A method for predicting the release amount of microplastics based on mulch aging characteristics and soil physicochemical properties

Abstract

The application discloses a kind of microplastic release amount prediction methods based on mulch aging characteristics and soil physicochemical properties, belong to environmental science and pollution control technical field.The method is by collecting mulch material type, oxygen-carbon ratio, carbon content, haze, light transmittance and soil pH, conductivity, total iron, total manganese, organic matter and so on multiple input variables, with microplastic release abundance as output variable, the nonlinear mapping relationship between input and output is established using random forest regression model, realize the high-precision prediction of microplastic release abundance.The test set is verified, and the model determination coefficient R² can reach more than 0.86.The application can also identify key control factors through feature importance and SHAP analysis, has the advantages of high prediction accuracy, strong interpretability, can handle multi-factor nonlinear coupling relationship, can provide effective tools for agricultural mulch source microplastic formation end evaluation, parameter extraction and environmental management.

Description

Technical Field

[0001] This invention relates to the fields of environmental science and pollution control technology, and in particular to a method for predicting the abundance of microplastic release from agricultural mulch film. Background Technology

[0002] Agricultural mulching technology plays a crucial role in increasing yields and ensuring profits in arid-region agriculture and is widely used. However, commonly used polyethylene (PE) mulch films and (biodegradable) materials such as polybutylene adipate / terephthalate (PBAT) are subject to aging and breakage under the influence of various environmental factors such as sunlight, temperature, and mechanical disturbance during long-term use, which can lead to the formation of mulch film-derived microplastics, posing a potential pollution risk to the soil environment.

[0003] Current research on microplastic pollution in farmland mainly focuses on content surveys, speciation descriptions, or single-factor impact analyses. However, the formation and release of microplastics is a complex process involving multiple coupled factors, influenced by membrane material properties, aging levels, and soil physicochemical properties, with significant nonlinear relationships among these factors. Traditional methods such as empirical judgment, single-factor analysis, or simple linear regression are insufficient to accurately characterize this complex relationship and effectively identify key controlling factors, resulting in low prediction accuracy and poor generalization ability.

[0004] Therefore, there is an urgent need for a method to predict microplastic release that can integrate multi-source data, handle nonlinear coupling relationships, and combine high-precision prediction with interpretability. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, this invention provides a method for predicting microplastic release based on the aging characteristics of plastic film and the physicochemical properties of soil.

[0006] The technical solution is as follows:

[0007] This invention provides a method for predicting microplastic release based on the aging characteristics of plastic film and the physicochemical properties of soil, comprising the following steps:

[0008] Step S1: Data Acquisition and Variable Determination

[0009] Data on agricultural mulch film and / or soil samples within the target area were collected to establish a model dataset. This dataset includes input and output variables. The input variables include at least: the mulch film's material type, carbonyl index, oxygen-to-carbon ratio, carbon content, haze, and transmittance; and the soil's pH, electrical conductivity, total iron content, total manganese content, and organic matter content. The output variable is the abundance of microplastic release. The carbonyl index is the ratio of carbonyl (C=O) absorbance to methylene (–CH2) absorbance under Fourier transform infrared spectroscopy.

[0010] Step S2: Data Preprocessing

[0011] The collected data is preprocessed, including at least missing value handling, outlier handling, and categorical variable encoding, and the dataset is divided into training set and test set.

[0012] Step S3: Construct the prediction model

[0013] The preprocessed training set data is imported into the random forest regression model for training, and a microplastic release abundance prediction model is established with the input variable as the independent variable and the output variable as the dependent variable.

[0014] Step S4: Model Training and Validation

[0015] The random forest model was trained using a training set, and its stability and generalization ability were evaluated using cross-validation.

[0016] Step S5: Output the prediction results

[0017] Input the input variables of the sample to be predicted into the trained random forest model, and output the corresponding predicted value of microplastic release abundance.

[0018] In a preferred embodiment of the present invention, the method further includes step S6: key feature identification and model interpretation. This step, based on feature importance analysis and / or SHAP (Shapley Additive exPlanations) interpretation methods, quantitatively analyzes the contribution of each input variable to the prediction results to identify key controlling factors affecting the abundance of microplastic release.

[0019] In a preferred embodiment of the present invention, the hyperparameters of the random forest model are tuned using a grid search method and are set as follows: bootstrap=False, max_depth=None, max_features='sqrt', min_samples_leaf=1, min_samples_split=10, n_estimators=500.

[0020] In a preferred embodiment of the present invention, the material type is polyethylene (PE) or polybutylene adipate / terephthalate (PBAT).

[0021] In a preferred embodiment of the present invention, the evaluation index of the model includes the coefficient of determination (R²). 2 One or more of the following: mean square error (MSE), root mean square error (RMSE), and mean absolute error (MAE).

[0022] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0023] High-precision prediction capability: This invention constructs a multivariate input variable system including mulch film properties, aging characteristics, and soil physicochemical properties, and utilizes the random forest algorithm to handle complex nonlinear relationships, achieving high-precision prediction of microplastic release abundance. Validated with 559 samples (447 training samples and 112 testing samples), the model achieved a coefficient of determination (R²) of 0.8655, a root mean square error (RMSE) of 50.68, a mean absolute error (MAE) of 31.73, and a best cross-validation R² score of 0.8482, indicating excellent prediction accuracy and generalization ability.

[0024] Key factor identification capability: This invention can not only make predictions, but also effectively identify the key controlling factors affecting microplastic release through feature importance analysis. The analysis shows that carbonyl index (importance 0.2863), transmittance (0.2480), and haze (0.1882) are the three most influential driving factors, with a total contribution of over 70%, providing a clear direction for subsequent mechanism research and management decisions.

[0025] The model boasts strong interpretability: This invention, combined with the SHAP interpretation method, can reveal the direction, intensity, and nonlinear characteristics of the key variables. For example, SHAP analysis shows that the carbonyl index has the highest positive impact on the model output, while the effects of transmittance and haze exhibit significant nonlinearity. Different membrane material types (PE vs PBAT) have a significant differentiating effect on microplastic release behavior, enhancing the model's credibility and transferability in practical applications.

[0026] Handling complex relationships: This invention uses a random forest model, which, through multi-tree ensemble and optimized hyperparameter combination (after 10,800 fitting optimizations), can effectively handle the nonlinear coupling and interaction between multiple factors, avoid the limitations of traditional linear models, and has higher predictive robustness under complex environmental conditions. Attached Figure Description

[0027] Figure 1 This is a flowchart illustrating the microplastic release prediction method based on the aging characteristics of plastic film and the physicochemical properties of soil according to the present invention.

[0028] Figure 2 This is a bar chart showing the feature importance of the random forest model in Embodiment 1 of the present invention;

[0029] Figure 3 This is a scatter plot of the predicted abundance and actual abundance of the model on the test set in Embodiment 1 of the present invention;

[0030] Figure 4This is a SHAP dependency graph of the "transmittance" variable in Embodiment 2 of the present invention;

[0031] Figure 5 This is the SHAP dependency graph of the "haze" variable in Embodiment 2 of the present invention;

[0032] Figure 6 This is a SHAP dependency graph of the "carbonyl index" variable in Example 2 of the present invention;

[0033] Figure 7 This is the SHAP beehive graph (global feature importance ranking) of each input variable in Embodiment 2 of the present invention.

[0034] Figure 8 This section compares the performance of the random forest model with other traditional models (support vector machine and linear regression model). Detailed Implementation

[0035] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0036] The terms used in the technical solution of this invention are defined as follows:

[0037] Haze:

[0038] Mulch film haze refers to the degree of light scattering caused by opacity or small particulate matter (such as water droplets, dust, coatings, etc.) on or inside the mulch film.

[0039] Measurement standard: Measured using a transmittance meter or haze meter, unit is %.

[0040] Light transmittance:

[0041] Definition: The light transmittance of mulch film refers to the ratio of the amount of light passing through the mulch film to the amount of incident light, reflecting the transparency of the mulch film.

[0042] Measurement standard: The light transmittance of plastic film is usually measured using a light transmittance meter, and the unit is %.

[0043] Carbonyl index:

[0044] Definition: The ratio of the absorbance of the carbonyl group (C=O) to the absorbance of the methylene group (–CH2).

[0045] Measurement standards: Absorbance at wavelengths measured by Fourier transform infrared spectrometer (AC=O / A–CH2): PE: A1715cm-1 / A1460cm-1; PBAT: A1754cm-1 / A1452cm-1.

[0046] Carbon content:

[0047] Definition: Percentage of carbon content on the surface of the plastic film.

[0048] Measurement standard: Elemental analyzer was used for measurement, and the unit is %.

[0049] Oxygen-to-carbon ratio:

[0050] Definition: The ratio of O to C elements in the mulch film.

[0051] Measurement standard: X-ray photoelectron spectrometer.

[0052] The flowchart of the microplastic release prediction method based on the aging characteristics of plastic film and soil physicochemical properties of this invention is shown below. Figure 1 As shown.

[0053] Example 1: Constructing a random forest prediction model for microplastic release abundance based on 12 input variables

[0054] This embodiment provides a method for constructing a random forest prediction model for microplastic release abundance based on 12 input variables.

[0055] 1. Data Sources and Dataset Construction

[0056] Data related to the abundance of microplastic release from agricultural mulch film were collected, resulting in 559 valid samples. Each sample included 12 input variables and 1 output variable.

[0057] Input variables: Material - polyethylene, Material - poly(butylene adipate / terephthalate) (uniquely thermally coded), carbonyl index, oxygen-carbon ratio (%), carbon content (%), haze (%), transmittance (%), pH, electrical conductivity (dS / m), total soil iron (g / kg), total soil manganese (g / kg), organic matter (g / kg). Some typical sample data are shown in Table 1.

[0058] Table 1. Examples of Partial Sample Data

[0059] 1 0 0.167 0.187 86.5236 14.4 91.6 8.27 173.8333333 29.26 1.34 10.60 0 0 1 0.629 0.354 53.5908 92.13333 85.4 8.27 173.8333333 29.26 1.34 10.60 0 0 1 0.629 0.354 53.5908 92.13333 85.4 8.47 234 25.36 0.52 7.03 0 0 1 0.629 0.354 53.5908 92.13333 85.4 8.2 429.3333333 33.89 0.67 18.17 0 0 1 0.629 0.354 53.5908 92.13333 85.4 8.723333 755 23.25 0.50 15.56 0 0 1 0.629 0.354 53.5908 92.13333 85.4 8.11 951.3333333 20.12 0.59 22.39 0 1 0 0.051 0.674 83.964 30.875 90.225 8.27 173.8333333 29.26 1.34 10.60 0 1 0 0.061 0.674 83.964 28.53333 90.33333 8.47 234 25.36 0.52 7.03 0 1 0 0.120 0.674 83.964 40.36667 88.1 8.2 429.3333333 33.89 0.67 18.17 0 1 0 0.174 0.674 83.964 52.875 85.375 8.723333 755 23.25 0.50 15.56 0 1 0 0.230 0.674 83.964 36.56667 88.3 8.11 951.3333333 20.12 0.59 22.39 0 0 1 0.662 1.120 46.09 95.05 79.55 8.27 173.8333333 29.26 1.34 10.60 0

[0060] Output variable: Microplastic release abundance (microplastics / g mulch film).

[0061] The sources of the sample data are as follows:

[0062] Provinces, regions and cities:

[0063] Shaanxi Province: Yan'an City, Yulin City

[0064] Gansu Province: Zhangye City, Jinchang City

[0065] Xinjiang: Shihezi City

[0066] These locations are in Northwest China, encompassing Shaanxi, Gansu, and Xinjiang provinces. These are all arid or semi-arid regions, with climates primarily classified as temperate arid and temperate semi-arid. This makes these areas suitable for studying the impacts of ecological and climatic conditions on crops, soils, and other organisms.

[0067] 2. Data partitioning and model training: The 559 samples were divided into a training set (447 sets) and a test set (112 sets) in an 8:2 ratio.

[0068] A random forest regression model was used for training, and hyperparameter optimization was performed using 10-fold cross-validation combined with grid search. A total of 1080 candidate parameter combinations were searched, and the total number of fitting iterations was 10800.

[0069] The final optimal combination of hyperparameters is:

[0070] bootstrap = False

[0071] max_depth = None

[0072] max_features = 'sqrt'

[0073] min_samples_leaf = 1

[0074] min_samples_split = 10

[0075] n_estimators = 500

[0076] 3. The model evaluation results were performed on the test set, and the results are as follows:

[0077] Mean squared error (MSE) = 2568.01

[0078] Root mean square error (RMSE) = 50.68

[0079] Mean Absolute Error (MAE) = 31.73

[0080] Coefficient of determination (R) 2 = 0.8655

[0081] The best R-value for 10-fold cross-validation 2 The score is 0.8482.

[0082] Figure 3 The scatter plot shows the predicted abundance and actual abundance of the model on the test set. It can be seen that the data points are basically evenly distributed near the diagonal of y=x, which further verifies the high prediction accuracy and generalization ability of the model.

[0083] 4. Feature Importance Analysis Results: Feature importance analysis was performed on the trained model. The importance scores (after normalization) of each variable are ranked as follows:

[0084] 1. Carbonyl index: 0.2863

[0085] 2. Light transmittance: 0.2480

[0086] 3. Haze: 0.1882

[0087] 4. Carbon content: 0.0792

[0088] 5. Oxygen-to-carbon ratio: 0.0390

[0089] 6. Material - PBAT: 0.0309

[0090] 7. Conductivity: 0.0298

[0091] 8. Material - PE: 0.0274

[0092] 9. Total iron in soil: 0.0223

[0093] 10. Organic matter: 0.0220

[0094] 11. pH: 0.0169

[0095] 12. Total manganese in soil: 0.0101 Figure 2 The importance of each variable is presented in the form of a bar chart, which visually shows that carbonyl index, transmittance, and haze are the three most contributing characteristics, with a cumulative importance exceeding 70%. These results indicate that the microplastic release process is closely related to the aging state and optical / structural properties of the mulch film.

[0096] Example 2: Interpretive Analysis of Random Forest Model Based on SHAP Method

[0097] Based on the random forest model obtained in Example 1, the SHAP (Shapley Additive ex Planations) method is further used to interpret and analyze the model in order to obtain explanations of the importance of global and local features.

[0098] Global explanation: Figure 7 The SHAP bee colony diagram summarizes the direction and intensity of the influence of each feature on the model output across all samples. Analysis results show that:

[0099] The carbonyl index has the strongest positive influence on the model output (SHAP value has the widest distribution range).

[0100] Both transmittance and haze contribute significantly to the prediction results;

[0101] Material type (PE / PBAT), carbon content, electrical conductivity, and organic matter also have a certain impact on model output;

[0102] The influence of each characteristic on the abundance of microplastic release exhibits a nonlinear relationship.

[0103] Local dependency analysis:

[0104] Combination Figure 6 The SHAP dependency plot shows that the carbonyl index and SHAP value (i.e., the positive and negative contributions to the predicted abundance) have a complex nonlinear relationship, with a significant positive contribution in a specific range.

[0105] Figure 4 The SHAP dependency plot shows that the effect of transmittance on the abundance of microplastic release has a significant threshold effect and nonlinear characteristics.

[0106] Figure 5 The SHAP dependency graph reveals the nonlinear driving effect of haze on the prediction results.

[0107] This embodiment demonstrates that the method of the present invention has good interpretability, can reveal the direction, intensity and sensitive range of the effects of key variables, and improves the credibility and transferability of the model in practical applications.

[0108] Example 3: Predictive applications of the model in different scenarios

[0109] This embodiment demonstrates the application of the model constructed in Example 1 to predict the abundance of microplastic release under different membrane material types, different aging states, and different soil physicochemical background conditions.

[0110] By inputting different combinations of sample features, the model can output corresponding predicted values ​​for microplastic release abundance. For example:

[0111] For PE material, under the characteristic combination of carbonyl index 0.222, carbon content 83.96%, haze 43.08%, light transmittance 87.1%, pH 8.11, electrical conductivity 951.3, and organic matter 22.39, the model predicts an abundance of 530 particles / g.

[0112] For PBAT material, under the characteristic combination of carbonyl index 0.612, carbon content 46.09%, haze 84.96%, transmittance 69.8%, pH 8.12, electrical conductivity 946.0, and organic matter 22.39, the model predicts an abundance of 260 particles / g.

[0113] This model can be used to compare the release potential of different film materials, different aging levels and different soil environmental conditions, providing data support and model foundation for the extraction of microplastic formation parameters from agricultural mulch film, regional differentiated management and full life cycle assessment.

[0114] Example 4: Comparative Example

[0115] like Figure 8 As shown, to verify the effectiveness of the hyperparameter-optimized random forest model in this invention, we use three control models for performance comparison: the unoptimized random forest model (Baseline RF), the support vector regression model (SVR), and the linear regression model (LR). The evaluation metrics for each model are as follows:

[0116] Baseline Random Forest Model:

[0117] R 2 0.81

[0118] RMSE: 64

[0119] MSE: 4066

[0120] MAE: 35

[0121] Support Vector Regression (SVR) model:

[0122] R 2 0.52

[0123] RMSE: 101

[0124] MSE: 10329

[0125] MAE: 57

[0126] Linear regression model (LR):

[0127] R 2 0.21

[0128] RMSE: 131

[0129] MSE: 17213

[0130] MAE: 99

[0131] As can be seen from the above evaluation metrics, Random Forest (RF) performed best in this study, exhibiting high R² and low RMSE, MSE, and MAE, indicating its ability to fit and predict microplastic abundance well. In contrast, Support Vector Regression (SVM) and Linear Regression (LR) models showed poor predictive performance, especially the Linear Regression model, whose R² was low. 2The value is low, and the error is large.

[0132] Climate zone:

[0133] Yan'an and Yulin in Shaanxi Province have a temperate semi-arid climate with low annual precipitation, mainly relying on winter rainfall and localized heavy rains in spring and summer to replenish water.

[0134] Zhangye and Jinchang in Gansu Province have a typical temperate arid climate. This region has extremely low annual precipitation, a dry climate, and large temperature differences between day and night, making it suitable for research on drought-tolerant plants or soils.

[0135] Shihezi City in Xinjiang also belongs to the temperate arid climate zone, a typical desert climate zone with very little precipitation, abundant sunshine, and often accompanied by strong sandstorms.

[0136] Data collection time span:

[0137] The data collection period spans from July 2025 to February 2026, covering the entire annual climate change cycle from summer to winter. This period is helpful for studying the impacts of seasonal climate change on ecosystems, crop growth, and soil environments.

Claims

1. A method for predicting microplastic release based on the aging characteristics of plastic film and soil physicochemical properties, characterized in that, Includes the following steps: Step S1: Collect relevant data on agricultural mulch film and / or soil samples within the target area to establish a model dataset; the dataset includes input variables and output variables; Step S2: Preprocess the collected data. The preprocessing includes at least missing value handling, outlier handling, and categorical variable encoding, and divides the dataset into training set and test set. Step S3: Import the preprocessed training set data into the random forest regression model for training, and establish a microplastic release abundance prediction model with the input variable as the independent variable and the output variable as the dependent variable; Step S4: Train the random forest model using the training set, and evaluate the stability and generalization ability of the model using cross-validation. Step S5: Input the input variables of the sample to be predicted into the trained random forest model, and output the corresponding predicted value of microplastic release abundance; The input variables include at least: the material type of the mulch film, carbonyl index, oxygen-carbon ratio, carbon content, haze, light transmittance, and soil pH, electrical conductivity, total iron content, total manganese content, and organic matter content; the output variable is the abundance of microplastic release.

2. The method for predicting microplastic release based on aging characteristics of plastic film and soil physicochemical properties according to claim 1, characterized in that, It also includes step S6: based on feature importance analysis and / or SHAP interpretation methods, quantitative analysis is performed on the contribution of each input variable to the prediction results in order to identify the key controlling factors affecting the abundance of microplastic release.

3. The method for predicting microplastic release based on aging characteristics of plastic film and soil physicochemical properties according to claim 1, characterized in that, The hyperparameters of the random forest model are set as follows: bootstrap=False, max_depth=None, max_features='sqrt', min_samples_leaf=1, min_samples_split=10, n_estimators=500.

4. The method for predicting microplastic release based on aging characteristics of plastic film and soil physicochemical properties according to claim 1, characterized in that, The material type is polyethylene or polybutylene adipate / terephthalate.

5. The method for predicting microplastic release based on aging characteristics of plastic film and soil physicochemical properties according to claim 1, characterized in that, The evaluation metrics for the model include the coefficient of determination R. 2 One or more of the following: mean square error (MSE), root mean square error (RMSE), and mean absolute error (MAE).

6. The method for predicting microplastic release based on aging characteristics of plastic film and soil physicochemical properties according to claim 1, characterized in that, In step S2, the dataset is divided into a training set and a test set in an 8:2 ratio, and one-hot encoding is used for the categorical variables.

7. The method for predicting microplastic release based on aging characteristics of plastic film and soil physicochemical properties according to claim 1, characterized in that, The unit for the microplastic release abundance is cells / g of mulch film.

Images