Machine learning-based soil drought risk early warning system and early warning method

Abstract

The application discloses a soil drought risk early warning system and method based on machine learning, belongs to the field of soil drought early warning, and comprises a data acquisition module, a data repair and fusion module, a space-time drought coupling prediction model construction module and a real-time early warning and model updating module; wherein the data acquisition module is used for acquiring wind cloud weather data and MODIS optical image data; the data repair and fusion module is used for generating all-weather soil humidity maps; the space-time drought coupling prediction model construction module is used for constructing a space-time drought coupling prediction model; and the real-time early warning and model updating module is used for obtaining a dynamic drought index and dynamically updating. The soil drought risk early warning system and method based on machine learning have the advantages that through multi-source data fusion, a space-time coupling prediction model and a dynamic grading early warning mechanism, high-precision drought risk prediction and decision support are realized.

Description

Technical Field

[0001] This invention relates to the field of soil drought early warning technology, and in particular to a soil drought risk early warning system and method based on machine learning. Background Technology

[0002] Traditional drought early warning methods rely on single meteorological indicators or static models, making it difficult to capture the spatiotemporal heterogeneity and suddenness of soil drought. These methods often focus only on simple meteorological parameters such as precipitation and temperature, neglecting multidimensional factors such as soil moisture and vegetation status, resulting in an insufficient understanding of the complexity of drought development. Existing technologies have the following limitations:

[0003] 1. Insufficient data integration: Traditional models fail to fully integrate remote sensing soil moisture, meteorological reanalysis data and vegetation dynamic indicators. The lack of effective integration of these multi-source data makes it difficult for prediction models to fully capture the complex mechanisms of drought evolution, thus limiting prediction accuracy.

[0004] 2. Weak model generalization ability: Drought is a complex dynamic process whose development is not only influenced by historical climate conditions but also closely related to neighboring regions in geographic space. However, while static machine learning models (such as random forests) perform well in processing structured data, they are less adaptable to the temporal dependence and spatial correlation of drought evolution.

[0005] 3. Poor Timeliness of Early Warnings: The occurrence and development of drought is a gradual but rapidly changing process, requiring timely monitoring and early warning to take effective measures. However, many early warning systems rely on fixed-period data updates, which cannot respond quickly to environmental changes, resulting in delayed warnings and missed opportunities for optimal intervention. Furthermore, the lack of real-time data stream support limits iterative model optimization, further weakening the system's predictive performance. In other words, existing systems generally lack dynamic update mechanisms, making it difficult to achieve real-time early warnings at the sub-seasonal (1-12 week) scale. Summary of the Invention

[0006] The purpose of this invention is to provide a soil drought risk early warning system and method based on machine learning, thereby solving the aforementioned technical problems.

[0007] To achieve the above objectives, the present invention provides a soil drought risk early warning system based on machine learning, including a data acquisition module, a data repair and fusion module, a spatiotemporal drought coupled prediction model construction module, and a real-time early warning and model update module;

[0008] The data acquisition module is used to collect wind and cloud meteorological data and MODIS optical image data.

[0009] The data repair and fusion module is used to repair cloud area data using generative adversarial networks and fuse wind and cloud meteorological data and MODIS optical image data to generate an all-weather soil moisture map.

[0010] The spatiotemporal drought coupled prediction model construction module is used to construct a spatiotemporal drought coupled prediction model based on all-weather soil moisture maps;

[0011] The real-time early warning and model update module is used to input real-time collected wind and cloud meteorological data and MODIS optical image data into the spatiotemporal drought coupling prediction model to obtain the dynamic drought index. The accuracy is verified by ground station data, and the spatiotemporal drought coupling prediction model is dynamically updated by combining the federated learning framework.

[0012] The early warning method for a soil drought risk early warning system based on machine learning includes the following steps:

[0013] S1. Collect wind and cloud meteorological data and MODIS optical image impact data, and extract features to form a dataset;

[0014] S2. Use generative adversarial networks to repair cloud area data, integrate the wind and cloud meteorological data collected in step S1 and MODIS optical image impact data, eliminate cloud pollution interference, and generate an all-weather soil moisture map.

[0015] S3. Based on the all-weather soil moisture map described in step S2, construct a spatiotemporal drought coupled prediction model;

[0016] S4. Input the real-time collected wind and cloud meteorological data and MODIS optical image impact data into the spatiotemporal drought coupling prediction model constructed in step S3 to obtain the dynamic drought index.

[0017] S5. Verify the accuracy of the dynamic drought index obtained in step S4 using ground station data, and combine it with the federated learning framework to realize the dynamic update of the spatiotemporal drought coupling prediction model.

[0018] Preferably, in step S1, the Fengyun meteorological satellite data includes microwave soil moisture data. ;

[0019] MODIS optical image data includes vegetation index. and surface temperature ;

[0020] Noise was eliminated and drought cycle features at different time scales were extracted using wavelet decomposition.

[0021] (1);

[0022] In the formula, This represents the decomposed drought abrupt change signal and drought cycle trend component; Indicates the total time; Represents the original time series; Represents the discrete wavelet basis functions. Indicates the number of decomposition layers. This represents the translation parameter.

[0023] Preferably, step S2 specifically includes the following steps:

[0024] S21. Data preprocessing: Unify the wind and cloud meteorological data and MODIS optical image impact data into the WGS84 coordinate system, and use MODIS cloud masks to mark the cloud-covered areas in the MODIS optical images.

[0025] S22. Construct a generative adversarial network and train it using historical data;

[0026] S23. Cloudy Area Restoration: Replace the original MODIS optical image data with cloudless optical images output by the generator of the generative adversarial network in the cloud-covered area, and at the same time, use Poisson fusion to eliminate stitching marks in the edge transition area of ​​the MODIS optical image data.

[0027] S24. Using fused and restored MODIS optical imagery data and Fengyun meteorological satellite data, a 1km resolution all-weather soil moisture map was generated through a multivariate regression model. :

[0028] (2);

[0029] In the formula, , , All represent regression coefficients; Indicates the meteorological-vegetation coupling index; Indicates the error term;

[0030] in,

[0031] (3);

[0032] In the formula, and These represent the maximum and minimum values ​​of the vegetation index, respectively. This represents the actual evapotranspiration, which is the sum of plant transpiration and surface water evaporation. Represents the potential maximum evapotranspiration;

[0033] S25. Verification and Iterative Optimization: Calculate the root mean square error between the measured soil moisture data from the weather station and the soil moisture predicted in step S24, and compare whether the root mean square error is within the set threshold range. If it is, it proves that the all-weather soil moisture map constructed in step S24 has passed the verification; otherwise, return to step S22 for update and iteration.

[0034] Preferably, the generator of the generative adversarial network adopts a U-Net structure, and the input of the generator is a cloud-covered MODIS optical image. Hefengyun meteorological data Output cloudless optical images ;

[0035] The discriminator in the generative adversarial network employs a convolutional neural network to distinguish between real, cloudless optical images. Output cloudless optical images Differences;

[0036] And the total loss function of the generative adversarial network The expression is as follows:

[0037] (4);

[0038] In the formula, Indicates resistance to loss; Indicates physical constraint loss; Indicates the weight of physical constraints;

[0039] in,

[0040] (5);

[0041] (6);

[0042] In the formula, This represents the actual data output by the discriminator; This represents the expected value operator; This represents the probability that false data is judged as true. Indicated based on vegetation index and surface temperature Soil moisture inversion function.

[0043] Preferably, the spatiotemporal coupled drought prediction model described in step S3 is constructed based on a temporal convolutional network and a graph attention network, wherein the temporal convolutional network is used to extract historical soil moisture. Long-range dependency:

[0044] (7);

[0045] In the formula, This indicates that the temporal convolutional network is at time step [time]. The feature vector obtained after feature extraction from the input sequence; This represents the extraction operation of a temporal convolutional network. Indicates a historical time window;

[0046] Graph attention networks are used to model the propagation of drought between regions:

[0047] (8);

[0048] In the formula, Indicates the area At any moment Drought prediction results; Indicates the area The set of neighboring regions; Indicates the area Its neighboring areas The intensity of the impact; Indicates the area At any moment Historical soil moisture data was extracted using a temporal convolutional network.

[0049] Furthermore, the primary objective of the spatiotemporally coupled drought prediction model is drought level classification, while the secondary objective is soil moisture numerical regression.

[0050] The preferred numerical regression expression for soil moisture is as follows:

[0051] (9);

[0052] In the formula, A measure of the degree of regression after soil moisture numerical values ​​have been regressed; Indicates the total number of regions; Indicates an index variable; This indicates the predicted soil moisture value; This indicates the actual measured soil moisture value.

[0053] Preferably, the dynamic drought index mentioned in step S4 The expression is as follows:

[0054] (10);

[0055] In the formula, Indicates the actual leaf area index; Indicates the potential leaf area index; Represents the logistic regression probability;

[0056] in,

[0057] (11);

[0058] (12);

[0059] In the formula, Indicates the maximum leaf surface area index; , , and All of these represent regression coefficients.

[0060] Preferably, the federated learning framework described in step S5 implements the following expression for dynamic model update:

[0061] (13);

[0062] In the formula, Indicates the global warning value; Indicates the total number of iterations; Indicates the first The number of data points used in the next iteration; Indicates the first The warning value calculated in the next iteration; This indicates the amount of weight update.

[0063] Therefore, the beneficial effects of the above-mentioned machine learning-based soil drought risk early warning system and method of the present invention are as follows:

[0064] 1. Comprehensive data integration: It integrates multi-source information such as wind and cloud meteorological data and MODIS optical image data, which effectively improves the prediction model's ability to capture the complex mechanisms of drought evolution;

[0065] 2. Strong model generalization ability: By constructing a spatiotemporally coupled drought prediction model and combining temporal convolutional networks and graph attention networks, it can simultaneously handle the temporal dependence and spatial correlation of drought, thereby enhancing the model's generalization performance.

[0066] 3. Good early warning timeliness: The federated learning framework enables dynamic model updates and supports real-time data stream processing, which helps to respond quickly to environmental changes and improve the timeliness of early warning.

[0067] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0068] Figure 1 This is a flowchart of a machine learning-based soil drought risk early warning system according to the present invention;

[0069] Figure 2 This is an all-weather soil moisture map generated from the simulation experiment described in this invention. Detailed Implementation

[0070] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the embodiments of the present invention and are not intended to limit the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of this application. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout.

[0071] It should be noted that the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, such as a process, method, system, product, or server that includes a series of steps or units, not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such process, method, product, or device.

[0072] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0073] A machine learning-based soil drought risk early warning system includes a data acquisition module, a data repair and fusion module, a spatiotemporal drought coupled prediction model construction module, and a real-time early warning and model update module. The data acquisition module collects wind and cloud meteorological data and MODIS optical imagery data. The data repair and fusion module uses generative adversarial networks to repair cloud area data and fuses wind and cloud meteorological data and MODIS optical imagery data to generate an all-weather soil moisture map. The spatiotemporal drought coupled prediction model construction module constructs a spatiotemporal drought coupled prediction model based on the all-weather soil moisture map. The real-time early warning and model update module inputs the real-time collected wind and cloud meteorological data and MODIS optical imagery data into the spatiotemporal drought coupled prediction model to obtain a dynamic drought index, verifies the accuracy using ground station data, and dynamically updates the spatiotemporal drought coupled prediction model using a federated learning framework.

[0074] like Figure 1 As shown, the early warning method of the soil drought risk early warning system based on machine learning includes the following steps:

[0075] S1. Collect wind and cloud meteorological data and MODIS optical image impact data, and extract features to form a dataset;

[0076] In step S1, the Fengyun meteorological satellite data includes microwave soil moisture data. ;

[0077] MODIS optical image data includes vegetation index. and surface temperature ;

[0078] Noise was eliminated and drought cycle features at different time scales were extracted using wavelet decomposition.

[0079] (1);

[0080] In the formula, This represents the decomposed drought abrupt change signal and drought cycle trend component; Indicates the total time; Represents the original time series; Represents the discrete wavelet basis functions. Indicates the number of decomposition layers. This represents the translation parameter.

[0081] S2. Use generative adversarial networks to repair cloud area data, integrate the wind and cloud meteorological data collected in step S1 and MODIS optical image impact data, eliminate cloud pollution interference, and generate an all-weather soil moisture map.

[0082] Step S2 specifically includes the following steps:

[0083] S21. Data preprocessing: Unify the wind and cloud meteorological data and MODIS optical image impact data into the WGS84 coordinate system, and use MODIS cloud masks to mark the cloud-covered areas in the MODIS optical images.

[0084] S22. Construct a generative adversarial network and train it using historical data;

[0085] S23. Cloudy Area Restoration: Replace the original MODIS optical image data with cloudless optical images output by the generator of the generative adversarial network in the cloud-covered area, and at the same time, use Poisson fusion to eliminate stitching marks in the edge transition area of ​​the MODIS optical image data.

[0086] S24. Using fused and restored MODIS optical imagery data and Fengyun meteorological satellite data, a 1km resolution all-weather soil moisture map was generated through a multivariate regression model. :

[0087] (2);

[0088] In the formula, , , All represent regression coefficients; Indicates the meteorological-vegetation coupling index; Indicates the error term;

[0089] in,

[0090] (3);

[0091] In the formula, and These represent the maximum and minimum values ​​of the vegetation index, respectively. This represents the actual evapotranspiration, which is the sum of plant transpiration and surface water evaporation. Represents the potential maximum evapotranspiration;

[0092] S25. Verification and Iterative Optimization: Calculate the root mean square error between the measured soil moisture data from the weather station and the soil moisture predicted in step S24, and compare whether the root mean square error is within the set threshold range. If it is, it proves that the all-weather soil moisture map constructed in step S24 has passed the verification; otherwise, return to step S22 for update and iteration.

[0093] Preferably, the generator of the generative adversarial network adopts a U-Net structure, and the input of the generator is a cloud-covered MODIS optical image. Hefengyun meteorological data Output cloudless optical images ;

[0094] The discriminator in the generative adversarial network employs a convolutional neural network to distinguish between real, cloudless optical images. Output cloudless optical images Differences;

[0095] And the total loss function of the generative adversarial network The expression is as follows:

[0096] (4);

[0097] In the formula, Indicates resistance to loss; Indicates physical constraint loss; Represents the physical constraint weights; in this embodiment, we take... ;

[0098] in,

[0099] (5);

[0100] (6);

[0101] In the formula, This represents the actual data output by the discriminator; This represents the expected value operator; This represents the probability that false data is judged as true. Indicated based on vegetation index and surface temperature Soil moisture inversion function.

[0102] S3. Based on the all-weather soil moisture map described in step S2, construct a spatiotemporal drought coupled prediction model;

[0103] The spatiotemporal coupled drought prediction model described in step S3 is constructed based on a temporal convolutional network and a graph attention network, wherein the temporal convolutional network is used to extract historical soil moisture. Long-range dependency:

[0104] (7);

[0105] In the formula, This indicates that the temporal convolutional network is at time step [time]. The feature vector obtained after feature extraction from the input sequence; This represents the extraction operation of a temporal convolutional network. Indicates a historical time window;

[0106] Graph attention networks are used to model the propagation of drought between regions:

[0107] (8);

[0108] In the formula, Indicates the region At any moment Drought prediction results; Indicates the region The set of neighboring regions; Indicates the region Its neighboring areas The intensity of the impact; Indicates the region At any moment Historical soil moisture data was extracted using a temporal convolutional network.

[0109] Furthermore, the primary objective of the spatiotemporally coupled drought prediction model is drought level classification, while the secondary objective is soil moisture numerical regression.

[0110] The preferred numerical regression expression for soil moisture is as follows:

[0111] (9);

[0112] In the formula, A measure of the degree of regression after soil moisture numerical values ​​have been regressed; Indicates the total number of regions; Indicates an index variable; This indicates the predicted soil moisture value; This indicates the actual measured soil moisture value.

[0113] S4. Input the real-time collected wind and cloud meteorological data and MODIS optical image impact data into the spatiotemporal drought coupling prediction model constructed in step S3 to obtain the dynamic drought index.

[0114] The dynamic drought index mentioned in step S4 The expression is as follows:

[0115] (10);

[0116] In the formula, Indicates the actual leaf area index; Indicates the potential leaf area index; Represents the logistic regression probability;

[0117] in,

[0118] (11);

[0119] (12);

[0120] In the formula, Indicates the maximum leaf surface area index; , , and All of these represent regression coefficients.

[0121] In this embodiment, it is set to when When the alert level is set to red, the corresponding response measures are: to initiate emergency irrigation and allocate water resources; when The alert level is set at orange, and the corresponding response measures are: restricting non-agricultural water use; when When the alert level is set to yellow, the corresponding response is to issue a drought warning.

[0122] S5. Verify the accuracy of the dynamic drought index obtained in step S4 using ground station data, and combine it with the federated learning framework to realize the dynamic update of the spatiotemporal drought coupling prediction model.

[0123] The federated learning framework described in step S5 implements the following expression for dynamic model update:

[0124] (13);

[0125] In the formula, Indicates the global warning value; Indicates the total number of iterations; Indicates the first The number of data points used in the next iteration; Indicates the first The warning value calculated in the next iteration; This indicates the amount of weight update.

[0126] Simulation Experiment

[0127] Data preparation: Collect wind and cloud meteorological data (including microwave soil moisture data and MODIS optical image data (vegetation index, surface temperature) from different regions and time periods (the time span and spatial range of the data should be representative, such as selecting data from multiple climate zones and different topographic regions, covering both dry and wet periods in time). At the same time, obtain the measured soil moisture data, actual leaf area index, accumulated temperature, and other data from the corresponding ground stations for model training, validation, and result comparison.

[0128] Environment Setup: Set up the necessary software environment (Python programming environment) for the experiment, including the programming language, deep learning framework, and libraries related to data processing and analysis. On the hardware side, equip the lab with a computer with sufficient computing power.

[0129] The all-weather soil moisture map generated using the early warning method described in this invention is as follows: Figure 2 As shown.

[0130] Table 1. Early Warning Accuracy Results

[0131]

[0132] Based on Table 1 Figure 2 It can be seen that the early warning method described in this invention has a minimum overlap of 85% with actual data, and its accuracy meets the requirements, proving the effectiveness of this invention.

[0133] Furthermore, during this process, as the number of training rounds increased, the accuracy of drought level classification gradually increased, with a rapid increase in the early stage and a tendency to stabilize in the later stage, reaching a stable level of over 85% at the end of training. This indicates that the model's ability to classify drought levels is continuously improving and becoming more stable.

[0134] Meanwhile, during the training process, the MSE gradually decreased, with a significant decrease in the early stage, followed by a slower rate of decrease, and eventually stabilized below 0.05, which means that the accuracy of the model in predicting soil moisture values ​​is continuously improving.

[0135] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A soil drought risk early warning method based on machine learning, characterized in that: Includes the following steps: S1. Collect wind and cloud meteorological data and MODIS optical image impact data, and extract features to form a dataset; S2. Use generative adversarial networks to repair cloud area data, integrate the wind and cloud meteorological data collected in step S1 and MODIS optical image impact data, eliminate cloud pollution interference, and generate an all-weather soil moisture map. S3. Based on the all-weather soil moisture map described in step S2, construct a spatiotemporal drought coupled prediction model; S4. Input the real-time collected wind and cloud meteorological data and MODIS optical image impact data into the spatiotemporal drought coupling prediction model constructed in step S3 to obtain the dynamic drought index. S5. Verify the accuracy of the dynamic drought index obtained in step S4 using ground station data, and combine it with the federated learning framework to realize the dynamic update of the spatiotemporal drought coupling prediction model. Step S2 specifically includes the following steps: S21. Data preprocessing: Unify the wind and cloud meteorological data and MODIS optical image impact data into the WGS84 coordinate system, and use MODIS cloud masks to mark the cloud-covered areas in the MODIS optical images. S22. Construct a generative adversarial network and train it using historical data; S23. Cloudy Area Restoration: Replace the original MODIS optical image data with cloudless optical images output by the generator of the generative adversarial network in the cloud-covered area, and at the same time, use Poisson fusion to eliminate stitching marks in the edge transition area of ​​the MODIS optical image data. S24. Using fused and restored MODIS optical imagery data and Fengyun meteorological satellite data, a 1km resolution all-weather soil moisture map was generated through a multivariate regression model. : （2）； In the formula, , , All represent regression coefficients; Indicates the meteorological-vegetation coupling index; Indicates the error term; This represents microwave soil moisture data; Indicates surface temperature; in, （3）； In the formula, and These represent the maximum and minimum values ​​of the vegetation index, respectively. This represents the actual evapotranspiration, which is the sum of plant transpiration and surface water evaporation. Represents the potential maximum evapotranspiration; Indicates vegetation index; S25. Verification and Iterative Optimization: Calculate the root mean square error between the measured soil moisture data from the weather station and the soil moisture predicted in step S24, and compare whether the root mean square error is within the set threshold range. If it is, it proves that the all-weather soil moisture map constructed in step S24 has passed the verification; otherwise, return to step S22 for update and iteration. The dynamic drought index mentioned in step S4 The expression is as follows: （10）； In the formula, Indicates the actual leaf area index; Indicates the potential leaf area index; Represents the logistic regression probability; in, （11）； （12）； In the formula, Indicates the maximum leaf surface area index; , , and All of these represent regression coefficients.

2. The soil drought risk early warning method based on machine learning according to claim 1, characterized in that: In step S1, the Fengyun meteorological satellite data includes microwave soil moisture data. ; MODIS optical image data includes vegetation index. and surface temperature ; Noise was eliminated and drought cycle features at different time scales were extracted using wavelet decomposition. （1）； In the formula, This represents the decomposed drought abrupt change signal and drought cycle trend component; Indicates the total time; Represents the original time series; Represents the discrete wavelet basis functions. Indicates the number of decomposition layers. This represents the translation parameter.

3. The soil drought risk early warning method based on machine learning according to claim 2, characterized in that: The generator of the generative adversarial network adopts a U-Net architecture, and the input of the generator is cloud-covered MODIS optical imagery. Hefengyun meteorological data Output cloudless optical images ; The discriminator in the generative adversarial network employs a convolutional neural network to distinguish between real, cloudless optical images. Output cloudless optical images Differences; And the total loss function of the generative adversarial network The expression is as follows: （4）； In the formula, Indicates resistance to loss; Indicates physical constraint loss; Indicates the weight of physical constraints; in, （5）； （6）； In the formula, This represents the actual data output by the discriminator; This represents the expected value operator; This represents the probability that false data is judged as true. Indicated based on vegetation index and surface temperature Soil moisture inversion function.

4. The soil drought risk early warning method based on machine learning according to claim 3, characterized in that: The spatiotemporal coupled drought prediction model described in step S3 is constructed based on a temporal convolutional network and a graph attention network, wherein the temporal convolutional network is used to extract historical soil moisture. Long-range dependency: （7）； In the formula, This indicates that the temporal convolutional network is at time step [time]. The feature vector obtained after feature extraction from the input sequence; This represents the extraction operation of a temporal convolutional network. Indicates a historical time window; Graph attention networks are used to model the propagation of drought between regions: （8）； In the formula, Indicates the area At any moment Drought prediction results; Indicates the area The set of neighboring regions; Indicates the area Its neighboring areas The intensity of the impact; Indicates the area At any moment Historical soil moisture data was extracted using a temporal convolutional network. Furthermore, the primary objective of the spatiotemporally coupled drought prediction model is drought level classification, while the secondary objective is soil moisture numerical regression.

5. The soil drought risk early warning method based on machine learning according to claim 4, characterized in that: The numerical regression expression for soil moisture is as follows: （9）； In the formula, A measure of the degree of regression after soil moisture numerical values ​​have been regressed; Indicates the total number of regions; Indicates an index variable; This indicates the predicted soil moisture value; This indicates the actual measured soil moisture value.

6. The soil drought risk early warning method based on machine learning according to claim 5, characterized in that: The federated learning framework described in step S5 implements the following expression for dynamic model update: （13）； In the formula, Indicates the global warning value; Indicates the total number of iterations; Indicates the first The number of data points used in the next iteration; Indicates the first The warning value calculated in the next iteration; This indicates the amount of weight update.

7. An early warning system for implementing the machine learning-based soil drought risk early warning method according to any one of claims 1-6, characterized in that: It includes a data acquisition module, a data repair and fusion module, a spatiotemporal drought coupled prediction model construction module, and a real-time early warning and model update module; The data acquisition module is used to collect wind and cloud meteorological data and MODIS optical image data. The data repair and fusion module is used to repair cloud area data using generative adversarial networks and fuse wind and cloud meteorological data and MODIS optical image data to generate an all-weather soil moisture map. The spatiotemporal drought coupled prediction model construction module is used to construct a spatiotemporal drought coupled prediction model based on all-weather soil moisture maps; The real-time early warning and model update module is used to input real-time collected wind and cloud meteorological data and MODIS optical image data into the spatiotemporal drought coupling prediction model to obtain the dynamic drought index. The accuracy is verified by ground station data, and the spatiotemporal drought coupling prediction model is dynamically updated by combining the federated learning framework.

Images