Soil moisture assimilation reconstruction and prediction method and system based on multi-source data collaboration

Abstract

This invention provides a method and system for multi-source data collaborative soil moisture assimilation reconstruction and prediction, belonging to the field of data processing technology. The method includes: obtaining multi-layered isometric soil moisture estimates based on satellite remote sensing imagery; constructing a prediction model; dividing historical periods into multiple time windows based on all satellite transit dates; generating daily soil moisture reconstruction values ​​within each time window based on daily soil moisture monitoring data and the prediction model; concatenating all soil moisture reconstruction values ​​in chronological order to generate a historical soil moisture sequence; and using the soil moisture reconstruction value at the end of the last time window as the initial state, generating a predicted soil moisture sequence based on the prediction model and meteorological forecast data. This invention constructs time windows with satellite transit dates as boundaries and employs a bidirectional assimilation algorithm to fuse discrete isometric observations and continuous point observations, thereby generating a multi-layered historical soil moisture reconstruction sequence and achieving high-precision short-term prediction.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and system for reconstructing and predicting soil moisture using multi-source data collaboration. Background Technology

[0002] Accurate acquisition of soil moisture information is crucial for agricultural production and water resource management. Currently, data is mainly obtained through ground stations and satellite remote sensing. Ground stations can provide continuous and accurate point data, but their spatial representativeness is insufficient; satellite remote sensing can acquire large-area spatial information, but its revisit cycle and weather influences result in temporal discontinuity, and it is difficult to detect deep moisture.

[0003] Existing technologies attempt to combine these two types of data, but most employ simple fusion strategies. These methods struggle to effectively reconcile the differences in spatiotemporal scales and physical mechanisms between the two types of data, resulting in data gaps in the generated soil moisture products, failing to achieve true spatiotemporal continuity and high accuracy. Therefore, how to systematically fuse multi-source data to achieve continuous reconstruction and accurate prediction of soil moisture with high spatiotemporal resolution has become an urgent technical problem to be solved. Summary of the Invention

[0004] This invention provides a method and system for soil moisture assimilation reconstruction and prediction based on multi-source data collaboration, in order to overcome the deficiencies in the prior art and achieve continuous reconstruction and accurate prediction of soil moisture with high spatiotemporal resolution.

[0005] This invention provides a method for soil moisture assimilation reconstruction and prediction based on multi-source data collaboration, comprising the following steps:

[0006] Based on satellite remote sensing images, the estimated values ​​of multi-layered soil moisture in the areal layer were obtained for each satellite transit day.

[0007] Based on soil moisture monitoring data from various monitoring stations over historical periods, a soil moisture change rate prediction model was constructed.

[0008] Based on all the aforementioned satellite transit dates, the historical period is divided into multiple time windows;

[0009] Within each time window, the estimated values ​​of multi-layered soil moisture on the start and end days of the time window are used as boundary constraints, and the reconstructed values ​​of soil moisture for each day within the time window are generated based on the soil moisture monitoring data for each day within the time window and the soil moisture change rate prediction model.

[0010] The reconstructed soil moisture values ​​of all the time windows are spliced ​​together in chronological order to generate the historical soil moisture sequence for the historical period.

[0011] Using the soil moisture reconstruction value at the end of the last time window as the initial state, a soil moisture prediction sequence for a future preset period is iteratively generated based on the soil moisture change rate prediction model and future weather forecast data.

[0012] According to the present invention, a method for multi-source data collaborative soil moisture assimilation reconstruction and prediction is provided. In each time window, the method uses the estimated areal multi-layer soil moisture values ​​of the start and end dates of the time window as boundary constraints, and generates daily soil moisture reconstruction values ​​within the time window based on the soil moisture monitoring data for each day within the time window and the soil moisture change rate prediction model. The method includes:

[0013] Within each time window, the estimated value of the multi-layered soil moisture on the starting day is used as the initial state. For each day in the forward time series, the predicted soil moisture change rate of each monitoring station output by the soil moisture change rate prediction model and the soil moisture monitoring data of the monitoring station on that day are used to obtain the soil moisture analysis value for that day. The forward time series is from the day after the starting day to the ending day.

[0014] For each day in the reverse time series, the soil moisture analysis value of the day is smoothed and optimized based on the soil moisture analysis value of the next day to obtain the reconstructed soil moisture value for each day within the time window; the reverse time series is from the end date to the next day after the start date.

[0015] According to the multi-source data collaborative soil moisture assimilation reconstruction and prediction method provided by the present invention, for each day in the positive time series, the soil moisture analysis value for that day is obtained based on the predicted soil moisture change rate of each monitoring station output by the soil moisture change rate prediction model and the soil moisture monitoring data of the monitoring stations on that day, including:

[0016] For each day from the day after the start date to the day before the end date, spatial interpolation is performed on the predicted soil moisture change rate for all the monitoring stations to obtain the areal soil moisture change rate field.

[0017] Based on the soil moisture status of the previous day and the isometric soil moisture change rate field, the predicted soil moisture value for the current day is calculated.

[0018] The predicted soil moisture value is assimilated with the soil moisture monitoring data from the monitoring stations on that day to obtain the soil moisture analysis value for that day.

[0019] For the end date, the predicted soil moisture value is simultaneously assimilated with the soil moisture monitoring data from the monitoring stations on that day and the estimated areal multilayer soil moisture value for the end date to obtain the soil moisture analysis value for the end date.

[0020] According to the present invention, a method for multi-source data collaborative soil moisture assimilation reconstruction and prediction is provided, wherein the method for obtaining multi-layered areal soil moisture estimates for each satellite transit day based on satellite remote sensing imagery includes:

[0021] Based on the satellite remote sensing images, the surface soil moisture of each of the satellite transit days is retrieved and generated.

[0022] For any of the satellite transit days, soil moisture at multiple depths is obtained based on the surface soil moisture.

[0023] The estimated values ​​of the multi-layered soil moisture are obtained based on the surface soil moisture and the soil moisture of all depth layers.

[0024] According to the present invention, a method for multi-source data collaborative soil moisture assimilation reconstruction and prediction is provided, wherein the method for constructing a soil moisture change rate prediction model based on soil moisture monitoring data from various monitoring stations within a historical time period includes:

[0025] For any of the monitoring stations, based on the daily soil moisture monitoring data within the historical period, calculate the daily soil moisture change rate label for each soil depth layer;

[0026] Obtain daily historical meteorological data within the specified historical period;

[0027] Based on the historical meteorological data and the daily soil moisture change rate labels of each soil depth layer, the machine learning model corresponding to each soil depth layer is trained to obtain the soil moisture change rate prediction model corresponding to each soil depth layer.

[0028] According to the present invention, a method for multi-source data collaborative soil moisture assimilation reconstruction and prediction includes training a machine learning model corresponding to each soil depth layer based on the historical meteorological data and the daily soil moisture change rate labels of each soil depth layer to obtain a soil moisture change rate prediction model corresponding to each soil depth layer, comprising:

[0029] For any of the soil depth layers, the historical meteorological data is input into the machine learning model to obtain a sample of the predicted soil moisture change rate output by the machine learning model;

[0030] The machine learning model is trained based on the error between the predicted soil moisture change rate sample and the daily soil moisture change rate label to obtain the soil moisture change rate prediction model.

[0031] This invention also provides a multi-source data collaborative soil moisture assimilation reconstruction and prediction system, comprising the following modules:

[0032] The first processing module is used to obtain the estimated values ​​of multi-layered soil moisture on each satellite transit day based on satellite remote sensing images.

[0033] The second processing module is used to construct a soil moisture change rate prediction model based on soil moisture monitoring data from each monitoring station during historical periods.

[0034] The third processing module is used to divide the historical period into multiple time windows based on all the satellite transit days;

[0035] The fourth processing module is used to generate daily soil moisture reconstruction values ​​within each time window, using the estimated values ​​of multi-layered soil moisture on the start and end days of the time window as boundary constraints, and based on the soil moisture monitoring data for each day within the time window and the soil moisture change rate prediction model.

[0036] The fifth processing module is used to splice the reconstructed soil moisture values ​​of all the time windows in chronological order to generate the historical soil moisture sequence for the historical period.

[0037] The sixth processing module is used to take the soil moisture reconstruction value at the end of the last time window as the initial state, and iteratively generate a soil moisture prediction sequence for a future preset period based on the soil moisture change rate prediction model and future weather forecast data.

[0038] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the multi-source data collaborative soil moisture assimilation reconstruction and prediction method as described above.

[0039] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the multi-source data collaborative soil moisture assimilation reconstruction and prediction method as described above.

[0040] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the multi-source data collaborative soil moisture assimilation reconstruction and prediction method as described above.

[0041] In summary, one or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0042] By obtaining multi-layered isometric soil moisture estimates for each satellite transit day based on satellite remote sensing imagery, high spatial resolution discrete observation anchor points are formed on the time axis, providing accurate boundary conditions for subsequent time-series reconstruction. A soil moisture change rate prediction model is constructed based on soil moisture monitoring data from various monitoring stations within historical time periods, establishing a dynamic evolution model driven by meteorological factors and possessing clear physical meaning, providing a foundation for subsequent state extrapolation and future prediction in the time dimension. By dividing the historical time period into multiple time windows based on all satellite transit days, a segmented assimilation framework is established in the time dimension, enabling each time window to independently complete optimal estimation under strong observational constraints. Within each time window, multi-layered isometric soil moisture estimates on the start and end days are used as boundary constraints, combined with monitoring data from each day within the window and the change rate prediction model to generate daily reconstructed soil moisture values. This integrates spatially continuous satellite observations with temporally continuous ground observations, significantly improving the spatiotemporal continuity and accuracy of the historical sequence. By stitching together the reconstructed values ​​from all time windows in chronological order, a complete historical soil moisture sequence is generated, thus achieving continuous soil moisture reconstruction across all time periods and spatial dimensions. Using the soil moisture reconstruction value at the end of the last time window as the initial state, and combining iterative generation of future soil moisture prediction sequences with a rate of change prediction model and future weather forecast data, a multi-layered short-term soil moisture forecast for the future is achieved while maintaining the physical consistency of the historical reconstruction, thereby improving the accuracy of the forecast. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0044] Figure 1 This is one of the flowcharts of the multi-source data collaborative soil moisture assimilation reconstruction and prediction method provided by the present invention.

[0045] Figure 2 This is the second flowchart of the multi-source data collaborative soil moisture assimilation reconstruction and prediction method provided by the present invention.

[0046] Figure 3 This is the third flowchart of the multi-source data collaborative soil moisture assimilation reconstruction and prediction method provided by the present invention.

[0047] Figure 4 This is the fourth flowchart of the multi-source data collaborative soil moisture assimilation reconstruction and prediction method provided by the present invention.

[0048] Figure 5 This is the fifth flowchart of the multi-source data collaborative soil moisture assimilation reconstruction and prediction method provided by the present invention.

[0049] Figure 6 This is the sixth flowchart of the multi-source data collaborative soil moisture assimilation reconstruction and prediction method provided by the present invention.

[0050] Figure 7 This is a schematic diagram of the structure of the multi-source data collaborative soil moisture assimilation reconstruction and prediction system provided by the present invention.

[0051] Figure 8 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0053] It should be noted that in the description of this invention, the terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element. The terms "upper," "lower," etc., indicating orientation or positional relationships according to the accompanying drawings, are only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the system or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0054] The terms "first," "second," etc., used in this invention are used to distinguish similar objects, not to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class, without limiting the number of objects; for example, a first object can be one or more. Furthermore, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0055] The following is combined Figures 1-8 This invention describes the method, system, electronic device, storage medium, and computer program product for multi-source data collaborative soil moisture assimilation reconstruction and prediction.

[0056] Reference Figure 1 , Figure 1 This is one of the flowcharts illustrating the multi-source data collaborative soil moisture assimilation reconstruction and prediction method provided by the present invention, such as... Figure 1 As shown, steps 101 to 106 are included:

[0057] Step 101: Based on satellite remote sensing imagery, obtain the estimated values ​​of multi-layered soil moisture on each satellite transit day.

[0058] In this embodiment, satellite remote sensing imagery refers to Earth surface image data covering the target study area acquired by satellite sensors, such as C-band synthetic aperture radar imagery and high-resolution multispectral imagery.

[0059] Satellite transit date refers to the specific date on which a satellite flies over the research area and acquires valid image data.

[0060] The multi-layered planar soil moisture estimate refers to a spatial distribution dataset of soil moisture content, characterized in two-dimensional raster data form, covering the entire study area and containing multiple preset soil depth layers, on a specified satellite transit date. Specifically, the multi-layered planar soil moisture estimate can be divided into surface soil moisture and depth layer soil moisture. The surface layer is preferably 0-10 cm deep, and the depth layers are preferably 10-20 cm, 20-30 cm, or 30-40 cm deep.

[0061] In one specific implementation, when performing step 101, satellite remote sensing images of the study area on different satellite transit days are first acquired.

[0062] First, a series of preprocessing operations are performed on the acquired raw multi-source satellite remote sensing images, including but not limited to thermal noise removal, radiometric calibration, atmospheric correction, terrain correction, and image filtering. The purpose is to eliminate various types of noise and distortion and obtain standardized surface reflectance or backscattering coefficient data.

[0063] Next, various remote sensing features related to soil moisture, such as backscattering coefficient, vegetation index, and moisture index, are extracted from the preprocessed multi-source satellite remote sensing imagery. Then, a comprehensive processing model is used, taking the extracted remote sensing features as input, to perform calculations and spatial extrapolation, thereby generating a multi-layered planar soil moisture estimate covering the entire study area on the current satellite transit day. This process is executed independently on each satellite transit day, ultimately forming a set of temporally discrete but spatially continuous multi-layered planar soil moisture estimates.

[0064] In one specific implementation, refer to Figure 2 , Figure 2 This is the second flowchart of the multi-source data collaborative soil moisture assimilation reconstruction and prediction method provided by the present invention. Step 101 specifically includes steps 201 to 203:

[0065] First, perform step 201: Based on satellite remote sensing images, invert and generate the areal surface soil moisture for each satellite transit day.

[0066] The surface soil moisture refers to the spatial distribution map of surface soil (e.g., 0-10 cm depth) moisture content, characterized in the form of two-dimensional raster data, covering the entire study area.

[0067] This step involves first acquiring and preprocessing Sentinel-1 radar imagery and Sentinel-2 optical imagery. For Sentinel-1 imagery, the preprocessing workflow includes thermal noise removal, radiometric calibration, terrain correction, and Frost filtering, followed by extraction of the incident angle, VV, and VH backscattering coefficient features. For Sentinel-2 imagery, the preprocessing workflow includes cloud removal, radiometric calibration, atmospheric correction, and orthorectification, followed by extraction and calculation of vegetation indices such as NDVI and NDWI.

[0068] Then, based on the preprocessed remote sensing features and surface soil moisture data collected from ground-based sampling points, an inversion model was selected and constructed. This inversion model employs a mechanism-machine learning coupled model, specifically a water cloud model coupled with a random forest model. This coupled model takes the intermediate variables of the water cloud model, along with features such as vegetation index and backscattering coefficient, as input, and trains and inverts them using the random forest algorithm to generate the areal surface soil moisture for each satellite transit day.

[0069] Next, step 202 is performed: for any satellite transit day, soil moisture at multiple depths is obtained based on the surface soil moisture.

[0070] Soil moisture at multiple depths refers to the spatial distribution map of soil moisture content at different soil depths (e.g., 10-20 cm, 20-30 cm, 30-40 cm) on a specified satellite transit date. This step requires establishing a surface-to-depth statistical relationship model for each satellite transit date, using the areal surface soil moisture retrieved on that day and the corresponding depth soil moisture data collected from all ground-based sampling points. This model incorporates auxiliary variables such as soil texture and vegetation index to improve the accuracy of the estimation. This statistical relationship model can be expressed as the following formula:

[0071]

[0072] In the formula, Represents deep soil moisture. Represents surface soil moisture. These represent auxiliary variables, such as soil texture, vegetation index, air temperature and humidity, and meteorological factors such as rainfall. This represents the established statistical model (e.g., a machine learning model). This represents the model error.

[0073] Train an independent statistical model for each depth layer. Then, by inputting the surface soil moisture (SSM) data for the day the satellite passes over and the corresponding auxiliary variable layer data into the model, the spatial distribution map of soil moisture at the corresponding depth on that day can be calculated, thus obtaining the soil moisture at multiple depth layers.

[0074] Finally, step 203 is performed: based on the surface soil moisture and the soil moisture of all depth layers, the estimated values ​​of the multi-layered surface soil moisture are obtained.

[0075] In this step, the areal surface soil moisture generated in step 201 is spatially integrated with the spatial distribution maps of soil moisture generated for each deep soil depth in step 202. Specifically, these two-dimensional raster data layers representing different depths are stacked vertically to form a multi-band raster data file. In this file, each band (or layer) represents the spatial distribution of moisture at a specific soil depth. After integration, the final dataset is the estimated areal multi-layer soil moisture for the satellite transit date.

[0076] Step 102: Based on soil moisture monitoring data from each monitoring station during historical periods, construct a soil moisture change rate prediction model.

[0077] In this embodiment, a monitoring station refers to a ground-based sensor station deployed within the study area that can continuously monitor and transmit the moisture content of multiple soil depth layers in real time. Soil moisture monitoring data refers to continuous, stratified time-series soil moisture data collected by the monitoring station and preprocessed (e.g., outlier removal and imputation). The soil moisture change rate prediction model is a computational model that can predict the relative change rate of soil moisture at each depth layer on a daily basis based on input meteorological data.

[0078] In one specific implementation, when performing step 102, soil moisture monitoring data from all monitoring stations covering the entire historical period are first collected and organized. Based on this continuous monitoring data, the daily changes in soil moisture at each soil depth are calculated, and these changes are quantified into a specific rate of change index. Simultaneously, meteorological data corresponding to the historical period, such as daily maximum temperature, daily minimum temperature, rainfall, and air humidity, are acquired.

[0079] Then, using the calculated daily soil moisture change rate as the training objective, and the corresponding historical meteorological data and other auxiliary information (such as accumulated days) as input features, a machine learning algorithm (such as a Long Short-Term Memory network, LSTM) is employed to train the model. This training process is conducted independently for each soil depth layer, aiming to learn and establish the dynamic response relationship between meteorological driving factors and the soil moisture change rate at each layer. After training, a set of soil moisture change rate prediction models for different soil depth layers is finally obtained.

[0080] In one specific implementation, refer to Figure 3 , Figure 3 This is the third flowchart of the multi-source data collaborative soil moisture assimilation reconstruction and prediction method provided by the present invention. Step 102 specifically includes steps 301 to 303:

[0081] First, perform step 301: For any monitoring station, calculate the daily soil moisture change rate label for each soil depth layer based on the daily soil moisture monitoring data within the historical time period.

[0082] The daily soil moisture change rate label is a supervised learning target value used to train machine learning models. The soil moisture change rate is the relative change in soil moisture value on a given day compared to the previous day. This step involves first obtaining the daily average soil moisture values ​​for each soil depth layer (e.g., layer i) from any monitoring station over a historical period. Then, the daily soil moisture change rate label is calculated using the following formula. :

[0083]

[0084] in, and These represent the soil moisture values ​​of the i-th soil layer at the monitoring station on historical day t and day t-1, respectively. This calculation is performed daily for each soil depth layer at each monitoring station throughout the entire historical period, ultimately generating a daily soil moisture change rate label sequence for each station and each depth layer, corresponding to the long-term series of soil moisture monitoring data.

[0085] Next, proceed to step 302: obtain the historical meteorological data for each day within the historical time period.

[0086] Historical meteorological data refers to daily recorded meteorological element information corresponding to historical periods. This step involves obtaining daily meteorological records for the study area within the historical period from meteorological departments or professional data platforms. This data includes, but is not limited to, daily maximum temperature, daily minimum temperature, daily average temperature, rainfall, air humidity, air pressure, and solar radiation. Because the study area is relatively small (e.g., a small farm), these meteorological elements are considered spatially consistent.

[0087] Finally, step 303 is executed: based on historical meteorological data and the daily soil moisture change rate labels of each soil depth layer, the machine learning model corresponding to each soil depth layer is trained to obtain the soil moisture change rate prediction model corresponding to each soil depth layer.

[0088] When performing this step, a separate machine learning model (e.g., a Long Short-Term Memory (LSTM) network model) is assigned to each soil depth layer (i-th layer), denoted as . Then, the historical meteorological data and auxiliary information such as Day of Year (DOY) obtained in step 302 are used as input features of the model, and the daily soil moisture change rate label sequence for the corresponding depth layer calculated in step 301 is used as the output target of the model. The machine learning model is evaluated by minimizing the error between the model's predicted values ​​and the true label values. Training is then performed. This training process is repeated for each soil depth layer to obtain the soil moisture change rate prediction model for each soil depth layer.

[0089] In one specific implementation, refer to Figure 4 , Figure 4 This is the fourth flowchart of the multi-source data collaborative soil moisture assimilation reconstruction and prediction method provided by the present invention. Step 303 specifically includes steps 401 to 402:

[0090] First, perform step 401: For any soil depth layer, input historical meteorological data into the machine learning model to obtain a sample of the predicted soil moisture change rate output by the machine learning model.

[0091] In this embodiment, the predicted soil moisture change rate sample refers to the unoptimized predicted soil moisture change rate generated by the machine learning model during the training process based on a set of input historical meteorological data. When performing this step, for any soil depth layer (e.g., the i-th layer), historical meteorological data for one day (e.g., day t) is selected from the historical dataset, including daily average temperature, daily maximum temperature, daily cumulative rainfall, and auxiliary features such as annual accumulated days.

[0092] These feature data are organized into an input vector and fed into the machine learning model currently in training for that depth layer. The model calculates the input vector and generates an output value, which is the predicted soil moisture change rate sample for that day. This process can be represented by the following formula:

[0093]

[0094] in, Indicates the average daily temperature. Indicates the highest daily temperature, Indicates the daily cumulative rainfall. For annual accumulated days, other relevant meteorological driving factors can also be included as input to the machine learning model, which can be selected according to the actual situation; This represents the model error.

[0095] Next, step 402 is performed: the machine learning model is trained based on the error between the predicted soil moisture change rate sample and the daily soil moisture change rate label to obtain the soil moisture change rate prediction model.

[0096] When performing this step, the predicted soil moisture change rate sample generated in step 401 will be used. The daily soil moisture change rate labels are the same as those calculated in step 301 for the same date (day t) and the same depth layer (layer i). The two inputs are compared, the error between them is calculated, and a loss function is constructed based on this error, such as mean squared error (MSE) or cross-entropy loss. The specific loss function can be selected according to the actual situation, and no restrictions are imposed in this application. Then, an optimization algorithm (such as backpropagation) is used to adjust the internal parameters of the machine learning model, so that when the machine learning model makes a prediction on the same input in the next time, its output can be closer to the true label value.

[0097] This process is iterated over and over again on the training dataset throughout the historical period until the overall error of the model on all samples converges to below a preset threshold. Once the training process is complete, the machine learning model becomes the final model for predicting the rate of change of soil moisture at that soil depth.

[0098] Step 103: Divide the historical period into multiple time windows based on all satellite transit dates.

[0099] In this embodiment, a time window refers to a continuous time interval defined by two consecutive satellite transit days as the start and end dates.

[0100] In one specific implementation, when performing step 103, firstly, all satellite transit dates that acquired high-quality, valid imagery throughout the entire historical period are statistically identified. Then, these satellite transit dates are arranged in chronological order and denoted as... Based on this arrangement, the entire historical period is divided into multiple consecutive and independent sub-intervals, forming multiple time windows. Specifically, the first time window is... The second time window is This continues until the last time window. Each time window represents an independent unit for subsequent data assimilation processing.

[0101] Step 104: Within each time window, the estimated values ​​of multi-layered soil moisture on the start and end days of the time window are used as boundary constraints. Based on the soil moisture monitoring data and the soil moisture change rate prediction model for each day within the time window, the reconstructed values ​​of soil moisture for each day within the time window are generated.

[0102] In this embodiment, the boundary constraint refers to the spatially continuous, multi-layered planar soil moisture estimate generated in step 101 on the start and end dates of each time window. This estimate serves as a mandatory reference for the initial and final states of the assimilation process within the window. The reconstructed soil moisture value refers to the optimized, diurnal, multi-layered planar soil moisture state estimate obtained within the time window after fusing the window boundary constraint and the daily soil moisture monitoring data within the window.

[0103] In one specific implementation, when executing step 104, a data assimilation algorithm is employed for any given time window. This data assimilation algorithm uses the estimated isomorphic multi-layer soil moisture value on the starting day of the time window as the initial state. Then, the state evolves day by day within the time window, driven by the soil moisture change rate prediction model constructed in step 102. On each day of evolution, the algorithm integrates soil moisture monitoring data from all monitoring stations on that day to correct the state predicted by the model. When calculating the final reconstructed soil moisture value for any day within the window (denoted as day t), the assimilation algorithm utilizes not only all observation information from the starting day to day t, but also all observation information from the day after day t to the ending day, especially the strong boundary constraint of the isomorphic multi-layer soil moisture estimate on the ending day.

[0104] By integrating and optimizing this two-way information, we can ensure that the soil moisture reconstruction value for each day within the window is the best comprehensive estimate of all available data within the entire window (including the starting boundary, ending boundary, and internal station data).

[0105] In one specific implementation, refer to Figure 5 , Figure 5 This is the fifth flowchart of the multi-source data collaborative soil moisture assimilation reconstruction and prediction method provided by the present invention. Step 104 specifically includes steps 501 to 502:

[0106] Step 501: Within each time window, using the estimated value of the multi-layered soil moisture on the starting day as the initial state, for each day in the forward time series, based on the predicted soil moisture change rate of each monitoring station output by the soil moisture change rate prediction model and the soil moisture monitoring data of the monitoring station on that day, the soil moisture analysis value for that day is obtained; the forward time series is from the next day after the starting day to the ending day.

[0107] In this embodiment, the initial state refers to the estimated isometric soil moisture value generated in step 101 on the starting date of the time window. To characterize the uncertainty of the initial state, a set of initial states is generated by introducing random perturbations into the initial state.

[0108] A forward time series is a date sequence that starts from the day after the beginning of the time window and increases day by day until the end of the time window.

[0109] Predicting the rate of change of soil moisture refers to the predicted value of the relative rate of change of soil moisture at each monitoring station on each day in the future, after inputting future meteorological forecast data into the soil moisture rate of change prediction model trained in step 102.

[0110] Soil moisture analysis value refers to the posterior estimate of soil moisture status obtained by combining the soil moisture monitoring data of that day, after two stages of prediction and data assimilation correction for each day of the positive time series.

[0111] In one specific implementation, refer to Figure 6 , Figure 6 This is the sixth flowchart of the multi-source data collaborative soil moisture assimilation reconstruction and prediction method provided by the present invention. Step 501 specifically includes steps 601 to 604:

[0112] First, perform step 601: For each day from the day after the start date to the day before the end date, spatially interpolate the predicted soil moisture change rate for all monitoring stations to obtain the areal soil moisture change rate field.

[0113] A planar soil moisture change rate field refers to a spatial distribution map of the predicted relative change rate of soil moisture, characterized in the form of two-dimensional raster data and covering the entire study area. In this step, for any soil depth layer, the predicted soil moisture change rate from all monitoring stations, output by the soil moisture change rate prediction model, is used as discrete point observations. Then, a spatial interpolation method, such as the inverse distance weighting (IDW) or Kriging, is used to interpolate these discrete point values, thereby generating a planar soil moisture change rate field covering the entire study area at that soil depth layer.

[0114] Next, step 602 is performed: based on the soil moisture status and the areal soil moisture change rate field of the previous day, the predicted soil moisture value for the current day is calculated.

[0115] Soil moisture prediction refers to a priori estimate of soil moisture status obtained based on a physical dynamic model before the fusion of actual observation data for the day. When performing this step, based on the soil moisture status of the previous day (for the day after the starting day, this status is the initial state; for other dates, it is the soil moisture analysis value of the previous day), and combined with the daily areal soil moisture change rate field generated in step 601, a dynamic update model is used to calculate the daily soil moisture prediction value pixel by pixel.

[0116] Then, for each day from the day after the start date to the day before the end date, step 603 is performed: the predicted soil moisture value is assimilated with the soil moisture monitoring data of the monitoring station on that day to obtain the soil moisture analysis value for that day.

[0117] In this step, the predicted values ​​for all monitoring stations are first extracted from the predicted soil moisture values ​​obtained in step 602. Then, these extracted predicted values ​​are compared with the actual soil moisture monitoring data collected by each monitoring station on that day. Finally, based on the differences between the two, a data assimilation algorithm (such as ensemble Kalman filtering) is used to correct the predicted soil moisture values ​​for the entire area, thereby obtaining the soil moisture analysis values ​​for that day.

[0118] Finally, for the end date, perform step 604: simultaneously assimilate the predicted soil moisture value with the soil moisture monitoring data from the monitoring stations on that day and the estimated areal multilayer soil moisture value for the end date to obtain the soil moisture analysis value for the end date.

[0119] In this step, the predicted soil moisture value for the end date calculated in step 602 is combined with the soil moisture monitoring data from all monitoring stations on that day and the estimated areal multi-layer soil moisture value for the end date for joint data assimilation. This joint assimilation process also employs a data assimilation algorithm, which corrects the predicted soil moisture value by simultaneously fusing point-based station observations and areal satellite inversion observations, thus obtaining the final soil moisture analysis value for the end date.

[0120] Step 502: For each day in the reverse time series, the soil moisture analysis value of the day is smoothed and optimized based on the soil moisture analysis value of the next day to obtain the reconstructed soil moisture value of each day within the time window; the reverse time series is from the end date to the next day after the start date.

[0121] In this embodiment, the reverse time series refers to a date sequence arranged in reverse chronological order, with the processing order starting from the day before the end of the time window and decreasing day by day until the beginning of the time window. The soil moisture reconstruction value refers to the final isometric multilayer soil moisture estimate for each day within the window after the reverse smoothing optimization process.

[0122] In one specific implementation, when performing this step, a backward smoothing operation is performed based on the soil moisture analysis values ​​generated for each day within the time window in step 501. This operation is performed on each day (denoted as day t) in the reverse time series.

[0123] When processing day t, the algorithm first calculates a smoothing gain matrix. This smoothing gain matrix, based on the covariance information of the state set saved during the forward filtering process, is used to quantitatively assess the influence weight of the state on the next day (i.e., day t+1) on the state on day t. Then, using this smoothing gain matrix, the smoothed and optimized state information of day t+1 (i.e., the reconstructed soil moisture value of day t+1) is fed back and used to correct the soil moisture analysis value on day t.

[0124] Specifically, the difference between the reconstructed soil moisture value on day t+1 and the predicted value on day t+1 based on the state prediction on day t is calculated. This difference represents the correction information introduced by the end date of the time window and subsequent observations. Then, this correction information is applied to the soil moisture analysis value on day t using a smoothing gain matrix to obtain the reconstructed soil moisture value on day t. This process is performed in reverse order, starting from the day before the end date.

[0125] In each step of the reverse iteration, the calculation of the day utilizes the results of the smoothing optimization completed on the next day, until the starting day is processed, and finally the daily soil moisture reconstruction values ​​for the entire time window are generated.

[0126] Step 105: Segment the reconstructed soil moisture values ​​of all time windows in chronological order to generate a historical soil moisture sequence for historical periods.

[0127] In this embodiment, the historical soil moisture sequence refers to a diurnal, areal, multi-layered soil moisture dataset that covers the entire historical period of the study and is seamlessly continuous in time. It constitutes a complete spatiotemporal reconstruction product of the historical soil moisture status of the study area.

[0128] In one specific implementation, when performing step 105, step 104 is for all independent time windows ( After generating daily soil moisture reconstruction values ​​for each time window, they were merged in chronological order. Specifically, the first time window was extracted first. The soil moisture reconstruction value is recorded daily. Then, the second time window is... The reconstructed soil moisture values ​​for each day are appended to the previous sequence. This stitching process is performed sequentially along the timeline, connecting the reconstructed sequences of all time windows one by one to form a single sequence covering the period from the initial satellite transit date. By the final satellite transit date The complete time series.

[0129] Step 106: Using the soil moisture reconstruction value at the end of the last time window as the initial state, generate a soil moisture prediction sequence for the future preset time period based on the soil moisture change rate prediction model and future weather forecast data.

[0130] In this embodiment, future weather forecast data refers to daily weather element forecast information obtained from meteorological departments that covers a preset future time period (e.g., the next 7 days), including daily maximum temperature, daily minimum temperature, and rainfall. The preset future time period refers to the length of time for which soil moisture prediction needs to be performed, such as the next week. The soil moisture prediction sequence refers to a areal, multi-layered soil moisture forecast dataset containing every day within the preset future time period. This sequence also includes quantitative information on the uncertainty of the forecast results.

[0131] In one specific implementation, when executing step 106, the reconstructed soil moisture value at the end of the last time window in step 105 is first used as the initial state for prediction. To quantify the uncertainty of the prediction, a set of initial states is generated by introducing random perturbations into this initial state. Then, for the first day of the future preset time period, the corresponding future weather forecast data is obtained and input into the soil moisture change rate prediction model constructed in step 102 to obtain the predicted soil moisture change rate for each monitoring station on that day. Next, spatial interpolation is performed on these point-like prediction values ​​to generate a surface soil moisture change rate field. Based on the initial state set and the change rate field, the soil moisture prediction set for the first day is calculated by dynamically updating the model. Then, the soil moisture prediction set generated on the first day is used as the new initial state, and the above process is repeated, i.e., the weather forecast data for the second day is obtained, the change rate field for the second day is calculated, and the soil moisture prediction set for the second day is updated. This process is repeated day by day until the calculation for the entire future preset time period is completed. Finally, the mean of the daily forecast set is used as the final soil moisture forecast value for that day, and the dispersion of the set is used as a measure of forecast uncertainty, thereby generating a complete soil moisture forecast sequence.

[0132] By executing step 106, this method, based on high-precision historical reconstruction, further provides forward prediction capabilities. Since the initial predicted state is an optimal estimate after complete bidirectional assimilation optimization, and the prediction process is driven by a rate-of-change model with clear physical meaning and future weather forecast data, the generated soil moisture prediction sequence has high accuracy and reliability.

[0133] Reference Figure 7 , Figure 7 This is a schematic diagram of the structure of the multi-source data collaborative soil moisture assimilation reconstruction and prediction system provided by the present invention. The system includes:

[0134] The first processing module is used to obtain the estimated values ​​of multi-layered soil moisture on each satellite transit day based on satellite remote sensing images.

[0135] The second processing module is used to construct a soil moisture change rate prediction model based on soil moisture monitoring data from each monitoring station during historical periods.

[0136] The third processing module is used to divide the historical period into multiple time windows based on all satellite transit days;

[0137] The fourth processing module is used to generate daily soil moisture reconstruction values ​​within each time window, using the estimated values ​​of multi-layered soil moisture on the start and end days of the time window as boundary constraints, and based on the soil moisture monitoring data and soil moisture change rate prediction model for each day within the time window.

[0138] The fifth processing module is used to stitch together the reconstructed soil moisture values ​​of all time windows in chronological order to generate historical soil moisture sequences for historical periods.

[0139] The sixth processing module is used to take the soil moisture reconstruction value at the end of the last time window as the initial state, and iteratively generate a soil moisture prediction sequence for the future preset period based on the soil moisture change rate prediction model and future weather forecast data.

[0140] In one possible implementation, the fourth processing module is further configured to:

[0141] Within each time window, the estimated value of the multi-layered soil moisture on the starting day is used as the initial state. For each day in the forward time series, the predicted soil moisture change rate of each monitoring station output by the soil moisture change rate prediction model and the soil moisture monitoring data of the monitoring station on that day are used to obtain the soil moisture analysis value for that day. The forward time series is from the next day after the starting day to the ending day.

[0142] For each day in the reverse time series, the soil moisture analysis value of that day is smoothed and optimized based on the soil moisture analysis value of the next day, so as to obtain the daily soil moisture reconstruction value within the time window; the reverse time series is from the end date to the next day after the start date.

[0143] In one possible implementation, the fourth processing module is further configured to:

[0144] For each day from the day after the start date to the day before the end date, spatial interpolation is performed on the predicted soil moisture change rate for all monitoring stations to obtain the areal soil moisture change rate field.

[0145] Based on the soil moisture status and areal soil moisture change rate field of the previous day, calculate the predicted soil moisture value for the current day.

[0146] The predicted soil moisture value is assimilated with the soil moisture monitoring data from the monitoring stations on the same day to obtain the soil moisture analysis value for the day.

[0147] For the end date, the predicted soil moisture value is assimilated with the soil moisture monitoring data from the monitoring stations on that day and the estimated isotropic multilayer soil moisture value for the end date to obtain the soil moisture analysis value for the end date.

[0148] In one possible implementation, the first processing module is further configured to:

[0149] Based on satellite remote sensing images, the surface soil moisture of each satellite transit day is retrieved and generated.

[0150] For any given satellite transit day, soil moisture at multiple depths is obtained based on the surface soil moisture.

[0151] Based on the surface soil moisture and the soil moisture at all depths, the estimated values ​​of multi-layered surface soil moisture are obtained.

[0152] In one possible implementation, the second processing module is further configured to:

[0153] For any monitoring station, based on the daily soil moisture monitoring data within the historical period, calculate the daily soil moisture change rate label for each soil depth layer; the soil moisture change rate is the relative change of the soil moisture value on the current day relative to the soil moisture value on the previous day.

[0154] Obtain daily historical meteorological data for a given historical period;

[0155] Based on historical meteorological data and daily soil moisture change rate labels for each soil depth layer, machine learning models corresponding to each soil depth layer are trained to obtain soil moisture change rate prediction models for each soil depth layer.

[0156] In one possible implementation, the second processing module is further configured to:

[0157] For any soil depth layer, historical meteorological data is input into the machine learning model to obtain a sample of the predicted soil moisture change rate output by the machine learning model.

[0158] The machine learning model is trained based on the error between the predicted soil moisture change rate samples and the daily soil moisture change rate labels to obtain the soil moisture change rate prediction model.

[0159] It should be noted that the multi-source data collaborative soil moisture assimilation reconstruction and prediction system provided by the present invention can execute the multi-source data collaborative soil moisture assimilation reconstruction and prediction method of any of the above embodiments during specific operation, which will not be elaborated in this embodiment.

[0160] Figure 8 This is a schematic diagram of the structure of the electronic device provided by the present invention, such as... Figure 8As shown, the electronic device may include: a processor 810, a communication interface 820, a memory 830, and a communication bus 840, wherein the processor 810, the communication interface 820, and the memory 830 communicate with each other through the communication bus 840. The processor 810 can call logic instructions in the memory 830 to execute a multi-source data collaborative soil moisture assimilation reconstruction and prediction method. This method includes: obtaining multi-layered areal soil moisture estimates for each satellite transit day based on satellite remote sensing imagery; constructing a soil moisture change rate prediction model based on soil moisture monitoring data from each monitoring station within a historical time period; dividing the historical time period into multiple time windows based on all satellite transit days; within each time window, using the multi-layered areal soil moisture estimates for the start and end days of the time window as boundary constraints, and generating daily soil moisture reconstruction values ​​within the time window based on the soil moisture monitoring data and the soil moisture change rate prediction model; concatenating the soil moisture reconstruction values ​​of all time windows in chronological order to generate a historical soil moisture sequence for the historical time period; and using the soil moisture reconstruction value of the last time window's end day as the initial state, iteratively generating a soil moisture prediction sequence for a future preset time period based on the soil moisture change rate prediction model and future weather forecast data.

[0161] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0162] On the other hand, the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions, and when the program instructions are executed by the computer, the computer is able to execute the multi-source data collaborative soil moisture assimilation reconstruction and prediction method provided in the above embodiments.

[0163] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the multi-source data collaborative soil moisture assimilation reconstruction and prediction method provided in the above embodiments.

[0164] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0165] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0166] 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 the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for reconstructing and predicting soil moisture assimilation using multi-source data collaboration, characterized in that, include: Based on satellite remote sensing imagery, the estimated values ​​of multi-layered soil moisture in the areal layer were obtained for each satellite transit day. Based on soil moisture monitoring data from various monitoring stations over historical periods, a soil moisture change rate prediction model was constructed. Based on all the aforementioned satellite transit dates, the historical period is divided into multiple time windows; Within each time window, the estimated values ​​of multi-layered soil moisture on the start and end days of the time window are used as boundary constraints, and the reconstructed values ​​of soil moisture for each day within the time window are generated based on the soil moisture monitoring data for each day within the time window and the soil moisture change rate prediction model. Within each time window, the estimated areal multi-layer soil moisture values ​​on the start and end dates of the time window are used as boundary constraints. Based on the soil moisture monitoring data for each day within the time window and the soil moisture change rate prediction model, daily reconstructed soil moisture values ​​are generated within the time window, including: Within each time window, the estimated value of the multi-layered soil moisture on the starting day is used as the initial state. For each day in the forward time series, the predicted soil moisture change rate of each monitoring station output by the soil moisture change rate prediction model and the soil moisture monitoring data of the monitoring station on that day are used to obtain the soil moisture analysis value for that day. The forward time series is from the day after the starting day to the ending day. For each day in the positive time series, based on the predicted soil moisture change rate of each monitoring station output by the soil moisture change rate prediction model and the soil moisture monitoring data of the monitoring stations on that day, the soil moisture analysis value for that day is obtained, including: For each day from the day after the start date to the day before the end date, spatial interpolation is performed on the predicted soil moisture change rate for all the monitoring stations to obtain the areal soil moisture change rate field. Based on the soil moisture status of the previous day and the isometric soil moisture change rate field, the predicted soil moisture value for the current day is calculated. The predicted soil moisture value is assimilated with the soil moisture monitoring data from the monitoring stations on that day to obtain the soil moisture analysis value for that day. For the end date, the predicted soil moisture value is simultaneously assimilated with the soil moisture monitoring data of the monitoring station on that day and the estimated isomorphic multilayer soil moisture value on the end date to obtain the soil moisture analysis value for the end date. For each day in the reverse time series, the soil moisture analysis value of that day is smoothed and optimized based on the soil moisture analysis value of the next day to obtain the reconstructed soil moisture value for each day within the time window; the reverse time series is from the end date to the next day after the start date. The reconstructed soil moisture values ​​of all the time windows are spliced ​​together in chronological order to generate the historical soil moisture sequence for the historical period. Using the soil moisture reconstruction value at the end of the last time window as the initial state, a soil moisture prediction sequence for a future preset period is iteratively generated based on the soil moisture change rate prediction model and future weather forecast data.

2. The method for soil moisture assimilation reconstruction and prediction based on multi-source data collaboration according to claim 1, characterized in that, The estimated values ​​of multi-layered areal soil moisture obtained based on satellite remote sensing imagery for each satellite transit day include: Based on the satellite remote sensing images, the surface soil moisture of each of the satellite transit days is retrieved and generated. For any of the satellite transit days, soil moisture at multiple depths is obtained based on the surface soil moisture. The estimated values ​​of the multi-layered soil moisture are obtained based on the surface soil moisture and the soil moisture of all depth layers.

3. The method for soil moisture assimilation reconstruction and prediction based on multi-source data collaboration according to claim 1, characterized in that, The soil moisture change rate prediction model is constructed based on soil moisture monitoring data from various monitoring stations over historical periods, including: For any of the monitoring stations, based on the daily soil moisture monitoring data within the historical period, calculate the daily soil moisture change rate label for each soil depth layer; Obtain daily historical meteorological data within the specified historical period; Based on the historical meteorological data and the daily soil moisture change rate labels of each soil depth layer, the machine learning model corresponding to each soil depth layer is trained to obtain the soil moisture change rate prediction model corresponding to each soil depth layer.

4. The method for soil moisture assimilation reconstruction and prediction based on multi-source data collaboration according to claim 3, characterized in that, The step of training a machine learning model corresponding to each soil depth layer based on the historical meteorological data and the daily soil moisture change rate labels of each soil depth layer to obtain a soil moisture change rate prediction model corresponding to each soil depth layer includes: For any of the soil depth layers, the historical meteorological data is input into the machine learning model to obtain a sample of the predicted soil moisture change rate output by the machine learning model; The machine learning model is trained based on the error between the predicted soil moisture change rate sample and the daily soil moisture change rate label to obtain the soil moisture change rate prediction model.

5. A multi-source data collaborative soil moisture assimilation reconstruction and prediction system, characterized in that, include: The first processing module is used to obtain the estimated values ​​of multi-layered soil moisture on each satellite transit day based on satellite remote sensing images. The second processing module is used to construct a soil moisture change rate prediction model based on soil moisture monitoring data from each monitoring station during historical periods. The third processing module is used to divide the historical period into multiple time windows based on all the satellite transit days; The fourth processing module is used to generate daily soil moisture reconstruction values ​​within each time window, using the estimated areal multi-layer soil moisture values ​​of the start and end dates of the time window as boundary constraints, and based on the soil moisture monitoring data of each day within the time window and the soil moisture change rate prediction model. The process of generating daily soil moisture reconstruction values ​​within each time window, using the estimated areal multi-layer soil moisture values ​​of the start and end dates of the time window as boundary constraints, and based on the soil moisture monitoring data of each day within the time window and the soil moisture change rate prediction model, includes: within each time window, using the estimated areal multi-layer soil moisture value of the start date as the initial state, for each day in the forward time series, obtaining the soil moisture analysis value for that day based on the predicted soil moisture change rate of each monitoring station output by the soil moisture change rate prediction model and the soil moisture monitoring data of the monitoring stations on that day; the forward time series is from the next day after the start date to the end date; for each day in the forward time series, based on the estimated areal multi-layer soil moisture values ​​of the start and end dates of the time window as boundary constraints, and based on the soil moisture monitoring data of each day within the time window and the soil moisture change rate prediction model, obtaining the soil moisture analysis value for that day. The soil moisture change rate prediction model outputs the predicted soil moisture change rate of each monitoring station and the soil moisture monitoring data of the monitoring stations on the current day to obtain the soil moisture analysis value for the current day. This includes: for each day from the day after the start date to the day before the end date, spatial interpolating the predicted soil moisture change rate of all the monitoring stations to obtain a planar soil moisture change rate field; calculating the predicted soil moisture value for the current day based on the soil moisture status of the previous day and the planar soil moisture change rate field; assimilating the predicted soil moisture value with the soil moisture monitoring data of the monitoring stations on the current day to obtain the soil moisture analysis value for the current day; for the end date, assimilating the predicted soil moisture value with the soil moisture monitoring data of the monitoring stations on the current day and the planar multi-layer soil moisture estimate for the end date to obtain the soil moisture analysis value for the end date; for each day in the reverse time series, smoothing and optimizing the soil moisture analysis value for the current day based on the soil moisture analysis value of the next day to obtain the reconstructed soil moisture value for each day within the time window; the reverse time series is from the end date to the day after the start date. The fifth processing module is used to splice the reconstructed soil moisture values ​​of all the time windows in chronological order to generate the historical soil moisture sequence for the historical period. The sixth processing module is used to take the soil moisture reconstruction value at the end of the last time window as the initial state, and iteratively generate a soil moisture prediction sequence for a future preset period based on the soil moisture change rate prediction model and future weather forecast data.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the multi-source data collaborative soil moisture assimilation reconstruction and prediction method as described in any one of claims 1 to 4.

7. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the multi-source data collaborative soil moisture assimilation reconstruction and prediction method as described in any one of claims 1 to 4.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the multi-source data collaborative soil moisture assimilation reconstruction and prediction method as described in any one of claims 1 to 4.

Images