being most severe in southern regions [2,11,12,13,14,15]. However, systematic investigations of the long-term spatiotemporal evolution remain lacking. Extensive research has indicated that soil heavy metal accumulation is closely linked to influencing factors such as the industry, agriculture, and transportation [1,13]. The spatial heterogeneity of contamination is associated with regional pollutant emission patterns and industrial characteristics [2]. For example, the number of industrial enterprises and industrial wastewater discharge significantly affect Hg levels [16], while the metal processing and smelting industries are the primary sources of Cd pollution [17]. However, prior research primarily focused on the effects of socioeconomic factors—such as industrial emissions and agricultural inputs—on heavy metal inputs. In reality, the environmental behavior of heavy metals is more dependent on the synergistic effects of soil physicochemical properties (e.g., pH and organic matter) and climatic factors (e.g., precipitation patterns), which collectively regulate their migration and transformation processes [18]. Although previous studies have assessed the hazards of potentially toxic elements in Chinese soils from various perspectives, comprehensive analyses of their long-term spatiotemporal dynamics at the national scale and the relative contribution of natural and anthropogenic factors remain insufficient.

Scholars have employed various analytical methods to investigate the factors affecting soil heavy metal pollution. Multivariate statistical methods such as cluster analysis, principal component analysis, and correlation analysis have been applied extensively in this field [19,20,21]. However, because soil heavy metal pollution systems typically involve complex nonlinear relationships, traditional methods often inadequate to capturing these dynamics. They are often limited to the qualitative identification of influencing factors without accurately quantifying the specific contributions of each driver to the spatial variation in pollution. In contrast, the Random Forest (RF) model. owing to its strong capacity to capture nonlinear relationships, has significant advantages in identifying the driving factors of heavy metal pollution [22].Moreover, traditional receptor models often rely on subjective experience and prior knowledge, rendering their outcomes susceptible to human intervention. In contrast, random forests employ a fully data-driven analytical strategy, with all processes from feature selection to model construction autonomously performed by the algorithm. This significantly enhances the reliability and reproducibility of source apportionment results [23]. RF demonstrate superior adaptability and stability when processing real-world environmental data. Traditional methods typically impose stringent requirements on data distribution characteristics (such as normality and homogeneity of variance), and their performance is constrained by limited sample sizes or suboptimal data structures. In contrast, random forests exhibit insensitivity to data distribution patterns and maintain robust performance under moderate sample conditions, rendering them particularly suitable for medium-to-low scale, high-dimensional, and complex soil contamination datasets [24]. In terms of Regarding predictive capability, traditional regression models are prone to inaccuracy when variable relationships are complex or multicollinearity exists. Random forests, however, effectively suppress overfitting by integrating multiple decision trees and employing voting or averaging predictions. This enhances the model’s generalization ability and predictive accuracy under diverse environmental conditions, providing a more reliable analytical tool for assessing heavy metal pollution under multifactorial interactions [25].

Therefore, this study conducted a meta-analysis of soil heavy metal data from 2003 to 2025, with the following primary objectives: (1) to assess the average concentration levels of five heavy metals (As, Cd, Cr, Hg, and Pb) nationwide and across major geographic regions, as well as their deviations from soil background values; (2) to analyze the spatiotemporal trends of cumulative pollution for these five heavy metals in national farmland soils over the past 23 years, assess the pollution severity across nine major agricultural regions, and reveal the spatiotemporal evolution of pollution hotspots; and (3) to establish nonlinear relationships between the spatial distribution of heavy metals and influencing factors to identify key drivers, providing scientific support for the development of soil heavy metal pollution risk prevention and control policies in China.

2. Materials and Methods

2.1. Data Collection and Analysis

Data of As, Cd, Cr, Hg, and Pb concentrations in farmland soils of mainland China (Hong Kong, Macau, and Taiwan were excluded due to limited data) from 2003 to 2025 were obtained from the Web of Science (https://www.webofscience.com, accessed on 10 March 2025) and China National Knowledge Infrastructure (CNKI) (https://www.cnki.net, accessed on 12 March 2025) databases. The literature search employed the following query: TS = (lead OR chromium OR arsenic OR cadmium OR mercury OR Cr OR Pb OR Cd OR As OR Hg OR toxic metal OR heavy metal OR toxic element) AND (agriculture soil OR arable land OR cropland OR paddy) AND (China).

To maximize analytical accuracy, the literature screening was conducted according to the following principles: (1) case data must originate from field sampling; (2) the geographic sampling area (province-city-county/district), sample size, and acceptable detection methods should be clearly specified. Sample pretreatment must employ a mixed acid digestion system (e.g., HNO₃–HClO₄–HF), with determination performed using methods such as ICP-MS, ICP-OES, or CV-AAS; (3) the sample size must be ≥100; and (4) statistical parameters, including the mean, extreme values, standard deviation, and coefficient of variation, must be fully reported. To ensure data reliability, heavy metal concentration data were preprocessed to identify and exclude outliers prior to analysis. In this study, we chose Studentized deleted residual to outlier diagnostic index [11], resulting in the elimination of 2 As, 6 Cd, 6 Cr, 2 Hg, and 4 Pb cases. The final dataset comprised 221 valid studies. This process ultimately yielded a comprehensive geographic database encompassing 31 provincial-level administrative regions and 241 district/county-level cases, including more than 250,000 soil samples (Figure 1).

2.2. Meta-Analysis

Meta-analysis is a statistical method that involves systematically combining the results of multiple studies with the same objective via statistical techniques and evaluating effectiveness through a quantitative metric. It is suitable for research on large-scale issues [26] and has been widely applied and validated in the field of environmental science [27,28,29]. The calculation formula is presented as follows:

$$E_s=\ln R=\ln \left({\displaystyle \frac{w_i}{S_i}}\right)=\ln \left(w_i\right)-\ln \left(S_i\right)$$

(1)

where $E_s$ represents the effect values of soil heavy metal data collected from the literature; $w_i$ denotes the measured values of soil heavy metals (mg·kg⁻¹) collected from the literature; $S_i$ indicates the evaluation standards (mg·kg⁻¹) for corresponding elements, which are based on regional background values of soil heavy metals. The background values of the soil heavy metals in the study area are listed in

Table A1. Background values of heavy metals in soils of various provinces in China [72].

Province	As	Cd	Cr	Hg	Pb
Beijing	9.70	0.074	68.10	0.069	25.40
Tianjin	9.60	0.090	84.20	0.084	21.00
Hebei	13.60	0.094	68.30	0.036	21.50
Shanxi	9.80	0.128	61.80	0.027	15.80
Liaoning	8.80	0.108	57.90	0.037	21.40
Jilin	8.00	0.099	46.70	0.037	28.80
Heilongjiang	7.30	0.086	58.60	0.037	24.20
Shanghai	9.19	0.138	70.20	0.095	25.00
Jiangsu	11.00	0.126	77.80	0.289	26.20
Zhejiang	9.20	0.070	52.90	0.086	23.70
Anhui	9.99	0.097	66.50	0.033	26.60
Fujian	5.78	0.054	41.30	0.081	34.90
Jiangxi	14.90	0.108	45.90	0.084	32.30
Shandong	9.30	0.084	66.00	0.019	25.80
Henan	11.40	0.074	63.80	0.034	19.60
Hubei	12.30	0.172	86.00	0.080	26.70
Hunan	15.70	0.126	71.40	0.116	29.70
Guangdong	8.99	0.056	50.50	0.078	36.00
Guangxi	20.50	0.267	82.10	0.152	24.00
Sichuan	10.40	0.079	79.00	0.061	30.90
Guizhou	20.00	0.659	95.90	0.110	35.20
Yunnan	18.40	0.218	65.20	0.058	40.60
Tibet	16.20	0.074	68.00	0.021	27.60
Shaanxi	11.10	0.094	62.50	0.030	21.40
Gansu	12.60	0.116	70.20	0.020	18.80
Ningxia	11.90	0.112	60.00	0.021	20.90
Xinjiang	11.20	0.120	49.30	0.017	19.40
Chongqing	5.82	0.133	76.10	0.053	25.50
Inner Mongolia	6.12	0.050	39.80	0.030	16.80
Hainan	1.14	0.050	15.20	0.030	22.30

.

2.3. Geoaccumulation Index

The geoaccumulation index ($l_{geo}$) reflects not only the effect of natural geological processes on background values but also the contribution of anthropogenic activities to heavy metal accumulation in the soil, making it suitable for assessing heavy metal pollution levels in farmland soils [30,31]. In this study, temporal trends in soil heavy metal pollution from 2003 to 2025 were evaluated. Based on the eight major economic zones and nine agricultural regions in China, the national territory was subdivided into nine subregions (Figure A1), and the spatial variation in soil heavy metal pollution were identified. The corresponding formula is as follows (2):

$$I_{\mathit{geo}}={\mathit{log}}_2\left({\displaystyle \frac{w_i}{1.5S_i}}\right)$$

(2)

where $W_i$ represents the measured value of heavy metal i in the sample (mg·kg⁻¹); $S_i$ denotes the evaluation standard for heavy metal i (mg·kg⁻¹), which is based on regional soil heavy metal background values; see Table A1 for background values in the study area; $l_{geo}$ indicates the geoaccumulation index. Soil contamination levels can be classified into five grades on the basis of the geoaccumulation index, as shown in

Table A2. Classification of the geoaccumulation index.

Pollution Classification Standard of the Geoaccumulation Index
	$3<I_{\mathrm{g}\mathrm{e}\mathrm{o}}$	$2<I_{\mathrm{g}\mathrm{e}\mathrm{o}}\le 3$	$1<I_{\mathrm{g}\mathrm{e}\mathrm{o}}\le 2$	$0<I_{\mathrm{g}\mathrm{e}\mathrm{o}}\le 1$	$I_{\mathrm{g}\mathrm{e}\mathrm{o}}\le 0$
Rang	4	3	2	1	0
Pollution level	Strongly polluted	Moderately to heavily contaminated	Moderately polluted	Unpolluted to moderately polluted	Unpolluted

.

2.4. Ecological Risk Assessment

The potential ecological risk index method integrates element-specific characteristics with environmental background parameters to achieve both a single ecological risk index for individual heavy metals and a comprehensive assessment of the combined effects of multiple metals [32,33]. This study employs this method to systematically evaluate the ecological risk levels of heavy metal contamination in agricultural soils. The calculation formula is presented as follows:

$$I_{r,i}=\sum _{i=1}^n\left(T_i\times {\displaystyle \frac{W_i}{S_i}}\right)$$

(3)

where $T_i$ represents the biological toxicity response factor for different metals. The toxicity coefficients for each heavy metal element are ranked as follows: Hg (40) > Cd (30) > As (10) > Pb (5) > Cr (2), $W_i$ represents the pollution index for pollutant i and $I_{r,i}$ denotes the potential ecological risk factor for a single metal. Here, $I_{r,i}$ ≤ 40 represents slight potential ecological risk; 40 < $I_{r,i}$ ≤ 80 denotes moderate potential ecological risk; is 80 < $I_{r,i}$ ≤ 160 indicates strong potential ecological risk; 160 < $I_{r,i}$ ≤ 320 represents very strong potential ecological risk; and is $I_{r,i}$ > 320 denotes extremely strong potential ecological risk.

2.5. Standard Deviational Ellipse

The standard deviation ellipse (SDE) approach is an effective method for quantifying spatial distribution characteristics [34,35]. Its main parameters include the major axis, minor axis, flattening degree, and centroid of the ellipse. The major axis indicates the overall spatial direction of the study object, whereas the minor axis reflects its spatial distribution extent. The eccentricity, defined as the ratio of the difference between the major and minor axes to the major axis, represents the degree of dispersion of the spatial distribution, and the center of gravity of the ellipse denotes the central location of the spatial distribution [36]. To identify the spatiotemporal patterns of heavy metals in agricultural soils, the equidistant time interval method was applied, in which the study period was divided into three phases on the basis of important policy implementation dates: Phase I (2003–2009), Phase II (2010–2017), and Phase III (2018–2025). The SDE for each phase was calculated using the average of each case as the weight. The formulas for the above parameters are as follows:

$$C_x=\sqrt{{\displaystyle \frac{\sum _{i=1}^n{\left(x_i-\overline{x}\right)}^2}{n}}}$$

(4)

$$C_y=\sqrt{{\displaystyle \frac{\sum _{i=1}^n{\left(y_i-\overline{y}\right)}^2}{n}}}$$

(5)

In Equations (3) and (4), $C_x$ and $C_y$ denote the center coordinates of the ellipse; $x_i$ and $y_i$ denote the coordinates of the object under study; and $\overline{x}$ and $\overline{y}$ denote the mean values of the coordinates of the object.

2.6. Machine Learning Model Development

The random forest (RF) model is an ensemble classification algorithm based on the construction of multiple decision trees to reduce overfitting. This approach significantly increases prediction accuracy and effectively addresses issues such as missing values and multicollinearity [37,38]. In this study, the RF was employed to investigate the mechanisms by which different factors affect soil heavy metal pollution. The independent variables incorporates into the model included socioeconomic factors (average per capita GDP, average output value of primary industry, and average output value of secondary industry), transportation factors (motor vehicle ownership and average highway mileage), soil physicochemical properties (total phosphorus, clay content, pH, organic matter, and total nitrogen), soil heavy metal input/output factors (irrigation water and fertilizer application rate), meteorological factors (average temperature and average precipitation), and environmental stress and pollutant emission factors (total volume of industrial wastewater discharge, total volume of noncompliant industrial wastewater discharge, and number of enterprises involving heavy metals) (

Table 1. Dataset and sources of the factors affecting the heavy metal distribution in farmland soils.

Variable Type	Variable	Abbreviation	Data Source
Socioeconomic factors	Average per capita GDP	APCGDP	Research Area Statistical Yearbook
	Average output value of primary industry	AOVPI
	Average output value of secondary industry	AOVSI
Transportation factors	Motor vehicle ownership	MVO
	Average highway mileage	AHM
Factors affecting heavy metal inputs and outputs in the soil	Fertilizer	Fertilizer
	Irrigation	Irrigation
Environmental stress and pollutant emission factors	Average comprehensive utilization rate of industrial solid waste	ACURISW	China Urban Statistical Yearbook, Regional Statistical Yearbook
	Total emissions of industrial particulate matter	TEIPM
	Total volume of industrial wastewater discharge	TVIWD
	Total volume of noncompliant industrial wastewater discharge	TVNCIWD
	Number of enterprises involving heavy metals	NEIHM
Climate factors	Average humidity	AH	China Meteorological Science Data Center
	Average wind speed	AWS
	Average precipitation	APR
	Average temperature	AT
Soil physicochemical properties	Clay	Clay	Chinese Academy of Sciences Science Data Center
	pH	pH
	Bulk density	BD
	Available phosphorus	AP
	Cation exchange capacity	CEC
	Total phosphorus	TP
	Total nitrogen	TN
	Soil organic matter	SOM

). Heavy metal concentrations were treated as the response variable. The dataset was partitioned using 10-fold cross-validation, with 90% of the data employed as the training set for model construction, and the remaining 10% served as the test set. The SHapley additive explanations (SHAP) approach can be used to decompose model predictions into contributions from each input feature [39]. Each prediction is expressed as the sum of the SHAP values corresponding to all input feature, as computed with Equation (6).

$$f\left(x\right)=f_0+\sum _{i=1}^N{f}_i$$

(6)

where $f\left(x\right)$ represents the model-predicted value, $f_0$ denotes the mean of all training sample predictions, $N$ is the number of input features, and $f_i$ represents the attribute value corresponding to each feature. The SHAP values for features $x_i$ in the model are derived from Equation (7).

$$\phi \left(X_i\right)={\sum }_{T\subseteq N_{\left\{i\right\}}}{\displaystyle \frac{\left|T\right|!\left(N-\left|T\right|-1\right)!}{\left|N!\right|}}\left(f\left(T\bigcup \left\{i\right\}\right)\right)-f\left(T\right)$$

(7)

where N denotes the entire set of features, $N_{\left\{i\right\}}$ represents the set of all possible combinations excluding ${ x}_i$, $T$ indicates the set of features in $N_{\left\{i\right\}}$, $f\left(T\right)$ denotes the model prediction for features in $T$, and $f\left(T\bigcup \left\{i\right\}\right)$ represents a model prediction that combines $T$ features with the feature $x_i$.

3. Results and Discussion

3.1. Publication Bias in the Chinese Literature on Heavy Metals in Agricultural Soils

This study employs regression analysis for systematic testing, establishing a regression equation with effect size (Es) as the response variable and sample size (n) as the independent variable. Small-sample studies are often characterized by high standard errors, resulting in scattered and dispersed data points along the y-axis. With increasing sample size, standard error decreases, precision improves and data points become more concentrated. When the slope of the linear fit approaches 0, it indicates low bias and high data reliability [40]. The bias assessment in this study, with a focus on publication bias (Figure 2), reveal that the slopes of the fitting curves for all five heavy metal elements are close to zero (slope range of −9.28466 × 10⁻⁵–9.4998 × 10⁻⁶). These results confirm that the possibility of publication bias in this study is minimal and that the data exhibits high reliability.

3.2. Characteristics of Heavy Metal Pollution in Farmland Soils

As shown in Table 2, the Cd content ranged from 0.03 to 1.17 mg/kg, with an average value of 0.25 mg/kg. The As content ranged from 1.78 to 45.32 mg/kg, averaging 11.19 mg/kg. The Cr content ranged from 0.05 to 139.50 mg/kg, averaging 64.80 mg/kg. The Hg content ranged from 0.018 to 1.18 mg/kg, with an average of 0.122 mg/kg. The Pb content ranged from 6.81 to 80.06 mg/kg, with an average of 31.80 mg/kg. The soil concentrations of the heavy metals Cd and Hg in China have significantly increased, whereas those of Cr and Pb have decreased, and the As content has remained stable. The pollution severity ranking is Cd > Hg > Pb > Cr > As. Chen et al. [30] also reached similar conclusions. Compared with the national soil background values, the Cd and Hg concentrations significantly increased, exceeding the background levels by 160.89% and 90%, respectively. Their average soil accumulation indices were 0.80 and 0.34, indicating mild contamination. The Cr and Pb concentrations exceeded the background values by 7.16% and 22.64%, with indices of −0.49 and −0.29, respectively. As exhibited a value of −0.59.

Table A5. Domestic research on soil heavy metal content (mg/kg).

	Research Period	As	Cd	Cr	Hg	Pb	Research Method	Reference
China agricultural soils	2003–2025	11.18	0.245	64.79	0.123	31.67	Review (224 articles)	This Article
China agricultural soils	2002–2020	10.40	0.24	59.97	0.118	32.73	Review (603 articles)	[82]
China agricultural soils	2000–2021	11.00	0.35	62.91	0.14	34.67	Meta analysis (449 articles)	[56]
Qatar	2020	27.60	0.20	85.70	—	18.20	Field (50 samples)	[83]
Australia	2012	3.00	0.04	48.00	—	13.00	Field (2211 samples)	[84]
Europe		7.00	0.18	64.00	—	21.00
United States of America	2024	—	0.32	—	—	26.50	Review (9183 points)	[85]
	1993	—	0.34	—	—	15.00	Field (3045 sites)	[86]
England and Wales	2012	15.00	0.33	—	—	49.00	Field (569 samples)	[87]

presents the current status of heavy metal levels in this study compared with relevant research. Based on sample means, the findings of this study are broadly consistent with domestic research conclusions, both indicating that cadmium and mercury are the primary pollutants requiring urgent remediation [40].

Compared with international soil heavy metal levels (Table A5), the average Cd concentration (0.25 mg/kg) in China was lower than those in the United States (0.34 mg/kg and 0.32 mg/kg) and England (0.33 mg/kg), but higher than those in Australia and Europe (0.04 mg/kg and 0.18 mg/kg, respectively). As levels (11.19 mg/kg) and Pb levels (31.80 mg/kg) in China were intermediate between those in Europe (As: 7 mg/kg; Pb: 21 mg/kg) and the more industrialized England (As: 15 mg/kg; Pb: 49 mg/kg). Notably, historical industrial pollution in England has resulted in more severe soil contamination than that in other parts of Europe [41]. Based on the risk screening values for farmland soils specified in GB 15618-2018 [42], Cd exceedance rates of 24.66% and Hg exceedance rates of 1.61% were observed in China. The Cd value exceeds global Cd contamination levels in agricultural soils, Hou et al. reported that approximately 9.0% of global agricultural soils exceed the agricultural safety threshold for Cd [3]. The current stringent soil risk management standards in China significantly overestimate soil heavy metal contamination levels. Taking Cd as an example, China’s standard (0.3 mg/kg) is stricter than those of Russia (0.76 mg/kg) and Finland (1.0 mg/kg) [43]. However, the soil background levels of Cd in China (Table 2) are comparable to those in Russia’s Arctic permafrost regions (0.053 mg/kg) and Finland in 1987 (0.123 mg/kg) [44,45]. Applying Russian or Finland standards would reduce, China’s soil Cd exceedance rate from 24.66% to 0.90–2.22%. Similarly, China’s Hg standard (0.5 mg/kg) is stricter than those of Sweden and Norway (1.0 mg/kg), despite the background Hg concentration in China (Table 2) being close to those in central Norway (approximately 0.06 mg/kg) [43,46]. Adoption of the Norway’s standard would reduce Hg exceedance rates from 1.61% to 0.54%. For other elements such as As and Pb, despite the similar soil background values to Japan, China’s regulatory standards (As: 30 mg/kg, Pb: 80 mg/kg) are significantly stricter than Japan’s (As: 150 mg/kg, Pb: 150 mg/kg) [43,47]. Those comparisons indicates that the relatively high exceedance rates of soil pollutants in China are largely attributable to the stringency of its regulatory standards. The underlying reasons for these divergent international standards are complex and often rooted in differences in soil environmental background values, land use patterns, and national risk assessment frameworks. Soil background values, which are influenced by regional geology and pedogenesis, form the natural baseline for contamination assessment. While this study notes comparable background levels for some elements between China and certain other regions, the integration of these baseline levels into health-based standards varies. Furthermore, land use types critically determine exposure pathways and risk scenarios [48]. Therefore, China’s stringent standards likely reflect a precautionary approach, synthesizing its specific environmental background, intensive land use, and public health priorities. This contextual understanding strengthens the scientific basis for policy recommendations, suggesting that while international benchmarking is informative, soil management policies must be tailored to national and regional circumstances.

Table 2. Descriptive statistics of the heavy metal concentrations in the soils of agricultural land (mg/kg).

Heavy Metal	As	Cd	Cr	Hg	Pb
Number of samples 1	235,643	248,386	219,834	229,085	239,561
Max 1	45.32	1.17	139.50	1.18	80.06
Min 1	1.78	0.03	0.05	0.018	6.81
Mean 1	11.19	0.25	64.80	0.122	31.80
Igeo	−0.59	0.80	−0.49	0.34	−0.29
Number of outliers	2	6	6	2	3
Background value 2	11.20	0.097	61.00	0.065	26.00
Standard value 3	30.00	0.30	250.00	0.50	80.00

¹ Outliers were excluded during the calculation process. ² The background values were selected from the Soil Element Background Values in China [49]. ³ The background values were selected from the Standard values were based on the secondary standard heavy metal concentration (paddy fields with a pH ≤ 5.5) (GB 15618-2018) [50].

Table 2. Descriptive statistics of the heavy metal concentrations in the soils of agricultural land (mg/kg).

Heavy Metal	As	Cd	Cr	Hg	Pb
Number of samples 1	235,643	248,386	219,834	229,085	239,561
Max 1	45.32	1.17	139.50	1.18	80.06
Min 1	1.78	0.03	0.05	0.018	6.81
Mean 1	11.19	0.25	64.80	0.122	31.80
Igeo	−0.59	0.80	−0.49	0.34	−0.29
Number of outliers	2	6	6	2	3
Background value 2	11.20	0.097	61.00	0.065	26.00
Standard value 3	30.00	0.30	250.00	0.50	80.00

¹ Outliers were excluded during the calculation process. ² The background values were selected from the Soil Element Background Values in China [49]. ³ The background values were selected from the Standard values were based on the secondary standard heavy metal concentration (paddy fields with a pH ≤ 5.5) (GB 15618-2018) [50].

3.3. Spatiotemporal Patterns of Heavy Metal Accumulation in Soil and Ecological Risk Assessment

Spatial variation analysis further reveals distinct patterns of pollutant distribution and ecological risk across the national scale (

Table A3. Geoaccumulation index for different regions of China.

Region	As	Cd	Cr	Hg	Pb
Northwest arid region (NAR)	−0.5	0.27	−0.47	1.08	−0.15
Loess plateau (LP)	−0.53	0.33	−0.29	1.23	−0.13
Northeast region (NR)	−0.34	0.3	−0.69	1.11	−0.5
East coast (EC)	−0.88	0.31	−0.43	−0.47	−0.3
South coast (SC)	−0.61	0.81	−0.47	0.34	−0.12
Huanghuai plain (HP)	−0.77	0.83	−0.75	−0.17	−0.36
middle reaches of the Yangtze River (YR)	−0.63	0.89	−0.52	−0.09	−0.31
Sichuan basin (SB)	0.05	1.53	−0.78	1.55	−0.33
Yunnan and Guizhou plateau (YGP)	−0.66	−0.89	−0.69	0.07	−0.16
Qinghai-Tibetan plateau (QTP)	−0.44	0.41	−0.68	1.34	−0.41

and

Table A4. Ecological risk assessment for different regions of China.

Region	As	Cd	Cr	Hg	Pb
Northwest arid region (NAR)	10.60	54.27	2.17	127.27	6.76
Loess plateau (LP)	10.39	56.76	2.44	140.35	6.85
Northeast region (NR)	11.88	55.29	1.87	129.73	5.30
East coast (EC)	8.18	56.59	2.22	43.40	6.09
South coast (SC)	9.81	78.75	2.16	76.19	6.88
Huanghuai plain (HP)	8.79	79.86	1.79	53.36	5.85
middle reaches of the Yangtze River (YR)	9.67	83.50	2.09	56.23	6.06
Sichuan basin (SB)	15.50	130.19	1.75	175.44	5.98
Yunnan and Guizhou plateau (YGP)	9.91	24.30	1.87	63.04	6.72
Qinghai-Tibetan plateau (QTP)	11.03	59.72	1.87	152.20	5.65

). The spatial distribution of Cd exhibited a clear macro-scale pattern, with pollution hotspots predominantly concentrated east of the Hu Huanyong Line, primarily in YR, HP, and SC (Abbreviations are shown in Figure A1). This finding is consistent with previous studies [12]. Notably, the strong correlation with the Hu Huanyong Line underscores the profound influence of socio-economic factors on Cd pollution, a pattern also observed by [40]. The ecological risk of Cd was generally moderate, but reached a strong level in the SB. In contrast, Hg pollution displayed more scattered spatial pattern, with core contaminated zones identified in western, northern, and plateau regions such as QTP, LP, and NAR. The observation of moderately to heavily contaminated spots in these regions, accompanied by high ecological risk. Ren has been partially documented by [40] in northwestern China, confirming the regional specificity of Hg threats.

Regarding other elements, As and Cr showed no significant nationwide pollution enrichment or pronounced spatial clustering, with ecological risks consistently slight. This stability and lack of strong spatial gradients are in agreement with Ren [40]. Similarly, Pb pollution was predominantly at an unpolluted to moderate level, with only localized enrichment in northwestern regions (Figure 3).

3.4. Spatial and Temporal Trends of Heavy Metal Pollution in Farmland Soils

The geoaccumulation index for the five heavy metals exhibited distinct temporal patterns between 2003 and 2025 (Figure 4). The $l_{geo}$ value for As remained negative, indicating an unpolluted state, but it showed an upward trend. In contrast, the $l_{geo}$ values for Cd and Hg were positive and demonstrated a clear upward trajectory over the study period, although the rate of increase appeared to slow after approximately 2017. This attenuation is likely attributable to the implementation of stringent environmental policies—such as the Soil Pollution Prevention and Control Action Plan—which have effectively reduced anthropogenic inputs of heavy metals into agricultural soils. The $l_{geo}$ for Cr fluctuated within a narrow range between −1.0 and −0.5, showing no significant trend, possibly reflecting its dominant natural origin and limited anthropogenic influence. The $l_{geo}$ for Pb increased from 2003 to around 2015, after which it began a marked decline, a reversal that coincides with the nationwide phase-out of leaded gasoline and enhanced industrial emission controls. Overall, the changes in concentration for these five heavy metals over the past two decades have been relatively modest. At the national scale, heavy metal pollution in China’s farmland soils remains generally mild, and the previously observed increasing trend has been preliminarily curbed. This improvement is likely attributable to the implementation of stringent environmental policies which have effectively reduced heavy metal inputs into agricultural soils. The findings of [11,40] also reported a decreasing trend of increasing Cd and Hg pollution in China in the past decade.

While the strict standards present a statistical challenge, they have been paralleled by a suite of aggressive policies that have successfully curbed the escalating trend of heavy metal pollution over the past decade. The escalating trend of heavy metal mercury and cadmium contamination in Chinese soils has been curbed. Compared with the 2006–2015 period, the comprehensive phase-out of leaded gasoline in 2000 and stringent post-2010 emission standards for lead and zinc industrial pollutants resulted in a 43.62% reduction in atmospheric Pb emissions during 2015–2020 [51]. Since 2012, China has implemented stringent industrial emission standards and fulfilled its obligations under the Minamata Convention on Mercury, which prompted multipollutant synergistic ultralow emission retrofits in coal-fired power plants, nonferrous metals, and cement industries, as well as the phased closure of primary mercury mines (except for special purposes). As a consequence, compared with the 2006–2015 period, atmospheric Hg emissions decreased by approximately 63.87% during 2015–2020 [51]. Following the nationwide ban on sewage irrigation in 2013, agricultural irrigation has relied primarily on surface runoff and groundwater. Since the implementation of the Water Pollution Prevention and Control Action Plan in 2015, the proportion of surface water sections meeting Class I–III standards nationwide has increased from 67.8% in 2016 to 89.4% in 2023, representing a 21.6% improvement, whereas the proportion of Class V or lower sections declined from 8.6% to 0.7% (a reduction of 7.9%). Heavy metal concentrations in surface water decreased by 1 to 19 times [52], reducing soil heavy metal inputs from irrigation water. China launched the “Zero Growth in Fertilizer Use Action” in 2015. By 2024, application rates of phosphate, nitrogen, and potassium fertilizers had decreased by 33%, 31%, and 14.7%, respectively, compared with the 2009 levels (Figure A2). Fertilizer registration and management measures have imposed strict controls on the heavy metal content in registered fertilizer products, thereby reducing fertilizer-derived inputs of certain heavy metals. Furthermore, continuous improvements in fertilization techniques and advances in production processes have also decreased heavy metal contributions from fertilizers. In 2017, the Hygiene Standard for Feeds (GB13078-2017) [53] systematically established maximum allowable limits for heavy metals, including Pb, Hg, As, Cd, and Cr in livestock, for the first time, effectively reducing heavy metal concentration in manure. Between 2015 and 2020, relative to 2006–2015, total inputs of As, Cd, Cr, Hg, and Pb from livestock manure decreased by 9.90%, 32.76%, 9.03%, 30.77%, and 18.49%, respectively [51]. However, the current Limit Requirements of Toxic and Harmful Substance in Fertilizers (GB 38400-2019) [54] still set higher thresholds for Cd and Hg than the soil pollution risk screening values defined in GB 15618-2018 [50], implying that fertilizer application remains a potential source of soil heavy metal contamination. In 2016, the State Council issued the Action Plan for Soil Pollution Prevention and Control, establishing the first comprehensive institutional framework for soil pollution management in China. This initiative has played a fundamental, systematic, and transformative role in reversing the trend of soil pollution deterioration.

The spatial evolution patterns of the five heavy metals in Chinese farmland soils across the three periods, as revealed by the SDE analysis, are shown in Figure 3. In Phase I, the 48 cases recorded exhibited a north-south axial banded distribution, with most cases located in eastern provinces and the SDE rotation angle was 19.32°. In Phase II, the number of cases increased to 75, forming a dual-core spatial pattern with expansion in the both the eastern coastal regions (Hebei, Zhejiang) and the central–southern inland areas (Hunan, Guangxi). During this phase, the SDE rotation angle reached 93.27°, predominantly oriented “east (slightly south) and west (slightly north)”. The SDE gradually shifted along a northeast-southwest axis, with its centroid moving westward by 3.67° in longitude from east to southwest. The east-west standard distance increased, with many cases distributed within the ellipse, indicating higher spatial density of heavy metal occurrences. In Phase III, the number of cases further increased to 121. The SDE major axis (12.51) and minor axis (7.2) expanded by 4.5% and 5.5%, respectively, compared with phase I and phase II. The SDE rotation angle increased to 109.87°, with its major axis oriented along the northeast–southwest direction and the centroid continued to drift westward 3.74° in longitude. Over the past two decades, the SDE centroid has shifted from the eastern coastal regions toward the western inland areas, accompanied by an expanding spatial distribution and heightened dispersion. This migration pattern aligns with the evolution of China’s economic landscape. Over the past decade, the central and southwestern regions have progressively emerged as the nation’s fastest-growing economic zones. From 2013 to 2023, the annual average growth rate of Gross Domestic Product (GDP) in the central and western regions (LP, YR, SB, YGP, QTP, NAR) stood at 7.66%, exceeding that of the eastern regions (EC, NR, HP, SC) by 0.94 percentage points [55]. Ren and Yang also reached similar conclusions [40,56].

3.5. The Drivers of Potentially Toxic Elements Pollution in Farmland Soils

Figure 5 illustrates the primary drivers for five heavy metals, with varying contributions from different influencing factors across the five heavy metals. The primary influencing factors for As were identified as Average Comprehensive Utilization Rate of Industrial Solid Waste (ACURISW), Average Per Capita GDP (APCGDP), Average Output Value of Primary Industry (AOVSI), Fertilizer, and Average Wind Speed (AWS). For Cd, the dominant factors included Soil Organic Matter (SOM), AWS, Total Nitrogen (TN), Irrigation, and Total Volume of Non-Compliant Industrial Wastewater Discharge (TVNCIWD). The key factors affecting Cr were Average Temperature (AT), Average Humidity (AH), Total Volume of Industrial Wastewater Discharge (TVIWD), Average Precipitation (APR), and AWS. In the case of Hg, the major influencing factors were TVIWD, pH, AH, Clay, and SOM. Regarding Pb, the most significant factors were pH, Clay, APR, AT, Available Phosphorus (AP), and SOM.

When interpreted in conjunction with existing literature on Chinese agricultural soils, our findings underscore the combined influence of climatic conditions and soil physicochemical properties on heavy metal distribution, with the exception of arsenic. For As, the model places greater emphasis on socio-economic and industrial indicators (e.g., ACURISW, APCGDP), aligning with mounting evidence linking arsenic accumulation to anthropogenic activities [2]. For the remaining heavy metals (Cd, Cr, Hg, and Pb), our results strongly support established consensus. For the heavy metals examined in this study (Hg, Cr, Pb, Cd), environmental behavior is predominantly governed by the interaction between soil physicochemical properties (e.g., pH, SOM) and climatic factors (e.g., AWS, AT, AH), which collectively form a key integrated driving framework [57,58]. Notably, for Hg, the combined influence of industrial sources (TVIWD) and soil pH highlights a typical scenario where pollutant emissions interact with soil geochemical conditions to determine final accumulation. This nuance refines the purely industrial emission source previously emphasized by Song [15].

The R² values for As, Cd, Cr, Hg, and Pb were 0.60, 0.61, 0.81, 0.85, and 0.60, respectively. These results indicate that the models for Cr and Hg explained over 80% of the spatial variance, demonstrating high reliability. The models for As, Cd, and Pb, while slightly lower, still accounted for approximately 60% of the variance, which is considered acceptable for complex environmental systems. The corresponding RMSE values (As: 2.97, Cd: 0.09, Cr: 8.09, Hg: 0.04, Pb: 7.80) provide context for the average prediction error relative to the actual concentration ranges. The slightly lower but still acceptable R² for As, Cd, and Pb (around 0.60) indicates that while our model captured the major influencing factors, there might be additional localized or unmeasured sources contributing to their variability. This favorable model performance underpins the credibility of the subsequent spatial distribution and driver analysis (Figure 6).

Broadly, the key drivers can be summarized as follows: anthropogenic activities play a dominant role in governing As accumulation while also exerting a partial yet significant influence on Cd and Hg. In contrast, Cr is primarily controlled by climatic conditions, and Hg and Pb are mainly regulated by soil physicochemical properties. Importantly, the environmental behavior of multiple heavy metals—particularly Hg, Cr, Pb, and Cd—is jointly shaped by the interactive effects of climatic factors and soil properties, highlighting the complexity of their driving mechanisms. This categorization provides a structured framework for discussing the underlying processes, which are elaborated in the subsequent sections according to these three factor categories.

Among anthropogenic factors, Cd accumulation is affected primarily by the irrigation volume, the discharge of noncompliant industrial wastewater, and fertilizer application rates. Owing to a 56-year history of sewage irrigation, heavy metal-contaminated areas accounting for 65% of the total sewage-irrigated land area, with Hg and Cd being the most severe contaminants [59]. Although sewage irrigation was officially banned in 2013, sporadic practices persists in several northern regions (e.g., the HP) [60]. Excessive irrigation further intensifies leaching, enhancing the downward migration of heavy metals into deeper soil and groundwater, thereby increasing the risks of secondary pollution diffusion [61]. In agriculture system, phosphate fertilizers are a major pathway for Cd and As, as these elements present as impurities in phosphate rock [29]. Moreover, excessive nitrogen fertilizer application decreases soil pH, creating acidic conditions that enhance heavy metal dissolution, migration and bioavailability [62]. This effect is particularly obvious in the major grain-producing and fertilizer-intensive regions such as HP, SB, YR, and SC [63,64], which account for 14.7%, 8.7%, 14%, and 7.3% of the national phosphorus fertilizer consumption and 21.5%, 7.3%, 15%, and 7.4% of the nitrogen fertilizer application, respectively [55]. Industrial solid waste utilization rates, per capita GDP, and fertilization practices significantly affect soil As concentration. Consequently, industrial zones and intensive farmed regions (e.g., HPs and YRs) may represent potential hotspots of arsenic accumulation.

Climate factors (temperature, wind speed, humidity, and precipitation) and soil physicochemical properties (pH, clay content, and soil organic matter (SOM)) jointly control the environmental behavior of heavy metals (Hg, Cr, Pb, and Cd), with significant interactions between these two categories of drivers. In terms of climate, elevated temperatures enhance the solubility and mobility of heavy metals in environmental media [65] and accelerate the SOM decomposition, thereby altering metals-binding forms and stability [66], which can lead to the release of previously bound metals or the formation of more stable organo-metal complexes, depending on the specific metal and soil conditions [67]. Moreover, increased temperature can stimulate microbial activity, which consumes oxygen and lowers the redox potential (Eh) in soil microsites [68,69]. This shift towards anaerobic conditions can promote the reductive dissolution of iron and manganese (oxyhydr) oxides, thereby releasing associated heavy metals such as As and Cd into the soil solution [69]. further indicated that heavy metal contamination is more severe in subtropical monsoon climate zones than in colder, humid subnorthern climate zones. It does this firstly through leaching, which lowers soil pH and enhances metal desorption from clay particles. Secondly, the resulting water saturation alters the soil’s redox state, triggering the reductive dissolution of iron and manganese oxides and the subsequent release of their associated heavy metals. These processes work in tandem to drive the desorption and remigration of heavy metals, ultimately controlling their vertical and lateral redistribution [70,71,72].

This mechanism is obvious in southern regions with high rainfall, especially in southwestern China, where extreme weather events are becoming more frequent [34]. Humidity also affects heavy metal mobility within the soil–plant system by altering evaporation pathways [73]. Wind speed primarily affects heavy metal redistribution and deposition: high-speed winds from uncontaminated directions typically dilute and remove pollutants, reducing soil contamination from atmospheric deposition, whereas low-speed or calm winds may cause pollutant accumulation and deposition, increasing soil heavy metal input and accumulation [74]. With respect to soil physicochemical properties, the soil pH affects the dissolution/precipitation equilibrium and adsorption behavior of heavy metals, whereas the adsorption rate is affected by factors such as precipitation and fertilization. Increased clay concentration increases soil adsorption and immobilization of heavy metal ions, reducing leaching losses and facilitating accumulation in surface soils [75,76,77]. SOM adsorbs heavy metals through complexation [78], but its stability is also affected by factors such as temperature and pH.

Compared with anthropogenic influences, climatic factors and soil physicochemical properties are the main drivers affecting the spatial distribution and accumulation of multiple heavy metals. This process essentially involves complex source–sink relationships and migration pathways (Figure 7). For example, climatic factors not only directly affect the migration and transformation of heavy metals but also indirectly affect their fate by altering critical soil properties. In the context of global climate change, rising temperatures and frequent extreme precipitation events may significantly alter regional soil environments, further disrupting heavy metal stability and elevating their ecological risks, ultimately posing threats to agricultural environmental safety [79,80,81]. With respect to Cd, the high-risk regions in China—such as the SB, SC, and YR—are located within the subtropical monsoon climate zone. The combination of elevated-temperatures and abundant rainfall not only increase weathering and the release of metals from parent materials but also facilitate Cd accumulation in surface soils by affecting SOM decomposition and the adsorption-desorption behavior of clay particles [3]. Therefore, it is recommended that future research systematically investigate the environmental behavior and ecological risks of heavy metals in the context of climate change, providing theoretical support for the formulation of strategies aimed at preventing and controlling food security risks.

4. Conclusions

This study transcends traditional static perspectives to systematically elucidate the spatio-temporal evolution patterns and driving mechanisms of heavy metal contamination in China’s soils. Findings reveal a pronounced dynamic shift in the distribution of heavy metal pollution within agricultural soils: contamination hotspots have migrated from coastal regions towards the southwestern inland areas. The emergence of the Sichuan Basin as a new pollution hotspot signifies a significant transformation in the spatial distribution of contamination. Regarding elemental distribution, cadmium and mercury exhibit persistent accumulation nationwide, contrasting sharply with the relatively stable pollution levels of arsenic, chromium, and lead. Since 2017, however, the accumulation rates of cadmium and mercury have markedly slowed, providing crucial evidence for the effectiveness of national environmental governance measures. Regarding driving mechanisms, industrial activities (industrial waste and wastewater discharge), levels of economic development, and agricultural practices (irrigation and fertilization) were identified as primary anthropogenic drivers. Concurrently, climatic conditions (temperature, wind speed, humidity, and precipitation) and soil properties (pH, clay content, and organic matter) constitute crucial natural modulating factors shaping the spatial distribution of pollution. This mechanistic understanding provides a theoretical basis for formulating targeted intervention strategies. In summary, this study not only elucidates the intrinsic patterns governing the spatiotemporal evolution of heavy metal pollution but also validates the positive outcomes of recent environmental regulatory policies. To effectively safeguard food security, we recommend prioritizing the advancement of regionally differentiated, precision prevention and control strategies in future work. Furthermore, enhanced research into predicting metal migration and transformation under climate scenarios is essential to address evolving environmental challenges. This study holds multiple practical values for advancing sustainable development: Firstly, through the precise identification of pollution hotspots and systematic analysis of their driving mechanisms, it provides scientific support for formulating differentiated control policies that balance ecological security with food production. Secondly, empirical analysis of governance outcomes demonstrates that effective environmental regulation can synergistically advance regional development processes, thereby providing crucial validation for sustainable soil management models. Furthermore, integrating source–sink mechanisms of soil heavy metal pollution with spatial regulation strategies into the national territorial governance framework can furnish systematic decision-making foundations for harmonizing human–land interactions and ensuring sustainable ecosystem management.