A daily stratified soil water content prediction method, device and medium

Abstract

The present application discloses a method, device and medium for predicting daily stratified soil water content, relating to the technical field of soil moisture monitoring and simulation. The method includes: dividing the soil in the research area into multiple soil layers according to depth intervals; based on the meteorological data, hydrological data and measured soil water content data of the current soil layer at different time periods, using HYDRUS-1D to simulate the soil water status, obtaining the water content distribution of the current soil layer for each time period, and further constructing a random walk state transition model. Based on the random walk state transition model, using the particle filter method to complete the prediction of the soil water content of the current soil layer. The present application combines HYDRUS-1D and the particle filter method to complete high-precision soil moisture stratified prediction, can overcome the errors of a single physical model, and provides data support for intelligent irrigation and precision agriculture.

Description

Technical Field

[0001] The present application relates to the technical field of soil moisture monitoring and simulation, and particularly to a method, device and medium for predicting daily stratified soil water content. Background Art

[0002] Soil water content is a key variable in agricultural production, water resource management and ecosystem research, and has an important impact on crop growth, groundwater recharge and surface hydrological cycle. Accurately monitoring and simulating the dynamic changes of soil water content is of great significance for optimizing irrigation strategies, improving water resource utilization efficiency and predicting extreme meteorological events such as droughts.

[0003] At present, the monitoring methods of soil water content mainly include two categories. The measured methods include artificial sampling methods (such as oven drying method, tensiometer method) and numerical simulation methods (that is, using sensors (such as TDR, FDR sensors) to obtain soil moisture data and performing large-scale soil moisture inversion through remote sensing technology (such as microwave remote sensing, thermal infrared remote sensing)). Among them, although the artificial sampling method has high accuracy, it is not suitable for large-scale real-time monitoring. The numerical simulation methods include using physical models (such as HYDRUS-1D, SWAP, Soil Water Balance Model) to calculate soil moisture changes, and using data-driven models (such as machine learning, neural network) to predict soil moisture. However, there are often large errors in the simulation of a single physical model. The main reasons are as follows: The uncertainty of soil hydraulic properties (such as saturated hydraulic conductivity, field capacity) will affect the simulation accuracy, and the inaccuracy of boundary conditions such as precipitation, evapotranspiration, and surface runoff will lead to simulation deviations. At the same time, soil water content has high spatial heterogeneity and temporal variability, and it is difficult for traditional numerical simulation to fully capture its dynamic characteristics. Summary of the Invention

[0004] The purpose of the present application is to provide a method, device and medium for predicting daily stratified soil water content, which can complete high-precision soil moisture stratified prediction, overcome the errors of a single physical model, and provide data support for intelligent irrigation and precision agriculture.

[0005] To achieve the above purpose, the present application provides the following solutions:

[0006] In the first aspect, the present application provides a method for predicting daily stratified soil water content, including:

[0007] Dividing the soil in the study area into multiple soil layers according to depth intervals;

[0008] Determining any soil layer as the current soil layer;

[0009] Obtaining the meteorological data, hydrological data and measured soil water content data of the current soil layer for each time period;

[0010] Based on the meteorological data, hydrological data, and measured soil water content data of the current soil layers at different time periods, use HYDRUS-1D to simulate the soil water status, and obtain the water content distribution of each time period for the current soil layers;

[0011] Based on the water content distribution of each time period for the current soil layers, construct a random walk state transition model;

[0012] Based on the random walk state transition model, use the particle filter method to complete the prediction of the soil water content of the current soil layers.

[0013] Optionally, the depth interval is 10 cm.

[0014] Optionally, the time period is 1 day.

[0015] Optionally, the process of using HYDRUS-1D to simulate the soil water status based on the meteorological data, hydrological data, and measured soil water content data of the current soil layers at different time periods to obtain the water content distribution of each time period for the current soil layers includes:

[0016] Based on the meteorological data, hydrological data, and measured soil water content data of the current soil layers at different time periods, initialize the layered state of the soil water content;

[0017] Set the soil water status simulation parameters; the soil water status simulation parameters include: initial time step, step size extreme value, upper boundary condition, and lower boundary condition;

[0018] Input the measured soil water content data of the current soil layers at different time periods, and run HYDRUS-1D to calculate the soil water movement until the error of HYDRUS-1D meets the set accuracy standard, and output the water content distribution of each time period for the current soil layers.

[0019] Optionally, the initial time step is: 0.001 day;

[0020] The step size extreme value is: 1×10 -5 days and 5 days;

[0021] The upper boundary condition is: the atmospheric boundary condition with runoff;

[0022] The lower boundary condition is: the free drainage boundary.

[0023] Optionally, the random walk state transition model is:

[0024] ;

[0025] In the formula, represents the Soil water content of soil layers Indicates the soil water content of the th soil layer at time t-1; Indicates the random disturbance at time t.

[0026] Optionally, based on the random walk state transition model, use the particle filter method to complete the prediction of the soil water content of the current soil layer, including:

[0027] Randomly sample the measured initial water content data of the current soil layer within the range of the measured initial water content data of the current soil layer to obtain the initial particle set of the current soil layer;

[0028] Calculate the soil water content of the current soil layer at the current moment through HYDRUS-1D;

[0029] According to the soil water content of the current soil layer at the current moment, use the soil water content of the current soil layer at the current moment to determine the predicted value of the soil water content of the current soil layer at the next moment;

[0030] Obtain the observed value of the soil water content of the current soil layer at the next moment;

[0031] Based on the predicted value of the soil water content of the current soil layer at the next moment and the observed value of the soil water content of the previous soil layer at the next moment, use the particle weight update formula to update the particle weights;

[0032] Based on the updated particle weights, resample the initial particle set of the current soil layer;

[0033] Assimilate the predicted value of the soil water content of the current soil layer at the next moment and the observed value of the soil water content of the previous soil layer at the next moment.

[0034] Optionally, the weight update formula is:

[0035] ;

[0036] Wherein, ;

[0037] In the formula, Indicates the updated particle weight of the i-th particle of the th soil layer at time t; Indicates the process noise of the i-th particle of the th soil layer at time t-1; Indicates an intermediate quantity; Indicates the square value of the observation noise; Indicates the observed value of the soil water content corresponding to the i-th particle at time t; Indicates the The predicted soil water content corresponding to the i-th particle of the layered soil

[0038] In a second aspect, the present application provides a computer device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor executes the computer program to implement the above-mentioned daily layered soil water content prediction method.

[0039] In a third aspect, the present application provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the above-mentioned daily layered soil water content prediction method.

[0040] According to the specific embodiments provided by the present application, the following technical effects are disclosed:

[0041] The present application provides a daily layered soil water content prediction method, device, and medium. To solve the HYDRUS-1D simulation error problem, the present application combines HYDRUS-1D and particle filtering to construct a daily layered soil water content prediction system, making full use of sensor data to improve the prediction accuracy. Especially in precision agriculture and intelligent irrigation applications, the soil moisture layered prediction method based on HYDRUS-1D and particle filtering can provide high-precision soil moisture layered prediction, overcome the errors of a single physical model, improve the moisture prediction accuracy, and provide data support for intelligent irrigation and precision agriculture.

[0042] Compared with the prior art, the present application: (1) combines a physical model with data assimilation to improve prediction accuracy. The present application comprehensively utilizes the HYDRUS-1D physical model to simulate soil water movement, and at the same time adopts particle filter (PF) to dynamically fuse observational data to correct model biases. This method fully considers the uncertainty of soil hydraulic parameters, enabling the prediction results to more accurately reflect the true soil water status. Compared with traditional methods that only rely on physical models or data-driven models, it can effectively reduce simulation errors and improve soil water prediction accuracy. (2) Adopts daily dynamic updates of soil water content status, enabling the model to respond in real time to the impacts of factors such as rainfall, evapotranspiration, and irrigation, and avoiding the problem that static models cannot adapt to rapidly changing environments. Compared with the single offline simulation method of traditional HYDRUS-1D, the present application can ensure real-time adjustment of soil water simulation through sensor data assimilation, improving the reliability of prediction. (3) Adopts the particle filter algorithm to optimize the calculation results of HYDRUS-1D, making the state transition model more conform to the actual soil water change law and avoiding the accumulation of HYDRUS-1D simulation errors. Compared with the traditional method of directly optimizing the numerical solution of the Richards equation, the present application uses PF for nonlinear state estimation, which can reduce unnecessary calculations and avoid a sharp increase in computational complexity caused by complex parameter optimization, thereby improving prediction efficiency. (4) Effectively addresses the uncertainty of soil hydraulic parameters and boundary conditions: The traditional HYDRUS-1D physical model relies strongly on soil hydraulic parameters (such as saturated hydraulic conductivity, field capacity) and boundary conditions (such as rainfall, evapotranspiration, and leakage at the lower boundary). If there are biases in the parameters, large errors may occur in the simulation results. The present application corrects through particle filter combined with observational data, which can compensate for the uncertainty of parameters and boundary conditions, making the model prediction more robust and applicable to different soil types, different climate environments, and various agricultural application scenarios. (5) Layer-by-layer soil water prediction to optimize precise irrigation decisions: The present application can predict the soil water content at different depths from 10 cm to 100 cm, providing accurate layered soil water information. Compared with traditional methods that only predict surface moisture (such as remote sensing inversion), the present application can better support precise agricultural irrigation management, help formulate layer-by-layer irrigation strategies, reduce water resource waste, and improve water use efficiency. (6) Avoids the "curse of dimensionality" problem in numerical optimization and improves computational feasibility: The present application uses the HYDRUS-1D physical constraint + particle filter data optimization method to avoid the "curse of dimensionality" problem brought about by the global optimization calculation directly based on the Richards equation, enabling daily soil water simulation to operate efficiently in a long time series. Compared with traditional water simulation methods based on dynamic programming or complex parameter optimization, the present application can reduce computational overhead and improve the practicality and generalizability of the prediction method. Brief Description of the Drawings

[0043] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the accompanying drawings required for use in the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0044] Figure 1 It is a flowchart of the daily stratified soil water content prediction method in an embodiment of the present application;

[0045] Figure 2 It is a schematic diagram of the principle of the daily stratified soil water content prediction method in an embodiment of the present application;

[0046] Figure 3 It is a schematic diagram of the particle filter solution in an embodiment of the present application;

[0047] Figure 4 It is a comparison chart of the prediction accuracy of predicting the soil water content in a tea garden in an embodiment of the present application;

[0048] Figure 5 It is a comparison chart of the prediction accuracy of predicting the soil water content in a grassland in an embodiment of the present application. Specific embodiments

[0049] The following will clearly and completely describe the technical solutions in the embodiments of the present application in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, rather than all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0050] To make the above objects, features, and advantages of the present application more obvious and understandable, the present application will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0051] The particle filter (PF) method is an effective non - linear and non - Gaussian filtering algorithm that can fuse observation data and model predictions in real - time to improve the simulation accuracy. However, how to fully combine the physical mechanism model, on the basis of high - precision simulation, achieve high - precision assimilation of stratified soil water content with a high - computational - efficiency particle filter method, and make it better applied to the real - time monitoring and accurate prediction of soil water content, is still the focus and difficulty in the applications of precision agriculture and water resource management. In an exemplary embodiment, as Figure 1 shown, a daily stratified soil water content prediction method is provided, including:

[0052] Step 101: Divide the soil in the study area into multiple soil layers according to a depth interval. The depth interval is 10 cm.

[0053] Step 102: Determine any soil layer as the current soil layer.

[0054] Step 103: Obtain the meteorological data, hydrological data, and measured soil water content data of the current soil layer for each time period. The time period is 1 day.

[0055] Step 104: Based on the meteorological data, hydrological data, and measured soil water content data of the current soil layer for different time periods, use HYDRUS-1D to simulate the soil water status and obtain the soil water content distribution of the current soil layer for each time period. Step 104 includes: Initializing the layered state of the soil water content based on the meteorological data, hydrological data, and measured soil water content data of the current soil layer for different time periods. Setting the simulation parameters for the soil water status. The simulation parameters for the soil water status include: initial time step, step size extreme value, upper boundary condition, and lower boundary condition. Input the measured soil water content data of the current soil layer for different time periods and run HYDRUS-1D to calculate the soil water movement until the error of HYDRUS-1D meets the set accuracy standard, and output the soil water content distribution of the current soil layer for each time period. The initial time step is: 0.001 days. The step size extreme value is: 1×10 -5 days and 5 days. The upper boundary condition is: atmospheric boundary condition with runoff. The lower boundary condition is: free drainage boundary.

[0056] Step 105: Based on the soil water content distribution of the current soil layer for each time period, construct a random walk state transition model. The random walk state transition model is:

[0057] 。

[0058] In the formula, represents the soil water content of the th soil layer at time t. represents the soil water content of the th soil layer at time t-1. represents the random perturbation at time t.

[0059] Step 106: Based on the random walk state transition model, use the particle filter method to complete the prediction of the soil water content of the current soil layer.

[0060] Step 106 includes: randomly sampling the measured initial water content data of the current soil layer within the range of the measured initial water content data of the current soil layer to obtain the initial particle set of the current soil layer. Calculate the soil water content at the current moment of the current soil layer through HYDRUS-1D. Determine the predicted value of the soil water content at the next moment of the current soil layer by using the soil water content at the current moment of the current soil layer. Obtain the observed value of the soil water content at the next moment of the current soil layer. Based on the predicted value of the soil water content at the next moment of the current soil layer and the observed value of the soil water content at the next moment of the previous soil layer, update the particle weights using the particle weight update formula. Resample the initial particle set of the current soil layer based on the updated particle weights. Assimilate the predicted value of the soil water content at the next moment of the current soil layer and the observed value of the soil water content at the next moment of the previous soil layer. The weight update formula is:

[0061] 。

[0062] Among them, 。

[0063] In the formula, represents the updated particle weight of the i-th particle in the th soil layer at time t. represents the process noise of the i-th particle in the th soil layer at time t - 1. represents an intermediate quantity. represents the square value of the observation noise. represents the observed value of the soil water content corresponding to the i-th particle at time t. represents the predicted value of the soil water content corresponding to the i-th particle in the th soil layer at time t.

[0064] To ensure the relative probability meaning of the particle weights, it is necessary to normalize the updated weights:

[0065] ;

[0066] is the normalized particle weight.

[0067] This application belongs to the technical field of soil moisture dynamic simulation and prediction. The daily layered soil water content prediction method includes the following steps: S1. Determine the modeling area and set model parameters. Divide the soil profile of the research area into multiple layers, and perform inverse inversion of soil hydraulic parameters based on HYDRUS-1D, including residual water content, saturated water content, saturated hydraulic conductivity, water potential-water curve parameters, and pore connectivity parameters, and optimize the parameters in combination with the measured data. S2. Use HYDRUS-1D to simulate soil water movement. Input meteorological and hydrological data (such as precipitation, evapotranspiration), set boundary conditions (such as atmospheric driving, free drainage), run HYDRUS-1D to calculate the soil water content changes in different soil layers daily, and initialize the soil water content state based on the measured sensor data (TDR / FDR) to optimize the model simulation effect. S3. Construct a particle filter estimation model. Based on a random walk or a physically driven model, establish a state transition equation and an observation model for soil moisture, and perform particle filter update in combination with the HYDRUS-1D simulation values and the measured sensor data. At the initial moment, randomly sample from the measured data to generate a particle set, and update the state variables daily, calculate the weights of each particle, and normalize them. Further, perform particle filter assimilation optimization. Adjust the credibility of the particles according to the observed data. Particles with higher weights are retained during the resampling process to avoid particle degradation. Optimize the soil water content calculated by HYDRUS-1D based on the assimilated particle set to make it closer to the measured data and correct possible systematic errors. Predict the soil water content in the next time period through the soil moisture state optimized by particle filter assimilation, adjust the physical simulation results calculated by HYDRUS-1D to make it more consistent with the measured data, and ensure the prediction accuracy of the water content in different soil layers. Perform reverse recursive optimization for the entire prediction cycle to ensure the continuity and smoothness of the soil moisture state in each time period, and finally complete the daily layered soil water content prediction process to obtain the optimal change trajectory of the soil moisture in each layer. This application solves the problems of large prediction errors in pure physical models and sensitivity of data-driven methods to measurement noise, realizes high-precision daily soil moisture simulation based on HYDRUS-1D, and performs optimization assimilation through particle filter, providing an efficient and stable prediction method for precise irrigation, eco-hydrological simulation, and climate change research.

[0068] Next, taking a typical grassland and tea garden as examples, the daily layered soil water content prediction method provided in this embodiment will be specifically described. For example, Figure 2 , this embodiment includes the following steps.

[0069] S1. Specify the research area, determine the depth of the calculation profile, and divide the soil layers at equal depth intervals. Set the soil depth of the calculation area to 0 - 100 cm, and divide it into layers at 10 cm intervals. The soil profile states are sequentially marked as 1, 2, 3, …, 10, corresponding to the soil layers at 10 cm, 20 cm, 30 cm, …, 100 cm respectively. For the time period t , all possible soil layer states are marked as j = 1, 2, 3, …, 10.

[0070] Determine the soil depth and stratification structure of the calculation area, determine the depth of the calculation profile, such as the 0 - 100 cm soil layer, and divide it into layers at 10 cm intervals.

[0071] Combine the soil texture (such as sandy soil, loam, clay) and its hydraulic characteristics in the study area, divide different soil layers, and determine the model parameters (residual water content , saturated water content , saturated hydraulic conductivity , water potential - water content curve parameters , , pore connectivity parameters ) through experimental determination and parameter inversion, and define the parameters. Set the boundary conditions of water movement and the initial water content. Run HYDRUS - 1D for water movement simulation based on the Richards equation:

[0072] .

[0073] In the formula, = is the soil water differential; is the soil water potential; is the vertical distance from the ground surface; is the unsaturated hydraulic conductivity, is the source - sink term, mainly considering the root water uptake, (s) is the time.

[0074] Further, in this embodiment, the monitoring data of soil water content in typical grasslands and tea gardens in Rizhao City, Shandong Province (N119.560577°, E35.462755°) are used for calculation. The test site belongs to the continental climate of the warm - temperate semi - humid monsoon region. The average annual precipitation is 840.3 mm, but most of the precipitation is concentrated in June - September, and droughts and floods alternate, which have a significant impact on the soil water dynamics of tea gardens and grasslands. The tThe time period is set to days. The data collected includes: the measured daily soil water content in the tea garden from June 3, 2022 to August 20, 2023 and in the grassland from March 26, 2021 to August 20, 2023 (soil moisture data of different soil layers from 0 to 100 cm measured by sensors); meteorological data (daily precipitation, temperature, wind speed, relative humidity, solar radiation) for calculating evapotranspiration (ETc); soil hydraulic parameters required by HYDRUS-1D (residual water content, saturated water content, saturated hydraulic conductivity, water potential-water curve parameters, pore connectivity parameters of different soil layers), obtained by combining experimental measurements and parameter inversion; soil profile characteristic data (soil texture, bulk density, organic matter content of different soil layers from 0 to 100 cm) for determining the hydraulic characteristic curve; irrigation management data of the tea garden and grassland (irrigation duration, irrigation method, irrigation amount) for analyzing the soil water recharge process; boundary condition data (surface runoff, subsurface drainage, potential evapotranspiration ET0, soil evaporation rate) for HYDRUS-1D water movement simulation; crop growth parameters (root depth and transpiration coefficient of tea trees and grassland) for calculating the plant water absorption process.

[0075] Furthermore, in this case, the soil water content is monitored by collecting data once an hour and calculating the daily average soil water content.

[0076] 。

[0077] In the formula, is the average water content of the th soil layer on the th day, and is the measured water content of the th soil layer at the th day and the th hour.

[0078] Furthermore, during the experimental measurement or numerical inversion process, if the parameters of a certain soil layer do not meet the requirements, then adjust its hydraulic characteristics to make it fit the measured data optimally. Finally, the hydraulic parameter sets of all soil layers are defined as:

[0079] 。

[0080] S2. HYDRUS-1D soil water status simulation.

[0081] Input meteorological and hydrological data, and initialize the layered state of soil water content based on historical measured data (such as TDR and FDR sensor data). In this implementation case, the initial time step is set to 0.001 d, and the extreme value of the time step is 1×10 -5d and 5d. The simulated upper boundary condition is the atmospheric boundary condition with runoff, and the lower boundary condition is the free drainage boundary. The calculated is used as the input of the HYDRUS-1D physical model, and HYDRUS-1D is run to calculate the soil water movement. After the model runs, when the overall error of HYDRUS-1D meets the set accuracy standard (such as R 2 > 0.7, and the specific accuracy standard varies according to the actual situation), it is considered that the simulation effect reaches a better state, and the daily water content distribution of different soil layers at different depths over time can be obtained. .

[0082] S3. Construct a particle filter estimation model.

[0083] Particle filter (PF) is used to assimilate the model calculation results to optimize the prediction of soil water content (SWC) in different soil layers within the range of 0 - 100 cm.

[0084] After analyzing and simulating the measured data of tea gardens and grasslands, by constructing a random walk state transition model, it is assumed that the change of soil moisture is based on the previous state plus a small random perturbation to simulate the dynamic changes of moisture caused by factors such as precipitation, evapotranspiration, and leakage.

[0085] , or .

[0086] In the formula, makes the change of soil moisture more adaptable.

[0087] Furthermore, due to the possible influence of factors such as conductivity, temperature, and soil salinity on sensor measurements, there are errors, and the observed values can be modeled as:

[0088] or .

[0089] In the formula, is the soil water content monitored by the sensor, is the observation matrix used to map the true state to the observed value, is the observation noise representing the uncertainty of sensor measurements. is the state of the system at time , is the state transition function, is the process noise representing the uncertainty factors that cannot be measured from the outside, is the observed value at time , is the observation function, is the observation noise. It is also possible to consider using the HYDRUS-1D to calculate the physically driven state transition model. If the computational load is large and it is difficult to meet the high-frequency or real-time calculation requirements, a simplified water balance equation or empirical formula can be used to replace the full physical simulation.

[0090] .

[0091] Among them is precipitation, is evaporation, is transpiration, is surface runoff, is deep percolation.

[0092] If data is lacking and it is impossible to accurately calculate the water balance equation, or the observational data itself has a large amount of noise, using an empirical model may lead to an amplification of errors. It is recommended to adopt a random walk to make the model more sensitive to data changes, rather than using complex physical equations.

[0093] Furthermore, there may be certain deviations in the soil moisture state calculated by HYDRUS-1D. The prediction effect can be improved through the diversity of particle filtering. For example Figure 3 , PF starts from the initial state and generates N = 2500 particles, and each particle represents the possible soil water content state at different layers:

[0094] .

[0095] In the formula, is the initial soil water content of the th layer corresponding to the th particle; Random sampling is performed within the range of the measured initial water content data according to the soil layer to form an initial particle set at different layers.

[0096] Furthermore, the state of the next moment of each soil layer is calculated through the state transition model:

[0097] .

[0098] In the formula, is the water content of different soil layers calculated by HYDRUS-1D, , representing the process noise.

[0099] Furthermore, the particle weights are updated according to the observational values of different soil layers:

[0100] .

[0101] Among them, .

[0102] To ensure the relative probability meaning of particle weights, it is necessary to normalize the updated weights:

[0103] ;

[0104] is the normalized particle weight; the sum of the normalized weights is 1.

[0105] Furthermore, based on the normalized weights, independent resampling is performed on the particles in different soil layers, low-weight particles are removed, and high-weight particles in different soil layers are retained to prevent particle degradation:

[0106] .

[0107] Furthermore, by combining prediction and observation data, the soil moisture status of different soils is finally corrected:

[0108] .

[0109] In the formula, is the gain matrix of the th layer, is the observation matrix, is the predicted value of soil water content by the HYDRUS-1D model, the th layer soil water content observation value at time t of the soil.

[0110] Through state transfer + weight update + resampling in PF, the simulation accuracy of soil moisture in different layers after assimilation is significantly improved. Assimilation combines prediction and observation data, making the simulated soil moisture closer to the actual situation. Reverse recursive optimization is performed on the entire prediction cycle to ensure the continuity and smoothness of the soil moisture status at each time period, and finally the daily stratified soil water content prediction process is completed to obtain the optimal change trajectory of soil moisture at each layer.

[0111] This method uses the HYDRUS-1D physical model to calculate the soil water movement in different soil layers. First, according to the soil texture, layered structure and hydraulic characteristics of the study area, a Richards equation model is established. Input meteorological data (precipitation, evapotranspiration), set boundary conditions (free drainage / meteorological drive), and run HYDRUS-1D to simulate the daily changes of soil water content at different depths (10 cm - 100 cm). Through inverse modeling optimization, the soil hydraulic parameters are adjusted to make the simulated values consistent with the measured values (TDR / FDR sensor data) to improve the accuracy of the initial soil moisture status setting. Based on the preliminary simulation results of HYDRUS-1D, the particle filter method is used to generate an initial particle set, simulate the daily dynamic changes of soil moisture, and combine the sensor observation data for real-time assimilation to improve the prediction accuracy.

[0112] The sensor can provide hourly or higher-frequency soil water content data. Through wireless transmission (LoRa, NB-IoT), the soil water content observation data is uploaded in real time. PF combined with HYDRUS-1D dynamically adjusts the soil water status every day to make the prediction more accurate. The traditional HYDRUS-1D calculation is complex and difficult to support high-frequency real-time prediction. However, it can perform a complete calculation in the initial state and only make a small-range update based on the latest observation data subsequently: In the initial stage, run the complete HYDRUS-1D to obtain the benchmark soil water status and provide physical-driven prediction on a long time scale; In the real-time stage, only perform PF prediction + observation assimilation on the latest time step t through high-frequency sensor data to correct the prediction value, taking into account both accuracy and computational efficiency, and significantly reducing the computational amount.

[0113] Furthermore, in this embodiment, two typical plots, namely a tea garden and a grassland, are selected respectively, and soil water simulation is carried out in the depth range of 0-100 cm, and data assimilation is performed using the HYDRUS-1D physical model combined with particle filter (PF). To evaluate the improvement effect of PF assimilation on the simulation accuracy of soil water, the following different calculation schemes are compared:

[0114] (1) Only use HYDRUS-1D (without assimilation) to simulate the soil water status of the entire profile.

[0115] (2) HYDRUS-1D+PF layer-by-layer assimilation (10 cm stratification), and use sensor data for dynamic correction.

[0116] (3) HYDRUS-1D+PF overall profile assimilation, that is, without distinguishing soil layers, directly correct the overall soil water status. To further analyze the simulation accuracy of different schemes, the coefficient of determination (R 2 )、root mean square error (RMSE), and mean absolute error (MAE) are calculated and compared and analyzed. The specific effects are shown in Table 1.

[0117] Table 1 Comparison table of real-time prediction accuracy of soil water in tea garden and grassland before and after using this method

[0118]

[0119] HYDRUS-1D provides daily physical predictions, but has a large computational amount, cannot be updated at high frequency, and has low accuracy. PF combined with sensor data performs high-frequency prediction and correction to improve short-term accuracy. The prediction results of the tea garden and the grassland are as Figure 4 and Figure 5 shown. In this case, when only HYDRUS-1D (without PF assimilation) is used, R 2Only 0.806 - 0.809, with large prediction errors. The combination of HYDRUS-1D and particle filter (PF) assimilation technology can effectively improve the simulation accuracy. In the depth range of 0 - 100 cm, the soil water content (SWC) models of the tea garden and the case study soil perform well overall, and the overall soil layer R 2 reaches 0.99995 (tea garden) and 0.99449 (grassland), and the errors are significantly reduced. The accuracy of layer-by-layer HYDRUS-1D (without assimilation) is low. The performance of each depth level of the present invention is relatively consistent, and the R 2 value is close to 1.0, and both the RMSE and MAE are low, indicating that PF has a strong ability to correct errors in different soil layers.

[0120] As can be seen from Table 1, through the HYDRUS-1D physical model + particle filter (PF) assimilation method of the present invention, the simulation accuracy of soil moisture in different soil layers has been significantly improved, not only improving the overall prediction accuracy, but also effectively reducing the simulation errors at different depth levels. Compared with the case of only using HYDRUS-1D calculation, the average RMSE of the method of the present invention in the range of 0 - 100 cm is reduced by more than 50%, the MAE is reduced by more than 60%, and the coefficient of determination (R 2 ) is increased to more than 0.99, indicating that the assimilated results are more in line with the actual soil moisture change trend.

[0121] In addition, through layer-by-layer PF assimilation (10 cm stratified correction), the prediction accuracy at different depth levels has been optimized. Especially in the area with large water content changes in the shallow layer (0 - 30 cm), the errors are significantly reduced, and the short-term prediction stability is improved. In the deep layer (80 - 100 cm), due to the slow water migration, PF assimilation can effectively correct the long-term deviation of HYDRUS-1D and improve the credibility of the model.

[0122] From the above analysis, it can be seen that the daily stratified soil moisture prediction method based on HYDRUS-1D + PF proposed by the present invention has a significant effect of improving the simulation accuracy. The operation results show that this method can effectively make up for the deficiency of the traditional HYDRUS-1D model in responding to short-term water dynamic changes, and at the same time avoid systematic errors caused by model parameter uncertainties. Compared with the non-assimilated treatment, the method of the present application can effectively reduce the accumulation of physical model errors and improve the reliability of soil moisture dynamic monitoring.

[0123] Applying the method of the present application to the fields of precision agriculture and intelligent irrigation can effectively optimize the irrigation strategy, improve the water resource utilization efficiency, and provide a scientific basis for eco-hydrological research and water management.

[0124] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, it implements a daily layered soil moisture content prediction method.

[0125] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0126] In an exemplary embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0127] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.

[0128] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memories. Non-volatile memories can include read-only memory (ROM), magnetic tapes, floppy disks, flash memories, optical memories, high-density embedded non-volatile memories, resistive random access memories (ReRAM), magnetoresistive random access memories (MRAM), ferroelectric random access memories (FRAM), phase change memories (PCM), graphene memories, etc. Volatile memories can include random access memory (RAM) or external cache memories, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0129] The databases involved in the embodiments provided in the present application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in the present application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logics, data processing logics based on quantum computing, etc., without limitation.

[0130] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0131] Specific examples are used in this article to elaborate on the principles and implementation manners of the present application. The description of the above embodiments is only used to help understand the method and its core idea of the present application; at the same time, for those of ordinary skill in the art, according to the idea of the present application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present application.

Claims

1. A daily stratified soil water content prediction method, characterized in that, Including: Dividing the soil in the research area into multiple soil layers according to the depth interval; Determining any one soil layer as the current soil layer; Obtaining the meteorological data, hydrological data and measured soil water content data of the current soil layer for each time period; Based on the meteorological data, hydrological data and measured soil water content data of the current soil layer for different time periods, using HYDRUS-1D to simulate the soil water state, and obtaining the water content distribution of the current soil layer for each time period; Based on the water content distribution of the current soil layer for each time period, constructing a random walk state transition model; Based on the random walk state transition model, using the particle filter method to complete the prediction of the soil water content of the current soil layer, including: Randomly sampling the measured initial water content data of the current soil layer within the range of the measured initial water content data of the current soil layer to obtain the initial particle set of the current soil layer; Calculating the soil water content of the current soil layer at the current moment through HYDRUS-1D; According to the soil water content of the current soil layer at the current moment, using the soil water content of the current soil layer at the current moment to determine the predicted value of the soil water content of the current soil layer at the next moment; Obtaining the observed value of the soil water content of the current soil layer at the next moment; Based on the predicted value of the soil water content of the current soil layer at the next moment and the observed value of the soil water content of the previous soil layer at the next moment, using the particle weight update formula to update the particle weights; the weight update formula is: ； Among them, ; In the formula, represents the updated particle weight of the i-th particle in the -th soil layer stratification at time t; represents the process noise of the i-th particle in the -th soil layer stratification at time t-1; represents an intermediate quantity; represents the squared value of the observation noise; represents the observed soil water content value corresponding to the i-th particle at time t; represents the predicted soil water content value corresponding to the i-th particle in the -th soil layer stratification at time t; Based on the updated particle weights, resampling the initial particle set of the current soil layer; Performing assimilation processing on the predicted value of the soil water content of the current soil layer at the next moment and the observed value of the soil water content of the previous soil layer at the next moment.

2. The daily stratified soil water content prediction method according to claim 1, wherein The depth interval is 10 cm.

3. The daily-layered soil water content prediction method according to claim 1, wherein, The time period is 1 day.

4. The daily-layered soil water content prediction method according to claim 1, characterized in that The method of using HYDRUS-1D to simulate the soil water state based on the meteorological data, hydrological data and measured soil water content data of the current soil layer for different time periods to obtain the water content distribution of the current soil layer for each time period includes: Initializing the layered state of the soil water content based on the meteorological data, hydrological data and measured soil water content data of the current soil layer for different time periods; Setting the soil water state simulation parameters; the soil water state simulation parameters include: initial time step, step size extreme value, upper boundary condition and lower boundary condition; Inputting the measured soil water content data of the current soil layer for different time periods, and running HYDRUS-1D to calculate the soil water migration until the error of HYDRUS-1D meets the set accuracy standard, and outputting the water content distribution of the current soil layer for each time period.

5. The daily layered soil water content prediction method according to claim 4, wherein The initial time step is: 0.001 day; The step extreme values are: 1×10 -5 days and 5 days; The upper boundary condition is: the atmospheric boundary condition with runoff; The lower boundary condition is: the free drainage boundary.

6. The daily stratified soil water content prediction method according to claim 4, wherein The random walk state transition model is: ； Wherein, represents the soil water content of the th soil layer at time t; represents the soil water content of the th soil layer at time t-1; represents the random perturbation at time t.

7. A computer device, comprising: A memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the daily layered soil water content prediction method according to any one of claims 1-6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the daily stratified soil water content prediction method described in any one of claims 1-6.

Images