A soil moisture movement inversion method and system based on soil moisture distribution

Abstract

The present invention discloses a soil moisture movement inversion method and system based on soil moisture distribution. The method includes: obtaining geological data and hydrological data of a watershed; constructing a physical watershed hydrological model based on the geological data and hydrological data; using the physical watershed hydrological model to simulate the soil moisture distribution time series results and soil moisture movement time series results of the entire watershed at any time under various working conditions and construct a data set; constructing a deep learning model for each grid node in the physical watershed hydrological model, and training the deep learning model using the data set and geological data; and using the trained deep learning model to realize soil moisture movement inversion. The present invention uses easily observable soil moisture distribution as input and uses a deep learning model as a function approximator to invert soil moisture movement that is difficult to observe for a long time at the watershed scale, providing a scientific theoretical basis for in-depth understanding of the soil moisture movement system at the watershed scale.

Description

Technical Field

[0001] The present invention belongs to the field of soil moisture movement prediction, and relates to a soil moisture movement inversion method and system, and in particular to a soil moisture movement inversion method and system based on soil moisture distribution. Background Art

[0002] Soil moisture movement is an important part of the surface water cycle. It changes the spatial and temporal distribution of soil moisture, forms river runoff, and further affects the water and heat cycle. However, due to the limitations of observation methods, it is impossible to conduct long-term observations of soil moisture movement. Existing monitoring methods are limited to small and medium scales such as profiles and slopes, and cannot take into account both revealing the path of moisture movement and estimating moisture movement flux. Models based on physical mechanisms can make up for these shortcomings, but their dual nonlinearity consumes huge computing resources, which greatly limits the research and development of three-dimensional soil moisture movement at the basin scale.

[0003] Soil moisture movement is a process corresponding to the state of soil moisture, and soil moisture is the result of water movement in the soil. Soil moisture is easy to observe and there are many means. Sensor monitoring, remote sensing observation and data assimilation methods provide long-term and variable-scale soil moisture distribution data. Deep learning can be used to build complex nonlinear models. Therefore, using soil moisture distribution data based on deep learning models can effectively predict the three-dimensional soil moisture movement speed. Summary of the invention

[0004] In order to solve the above technical problems, the present invention overcomes the deficiencies in the prior art and provides a soil moisture movement inversion method and system based on soil moisture distribution. The present invention selects soil moisture data that is easy to observe, uses a physical watershed hydrological model output data set based on the physical relationship between soil moisture and soil moisture movement, trains a deep learning model based on in-situ observation data and physical watershed hydrological model output data to grasp the mapping relationship between the two, reveals the water movement of the watershed from a three-dimensional watershed scale, and predicts the three-dimensional soil moisture movement speed at any point in the watershed at any time.

[0005] The technical solution adopted by the present invention is as follows:

[0006] A soil moisture movement inversion method based on soil moisture distribution includes the following steps:

[0007] S1. Obtain geological and hydrological data of the basin;

[0008] S2. Construct a physical basin hydrological model based on geological and hydrological data;

[0009] S3. Use the physical watershed hydrological model to simulate the soil moisture distribution time series and soil moisture movement time series results of the entire watershed at any time under various working conditions and construct a data set;

[0010] S4. constructing a deep learning model for each grid node in the physical watershed hydrological model, and training the deep learning model using the data set and geological data;

[0011] S5. Use the trained deep learning model to invert soil moisture movement.

[0012] Furthermore, the physical watershed hydrological model and the deep learning model are calibrated using measured soil moisture distribution data and hydrological data to improve the accuracy of the numerical model and reduce the errors in long-term numerical simulation results. The predicted results of soil moisture movement speed are input into the hydrological model to calculate the soil moisture distribution results and compared with the measured data to achieve the optimization of the deep learning model.

[0013] Furthermore, in step S1, the geological data includes soil property parameters at different depths, and the hydrological data includes precipitation, runoff and previous soil moisture. The specific collection steps are:

[0014] S1.1. According to the historical runoff, soil and vegetation distribution of the basin, select appropriate locations to deploy monitoring instruments to monitor relevant hydrological data such as rainfall, runoff and soil moisture at the measuring station in real time;

[0015] S1.2. Collect soil samples at different depths at various measuring points in the watershed, covering various vegetation cover types, and sample the soil in both horizontal and vertical profiles to obtain comprehensive soil property parameters in the watershed;

[0016] Furthermore, in step S2, the specific steps of constructing the physical watershed hydrological model are:

[0017] S2.1. Determine the basin boundary based on the digital elevation model, construct the three-dimensional terrain grid of the physical basin hydrological model using irregular triangulated networks, and refine the grid around the river channel and near the basin outlet section;

[0018] S2.2. Define the governing equations for variably saturated groundwater flow, surface water flow, and surface-groundwater exchange in a physical basin hydrological model based on Darcy's law;

[0019] S2.3. Construct typical precipitation or randomly generate precipitation based on historical data in the basin and input it into the physical basin hydrological model, and then input the geological data into the grid nodes of the physical basin hydrological model;

[0020] S2.4. Use hydrological data to calibrate and verify relevant parameters of the physical watershed hydrological model.

[0021] Furthermore, in step S3, the soil moisture distribution time series results include the soil moisture of all grid nodes in the watershed, and the soil moisture movement time series results include the soil moisture movement speed in three directions in the three-dimensional space coordinates of all grid nodes.

[0022] Further, in step S4, a deep learning model is constructed for each grid node in the physical watershed hydrological model, and the deep learning model is trained using the data set and geological data, specifically:

[0023] A deep learning model is constructed for the three directions of the spatial rectangular coordinate system of each grid node in the physical watershed hydrological model. The time series results of soil moisture distribution at the grid nodes and geological data, namely soil attribute data, such as soil depth, soil lateral saturated hydraulic conductivity, soil longitudinal saturated hydraulic conductivity, soil porosity, soil layer thickness, etc. are used as model inputs. The soil moisture movement speed in the three directions of the spatial rectangular coordinate system in the time series results of soil moisture movement at the grid nodes is used as model output to train the deep learning model.

[0024] Furthermore, in step S4, a suitable loss function, optimization algorithm, early stopping scheme, and hyperparameter search scheme are set for the deep learning model, and the deep learning model is fully trained until the model performance meets the requirements for inversion.

[0025] A soil moisture movement inversion system based on soil moisture distribution, comprising:

[0026] Data acquisition module: used to obtain geological data and hydrological data of the basin;

[0027] Model building module: used to build a physical basin hydrological model based on geological data and hydrological data;

[0028] Data simulation module: used to use the physical watershed hydrological model to simulate the soil moisture distribution time series results and soil moisture movement time series results of the entire watershed at any time under various working conditions and construct a data set;

[0029] Model training module: used to construct a deep learning model for each grid node in the physical watershed hydrological model, and train the deep learning model using the data set and geological data;

[0030] Model inversion module: used to realize soil moisture movement inversion using the trained deep learning model.

[0031] A computer device, comprising:

[0032] one or more processors;

[0033] A memory for storing one or more programs;

[0034] When the one or more programs are executed by the one or more processors, the one or more processors implement the above-mentioned soil moisture movement inversion method based on soil moisture distribution.

[0035] A computer-readable storage medium storing computer instructions, when the computer instructions are executed by one or more processors, causes the one or more processors to execute the steps in the above method.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] 1. The present invention solves the problems that the existing monitoring methods cannot simultaneously reveal the water movement path and estimate the water movement flux, and the traditional numerical simulation methods have a large amount of calculation, and has high temporal and spatial resolution.

[0038] 2. The data used for numerical simulation in the present invention are completely based on measured data, which can reflect the hydrogeological characteristics of the current basin and the hydrophysical processes involved.

[0039] 3. The data set used in the present invention is output by numerical simulation of a strictly physically based distributed hydrological model. The data set covers the entire three-dimensional basin and the data is accurate, comprehensive and consistent.

[0040] 4. The soil moisture movement inversion method based on soil moisture distribution proposed in the present invention can quickly invert the process and spatiotemporal differentiation of soil moisture movement in a three-dimensional watershed, which has positive significance for strengthening the scale and method of soil moisture monitoring, and revealing and simulating the transport and circulation process of coupled carbon, nitrogen and other solutes. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is a flow chart of the soil moisture inversion method in an embodiment of the present invention;

[0042] Figure 2 A schematic diagram of data composition for constructing an inversion model in a typical watershed according to an embodiment of the present invention;

[0043] Figure 3 It is the overall model performance in the embodiment of the present invention;

[0044] Figure 4 It is the model performance of a certain node in a certain rainfall in an embodiment of the present invention. DETAILED DESCRIPTION

[0045] The technical solution of the present invention is further described in detail below with reference to the accompanying drawings and specific examples.

[0046] Example 1

[0047] like Figure 1 As shown, a soil moisture movement inversion method based on soil moisture distribution includes the following steps:

[0048] S1. Obtain the geological data and hydrological data of the basin; the geological data includes soil property parameters at different depths, and the hydrological data includes precipitation, runoff and soil moisture content. The specific collection steps are:

[0049] S1.1. According to the historical runoff, soil and vegetation distribution of the basin, select appropriate locations to deploy monitoring instruments to monitor relevant hydrological data such as rainfall, runoff and soil moisture at the measuring points in the basin in real time;

[0050] S1.2. Collect soil samples at different depths at various measuring points in the watershed, covering various vegetation cover types, and sample the soil in both horizontal and vertical profiles to obtain comprehensive soil property parameters in the watershed;

[0051] S2. Based on geological data and hydrological data, construct a physical basin hydrological model; the specific steps are:

[0052] S2.1. Determine the basin boundary based on the digital elevation model, construct the three-dimensional terrain grid of the physical basin hydrological model using irregular triangulated networks, and refine the grid around the river channel and near the basin outlet section;

[0053] S2.2. Define the governing equations for variably saturated groundwater flow, surface water flow, and surface-groundwater exchange in a physical basin hydrological model based on Darcy's law;

[0054] S2.3. Construct typical precipitation or randomly generate precipitation based on historical data in the basin and input it into the physical basin hydrological model, and then input the geological data into the grid nodes of the physical basin hydrological model;

[0055] S2.4. Use hydrological data to calibrate and verify relevant parameters of the physical watershed hydrological model.

[0056] S3. Use a physical watershed hydrological model to simulate the soil moisture distribution time series results and soil moisture movement time series results of the entire watershed at any time under various working conditions and construct a data set; the soil moisture distribution time series results include the soil moisture of all grid nodes in the watershed, and the soil moisture movement time series results include the soil moisture movement speed in three directions in the three-dimensional spatial coordinates of all grid nodes, as well as the cross-sectional runoff process line and surface water depth.

[0057] S4. construct a deep learning model for each grid node in the physical watershed hydrological model, and train the deep learning model using the data set and geological data; specifically:

[0058] A deep learning model is constructed for each grid node in the three directions of the spatial rectangular coordinate system in the physical watershed hydrological model. The time series results of soil moisture distribution at the grid nodes and geological data, i.e., soil attribute data, such as soil depth, soil lateral saturated hydraulic conductivity, soil longitudinal saturated hydraulic conductivity, soil porosity, soil layer thickness, etc. are used as input. The soil moisture movement speed in the three directions of the spatial rectangular coordinate system in the time series results of soil moisture movement at the grid nodes is used as output. Appropriate loss function, optimization algorithm, early stopping scheme, and hyperparameter search scheme are set for the deep learning model, and the deep learning model is fully trained until the model performance meets the requirements of inversion.

[0059] S5. Use the trained deep learning model to invert soil moisture movement.

[0060] Furthermore, the measured soil moisture data is used to calibrate the physical watershed hydrological model and the deep learning model to improve the accuracy of the numerical model and reduce the errors in the long-term numerical simulation results. The predicted results of soil moisture movement speed are input into the hydrological model to calculate the soil moisture distribution results and compare them with the measured data to achieve the optimization of the deep learning model.

[0061] Example 2

[0062] A typical small watershed was selected as the research object, and the soil moisture distribution data in the watershed was used to invert the soil moisture movement. By outputting the data set through the physical hydrological model and building a deep learning model, the process and spatiotemporal differentiation of soil moisture movement in the three-dimensional watershed can be quickly and accurately inverted, providing a more optimal method for revealing the flow path and flux of soil flow in the watershed.

[0063] S1: According to the geological and hydrological conditions of the basin, obtain the geological data and hydrological data of the basin; the geological data includes soil attribute parameters at different depths, and the hydrological data includes precipitation, runoff and soil moisture content. The specific data collection method is as follows:

[0064] (1) Basin Overview

[0065] Select a typical small watershed, such as Figure 2 As shown in (a), this small watershed is located in the Yangtze River Basin of China and has an area of ​​3.5 km 2 The altitude ranges from 1051 to 2199m, the slope ranges from 0° to 63.3°, and the average slope is 32.4°. The average longitudinal gradient of the basin is 36.3%, and the main ditch is 1.7km long, with a northeast-southwest orientation. The basin is affected by earthquake activities and debris flows, and the surface soil of the basin is mainly loose sediments with high permeability.

[0066] (2) Measured data collection

[0067] Flow monitoring equipment and rainfall monitoring equipment are installed at the outlet of the basin to monitor and record the basin precipitation and runoff data in real time. Considering the terrain and vegetation conditions, soil moisture sensors are installed at representative points at the bottom and surface of the slope near the outlet of the basin, and the soil moisture at each point is observed. Soil is collected layer by layer at representative points, and the soil type, soil porosity, and lateral and longitudinal saturated hydraulic conductivity of the soil samples are measured.

[0068] S2. Based on geological data and hydrological data, construct a physical basin hydrological model; the specific steps are:

[0069] S2.1. Define the three-dimensional grid of the distributed hydrological model. Figure 2 As shown in (b), the river channel and boundary of the basin are extracted through the digital elevation model (DEM), and the plane triangulated mesh is generated by irregular triangulation. The soil layer is set to M layer according to the soil properties in the vertical direction, and the mesh accuracy is reasonably adjusted. Specifically, it increases from the basin boundary to the river channel in the horizontal direction, and the upper layer is dense and the lower layer is sparse in the vertical direction.

[0070] S2.2. Define the variable saturated soil water movement of the hydrological model based on Darcy's theorem and Richards' formula:

[0071]

[0072] in, is the Darcy flow [LT -1 ], q b is the source / sink term of each boundary [T -1 ], q e is the water exchange rate between the surface and underground [T -1 ],n s is the porosity [-], S w is the soil saturation [-], t is the time [T], f a is the area percentage corresponding to the surface / underground soil [-], f v is the volume percentage corresponding to the surface / underground soil [-], where L, T, - are all dimensions of physical quantities.

[0073] Darcy flow Calculated according to the following formula:

[0074]

[0075] Among them, k rw is the relative permeability [-], which represents the ratio of the permeability of water in the soil to the permeability under saturated conditions, ρ w is the density of water [ML -3 ], g is the acceleration due to gravity [LT -1 ],μw is the viscosity of water [ML -1 T -1 ], is the intrinsic permeability vector [L 2 ], z is the elevation head [L], ψ is the pressure head [L].

[0076] The soil moisture characteristic curve is expressed by the Van Genuchten model:

[0077] S w (ψ) = S wr +(1-S wr )[1+|αψ| n ] -m ; m = 1-1 / n

[0078] In the formula, α[L -1 ] and n[L -1 ] is a parameter related to soil characteristics, α is usually related to the size and distribution of soil pores and determines the horizontal position of the curve, and the two parameters n and m jointly determine the shape of the function curve. n is a shape parameter and usually n>1, m=1-1 / n is a unique parameter relationship in the Van Genuchten model, which simplifies the model and reduces the number of free parameters. Adjusting α will shift the curve horizontally, and adjusting n and m will change the steepness of the curve. ψ is the pressure head [L], S w (ψ) represents the soil saturation under the pressure head of ψ, S wr is the residual saturation of the soil [-].

[0079] The hydraulic conductivity function of the soil is also expressed by the Van Genuchten model:

[0080]

[0081] In the formula, S w is the soil saturation, k rw (S w ) indicates saturation is S w The relative permeability under the condition of m is the same as above, S e is the effective saturation of soil [-], defined as:

[0082]

[0083] S2.3. Input multiple rainfall events designed according to the local storm and flood manual and multiple measured rainfall events as boundary conditions into the physical watershed hydrological model, input soil property data and other data into the grid nodes of the physical watershed hydrological model, and set the model calculation mode;

[0084] S2.4. Use hydrological data such as measured rainfall and runoff data to calibrate and verify relevant parameters (such as α and n) of the physical watershed hydrological model.

[0085] S3. Use the physical watershed hydrological model to numerically simulate the soil moisture distribution time series results and soil moisture movement time series results of the entire watershed at any time under various working conditions and construct a data set; the specific steps are:

[0086] S3.1. Use the physical watershed hydrological model to output the distributed results of all sampling times and aggregate them, numbering the soil layers as Layer i , i∈[1,M], where M is the total number of soil layers defined in the physical watershed hydrological model. All variables in the soil moisture movement system are arranged according to the grid node number, and the time series results of each soil layer depth at each node are output in turn.

[0087] S3.2 According to the configuration of irregular triangulated network, such as Figure 2 As shown in (c), a grid node with 6 adjacent nodes is selected as the central node O0. The central node and its 6 adjacent nodes are sorted from high to low in terms of elevation (O1 to O6). The nodes directly above and below the central node (U0 / L0) and their adjacent nodes are also sorted in this way (U1 to U6 / L1 to L6). The three layers of 21 nodes are combined and the soil moisture data at the nodes is output. Because Layer1 has no corresponding node directly above it according to the above definition, starting from Layer2, the soil moisture time series data of the above 21 nodes and the dynamic meteorological data (i.e., precipitation, evapotranspiration, etc.) and static soil attribute data (i.e., lateral saturated hydraulic conductivity, vertical saturated hydraulic conductivity, soil layer depth, soil layer thickness, relative hydraulic conductivity, saturation, etc.) at the central node are aggregated to generate the data set x 2,n , Layer is generated in sequence along the soil layer depth i The dataset x i,n , and aggregate the data of the entire soil layer into x n , as the data set at node n. The remaining nodes construct data sets in the same way as shown at node n.

[0088] 4) constructing a deep learning model for each grid node in the physical watershed hydrological model, and training the deep learning model using the data set and geological data; the specific steps are:

[0089] S4.1. Dataset preprocessing: Use the time series segmentation method to split the dataset into training set and test set, and perform normalization.

[0090] S4.2. Feature screening. Feature screening includes but is not limited to the following steps: 1. Use statistical analysis methods to gain a preliminary understanding of the data distribution; 2. Preliminary screening of features by calculating the correlation between the features in the data set and the speed of soil moisture movement; 3. Train a simple tree model to output feature importance and evaluate it; 4. Recursively select feature data to train the tree model and gradually eliminate unimportant features. After preliminary screening, the data set used for the typical watershed is characterized by the soil moisture data of the aforementioned 21 nodes, dynamic meteorological data (precipitation data) and static soil properties at the center point (soil lateral saturated hydraulic conductivity, longitudinal saturated hydraulic conductivity, depth and soil thickness), see Figure 2 (c).

[0091] S4.3. Define a deep learning model. Taking LSTM as an example, import the necessary framework library, define the LSTM class, create a model instance, and determine the loss function. The loss function includes but is not limited to mean square error, mean absolute error, sequence classification loss, etc.

[0092] S4.4. Determine the optimization algorithm and hyperparameter search scheme. Select optimization algorithms including but not limited to gradient descent, adaptive moment estimation, root mean square propagation, etc. to improve the convergence speed, and use hyperparameter search schemes including but not limited to grid search, random search, Bayesian optimization, ultra-wideband optimization, Optuna, HyperOpt, etc. for parameters including but not limited to training batches, training rounds, learning rate, learning rate scheduling related parameters, temporary withdrawal probability, etc.

[0093] S4.5. Fully train the model until it meets the inversion requirements. To prevent the model from overfitting, set an early stopping scheme. If the verification error does not improve in multiple consecutive training rounds, stop training and save the model parameters with the minimum verification error. To prevent the model from underfitting, increase the model complexity, increase the number of training rounds, and change the optimization algorithm. Ensure that all soil moisture movement inversion models in the entire basin reach a good level.

[0094] S4.6. Output the soil moisture movement speed Vx / Vy / Vz in three directions of spatial coordinates. The overall performance of the model is as follows Figure 3 As shown in Figure 2, the model is specifically expressed at a certain point as follows: Figure 4 shown.

[0095] S5. Correct and optimize the deep learning model based on measured data: specifically:

[0096] S5.1. Obtain a new data set. According to step S1, collect new observation data, calculate new soil moisture distribution and soil moisture movement time series data according to the numerical simulation described in steps S2 and S3, perform the same preprocessing steps as in step S4 on the new data, and obtain a new data set.

[0097] S5.2. Mix and generate a data set. Merge the new data set with the original data set, and retrain the model according to the deep learning training method described in step S4.

[0098] S5.3. Incremental learning optimization model. Divide the new data set into multiple small batches, gradually input and update the model, and set the inspection time according to the measured data sampling period to regularly evaluate the steady improvement of the model performance under incremental learning optimization.

[0099] S5.4. Fine-tune the model to optimize performance. Freeze some layers of the deep learning model separately, and retrain the parameters of the unfrozen layers according to the optimization algorithm described in step S4.

[0100] The method proposed in the present invention for inverting soil moisture movement speed based on soil moisture distribution, meteorological dynamic data and static soil data can quickly invert the process and spatiotemporal differentiation of soil moisture movement in a three-dimensional watershed. The present invention relies on field measured data and numerical simulation results of a physical hydrological model, but the model scale can cover the entire three-dimensional watershed with high spatiotemporal accuracy, providing a practical solution for strengthening the scale and method of soil moisture monitoring.

[0101] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0102] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 Process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0103] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0104] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0105] Those skilled in the art will readily appreciate other embodiments of the present application after considering the description and practicing the contents disclosed herein. The present application is intended to cover any modification, use or adaptation of the present application, which follows the general principles of the present application and includes common knowledge or customary techniques in the art that are not disclosed in the present application. The description and examples are intended to be exemplary only, and the true scope and spirit of the present application are indicated by the claims.

[0106] It should be understood that the present application is not limited to the calculation flow scheme described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present application is limited only by the attached claims.

Claims

1. A soil moisture movement inversion method based on soil moisture distribution, characterized in that: The following steps are involved: S1. Obtain geological and hydrological data of the basin; S2. Construct a physical basin hydrological model based on geological and hydrological data; S3. Use the physical watershed hydrological model to simulate the soil moisture distribution time series and soil moisture movement time series results of the entire watershed at any time under various working conditions and construct a data set; S4. Construct a deep learning model for each grid node in the physical watershed hydrological model, and train the deep learning model using the data set and geological data; specifically: construct a deep learning model for each grid node in the physical watershed hydrological model in three directions of the spatial rectangular coordinate system, use the soil moisture distribution time series results and geological data at the grid node and its surrounding nodes as input, and use the soil moisture movement speed in the three directions of the spatial rectangular coordinate system in the soil moisture movement time series results of the grid node as output to train the deep learning model; S5. Use the trained deep learning model to invert soil moisture movement.

2. A soil moisture movement inversion method based on soil moisture distribution according to claim 1, characterized in that: The physical watershed hydrological model and the deep learning model are corrected using measured soil moisture distribution data and hydrological data to achieve optimization of the deep learning model.

3. The soil moisture movement inversion method based on soil moisture distribution according to claim 1 is characterized in that: In step S1, the geological data includes soil property parameters at different depths, and the hydrological data includes precipitation, runoff and previous soil moisture.

4. The soil moisture movement inversion method based on soil moisture distribution according to claim 1 is characterized in that: In step S2, the specific steps of constructing the physical watershed hydrological model are: S2.

1. Determine the basin boundary based on the digital elevation model, construct the three-dimensional terrain grid of the physical basin hydrological model using irregular triangulated networks, and refine the grid around the river channel and near the basin outlet section; S 2.

2. Define the governing equations for variably saturated groundwater flow, surface water flow, and surface-groundwater exchange in the physical basin hydrological model based on Darcy's law; S 2.

3. Construct typical precipitation or randomly generate precipitation based on historical data in the basin and input it into the physical basin hydrological model as boundary conditions. Then input geological data into the grid nodes of the physical basin hydrological model. S2.

4. Use hydrological data to calibrate and validate relevant parameters of the physical watershed hydrological model.

5. The soil moisture movement inversion method based on soil moisture distribution according to claim 1 is characterized in that: In step S3, the soil moisture distribution time series result includes the soil moisture of all grid nodes in the watershed, and the soil moisture movement time series result includes the soil moisture movement speed in three directions in the three-dimensional space coordinates of all grid nodes.

6. The soil moisture movement inversion method based on soil moisture distribution according to claim 1 is characterized in that: In step S4, a suitable loss function, optimization algorithm, early stopping scheme, and hyperparameter search scheme are set for the deep learning model, and the deep learning model is fully trained until the model performance meets the requirements for inversion.

7. A soil moisture movement inversion system based on soil moisture distribution, characterized in that: include: Data acquisition module: used to obtain geological data and hydrological data of the basin; Model building module: used to build a physical basin hydrological model based on geological data and hydrological data; Data simulation module: used to use the physical watershed hydrological model to simulate the soil moisture distribution time series results and soil moisture movement time series results of the entire watershed at any time under various working conditions and construct a data set; Model training module: used to construct a deep learning model for each grid node in the physical watershed hydrological model, and train the deep learning model using the data set and geological data; specifically: construct a deep learning model for each grid node in the physical watershed hydrological model in three directions of the spatial rectangular coordinate system, take the soil moisture distribution time series results and geological data at the grid node and its surrounding nodes as input, and take the soil moisture movement speed in the three directions of the spatial rectangular coordinate system in the soil moisture movement time series results of the grid node as output, and train the deep learning model; Model inversion module: used to realize soil moisture movement inversion using the trained deep learning model.

8. A computer device, characterized in that: The computer device comprises: one or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the soil moisture movement inversion method based on soil moisture distribution as described in any one of claims 1-6.

9. A computer-readable storage medium storing computer instructions, characterized in that: When the computer instructions are executed by one or more processors, the one or more processors are caused to execute the steps in the method according to any one of claims 1 to 6.

Images