Soil or sediment property mapping methods that integrate multi-source data and uncertain information

Abstract

The present invention discloses a soil or sediment attribute mapping method integrating multi-source data and uncertain information, and belongs to the field of soil sediment attribute prediction and mapping. The main steps are: classifying and sorting the acquired soil or sediment attribute data, distinguishing accurate soil attribute hard data and soil attribute soft data with uncertainty, wherein the soil attribute soft data can be derived from the transformation of historical data, model estimation of related soil sediment attributes, etc.; combining the Bayesian maximum entropy method framework, using the maximum entropy theorem to combine the soil attribute hard data and the soil attribute soft data to obtain the prior probability density function, and then searching for the soil attribute hard data and the soil attribute soft data around the prediction space point, and using the Bayesian theorem to update the prior probability density function, obtaining the posterior probability density function, and obtaining the predicted value and prediction uncertainty of the soil attribute by integrating the posterior probability density function.

Description

Technical Field

[0001] The invention belongs to the field of soil or sediment property prediction and mapping, and in particular relates to a soil sediment property mapping method integrating multi-source data and uncertain information. Background Art

[0002] Traditional soil sediment attribute mapping often requires a large number of field samples. After laboratory chemical analysis, geoscience methods are used to predict one or more soil sediment attributes in the entire study area, such as Kriging, inverse distance weighting, etc. Taking Kriging as an example, it has strict assumptions, second-order stationary and data distribution follows normal distribution, and it is impossible to include uncertain data in soil sediment attribute prediction. However, with the development of data science, more and more data and information have emerged, such as historical data, various related soil sediment attribute survey data, model estimation, etc. Traditional soil sediment attribute mapping technology cannot make full use of known multi-source data and process uncertain information, resulting in data information waste. Sometimes there is a situation where the amount of soil sediment attribute data of concern is relatively small, and the amount of other related soil sediment attribute data is large. Traditional soil sediment attribute prediction methods cannot perform soil sediment attribute prediction well.

[0003] Therefore, it is necessary to design a new soil sediment property mapping method with a rigorous mathematical framework to absorb and fully utilize known multi-source data and uncertain information to achieve high-precision prediction and mapping of soil sediment properties. Summary of the invention

[0004] The purpose of the present invention is to solve the problems existing in the prior art and provide a soil or sediment property mapping method that integrates multi-source data and uncertain information. The method of the present invention fully integrates historical data, a variety of related soil or sediment property survey data, model estimation, etc., represents these uncertain information in the form of probability density function, incorporates it into the framework of Bayesian maximum entropy method, and performs high-precision soil or sediment property prediction and mapping.

[0005] In order to achieve the above-mentioned invention object, the present invention specifically adopts the following technical solutions:

[0006] A soil or sediment attribute mapping method integrating multi-source data and uncertain information comprises the following steps:

[0007] S1. When the study area only contains soil attribute data of the target attribute, obtain the historical soil attribute data and the current soil attribute data in the study area, determine whether there is any overlap in spatial position between the historical soil attribute data and the current soil attribute data, integrate the historical soil attribute data according to the determination result, construct soil attribute soft data, and regard the current soil attribute data as soil attribute hard data;

[0008] S2. Incorporate the soft data and hard data of soil properties into the framework of the Bayesian maximum entropy method, set the constraints of the general knowledge function, construct the equation with the maximum information entropy as the optimization goal, solve the equation to obtain the prior probability density function, set the search radius at the spatial point where the soil property needs to be predicted, use the soft data and hard data of soil properties within the search radius to update the prior probability density function, obtain the posterior probability density function, and finally integrate the posterior probability density function to calculate the predicted value of the soil property and the uncertainty of the prediction and visualize them to complete the mapping;

[0009] S3. When the study area contains both soil attribute data of the target attribute and soil attribute data of the related attribute, the target soil attribute data and the related soil attribute data in the study area are obtained, and a multivariate model is constructed by taking the target soil attribute data as the dependent variable and the related soil attribute data as the independent variable, and the target soil attribute data and the related soil attribute data at the same spatial position are matched, and the spatial position without the target soil attribute data is taken as the target position, and the related soil attribute data at the target position is input into the multivariate model, and the target soil attribute data is predicted, and the predicted value output by the multivariate model is taken as the mean of the Gaussian probability density function, and the predicted variance output by the multivariate model is taken as the variance of the Gaussian probability density function, so as to obtain the soft data of the target soil attribute, and the target soil attribute data is taken as the hard data of the target soil attribute;

[0010] S4. Incorporate the target soil property soft data and the target soil property hard data into the framework of the Bayesian maximum entropy method, reset the constraints of the general knowledge function, reconstruct the equation with the maximum information entropy as the optimization goal, and obtain a new prior probability density function. Set a search radius at the spatial point where the soil property needs to be predicted, and use the target soil property soft data and the target soil property hard data within the search radius to update the new prior probability density function to obtain a new posterior probability density function. Finally, the new posterior probability density function is integrated to calculate the predicted value of the soil property and the uncertainty of the prediction and visualized to complete the mapping.

[0011] Based on the above scheme, each step can be implemented in the following preferred specific manner.

[0012] Preferably, in step S1, the specific process of constructing soil property soft data is as follows:

[0013] If the spatial position of the historical soil attribute data overlaps with the current soil attribute data, the overlapping data are matched, and the historical soil attribute data are used as the independent variable and the current soil attribute data as the dependent variable to build a linear regression model. The historical soil attribute data that does not overlap with the current soil attribute data are input into the linear regression model to obtain the model estimation value and the model estimation variance. Finally, the model estimation value is used as the mean of the Gaussian probability density function, and the model estimation variance is used as the variance of the Gaussian probability density function to obtain the soil attribute soft data in the form of Gaussian probability density function.

[0014] If there is no overlap in spatial position between the historical soil attribute data and the soil attribute data at the current moment, the mean of the soil attribute data at the current moment is calculated and used as the first mean, the variance of the soil attribute data at the current moment is calculated and used as the first variance, the mean of the historical soil attribute data is calculated and used as the second mean, the difference between the first mean and the second mean is calculated, the historical soil attribute data and the mean difference are added as the soil attribute update value, and finally the soil attribute update value is used as the mean of the Gaussian probability density function, and the first variance is used as the variance of the Gaussian probability density function to obtain soil attribute soft data in the form of a Gaussian probability density function.

[0015] Preferably, in steps S2 and S4, the general knowledge functions include mean, variance, covariance, variogram and high-order moments.

[0016] Preferably, the functional form of the equation is:

[0017]

[0018]

[0019] in, is information entropy; represents a random vector in the study area The realization value, for step S2, includes the hardened values ​​of all soil attribute hard data and soil attribute soft data, and for step S4, includes the hardened values ​​of all target soil attribute soft data and target soil attribute hard data; represents a random vector in the study area information; Represents a random vector expectations of information; Represents the prior probability density function associated with general knowledge; For the A general knowledge function of soil properties, if but , A normalizing constant representing general knowledge, The expectation of a normalizing constant representing general knowledge; Represents the number of general knowledge functions.

[0020] Furthermore, the prior probability density function obtained by solving the equation is in the form of:

[0021]

[0022]

[0023] in, represents the coefficient of normalized correlation; represents an intermediate variable; Represents the coefficients containing normalized correlation The Lagrange multiplier of .

[0024] Furthermore, in steps S2 and S4, the functional form of the posterior probability density function is:

[0025]

[0026] in, For spatial prediction points The posterior probability density function at ; Refers to the soil property value at the unknown point; represents the normalization constant; represents a data collection, which for step S2 is composed of soil attribute soft data and soil attribute hard data, and for step S4 is composed of target soil attribute hard data and target soil attribute soft data; represents the joint probability density function of the study area, which simultaneously considers soil attribute hard data, soil attribute soft data and unknown spatial location data for step S2, and simultaneously considers target soil attribute hard data, target soil attribute soft data and unknown spatial location data for step S4; Represents the probability density function of soil property soft data or the probability density function of target soil property soft data.

[0027] Furthermore, in steps S2 and S4, the soil property prediction value is calculated as follows:

[0028]

[0029] in, Represents the predicted value of soil properties.

[0030] Furthermore, in steps S2 and S4, the uncertainty of the prediction is calculated as follows:

[0031]

[0032] in, The uncertainty of the prediction.

[0033] Preferably, the soil property data include metal elements and non-metal elements, electrochemical parameters and physical parameters; wherein the electrochemical parameters specifically include conductivity, pH and redox potential, and the physical parameters specifically include particle size and moisture content.

[0034] Preferably, the multivariate model is a multivariate linear regression model or a support vector regression model.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] The method of the present invention can integrate multi-source uncertain data including historical data, relevant soil attribute data, model estimation, etc. The method of the present invention can capture the temporal and spatial variation law of soil attributes, and match the historical soil attribute data with the current soil attribute data as the basis for constructing a trend model, giving full play to the role of historical data, and then assisting in the construction of soil attribute soft data through a linear regression model. When the historical soil attribute data and the current soil attribute data cannot be matched, the method of the present invention can also capture the temporal variation law of soil attributes through the mean of soil attributes in different periods, and use the mean difference and variance to construct soil attribute soft data in the form of a Gaussian probability density function. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 Schematic diagram of the specific steps of the method of the present invention;

[0038] Figure 2 Schematic diagram of the spatial distribution of soil property A in scenario 1 of an embodiment of the present invention; wherein (a) represents the objective distribution of soil (or sediment) property A at the current moment, and (b) represents the objective distribution of soil (or sediment) property A at a certain moment in history;

[0039] Figure 3 The sampling point distribution diagram of scenario 1 at the current moment and a certain moment in history in the embodiment of the present invention;

[0040] Figure 4 Schematic diagram of the linear regression results of soil property A of the spatially coincident sampling points at the current moment and at a certain moment in history in an embodiment of the present invention;

[0041] Figure 5 Schematic diagram of mapping results of soil attribute A under different circumstances in scenario 1 in an embodiment of the present invention combined with the Bayesian maximum entropy method; wherein (a) represents the mapping result obtained using only soil attribute hard data, and (b) represents the mapping result obtained using both soil attribute hard data and soil attribute soft data;

[0042] Figure 6 Schematic diagram of the spatial distribution of soil property A in scenario 2 of an embodiment of the present invention; wherein (a) represents the objective distribution of soil (or sediment) property A at the current moment, and (b) represents the objective distribution of soil (or sediment) property A at a certain moment in history;

[0043] Figure 7 The sampling point distribution diagram of the current moment and a certain moment in history of scenario 2 in the embodiment of the present invention;

[0044] Figure 8 Schematic diagram of mapping results of soil attribute A under different circumstances in scenario 2 in an embodiment of the present invention combined with the Bayesian maximum entropy method; wherein (a) represents the mapping result obtained using only soil attribute hard data, and (b) represents the mapping result obtained using both soil attribute hard data and soil attribute soft data;

[0045] Fig. 9 Schematic diagram of sampling point distribution in scenario 3 of the embodiment of the present invention;

[0046] Fig.10 Schematic diagram of the spatial distribution of four soil properties at the current moment in scenario 3 of an embodiment of the present invention; wherein (a) represents the objective distribution of soil (or sediment) property A, (b) represents the objective distribution of soil (or sediment) property B, (c) represents the objective distribution of soil (or sediment) property C, and (d) represents the objective distribution of soil (or sediment) property D;

[0047] Fig.11 It is a scatter plot of the actual value of soil attribute A and the predicted value of the linear regression model in the embodiment of the present invention;

[0048] Fig.12 Schematic diagram of the mapping results of soil attribute A under different circumstances in scenario 3 combined with the Bayesian maximum entropy method in an embodiment of the present invention; wherein (a) represents the mapping result obtained using only soil attribute hard data, and (b) represents the mapping result obtained using both soil attribute hard data and soil attribute soft data. DETAILED DESCRIPTION

[0049] In order to make the above-mentioned purpose, features and advantages of the present invention more obvious and easy to understand, the specific implementation mode of the present invention is described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below. The technical features in each embodiment of the present invention can be combined accordingly without conflicting with each other.

[0050] In the description of the present invention, it should be understood that the terms "first" and "second" are only used for the purpose of distinguishing descriptions, and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features.

[0051] It should be noted that, in the present invention, sediment and soil are the two most common types of strata on the earth's surface, and the method of the present invention can be used for both soil attribute mapping and sediment attribute mapping. In this embodiment, taking soil attributes as an example, the implementation process of the soil or sediment attribute mapping method integrating multi-source data and uncertain information is discussed in detail.

[0052] like Figure 1 As shown, in a preferred implementation of the present invention, the soil or sediment attribute mapping method integrating multi-source data and uncertain information comprises the following steps S1 to S4. The specific implementation process is described in detail below.

[0053] 1. Soil property prediction and mapping by integrating historical and current data

[0054] 1. Data acquisition and preprocessing

[0055] S1. When the study area only contains soil attribute data of the target attribute, obtain the historical soil attribute data and the current soil attribute data in the study area, determine whether there is any overlap in spatial position between the historical soil attribute data and the current soil attribute data, and integrate the historical soil attribute data according to the determination result to construct soil attribute soft data in the form of Gaussian probability density function. The current soil attribute data is regarded as soil attribute hard data, and the constructed soil attribute soft data is merged with the soil attribute hard data.

[0056] It should be noted that in the present invention, soil attribute data is a general term, which includes but is not limited to various metal and non-metal elements (copper, zinc, chromium, cadmium, nitrogen, phosphorus, sulfur, etc.), electrochemical parameters (conductivity, pH, redox potential, etc.), physical parameters (particle size, moisture content, etc.). When classifying soil attribute data, it can be divided into historical soil attribute data and current soil attribute data according to the collection time of soil samples. In addition to this classification method, it can also be divided into soil attribute hard data and soil attribute soft data according to the accuracy or uncertainty of the data. Among them, soil attribute hard data refers to the latest soil attribute data obtained through sampling and chemical analysis, and the present invention believes that this part of the data is very accurate and has no uncertainty; soil attribute soft data refers to soil attribute data in historical soil surveys, and the present invention believes that this part of the data has uncertainty. In addition, the present invention takes into account the long age and the trend of soil attributes changing over time, so the historical soil attribute data is further integrated and processed, that is, the historical soil attribute data is combined with the current soil attribute data, and then converted into soil attribute soft data, and expressed in the form of a probability density function. The soil property data at the current moment is considered as soil property hard data in the present invention because of its high detection accuracy and lack of uncertainty and its ability to represent the current soil property state.

[0057] It should be noted that, in the present invention, the specific process of constructing soil attribute soft data is as follows:

[0058] 1) If the spatial position of the historical soil attribute data overlaps with the current soil attribute data, the overlapping data are matched, and the historical soil attribute data are used as the independent variable and the current soil attribute data as the dependent variable to construct a linear regression model. The historical soil attribute data that does not overlap with the current soil attribute data are input into the linear regression model to obtain the model estimation value and the model estimation variance. Finally, the model estimation value is used as the mean of the Gaussian probability density function, and the model estimation variance is used as the variance of the Gaussian probability density function to obtain the soil attribute soft data in the form of Gaussian probability density function.

[0059] 2) If there is no overlap in spatial position between the historical soil attribute data and the soil attribute data at the current moment, the mean of the soil attribute data at the current moment is calculated and used as the first mean, the variance of the soil attribute data at the current moment is calculated and used as the first variance, the mean of the historical soil attribute data is calculated and used as the second mean, the variance of the historical soil attribute data is calculated and used as the second variance, the difference between the first mean and the second mean is calculated, and the historical soil attribute data and the mean difference are added as the soil attribute update value. Finally, the soil attribute update value is used as the mean of the Gaussian probability density function, and the first variance is used as the variance of the Gaussian probability density function to obtain the soil attribute soft data in the form of a Gaussian probability density function.

[0060] In this embodiment, a linear regression model constructed with historical soil attribute data as independent variables and current soil attribute data as dependent variables is a preferred method, including but not limited to linear regression models. In addition, polynomial models, local regression models, etc. are also possible; the form of soil attribute soft data includes but is not limited to Gaussian probability density function. In addition, uniform probability density function is also possible.

[0061] 2. Attribute prediction

[0062] S2. Incorporate the acquired soft data and hard data of soil properties into the framework of the Bayesian maximum entropy method, set the constraints of the general knowledge function, construct the equation with the maximum information entropy as the optimization goal, solve the equation to obtain the prior probability density function, set the search radius at the spatial point where the soil property needs to be predicted, use the soft data and hard data of soil properties within the search radius to update the prior probability density function, and obtain the posterior probability density function. Finally, the posterior probability density function is integrated to calculate the predicted value of the soil property and the uncertainty of the prediction and visualized to complete the mapping.

[0063] It should be noted that in step S2 of the present invention, the Bayesian maximum entropy method framework is used to absorb multi-source data and uncertain information for soil property prediction. The Bayesian maximum entropy method framework includes two stages: maximum entropy and Bayesian. The maximum entropy stage involves target optimization. Under the premise of setting constraints using the general knowledge function obtained by hardening and induction of soil property hard data and soil property soft data, the information entropy is maximized, and the prior probability density function is obtained by solving; the Bayesian stage uses Bayesian theorem to set a search radius at the soil property prediction space point, and combines the soil property hard data and soil property soft data in the search space to update the prior probability density function to obtain the posterior probability density function. Finally, the soil property prediction value and the uncertainty of the prediction can be calculated based on the posterior probability density function. The variance or standard deviation of the soil property prediction value can be used as the uncertainty of the soil property prediction. The general knowledge function of the present invention includes but is not limited to the mean, variance, covariance, variogram and high-order moment.

[0064] It should be noted that, in step S2 of the present invention, the functional form of the above equation is:

[0065]

[0066]

[0067] in, is information entropy; represents the realized value of the random vector in the study area, which contains the values ​​of all soil property hard data and soil property soft data hardening; represents a random vector in the study area information; Represents a random vector expectations of information; Represents the prior probability density function (joint probability density function) associated with general knowledge; For the A general knowledge function of soil properties, if but ; A normalizing constant representing general knowledge, The expectation of a normalizing constant representing general knowledge; Represents the number of general knowledge functions.

[0068] Furthermore, the prior probability density function obtained by solving the equation is in the form of:

[0069]

[0070]

[0071] in, represents the coefficient of normalized correlation; represents an intermediate variable; Represents the coefficients containing normalized correlation The Lagrange multiplier of .

[0072] It should be noted that, in step S2 of the present invention, the functional form of the posterior probability density function is:

[0073]

[0074] in, For spatial prediction points The posterior probability density function at ; Refers to the soil property value at the unknown point; represents the normalization constant; It represents a data set consisting of soil attribute soft data and soil attribute hard data; represents the joint probability density function of the study area, which takes into account soil property hard data, soil property soft data and unknown spatial location data; Represents the probability density function of soil property soft data, that is, Gaussian probability density function.

[0075] Furthermore, the predicted value of soil properties and the uncertainty of prediction are calculated by the following two formulas:

[0076]

[0077]

[0078] in, is the predicted value of soil properties, The uncertainty of the prediction.

[0079] 2. Soil Property Prediction and Mapping Integrating Related Soil Properties

[0080] 1. Data acquisition and preprocessing

[0081] S3. When the study area contains both soil attribute data of the target attribute and soil attribute data of the related attribute, obtain the target soil attribute data and the related soil attribute data in the study area, build a multivariate model with the target soil attribute data as the dependent variable and the related soil attribute data as the independent variable, match the target soil attribute data with the related soil attribute data at the same spatial position, take the spatial position without the target soil attribute data as the target position, input the related soil attribute data at the target position into the multivariate model, predict the target soil attribute data, take the predicted value output by the multivariate model as the mean of the Gaussian probability density function, take the predicted variance output by the multivariate model as the variance of the Gaussian probability density function, obtain the target soil attribute soft data, and take the target soil attribute data as the target soil attribute hard data.

[0082] It should be noted that in step S3 of the present invention, the target soil attribute data refers to the soil attribute data belonging to the target attribute that is newly obtained through sampling and high-cost chemical analysis, and the related soil attribute data refers to the soil attribute data that is related to the target soil attribute data and has a low analysis cost.

[0083] It should be noted that, in step S3 of the present invention, when there are a large amount of other attribute data related to the target soil attribute data (ie, related soil attribute data) in the study area, the number of types of the data is not limited.

[0084] It should be noted that in step S3 of the present invention, the multivariate model may be a multivariate linear regression model and a support vector regression model, which may be selected by those skilled in the art according to actual conditions and is not limited in the present invention.

[0085] 2. Attribute prediction

[0086] S4. Incorporate the target soil property soft data and the target soil property hard data into the framework of the Bayesian maximum entropy method, reset the constraints of the general knowledge function, reconstruct the equation with the maximum information entropy as the optimization goal, and obtain a new prior probability density function. Set a search radius at the spatial point where the soil property needs to be predicted, and use the target soil property soft data and the target soil property hard data within the search radius to update the new prior probability density function to obtain a new posterior probability density function. Finally, the new posterior probability density function is integrated to calculate the predicted value of the soil property and the uncertainty of the prediction and visualized to complete the mapping.

[0087] It should be noted that in step S4 of the present invention, the method of calculating the predicted value of soil properties and the uncertainty of the prediction is similar to that of step S2, the only difference being that the soft data of soil properties in S2 is replaced by the soft data of target soil properties in S4, and the hard data of soil properties in S2 is replaced by the hard data of target soil properties in S4. In addition, when reconstructing the equation, The corresponding value includes all the soft data of target soil attributes and the hardened value of target soil attribute hard data; in the new posterior probability density function, the joint probability density function of the study area is The hard data of target soil properties, soft data of target soil properties and unknown spatial location data are considered at the same time. Then it represents the probability density function of the target soil property soft data. The soil property prediction value and the prediction uncertainty calculation can refer to step S2, so the present invention will not repeat its implementation process.

[0088] It should also be noted that in step S4 of the present invention, the form of the target soil property soft data includes but is not limited to a Gaussian probability density function. In addition, a uniform probability density function may also be selected.

[0089] In order to better demonstrate the specific implementation and technical effects of the present invention, the soil or sediment attribute mapping method for fusing multi-source data and uncertain information shown in steps S1 to S4 in the above preferred implementation is applied to a specific example.

[0090] Example

[0091] The specific implementation process of the soil or sediment attribute mapping method integrating multi-source data and uncertain information used in this embodiment is as described above and will not be repeated here.

[0092] Scenario 1: Random simulation of the objective distribution of soil (or sediment) property A at the current moment in the 100×50 study area (e.g. Figure 2 (a) in the figure) and the objective distribution of soil (or sediment) property A at a certain moment in history (e.g. Figure 2(b) in the figure), and “sediment” is omitted below. Several soil samples are collected in the study area at the current time and at a certain time in history, and the attribute values ​​of soil attribute A are obtained respectively. The sampling points are distributed as follows Figure 3 shown. Figure 3 In the example, U4 represents the sampling point that coincides with a certain moment in history at the current moment, U3 represents the 40 selected verification sampling points at the current moment, the combination of U1, U3 and U4 represents 200 sampling points at the current moment, and the combination of U2 and U4 represents 150 sampling points at a certain moment in history. Scenario 1 considers the situation where the current sampling point and the historical sampling point overlap in space. 40 points that do not overlap with the historical sampling points are randomly selected from the current sampling points as model verification points, and the soil attribute values ​​of the remaining 160 current sampling points are regarded as soil attribute A hard data.

[0093] Considering that soil properties change with time, the attribute values ​​of the sampling points that overlap with the current and historical moments are matched, and the historical soil attribute A data is used as the independent variable and the current soil attribute A data is used as the dependent variable to construct a linear regression model. The mean absolute error of the linear regression model is 0.0329. Figure 4 Then, the attribute values ​​of the remaining sampling points at a certain moment in history are input into the linear regression model to obtain the predicted value and predicted variance, which are then constructed into the form of a Gaussian probability density function, which is the soft data of soil attribute A. Figure 4 In the example, the number of overlapping sampling points is 50.

[0094] The Bayesian maximum entropy method was used to compare the estimation accuracy of the soil attribute A value at the model verification point using only the hard data of the soil attribute A and using both the hard and soft data of the soil attribute A as known data. The results are shown in Table 1. The Bayesian maximum entropy method that considers both hard and soft data has a higher model estimation accuracy for the soil attribute A than the Bayesian maximum entropy method that uses only hard data. Among them, the Bayesian maximum entropy method that uses only hard data is also equivalent to the traditional kriging method. The mapping results of soil attribute A in the two cases are shown in Figure 5 shown.

[0095] Table 1. Scenario 1 model validation results

[0096] index Use both hard and soft data Use only hard data Mean absolute error 0.0397 0.0481 Mean square error 0.0039 0.0062

[0097] Scenario 2: Random simulation of the objective distribution of soil (or sediment) property A at the current moment in the 100×50 study area (e.g. Figure 6 (a) in the figure) and the objective distribution of soil (or sediment) property A at a certain moment in history (e.g. Figure 6(b) in the figure), and “sediment” is omitted below. Several soil samples are collected in the study area at the current time and at a certain time in history, and the attribute values ​​of soil attribute A are obtained respectively. The sampling points are distributed as follows Figure 7 shown. Figure 7 In the example, U3 represents the 40 selected verification sampling points at the current moment, the combination of U1 and U3 represents the 200 sampling points at the current moment, and U2 represents the 150 sampling points at a certain moment in history. Scenario 2 considers the situation where there is no spatial overlap between the current sampling points and the historical sampling points; 40 sampling points are randomly selected from the current sampling points as model verification points, and the soil attribute values ​​of the remaining 160 current sampling points are regarded as soil attribute A hard data.

[0098] Considering that soil properties change with time, the mean values ​​of soil property A at sampling points other than the verification point at the current moment and at sampling points at a certain moment in history are calculated as m0=3.1504, m1=2.9111, and the variances are v1=0.4204, v2=0.4236; the mean difference between the two moments is m2=0.2393. The soil property A values ​​of all sampling points at a certain moment in history are added with the calculated mean difference m2 as the mean and v1 as the variance to construct the Gaussian probability density function distribution of soil property A soft data. The Bayesian maximum entropy method is used to compare the estimation accuracy of soil property A values ​​at the model verification point using only the hard data of soil property A and using both the hard and soft data of soil property A as known data. The results are shown in Table 2. The Bayesian maximum entropy method that considers both hard and soft data has a higher model estimation accuracy for soil property A than the Bayesian maximum entropy method that uses only hard data. Among them, the Bayesian maximum entropy method that uses only hard data is also equivalent to the traditional kriging method. The soil property A mapping results for the two cases are as follows: Figure 8 shown.

[0099] Table 2. Scenario 2 model validation results

[0100] index Use both hard and soft data Use only hard data Mean absolute error 0.0606 0.0671 Mean square error 0.0201 0.0217

[0101] Scenario 3: Random simulation of the objective distribution of soil (or sediment) property A at the current moment in the 100×50 study area (e.g. Fig.10 (a) in the figure) and the objective distribution of soil properties B, C, and D related to soil (or sediment) property A (as shown in Fig.10 (b)(c)(d) in Figure 2), and “sediment” is omitted below. Several soil samples were collected in the study area, with a total of 300 sampling points, such as Fig. 9 shown. Fig. 9In the figure, U4 represents 50 sampling points that simultaneously detect soil properties A, B, C, and D, U3 represents 40 randomly selected validation sampling points that only detect soil property A, the combination of U1 and U3 represents 150 sampling points that only detect soil property A, and U2 represents 100 sampling points that only detect soil properties B, C, and D. 40 sampling points are randomly selected from the sampling points that only detect soil property A as model validation points; the remaining 160 soil property A detection values ​​are regarded as hard data.

[0102] Considering that soil attribute A is correlated with soil attributes B, C, and D, soil attributes B, C, and D in the sampling points with four soil attribute data are used as independent variables, and soil attribute A is used as the dependent variable to construct a linear regression model. The mean absolute error of the model accuracy is 0.1866, as shown in Fig.11 Then, the attribute values ​​of only the sampling points of soil attributes B, C, and D are input into the linear regression model to obtain the predicted value and predicted variance of soil attribute A, and then they are constructed into the form of Gaussian probability density function, which is the soft data of soil attribute A.

[0103] The Bayesian maximum entropy method was used to compare the estimation accuracy of the soil attribute A value at the model verification point using only the hard data of the soil attribute A and using both the hard and soft data of the soil attribute A as known data. The results are shown in Table 3. The Bayesian maximum entropy method that considers both hard and soft data has a higher model estimation accuracy for the soil attribute A than the Bayesian maximum entropy method that uses only hard data. Among them, the Bayesian maximum entropy method that uses only hard data is also equivalent to the traditional kriging method. The mapping results of soil attribute A in the two cases are shown in Fig.12 shown.

[0104] Table 3. Scenario 3 model validation results

[0105] index Use both hard and soft data Use only hard data Mean absolute error 0.0882 0.1701 Mean square error 0.0199 0.0663

[0106] The above-described embodiment is only a preferred solution of the present invention, but it is not intended to limit the present invention. A person skilled in the relevant technical field may make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, any technical solution obtained by equivalent replacement or equivalent transformation falls within the protection scope of the present invention.

Claims

1. A soil or sediment attribute mapping method integrating multi-source data and uncertain information, characterized in that: The following steps are involved: S1. When the study area only contains soil attribute data of the target attribute, obtain the historical soil attribute data and the current soil attribute data in the study area, determine whether there is any overlap in spatial position between the historical soil attribute data and the current soil attribute data, integrate the historical soil attribute data according to the determination result, construct soil attribute soft data, and regard the current soil attribute data as soil attribute hard data; S2. Incorporate the soft data and hard data of soil properties into the framework of the Bayesian maximum entropy method, set the constraints of the general knowledge function, construct the equation with the maximum information entropy as the optimization goal, solve the equation to obtain the prior probability density function, set the search radius at the spatial point where the soil property needs to be predicted, use the soft data and hard data of soil properties within the search radius to update the prior probability density function, obtain the posterior probability density function, and finally integrate the posterior probability density function to calculate the predicted value of the soil property and the uncertainty of the prediction and visualize them to complete the mapping; S3. When the study area contains both soil attribute data of the target attribute and soil attribute data of the related attribute, the target soil attribute data and the related soil attribute data in the study area are obtained, and a multivariate model is constructed by taking the target soil attribute data as the dependent variable and the related soil attribute data as the independent variable, and the target soil attribute data and the related soil attribute data at the same spatial position are matched, and the spatial position without the target soil attribute data is taken as the target position, and the related soil attribute data at the target position is input into the multivariate model, and the target soil attribute data is predicted, and the predicted value output by the multivariate model is taken as the mean of the Gaussian probability density function, and the predicted variance output by the multivariate model is taken as the variance of the Gaussian probability density function, so as to obtain the soft data of the target soil attribute, and the target soil attribute data is taken as the hard data of the target soil attribute; S4. Incorporate the target soil property soft data and the target soil property hard data into the Bayesian maximum entropy method framework, reset the constraints of the general knowledge function, reconstruct the equation with the maximum information entropy as the optimization goal, and obtain a new prior probability density function. Set a search radius at the spatial point where the soil property needs to be predicted, and use the target soil property soft data and the target soil property hard data within the search radius to update the new prior probability density function to obtain a new posterior probability density function. Finally, the new posterior probability density function is used to integrate and calculate the predicted value of the soil property and the uncertainty of the prediction and visualize it to complete the mapping; In step S1, the specific process of constructing soil attribute soft data is as follows: If the spatial position of the historical soil attribute data overlaps with the current soil attribute data, the overlapping data are matched, and the historical soil attribute data are used as the independent variable and the current soil attribute data as the dependent variable to build a linear regression model. The historical soil attribute data that does not overlap with the current soil attribute data are input into the linear regression model to obtain the model estimation value and the model estimation variance. Finally, the model estimation value is used as the mean of the Gaussian probability density function, and the model estimation variance is used as the variance of the Gaussian probability density function to obtain the soil attribute soft data in the form of Gaussian probability density function. If there is no overlap in spatial position between the historical soil attribute data and the soil attribute data at the current moment, the mean of the soil attribute data at the current moment is calculated and used as the first mean, the variance of the soil attribute data at the current moment is calculated and used as the first variance, the mean of the historical soil attribute data is calculated and used as the second mean, the difference between the first mean and the second mean is calculated, the historical soil attribute data and the mean difference are added as the soil attribute update value, and finally the soil attribute update value is used as the mean of the Gaussian probability density function, and the first variance is used as the variance of the Gaussian probability density function to obtain the soil attribute soft data in the form of a Gaussian probability density function; The functional form of the equation is: ； ； in, is information entropy; represents a random vector in the study area The realization value, for step S2, includes the hardened values ​​of all soil attribute hard data and soil attribute soft data, and for step S4, includes the hardened values ​​of all target soil attribute hard data and target soil attribute soft data; represents a random vector in the study area information; Represents a random vector expectations of information; Represents the prior probability density function associated with general knowledge; For the A general knowledge function of soil properties, if but , A normalizing constant representing general knowledge, The expectation of a normalizing constant representing general knowledge; represents the number of general knowledge functions; The prior probability density function obtained by solving the equation is in the form of: ； ； in, represents the coefficient of normalized correlation; represents an intermediate variable; Represents the coefficients containing normalized correlation The Lagrange multiplier of ; In steps S2 and S4, the functional form of the posterior probability density function is: ； in, For spatial prediction points The posterior probability density function at ; Refers to the soil property value at the unknown point; represents the normalization constant; represents a data collection, which for step S2 is composed of soil attribute soft data and soil attribute hard data, and for step S4 is composed of target soil attribute hard data and target soil attribute soft data; represents the joint probability density function of the study area, which simultaneously considers soil attribute hard data, soil attribute soft data and unknown spatial location data for step S2, and simultaneously considers target soil attribute hard data, target soil attribute soft data and unknown spatial location data for step S4; Represents the probability density function of soil property soft data or the probability density function of target soil property soft data.

2. The soil or sediment attribute mapping method integrating multi-source data and uncertain information according to claim 1, characterized in that: In steps S2 and S4, the general knowledge functions include mean, variance, covariance, variogram and high-order moments.

3. The soil or sediment attribute mapping method integrating multi-source data and uncertain information according to claim 1, characterized in that: In steps S2 and S4, the soil property prediction values ​​are calculated as follows: ； in, Represents the predicted value of soil properties.

4. The soil or sediment attribute mapping method integrating multi-source data and uncertain information according to claim 3, characterized in that: In steps S2 and S4, the uncertainty of the prediction is calculated as follows: ； in, The uncertainty of the prediction.

5. The soil or sediment attribute mapping method integrating multi-source data and uncertain information according to claim 1, characterized in that: The soil attribute data includes metal elements and non-metal elements, electrochemical parameters and physical parameters; wherein the electrochemical parameters are specifically conductivity, pH and redox potential, and the physical parameters are specifically particle size and moisture content.

6. The soil or sediment attribute mapping method integrating multi-source data and uncertain information according to claim 1, characterized in that: The multivariate model is a multivariate linear regression model or a support vector regression model.

Images