CN108764527B - Screening method for soil organic carbon library time-space dynamic prediction optimal environment variables - Google Patents

Abstract

The invention relates to a method for screening optimal environmental variables for the temporal and spatial dynamic prediction of soil organic carbon pools, which collects and sorts soil databases and sets of dynamic and static environmental variables in different periods, and specifies the effective time for soil carbon pool simulation according to the timeliness of soil data. Segments - "time slices". According to the soil organic carbon density information in different landscape areas, the attribute information of various dynamic and static environmental variables is analyzed, and then the analytic hierarchy process and expert knowledge are used to construct an effective optimal set of environmental variables in a specific time slice.

Description

A screening method for optimal environmental variables for spatiotemporal dynamic prediction of soil organic carbon pools

technical field

The invention relates to a method for screening optimal environmental variables for temporal and spatial dynamic prediction of soil organic carbon pools, and belongs to the technical field of quantitative soil science.

Background technique

Soil organic carbon is a collective term for the humus produced by the action of animal and plant residues, microorganisms and soil microorganisms. Compared with soil inorganic carbon pools, soil organic carbon content is more sensitive to soil cultivation and land use changes. Global warming since the Industrial Revolution has forced governments around the world to pay increasing attention to reducing greenhouse gas emissions. According to relevant statistics, the organic carbon storage within one meter of soil is the most important component of soil carbon pool. As the largest carbon pool in the global terrestrial ecosystem, the storage of soil carbon pool is about the sum of atmospheric carbon pool and terrestrial ecosystem carbon pool (1200-2500Pg). Therefore, small changes in soil carbon stock will have a huge impact on the carbon stock of the entire terrestrial ecosystem. Considering that soil organic carbon content is also one of the most important indicators for soil fertility evaluation, spatial analysis techniques for soil organic carbon content/storage have received extensive attention in many sectors such as agriculture, environment, and industry.

Related research on soil carbon pool estimation technology includes carbon sequestration potential, storage distribution, influencing factors and uncertainty assessment. According to the size of the study area, the scale of the spatial distribution map of soil organic carbon can be divided into global scale, national scale, provincial scale, and watershed scale. The production of soil maps at different scales has different requirements for the accuracy of soil samples and soil maps. At present, the estimation of soil organic carbon storage at the provincial scale in my country focuses more on the medium scale (1:200,000), and the work on the large scale (≥1:50,000) is rarely reported. In terms of soil organic carbon storage estimation methods, domestic and foreign scholars have divided them into two categories: map method and model method. The map spot method is to assign the soil properties (bulk density, gravel content, organic carbon content) of the soil profile to a planar geographic information layer (such as soil type map, land use type map, vegetation type map), and then assign the The area of a single plot is multiplied by the organic carbon density of a single or multiple soils, and finally the soil organic carbon storage in this area is obtained. The graph spot method is simple and efficient, and is one of the most commonly used methods in related industries. The basic data sources such as soil type and area mainly come from the second national soil census. The patch method assumes that the soil within the patch type is homogeneous, which contradicts the high spatial heterogeneity of soils in the real world. With the development of geographic information system and computer science technology, soil scientists are gradually tending to use the model method to estimate soil organic carbon storage. The model method is to use a data-driven or process-driven prediction model to construct a prediction model in which other soil properties such as soil organic carbon are affected by soil-forming factors (climate, vegetation, topography, parent material) or various factors related to the decomposition rate of soil organic matter, typically Carbon cycle models such as Century model and DeNitrification-DeComposition model, typical data-driven models such as random forest, artificial neural network, geographically weighted regression and other methods.

The spatial variability of soil physical and chemical properties means that it is in an eternal process of change under different time and space conditions. This temporal and spatial variability is closely related to the man-made and natural factors involved in soil evolution. Relevant researchers have proved that soil organic carbon is closely related to soil physical properties (soil texture, bulk density) and chemical properties (nitrogen storage, inorganic carbon, heavy metal content), and this relationship is also reflected in the extensive application of the PTF function in hydrology, etc. aspect. In addition, in view of the zonal distribution characteristics of soil organic carbon storage, the distribution of soil organic carbon storage also has a high correlation with location information such as longitude and latitude. Some scholars have proved that the mineralization rate of soil organic carbon also has a certain quantitative coupling relationship with factors such as vegetation restoration, climate change, and conservation tillage. Based on remote sensing images in different periods and GIS analysis, the surface cover of each landscape type can be drawn, which can compare and analyze the impact of soil organic carbon storage on dynamic land use changes. Based on this, in the process of simulating soil organic carbon storage by the Century model, it is necessary to integrate the dynamic simulation mechanism of land use in related modules. In view of this, the spatiotemporal dynamic estimation technology of multi-scale soil organic carbon storage urgently needs to quantify the relationship between the spatial distribution of soil organic carbon and various environmental factors.

However, the spatiotemporal dynamic estimation technology of soil organic carbon storage still faces a series of technical problems in the implementation process of different landscape areas:

(1) Although digital soil mapping and geographic information system technologies are relatively mature, they mainly focus on research areas at the river basin and county level. The spatial heterogeneity of soil-landscape relationships in these areas is low, and there are often obvious gradient sequences or landforms. Characteristics, soil organic carbon content and other attributes generally have a high correlation with environmental variables, resulting in very low portability and robustness of the constructed prediction model, and it is difficult to directly implement it in adjacent or similar areas. The low portability of this model also indirectly causes serious regional uncertainties in the estimation of soil carbon stocks in different regions. International first-class soil research institutions such as the Institute have confirmed that the construction of national and global scale prediction models is one of the most important tasks in contemporary soil geography.

(2) Existing spatial prediction technologies pay more attention to the construction of models and their uncertainty assessment, while relatively ignoring the selection of model input parameters. Some digital soil mapping work only selects the input data of the model based on expert knowledge or data availability, and fails to comprehensively consider the spatial coupling relationship between the spatial variation characteristics of soil and environmental variables. If a specific variable screening mechanism is used to select the optimal environmental variables, the soil landscape information covered by all environmental variables may not be mined to the greatest extent in the prediction models based on different assumptions. If only using the functional theory-based precision evaluation mechanism to evaluate the set of environmental variables, it will reduce the stability of the construction of the prediction model and the derived analysis results, and thus cannot clearly indicate the spatial representation relationship between environmental variables and soil organic carbon content.

(3) The key link of digital soil mapping is to select appropriate data sources and environmental variables. There are many environmental variables that affect soil spatial variability. For example, there are no less than 50 commonly used topographic factors, no less than 20 commonly used climate variables, and no less than 30 commonly used remote sensing variables. At the same time, the research results of different extraction algorithms for the same variable are also continuously presented. These environmental variables have certain uncertainties and influence each other, so it is difficult to accurately define them with the traditional variable screening mechanism.

To sum up, the technical deficiencies mentioned above have seriously affected the spatiotemporal dynamic estimation of soil carbon pools in different industries, which not only reduces the efficiency and accuracy of the application departments in the production and processing of soil information products, but also brings great challenges to national economic planning and regulation. potential losses.

SUMMARY OF THE INVENTION

The technical problem to be solved by the present invention is to provide an optimal environmental variable screening method for the spatiotemporal dynamic prediction of soil organic carbon pool, which can not only provide a detailed optimal environmental variable selection method, but also quantitatively express the relationship between soil attributes and environmental variables. The spatial coupling relationship, based on the set of environmental variables at different time points, can effectively screen and obtain the impact of environmental variables on the changes of regional soil carbon pools in different periods.

In order to solve the above-mentioned technical problems, the present invention adopts the following technical solutions: The present invention designs a method for screening the optimal environmental variables for the temporal and spatial dynamic prediction of soil organic carbon pools, and obtains the optimal environmental variables corresponding to the prediction of soil organic carbon density in the target soil area. Collection, including the following steps:

Step A. Collect each soil sample point in the target soil area corresponding to the preset total time span, and based on the preset sub-time period, obtain each soil sample point in different periods, and each soil sample point within the preset sub-time period The projection coordinates of , are different from each other, and then proceed to step B; wherein, the total time period is composed of sub-time periods with equal time intervals, and the sub-time period is smaller than the total time span, and based on the sub-time period corresponding to the soil sample point, the total time span The number of next time periods is not less than 3;

Step B. Obtain the respective designated soil attribute data corresponding to each soil sample point at a preset depth in different periods, and the corresponding environmental variable data of each designated type, and then enter step C;

Step C. According to the respective designated soil attribute data corresponding to each soil sample point at a preset depth in different periods, obtain the soil organic carbon density of each soil sample point corresponding to the preset depth in different periods, and then proceed to step D ;

Step D. Preset the relative importance of different selection index attributes in the AHP, and then enter Step E;

Step E. According to the soil organic carbon density of each soil sample point corresponding to the preset depth in different periods, and the corresponding environmental variable data of each designated type, obtain the selection index attribute data of each designated type of environmental variable in the AHP, and then Go to step F;

Step F. According to the attribute data of each selection index relative to the target soil area for each designated type of environmental variables in the AHP, and the relative importance between the preset attributes of different selection indicators, calculate and obtain the different designated types in the AHP. The relative importance of each selection index attribute among environmental variables, and then enter step G;

Step G. According to the analytic hierarchy process, the relative importance of each selected index attribute among different designated types of environmental variables, and the data of each designated type of environmental variables corresponding to each soil sample point in different periods, construct the corresponding soil area corresponding to the target soil. The analytic hierarchy process tree of soil organic carbon density is used to obtain the optimal set of environmental variables corresponding to the prediction of soil organic carbon density in the target soil area.

Compared with the prior art, the method for screening the optimal environmental variables for the spatio-temporal dynamic prediction of soil organic carbon pool according to the present invention has the following technical effects:

(1) The optimal environmental variable screening method for temporal-spatial dynamic prediction of soil organic carbon pools designed by the present invention selects hundreds or thousands of environmental variables involved in the soil carbon pool estimation problem as a systematic problem, and according to its implicit information It decomposes many variables into several levels. While retaining the influence of each variable on the calculation results, it also clearly and clearly quantifies the degree of influence of each environmental variable on the spatial and temporal variation of soil carbon pool. The multi-criteria, high-efficiency general soil carbon pool estimation technology has important practical significance for soil carbon pool estimation without obvious structural characteristics. Sustainable development, ecological environment protection and other related departments have very important practical significance and application value;

(2) The optimal environmental variable screening method for temporal-spatial dynamic prediction of soil organic carbon pools designed by the present invention is simple and practical, not only integrates the expert knowledge of soil geography and other disciplines, but also systematically transforms complex information screening problems into easy-to-implement Reasoning problem; the pairwise comparison method used can determine the importance of the same type of environmental variables relative to other types of environmental variables, and the present invention is easier to generalize than complex mathematical methods in the fields of machine learning and artificial intelligence;

(3) The optimal environmental variable screening method for temporal-spatial dynamic prediction of soil organic carbon pools designed in the present invention can seamlessly connect a variety of spatial prediction techniques, and to the greatest extent guarantees the information content of environmental variables and the model assumptions of the prediction techniques "high internal" Poly, low coupling". The classification mechanism of static and dynamic variables not only effectively ensures the efficient use of historical soil data and basic environmental information, but also reduces the defect that the general spatiotemporal prediction model splits the coupling relationship between soil data and dynamic environmental variables in different periods.

Description of drawings

Fig. 1 is a schematic flow chart of the optimal environmental variable screening method for the spatiotemporal dynamic prediction of soil organic carbon pools designed by the present invention;

Figure 2 is the environmental variable attributes and main environmental variables involved in the AHP;

Figure 3 is a flow chart of the steps for comparing the Pearson correlation coefficient of environmental variables with soil organic carbon density;

Figure 4 is a flow chart of the steps for comparing the significant degree of difference between environmental variables and soil organic carbon density;

Fig. 5 is the flow chart of the steps that the numerical value type of environment variable is compared;

Fig. 6 is a flow chart of the steps of comparing whether the environment variable is a dynamic environment variable;

Figure 7 is a flow chart of steps for comparing the pedogenetic significance of environmental variables to soil organic carbon density;

Fig. 8 is a flow chart of the steps of comparing the collinear proportions of environmental variables;

Fig. 9 is a flow chart of the steps of comparing the absolute values of the linear regression coefficients of environmental variables;

Figure 10 is a flow chart of the steps for comparing the mean precision drop values of environmental variables;

Fig. 11 is the spatial distribution map of soil organic carbon density at 0-20cm depth predicted for the study area in 2010;

Figure 12 is the spatial distribution map of soil organic carbon density at 0-20cm depth predicted for the study area in 2020;

Figure 13 is the spatial distribution map of soil organic carbon density at 0-20 cm depth predicted for the study area in 2030.

Detailed ways

The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

The basic idea of the optimal environmental variable screening method for temporal and spatial dynamic prediction of soil organic carbon pools designed in the present invention is to collect and sort out soil databases and dynamic and static environmental variable sets in different periods, and to specify the simulation of soil carbon pools according to the timeliness of soil data. Valid time period - "time slice". According to the soil organic carbon density information in different landscape areas, the attribute information of various dynamic and static environmental variables is analyzed, and then the analytic hierarchy process and expert knowledge are used to construct an effective optimal set of environmental variables in a specific time slice.

As shown in Figure 1, the present invention designs a method for selecting optimal environmental variables for the spatiotemporal dynamic prediction of soil organic carbon pools, and obtains the optimal environmental variable set corresponding to the prediction of soil organic carbon density in the target soil area. , including the following steps:

Step A. Collect each soil sample point in the target soil area corresponding to the preset total time span, and based on the preset sub-time period, obtain each soil sample point in different periods, and each soil sample point within the preset sub-time period The projection coordinates of , are different from each other, and then proceed to step B; wherein, the total time period is composed of sub-time periods with equal time intervals, and the sub-time period is smaller than the total time span, and based on the sub-time period corresponding to the soil sample point, the total time span The number of the next time period shall not be less than 3.

The above-mentioned step A, in practical application, specifically includes the following steps:

Step A1. Collect each soil sample point in the target soil area corresponding to the preset total time span, and standardize the location information of all soil sample points to generate latitude and longitude location information with geographic coordinates and projected coordinates. Among them, the geographic coordinates are uniformly converted to GCS_WGS_1984, and the projected coordinates are uniformly converted to WGS_1984_Albers; the standardized projected coordinates are marked as (LongX, LatiY), based on the standardized coordinate positions, so that the projected coordinates of each soil sample point in the preset sub-time period are mutually If not, then go to step A2; wherein, the sub time period is less than the total time span.

Step A2. Take the sub-time period t, according to the following formula:

T=int(Year/t)*t

Set time slice labels for each soil sample point, that is, to obtain each soil sample point in different periods, and then enter step A3; where T represents the time slice label set for the soil sample point, and Year represents the collection of soil sample points time, int() represents the round down function.

Note: If there is very little soil data in a certain time period, some sample points can be extracted from the soil map in recent years. The specific number can refer to the number of soil sample points in the time interval before and after, and the difference does not exceed 50%.

Step A3. Determine whether the number of different time slice labels corresponding to each soil sample point is less than 3, if yes, go to Step A4, otherwise go to Step B.

Step A4. Expand the total time span according to the preset increment to realize the update of the total time span, and then return to step A1.

Step B. Obtain each designated soil attribute data corresponding to each soil sample point at a preset depth in different periods and corresponding environmental variable data of each designated type, and then proceed to step C.

In practical applications, in the above step B, the following steps B1 to B4 are specifically executed, that is, the specified soil attribute data corresponding to each soil sample point at the preset depth in different periods and the corresponding environmental variable data of each specified type are obtained. .

Step B1. Obtain each designated soil attribute data corresponding to each soil sample point under each time slice label, as shown in Table 1 below, and then proceed to Step B2.

serial number name abbreviation unit 1 Soil organic carbon content SOC g/kg 2 Test weight BD g/cm³ 3 Volume percentage of gravel content greater than 2mm Gr % 4 soil thickness T m 5 total nitrogen TN g/kg 6 total phosphorus TP g/kg 7 total potassium TK g/kg 8 Cation exchange capacity CEC cmol(+)/kg/clay particle 9 Hydrogen Ion Concentration Index pH - 10 soil sampling depth Depth m

Table 1

Step B2. For each designated soil attribute data corresponding to each soil sample point under each time slice label, perform a standardization operation to update, and then proceed to step B3.

Wherein, step B2 is specifically performed, and the following steps B2-1 to B2-2 are performed for each designated soil attribute data corresponding to each soil sample point under each time slice label to realize a standardized operation for updating.

Step B2-1. Standardize the units of each designated soil attribute data, that is, transform and unify the units of the soil attribute data.

Step B2-2. Perform outlier processing for each designated soil attribute data, wherein, according to the judgment of whether the deviation of the designated soil attribute data and its corresponding average value exceeds twice its corresponding standard deviation, different operations are performed as follows;

If s_i≥s_mean-2×s_std and s_i≤s_mean+2×s_std, then s_i is not an outlier;

If s_i<s_mean-2×s_std, then s_i is an outlier, and the value of s_i is defined as s_mean-2×s_std;

If s_i>s_mean+2×s_std, then s_i is an outlier, and the value of s_i is defined as s_mean+2×s_std;

Among them, s_i, s_mean, and s_std are the ith value, mean and standard deviation of soil attribute s, respectively.

Note: The method for detecting outliers in soil attribute data designed in the above steps can also adopt methods such as probability and statistics method, PCA method, and isolated forest method.

Step B3. Collect each designated type of environmental variable layer covering the target soil area, and then proceed to Step B4.

Layers of specified types of environmental variables covering the target soil area, including raster and vector formats. Common environmental variables include five soil-forming factors: time, terrain, organisms, parent material and climate. Terrain variables can be derived from digital elevation models, including slope, aspect, curvature, terrain humidity index and other variables; climatic variables include rainfall, temperature, wind speed, sunshine time and other variables in different periods; biological variables include surface vegetation types in different periods, Land use and various variables extracted from remote sensing images in different periods; parent material variables include variables such as hydrogeological maps, regolith types, etc. Some data can be applied for by relevant data platforms, such as the National Basic Geographic Information Center, the National Bureau of Surveying and Mapping Geographic Information, and the National Earth System Science Data Sharing Platform.

Step B4. Perform a normalization operation to update the environmental variable data in each designated type of environmental variable layer.

Wherein, step B4 is specifically executed, and the following steps B4-1 to B4-5 are executed for the environmental variable data in each designated type of environmental variable layer to realize the standardization operation for updating.

Step B4-1. Using the preset resampling method, convert each specified type of environment variable layer into each specified type of environmental variable raster layer with a specified resolution, and then go to step B4-2.

The above grid data resampling method is: bilinear interpolation method. The specified resolution can be set to 50m, 100m, 250m, 1000m and 2000m.

Note: The raster data resampling method can also use the nearest neighbor method, cubic convolution interpolation method, mode method and other methods.

Step B4-2. Uniformly transform the coordinates of each specified type of environment variable raster layer into the pre-set standard projection coordinates, and then proceed to step B4-3.

Step B4-3. Normalize the environmental variable data in each specified type of environmental variable layer to eliminate the influence of the difference of different environmental variable dimensions on the dependent variable. The normalization method can choose z-score standardization, Min -max normalization and other methods, then go to step B4-4.

Step B4-4. Determine whether there is missing data in each designated type of environment variable layer, if yes, select resampling to supplement or delete the designated type of environment variable, otherwise go to step B4-5.

Step B4-5. Superimpose each soil sample point on the surface-shaped environmental variable raster layer of each designated type, and obtain the environmental variable data of each designated type corresponding to each soil sample point.

Step C. According to the specified soil attribute data corresponding to each soil sample point at a preset depth in different periods, according to the following formula:

Obtain the soil organic carbon density corresponding to the preset depth of each soil sample point in different periods, and then proceed to step D. Among them, n∈{1,...,N}, N represents the number of soil sample points, SOCS _n represents the soil organic carbon density (kg/m ² ) at the nth soil sample point location, I represents the soil sample location of the soil sample point The number of layers, _{SOC_in} represents the soil organic carbon content in the soil attribute data corresponding to the i-th layer of the _n -th soil sample point, BD_in represents the soil in the soil attribute data corresponding to the i-th layer of the n-th soil sample point Bulk density, _{Gr_in} represents the volume percentage of the gravel content greater than 2 mm in the soil attribute data corresponding to the i-th layer of the _n -th soil sample point, T_in represents the soil attribute data of the i-th layer corresponding to the n-th soil sample point soil thickness.

Note: If the sampling depth of some soil sample points is less than the preset depth, the spline function and weighting function can be used to obtain the soil organic carbon density corresponding to the preset depth at the soil sample position.

Note: Preset depth refers to a fixed depth value below the soil surface.

Step D. Preset to define the relative importance of different selection index attributes in the AHP, and then proceed to Step E.

The attribute information of environmental variables is classified as the calculation information in the AHP model. A total of six selection indicators of AHP are considered: Pearson correlation coefficient, significant degree of difference, numerical type of environmental variables, whether it is a dynamic environmental variable, the proportion of collinearity and the importance of variables. Define the following variables to express the property types of environment variables, as shown in Table 2 below.

Table 2

The relationship of the above variables is shown in Figure 2.

In practical applications, in the above step D, the following steps D1 to D3 are specifically performed, and the relative importance of different selection index attributes in the AHP is preset and defined.

Step D1. Define a formal expression [a, b, va] or [a, b, 1/va], indicating the relative importance between the selection index attribute a and the selection index attribute b, where va is a range of 1 The integer value of ~9, [a, b, va] indicates that the selection index attribute a is more important than the selection index attribute b, [a, b, 1/va] indicates that the selection index attribute b is more important than the selection index attribute a The greater the value of va, the greater the importance.

Step D2. Based on the formal expression, define the relative importance of the first category of selection index attributes as follows:

[Pearson, Significance, 3]

[Pearson, VariableType, 9]

[Pearson, Soilforming, 3]

[Pearson, ConMulticoll, 3]

[Pearson, Importance, 1/3]

[Significance, VariableType, 9]

[Significance, Soilforming, 1]

[Significance, ConMulticoll, 1]

[Significance, Importance, 1/9]

[VariableType, Soilforming, 1/7]

[VariableType, ConMulticoll, 1/7]

[VariableType, Importance, 1/9]

[Soilforming, ConMulticoll, 1/7]

[Soilforming, Importance, 1/7]

[ConMulticoll, Importance, 1/5]

Among them, Pearson represents the Pearson correlation coefficient, Significance represents the significant degree of difference, VariableType represents the variable type, Soilforming represents the significance of soil genetics, ConMulticoll represents the proportion of collinearity, and Importance represents the variable importance.

Since the type of environment variable involves two types of attributes: numerical type and whether it is a dynamic environment variable, the importance of the variable also includes two types of attributes: linear regression coefficient and average precision reduction, so the following step D3 is performed.

Step D3. Based on the formal expression, define the relative importance of the second largest category of selection index attributes as follows:

[Scenario, Numeric, 3]

[Regression, MDA, 1]

Among them, Scenario and Numeric belong to the variable type VariableType, Scenario indicates whether it is a dynamic environment variable, Numeric indicates the numerical type, Regression and MDA belong to the variable importance Importance, Regression indicates the absolute value of the linear regression coefficient, and MDA indicates the average precision drop.

Step E. According to the soil organic carbon density corresponding to the preset depth of each soil sample point in different periods, and the corresponding environmental variable data of each designated type, perform the following steps E1 to E8 to obtain the environment of each designated type in the analytic hierarchy process. Variable selection index attribute data, and then enter step F.

Step E1. Obtain the Pearson correlation coefficient (Pearson) of each designated type of environmental variable relative to the soil organic carbon density, and the specific calculation method is as follows:

In the formula: ai is the ith environment variable.

Step E2. Obtain the significant degree of difference of each designated type of environmental variables relative to soil organic carbon density respectively.

Step E3. Confirm that the type of each designated type of environmental variable relative to soil organic carbon density is a categorical variable or a continuous variable.

Step E4. Confirm that the type of each designated type of environmental variable relative to the soil organic carbon density is a static environmental variable or a dynamic environmental variable.

Step E5. Confirm whether the relative soil organic carbon density and soil organic carbon content have pedogenetic significance for each specified type of environmental variable.

Step E6. Obtain the collinear proportion (ConMulticoll) of each designated type of environmental variable relative to soil organic carbon density. Among them, the calculation method of the variance inflation factor of the environmental variable ai is as follows:

Among them, R ² represents the coefficient of determination of ai after the regression of the constant term and other environmental variables.

Step E7. Obtain the absolute value (Regression) of the linear regression coefficient of each designated type of environmental variable relative to the soil organic carbon density.

Step E8. Obtain the average precision drop (MDA) of each designated type of environmental variable relative to soil organic carbon density. The specific calculation method is as follows.

Step E8-1. Using a random forest model, build a prediction model for environmental variables and soil organic carbon density (SOCS).

Step E8-2. Randomly modify the value of ai and record the error of the prediction model.

Step E8-3. According to the error corresponding to each environmental variable, set the average precision drop. The larger the MDA value, the more important the corresponding environmental variable is to the prediction model.

Step F. According to the analytic hierarchy process, the relative importance of each selection index attribute of each designated type of environmental variable relative to the target soil area, and the relative importance between the preset different selection index attributes, execute the following steps F1 to F8 in sequence, calculate Obtain the relative importance of each selection index attribute between different specified types of environmental variables in the AHP, and then enter step G.

Step F1. Based on the comparison between the Pearson correlation coefficient Pearson_ak of the kth specified type environmental variable relative to the target soil area, and the Pearson correlation coefficient Pearson_aj of the jth specified type environmental variable relative to the target soil area, if Pearson_ak≥Pearson_aj, Then go to step F1-1; otherwise go to step F1-2; where k∈{1, 2,...,K}, j∈{1, 2,...,K}, K represents the number of environment variables of the specified type, ak Represents the k-th specified type of environment variable, and aj represents the j-th specified type of environment variable.

Step F1-1. As shown in Figure 3, based on the Pearson correlation coefficient (Pearson), the importance of the kth specified type environmental variable ak relative to the jth specified type environmental variable aj is calculated in va [ak, aj, va] as follows:

If Indicator ₁ >10, and Pearson_ak>0.2, then va=4;

If Indicator ₁ >5, and Pearson_ak>0.1, then va=2;

otherwise, va = 1;

Wherein, Indicator ₁ =Pearson_ak/Pearson_aj.

Step F1-2. Based on the Pearson correlation coefficient, the degree of importance [ak, aj, va] of the environmental variable ak of the kth designated type relative to the environmental variable aj of the jth designated type [ak, aj, va] is calculated as follows:

If Indicator ₂ >10, and Pearson_aj>0.2, then va=1/4;

If Indicator ₂ >5, and Pearson_aj>0.1, then va=1/2;

otherwise, va = 1;

In the formula: Indicator ₂ =Pearson_aj/Pearson_ak.

Step F2. As shown in Figure 4, based on the comparison between the significance level Significance_ak of the environmental variable of the kth specified type relative to the target soil area, and the significance level Significance_aj of the environmental variable of the jth specified type relative to the target soil area , as follows, to achieve the importance of the kth specified type of environment variable ak relative to the jth of the jth of the specified type of environmental variable aj based on the degree of significance of the difference [ak, aj, va] obtained in va;

If Indicator ₃ =2, then va=9;

If Indicator ₃ =1, then va=6;

If Indicator ₃ =0, then va=1;

If Indicator ₃ = -1, then va = 1/6;

If Indicator ₃ =-2, then va=1/9;

Wherein, Indicator ₃ =Significance_ak-Significance_aj.

Step F3. As shown in FIG. 5, based on the comparison between the numerical type Numeric_ak of the kth specified type environmental variable relative to the target soil area, and the jth specified type environmental variable relative to the numerical type Numeric_aj of the target soil area, as follows, Based on the numerical type, the importance of the kth specified type environment variable ak relative to the jth specified type environment variable aj is obtained from va in [ak, aj, va];

If Indicator ₄ =1, then va=3;

If Indicator ₄ =0, then va=1;

otherwise, va = 1/3;

Wherein, Indicator ₄ =Numeric_ak-Numeric_aj.

Step F4. As shown in Figure 6, based on whether the kth specified type environmental variable relative to the target soil area is a dynamic environmental variable Scenario_ak, and the jth specified type environmental variable relative to whether the target soil area is a dynamic environmental variable Scenario_aj comparison between , as follows, the realization is based on whether it is a dynamic environment variable, the importance of the kth specified type environment variable ak relative to the jth specified type environment variable aj is obtained in va [ak, aj, va];

If Indicator ₅ =1, then va=6;

If Indicator ₅ = 0, then va = 1;

Otherwise, va=1/6.

Wherein, Indicator ₅ =Scenario_ak-Scenario_aj.

Step F5. As shown in Figure 7, based on the comparison between the soil genetic significance Soilforming_ak of the kth designated type environmental variable relative to the target soil area, and the soil genetic significance Soilforming_aj of the jth designated type environmental variable relative to the target soil area , as follows, to achieve the importance of the environmental variable ak of the kth specified species relative to the environmental variable aj of the jth specified species based on the significance of soil genetics [ak, aj, va] obtained in va;

If Indicator ₆ =1, then va=6;

If Indicator ₆ =2, then va=9;

If Indicator ₆ =1, then va=6;

If Indicator ₆ =0, then va=1;

If Indicator ₆ = -1, then va = 1/6;

If Indicator ₆ = -2, then va = 1/9;

Wherein, Indicator ₆ =Soilforming_ak-Soilforming_aj.

Step F6. As shown in Figure 8, based on the comparison between the collinear proportion ConMulticoll_ak of the kth specified type environmental variable relative to the target soil area, and the collinear proportion ConMulticoll_aj of the jth specified type environmental variable relative to the target soil area, if ConMulticoll_ak≥ConMulticoll_aj, go to step F6-1; otherwise go to step F6-2.

Step F6-1. Based on the proportion of collinearity, the importance of the kth specified type environmental variable ak relative to the jth specified type environmental variable aj [ak, aj, va] in va is calculated as follows:

If Indicator ₇ >10, and ConMulticoll_ak>10, then va=1/4;

If Indicator ₇ > 5, then va=1/3;

otherwise, va = 1;

Wherein, Indicator ₇ =ConMulticoll_ak-ConMulticoll_aj.

Step F6-2. Based on the proportion of collinearity, the importance of the kth specified type environmental variable ak relative to the jth specified type environmental variable aj [ak, aj, va] in va is calculated as follows:

If Indicator ₈ >10, and ConMulticoll_aj>10, then va=4;

If Indicator ₈ > 5, then va=3;

otherwise, va = 1;

Wherein, Indicator ₈ =ConMulticoll_aj-ConMulticoll_ak.

F7. As shown in Figure 9, based on the absolute value Regression_ak of the linear regression coefficient of the kth specified type of environmental variable relative to the target soil area, and the absolute value of the linear regression coefficient of the jth specified type of environmental variable relative to the target soil area Regression_aj If Regression_ak≥Regression_aj, go to step F7-1; otherwise, go to step F7-2.

Step F7-1. Based on the absolute value of the linear regression coefficient, the degree of importance [ak, aj, va] of the kth specified type environmental variable ak relative to the jth specified type environmental variable aj [ak, aj, va] is calculated as follows:

If Indicator ₉ > 10, then va=9;

If Indicator ₉ > 5, then va=6;

Otherwise, va=3.

Wherein, Indicator ₉ =Regression_ak/Regression_aj.

Step F7-2. Based on the absolute value of the linear regression coefficient, the degree of importance [ak, aj, va] of the kth specified type environmental variable ak relative to the jth specified type environmental variable aj [ak, aj, va] is calculated as follows:

If Indicator ₁₀ > 10, then va=1/9;

If Indicator ₁₀ >5, then va=1/6;

Otherwise, va=1/3.

Wherein, Indicator ₁₀ =Regression_aj/Regression_ak.

Step F8. As shown in Figure 10, based on the comparison between the average precision drop value MDA_ak of the kth specified type environmental variable relative to the target soil area, and the jth specified type environmental variable relative to the target soil area The comparison between the average precision drop value MDA_aj , if MDA_ak≥MDA_aj, go to step F8-1; otherwise, go to step F8-2.

Step F8-1. Based on the average precision drop value, the degree of importance [ak, aj, va] of the kth specified type environment variable ak relative to the jth specified type environment variable aj [ak, aj, va] is calculated as follows:

If Indicator ₁₁ >15, then va=9;

If Indicator ₁₁ >10, then va=7;

If Indicator ₁₁ >5, then va=5;

If Indicator ₁₁ >2, then va=3;

Otherwise, va=1.

Wherein, Indicator ₁₁ =MDA_ak/MDA_aj.

Step F8-2. Based on the average precision drop value, the degree of importance [ak, aj, va] of the kth specified type environment variable ak relative to the jth specified type environment variable aj [ak, aj, va] is calculated as follows:

If Indicator ₁₂ >15, then va=1/9;

If Indicator ₁₂ >10, then va=1/7;

If Indicator ₁₂ >5, then va=1/5;

If Indicator ₁₂ > 2, then va=1/3;

Otherwise, va=1.

In the formula: Indicator ₁₂ =MDA_aj/MDA_ak.

Step G. According to the analytic hierarchy process, the relative importance of each selected index attribute among different designated types of environmental variables, and the data of each designated type of environmental variables corresponding to each soil sample point in different periods, construct the corresponding soil area corresponding to the target soil. The analytic hierarchy process tree of soil organic carbon density is used to obtain the optimal set of environmental variables corresponding to the prediction of soil organic carbon density in the target soil area.

In practical applications, the above-mentioned step G is performed, and specifically includes the following steps:

Step G1. According to the analytic hierarchy process, the relative importance of each selection index attribute among different designated types of environmental variables, and the data of each designated type of environmental variables corresponding to each soil sample point in different periods, construct the corresponding soil area corresponding to the target soil. Analytic Hierarchy Tree for Soil Organic Carbon Density.

Step G2. Sort according to the overall contribution rate of the environmental variables, and select a pre-set percentage, such as 80% of the environmental variables, for the optimal set of environmental variables corresponding to the prediction of soil organic carbon density in the target soil area.

Step G3. According to the optimal set of environmental variables corresponding to the prediction of soil organic carbon density in the target soil area, a preset specified prediction model is used to predict and obtain the spatial distribution map of historical soil organic carbon density and future soil organic carbon density under different time slice labels Spatial distribution map.

The following is an example of the optimal environmental variable screening for the spatiotemporal dynamic estimation of organic carbon pools in Danyang City

Danyang City is a county-level city under the jurisdiction of Zhenjiang City, Jiangsu Province. Its comprehensive economic strength ranks seventh among the top ten counties (cities) in Jiangsu Province. The terrain is high in the northwest and low in the southeast, with low mountains and plains, mainly plains. Danyang is located in the mid-latitude north subtropical zone and has an oceanic climate. Due to the fast pace of economic development, industrial and agricultural land use has changed rapidly. Different land use conditions led to rapid changes in soil organic carbon pools. Taking 1200 soil sample points and environmental variables collected by different departments since 1980 as input data, the soil sampling depth is 0-20cm, and the optimal set of environmental variables for the spatiotemporal dynamic estimation of organic carbon pools is selected.

A total of 1200 soil sample points were collected from 1980 to 2015, and the time slice label of each soil sample point data was set at a time interval of 10 years. A total of four time slice labels are 1980, 1999, 2000, and 2010.

Based on the obtained soil sample points, the above-mentioned optimal environmental variable screening method for the spatiotemporal dynamic prediction of soil organic carbon pools designed by the present invention is performed, wherein, step G1 in step G is to construct a hierarchical analysis tree of soil organic carbon density corresponding to the target soil area , and the specific results are shown in Table 3 below.

table 3

Step G2. Sort according to the overall contribution rate of the variables, and select the top 80% as the set of environmental variables for predicting soil organic carbon density. The specific environmental variables selected are: elevation, river distance, terrain humidity index, annual average temperature, sunshine, annual Average rainfall, land use, slope position, neighborhood type, NDVI, plane curvature, soil type.

Step G3. According to the optimal set of environmental variables corresponding to the prediction of soil organic carbon density in the target soil region, a cellular automata model based on random forest is used. The spatial distribution map of soil organic carbon density is predicted in 2010, 2020 and 2030, as shown in Figure 11, Figure 12, and Figure 13, respectively.

For the spatiotemporal dynamic estimation of soil carbon pool, the screening of environmental variables is of great significance to the prediction accuracy. Different from the conventional data-based environmental variable screening technology, the present invention fully considers the comprehensive influence of environmental variable attributes on the prediction model, such as the environmental variable numerical type, whether it is a dynamic environmental variable, collinearity proportion and variable importance. The comprehensive consideration of these information provides an optimal set of environmental variables for accurate engineering measurement of soil carbon pools in the study area, effectively overcoming the uncertainty of the multi-period effects of dynamic environmental variables and soil organic carbon density. This technology has very important guiding significance for scientifically formulating reasonable engineering plans, realizing multi-scale soil survey and ecosystem service planning.

The embodiments of the present invention have been described in detail above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned embodiments, and can also be made within the scope of knowledge possessed by those of ordinary skill in the art without departing from the purpose of the present invention. Various changes.

Claims (10)

1. an optimal environmental variable screening method for the spatiotemporal dynamic prediction of soil organic carbon pool, obtains the optimal environmental variable set corresponding to the prediction of soil organic carbon density in the target soil region, and is characterized in that, comprising the steps: Step A. Collect each soil sample point in the target soil area corresponding to the preset total time span, and based on the preset sub-time period, obtain each soil sample point in different periods, and each soil sample point within the preset sub-time period The projection coordinates of , are different from each other, and then proceed to step B; wherein, the total time period is composed of sub-time periods with equal time intervals, and the sub-time period is smaller than the total time span, and based on the sub-time period corresponding to the soil sample point, the total time span The number of next time periods is not less than 3; Step B. Obtain the respective designated soil attribute data corresponding to each soil sample point at a preset depth in different periods, and the corresponding environmental variable data of each designated type, and then enter step C; Step C. According to the respective designated soil attribute data corresponding to each soil sample point at a preset depth in different periods, obtain the soil organic carbon density of each soil sample point corresponding to the preset depth in different periods, and then proceed to step D ; Step D. Preset the relative importance of different selection index attributes in the AHP, and then enter Step E; Step E. According to the soil organic carbon density of each soil sample point corresponding to the preset depth in different periods, and the corresponding environmental variable data of each designated type, obtain the selection index attribute data of each designated type of environmental variable in the AHP, and then Go to step F; Step F. According to the attribute data of each selection index relative to the target soil area for each designated type of environmental variables in the AHP, and the relative importance between the preset attributes of different selection indicators, calculate and obtain the different designated types in the AHP. The relative importance of each selection index attribute among environmental variables, and then enter step G; Step G. According to the analytic hierarchy process, the relative importance of each selected index attribute among different designated types of environmental variables, and the data of each designated type of environmental variables corresponding to each soil sample point in different periods, construct the corresponding soil area corresponding to the target soil. The analytic hierarchy process tree of soil organic carbon density is used to obtain the optimal set of environmental variables corresponding to the prediction of soil organic carbon density in the target soil area. 2. a kind of soil organic carbon pool space-time dynamic prediction optimal environmental variable screening method according to claim 1, is characterized in that, described step A comprises the steps: Step A1. Collect each soil sample point corresponding to the preset total time span in the target soil area, and the projection coordinates of each soil sample point in the preset sub-time period are different from each other, and then enter step A2; wherein, the sub-time period is less than total time span; Step A2. Take the sub-time period t, according to the following formula: T=int(Year/t)*t Set time slice labels for each soil sample point, that is, to obtain each soil sample point in different periods, and then enter step A3; where T represents the time slice label set for the soil sample point, and Year represents the collection of soil sample points Time, int() represents the round-down function; Step A3. Determine whether the number of different time slice labels corresponding to each soil sample point is less than 3, if yes, then enter step A4, otherwise enter step B; Step A4. Expand the total time span according to the preset increment to realize the update of the total time span, and then return to step A1. 3. A kind of soil organic carbon pool space-time dynamic prediction optimal environmental variable screening method according to claim 2, is characterized in that, in described step B, execute following step B1 to step B4, namely obtain under different periods, each soil Specified soil attribute data corresponding to each sample point at the preset depth, and corresponding environmental variable data of each specified type; Step B1. Obtain each designated soil attribute data corresponding to each soil sample point under each time slice label, and then enter step B2; Step B2. For each designated soil attribute data corresponding to each soil sample point under each time slice label, perform a standardized operation to update, and then enter step B3; Step B3. Collect the environmental variable layers of each designated type covering the target soil area, and then enter step B4; Step B4. Perform a normalization operation to update the environmental variable data in each designated type of environmental variable layer. 4. A kind of optimal environmental variable screening method for temporal and spatial dynamic prediction of soil organic carbon pool according to claim 3, is characterized in that, in described step B2, under each time slice label, each soil sample point corresponds to each designation respectively For soil attribute data, perform the following steps B2-1 to B2-2 to implement standardized operations for updating; Step B2-1. Standardize the units of each designated soil attribute data; Step B2-2. Perform outlier processing for each designated soil attribute data, wherein, according to the judgment of whether the deviation of the designated soil attribute data and its corresponding average value exceeds twice its corresponding standard deviation, different operations are performed as follows; If s_i≥s_mean-2×s_std and s_i≤s_mean+2×s_std, then s_i is not an outlier; If s_i<s_mean-2×s_std, then s_i is an outlier, and the value of s_i is defined as s_mean-2×s_std; If s_i>s_mean+2×s_std, then s_i is an outlier, and the value of s_i is defined as s_mean+2×s_std; Among them, s_i, s_mean, and s_std are the ith value, mean and standard deviation of soil attribute s, respectively. 5. a kind of soil organic carbon pool space-time dynamic prediction optimal environmental variable screening method according to claim 3, is characterized in that, in described step B4, for the environmental variable data in each designated type environmental variable layer, execute as follows Step B4-1 to Step B4-5, realize standardized operation to update; Step B4-1. Using the preset resampling method, convert each specified type of environment variable layer into each specified type of environmental variable raster layer of specified resolution, and then enter step B4-2; Step B4-2. Uniformly transform the coordinates of each specified type of environment variable raster layer into the projected coordinates of the preset standard, and then enter step B4-3; Step B4-3. Normalize the environmental variable data in each specified type of environmental variable layer, and then enter step B4-4; Step B4-4. Determine whether there is missing data in each designated type of environment variable layer, if yes, select resampling to supplement or delete the designated type of environment variable, otherwise go to step B4-5; Step B4-5. Superimpose each soil sample point on the surface-shaped environmental variable raster layer of each designated type, and obtain the environmental variable data of each designated type corresponding to each soil sample point. 6. A kind of soil organic carbon pool space-time dynamic prediction optimal environmental variable screening method according to claim 3, is characterized in that, in described step C, according to different periods, each soil sample point is under preset depth respectively. Corresponding to each specified soil attribute data, according to the following formula:

Obtain the soil organic carbon density corresponding to the preset depth of each soil sample point in different periods, where n∈{1,…,N}, N represents the number of soil sample points, SOCS _n represents the nth soil sample point Soil organic carbon density at the location, I represents the number of soil layers at the soil sample location, SOC_in represents the soil organic carbon content in the soil attribute data corresponding to the i-th layer of the _n -th soil sample location, BD_in represents the _n -th soil The soil bulk density in the soil attribute data corresponding to the i-th layer of the sample location, Gr_in represents the volume percentage of the gravel content greater than 2 mm in the soil attribute data corresponding to the i-th layer of the _n -th soil sample location, and T_in represents the _n -th The soil thickness in the soil attribute data corresponding to the i-th layer of the soil sample location. 7. A kind of optimal environmental variable screening method for soil organic carbon pool spatiotemporal dynamic prediction according to claim 6, characterized in that, in the step D, the following steps D1 to D3 are performed, and in the preset definition analytic hierarchy process, the relative importance between attributes of different selection indicators; Step D1. Define a formal expression [a, b, va] or [a, b, 1/va], indicating the relative importance between the selection index attribute a and the selection index attribute b, where va is a range of 1 The integer value of ~9, [a, b, va] indicates that the selection index attribute a is more important than the selection index attribute b, [a, b, 1/va] indicates that the selection index attribute b is more important than the selection index attribute a degree, the greater the value of va, the greater the degree of importance; Step D2. Based on the formal expression, define the relative importance of the first category of selection index attributes as follows: [Pearson, Significance, 3] [Pearson, VariableType, 9] [Pearson, Soilforming, 3] [Pearson, ConMulticoll, 3] [Pearson, Importance, 1/3] [Significance, VariableType, 9] [Significance, Soilforming, 1] [Significance, ConMulticoll, 1] [Significance, Importance, 1/9] [VariableType, Soilforming, 1/7] [VariableType, ConMulticoll, 1/7] [VariableType, Importance, 1/9] [Soilforming, ConMulticoll, 1/7] [Soilforming, Importance, 1/7] [ConMulticoll, Importance, 1/5] Among them, Pearson represents the Pearson correlation coefficient, Significance represents the significant degree of difference, VariableType represents the variable type, Soilforming represents the significance of soil genetics, ConMulticoll represents the proportion of collinearity, and Importance represents the variable importance; Step D3. Based on the formal expression, define the relative importance of the second largest category of selection index attributes as follows: [Scenario, Numeric, 3] [Regression, MDA, 1] Among them, Scenario and Numeric belong to the variable type VariableType, Scenario indicates whether it is a dynamic environment variable, Numeric indicates the numerical type, Regression and MDA belong to the variable importance Importance, Regression indicates the absolute value of the linear regression coefficient, and MDA indicates the average precision drop. 8. The method for screening optimal environmental variables for temporal-spatial dynamic prediction of soil organic carbon pools according to claim 7, characterized in that, in the step E, according to the soil samples at different times and corresponding to preset depths of soil sample locations For the organic carbon density and the corresponding environmental variable data of each designated type, perform the following steps E1 to E8 to obtain the selection index attribute data of each designated type of environmental variable in the AHP; Step E1. Obtain the Pearson correlation coefficients of each designated type of environmental variable relative to the soil organic carbon density; Step E2. Obtain the significant degree of difference of each designated type of environmental variables relative to soil organic carbon density; Step E3. Confirm that the type of each designated type of environmental variable relative to the soil organic carbon density is a categorical variable or a continuous variable; Step E4. Confirm that the type of each designated type of environmental variable relative to the soil organic carbon density is a static environmental variable or a dynamic environmental variable; Step E5. Confirm whether each designated type of environmental variable has pedogenetic significance relative to soil organic carbon density and soil organic carbon content; Step E6. Obtain the collinear proportions of each designated type of environmental variable relative to the density of soil organic carbon; Step E7. Obtain the absolute value of the linear regression coefficient of each designated type of environmental variable relative to the soil organic carbon density; Step E8. Obtain the average precision drop of the respective relative soil organic carbon density for each specified type of environmental variable. 9. A kind of soil organic carbon pool space-time dynamic prediction optimal environmental variable screening method according to claim 8, is characterized in that, in described step F, according to in the analytic hierarchy process, each designated type of environmental variables are respectively relative to the target soil area The attribute data of each selection index, as well as the relative importance between the preset attributes of different selection indicators, perform the following steps F1 to F8 in sequence to obtain the relative importance of each selection index attribute between different specified types of environmental variables in the AHP method. Importance; Step F1. Based on the comparison between the Pearson correlation coefficient Pearson_ak of the kth specified type environmental variable relative to the target soil area, and the Pearson correlation coefficient Pearson_aj of the jth specified type environmental variable relative to the target soil area, if Pearson_ak ≥Pearson_aj, then go to step F1-1; otherwise, go to step F1-2; where k∈{1, 2,…,K}, j∈{1, 2,…,K}, K represents the specified type of environment variable Quantity, ak represents the kth specified type of environment variable, aj represents the jth of the specified type of environment variable; Step F1-1. Based on the Pearson correlation coefficient, the degree of importance [ak, aj, va] of the environmental variable ak of the kth specified type relative to the environmental variable aj of the jth specified type is calculated as follows: If Indicator ₁ >10, and Pearson_ak>0.2, then va=4; If Indicator ₁ >5, and Pearson_ak>0.1, then va=2; otherwise, va = 1; Wherein, Indicator ₁ =Pearson_ak/Pearson_aj; Step F1-2. Based on the Pearson correlation coefficient, the degree of importance [ak, aj, va] of the environmental variable ak of the kth designated type relative to the environmental variable aj of the jth designated type [ak, aj, va] is calculated as follows: If Indicator ₂ >10, and Pearson_aj>0.2, then va=1/4; If Indicator ₂ >5, and Pearson_aj>0.1, then va=1/2; otherwise, va = 1; In the formula: Indicator ₂ =Pearson_aj/Pearson_ak; Step F2. Based on the comparison between the significance level Significance_ak of the environmental variable of the kth designated type relative to the target soil area, and the significance level Significance_aj of the environmental variable of the jth designated type relative to the target soil area, as follows, to achieve based on The degree of significance of the difference, the importance of the kth specified type environmental variable ak relative to the jth specified type environmental variable aj [ak, aj, va] obtained in va; If Indicator ₃ =2, then va=9; If Indicator ₃ =1, then va=6; If Indicator ₃ =0, then va=1; If Indicator ₃ = -1, then va = 1/6; If Indicator ₃ =-2, then va=1/9; Wherein, Indicator ₃ =Significance_ak-Significance_aj; Step F3. Based on the comparison between the numerical type Numeric_ak of the kth specified type environmental variable relative to the target soil area, and the jth specified type environmental variable relative to the numerical type Numeric_aj of the target soil area, as follows, based on the numerical type, the first The importance of k environmental variables ak of specified types relative to the environmental variables aj of the jth specified type is obtained from va in [ak, aj, va]; If Indicator ₄ =1, then va=3; If Indicator ₄ =0, then va=1; otherwise, va = 1/3; Wherein, Indicator ₄ =Numeric_ak-Numeric_aj; Step F4. Based on the comparison between whether the kth specified type environmental variable relative to the target soil area is a dynamic environmental variable Scenario_ak, and whether the jth specified type environmental variable relative to the target soil area is a dynamic environmental variable Scenario_aj, as follows, to achieve based on Whether it is a dynamic environment variable, the importance of the k-th specified type of environment variable ak relative to the j-th specified type of environment variable aj is obtained from va in [ak, aj, va]; If Indicator ₅ =1, then va=6; If Indicator ₅ = 0, then va = 1; otherwise, va = 1/6; Wherein, Indicator ₅ =Scenario_ak-Scenario_aj; Step F5. Based on the comparison between the soil genetic significance Soilforming_ak of the kth designated type environmental variable relative to the target soil area, and the soil genetic significance Soilforming_aj of the jth designated type environmental variable relative to the target soil area, as follows, to achieve based on Soil genetic significance, the importance of the kth specified species environmental variable ak relative to the jth specified species environmental variable aj [ak, aj, va] obtained in va; If Indicator ₆ =1, then va=6; If Indicator ₆ =2, then va=9; If Indicator ₆ =1, then va=6; If Indicator ₆ =0, then va=1; If Indicator ₆ = -1, then va = 1/6; If Indicator ₆ = -2, then va = 1/9; Wherein, Indicator ₆ =Soilforming_ak-Soilforming_aj; Step F6. Based on the comparison between the collinear proportion ConMulticoll_ak of the kth specified type environmental variable relative to the target soil area, and the collinear proportion ConMulticoll_aj of the jth specified type environmental variable relative to the target soil area, if ConMulticoll_ak ≥ ConMulticoll_aj, enter Step F6-1; otherwise, go to Step F6-2; Step F6-1. Based on the proportion of collinearity, the degree of importance of the kth specified type environmental variable ak relative to the jth specified type environmental variable aj [ak, aj, va] va is calculated as follows: If Indicator ₇ >10, and ConMulticoll_ak>10, then va=1/4; If Indicator ₇ > 5, then va=1/3; otherwise, va = 1; Wherein, Indicator ₇ =ConMulticoll_ak-ConMulticoll_aj; Step F6-2. Based on the proportion of collinearity, the importance of the kth specified type environmental variable ak relative to the jth specified type environmental variable aj [ak, aj, va] in va is calculated as follows: If Indicator ₈ >10, and ConMulticoll_aj>10, then va=4; If Indicator ₈ > 5, then va=3; otherwise, va = 1; Wherein, Indicator ₈ =ConMulticoll_aj-ConMulticoll_ak; F7. Based on the comparison between the absolute value Regression_ak of the linear regression coefficient of the kth specified type environmental variable relative to the target soil area, and the absolute value Regression_aj of the jth specified type environmental variable relative to the linear regression coefficient of the target soil area, if Regression_ak ≥Regression_aj, then go to step F7-1; otherwise go to step F7-2; Step F7-1. Based on the absolute value of the linear regression coefficient, the degree of importance [ak, aj, va] of the kth specified type environmental variable ak relative to the jth specified type environmental variable aj [ak, aj, va] is calculated as follows: If Indicator ₉ > 10, then va=9; If Indicator ₉ > 5, then va=6; otherwise, va = 3; Wherein, Indicator ₉ =Regression_ak/Regression_aj; Step F7-2. Based on the absolute value of the linear regression coefficient, the degree of importance [ak, aj, va] of the kth specified type environmental variable ak relative to the jth specified type environmental variable aj [ak, aj, va] is calculated as follows: If Indicator ₁₀ > 10, then va=1/9; If Indicator ₁₀ >5, then va=1/6; otherwise, va = 1/3; Wherein, Indicator ₁₀ =Regression_aj/Regression_ak; Step F8. Based on the comparison between the average precision decline value MDA_ak of the kth designated type environmental variable relative to the target soil area, and the jth designated type environmental variable relative to the average precision reduction value MDA_aj of the target soil area, if MDA_ak≥MDA_aj, Then go to step F8-1; otherwise go to step F8-2; Step F8-1. Based on the average precision drop value, the degree of importance [ak, aj, va] of the kth specified type environment variable ak relative to the jth specified type environment variable aj [ak, aj, va] is calculated as follows: If Indicator ₁₁ >15, then va=9; If Indicator ₁₁ >10, then va=7; If Indicator ₁₁ >5, then va=5; If Indicator ₁₁ >2, then va=3; otherwise, va = 1; Wherein, Indicator ₁₁ =MDA_ak/MDA_aj; Step F8-2. Based on the average precision drop value, the degree of importance [ak, aj, va] of the kth specified type environment variable ak relative to the jth specified type environment variable aj [ak, aj, va] is calculated as follows: If Indicator ₁₂ >15, then va=1/9; If Indicator ₁₂ >10, then va=1/7; If Indicator ₁₂ >5, then va=1/5; If Indicator ₁₂ > 2, then va=1/3; otherwise, va = 1; In the formula: Indicator ₁₂ =MDA_aj/MDA_ak. 10. a kind of soil organic carbon pool space-time dynamic prediction optimal environmental variable screening method according to claim 9, is characterized in that, described step G, comprises the steps: Step G1. According to the analytic hierarchy process, the relative importance of each selection index attribute among different designated types of environmental variables, and the data of each designated type of environmental variables corresponding to each soil sample point in different periods, construct the corresponding soil area corresponding to the target soil. Analytic Hierarchy Tree for soil organic carbon density; Step G2. Sorting is performed according to the overall contribution rate of the environmental variables, and selects the environmental variables of the previous preset percentage, which are used for the optimal set of environmental variables corresponding to the prediction of soil organic carbon density in the target soil area; Step G3. According to the optimal set of environmental variables corresponding to the prediction of soil organic carbon density in the target soil area, a preset specified prediction model is used to predict and obtain the spatial distribution map of historical soil organic carbon density and future soil organic carbon density under different time slice labels Spatial distribution map.

Images