Land desertification decision-making method based on multi-source data

Abstract

The invention discloses a land desertification decision-making method based on multi-source data. The method comprises the following steps: S1, collecting land desertification area data; s2, performing data preprocessing on the land desertification area data; s3, applying a geographic detector model to the land desertification area data subjected to data preprocessing, determining a factor contribution rate, and screening land desertification data key factors through the factor contribution rate; s4, the screened land desertification data key factors are used for model construction of a generative adversarial network GAN, and simulated land desertification area data are generated; s5, fusing the simulated land desertification area data and the land desertification data key factors to generate desertification fusion data; and S6, evaluating the desertification risk levels of different areas, determining the desertification risk levels of different areas, and performing land desertification decision making. According to the method, the limitation of a traditional single data source and static analysis is overcome, and more comprehensive and intelligent desertification decision support is provided.

Description

Technical Field

[0001] The invention relates to the field of land desertification, and in particular to a land desertification decision-making method based on multi-source data. Background Art

[0002] Land desertification is one of the major environmental problems facing the world today. It not only causes serious damage to the ecological balance, but also threatens human survival and development. In existing technologies, the monitoring and assessment of land desertification mainly rely on traditional field surveys and analysis methods based on a single data source, which have many limitations:

[0003] 1. Land desertification is caused by the combined effects of multiple factors. A single data source is difficult to reflect the multi-dimensional changes in land desertification.

[0004] 2. The timeliness of desertification monitoring is limited and the monitoring accuracy is insufficient. Many monitoring measures are mainly based on regular data updates, which makes it difficult to reflect the dynamic changes of desertification in a timely manner;

[0005] 3. The application of land desertification decision-making based on advanced technologies is insufficient. Decision-making still mainly relies on traditional experience and manual analysis. There is a lack of intelligent decision-making support systems, which limits the efficiency and accuracy of scientific decision-making. Summary of the invention

[0006] In order to solve the above technical problems, the present invention proposes a land desertification decision-making method based on multi-source data.

[0007] A land desertification decision-making method based on multi-source data comprises the following steps:

[0008] S1: Collect data on land desertification areas and aggregate them into multi-source data;

[0009] S2: Data preprocessing of land desertification area data by data cleaning, format conversion, classification and discretization;

[0010] S3: Apply the geographic detector model to the land desertification area data after data preprocessing, determine the factor contribution rate, and screen the key factors of land desertification data through the factor contribution rate;

[0011] S4: The selected key factors of land desertification data are used to construct the model of generative adversarial network (GAN) to generate simulated land desertification area data;

[0012] S5: Fusing the simulated land desertification area data and the key factors of land desertification data to generate desertification fusion data;

[0013] S6: Assess the desertification risk level in different regions, determine the desertification risk grade in different regions, and make decisions on land desertification.

[0014] Furthermore, a land desertification decision-making method based on multi-source data, said step S1 comprises the following sub-steps:

[0015] S11: Collect climate data of the land;

[0016] The climate data include soil precipitation, air temperature, and wind speed;

[0017] S12: Collect geological data of the land;

[0018] The geological data include soil type, parent rock, topography and landform;

[0019] S13: Collect GDP and population density data for the land study area;

[0020] S14: Collect remote sensing image data of the land study area to supplement and strengthen the spatial and temporal information of land desertification analysis.

[0021] Furthermore, a land desertification decision-making method based on multi-source data, said step S2 comprises the following sub-steps:

[0022] S21: Data cleaning:

[0023] The Z-scor method and IQR method are used to process the outliers of climate data;

[0024] Use rule verification to correct erroneous information in geological data;

[0025] Linear interpolation was used to fill in missing values ​​in GDP and population density data;

[0026] Use the Fmask algorithm to process noise and outliers in remote sensing image data;

[0027] The noise and outliers include cloud cover and shadows;

[0028] S22: Format conversion:

[0029] Convert meteorological data and geological data into a unified raster format through ArcGIS;

[0030] Convert remote sensing image data from different sources into a unified format;

[0031] S23: Classification and discretization:

[0032] S231: Classify the continuous variables in the collected multi-source data;

[0033] S232: Discretize geological and economic continuous attribute data and map continuous values ​​into discrete categories.

[0034] Furthermore, a land desertification decision-making method based on multi-source data, said S3 comprises the following sub-steps:

[0035] S31: Calculate the q of each data of land desertification area data through the geographic detector model s Value, non-interactive detection calculation formula is:

[0036]

[0037] Among them, q s It represents the contribution rate of the factor to land desertification, and the value range is between 0 and 1. The magnitude of the q value is positively correlated with the impact of the factor on desertification.

[0038] Among them, K represents the total number of factors, that is, the number of independent factors considered in the model, and N i represents the number of observed samples at a certain factor level, N represents the number of all observed samples included in the model, var i represents the variance of land desertification degree at the i-th factor level, and var represents the variance of land desertification degree of all samples;

[0039] S32: Calculate the q of each data of land desertification area data through the geographic detector model interract Value, interactive detection calculation formula is:

[0040]

[0041] Among them, q interract represents the contribution rate of factor interaction to land desertification, n ij represents the number of observed samples under the interaction between the i-th factor and the j-th factor, var ij Represents the variance of land desertification degree under the interaction conditions of the ith factor and the ith factor;

[0042] S33: screening is performed according to the calculated q value, setting the q value threshold to ε, retaining factors with q>=ε, and removing factors with q<ε;

[0043] The q includes q s ,q interract .

[0044] Furthermore, a land desertification decision-making method based on multi-source data, said S4 comprises the following sub-steps:

[0045] S41: The key factors of land desertification data are used as inputs of the generator, and the generator generates simulated land desertification area data. The formula is:

[0046] G(z)→x′;

[0047] Among them, G(z) represents the generator, z represents the combination of the noise vector and the retained factor features, and x′ represents the simulated data for generating the desertified area;

[0048] S42: Use the loss function to make the generated data gradually approach the real data, and use the discriminator D to distinguish the real data x and the simulated data x′ generated by the generator G. The loss function of the discriminator D is L D The formula is:

[0049]

[0050] Among them, D(x) represents the output probability of the discriminator D for the real data x, and D(G(z)) represents the output probability of the discriminator D for the data G(z) generated by the generator G. Represents the mathematical expectation of the probability distribution of the real data x, Represents the mathematical expectation under the noise probability distribution;

[0051] S43: The generator G deceives the discriminator D, making it impossible for the discriminator D to distinguish between simulated data and real data. The generator loss function formula is for:

[0052]

[0053] S44: Introducing the auxiliary loss function L of the contribution rate of key factors of land desertification aux , to ensure that the generated data is more consistent with the characteristics of factors affecting desertification. The formula is:

[0054]

[0055] Among them, q i represents the contribution rate q value of the i-th factor, x′i represents the simulated value corresponding to the i-th factor feature in the data generated by the generator G, and x i represents the actual value of the i-th factor feature in the real data, n represents the number of factor features involved in the calculation, and MSE represents the mean square error, which is used to measure the difference between the generated data and the real data;

[0056] S45: Calculate the final generator loss function The formula is:

[0057]

[0058] Among them, λ represents the weight parameter, which is used to balance the adversarial loss and auxiliary loss;

[0059] S46: By maximizing the discriminator loss function L D Improve the discriminator D's ability to distinguish by minimizing the generator loss function Optimize the generator G so that the generated desertified area data is closer to the real data and more consistent with the characteristics of factors with higher values.

[0060] Furthermore, a land desertification decision-making method based on multi-source data, said S5 comprises the following sub-steps:

[0061] S51: Under the condition that the key factors of land desertification data and the simulated land desertification area data have the same spatial and temporal resolution, interpolate or resample the data from different sources to make the data consistent on the same grid or time step, and merge the key factors of land desertification data and the simulated land desertification area data into a complete input data set;

[0062] S52: Assign weights to the data sources generated by the generative adversarial network GAN to solve the existing errors. The weighted fusion formula is:

[0063] X final =α·x+(1-α)·x′;

[0064] Among them, x represents real data, x′ represents simulated data generated by the generative adversarial network GAN, and X final represents desertification fusion data, α represents the weight parameter, and α is dynamically adjusted according to the contribution rate of different factors;

[0065] S53: Use desertification fusion data to predict future desertification progress and fill in areas where there are gaps or scarcity in real data.

[0066] Furthermore, a land desertification decision-making method based on multi-source data, said S6 comprises the following sub-steps:

[0067] S61: Calculate the land desertification risk index of each region through desertification fusion data to evaluate the desertification risk level of different regions. The formula is:

[0068]

[0069] Among them, DRI represents the land desertification risk index, and the size of DRI is positively correlated with desertification risk;

[0070] Among them, w i Represents the weight of the i-th factor, which is determined based on the contribution rate q value of the i-th factor. represents the fused data of the ith factor, represents the standardized value of the factor, indicating the impact of the factor on desertification risk;

[0071] S61: Classify the desertification risk level by land desertification risk index and determine the desertification risk level R of different regions t , the formula is:

[0072]

[0073] Among them, a and b represent the thresholds of division, which are determined based on historical data and expert opinions;

[0074] S63: Optimize decisions through risk levels, set feedback indicators, and implement an iterative feedback mechanism until the decision effect is optimal.

[0075] Furthermore, in a land desertification decision-making method based on multi-source data, the S63 comprises the following sub-steps:

[0076] S631: According to the current risk level R t Perform decision optimization and set the decision function D(R t ):

[0077] D(R t )=g(R t );

[0078] Among them, g(R t ) represents the risk level R t The decision function is based on different risk levels R t Take different management measures;

[0079] S632: After the decision is implemented, the monitoring effect is fed back to the model to form a closed loop and set the feedback indicator F t :

[0080] F t =h(D(R t ), DRI t+1 );

[0081] Among them, h represents the decision effect evaluation function, DRI t+1 Represents the next land desertification risk index DRI value calculated after the decision is implemented;

[0082] S633: Check risk index R t and feedback index F t The change of is:

[0083] If|DRI t+1 -DRI t |<∈ and |F t+1 -F t |<∈′;

[0084] Among them, F t+1 Represents the next risk level R calculated after the decision is implemented t Value, ∈ represents the threshold for judging whether the land desertification risk index converges, and ∈′ represents the threshold for judging whether the feedback index converges;

[0085] S633: Determine the land desertification risk index DRI and feedback index F t Convergence:

[0086] If |DRI is satisfied t+1 -DRI t |<∈, the land desertification risk index DRI converges, indicating that the risk index calculation of the model has become stable and the risk prediction results have become accurate;

[0087] If |F t+1 -F t |<∈′), then the feedback index F t Convergence means that under the current decision-making strategy, the management effect is basically stable;

[0088] If there is a land desertification risk index DRI and a feedback index F t If the convergence condition is not met, the next iteration needs to be continued, using the updated desertification fusion data X final (t) Update risk index and decision, the formula is:

[0089]

[0090] Among them, ΔO k Represents the deviation between the indicator and the target value, indicating the difference in the current governance effect, ΔO k =O k (t)-T k , O k (t) represents the actual observed value of the kth indicator at time t, T k represents the expected target value of the kth indicator;

[0091] in, represents the updated weight of the ith factor, and β represents the adjustment step size, which controls the amplitude of the feedback update;

[0092] S634: When the land desertification risk index DRI and feedback index F t When the convergence continues, it is considered that the decision effect has reached the optimal one.

[0093] The beneficial effects of the present invention are:

[0094] 1. The accuracy and data integrity of land desertification monitoring are improved through multi-source data fusion and simulation data supplementation;

[0095] 2. By collecting multi-dimensional data such as climate, geology, socio-economics and remote sensing images, pre-processing technology is used to eliminate data noise and outliers, thus achieving standardized processing of various data;

[0096] 3. The geographic detector model screens out the key factors that have a significant impact on desertification, and dynamically adjusts the data weight based on the contribution rate, so that multi-source data can be reasonably integrated to generate high-quality desertification data;

[0097] 4. The desertification risk index calculated based on fusion data is used to classify risk levels and provide scientific decision-making support for desertification management in different regions;

[0098] 5. The generated simulation data fills the areas with insufficient data, realizes the dynamic optimization of desertification management through iterative feedback mechanism, and enhances the accuracy and responsiveness of desertification prevention and control;

[0099] 6. It overcomes the limitations of traditional single data source and static analysis, and provides more comprehensive and intelligent desertification decision-making support. BRIEF DESCRIPTION OF THE DRAWINGS

[0100] Figure 1 is a flow chart of the method. DETAILED DESCRIPTION

[0101] The present invention is further described below, but the protection scope of the present invention is not limited to the following description.

[0102] As attached Figure 1 As shown, a land desertification decision-making method based on multi-source data includes the following steps:

[0103] S1: Collect data on land desertification areas and aggregate them into multi-source data;

[0104] S2: Data preprocessing of land desertification area data by data cleaning, format conversion, classification and discretization;

[0105] S3: Apply the geographic detector model to the land desertification area data after data preprocessing, determine the factor contribution rate, and screen the key factors of land desertification data through the factor contribution rate;

[0106] S4: The selected key factors of land desertification data are used to construct the model of generative adversarial network (GAN) to generate simulated land desertification area data;

[0107] The simulated land desertification area data are used to supplement areas with insufficient actual data, helping to improve the completeness and accuracy of data in model evaluation and desertification prediction, thereby optimizing desertification risk assessment and decision support, and enabling the model to learn the characteristics and trends of land desertification.

[0108] S5: Fusing the simulated land desertification area data and the key factors of land desertification data to generate desertification fusion data;

[0109] S6: Assess the desertification risk level in different regions, determine the desertification risk grade in different regions, and make decisions on land desertification.

[0110] Furthermore, a land desertification decision-making method based on multi-source data, said step S1 comprises the following sub-steps:

[0111] S11: Collect climate data of the land;

[0112] The climate data include soil precipitation, air temperature, and wind speed;

[0113] S12: Collect geological data of the land;

[0114] The geological data include soil type, parent rock, topography and landform;

[0115] S13: Collect GDP and population density data for the land study area;

[0116] S14: Collect remote sensing image data of the land study area to supplement and strengthen the spatial and temporal information of land desertification analysis.

[0117] Furthermore, a land desertification decision-making method based on multi-source data, said step S2 comprises the following sub-steps:

[0118] S21: Data cleaning:

[0119] The Z-scor method and IQR method are used to process the outliers of climate data;

[0120] Among them, climate data here include precipitation and temperature;

[0121] Use rule verification to correct erroneous information in geological data;

[0122] Linear interpolation was used to fill in missing values ​​in GDP and population density data;

[0123] Use the Fmask algorithm to process noise and outliers in remote sensing image data;

[0124] The noise and outliers include cloud cover and shadows;

[0125] S22: Format conversion:

[0126] Convert meteorological data and geological data into a unified raster format through ArcGIS;

[0127] Convert remote sensing image data from different sources into a unified format;

[0128] S23: Classification and discretization:

[0129] S231: Classify the continuous variables in the collected multi-source data;

[0130] Among them, the multi-source data here include climate, geology, and socio-economics;

[0131] S232: Discretize geological and economic continuous attribute data and map continuous values ​​into discrete categories.

[0132] Furthermore, a land desertification decision-making method based on multi-source data, said S3 comprises the following sub-steps:

[0133] S31: Calculate the q of each data of land desertification area data through the geographic detector model s Value, non-interactive detection calculation formula is:

[0134]

[0135] Among them, q s It represents the contribution rate of the factor to land desertification, and the value range is between 0 and 1. The magnitude of the q value is positively correlated with the impact of the factor on desertification.

[0136] Among them, K represents the total number of factors, that is, the number of independent factors considered in the model, and N i represents the number of observed samples at a certain factor level, N represents the number of all observed samples included in the model, var i represents the variance of land desertification degree at the i-th factor level, and var represents the variance of land desertification degree of all samples;

[0137] S32: Calculate the q of each data of land desertification area data through the geographic detector model interract Value, interactive detection calculation formula is:

[0138]

[0139] Among them, q interract represents the contribution rate of factor interaction to land desertification, n ij represents the number of observed samples under the interaction between the i-th factor and the j-th factor, var ij Represents the variance of land desertification degree under the interaction condition of the i-th factor and the j-th factor;

[0140] S33: screening is performed according to the calculated q value, setting the q value threshold to ε, retaining factors with q>=ε, and removing factors with q<ε;

[0141] The q includes q s ,q interract .

[0142] Furthermore, a land desertification decision-making method based on multi-source data, said S4 comprises the following sub-steps:

[0143] S41: The key factors of land desertification data are used as inputs of the generator, and the generator generates simulated land desertification area data. The formula is:

[0144] G(z)→x′;

[0145] Among them, G(z) represents the generator, z represents the combination of the noise vector and the retained factor features, and x′ represents the simulated data for generating the desertified area;

[0146] S42: Use the loss function to make the generated data gradually approach the real data, and use the discriminator D to distinguish the real data x and the simulated data x′ generated by the generator G. The loss function of the discriminator D is L D The formula is:

[0147]

[0148] Among them, D(x) represents the output probability of the discriminator D for the real data x, and D(G(z)) represents the output probability of the discriminator D for the data G(z) generated by the generator G. Represents the mathematical expectation of the probability distribution of the real data x, Represents the mathematical expectation under the noise probability distribution;

[0149] Among them, the loss function ensures the quality and authenticity of the generated data;

[0150] S43: The generator G deceives the discriminator D, making it impossible for the discriminator D to distinguish between simulated data and real data. The generator loss function formula is for:

[0151]

[0152] S44: Introducing the auxiliary loss function L of the contribution rate of key factors of land desertification aux , to ensure that the generated data is more consistent with the characteristics of factors affecting desertification. The formula is:

[0153]

[0154] Among them, q i represents the contribution rate q value of the i-th factor, x′i represents the simulated value corresponding to the i-th factor feature in the data generated by the generator G, and x i represents the actual value of the i-th factor feature in the real data, n represents the number of factor features involved in the calculation, and MSE represents the mean square error, which is used to measure the difference between the generated data and the real data;

[0155] Among them, the auxiliary loss function Laux Let the generator G better capture the impact of factors with higher q values ​​on desertification;

[0156] S45: Calculate the final generator loss function The formula is:

[0157]

[0158] Among them, λ represents the weight parameter, which is used to balance the adversarial loss and auxiliary loss;

[0159] S46: By maximizing the discriminator loss function L D Improve the discriminator D's ability to distinguish by minimizing the generator loss function Optimize the generator G so that the generated desertified area data is closer to the real data and more consistent with the characteristics of factors with higher values.

[0160] Furthermore, a land desertification decision-making method based on multi-source data, said S5 comprises the following sub-steps:

[0161] S51: Under the condition that the key factors of land desertification data and the simulated land desertification area data have the same spatial and temporal resolution, interpolate or resample the data from different sources to make the data consistent on the same grid or time step, and merge the key factors of land desertification data and the simulated land desertification area data into a complete input data set;

[0162] S52: Assign weights to the data sources generated by the generative adversarial network GAN to solve the existing errors. The weighted fusion formula is:

[0163] X final =α·x+(1-α)·x′;

[0164] Among them, x represents real data, x′ represents simulated data generated by the generative adversarial network GAN, and X final represents desertification fusion data, α represents the weight parameter, and α is dynamically adjusted according to the contribution rate of different factors;

[0165] S53: Use desertification fusion data to predict future desertification progress and fill in areas where there are gaps or scarcity in real data.

[0166] Furthermore, a land desertification decision-making method based on multi-source data, said S6 comprises the following sub-steps:

[0167] S61: Calculate the land desertification risk index of each region through desertification fusion data to evaluate the desertification risk level of different regions. The formula is:

[0168]

[0169] Among them, DRI represents the land desertification risk index, and the size of DRI is positively correlated with desertification risk;

[0170] Among them, the calculation of DRI is based on the weighted values ​​of various environmental factors, such as precipitation, wind speed, temperature, and vegetation cover;

[0171] Among them, w i Represents the weight of the i-th factor, which is determined based on the contribution rate q value of the i-th factor. represents the fused data of the ith factor, represents the standardized value of the factor, indicating the impact of the factor on desertification risk;

[0172] S61: Classify the desertification risk level by land desertification risk index and determine the desertification risk level R of different regions t , the formula is:

[0173]

[0174] Among them, a and b represent the thresholds of division, which are determined based on historical data and expert opinions;

[0175] Among them, a and b can also be determined according to other standards;

[0176] S63: Optimize decisions through risk levels, set feedback indicators, and implement an iterative feedback mechanism until the decision effect is optimal.

[0177] Furthermore, in a land desertification decision-making method based on multi-source data, the S63 comprises the following sub-steps:

[0178] S631: According to the current risk level R t Perform decision optimization and set the decision function D(R t ):

[0179] D(R t )=g(R t );

[0180] Among them, g(R t ) represents the risk level R t The decision function is based on different risk levels R t Take different management measures;

[0181] S632: After the decision is implemented, the monitoring effect is fed back to the model to form a closed loop and set the feedback indicator F t :

[0182] F t =h(D(R t ), DRI t+1);

[0183] Among them, h represents the decision effect evaluation function, DRI t+1 Represents the next land desertification risk index DRI value calculated after the decision is implemented;

[0184] S633: Check risk index R t and feedback index F t The change of is:

[0185] If|DRI t+1 -DRI t |<∈ and |F t+1 -F t |<∈′;

[0186] Among them, F t+1 Represents the next risk level R calculated after the decision is implemented t Value, ∈ represents the threshold for judging whether the land desertification risk index converges, and ∈′ represents the threshold for judging whether the feedback index converges;

[0187] S633: Determine the land desertification risk index DRI and feedback index F t Convergence:

[0188] If |DRI is satisfied t+1 -DRI t |<∈, the land desertification risk index DRI converges, indicating that the risk index calculation of the model has become stable and the risk prediction results have become accurate;

[0189] If |F t+1 -F t |<∈′), then the feedback index F t Convergence means that under the current decision-making strategy, the management effect is basically stable;

[0190] If there is a land desertification risk index DRI and a feedback index F t If the convergence condition is not met, the next iteration needs to be continued, using the updated desertification fusion data X final (t) Update risk index and decision, the formula is:

[0191]

[0192] Among them, ΔO k Represents the deviation between the indicator and the target value, indicating the difference in the current governance effect, ΔO k =O k (t)-T k , O k (t) represents the actual observed value of the kth indicator at time t, T krepresents the expected target value of the kth indicator;

[0193] Among them, the expected target values ​​are such as NDVI and precipitation;

[0194] in, represents the updated weight of the ith factor, and β represents the adjustment step size, which controls the amplitude of the feedback update;

[0195] S634: When the land desertification risk index DRI and feedback index F t When the convergence continues, it is considered that the decision effect has reached the optimal one.

[0196] This solution uses a land desertification decision-making method based on multi-source data to achieve multi-source data fusion and simulation data completion, thereby improving the accuracy and data integrity of land desertification monitoring; collects multi-dimensional data such as climate, geology, socio-economics and remote sensing images, uses preprocessing technology to eliminate data noise and outliers, and realizes standardized processing of various data; the geographic detector model screens out key factors that have a significant impact on desertification, and dynamically adjusts data weights based on contribution rates, so that multi-source data can be reasonably integrated to generate high-quality desertification data; the desertification risk index calculated based on fused data is used for risk level classification, providing scientific decision-making support for desertification management in different regions; the generated simulation data fills the areas with insufficient data, realizes dynamic optimization of desertification management through an iterative feedback mechanism, and enhances the accuracy and responsiveness of desertification prevention and control; overcomes the limitations of traditional single data source and static analysis, and provides more comprehensive and intelligent desertification decision-making support.

[0197] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A land desertification decision-making method based on multi-source data, characterized in that: The following steps are involved: S1: Collect data on land desertification areas and aggregate them into multi-source data; S2: Data preprocessing of land desertification area data by data cleaning, format conversion, classification and discretization; S3: Apply the geographic detector model to the land desertification area data after data preprocessing, determine the factor contribution rate, and screen the key factors of land desertification data through the factor contribution rate; S4: The selected key factors of land desertification data are used to construct the model of generative adversarial network (GAN) to generate simulated land desertification area data; S5: Fusing the simulated land desertification area data and the key factors of land desertification data to generate desertification fusion data; S6: Assess the desertification risk level in different regions, determine the desertification risk grade in different regions, and make decisions on land desertification.

2. The land desertification decision-making method based on multi-source data according to claim 1 is characterized in that: The step S1 comprises the following sub-steps: S11: Collect climate data of the land; The climate data include soil precipitation, air temperature, and wind speed; S12: Collect geological data of the land; The geological data include soil type, parent rock, topography and landform; S13: Collect GDP and population density data for the land study area; S14: Collect remote sensing image data of the land study area to supplement and strengthen the spatial and temporal information of land desertification analysis.

3. The land desertification decision-making method based on multi-source data according to claim 1 is characterized in that: The step S2 comprises the following sub-steps: S21: Data cleaning: The Z-scor method and IQR method are used to process the outliers of climate data; Use rule verification to correct erroneous information in geological data; Linear interpolation was used to fill in missing values ​​in GDP and population density data; Use the Fmask algorithm to process noise and outliers in remote sensing image data; The noise and outliers include cloud cover and shadows; S22: Format conversion: Convert meteorological data and geological data into a unified raster format through ArcGIS; Convert remote sensing image data from different sources into a unified format; S23: Classification and discretization: S231: Classify the continuous variables in the collected multi-source data; S232: Discretize geological and economic continuous attribute data and map continuous values ​​into discrete categories.

4. The land desertification decision-making method based on multi-source data according to claim 1 is characterized in that: The S3 comprises the following sub-steps: S31: Calculate the q of each data of land desertification area data through the geographic detector model s Value, non-interactive detection calculation formula is: Among them, q s It represents the contribution rate of the factor to land desertification, and the value range is between 0 and 1. The magnitude of the q value is positively correlated with the impact of the factor on desertification. Among them, K represents the total number of factors, that is, the number of independent factors considered in the model, and N i represents the number of observed samples at a certain factor level, N represents the number of all observed samples included in the model, var i represents the variance of land desertification degree at the i-th factor level, and var represents the variance of land desertification degree of all samples; S32: Calculate the q of each data of land desertification area data through the geographic detector model interract Value, interactive detection calculation formula is: Among them, q interract represents the contribution rate of factor interaction to land desertification, n ij represents the number of observed samples under the interaction between the i-th factor and the j-th factor, var ij Represents the variance of land desertification degree under the interaction condition of the i-th factor and the j-th factor; S33: screening is performed according to the calculated q value, setting the q value threshold to ε, retaining factors with q>=ε, and removing factors with q<ε; The q includes q s ,q interract .

5. The land desertification decision-making method based on multi-source data according to claim 1 is characterized in that: The S4 comprises the following sub-steps: S41: The key factors of land desertification data are used as inputs of the generator, and the generator generates simulated land desertification area data. The formula is: G(z)→x′; Among them, G(z) represents the generator, z represents the combination of the noise vector and the retained factor features, and x′ represents the simulated data for generating the desertified area; S42: Use the loss function to make the generated data gradually approach the real data, and use the discriminator D to distinguish the real data x and the simulated data x′ generated by the generator G. The loss function of the discriminator D is L D The formula is: Among them, D(x) represents the output probability of the discriminator D for the real data x, and D(G(z)) represents the output probability of the discriminator D for the data G(z) generated by the generator G. Represents the mathematical expectation of the probability distribution of the real data x, Represents the mathematical expectation under the noise probability distribution; S43: The generator G deceives the discriminator D, making it impossible for the discriminator D to distinguish between simulated data and real data. The generator loss function formula is for: S44: Introducing the auxiliary loss function L of the contribution rate of key factors of land desertification aux , to ensure that the generated data is more consistent with the characteristics of factors affecting desertification. The formula is: Among them, q i represents the contribution rate q value of the i-th factor, x′ i Represents the simulated value corresponding to the i-th factor feature in the data generated by the generator G, x i represents the actual value of the i-th factor feature in the real data, n represents the number of factor features involved in the calculation, and MSE represents the mean square error, which is used to measure the difference between the generated data and the real data; S45: Calculate the final generator loss function The formula is: Among them, λ represents the weight parameter, which is used to balance the adversarial loss and auxiliary loss; S46: By maximizing the discriminator loss function L D Improve the discriminator D's ability to distinguish by minimizing the generator loss function Optimize the generator G so that the generated desertified area data is closer to the real data and more consistent with the characteristics of factors with higher values.

6. The land desertification decision-making method based on multi-source data according to claim 1 is characterized in that: The S5 comprises the following sub-steps: S51: Under the condition that the key factors of land desertification data and the simulated land desertification area data have the same spatial and temporal resolution, interpolate or resample the data from different sources to make the data consistent on the same grid or time step, and merge the key factors of land desertification data and the simulated land desertification area data into a complete input data set; S52: Assign weights to the data sources generated by the generative adversarial network GAN to solve the existing errors. The weighted fusion formula is: X final =α·x+(1-α)·x′; Among them, x represents real data, x′ represents simulated data generated by the generative adversarial network GAN, and X final represents desertification fusion data, α represents the weight parameter, and α is dynamically adjusted according to the contribution rate of different factors; S53: Use desertification fusion data to predict future desertification progress and fill in areas where there are gaps or scarcity in real data.

7. The land desertification decision-making method based on multi-source data according to claim 1 is characterized in that: The S6 comprises the following sub-steps: S61: Calculate the land desertification risk index of each region through desertification fusion data to evaluate the desertification risk level of different regions. The formula is: Among them, DRI represents the land desertification risk index, and the size of DRI is positively correlated with desertification risk; Among them, w i Represents the weight of the i-th factor, which is determined based on the contribution rate q value of the i-th factor. represents the fused data of the ith factor, represents the standardized value of the factor, indicating the impact of the factor on desertification risk; S61: Classify the desertification risk level by land desertification risk index and determine the desertification risk level R of different regions t , the formula is: Among them, a and b represent the thresholds of division, which are determined based on historical data and expert opinions; S63: Optimize decisions through risk levels, set feedback indicators, and implement an iterative feedback mechanism until the decision effect is optimal.

8. The land desertification decision-making method based on multi-source data according to claim 7 is characterized in that: The S63 includes the following sub-steps: S631: According to the current risk level R t Perform decision optimization and set the decision function D(R t ): D(R t )=g(R t ); Among them, g(R t ) represents the risk level R t The decision function is based on different risk levels R t Take different management measures; S632: After the decision is implemented, the monitoring effect is fed back to the model to form a closed loop and set the feedback indicator F t : F t =h(D(R t ),DRI t+1 ); Among them, h represents the decision-making effect evaluation function, DRI t+1 Represents the next land desertification risk index DRI value calculated after the decision is implemented; S633: Check risk index R t and feedback index F t The change of is: If|DRI t+1 -DRI t |<∈ and |F t+1 -F t |<∈′; Among them, F t+1 Represents the next risk level R calculated after the decision is implemented t Value, ∈ represents the threshold for judging whether the land desertification risk index converges, and ∈′ represents the threshold for judging whether the feedback index converges; S633: Determine the land desertification risk index DRI and feedback index F t Convergence: If |DRI is satisfied t+1 -DRI t |<∈, the land desertification risk index DRI converges, indicating that the risk index calculation of the model has become stable and the risk prediction results have become accurate; If |F t+1 -F t |<∈′), then the feedback index F t Convergence means that under the current decision-making strategy, the management effect is basically stable; If there is a land desertification risk index DRI and a feedback index F t If the convergence condition is not met, the next iteration needs to be continued, using the updated desertification fusion data X final (t) Update risk index and decision, the formula is: Among them, ΔO k Represents the deviation between the indicator and the target value, indicating the difference in the current governance effect, ΔO k =O k (t)-T k , O k (t) represents the actual observed value of the kth indicator at time t, T k represents the expected target value of the kth indicator; in, represents the updated weight of the ith factor, and β represents the adjustment step size, which controls the amplitude of the feedback update; S634: When the land desertification risk index DRI and feedback index F t When the convergence continues, it is considered that the decision effect has reached the optimal one.

Images