Method for predicting soil acidification sensitivity based on regional big data

Abstract

The invention belongs to the field of acidified soil improvement and soil health, and provides a method for predicting soil acidification sensitivity based on regional big data, and the method comprises the steps: obtaining to-be-detected data and predicting a soil acidification sensitivity prediction model; the training process of the soil acidification sensitivity prediction model comprises the steps of basic data set construction, data set preprocessing, data set division and model training and screening. According to the method, the machine learning model is introduced, so that the defects of a traditional method for measuring the acidification sensitivity by adopting acid-base continuous titration in a laboratory are overcome, and rapid and accurate prediction of the acidification sensitivity of batch soil sample points under the regional scale is realized; the method can be effectively applied to research on soil acidification, determination and mechanism research on soil acidification sensitivity, soil health evaluation and the like.

Description

Technical Field

[0001] The present invention relates to the field of acidified soil improvement and soil health, and in particular to a method for predicting soil acidification sensitivity based on regional big data. Background Art

[0002] Appropriate soil pH is a crucial prerequisite for sustainable agricultural development, and acidification has become a significant obstacle to soil degradation. Numerous studies have shown that when soil pH is below 5.5, crop diseases are more frequent, leading to reduced yields and incomes. For example, tobacco, a cash crop, is susceptible to root and stem diseases such as wilt when soil pH is below 5.5. Once these diseases develop, they are difficult to control with chemical pesticides. Remediation of acidified soils involves two aspects: first, determining the extent of soil acidification, as indicated by soil pH. For example, a pH below 6.5 is considered acidic, and below 4.5 is considered strongly acidic. Second, understanding which soils are susceptible to acidification and which are less susceptible is crucial for preventing and remediating soil acidification. Soil acidification sensitivity, often mentioned in the literature, refers to the amount of acid required to change a given mass of soil by one pH unit. This indicator provides a scientific measure of a soil's vulnerability to acidification. Soil acidification sensitivity is also known as soil acidification sensitivity. Currently, the continuous titration method is used, that is, the change process of soil acidity is measured by neutralization titration with an alkaline solution. The alkali used can be one of sodium hydroxide, calcium hydroxide or potassium hydroxide. Continuous titration requires continuous recording of the amount of alkaline solution used and the measured soil pH value. Based on these one-to-one values, a titration curve between the amount of titration base added and the soil pH is established, that is, the soil acid buffer curve. Then, according to the changing characteristics of the curve, its inflection point is found, and the soil acidification sensitivity is obtained.

[0003] This field measurement method requires measuring soil samples one by one, and each soil sample takes at least 72 hours to complete. This method is time-consuming, costly, and time-consuming, resulting in extremely low measurement efficiency. Literature statistics show that the minimum number of soil samples in a single article is only 2, and the maximum is no more than 52, indicating that measuring soil acidification sensitivity is very difficult. Soil acidification sensitivity indicators are important, but the time required to conduct regional soil surveys with massive amounts of soil samples is extremely long. Previous literature reports have used multiple linear regression equations based on the physical and chemical properties of historical soil samples for prediction. This method is relatively simple, but due to the complex nature of soils and the large number of soil types, the inherent mechanism of spatial variation in soil acidification sensitivity is nonlinear and difficult to describe quantitatively. Therefore, for slightly larger areas, such as prefecture-level cities, there are still problems such as low prediction accuracy and difficulty in generalizing prediction results. Summary of the Invention

[0004] In order to overcome the shortcomings of the existing technology, the purpose of the present invention is to provide a method for predicting soil acidification sensitivity based on regional big data, so as to solve the defect of low efficiency of traditional laboratory acid-base titration curve fitting method.

[0005] To achieve the above object, the present invention provides the following solutions:

[0006] A method for predicting soil acidification sensitivity based on regional big data, comprising:

[0007] Obtain the soil type and several preset soil physical and chemical properties of the soil sample points for which acidification sensitivity is to be predicted;

[0008] The soil type and all the soil physical and chemical properties are input into a pre-trained soil acidification sensitivity prediction model to perform prediction to obtain an acidification sensitivity prediction value. The training process of the soil acidification sensitivity prediction model is as follows:

[0009] Collect and organize existing acidic soil data, and establish an acidic soil acidification sensitivity database based on the acidic soil data to obtain a basic data set; the acidic soil data includes: acidification sensitivity values;

[0010] Preprocessing the basic data set to obtain a soil acidification sensitivity preprocessed data set;

[0011] Dividing the soil acidification sensitivity preprocessing dataset into a soil acidification sensitivity training set and a soil acidification sensitivity test set;

[0012] Several machine learning models are modeled and trained based on the soil acidification sensitivity training set, and each trained machine learning model is evaluated based on the soil acidification sensitivity test set to obtain a comprehensive model accuracy score, and the machine learning model with the highest comprehensive model accuracy score is output to obtain the soil acidification sensitivity prediction model; the machine learning models include: a random forest model, a support vector machine model, an extreme gradient boosting model, a neural network model, and a decision tree model.

[0013] Preferably, existing acidic soil data are collected and collated, and an acidic soil acidification sensitivity database is established based on the acidic soil data to obtain a basic data set, including:

[0014] Different types of acidic soil sample data of multiple target representative areas are collected, and environmental information of each target representative area is recorded. The acidic soil sample data and the environmental information are integrated and analyzed to obtain the basic data set.

[0015] Preferably, preprocessing the basic data set to obtain a soil acidification sensitivity preprocessed data set includes:

[0016] Performing data set cleaning on the basic data set to obtain a soil acidification sensitivity cleaned data set; the data set cleaning includes: filling missing values ​​of soil attributes, smoothing noise of soil attribute data, and deleting outlier soil samples;

[0017] Performing data set normalization on the soil acidification sensitivity cleaned data set to obtain a soil acidification sensitivity normalized data set; the data set normalization includes: normalization processing, discretization processing, and sparse processing; the normalization processing includes: standard deviation normalization, minimum-maximum normalization, and decimal scaling normalization;

[0018] The soil acidification sensitivity standard dataset is subjected to dataset dimensionality reduction to obtain the soil acidification sensitivity preprocessing dataset.

[0019] The importance analysis of the soil acidification sensitivity preprocessing dataset was performed using soil science expert scoring and principal component analysis methods to obtain the important values ​​between soil properties and acidification sensitivity. The soil properties corresponding to a preset number of the important values ​​were extracted in descending order to obtain the soil physical and chemical properties.

[0020] Preferably, the soil acidification sensitivity training set and the soil acidification sensitivity test set are divided by stratified random sampling.

[0021] Preferably, the method for determining the hyperparameters during the training of the machine learning model includes: a grid search method and a Bayesian search method; the method for optimizing the hyperparameters is cross-validation.

[0022] Preferably, the calculation formula for the comprehensive accuracy score of the model is as follows:

[0023] S=0.7*R 2 1+0.1*RMSE1+0.2*MAE1;

[0024] in,

[0025]

[0026] max() is the maximum value function; min() is the minimum value function; S is the comprehensive accuracy score of the model; R 2 is the coefficient of determination; RMSE is the root mean square error; MAE is the mean absolute error; R 2 1 is the normalized value of the determination coefficient; RMSE1 is the normalized value of the root mean square error; MAE1 is the normalized value of the mean absolute error; m is the number of soil sample points in the soil acidification sensitivity training set; with y iare the predicted value of acidification sensitivity and the numerical value of acidification sensitivity of the i-th soil sample point respectively; is the mean of the acidification sensitivity prediction values ​​of the soil acidification sensitivity training set.

[0027] Preferably, the soil types include: red soil, yellow soil, yellow-brown soil, brown soil, black soil, chernozem, purple soil, fluvo-aquic soil, meadow soil, white soil and paddy soil; the soil physical and chemical properties include: clay content, pH value, organic matter, cation exchange capacity and base saturation.

[0028] Preferably, the normalization methods of the coefficient of determination, the root mean square error and the mean absolute error include: deviation normalization and Z-score normalization.

[0029] The present invention discloses the following technical effects:

[0030] The present invention provides a method for predicting soil acidification sensitivity based on regional big data. By selecting multiple machine learning models for training and optimization, and comparing and comprehensively evaluating the acidification sensitivity prediction values ​​of different machine learning models, the present invention overcomes the low efficiency of traditional laboratory acid-base titration curve fitting methods, and realizes rapid and accurate prediction of the acidification sensitivity of batch soil samples at a regional scale. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0032] Figure 1 A schematic diagram of a process for predicting soil acidification sensitivity based on regional big data provided by an embodiment of the present invention;

[0033] Figure 2 Schematic diagram of fitting curves of different models provided in the embodiment of the present invention, Figure 2 (a) is a schematic diagram of the fitting curve of the random forest model. Figure 2 (b) is a schematic diagram of the support vector machine model fitting curve. Figure 2 (c) Schematic diagram of the fitting curve of the extreme gradient boosting model. DETAILED DESCRIPTION

[0034] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0035] The purpose of the present invention is to provide a method for predicting soil acidification sensitivity based on regional big data, so as to solve the defect of low efficiency of traditional laboratory acid-base titration curve fitting method.

[0036] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0037] Figure 1 A schematic diagram of a process for predicting soil acidification sensitivity (acidification sensitivity) based on regional big data provided by an embodiment of the present invention is shown in FIG. Figure 1 As shown, the present invention provides a method for predicting soil acidification sensitivity based on regional big data, comprising:

[0038] Step 100: Obtain the soil type and several preset soil physical and chemical properties of the soil sample point whose acidification sensitivity is to be predicted;

[0039] Step 200: Input the soil type and all the soil physical and chemical properties into a pre-trained soil acidification sensitivity prediction model to perform prediction to obtain an acidification sensitivity prediction value. The training process of the soil acidification sensitivity prediction model is as follows:

[0040] Step 201: collecting and organizing existing acidic soil data, and establishing an acidic soil acidification sensitivity database based on the acidic soil data to obtain a basic data set; the acidic soil data includes: acidification sensitivity values;

[0041] Step 202: preprocessing the basic data set to obtain a soil acidification sensitivity preprocessed data set;

[0042] Step 203: dividing the soil acidification sensitivity preprocessing dataset into a soil acidification sensitivity training set and a soil acidification sensitivity test set;

[0043] Step 204: Modeling and training are performed on several machine learning models according to the soil acidification sensitivity training set, and each trained machine learning model is evaluated according to the soil acidification sensitivity test set to obtain a comprehensive model accuracy score, and the machine learning model with the highest comprehensive model accuracy score is output to obtain the soil acidification sensitivity prediction model; the machine learning models include: a random forest model, a support vector machine model, an extreme gradient boosting model, a neural network model, and a decision tree model.

[0044] Preferably, existing acidic soil data are collected and collated, and an acidic soil acidification sensitivity database is established based on the acidic soil data to obtain a basic data set, including:

[0045] Different types of acidic soil sample data of multiple target representative areas are collected, and environmental information of each target representative area is recorded. The acidic soil sample data and the environmental information are integrated and analyzed to obtain the basic data set.

[0046] Specifically, the basic data set is preprocessed to obtain a soil acidification sensitivity preprocessed data set, including:

[0047] Performing data set cleaning on the basic data set to obtain a soil acidification sensitivity cleaned data set; the data set cleaning includes: filling missing values ​​of soil attributes, smoothing noise of soil attribute data, and deleting outlier soil samples;

[0048] Performing data set normalization on the soil acidification sensitivity cleaned data set to obtain a soil acidification sensitivity normalized data set; the data set normalization includes: normalization processing, discretization processing, and sparse processing; the normalization processing includes: standard deviation normalization, minimum-maximum normalization, and decimal scaling normalization;

[0049] The soil acidification sensitivity standard dataset is subjected to dataset dimensionality reduction to obtain the soil acidification sensitivity preprocessing dataset.

[0050] The importance analysis of the soil acidification sensitivity preprocessing dataset was performed using soil science expert scoring and principal component analysis methods to obtain the important values ​​between soil properties and acidification sensitivity. The soil properties corresponding to a preset number of the important values ​​were extracted in descending order to obtain the soil physical and chemical properties.

[0051] Furthermore, the soil acidification sensitivity training set and the soil acidification sensitivity test set are divided by stratified random sampling.

[0052] Preferably, the method for determining the hyperparameters during the training of the machine learning model includes: a grid search method and a Bayesian search method; the method for optimizing the hyperparameters is cross-validation.

[0053] Specifically, the calculation formula for the comprehensive accuracy score of the model is as follows:

[0054] S=0.7*R 2 1+0.1*RMSE1+0.2*MAE1;

[0055] in,

[0056]

[0057]

[0058] max() is the maximum value function; min() is the minimum value function; S is the comprehensive accuracy score of the model; R 2 is the coefficient of determination; RMSE is the root mean square error; MAE is the mean absolute error; R 2 1 is the normalized value of the determination coefficient; RMSE1 is the normalized value of the root mean square error; MAE1 is the normalized value of the mean absolute error; m is the number of soil sample points in the soil acidification sensitivity training set; with y i are the predicted value of acidification sensitivity and the numerical value of acidification sensitivity of the i-th soil sample point respectively; is the mean of the acidification sensitivity prediction values ​​of the soil acidification sensitivity training set.

[0059] Optionally, the soil types include: red soil, yellow soil, yellow-brown soil, brown soil, black soil, chernozem, purple soil, fluvo-aquic soil, meadow soil, white soil and paddy soil; the soil physical and chemical properties include: clay content, pH value, organic matter, cation exchange capacity and base saturation.

[0060] Preferably, the normalization methods of the coefficient of determination, the root mean square error and the mean absolute error include: deviation normalization and Z-score normalization.

[0061] Optionally, a large number of regionally representative acidic soil sample data should be collected and compiled, with detailed records of each sample's location, vegetation cover, topography, and other environmental information to ensure sample diversity and data reliability. Basic soil physical and chemical properties, including acidification sensitivity, should be measured in a specialized laboratory using national or industry-recognized standard methods. The results should be rigorously quality-controlled to ensure data accuracy and reliability. The data will be collated and analyzed to form the basic soil acidification sensitivity dataset.

[0062] Preferably, the soil acidification sensitivity basic dataset is cleaned to resolve inconsistencies in the soil attribute data, thereby obtaining a soil acidification sensitivity cleaned dataset. The soil dataset cleaning method includes filling missing values ​​in soil attributes, smoothing noise in soil attribute data, or removing outlier soil samples.

[0063] Furthermore, the soil acidification sensitivity preprocessing data set is divided by stratified random sampling to ensure that the sample distribution of each soil type in the soil acidification sensitivity training set and the soil acidification sensitivity test set is as consistent as possible with the sample distribution in the soil acidification sensitivity preprocessing data set; the division ratio of the training set and the test set is selected based on the number of soil samples.

[0064] Specifically, the coefficient of determination (R 2 ), root mean square error (RMSE) and mean absolute error (MAE) are used as evaluation indicators to comprehensively judge the performance and reliability of the machine learning model. The calculation formula of the evaluation indicators is as follows:

[0065]

[0066] Preferably, normalization is performed on the three evaluation indicators to eliminate the dimensional differences between the different evaluation indicators and make them comparable. Deviation normalization and Z-score normalization are generally used for normalization.

[0067] Optionally, determine the weights of the evaluation indicators: determine the weights of each evaluation indicator based on the opinions of experts in the field of soil science and existing literature references to reflect its importance in the comprehensive score: the coefficient of determination (R 2 ) with a weight of 70%, the root mean square error (RMSE) with a weight of 10%, and the mean absolute error (MAE) with a weight of 20%. Calculate the weighted comprehensive accuracy score of the soil acidification sensitivity prediction model: multiply the normalized value of each evaluation indicator by its corresponding weight, and then add the results to obtain the comprehensive accuracy score of the soil acidification sensitivity prediction model. Mathematically expressed as:

[0068] S=0.7*R 2 1+0.1*RMSE1+0.2*MAE1

[0069] By comparing the S values ​​of each prediction model, we can intuitively understand and analyze the performance of the prediction model under different evaluation indicators, and directly select the optimal prediction model for soil acidification sensitivity.

[0070] Furthermore, the soil type of the unknown soil acidification sensitivity sample point and the five easily obtained soil physical and chemical properties selected above are obtained, and placed into the soil acidification sensitivity prediction model to obtain the predicted value of acidification sensitivity, thereby completing the calculation of acidification sensitivity.

[0071] Specifically, acidic soil data were screened from actual data collection and existing literature databases to construct an acidic soil acidification sensitivity database as a basic dataset for soil acidification sensitivity. To be included in the database, soil samples must have surface soil data (collected at depths of 0 to 10 cm, 0 to 15 cm, or 0 to 20 cm) and a pH value ≤ 6.5. In addition to acidification sensitivity, the database also includes 12 soil properties, including latitude and longitude, soil bulk density, mechanical composition, organic matter, available phosphorus, and available potassium.

[0072] Optionally, the operation of missing soil attribute values ​​can fill the missing attribute values ​​in the soil acidification sensitivity basic dataset with the median or mean of the valid features corresponding to the missing values, and delete duplicate data in the soil acidification sensitivity basic dataset.

[0073] Optionally, outlier soil sample points can be judged based on box plots, quantile judgments, and the 3σ principle. In the dataset cleaning stage, outlier soil sample points need to be considered as abnormal sample points that affect the quality of the soil dataset.

[0074] Furthermore, since soil type is a categorical variable, it was converted into a dummy variable for use in regression analysis. Since the ranges (dimensions) and degrees of variation of soil physical and chemical properties vary widely, the soil acidification sensitivity basic dataset was normalized and the soil properties were logarithmized based on actual conditions.

[0075] Preferably, dataset normalization is designed to address the problem of inconsistency in soil data and eliminate the impact of different soil indicators due to differences in properties, dimensions, and orders of magnitude, thereby reducing errors that may occur when soil data flows through machine learning models.

[0076] Specifically, studies have shown that soil acidification sensitivity is primarily influenced by carbonate (pH 6.2 to 8.6), silicate (pH > 5.0), cation exchange (pH 4.2 to 5.0), and aluminum-iron (pH < 4.2) buffer systems, with these buffer systems interacting with each other. Furthermore, oxygen-containing functional groups in soil organic matter and clay also play a role. Using a combination of soil science expert scoring and principal component analysis, the top five key soil properties were identified from a pool of 12 soil attributes. These properties include clay content, pH, organic matter (SOM), cation exchange capacity (CEC), and base saturation (BSP).

[0077] Optionally, the measured acidification sensitivity data needs to be determined using acid-base titration. Soil organic matter content is determined using the potassium dichromate volumetric method and external heating method according to standard GB 9834-1988; soil pH is determined using the electrode potentiometer; soil cation exchange capacity and total exchangeable base capacity (EBC) are determined using the ammonium acetate method and Kjeldahl nitrogen determination method, and base saturation is calculated; soil mechanical composition is determined using the pipette method or laser particle analyzer, and clay content is calculated. The calculation formula for base saturation is as follows:

[0078]

[0079] Where BSP is base saturation, EBC is the total amount of exchangeable bases, and CEC is the cation exchange capacity.

[0080] Preferably, based on the relevant achievements of soil acidification sensitivity in this field, combined with the guidance of experts in the field and the richness of data, the accuracy of the acidification sensitivity prediction model established based on effective characteristic variables will be greatly improved.

[0081] Specifically, the soil acidification sensitivity test set mainly verifies the generalization ability of the acidification sensitivity prediction model and then evaluates the predictive ability of the model.

[0082] Optionally, the acidification sensitivity prediction model may be a support vector machine, and the optimized hyperparameters include the selection of the kernel function, the Gamma value for controlling the influence spacing of the training points, and the penalty parameter C for controlling the tolerance of the balance control model to misclassification.

[0083] Furthermore, the soil acidification sensitivity test set is used to conduct a comprehensive performance evaluation of the optimized machine learning model with determined hyperparameters, and the machine learning model with the best evaluation result is used as the prediction method for the soil acidification sensitivity.

[0084] Specifically, in order to facilitate those skilled in the art to further understand the present invention, examples are given to objectively illustrate:

[0085] S1: All relevant literature was searched in CNKI and Web of Science using keywords such as “acidic soil,” “soilacidification,” “buffer performance,” or “buffer capacity.” The acidification sensitivity data of acidic soils located in the territory were extracted from the references of all relevant literature, along with relevant information including specific site information, soil types, and basic physical and chemical properties of the soil.

[0086] To be included in the Acidic Soil Acidification Sensitivity Database, soil samples must have surface soil data (collected at depths of 0 to 10 cm, 0 to 15 cm, or 0 to 20 cm) and a pH ≤ 6.5. In addition to acidification sensitivity, this includes 12 soil attributes, including geographic latitude and longitude, bulk density, mechanical composition, organic matter, available phosphorus, and available potassium. Using ArcGIS and Excel, this database was compiled, containing data on 1,056 valid soil samples, covering acidic soils across all major soil types in 13 provinces (municipalities) within China.

[0087] S2: Preprocess the acidic soil acidification sensitivity database to obtain a soil acidification sensitivity preprocessed dataset.

[0088] S21: Clean the acid soil acidification sensitivity database. This process includes filling in missing soil attribute values, smoothing out noise in the soil attribute data, and smoothing or deleting outlier soil samples. This results in a cleaned soil acidification sensitivity dataset, reducing the number of valid soil samples from 1056 to 376.

[0089] S22: performing data set normalization, discretization, and thinning processing on the obtained soil acidification sensitivity cleaned data set to obtain a soil acidification sensitivity normalized data set.

[0090] Among them, the soil type (categorical variable) in the soil acidification sensitivity cleanup dataset was assigned a value in the form of a dummy variable so that it can be used in subsequent regression analysis.

[0091] The soil acidification sensitivity cleaned dataset was normalized using standard deviation normalization, minimum-maximum normalization, and decimal scaling normalization to eliminate the effects of different properties, dimensions, and magnitudes among different soil indicators, thus obtaining a standardized soil acidification sensitivity dataset.

[0092] S23: Dimensionality reduction was performed on the soil acidification sensitivity standard dataset to obtain a preprocessed soil acidification sensitivity dataset. The top five key soil properties were selected from 12 soil attributes using a combination of soil science expert scoring and principal component analysis. These properties included clay content, pH, organic matter (SOM), cation exchange capacity (CEC), and base saturation percentage (BSP).

[0093] S3: Divide the soil acidification sensitivity preprocessing dataset into a soil acidification sensitivity training set and a soil acidification sensitivity test set, and identify the optimal prediction model from the machine learning model based on the soil acidification sensitivity training set and the soil acidification sensitivity test set to obtain a prediction method for soil acidification sensitivity.

[0094] S31: Divide the soil acidification sensitivity preprocessing dataset into a soil acidification sensitivity training set and a soil acidification sensitivity test set through stratified random sampling, ensuring that the sample point distribution of each soil type in the soil acidification sensitivity training set and the soil acidification sensitivity test set is as consistent as possible with the sample point distribution in the soil acidification sensitivity preprocessing dataset. In this example, due to the small number of soil samples in the soil acidification sensitivity preprocessing dataset, it is advisable to divide 70% of the soil acidification sensitivity preprocessing dataset into the soil acidification sensitivity training set and 30% into the soil acidification sensitivity test set.

[0095] S32: Train the machine learning model on the training set and optimize and determine the hyperparameters during the training process.

[0096] The machine learning models used in this example are random forest (RF), support vector machine (SVM), and extreme gradient boosting (XGBoost). Ten-fold cross-validation and grid search were used to optimize and determine the hyperparameters of each machine learning model. The optimal parameter combination was selected based on the minimum RMSE value in the 10-fold cross-validation.

[0097] A random forest model was trained on the soil acidification sensitivity training set. Ten-fold cross-validation and grid search were used to optimize and determine the final hyperparameters. The final hyperparameters for the random forest model after training and optimization were: number of decision trees (num.trees) = 250, number of variables extracted at each node (mtry) = 5, minimum number of nodes (min.node.size) = 1, and maximum depth of the decision tree (max.depth) = 15. All other hyperparameters were kept at their default values.

[0098] A support vector machine model was trained on the soil acidification sensitivity training set. Ten-fold cross-validation and grid search were used to optimize and determine the final hyperparameters. The final hyperparameters of the support vector machine model after training and optimization were: Gaussian radial basis function (RBF) kernel function, kernel coefficient (Gamma) = 0.325, penalty parameter (C) = 2.575. All other hyperparameters were kept at their default values.

[0099] The soil acidification sensitivity training set was trained using the extreme gradient boosting model. Ten-fold cross-validation and grid search were used to optimize and determine the final hyperparameters. The final hyperparameters of the extreme gradient boosting model after training were: learning rate (eta) = 0.33, post-pruning parameter (gamma) = 1.25, maximum decision tree depth (max_depth) = 4, maximum step size for each tree weight change (min_child_weight) = 3.386, sample size (subsample) = 0.885, feature column sampling (colsample_bytree) = 0.75, and number of decision trees (nrounds) = 49. All other hyperparameters were kept at their default values.

[0100] S33: Use the soil acidification sensitivity test set to predict the acidification sensitivity of the trained and optimized machine learning model, and then select the machine learning model with the best performance as the soil acidification sensitivity prediction method based on the evaluation indicators.

[0101] Among them, the evaluation indicators of machine learning models include the coefficient of determination (R 2 ), root mean square error (RMSE) and mean absolute error (MAE). In this example, after repeated training (fitting curve reference Figure 2 (a) to Figure 2 (c)), the support vector machine model has the best prediction performance (the weighted comprehensive score S is the largest), so the support vector machine model is selected as the final soil acidification sensitivity prediction system. Figure 2 (b) shows the coefficient of determination (R) between the predicted and measured values ​​of acidification sensitivity of the support vector machine model on the test set. 2 ) can reach 0.72, and the fitting effect is good in the entire range of acidification sensitivity. This shows that the prediction model of soil acidification sensitivity has good generalization performance on the test set, that is, it can have good prediction ability for soil samples with unknown acidification sensitivity. Therefore, it shows that the method of predicting acidification sensitivity based on machine learning according to the selected key soil physical and chemical properties is feasible and reliable and can withstand actual testing.

[0102] S4: When there is a sample point with unknown soil acidification sensitivity, it is only necessary to obtain the soil type of the sample point and the five soil attribute data information of clay content, soil pH, organic matter, cation exchange capacity and base saturation selected in the above steps. The acidification sensitivity corresponding to the soil sample point can be quickly calculated using the soil acidification sensitivity prediction method.

[0103] Advantageously, the method provided in this embodiment can meet the requirements for efficient and accurate prediction of soil acidification sensitivity at large data volumes and regional scales. The entire technical solution can be effectively applied to soil acidification research, soil acidification sensitivity measurement and mechanism research, and soil health assessment.

[0104] The beneficial effects of the present invention are as follows:

[0105] By introducing a machine learning model, the present invention supplements the shortcomings of the traditional laboratory method of measuring acidification sensitivity using continuous acid-base titration, and achieves rapid and accurate prediction of the acidification sensitivity of batch soil samples at a regional scale. It can be effectively applied to research on soil acidification, determination and mechanism research of soil acidification sensitivity, and soil health assessment.

[0106] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0107] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the method and core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims

1. A method for predicting soil acidification sensitivity based on regional big data, characterized in that: include: Obtain the soil type and several preset soil physical and chemical properties of the soil sample points whose acidification sensitivity is to be predicted; The soil type and all the soil physical and chemical properties are input into the pre-trained soil acidification sensitivity prediction model for prediction to obtain the acidification sensitivity prediction value; the training process of the soil acidification sensitivity prediction model is as follows: Collect and organize existing acidic soil data, and establish an acidic soil acidification sensitivity database based on the acidic soil data to obtain a basic data set; the acidic soil data includes: acidification sensitivity values; Preprocessing the basic data set to obtain a soil acidification sensitivity preprocessing data set; Dividing the soil acidification sensitivity preprocessing data set into a soil acidification sensitivity training set and a soil acidification sensitivity test set; Modeling and training are performed on several machine learning models according to the soil acidification sensitivity training set, and each of the trained machine learning models is evaluated according to the soil acidification sensitivity test set to obtain a comprehensive model accuracy score, and the machine learning model with the highest comprehensive model accuracy score is output to obtain the soil acidification sensitivity prediction model; the machine learning models include: a random forest model, a support vector machine model, an extreme gradient boosting model, a neural network model, and a decision tree model.

2. The method for predicting soil acidification sensitivity based on regional big data according to claim 1, characterized in that: Collect and organize existing acidic soil data, and establish an acidic soil acidification sensitivity database based on the acidic soil data to obtain a basic data set, including: Different types of acidic soil sample data of multiple target representative areas are collected, and the environmental information of each target representative area is recorded. The acidic soil sample data and the environmental information are integrated and analyzed to obtain the basic data set.

3. The method for predicting soil acidification sensitivity based on regional big data according to claim 1, characterized in that: The basic data set is preprocessed to obtain a soil acidification sensitivity preprocessed data set, including: Performing data set cleaning on the basic data set to obtain a soil acidification sensitivity cleaned data set; the data set cleaning includes: filling missing values ​​of soil attributes, smoothing noise of soil attribute data, and deleting outlier soil samples; Performing data set normalization on the soil acidification sensitivity cleaned data set to obtain a soil acidification sensitivity normalized data set; the data set normalization includes: normalization processing, discretization processing and sparse processing; the normalization processing includes: standard deviation normalization, minimum-maximum normalization and decimal calibration normalization; Performing data set dimensionality reduction on the soil acidification sensitivity standard data set to obtain the soil acidification sensitivity preprocessing data set; The importance analysis of the soil acidification sensitivity preprocessed data set was performed through expert scoring in the field of soil science and principal component analysis methods to obtain the important values ​​between soil properties and acidification sensitivity, and the soil properties corresponding to a preset number of the important values ​​were extracted in descending order to obtain the soil physical and chemical properties.

4. The method for predicting soil acidification sensitivity based on regional big data according to claim 1, characterized in that: The soil acidification sensitivity training set and the soil acidification sensitivity test set are divided by stratified random sampling.

5. The method for predicting soil acidification sensitivity based on regional big data according to claim 1, characterized in that: The method for determining the hyperparameters during the training of the machine learning model includes: a grid search method and a Bayesian search method; the method for optimizing the hyperparameters is cross-validation.

6. The method for predicting soil acidification sensitivity based on regional big data according to claim 1, characterized in that: The calculation formula for the comprehensive accuracy score of the model is as follows: S=0.7*R 2 1+0.1*RMSE1+0.2*MAE1; in, max() is the maximum value function; min() is the minimum value function; S is the comprehensive accuracy score of the model; R 2 is the coefficient of determination; RMSE is the root mean square error; MAE is the mean absolute error; R 2 1 is the normalized value of the determination coefficient; RMSE1 is the normalized value of the root mean square error; MAE1 is the normalized value of the mean absolute error; m is the number of soil sample points in the soil acidification sensitivity training set; With y i are the predicted value of acidification sensitivity and the numerical value of acidification sensitivity of the i-th soil sample point respectively; is the mean of the predicted values ​​of acidification sensitivity of the soil acidification sensitivity training set.

7. The method for predicting soil acidification sensitivity based on regional big data according to claim 1, characterized in that: The soil types include: red soil, yellow soil, yellow-brown soil, brown soil, black soil, chernozem, purple soil, tidal soil, meadow soil, white soil and paddy soil; the soil physical and chemical properties include: clay content, pH value, organic matter, cation exchange capacity and base saturation.

8. The method for predicting soil acidification sensitivity based on regional big data according to claim 6, characterized in that: The normalization methods of the determination coefficient, the root mean square error and the mean absolute error include: deviation normalization and Z-score normalization.

Images