a Key Laboratory of Pollution Processes and Environmental Criteria (Ministry of Education), Tianjin Key Laboratory of Environmental Remediation and Pollution Control, College of Environmental Science and Engineering, Nankai University, Tianjin 300350, China
b Xinjiang Key Laboratory of Crop Gene Editing and Germplasm Innovation, Institute of Western Agricultural of CAAS, Changji 831100, China
c Key Laboratory for Environmental Factors Control of Agro-product Quality Safety (Ministry of Agriculture and Rural Affairs), Tianjin Key Laboratory of Agro-environment and Safe-product, Agro-environmental Protection Institute, Ministry of Agriculture and Rural Affairs, Tianjin 300191, China
d School of Materials Science and Engineering, Nanyang Technological University, Singapore 639798, Singapore
e Stockbridge School of Agriculture, University of Massachusetts, Amherst, MA 01003, USA
The continuous increase in microplastic (MP) pollution poses significant risks to human health and environmental sustainability, especially in agroecosystems. This study focused on identifying and managing MP risk to crops in agricultural soils in China, which is among the world’s largest consumers of plastic. Via the use of 3243 site-year field observations, we developed intelligent agriculture models to predict MP-related crop risks and identify key drivers, such as climate, livestock density, and fertilizer application, other than the use of agricultural plastic film. Rice was most sensitive to MPs, with an average risk quotient (RQ; unitless) of (3.76 ± 1.95), which is 2.19 and 1.93 times greater than those of maize and wheat, respectively. Climate factors are closely related to livestock density and agricultural management practices, potentially exacerbating MP risk under future conditions. Optimizing livestock density and fertilizer use levels reduced MP risk by 20.9%, 22.9%, and 20.3% and increased crop yields by 9.0%, 6.0%, and 5.6% for maize, rice, and wheat, respectively. Despite limitations related to model uncertainty and policy implementation, the proposed intelligent agriculture model provides a comprehensive basis and potential solutions for assessing and managing MP risk to crops.
Microplastic (MP; particle size < 5 mm) pollution is becoming a central issue in environmental science [1,2] and is widely present in various compartments worldwide, such as soils, inland waters, oceans, and the atmosphere [3], [4], [5], [6]. MPs can enter agroecosystems through various pathways, such as organic fertilizer application, wastewater irrigation, plastic mulching, and atmospheric deposition [7], [8], [9], [10], making agroecosystems the primary sinks of various plastics. Global plastic production reached more than 430 million tonnes in 2023 and is projected to double within two decades, exacerbating MP pollution in agriculture (United Nations Environment Programme (UNEP), 2023) [11]. Previous studies have revealed that MPs may adversely influence crops (e.g., by inhibiting seed germination, disrupting metabolism, and reducing yields) [12,13]. Therefore, it is crucial to manage agricultural pollution caused by MPs without compromising crop productivity, which is closely related to sustainable agricultural development [14].
Although the MP footprint in agricultural soils has received considerable attention in recent years, the effects of MP pollution in field soils on crops remain largely unknown [14], [15], [16]. Researchers have assessed potential MP risk to crops on the basis of the spatial distribution of MPs, but have not considered differences among specific crop types, nor have they adopted spatiotemporally invariant risk indices [17,18]. In addition, the lack of methods to balance the diversity of the properties of MPs (e.g., size, type, and shape) in agricultural soils in the field poses a challenge for risk assessment [19]. The risk quotient (RQ, unitless) is an index that can be used to quantify pollutant risk, but the lack of a quantitative link between MPs and crop growth risk currently impedes risk assessment [20,21]. Integrating data on MP properties and toxicity effects to determine RQs may provide a solution for quantitatively assessing MP risk to crops. Previous studies have indicated that the response of crops to MPs is nonlinearly influenced by MP properties, growth conditions, livestock production, and fertilization, as well as by their interactions [9,22]. For example, under high-precipitation and high-temperature conditions, MPs reduce soil total nitrogen and organic carbon levels, thus affecting crop growth [23]. The application of fertilizer can introduce MPs into croplands, and these MPs can alter soil properties and disrupt microbial communities, thus affecting crop growth [24]. However, MP risk mapping accounting for crop types and local environmental factors is lacking [25]. Without comprehensive and localized MP pollution analysis for croplands, accurate measurement of agricultural MP risk to crops remains challenging, limiting the ability of policymakers to develop appropriate strategies. Previous studies have highlighted a disconnect between MP control and on-farm practices [14,26]. There is an urgent need for practical strategies to reduce MP-related crop damage. To address this challenge, artificial intelligence (AI)-based agriculture, in which data sensing, modeling, and algorithm-based decision-making are integrated, offers a promising pathway for managing crop production under pollution constraints [27], [28], [29], [30], [31]. For example, Internet of Things-based agricultural systems encompassing a deep belief network can be used to accurately predict crop prices and support income analysis, highlighting the potential of intelligent decision-making [32]. Data-driven multiobjective optimization algorithms (e.g., the nondominated sorting genetic algorithm II (NSGA-II)) that aim to examine the potential to reduce MP risk might provide an opportunity for the development of management practices [33]. Embedding such tools into intelligent agricultural systems could enable site-specific decision support for managing MP risk in real-world cropland scenarios.
Here, with the selection of China, one of the largest consumers of plastics worldwide, as a case study, we explored the effects of natural and anthropogenic environmental factors on soil MP pollution and crop risk. We employed 3243 site-year field observation data for MPs (e.g., abundance, size, shape, and type) to develop intelligent agricultural models on the basis of various factors, such as climate, soil, anthropogenic activities, and agricultural management. With the developed models, we predicted crop risks from MPs and identified the main drivers of pollution. Additionally, win-win management practices (e.g., livestock density, manure application, and chemical fertilizer application control methods) were proposed via multiobjective optimization to reduce the crop risk from MPs and increase yields. In contrast to traditional methods, such as manual soil sampling and laboratory filtration, which are often labor intensive, time consuming, and spatially constrained, our data-driven framework provides the integration of large-scale field observations, machine learning, and multiobjective optimization into AI-enabled agricultural systems. This facilitates scalable and adaptive crop-specific risk assessment and management of MP pollution under assumed or real-world conditions.
2. Materials and method
2.1. Compiling a dataset of the distribution of MPs in Chinese agricultural soils
To determine the risks of agricultural MP pollution, a comprehensive MP dataset was compiled. Further details are provided in the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flowchart shown in Fig. S1 in Appendix A. We systematically reviewed the literature concerning the distribution of MPs in agricultural areas in China from 2017 to 2024. Ultimately, 3243 datasets were extracted and screened to create a database documenting the occurrence and distribution of MPs in Chinese agricultural areas. The sampling sites cover 31 provinces across China and encompass the 2016-2022 period. The geographical distribution of the samples is shown in Fig. 1. Detailed information on the 108 studies is provided in Dataset S1 in Appendix A. The data related to MPs are complex and were derived from the text, tables, and graphs of the screened publications. More details are provided in Text S1 in Appendix A.
2.2. Alignment of MP data via continuous probability distributions for size, shape, and type
Owing to the nonuniformity of the data in publications, such as the size detection range and the diversity of shapes and types, a certain degree of uncertainty is often inherent in risk assessments of MPs at large scales [26]. Although the abundance of MPs is highly related to particle-related risk, this is not a characteristic of the material. We captured environmental variability via empirical data for the abundance, size, shape, and type of MPs in Chinese agricultural soils. We developed an abundance-mass concentration conversion method to accurately reflect the true material characteristics of MPs. In publications, varying size ranges are adopted to define MPs, typically between 1 and 5000 μm. An often-used pragmatic interval is that between approximately 20 and 5000 μm, with a range of 20 μm related to the practical detection limit of common Fourier transform infrared (FTIR) spectroscopy methods [34]. Previous research has demonstrated that the distributions of MP sizes typically follow a power law [19], and we fitted the following relationship between the relative abundance and size of MPs on the basis of 25 environmental distributions of the sizes of MPs in soil (Fig. S2 in Appendix A):
$ y=0.5458 \cdot x^{-1.2546}$
where 𝑦 denotes the relative abundance, and 𝑥 denotes the particle size (mm) based on the longest dimension. Thus, the number of particles within any size range can be converted to expected values ranging from 20 to 5000 μm via Eq. (1).
Currently, conversions between volume and mass concentration on the basis of spherical assumptions (or vice versa) are often inaccurate owing to the wide range of shapes and densities of particles in actual environments. MPs reported in the literature are generally classified into five shape categories, namely, fragments, fibers, films, foams, and beads, but the actual shapes are highly diverse [35]. We considered the length (L):width (W):height (H) ratio range (Fig. S3(a) in Appendix A) for common MP shapes to quantify the continuous distribution of the shape properties of MPs. The L value of each MP shape was set to the maximum size (an assigned value of 1), with W and H ≤ 1. A triangular distribution was assumed to capture the variability in each shape [19], with the minimum and maximum values adopted as upper and lower limits, respectively. The shape with the highest abundance at the center of these limits is shown in Fig. S3(a). On the basis of the obtained L:W:H ratio distribution, the distribution of dimensionless Corey shape factor (CSF) values (with $ \mathrm{CSF}=\frac{\mathrm{H}}{\sqrt{L \times W}}$) was obtained [36]. Through the use of the CSF, properties such as the aspect ratio of MPs were incorporated into the distribution simulation process, thereby preserving the key influence of the shape of MPs on their fate, including how the three-dimensional configuration affects the degradation process in soil [37]. To reflect the relative abundance within different shape categories, we employed the average abundance based on 752 soil samples. Given the number and diversity of studies selected, the data covered wide ranges of locations, ages, and weathering stages for MPs in Chinese agricultural soils. Via the use of a triangular distribution and the relative abundance of MP types as inputs (random number generator; n = 107), the general distribution of the shapes of MPs in agricultural soils was obtained, which exhibited a bimodal shape:
The eight common types of MPs in agricultural soils reported in the literature were considered to quantify the continuous density distribution on the basis of the combination of the relative abundance and density of the different types of MPs shown in Fig. S3(b). The average abundance in each crop category was determined on the basis of 332 actual soil samples. The effects of weathering and degradation on the density of MPs in soils have been considered in previous works [38]. In this study, the density of each MP type was assumed to follow a triangular distribution, with the upper and lower limits representing the minimum and maximum values, respectively, as shown in Fig. S3(b). By combining triangular density factor distributions with the relative abundance of the various MP types, we generated a general density distribution of MPs in agricultural soils (random number generator; n = 107), with a normal-inverse Gaussian shape:
On the basis of the continuous distribution functions of L, CSF, and density obtained through simulations, the MP abundance data for agricultural soil can be converted into mass concentration data. Specifically, 107 MPs (much larger than the abundance values observed for agricultural soils) were obtained via Monte Carlo simulations, and the corresponding L, CSF, and d values were randomly sampled via Eqs. (1), (2), (3), respectively. Here, we assumed L/W = L/H, thus simplifying the conversion equation to the following:
where 𝑖 denotes the index of an individual MP in the simulations. The MP mass (M) concentration in each soil sample was obtained by randomly selecting and summing n (i.e., the abundance of each sample) additive values from the 107 calculated MP mass values. The detailed procedure for converting MP abundance values into mass concentration values is shown in Fig. S4 in Appendix A.
2.3. Risk assessment for the effects of MPs on crop growth
Given the varying tolerance levels of crops (e.g., maize, rice, and wheat) to different MPs, the risk of MPs to crop growth more accurately represents the consequences of MP pollution than does their concentration [39]. In this study, we performed a risk assessment of MPs via the RQ approach, with RQs for MPs determined from measured environmental concentrations (MECs) and predicted no-effect concentrations (PNECs) [40]. To assess the risks of MPs, we first estimated PNECs via the assessment factor (AF) approach [21]. The PNEC was calculated by applying the AF to the no-observed-effect concentration (NOEC) (AF = 100), such as PNEC = NOEC/100 [41]. The NOEC was defined as the concentration at which no effect was observed in the test, and for those concentrations at which the lowest-observed-effect concentration (LOEC) showed effects greater than 10% but less than 20%, the NOEC was calculated by dividing the LOEC by 2 [41]. Toxicity data for relevant crops (namely, maize, rice, and wheat) derived from the soil samples are provided in Table S1 in Appendix A. We subsequently applied a species sensitivity distribution (SSD)-like method [42] to determine the probability distribution of PNEC values for each crop and calculated hazard concentrations (HCs) for 5%, 10%, 15%, and 20% of the samples (HC5, HC10, HC15, and HC20, respectively). The results are shown in Fig. 1(d). More details are provided in Text S2 in Appendix A.
2.4. Collection and preprocessing of variables as predictors
We selected predictor variables to comprehensively capture the complexity of soil MP sources, dynamics, and interactions with crops and categorized them into five categories: climate factors, soil properties, agricultural activities, anthropogenic activities, and others [26]. The climate factors included the annual mean temperature (AMT), annual mean precipitation (AMP), aridity index, and particulate matter ≤ 10 μm in diameter. The soil properties that substantially influence the fate and behavior of MPs in soil [43] are pH, total nitrogen content, soil organic carbon content, clay content, sand content, cation exchange capacity, bulk density, and electrical conductivity. Agricultural management can be considered a potential driver of MP introduction into soils [9,44], encompassing the type of agricultural land use, chemical fertilizer application (nitrogen and phosphorus), manure application, pesticide use, use of agricultural plastic film, irrigation water use, and tillage system. Anthropogenic activities, which are related to the population density, gross domestic product (GDP) density, and livestock density (cattle, pigs, ducks, and chickens), are considered to notably influence agriculture [22]. Other factors include the groundwater depth and slope, which influence crop leaching and MP movement [45,46]. Among the 28 predictors, the land use type (farmland, greenhouses, and mulch film land) and tillage system (no tillage, conventional annual tillage, and traditional annual tillage) are qualitative variables. To ensure compatibility between the datasets and variables, we processed all the data to a spatial resolution of 1 km. The predictors were extracted from the corresponding databases of the sampling locations and years. To avoid bias caused by unequal sample sizes from different literature sources, stratified sampling was applied to obtain the same proportion (60%) of samples from each literature source for model training, with 40% of the samples reserved for model validation. The detailed sources of the data are listed in Table S2 in Appendix A.
2.5. Model development and performance
The random forest (RF) algorithm is an efficient algorithm built on the basis of decision trees and is well known for its high accuracy, high robustness, and exceptional anti-interference capabilities [47]. We selected 28 variables, such as climate factors, soil properties, agricultural management practices, and anthropogenic activities, as features (input data). The RQ of MPs for each crop was adopted as a label (output data). The datasets were split into three subsets (maize, rice, and wheat) on the basis of the different crops, and three RF regression models were established. The RandomForestRegressor, TensorFlow Keras, SVM, and XGBRegressor packages in Python 3.8 were employed to train four models (RF; artificial neural network (ANN), support vector machine (SVM), and extreme gradient boosting (XGBoost) models). Furthermore, the 10-fold cross-validation method was applied to the three RF models to assess their stability and accuracy. The coefficient of determination (R2) and the root mean square error (RMSE) between the observed and predicted values were chosen to measure model performance [48]. As a validation set prior to model construction, 15% of the samples (measured field data) from each subset were randomly selected. This validation set was not included in model construction to ensure that the model did not learn from it during cross-validation. More details are provided in Text S3 in Appendix A.
2.6. Feature interaction and covariance analysis
The number of occurrences and the mean minimum depth were determined via the RandomForestExplainer package in R 4.0.5. These indicators reflect the magnitude of the interaction between two features on the basis of the structure of each tree in the RF model. A high occurrence frequency and a small minimum depth indicate a strong interaction between two features. The normalized average of the two indicators was employed to comprehensively reflect the interaction strength between the two features. Subsequently, 100 feature pairs with the highest interaction strength were identified, and a feature interaction network was established. The feature interaction network was visualized via the Gephi 0.9.2 software package. More details are provided in Text S4 in Appendix A.
2.7. Spatial patterns of RQs in 2020
As the RF model performed the best among the four models, we applied this model to map the RQ values of MPs for maize, rice, and wheat at a resolution of 1 km for the year 2020 in China. The input data included 28 variables (climate factors, soil properties, agricultural management, anthropogenic activities, and others). Planting distribution data for the three crops were obtained from the literature [49]. The detailed sources of the databases are listed in Table S3 in Appendix A. We mapped the predicted RQ values directly using ArcGIS 10.8. Unlike the previous risk classification method [41], we classified RQ values into five levels on the basis of the calculated HC5, HC10, HC15, and HC20 values, such as, low, low-medium, medium, medium-high, and high risk, respectively. More details are provided in Text S5 in Appendix A.
2.8. Co-optimization of multiple management practices
On the basis of the above importance analysis, livestock density, chemical fertilizer application, and manure application were identified as key drivers influencing RQs. Compared with the low feasibility of human intervention in terms of controlling climate factors, it may be feasible to reduce crop RQs by adjusting livestock breeding structures and fertilizer/manure application strategies [28,50]. A multi-objective optimization algorithm (NSGA-II) was employed to identify global optimal management practices [51] to achieve a reduction in crop RQ values without compromising yield. The NSGA-II model is a fast-sorting and elite multi-objective genetic algorithm used to obtain the Pareto-optimal frontiers (a set of solutions representing the best trade-offs among competing objectives). To construct the objective function, the RF models were wrapped to predict both crop RQ values and yields. Crop yield data for 2020 were obtained from Global Spatially Disaggregated Crop Production Statistics Data for 2020, version 1.0. To increase the feasibility of the optimization strategy, we restricted the adjustments to livestock density and chemical fertilizer/manure application, and we ignored areas where the optimization goals could not be achieved. The changes in livestock density and chemical fertilizer/manure application (with a judgment interval based on the optimization results of the original data) were constrained to 20%, for instance, 20% higher or lower than the unadjusted values, on the basis of the grid cell values. The other variables remained unchanged. The Geatpy package in Python 3.8 was employed to implement the NSGA-II model for each cell. The sum of the livestock units (LSUs) for the four types of livestock (cattle, pigs, chickens, and ducks) was calculated, with the variation constrained to 20% to ensure the sustainability of livestock farming in the optimized area, where 1 cow = 1 LSU, 1 pig = 0.35 LSU, 1 chicken = 0.012 LSU, and 1 duck = 0.018 LSU [52]. On the basis of the above constraints, some cells did not meet the optimization requirements, and these cells were ignored.
The population size (NIND) in the NSGA-II model was set to 100, and the maximum number of evolutionary generations (MAXGEN) was set to 500. By assessing the objective function values, the Pareto solution set was obtained through multiple iterations of the NSGA-II model. Assuming equal importance of the two goals (RQs and yield), the following search model was applied to filter the solution set on the basis of human decision-making:
$ \frac{O_{\mathrm{R}}-R_{\mathrm{R}}}{\mathrm{abs}\left(R_{\mathrm{R}}\right)}<0 \text { and } \frac{O_{\mathrm{Y}}-R_{\mathrm{Y}}}{\mathrm{abs}\left(R_{\mathrm{Y}}\right)}>0$
where 𝑂R and 𝑂Y denote the optimized values of the two outputs, and 𝑅R and 𝑅Y denote the original values of the two outputs. The percentage of the two objectives relative to the original value was calculated via this equation, and the effectiveness of the two optimization objectives was guaranteed. The sum of the two percentages was obtained, and the optimal single solution at the Pareto boundary was derived with optimization effect maximization as the screening principle, such as the Pareto-optimal solution in this study. After each grid was optimized, digital maps of the management strategy were obtained at a resolution of 1 km, as were the corresponding yields and RQ values.
2.9. Statistical analysis
To determine whether climate factors (AMT and AMP) influence the livestock/manure/chemical fertilizer-RQ relationship, linear mixed effects models were established. The interaction terms of AMT/AMP × RQs and AMT/AMP × livestock/manure/chemical fertilizer for each crop type, as well as AMT/AMP data for future years (2040, 2060, 2080, and 2100) under shared socioeconomic pathways (SSP126, SSP245, SSP370, and SSP585), were included, and the coefficients of these interactions were subsequently estimated in relation to crop RQ values. Since AMT/AMP, livestock/manure/fertilizer, and RQs are continuous variables, analyzing their interactions could lead to collinearity between the interaction predictor and the main effects [53]. We centered all the input variables by subtracting the sample mean and scaling the results by dividing by their standard deviation, thus reducing collinearity and facilitating comparable interpretations of the model parameter estimates. To clearly interpret interaction effects, three scaled values of AMT/AMP (thresholds) were selected: the mean (0), one standard deviation above the mean (+1), and one standard deviation below the mean (−1). This approach is typically adopted for analyzing the interactions between continuous predictors [54]. The three values of AMT/AMP represented three levels of AMT/AMP intensity: low (below average), medium (average), and high intensity (above average).
The Kruskal-Wallis test was employed to test for significant differences in the properties of MPs in different samples. The Wilcoxon rank-sum test was applied to test for significant differences in RQ values among provinces. The Wilcoxon rank-sum test was used to determine significant differences between the original and optimized RQ values. These tests were performed in the stats package in R 4.0.5.
3. Results and discussion
3.1. MP risk to crops considering livestock density, climate, and fertilization scheme
A dataset of 3243 observations of the abundance and properties (e.g., size, shape, and type) of MPs in agricultural soils in China was compiled and analyzed (Fig. 1(a)). On the basis of the locations of the sampling points and multisource data, a dataset encompassing climate factors, soil properties, agricultural activities, anthropogenic activities, and other factors was constructed (described in Section 2). The multidimensional factor distribution for the dataset is provided in Table S2. The abundance of MPs ranged from 7.00 to 3.38 × 105 items·kg−1, with an average abundance of 2514.56 items·kg−1 (Fig. 1(a)). The MP abundance in greenhouses was lower (P < 0.05) than that in farmland and mulch film areas (Fig. 1(a)), possibly because greenhouse films (with thicknesses ranging from 8 to 50 μm) are more durable and recyclable than mulch films (with thicknesses ranging from 6 to 8 μm) [26]. Notably, the MPs in agricultural soils are generally waste fragments, fibers, and films, accounting for approximately 88.2% of all the MPs shown in Fig. 1(b). The main components of MPs in soils are polyethylene (PE) and polypropylene (PP), accounting for > 50% of all MPs. The distributions of the properties of MPs in agricultural soils were fitted on the basis of field data (Fig. 1(c), described in Section 2), and the heterogeneity in MPs was considered, with the objective of limiting the nonalignment of methods adopted in MP research [34]. The differences in size, shape, and density were corrected with MP property distribution functions for converting abundance values into mass values of MPs (Fig. 1(c)) to determine concentration thresholds of MPs for crops. On the basis of the laboratory data and the SSD-like method (described the Methods section) [42], we determined the HC5 value for the effects of MPs on the growth of three crops, with values of 0.9391, 0.5153, and 0.7757 mg·kg−1 for maize, rice, and wheat, respectively (Fig. 1(d)). These values were subsequently applied as concentration thresholds for the MP risk of each crop type. Furthermore, the MEC/HC5 ratio was adopted as the RQ, which was used to characterize the risk of MPs to crop growth. The above RQ captures the combined influence of MPs in croplands and their impact on crop growth and serves as a quantitative indicator for assessing MP-related risks to crops (Fig. 1).
To establish a quantitative relationship between the predictor variables and RQ values, we employed three RF models, as shown in Fig. 2(a) and Fig. S5 in Appendix A. The prediction errors of the models for the validation set are shown in Fig. S5, and the RMSEs are 1.148 for maize, 2.143 for rice, and 1.551 for wheat. Three-way feature importance analysis was performed, and livestock density, AMT, AMP, aridity, phosphate fertilizer application, and manure application were identified as the major drivers influencing RQ values (Fig. 2(b) and Fig. S6 in Appendix A). On the basis of the interaction network, fertilizer use and climate factors strongly interacted (Fig. 2(c) and Figs. S6(c)-(d) in Appendix A). According to the continuous response and partial dependence analysis results, there was a positive correlation between livestock density and crop RQ (Fig. 2(d) and Fig. S7 in Appendix A). For example, increases in AMT and AMP increased the RQ of maize, especially when the AMT and AMP values exceeded approximately 19 °C and 1100 mm, respectively (Fig. 2, Fig. 2). Increased application rates of nitrogen fertilizer (>30.15 g·m−2·a−1), phosphate fertilizer (> 4.46 g·m−2·a−1), and manure (> 2.80 gN·m−2·a−1) also increased the RQ (Figs. S7 and S8 in Appendix A). Strong interactions were obtained via double-variable partial dependence analysis (Figs. 2(d) and (e) and Fig. S9 in Appendix A). In regions with AMP < 1000 mm, manure application was the primary driver of MP risk, and the RQs notably changed under high-AMP and high-manure conditions (Fig. 2(d)). AMT and phosphate fertilizer application exerted positive additive effects on crop yields (Fig. 2(e)). Unlike the methods employed in previous qualitative studies [55], importance analysis was conducted to identify the rank of the driving factors. Prioritizing livestock density and fertilizer application over the use of agricultural plastic film may provide a perspective for mitigating MP risk to crops [56]. Although quantitative interactions between climate factors and fertilizer application have seldom been documented in Food and Agriculture Organization (FAO) reports, our results (Fig. 2) highlight the importance of tailoring fertilization strategies to specific climatic conditions to mitigate the risk of MPs to crops.
3.2. RQs of MPs driven by agriculture, anthropogenic activities, and climate
Previous studies focused on specific cases or local scales [57], whereas assessment studies of the risk and driving patterns of MPs affecting crops at the national scale at a 1 km resolution are lacking. The average RQs of MPs in Chinese agricultural soils were (1.72 ± 0.79), (3.76 ± 1.95), and (1.95 ± 0.69) for maize, rice, and wheat, respectively (Figs. 3(a)-(c) and Fig. S10 in Appendix A). Maize was mainly subject to low and low-medium risk (81.2%) levels nationwide (Fig. 3(a)). Fig. 3(d) and Fig. S11(a) in Appendix A show that the high RQ values in three regions (northeast, south central, and east China) are driven mainly by climate factors (with an approximately 50% contribution). MPs may change the soil structure to exacerbate nitrogen loss through leaching in areas of heavy rainfall (e.g., south central and east China), thereby inhibiting crop growth [58,59]. Compared with maize, rice is more sensitive to MPs, with 52.1% of rice areas exhibiting low, medium, or medium risk (Fig. 3(b)). The data in Fig. 3(e) and Fig. S11(b) reveal that poor agricultural management was the main driver of risk. For example, excessive fertilizer application could increase the inhibitory effect of MPs on crop growth [60]. Six provinces (Hebei, Jiangsu, Anhui, Shandong, Henan, and Xinjiang) accounted for 85.1% of the total wheat yield, with only 4.0% of areas exhibiting high and medium-high MP risks (Fig. 3(c) and Fig. S11(c)). Compared with maize and rice, wheat demonstrated a distribution pattern of low risk in high-yield areas and high risk in low-yield areas. Notably, wheat was associated with lower MP risk (Fig. 3). The risk of MPs in wheat areas was clearly influenced by climate factors (> 50%), as shown in Fig. 3(f). The created maps of MP risk for maize, rice, and wheat in China can be used to identify patterns of climate factors, agricultural management, and anthropogenic drivers of MP risk and provide a systematic basis for the control of MP pollution. These results (Fig. 3) offer a spatial framework for guiding region-specific policy interventions, such as enhanced fertilizer regulation in rice cultivation-intensive zones and climate-adaptive strategies in areas where MP risk is primarily driven by climate.
3.3. Mitigation potential for MP-based RQs of various crops
To assess how climate factors influence the relationship between RQs and LSU/chemical fertilizer/manure application, the interactions of climate factors (AMT and AMP) with these variables were examined on the basis of climate projection data from 2020 to 2100 under different scenarios (Figs. 4(a) and (b)). The positive effects of LSU/chemical fertilizer/manure application on RQs became more notable with increasing AMT (Fig. 4(a)), indicating the potential amplification of MP-related risks under warming conditions. In contrast, higher AMP could mitigate the risk (Fig. 4(b)). Combined with the conclusions of the 6th Assessment Report of the Intergovernmental Panel on Climate Change (IPCC) [61], which emphasized the roles of climate change and water availability in shaping agricultural productivity, these findings suggest that future climatic shifts may increase MP risk by influencing the interaction between LSU/chemical fertilizer/manure application and MPs.
The predicted RQs were not highly related to the crop yield (adjusted R2 < 0.2), as shown in Fig. S12 in Appendix A, which may be due to other complex factors overshadowing this effect in local areas. Climate change may increase the RQs associated with agricultural management and anthropogenic activities (Figs. 4(a) and (b)). To increase the sustainability of agriculture within the context of climate change, we employed multi-objective optimization (NSGA-II) to identify optimal management practices to reduce RQs and increase yields (Fig. 4(c)). Optimized management practices, such as LSU and chemical fertilizer/manure application adjustments, aim to prevent the one-sided pursuit of a certain goal while ignoring other important factors [33]. The estimated RQs of maize, rice, and wheat in China decreased significantly by 20.9%, 22.9%, and 20.3%, respectively, with P < 0.01 for all the crops (Fig. 4(d)). Moreover, management increased the average unit yield of crops by 9.0%, 6.0%, and 5.6%, with P < 0.01 for all the crops (Figs. 4(c) and (d)). Our findings highlight opportunities for the agricultural sector to reduce the RQs of MPs and increase yields, which can help achieve the Sustainable Development Goals (SDGs) related to zero hunger and life on land [62].
3.4. Reducing MP risk through livestock and fertilizer management
Fig. 4, Fig. 5 reveal that adjusting the livestock density and manure/chemical fertilizer application is an effective strategy for reducing RQs and increasing yields at both the regional and national scales. The produced map of management practices can guide adjustments in livestock and agricultural management to mitigate the risks of MPs to crops (Fig. 5), which was previously impossible when process-based and empirical approaches were adopted. The spatial optimization mechanism is shown in Fig. 5(a). Specifically, with this strategy, 67.6%, 62.6%, and 83.1% of the grid cells should be targeted for maize, rice, and wheat growth, respectively, and the number of LSUs should be reduced (Figs. 5(b)-(d)). Wheat exhibited the greatest LSU replacement rate, at (−1655.04 ± 1773.38) km−2, to reduce RQs, followed by maize ((678.49 ± 1409.47) km−2) and rice ((340.08 ± 1277.14) km−2). A low LSU value corresponds to low levels of MP-containing manure applied in farmland areas, thus decreasing the risk of agricultural water pollution from livestock farming [22]. Increasing the removal of MPs from livestock manure, which has traditionally been neglected, could effectively reduce the number of MPs entering cropland areas [22,63]. Overall, reducing the LSU value in agricultural zones in regions such as east and southwest China could effectively reduce the risk of local MPs to crops (Figs. 5(b)-(d)). Under the optimal management practices, we estimated adjustments in manure and fertilizer application for the three crops (Figs. 5(e)-(g) and Fig. S13 in Appendix A), with an average manure application reduction of (0.08 ± 0.07) gN·m−2·a−1, a nitrogen fertilizer application reduction of (1.07 ± 0.52) g·m−2·a−1 and a phosphate fertilizer application reduction of (0.23 ± 0.03) g·m−2·a−1. The strategies of increased manure application and decreased chemical fertilizer application (a reduction in at least one chemical fertilizer applied) were implemented in 29.1%, 22.1%, and 30.5% of the grid cells for maize, rice, and wheat, respectively (Figs. 5(e)-(g)). To better capture regional heterogeneity, we examined the distribution of optimal management strategies for maize, rice, and wheat across six agricultural regions in China (Tables S4-S6 in Appendix A), which revealed notable spatial differences. Southwest China exhibited a preference for the “reduced manure, phosphorus fertilizer and nitrogen fertilizer application, and LSU” strategy, adopted in 44.5% of the maize areas, 36.1% of the rice areas, and 51.9% of the wheat areas, indicating a notable tendency toward comprehensive input reduction (Tables S4-S6). In contrast, northeast and northwest China exhibited a preference for diverse and input-intensive strategies, such as the “decreased manure application, nitrogen fertilizer application, and LSU and increased phosphorus fertilizer application” strategy and the “increased manure, phosphorus fertilizer, and nitrogen fertilizer application and decreased LSU” strategy, respectively, reflecting local nutrient deficiencies and large-scale mechanized farming (Tables S4-S6). Central and east China exhibited a preference for mixed strategies in which phosphorus compensation was combined with reductions in other inputs; for example, the “decreased manure application, nitrogen fertilizer application, and LSU and increased phosphorus fertilizer application” strategy accounted for 17.6% of the maize areas and 22.9% of the wheat areas in central south China, Tables S4 and S6, respectively. These patterns highlight the need for crop- and region-specific strategies for managing MP risk and demonstrate the ability of the proposed model to support adaptive, spatially informed agricultural policies.
The government regularly establishes innovative policies or provides subsidies and incentives to encourage the widespread adoption and implementation of best practices, such as increasing the use of manure instead of chemical fertilizers in local areas [64]. In regions with abundant manure production, while decreasing the use of chemical fertilizer (e.g., in rice cultivation areas in south central and east China, as shown in Fig. 5(f)), it is essential to implement appropriate strategies through manure treatment systems (e.g., reducing the MP content) to reduce the risk of manure-based MPs to crops [65]. Notably, the utilization efficiency of manure from different sources varies (e.g., cattle manure exhibits a higher C:N ratio and lower total ammoniacal nitrogen concentration than those of pig, chicken, and duck manure [66]), and local adjustments in the livestock density and manure resource management could constitute feasible solutions for mitigating agricultural MP risk. Although the optimization of livestock density and fertilization regimes offers clear environmental benefits, the implementation of this strategy may face practical constraints [67]. In high-yield agricultural zones, such adjustments could lead to short-term income losses, thus deterring adoption by smallholder farmers. Nevertheless, pilot initiatives in provinces such as Jiangxi have demonstrated that integrated crop-livestock systems, supported by public investment in manure treatment infrastructure, can mitigate environmental risks without compromising productivity [65,68]. Moreover, the integration of MP-related objectives into national programs (e.g., knowledge-based nitrogen fertilizer management initiatives [69]) could increase the practical feasibility of these strategies in real agricultural settings. According to the FAO, alternatives to agricultural plastics are lacking, and strategies to reduce MP risk by limiting the use of agricultural plastics may be difficult to implement on a large scale [70]. Our proposed livestock and fertilizer management method provides feasible solutions for reducing MP risk, thereby accounting for the impact on crop yields. Initiatives such as targeted subsidies, demonstration zones, and technical extension services are critical for bridging the gap between scientific recommendations and field-level adoption [28,71]. Overall, it is necessary and important to prioritize the management of livestock density and manure/chemical fertilizer regimes to achieve a win-win outcome (potential risk control for MPs and increased crop production) and ensure sustainable agricultural development.
3.5. Limitations and recommendations
Most of the model input factors exhibited low linear correlations (Fig. S14 in Appendix A). Thus, model overfitting caused by multicollinearity was not an issue. The RF model, which yielded a high average prediction accuracy (test R2 = 0.69), was superior to the other machine learning models, such as the ANN (R2 = 0.47), SVM (R2 = 0.40), and XGBoost models (R2 = 0.67), as shown in Fig. 2(a). The permutation test results verified that the established RF models were not influenced by overfitting (intercept < 0.05) (Fig. S15 in Appendix A) [72], thus confirming the robustness of the models. Our study aimed to assess the risk of MP pollution in agricultural soils across various spatial scales. However, further consideration of the following uncertainties and limitations is needed. These uncertainties arise from the RF models and prediction data adopted. Monte Carlo simulations were employed to assess the relative uncertainty related to the prediction data and the unexplained variability not captured by the models (described in Section 2). The relative uncertainties in MP risk for maize, rice, and wheat were (3.47% ± 0.86%), (3.23% ± 1.17%), and (3.59% ± 1.03%), respectively (Fig. S16 in Appendix A). Although the average uncertainty in the predicted RQ values was low, the use of fixed thresholds (e.g., HC5 and HC10) could exacerbate classification-related uncertainty. Grid cells with RQ values near these threshold values are particularly sensitive, as small deviations resulting from model stochasticity or input variability may shift them between adjacent risk categories, thus potentially influencing land management decisions. Future studies should aim to investigate threshold sensitivity, particularly in contexts where classification-based policy frameworks or spatial decision-making processes are applied. In RF models, feature importance metrics such as an increase in MSE typically favor continuous or high-variance variables, which may overemphasize the influence of predictors such as livestock density. To reduce feature importance bias in RF models, three-way feature importance analysis was adopted, combining SHAP values, increases in MSE, and increases in node purity for more robust evaluation. In regions with limited data coverage (e.g., western China; Fig. 1(a)), model accuracy may be compromised by sampling heterogeneity (e.g., sampling density and depth) and by spatial aggregation at a 1 km resolution, which may obscure fine-scale environmental heterogeneity and introduce localized uncertainty in the predictions.
Although long-term or subsurface risks may occur, only MPs in the topsoil layer (0-30 cm) were considered, as this depth interval is both the most reported and most relevant to crop root zones [73]. Therefore, the potential vertical migration and accumulation of MPs in deeper soil layers were not incorporated into our modeling framework. In addition to implementation- and modeling-related uncertainties, optimizing livestock density and fertilizer use regimes may introduce trade-offs in nutrient cycling, emissions, and labor dynamics [74]. Although the use of the proposed optimization framework can successfully reduce the risk of MP pollution, the consideration of the potential for unintended secondary environmental consequences resulting from altered fertilizer regimes is essential. For example, increased application of chemical fertilizers may exacerbate nitrogen and phosphorus runoff, thereby causing eutrophication in surface water bodies [75]. Elevated nitrogen inputs can also increase nitrous oxide emissions [76], potentially undermining ongoing efforts to mitigate climate change. In addition, excessive manure application, especially without proper treatment, may introduce antibiotics, pathogens, and heavy metals into soils, potentially altering microbial resistance and influencing co-contaminant dynamics. Interactions between MPs and these pollutants could increase their environmental persistence and toxicity [77]. These complexities underscore the need for integrated systems approaches such as life cycle models that evaluate MP mitigation, nutrient management, and socioeconomic outcomes simultaneously under real-world conditions. Although the increased use of manure offers a promising strategy for reducing MP risk and advancing sustainable agriculture, its large-scale adoption may be constrained by several practical challenges. In particular, manure pretreatment technologies intended to remove embedded MPs remain at the early stages of development and lack standardization, and their removal efficiency has not yet been validated under diverse field conditions [78]. The absence of comprehensive and reliable data on treatment performance, implementation costs, and long-term agronomic impacts impedes thorough cost-benefit assessments [79]. Moreover, smallholder farmers may face structural barriers, such as limited access to treatment infrastructure, insufficient technical support, and uncertainty regarding the economic viability of such investments [80]. While these constraints do not negate the feasibility of the optimized strategies proposed via our model, they highlight the importance of implementing targeted subsidies, strengthening technical extension services, and conducting further empirical investigations. Our results, therefore, provide a foundational framework that can be progressively improved as technical and empirical gaps are addressed.
The projected increase in population is expected to increase the food demand [81], with production increases potentially prioritized while MP accumulation is neglected, which may restrict sustainable agricultural development. To implement the optimized management practices proposed, we recommend actions from three perspectives: scientific research, farmer engagement, and government policies. First, more field research is needed to identify MP sources and better understand the complex processes and interactions among climate, soil, livestock, and agricultural management, as well as the corresponding relationships with MPs in crop areas. Improving the quality of data and expanding field research can help reduce the uncertainty in models. Second, effectively engaging farmers in policy implementation is challenging because of their limited understanding of MP pollution and their economic vulnerability. Future economic growth might negate these obstacles, similar to the effect of the Chinese governmental subsidy of approximately 106 USD·ha−1 in 2020 for rice to support the adoption of knowledge-based nitrogen fertilizer management practices [69]. Nonetheless, more attention is needed to address the environmental effects and risks of emerging pollutants such as MPs by improving management practices. Finally, implementing appropriate land policies and economic incentives is equally important for the government. For example, the 15th Five-Year Plan for Controlling Plastic Pollution in China promotes the control of plastic pollution in farmland areas and the development of sustainable agriculture. Moreover, the AI-based framework developed in this study exhibits broad applicability beyond MP risk management. The framework can be extended to address other emerging agricultural challenges involving pollutant exposure and resource management. For example, identifying key environmental drivers and proposing targeted interventions on the basis of this framework could support data-driven decision-making for mitigating the effects of antibiotics, pesticide residues, or other persistent organic pollutants in agroecosystems.
4. Conclusions
This study focused on the widespread presence of MPs in agricultural soils and their potential risks to crops in China. Via the use of extensive field observations and RF models, we established intelligent agricultural models to identify and manage the risks of MPs to crops. The feature importance and interaction analysis results can be used to increase the understanding of the drivers of MP risk, such as climate factors, livestock density, and agricultural management. Our findings suggest that climate factors are highly related to livestock density and agricultural management practices, potentially exacerbating MP risk under future conditions. To increase the sustainability of agriculture within the context of climate change, we proposed optimized management practices by adjusting the livestock density and fertilizer application rate rather than reducing agricultural plastic use. When optimized management practices are applied, the results highlight the excellent potential for reducing the risk of MPs to crops without sacrificing yields. Our findings indicate the synergy of combining advanced data-driven techniques and novel management practices, which can increase our ability to balance MP risks to crops and yields. Together with advancing MP characterization techniques, accelerating the accumulation of MP data in farmland areas, and increasing the understanding of crop-MP interactions, our estimation and management methods for MP risks can be refined to promote sustainable agricultural development.
CHANGE HISTORY
A Correction to this paper has been published: https://doi.org
The Retraction Note was updated to reflect reception of statements of disagreement from all authors.
This article has been retracted.Please see the Retraction Notice for more detail: https://doi.org
ThompsonRC, OlsenY, MitchellRP, DavisA, RowlandSJ, JohnAWG, et al. Lost at sea: where is all the plastic? Science 2004; 304(5672):838.
[3]
TrainicM, FloresJM, PinkasI, PedrottiML, LombardF, BourdinG, et al. Airborne microplastic particles detected in the remote marine atmosphere. Commun Earth Environ2020; 1(1):64.
[4]
StrokalM, VriendP, BakMPP, KroezeC, van WijnenJ, van EmmerikT. River export of macro- and micro-plastics to seas by sources worldwide. Nat Commun2023; 14(1):4842.
[5]
HamdhaniH, EppehimerDE, KhusmiadiA, JailaniJ. The abundance of microplastics in the digestive system of silver barb (Barbonymus gonionotus) from the waters of the Karang Mumus River, Samarinda city, Indonesia. Water Conserv Manag2023; 8(2):179-84.
[6]
AkhtarS, PranayK, KumariK. Personal protective equipment and micro-nano plastics: a review of an unavoidable interrelation for a global well-being hazard. Hyg Environ Health Adv2023; 6:100055.
[7]
SalehiZ, HashemiSH, FluryM. Micro- and meso-plastics in farmlands with different irrigation water sources. Water Air Soil Pollut2023; 234(4):267.
[8]
TayyabM, KazmiSSUH, PastorinoP, SaqibHSA, YaseenZM, HanifMS, et al. Microplastics in agroecosystems: soil-plant dynamics and effective remediation approaches. Chemosphere2024; 362:142641.
[9]
WeithmannN, MöllerJN, LöderMGJ, PiehlS, LaforschC, FreitagR. Organic fertilizer as a vehicle for the entry of microplastic into the environment. Sci Adv2018; 4(4):eaap8060.
[10]
HeeYY, HanifNM, WestonK, LatifMT, SuratmanS, RusliMU, et al. Atmospheric microplastic transport and deposition to urban and pristine tropical locations in Southeast Asia. Sci Total Environ2023; 902:166153.
[11]
United Nations Environment Programme (UNEP). UNEP 2023 annual report:keeping the promise. Report. Nairobi: UNEP; 2023.
[12]
ZantisLJ, BorchiC, VijverMG, PeijnenburgW, Di LonardoS, BoskerT. Nano- and micro-plastics commonly cause adverse impacts on plants at environmentally relevant levels: a systematic review. Sci Total Environ2023; 867:161211.
[13]
GuoS, MuL, SunS, HouX, YaoM, HuX. Concurrence of microplastics and heat waves reduces rice yields and disturbs the agroecosystem nitrogen cycle. J Hazard Mater2023; 452:131340.
NizzettoL, LangaasS, FutterM. Pollution: do microplastics spill on to farm soils? Nature 2016; 537(7621):488.
[16]
LinJ, ChengQ, KumarA, ZhangW, YuZ, HuiD, et al. Effect of degradable microplastics, biochar and their coexistence on soil organic matter decomposition: a critical review. TrAC Trends Analyt Chem2025; 183:118082.
[17]
SinghS, ChakmaS, AlawaB, KalyanasundaramM, DiwanV. Microplastic pollution in terrestrial environment: identification, characterization, and risk assessment in Indore, Central India. Soil Use Manag2024; 40(2):e13053.
[18]
LozanoYM, Aguilar-TriguerosCA, OnandiaG, MaaßS, ZhaoT, RilligMC. Effects of microplastics and drought on soil ecosystem functions and multifunctionality. J Appl Ecol2021; 58(5):988-96.
[19]
KooiM, KoelmansAA. Simplifying microplastic via continuous probability distributions for size, shape, and density. Environ Sci Technol Lett2019; 6(9):551-7.
[20]
EveraertG, Van CauwenbergheL, De RijckeM, KoelmansAA, MeesJ, VandegehuchteM, et al. Risk assessment of microplastics in the ocean: modelling approach and first conclusions. Environ Pollut2018; 242:1930-8.
[21]
ZhaoF, YangL, YenH, FengQ, LiM, ChenL. Reducing risks of antibiotics to crop production requires land system intensification within thresholds. Nat Commun2023; 14(1):6094.
[22]
LongY, ZhangY, ZhouZ, LiuR, QiuZ, QiuY, et al. Are microplastics in livestock and poultry manure an emerging threat to agricultural soil safety? Environ Sci Pollut Res Int 2024; 31(8):11543-58.
[23]
IqbalS, XuJ, ArifMS, WorthyFR, JonesDL, KhanS, et al. Do added microplastics, native soil properties, and prevailing climatic conditions have consequences for carbon and nitrogen contents in soil? A global data synthesis of pot and greenhouse studies. Environ Sci Technol2024; 58 (19):8464-79.
[24]
LiHZ, ZhuD, LindhardtJH, LinSM, KeX, CuiL. Long-term fertilization history alters effects of microplastics on soil properties, microbial communities, and functions in diverse farmland ecosystem. Environ Sci Technol2021; 55 (8):4658-68.
[25]
HuQ, ZhaoY, HuX, QiJ, SuoL, PanY, et al. Effect of saline land reclamation by constructing the ‘‘raised field-shallow trench” pattern on agroecosystems in Yellow River Delta. Agric Water Manag2022; 261:107345.
[26]
RenS, WangK, ZhangJ, LiJ, ZhangH, QiR, et al. Potential sources and occurrence of macro-plastics and microplastics pollution in farmland soils: a typical case of China. Crit Rev Environ Sci Technol2024; 54(7):533-56.
[27]
ZhuH, LinC, LiuG, WangD, QinS, LiA, et al. Intelligent agriculture: deep learning in UAV-based remote sensing imagery for crop diseases and pests detection. Front Plant Sci2024; 15:1435016.
[28]
XiaoL, WangG, WangE, LiuS, ChangJ, ZhangP, et al. Spatiotemporal co- optimization of agricultural management practices towards climate-smart crop production. Nat Food2024; 5(1):59-71.
[29]
LiH, TaoY, XuT, WangH, YangM, ChenY, et al. Real-time quantification of activated sludge concentration and viscosity through deep learning of microscopic images. Environ Sci Ecotechnol2025; 24:100527.
[30]
ChuY, ZhangX, TangX, JiangL, HeR. Uncovering anaerobic oxidation of methane and active microorganisms in landfills by using stable isotope probing. Environ Res2025; 271:121139.
[31]
OlawadeDB, WadaOZ, IgeAO, EgbewoleBI, OlojoA, OladapoBI. Artificial intelligence in environmental monitoring: advancements, challenges, and future directions. Hyg Environ Health Adv2024; 12:100114.
[32]
LuoJ, ZhaoC, ChenQ, LiG. Using deep belief network to construct the agricultural information system based on Internet of Things. J Supercomput2022; 78(1):379-405.
[33]
KroppI, NejadhashemiAP, DebK, AboualiM, RoyPC, AdhikariU, et al. A multi- objective approach to water and nutrient efficiency for sustainable agricultural intensification. Agr Syst2019; 173:289-302.
[34]
KoelmansAA, Redondo-HasselerharmPE, Mohamed NorNH, KooiM. Solving the nonalignment of methods and approaches used in microplastic research to consistently characterize risk. Environ Sci Technol2020; 54(19):12307-15.
[35]
ChiaRW, LeeJY, KimH, JangJ. Microplastic pollution in soil and groundwater: a review. Environ Chem Lett2021; 19(6):4211-24.
[36]
WaldschlägerK, SchuttrumpfH. Effects of particle properties on the settling and rise velocities of microplastics in freshwater under laboratory conditions. Environ Sci Technol2019; 53(4):1958-66.
[37]
ZhaoS, ZhangZ, ChenL, CuiQ, CuiY, SongD, et al. Review on migration, transformation and ecological impacts of microplastics in soil. Appl Soil Ecol2022; 176:104486.
[38]
LiJ, SongY, CaiY. Focus topics on microplastics in soil: analytical methods, occurrence, transport, and ecological risks. Environ Pollut2020; 257:113570.
[39]
JinT, TangJ, LyuH, WangL, GillmoreAB, SchaefferSM. Activities of microplastics (MPs) in agricultural soil: a review of MPs pollution from the perspective of agricultural ecosystems. J Agric Food Chem2022; 70 (14):4182-201.
[40]
ZhangYQ, LykakiM, MarkiewiczM, AlrajoulaMT, KraasC, StolteS. Environmental contamination by microplastics originating from textiles: emission, transport, fate and toxicity. J Hazard Mater2022; 430:128453.
[41]
European chemicals agency (ECHA). ECHA adds three hazardous chemicals to the Candidate List [Internet]. Helsinki: ECHA; 2025 Jun 25 [cited 2025 Aug 20]. Available from: https://echa.europa.eu/.
[42]
KimSW, RilligMC. Research trends of microplastics in the soil environment: comprehensive screening of effects. Soil Ecol Lett2022; 4(2):109-18.
[43]
ChiaRW, LeeJY, JangJ, KimH, KwonKD. Soil health and microplastics: a review of the impacts of microplastic contamination on soil properties. J Soil Sediment2022; 22(10):2690-705.
[44]
HarmsIK, DiekötterT, TroegelS, LenzM. Amount, distribution and composition of large microplastics in typical agricultural soils in Northern Germany. Sci Total Environ2021; 758:143615.
[45]
HuangJ, ChenH, ZhengY, YangY, ZhangY, GaoB. Microplastic pollution in soils and groundwater: characteristics, analytical methods and impacts. Chem Eng J2021; 425:131870.
[46]
GulerHG, LarsenBE, QuintanaO, GoralKD, CarstensenS, ChristensenED, et al. Experimental study of non-buoyant microplastic transport beneath breaking irregular waves on a live sediment bed. Mar Pollut Bull2022; 181:113902.
[47]
GeorgiouK, JacksonRB, VinduskovaO, AbramoffRZ, AhlstromA, FengW, et al. Global stocks and capacity of mineral-associated soil organic carbon. Nat Commun2022; 13(1):3797.
[48]
DengP, GaoY, MuL, HuX, YuF, JiaY, et al. Development potential of nanoenabled agriculture projected using machine learning. Proc Natl Acad Sci USA2023; 120(25):2301885120.
[49]
QiuB, HuX, ChenC, TangZ, YangP, ZhuX, et al. Maps of cropping patterns in China during 2015-2021. Sci Data2022; 9(1):479.
[50]
KangJ, DingW, ChangN, YiX, ZhangJ, LiH. Optimized crop-livestock coupling to reduce agricultural manure-N surplus and greenhouse gas emissions in China. J Clean Prod2024; 467:142835.
[51]
DebK, PratapA, AgarwalS, MeyarivanT. A fast and elitist multiobjective genetic algorithm: NSGA-II. IEEE Trans Evol Comput2002; 6(2):182-97.
[52]
LiuQ, WangJ, BaiZ, MaL, OenemaO. Global animal production and nitrogen and phosphorus flows. Soil Res2017; 55(6):451.
[53]
SchielzethH. Simple means to improve the interpretability of regression coefficients. Methods Ecol Evol2010; 1(2):103-13.
[54]
MoiDA, Lansac-TôhaFM, RomeroGQ, Sobral-SouzaT, CardinaleBJ, KratinaP, et al. Human pressure drives biodiversity-multifunctionality relationships in large Neotropical wetlands. Nat Ecol Evol2022; 6(9):1279-89.
[55]
QiuG, WangQ, WangT, ZhangS, SongN, YangX, et al. Microplastic risk assessment and toxicity in plants: a review. Environ Chem Lett2023; 22(1):209-26.
[56]
CusworthSJ, DaviesWJ, McAinshMR, StevensCJ. A nationwide assessment of microplastic abundance in agricultural soils: the influence of plastic crop covers within the United Kingdom. Plants People Planet2023; 6(2):304-14.
[57]
AzeemI, AdeelM, AhmadMA, ShakoorN, ZainM, YousefN, et al. Microplastic and nanoplastic interactions with plant species: trends, meta-analysis, and perspectives. Environ Sci Technol Lett2022; 9(6):482-92.
[58]
IngraffiaR, AmatoG, IovinoM, RilligMC, GiambalvoD, FrendaAS. Polyester microplastic fibers in soil increase nitrogen loss via leaching and decrease plant biomass production and N uptake. Environ Res Lett2022; 17(5):054012.
[59]
MinM, PuHF, HeX, DengSY. Anti-seepage performance and oxygen barrier performance of the three-layered landfill cover system comprising neutralized slag under extreme climate conditions. Eng Geol2024; 342:107750.
[60]
Meizoso-RegueiraT, FuentesJ, CusworthSJ, RilligMC. Prediction of future microplastic accumulation in agricultural soils. Environ Pollut2024; 359:124587.
[61]
The UN Intergovernmental Panel on Climate Change (IPCC). Climate change 2021-the physical science basis. Report. Cambridge: Cambridge University Press; 2023.
[62]
United Nations Transforming our world: the 2030 agenda for sustainable development. Report. Agenda. New York City: United Nations; 2015.
[63]
RuggeroF, GoriR, LubelloC. Methodologies for microplastics recovery and identification in heterogeneous solid matrices: a review. J Polym Environ2020; 28(3):739-48.
[64]
ZhangT, HouY, MengT, MaY, TanM, ZhangF, et al. Replacing synthetic fertilizer by manure requires adjusted technology and incentives: a farm survey across China. Resour Conserv Recycl2021; 168:105301.
[65]
PanD, TangJ, ZhangL, HeM, KungCC. The impact of farm scale and technology characteristics on the adoption of sustainable manure management technologies: evidence from hog production in China. J Clean Prod2021; 280:124340.
[66]
KacprzakM, MalinskaK, GrosserA, Sobik-SzoltysekJ, WystalskaK, DrozdzD, et al. Cycles of carbon, nitrogen and phosphorus in poultry manure management technologies—environmental aspects. Crit Rev Environ Sci Technol2023; 53(8):914-38.
[67]
JiangC, WangY, YangZ, ZhaoY. Do adaptive policy adjustments deliver ecosystem-agriculture-economy co-benefits in land degradation neutrality efforts? Evidence from southeast coast of China. Environ Monit Assess2023; 195(10):1215.
[68]
XingJ, SongJ, LiuC, YangW, DuanH, YabarH, et al. Integrated crop-livestock- bioenergy system brings co-benefits and trade-offs in mitigating the environmental impacts of Chinese agriculture. Nat Food2022; 3(12):1052-64.
[69]
CaiS, ZhaoX, PittelkowCM, FanM, ZhangX, YanX. Optimal nitrogen rate strategy for sustainable rice production in China. Nature2023; 615(7950):73-9.
[70]
United Nations Food and Agriculture Organization. Assessment of agricultural plastics and their sustainability: a call for action. Report. Rome: United Nations Food and Agriculture Organization; 2021.
[71]
WangJ, ZhangL, LiuK, ZhouB, GaoH, HanX, et al. Minimizing the potential risk of soil nitrogen loss through optimal fertilization practices in intensive agroecosystems. Agron Sustain Dev2025; 45(1):9.
[72]
YuF, WeiC, DengP, PengT, HuX. Deep exploration of random forest model boosts the interpretability of machine learning studies of complicated immune responses and lung burden of nanoparticles. Sci Adv2021; 7(22):eabf4130.
[73]
YanP, ZhangS, XingH, YanS, NiuX, WangJ, et al. Spatial distribution of microplastics in Mollisols of the farmland in Northeast China: the role of field management and plastic sources. Geoderma2025; 459:117367.
[74]
VingerhoetsR, SpillerM, De BackerJ, AdriaensA, VlaeminckSE, MeersE, et al. Detailed nitrogen and phosphorus flow analysis, nutrient use efficiency and circularity in the agri-food system of a livestock-intensive region. J Clean Prod2023; 410:137278.
[75]
LiS, WuX, SongX, LiuX, GaoH, LiangG, et al. Long-term nitrogen fertilization enhances crop yield potential in no-tillage systems through enhancing soil fertility. Resour Conserv Recycl2024; 206:107622.
[76]
LiangY, XueL, JiaP, ZhangS, HuY, ZamanianK, et al. Fertilization management and greenhouse gases emissions from paddy fields in China: a meta-analysis. Field Crops Res2024; 315:109490.
[77]
ZhangG, RenR, YanX, ZhangH, ZhuY. Effects of microplastics on dissipation of oxytetracycline and its relevant resistance genes in soil without and with Serratia marcescens: comparison between biodegradable and conventional microplastics. Ecotoxicol Environ Saf2024; 287:117235.
[78]
RafiqA, XuJL. Microplastics in waste management systems: a review of analytical methods, challenges and prospects. Waste Manag2023; 171:54-70.
[79]
TanT, ZhangZ, HuangZ, MaC, LiuG, HuheT, et al. Evaluating the nutrient and pollutant flows of the Chinese livestock manure management system from 1949 to 2050. Resour Conserv Recycl2025; 215:108092.
[80]
WangY, YanJ, WuY. Impact of policy measures on smallholders’ livelihood resilience: evidence from Hehuang Valley. Tibetan Plateau Ecol Indic2024; 158:111351.
[81]
GattoA, ChepelievM. Global food loss and waste estimates show increasing nutritional and environmental pressures. Nat Food2024; 5(2):136-47.