Method for estimating optimal soil moisture by using vegetation-penetrating microwaves

Abstract

The present invention relates to a method for estimating optimal soil moisture by using vegetation-penetrating microwaves. According to the present invention, the method for estimating optimal soil moisture by using vegetation-penetrating microwaves comprises the steps of: simulating soil moisture and permittivity information by using a permittivity model; extracting physical information related to the soil moisture by using the permittivity model; extracting a permittivity from a brightness temperature, generating a machine learning model, and generating error information from the machine learning model by using the extracted physical information; and parameterizing the permittivity model by using the error information generated through the machine learning model. As described above, the present invention, unlike a conventional method of merely using machine learning to accurately estimate soil moisture, improves a physical model through machine learning and thus can provide even wilting point and saturation point information, which cannot be identified only by machine learning, on the basis of target area data.

Description

Soil moisture estimation method using vegetation-penetrating microwave

[0001] The present invention relates to a method for estimating soil optimum moisture using vegetation-penetrating microwaves, and more particularly, to a method for estimating soil optimum moisture using vegetation-penetrating microwaves that derives in real time the optimal soil moisture range for crops from a ground soil moisture sensor by utilizing microwave satellite brightness temperature data and machine learning.

[0002] Accurate soil moisture management is essential for optimizing irrigation systems and maintaining crop health.

[0003] However, conventional soil moisture sensors can only measure the moisture content in the soil, and this information alone has limitations in estimating the optimal moisture irrigation essential for crop growth.

[0004] This makes it very difficult for agricultural managers to determine the exact timing and amount of irrigation needed to maximize crop health and productivity.

[0005] Additionally, conventional ground-based soil moisture measurement and estimation methods for optimal irrigation moisture ranges have limitations in that they are time-consuming, labor-intensive, and have limited spatial coverage.

[0006] Satellite surveys can provide information on soil moisture as well as irrigation moisture range, but optimizing the parameters used in radiative transfer models is practically difficult.

[0007] On the other hand, using machine learning can obtain accurate soil moisture, but there is a problem that observation data is required to learn the optimal range of soil moisture, or if observation is not possible, it must be estimated within a black box.

[0008] According to the present invention, the purpose is to provide a method for estimating optimum soil moisture using vegetation-penetrating microwaves that derives the optimal soil moisture range of crops in real time from a ground soil moisture sensor by utilizing microwave satellite brightness temperature data and machine learning.

[0009] According to one embodiment of the present invention for achieving such technical task, a method for estimating soil optimum moisture using vegetation-penetrating microwaves includes the steps of: simulating soil moisture and permittivity information using a permittivity model; extracting physical information related to the soil moisture using the permittivity model; extracting permittivity from brightness temperature, generating a machine learning model, and generating error information of the machine learning model using the extracted physical information; and parameterizing the permittivity model using the error information generated through the machine learning model.

[0010] Thus, according to the present invention, unlike conventional methods that simply estimate accurate soil moisture through machine learning, by improving a physical model through machine learning, it is possible to provide information on false points and saturation points that cannot be identified through machine learning based on target area data.

[0011] In addition, by applying the target area's fading point and saturation point values ​​found by machine learning when estimating soil moisture to a physical model, the accuracy of soil moisture estimated from satellites can be improved, and non-invasive, large-scale monitoring of irrigation appropriate ranges based on remote sensing can be enabled. By providing accurate soil moisture information and real-time irrigation appropriate range information, the optimal irrigation schedule can be determined to minimize water usage and maximize crop yields.

[0012] FIG. 1 is a diagram illustrating a flow chart of a soil moisture estimation method using vegetation-penetrating microwaves according to one embodiment of the present invention.

[0013] FIG. 2 is a diagram illustrating an example of simulating dielectric constant using a dielectric constant model according to one embodiment of the present invention.

[0014] FIG. 3 is a diagram illustrating an example of setting an error of a dielectric constant model according to one embodiment of the present invention.

[0015] FIG. 4 is a diagram illustrating an example of falsification point and saturation point extreme point information appearing in a dielectric constant error according to one embodiment of the present invention.

[0016] FIG. 5 is a diagram illustrating an example of generating an error of a machine learning model according to one embodiment of the present invention.

[0017] FIG. 6 is a diagram illustrating an example of parameterizing a dielectric constant model according to one embodiment of the present invention.

[0018] FIG. 7 is a diagram illustrating an example of machine learning error information for a monitoring target area according to one embodiment of the present invention.

[0019] FIG. 8 is a diagram illustrating an example of analyzing changes in Alaska forgery points and saturation points in machine learning error information by year according to one embodiment of the present invention.

[0020] FIG. 9 is a diagram showing an example of soil moisture estimation results before and after applying machine learning to parameterize the microwave radiation transfer model falsification point saturation point according to one embodiment of the present invention.

[0021] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the attached drawings. In this process, the thickness of lines and the sizes of components depicted in the drawings may be exaggerated for clarity and convenience of explanation.

[0022] Furthermore, the terms described below are defined based on their functions within the present invention, and may vary depending on the intent or custom of the user or operator. Therefore, the definitions of these terms should be based on the overall content of this specification.

[0023] Hereinafter, a soil moisture estimation method using vegetation penetration microwave according to one embodiment of the present invention will be specifically described with reference to FIGS. 1 to 6.

[0024] FIG. 1 is a diagram illustrating a flow of a soil optimum moisture estimation method using vegetation-penetrating microwaves according to an embodiment of the present invention, FIG. 2 is a diagram illustrating an example of simulating permittivity through a permittivity model according to an embodiment of the present invention, FIG. 3 is a diagram illustrating an example of setting an error of a permittivity model according to an embodiment of the present invention, FIG. 4 is a diagram illustrating an example of falsification point and saturation point extreme point information appearing in a permittivity error according to an embodiment of the present invention, FIG. 5 is a diagram illustrating an example of generating an error of a machine learning model according to an embodiment of the present invention, and FIG. 6 is a diagram illustrating an example of parameterizing a permittivity model according to an embodiment of the present invention.

[0025] As illustrated in FIGS. 1 and 2, a soil moisture estimation method using vegetation-penetrating microwaves according to one embodiment of the present invention first simulates soil moisture and permittivity information using a permittivity model (S110).

[0026] At this time, the dielectric constant model varies depending on the ratio of free and bound water, and the ratio of free and bound water is determined by the forging point and saturation point.

[0027] Because the bound water has a lower value than the free water, there may be a large error in the soil moisture value calculated from the permittivity without knowing the exact ratio.

[0028] At this time, the effective permittivity of the soil is calculated using the following mathematical formula 1.

[0029]

[0030] At this time, SM (Soil Moisture) is the soil moisture value, is the free water permittivity, is the coupling permittivity, is the air permittivity, is the soil mineral permittivity, and p is the soil porosity.

[0031] also, is the proportion of the combined number in SM, is the ratio of free numbers in SM, and the sum of the two ratios is always 1.

[0032] Additionally, the effective permittivity formula can be divided into three models according to the phase change of soil moisture.

[0033] More specifically, when only bound water exists in the soil without free water, the effective permittivity of the soil can be calculated using the following mathematical equation 2.

[0034]

[0035] On the other hand, when only free water exists in the soil without bound water, the effective permittivity of the soil can be calculated using the following mathematical equation 3.

[0036]

[0037] On the other hand, when bound water and free water coexist in the soil, the effective permittivity of the soil can be calculated using the following mathematical equation 4.

[0038]

[0039] That is, the ratio of bound water to free water is determined by the value of soil moisture between the wilting point and the saturation point, and this ratio is calculated as a function of soil moisture and the wilting point / saturation point, or when the free and bound water are 0 or 1.

[0040] Because it is difficult to directly measure the combined water content, the moisture content is calculated using the soil's wilting point and saturation point.

[0041] However, since it is difficult to measure accurate information on the fading point and saturation point over a wide area, the fading point (wp) and saturation point (p) are parameterized using mathematical equations 5 and 6 below based on key input data such as clay content and organic matter content.

[0042]

[0043]

[0044] The various previously proposed saturation point and saturation point formulas mean that the parameters to be optimized vary from region to region. Therefore, it is impossible to accurately estimate soil moisture across all regions using a single formula.

[0045] Currently, many studies are trying to minimize the error with the above-ground soil moisture measurements by finding the optimal parameters at various sites through empirical methods. However, this process relies on subjective judgment, requires a lot of time and money, and has limitations such as a lack of performance reproducibility under different conditions and regions.

[0046] Therefore, to solve these problems, the present invention proposes a new method of a machine learning-physical model bidirectional system that can parameterize unique forging points and saturation points in various regions and conditions through bidirectional modeling of physical models and machine learning.

[0047] After simulating soil moisture and permittivity information, physical information related to soil moisture is extracted using a permittivity model (S120).

[0048] At this time, the physical information is the dielectric simulation error due to the error in the input variables on the response side of the physical model according to the change in the input value.

[0049] At this time, the physical model is a microwave permittivity model that considers organic matter, as shown in mathematical equation 7 below.

[0050]

[0051] Here, T is the soil temperature (℃), CL is the clay volume ratio (cm 3 / cm 3 ), OM means the weight ratio of organic matter (kg / kg), and in the physical model, the error of the permittivity model due to organic matter in the monitoring target area ( ) is calculated as in the following mathematical formula 8, and the dielectric constant model error due to clay in the monitoring target area ( ) is calculated as shown in the following mathematical formula 9.

[0052]

[0053]

[0054] At this time, as shown in Fig. 3, the minimum value of the falsification point, the maximum value of the falsification point, the minimum value of the saturation point, and the maximum value of the saturation point in Fig. 2 are set as the extreme points of the prediction error of soil moisture.

[0055] More specifically, the minimum value of the fading point is set to the minimum value of the negative absolute error of soil moisture through the following mathematical expression 10, and the maximum value of the fading point is set to the maximum value of the positive absolute error of soil moisture through the following mathematical expression 11.

[0056]

[0057]

[0058] In addition, the minimum value of the saturation point is set to the maximum value of the negative absolute error of soil moisture in a range greater than the minimum value of the fading point through the following mathematical expression 12, and the maximum value of the saturation point is set to the minimum value of the positive absolute error of soil moisture in a range greater than the maximum value of the fading point through the following mathematical expression 13.

[0059]

[0060]

[0061] Through this, it is possible to obtain extreme information of the saturation point and the forgery point that appear in the dielectric constant error as shown in Fig. 4.

[0062] At this time, (a) of Fig. 4 is a diagram showing an example of a dielectric constant model prediction error extreme point, and (b) of Fig. 4 is a diagram showing an example of a random forest prediction error simulation result that learned a dielectric constant model.

[0063] After extracting physical information related to soil moisture, the permittivity is extracted from the brightness temperature, a machine learning model is created, and error information of the machine learning model is created using the extracted permittivity simulation error value (S130).

[0064] At this time, the dielectric constant observation used for learning extracts the dielectric constant that most closely simulates the brightness temperature.

[0065] Also, brightness temperature simulation is calculated through the following mathematical formula 14.

[0066]

[0067] At this time, is the optical depth of vegetation, is the scattering albedo of vegetation, determined by the NDVI (normalized difference vegetation index) and MPDI (Microwave Polarization Difference Index) coefficients, and T is the MODIS (Moderate Resolution Imaging Spectroradiometer) surface temperature.

[0068] Also, soil emissivity is calculated using the following mathematical formula 15.

[0069]

[0070] To estimate the permittivity at microwave brightness temperatures, the polarized reflectance used in calculating emissivity is ) real part of the permittivity ( ) and imaginary part( ) must be derived into a mathematical expression expressed as a single variable, that is, a mathematical expression that expresses only the real part of the permittivity. To this end, the imaginary part of the permittivity is parameterized into the real part as in the following mathematical expression 16.

[0071]

[0072] At this time, the imaginary and real parts of the dielectric constant fluctuate greatly depending on soil moisture and organic matter, but the variability of the ratio of the two values ​​is minimal, so the median of the ratio is used. can be determined, and in the embodiment of the present invention is set to .

[0073] Accordingly, by expressing the horizontal polarization reflectance as in Equation 17 below using only the real part of the permittivity, the real part of the permittivity can be estimated from Equation 14 at the brightness temperature.

[0074]

[0075] More specifically, the value at which the microwave observed brightness temperature minus the simulated brightness temperature is minimized is obtained through the following mathematical expression 18. The observed value of the real part of the permittivity can be estimated from .

[0076]

[0077] At this time, and is determined by NDVI and MPDI coefficients, is the value with the minimum brightness temperature simulation error among the values ​​in the preset range (e.g., 0 to 65). is decided by

[0078] Also, brightness temperature observation extracted from is used as learning data to be generated by applying the random forest technique as in the following mathematical formula 19 of the machine learning model.

[0079]

[0080] In addition, the error information of the machine learning model is generated as in the following mathematical expression 20 for the machine learning error due to organic matter in the target area, and as in the following mathematical expression 21 for the machine learning error due to clay.

[0081]

[0082]

[0083] At this time, as shown in Fig. 5, the extreme points of the foreclosure point or saturation point of the monitoring target area can be calculated using the extreme point information of the soil moisture error of the machine learning model.

[0084] More specifically, the minimum value of the actual wilting point in the target area can be calculated using the minimum value of the negative absolute error of the soil moisture of the machine learning model.

[0085] At this time, the minimum value of the negative absolute error of soil moisture due to organic matter is used as in the following mathematical expression 22 to obtain the minimum value of the withering point at the minimum of organic matter ( ) can be calculated, and the minimum value of the negative absolute error of soil moisture due to clay can be used to obtain the minimum value of the fading point at the minimum of clay, as in the following mathematical expression 23. ) can be produced.

[0086]

[0087]

[0088] Additionally, the maximum absolute error value of the soil moisture content of the machine learning model can be used to calculate the maximum actual withering point value of the target area.

[0089] At this time, the maximum value of the absolute error of the soil moisture content due to organic matter is used as in the following mathematical expression 24 to obtain the maximum value of the wilting point at the maximum of organic matter ( ) can be calculated, and the maximum value of the maximum value of the wilting point at the maximum of the clay can be calculated using the maximum value of the absolute error of the soil moisture content due to the clay as in the following mathematical expression 25. ) can be produced.

[0090]

[0091]

[0092] Additionally, the minimum value of the actual saturation point of the target area can be calculated by using the maximum value of the negative absolute error of the soil moisture of the machine learning model in a range greater than the minimum value of the saturation point.

[0093] At this time, the minimum value of the saturation point at the minimum of organic matter is obtained by using the maximum value of the negative absolute error of soil moisture in a range greater than the minimum value of the saturation point due to organic matter, as in the following mathematical expression 26. ) can be calculated, and the minimum value of the saturation point at the minimum of the clay is obtained by using the maximum value of the negative absolute error of the soil moisture in a range greater than the minimum value of the saturation point due to the clay, as in the following mathematical expression 27. ) can be produced.

[0094]

[0095]

[0096] Additionally, the maximum value of the actual saturation point of the target area can be calculated by using the minimum absolute error of the soil moisture content of the machine learning model in a range greater than the maximum value of the saturation point.

[0097] At this time, the minimum absolute error value of the soil moisture content in a range greater than the maximum value of the saturation point due to organic matter, as in the following mathematical expression 28, is used to obtain the maximum value of the saturation point at the maximum of organic matter ( ) can be calculated, and the minimum absolute error value of soil moisture in a range greater than the maximum value of the saturation point due to clay is used to calculate the maximum value of the saturation point at the maximum of the clay, as in the following mathematical expression 29. ) can be produced.

[0098]

[0099]

[0100] After generating error information of a machine learning model, the permittivity model is parameterized using the error information generated through the machine learning model (S140).

[0101] More specifically, as illustrated in Fig. 6, a parameterization function is determined that can linearize the forging point and saturation point to changes in input data (e.g., organic matter and clay), and the parameters required for parameterization are optimized using error information obtained through machine learning in the target area.

[0102] At this time, the sensitivity of the counterfeit point parameterization function as in the following mathematical expression 30 can be approximated to the maximum and minimum counterfeit points of machine learning to parameterize the target area counterfeit points.

[0103]

[0104] In addition, the sensitivity of the saturation point parameterization function can be approximated to the maximum and minimum machine learning saturation points as in the following mathematical expression 31 to parameterize the target area saturation point.

[0105]

[0106] At this time, the parameters required for the basic parameterization function before optimization, as in the following mathematical expression 32, can be determined by physical values ​​discovered through machine learning.

[0107]

[0108] FIG. 7 is a diagram illustrating an example of machine learning error information for a monitoring target area according to one embodiment of the present invention, FIG. 8 is a diagram illustrating an example of analyzing changes in Alaska fading points and saturation points in annual machine learning error information according to one embodiment of the present invention, and FIG. 9 is a diagram illustrating an example of soil moisture estimation results before and after applying machine learning to parameterization of microwave radiation transfer model fading points and saturation points according to one embodiment of the present invention.

[0109] According to one embodiment of the present invention, as illustrated in FIG. 7, the optimal parameterization parameters for each of Alaska and North America can be extracted from the soil moisture error of a random forest generated according to the dielectric model physics guidelines.

[0110] Furthermore, as illustrated in Figure 8, when analyzing machine learning error information by year, it can be seen that the forgery point and saturation point, which take organic matter variability into account, change annually. This implies that the parameterization of the microwave radiation transfer model must be optimized not only spatially but also temporally.

[0111] As a result, as shown in Fig. 9, the extracted forging point and saturation point values ​​can be reapplied to the physical model to improve the accuracy of soil moisture estimation.

[0112] In addition, according to the present invention, unlike conventional methods that simply estimate accurate soil moisture through machine learning, by improving a physical model through machine learning, it is possible to provide information on saturation points and false points that cannot be identified through machine learning based on target area data.

[0113] In addition, by applying the target area's fading point and saturation point values ​​found by machine learning when estimating soil moisture to a physical model, the accuracy of soil moisture estimated from satellites can be improved, and non-invasive, large-scale monitoring of irrigation appropriate ranges based on remote sensing can be enabled. By providing accurate soil moisture information and real-time irrigation appropriate range information, the optimal irrigation schedule can be determined to minimize water usage and maximize crop yields.

[0114] While the present invention has been described with reference to the embodiments illustrated in the drawings, these are merely exemplary, and those skilled in the art will understand that various modifications and equivalent embodiments are possible. Therefore, the true technical protection scope of the present invention should be determined by the technical spirit of the following claims.

Claims

1. Step of simulating soil moisture and dielectric constant information using a dielectric constant model; A step of extracting physical information related to the soil moisture using the above dielectric constant model; A step of extracting the permittivity from the brightness temperature, creating a machine learning model, and using the extracted physical information to create error information of the machine learning model; and A method for estimating soil optimum moisture using vegetation-penetrating microwaves, comprising a step of parameterizing a dielectric constant model using error information generated through the above machine learning model.

2. In paragraph 1, The above physical information is the dielectric simulation error according to the error of the input variable on the response side of the physical model according to the change in the input value, The above physical model is a microwave permittivity model that considers organic matter and is expressed by the following mathematical formula: Method for estimating soil moisture using vegetation-penetrating microwaves: At this time, is the real part of the permittivity, T is the soil temperature (℃), and CL is the clay volume fraction (cm 3 / cm 3 ), OM means the weight ratio of organic matter (kg / kg).

3. In paragraph 1, A continuous machine learning model is a method for estimating soil moisture using vegetation-penetrating microwaves, which is generated by applying the random forest technique according to the following mathematical formula: At this time, is the brightness temperature It is extracted from, T obtains soil temperature Celsius information from MODIS, CL obtains volume ratio information from soil type map, and OM obtains weight ratio information from organic matter map as input values.

4. In paragraph 3, The error information of the above machine learning model is generated according to the following mathematical formula: Method for estimating soil moisture using vegetation-penetrating microwaves: At this time, is the machine learning error due to organic matter in the target area, is a machine learning error due to clay.

5. In paragraph 1, The step of parameterizing the above dielectric constant model is: A method for estimating soil optimum moisture using vegetation-penetrating microwaves, which determines a parameterization function that can linearize the fading point and saturation point to changes in input data, and optimizes the parameters required for parameterization with error information obtained through machine learning in the target area.

6. In paragraph 5, The step of parameterizing the above dielectric constant model is: A method for estimating soil moisture using vegetation-penetrating microwaves, which approximates the sensitivity of the parametrization function of the parametrization point to the maximum and minimum parametrization points of machine learning according to the following mathematical formula: At this time, is the maximum value of the forgery point in the organic material found by machine learning, is the minimum value of the forgery point at the minimum organic matter, is the maximum value of the forging point at the maximum of the clay, is the minimum value of the forging point at the clay minimum, is the maximum organic matter in the machine learning target area, is the minimum organic matter value in the machine learning target area, OM is the organic matter value at the sampling point, is the maximum clay value in the machine learning target area, is the minimum clay value in the machine learning target area, CL is the clay value at the sampling point, is the theoretical minimum value of the forgery point (0.01 cm 3 / cm 3 ) am.

7. In paragraph 5, The step of parameterizing the above dielectric constant model is: A method for estimating soil optimum moisture using vegetation-penetrating microwaves, which approximates the sensitivity of the saturation point parameterization function to the maximum and minimum saturation points of machine learning according to the following mathematical formula: At this time, is the maximum saturation point of the organic matter found by machine learning, is the minimum saturation point at the minimum organic matter, is the maximum saturation point at the maximum of the clay, is the minimum saturation point at the clay minimum, is the maximum organic matter in the machine learning target area, is the minimum organic matter value in the machine learning target area, OM is the organic matter value at the sampling point, is the maximum clay value in the machine learning target area, is the minimum clay value in the machine learning target area, CL is the clay value at the sampling point, is the theoretical saturation point minimum (0.1 cm 3 / cm 3 ) am.

8. In paragraph 5, The step of parameterizing the above dielectric constant model is: A method for estimating soil moisture using vegetation-penetrating microwaves, where the parameters required for the basic parameterization function before optimization are determined by physical values ​​discovered through machine learning according to the following mathematical formula: At this time, is the maximum value of the forgery point in the organic material found by machine learning, is the minimum value of the forgery point at the minimum organic matter, is the maximum value of the forging point at the maximum of the clay, is the minimum value of the forging point at the clay minimum, is the maximum organic matter in the machine learning target area, is the minimum organic matter value in the machine learning target area, OM is the organic matter value at the sampling point, is the maximum clay value in the machine learning target area, is the minimum clay value in the machine learning target area, CL is the clay value at the sampling point, is the theoretical minimum value of the forgery point (0.01 cm 3 / cm 3 ) and, is the maximum saturation point of the organic matter found by machine learning, is the minimum saturation point at the minimum organic matter, is the maximum saturation point at the maximum of the clay, is the minimum saturation point at the clay minimum, is the maximum organic matter in the machine learning target area, is the minimum organic matter value in the machine learning target area, OM is the organic matter value at the sampling point, is the maximum clay value in the machine learning target area, is the minimum clay value in the machine learning target area, CL is the clay value at the sampling point, is the theoretical saturation point minimum (0.1 cm 3 / cm 3 ) am.

Images