Soil salt content estimation method based on Hapke model

Abstract

The invention provides a soil salt content estimation method based on a Hapke model, relates to the field of remote sensing image processing and computers, and provides the soil salt content estimation method. The invention aims to solve the problem that the mobility of a model is poor as an incident angle, an azimuth angle and surface roughness are not considered in the conventional optical remote sensing soil salinity method. The method comprises the following steps: obtaining reflectivity of a plurality of zenith angles, reflectivity of azimuth angles and bidirectional reflectivity factors; an average slope angle # imgabs0 # is calculated, a Hapke expansion model is established, and omega inversion and model optimization are achieved; obtaining a single scattering albedo omega, and calculating a directional hemispherical reflectivity DHRF; calculating the specular reflectance of the soil; establishing an empirical relation model between the measured data sr and the soil salt content; calculating the salt content of the soil; the method is high in calculation speed, high in mobility and low in cost, does not need sampling measurement, can obtain a large-area and long-time sequence soil salt content estimation result, and saves a large amount of manpower and material resources.

Description

Technical Field

[0001] The invention relates to the fields of remote sensing image processing and computers, and provides a soil salt content estimation method. Background Art

[0002] According to statistics, the area of ​​saline-alkali land in the world is 9.5×10 9 hm 2 Timely and accurate prediction of soil water-salt dynamics is crucial for regional ecological protection and sustainable agricultural development. Soil salinization is mainly due to the migration and redistribution of water and salt in the soil under the influence of environmental factors and human activities. With the repeated cycles of climate change, saline-alkali soil has become an unfavorable geological soil with unstable structure, poor mechanical properties, low productivity and fragile ecosystem, which restricts the development of economy and infrastructure in arid and semi-arid areas of my country.

[0003] In the study of soil water and salt monitoring, although traditional field sampling is highly accurate, it is time-consuming, laborious, and discontinuous, and cannot achieve large-scale, real-time monitoring of soil water and salt. The development of remote sensing technology provides an effective technical means for monitoring the spatiotemporal changes of saline-alkali land characteristics in large areas and multiple phases. Compared with non-saline soil, saline soil usually exhibits higher reflectivity, which provides a basis for optical remote sensing inversion of soil salinity. At present, soil salinity inversion is mainly achieved by screening the best spectral bands and indices, combining machine learning algorithms (including neural network (BPNN), gradient boosted regression tree (GBRT), partial least squares regression (PLSR), random forest regression (RFR), support vector regression (SVR), etc.) to build empirical models for inversion. For example, the PLSR-PCC model was used to draw a soil salinity distribution map of rice planting areas in the coastal area of ​​the Jaffna Peninsula. However, this type of method does not consider information such as the angle of incidence, azimuth, and surface roughness, which makes the model poorly transferable.

[0004] Bidirectional reflectance distribution function (BRDF) has certain advantages in characterizing soil directional and spectral characteristics and generating regional or global land surface products. Therefore, BRDF is used on optical data to invert soil water and salt content. The Hapke model (1984) describes the bidirectional reflectance of bare soil. On this basis, Jacquemoud (1992) proposed a phase function P (g, g') that is approximate to the Legendre polynomials, which can be used to explain the backscattering and forward scattering of smooth soil (ie, SOILSPECT model). In 2012, Hapke proposed the concept of shadow model, which takes the average slope angle parameter Taking this into account, the Hapke extended model was formed. Scholars used the Hapke model and the extended model to invert the macroscopic roughness and single scattering albedo ω of the surface, and extended the Hapke model to the spectral domain to characterize the physical properties of the soil. For example: the equivalent water thickness was introduced into the SOILSPECT model to estimate the soil moisture content and the Hapke model was used for photometric measurement. These studies have shown that the Hapke model has great application potential in the characterization of soil hyperspectral bidirectional features. However, the current Hapke model does not consider the influence of soil salinity. Summary of the invention

[0005] The purpose of the present invention is to solve the technical problem that the current optical remote sensing soil salinity method does not take into account the incident angle, azimuth and surface roughness, resulting in poor model transferability, and to provide a soil salinity estimation method based on the Hapke model.

[0006] The soil salt content estimation method based on the Hapke model is as follows:

[0007] Step 1: For soils with different salt contents, obtain the reflectivity at multiple zenith angles, the reflectivity at azimuth angles and the bidirectional reflectivity factor (BRF);

[0008] Step 2: Obtain soil surface roughness through field measurements or public data sets to calculate the average slope angle

[0009]

[0010] Where a(θ) is a function of the inclination angle θ, and A and B are constants;

[0011] Step 3: Based on the average slope angle Establish the Hapke extension model;

[0012]

[0013] Among them, i' and e' represent the average slope Corrected incident angle and exit angle; is the roughness correction function, obtained by formula 4; ω is the single scattering albedo;

[0014] Step 4: Implement ω inversion and model optimization based on Hapke extended model;

[0015] The bidirectional reflectivity factor BRF measured in step 1 and the reflectivity at the zenith angle, the reflectivity at the azimuth angle and the average slope angle are used. As parameter input, (ω, B0, h, b, c, b', c') as output, after the initial value of the parameter is given, based on the Hapke expansion model in step 3, the nonlinear least squares fitting algorithm is used to obtain the optimal solution of each parameter; where the error function is:

[0016]

[0017] Where n is the number of multi-angle spectrum observation angle combinations, r k is the viewing angle (θ s ,θ0,φ) of multi-angle spectrum observations, r is the simulated value of the model; the algorithm for finding the minimum value can choose Levenberg-Marquardt, Newton method or gradient method, and the determination coefficient R 2 , mean square error MSE represents model accuracy:

[0018]

[0019] In the formula, y i represents the true value, f i represents the predicted value, represents the mean, n represents the sample size; the coefficient of determination R 2 The larger the value, the smaller the mean square error (MSE) value, and the higher the model accuracy. The accuracy results will be compared and the optimal algorithm will be selected.

[0020] Finally, depending on the sensitivity of the parameters (ω, B0, h, b, c, b', c', ξ, rs), some insensitive parameters are discarded to simplify the model;

[0021] Step 5: Obtain the single scattering albedo ω according to the inversion model of step 4, and calculate the directional hemisphere reflectivity DHRF (Directional Hemisphere Reflectivity Factor) according to ω;

[0022]

[0023] The bi-hemispherical reflectivity, r0, is the ratio of light scattered from the surface in all directions to the incident uncollimated light:

[0024]

[0025] From formulas (3), (8) and (9), we can know that

[0026]

[0027] Step 6: Calculate the soil specular reflectance sr (Specular Reflectance);

[0028] sr=BRF-DHRF (11)

[0029] Where BRF is obtained by spectrometer measurement, and DHRF is obtained by step five;

[0030] Step 7: Based on the measured data SR and soil salinity, establish an empirical relationship model between the two;

[0031] Step 8: Calculate the soil salt content from the SR in step 6 according to the relationship between the SR in step 7 and the soil salt content;

[0032] Step 9: Based on any satellite spectrum and surface roughness, repeat steps 4 to 8 to estimate the soil salt content.

[0033] The method includes two parts: model construction and application. When constructing the model, the multi-angle reflectance and BRF of soils with different salinity contents are first tested, and the extended Hapke model is input with the roughness parameter to obtain the single scattering albedo ω. The DHRF is calculated based on ω, and the specular reflectance sr is calculated in combination with BRF. Finally, based on the measured sr and salinity, an empirical relationship model between the two is constructed. When applying the model, satellite spectral data is brought into the extended Hapke model, and the single scattering albedo ω is obtained by the optimization method, and the DHRF is calculated. The specular reflectance sr is calculated in combination with BRF. Finally, the soil salinity is obtained based on the empirical relationship model between the measured sr and salinity.

[0034] The beneficial effects of the present invention are:

[0035] Compared with the traditional indoor soil salinity measurement method, the present invention provides a large-area, fast soil salinity estimation method; compared with the commonly used soil salinity estimation method based on spectral inversion, the present invention provides a transferable soil salinity estimation method. This method has fast calculation speed, strong transferability, low cost, and does not require sampling measurement. It can obtain large-area, long-term series soil salinity estimation results, saving a lot of manpower and material resources, and is of great significance for studying the spatiotemporal changes of soil salinity. The present invention will construct a soil salinity estimation method based on the Hapke model based on the measured multi-angle BRDF data, which has important application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is the algorithm flow chart of the present invention;

[0037] Figure 2 It is the Sentinel-2 satellite image data of saline soil in western Jilin Province in Experiment 1;

[0038] Figure 3 This is the empirical relationship diagram between soil salinity and specular reflectance sr in Experiment 1;

[0039] Figure 4 This is the result of estimating the salt content of saline soil in western Jilin Province using this method in Experiment 1;

[0040] Figure 5 This is the result of estimating the salt content of saline soil in western Jilin Province using this method in Experiment 1. DETAILED DESCRIPTION

[0041] The technical solution of the present invention is not limited to the specific implementation modes listed below, but also includes any combination of the specific implementation modes.

[0042] Specific implementation method 1: This implementation method

[0043] The soil salt content estimation method based on the Hapke model is as follows:

[0044] Step 1: For soils with different salt contents, obtain the reflectivity at multiple zenith angles, the reflectivity at multiple azimuth angles, and the bidirectional reflectivity factor (BRF);

[0045] Step 2: Obtain soil surface roughness through field measurements or public data sets to calculate the average slope angle

[0046] Where a(θ) is a function of the inclination angle θ, and A and B are constants;

[0047] Step 3: Based on the average slope angle Establish the Hapke extension model;

[0048]

[0049] Among them, i' and e' represent the average slope Corrected incident angle and exit angle; is the roughness correction function, obtained by formula 4; ω is the single scattering albedo;

[0050] Step 4: Implement ω inversion and model optimization based on Hapke extended model;

[0051] The bidirectional reflectivity factor BRF measured in step 1 and the reflectivity at the zenith angle, the reflectivity at the azimuth angle and the average slope angle are used. The parameters are taken as input and (ω, B0, h, b, c, b', c') are taken as output. After the initial values ​​of the parameters are given, the nonlinear least squares fitting algorithm is used to obtain the optimal solution of each parameter based on the Hapke extended model in step 3; the error function is:

[0052]

[0053] Where n is the number of multi-angle spectrum observation angle combinations, r k is the viewing angle (θ s ,θ0,φ) of multi-angle spectrum observations, r is the simulated value of the model; the algorithm for finding the minimum value can choose Levenberg-Marquardt, Newton method or gradient method, and the determination coefficient R 2 , mean square error MSE represents model accuracy:

[0054]

[0055] In the formula, y i represents the true value, f i represents the predicted value, represents the mean, n represents the sample size; the coefficient of determination R 2 The larger the value, the smaller the mean square error (MSE) value, and the higher the model accuracy. The accuracy results will be compared and the optimal algorithm will be selected.

[0056] Finally, depending on the sensitivity of the parameters (ω, B0, h, b, c, b', c', ξ, rs), some insensitive parameters are discarded to simplify the model;

[0057] Step 5: Obtain the single scattering albedo ω according to the inversion model of step 4, and calculate the directional hemisphere reflectivity DHRF (Directional hemisphere reflectivity factor) according to ω;

[0058]

[0059] The bi-hemispherical reflectivity, r0, is the ratio of light scattered from the surface in all directions to the incident uncollimated light:

[0060]

[0061] From formulas (3), (8) and (9), we can know that

[0062]

[0063] Step 6: Calculate the soil specular reflectance sr (Specular reflectance);

[0064] sr=BRF-DHRF (11)

[0065] Where BRF is obtained by spectrometer measurement, and DHRF is obtained by step five;

[0066] Step 7: Based on the measured data SR and soil salinity, establish an empirical relationship model between the two;

[0067] Step 8: Calculate the soil salt content from the SR in step 6 according to the relationship between the SR in step 7 and the soil salt content;

[0068] Step 9: Based on any satellite spectrum and surface roughness, repeat steps 4 to 8 to estimate the soil salt content.

[0069] The following experiments were used to verify the effect of the present invention:

[0070] like Figure 1 As shown in the figure, the soil salt content estimation method based on the Hapke model includes two parts: model construction and application. When constructing the model, the multi-angle reflectance and BRF of soils with different salt contents are first tested, and the extended Hapke model is input with the roughness parameter to invert the single scattering albedo ω, and the DHRF is calculated based on ω, and the specular reflectance sr is calculated in combination with BRF. Finally, based on the measured sr and salt content, an empirical relationship model between the two is constructed. When applying the model, satellite spectral data is brought into the extended Hapke model, the single scattering albedo ω is obtained by the optimization method, and the DHRF is calculated, and the specular reflectance sr is calculated in combination with BRF. Finally, the soil salt content is obtained based on the empirical relationship model between the measured sr and salt content.

[0071] Experiment 1: Taking the estimation of salinity of saline soil in western Jilin Province based on the Sentinel-2 satellite as an example, the soil salinity estimation method based on the Hapke model is as follows:

[0072] Step 1: Collect 57 soil samples with different salt contents, and obtain the reflectance at azimuth angles of 0°, 90° and 180°, the reflectance at zenith angles of -60° to 60° with an interval of 30°, and the bidirectional reflectance factor (BRF) in a dark room;

[0073] Step 2: Use Leica Total Station (Model TZ12) to obtain the surface roughness of the soil in Step 1 and calculate the average slope angle

[0074]

[0075] The results show that the soil surface is Gaussian distributed, so a(θ) uses the Gaussian distribution function:

[0076]

[0077] The present invention uses the isotropic undecimated wavelet transform (IUWT) algorithm, and the mean of the curve represents the center position of the data, that is, the average slope angle

[0078] Step 3: Based on the average slope angle Establish the Hapke extension model;

[0079]

[0080] Among them, i' and e' represent the average slope Corrected incident angle and exit angle; is the roughness correction function, obtained by formula 4; ω is the single scattering albedo;

[0081] Step 4: Implement ω inversion and model optimization based on Hapke extended model;

[0082] The bidirectional reflection factor BRF measured in step 1 and the reflectivity at the zenith angle, the reflectivity at the azimuth angle and the average slope angle are used. Parameters are used as input and parameters (ω, B0, h, b, c, b', c') are used as output. After the initial values ​​of the parameters are given, the optimal solution of each parameter is obtained based on the Hapke extended model in step 3 using the nonlinear least squares fitting algorithm. The error function is:

[0083]

[0084] When finding the minimum value, you can choose Levenberg-Marquardt, Newton method, gradient method, etc., and use the determination coefficient R 2 , mean square error MSE represents model accuracy:

[0085]

[0086] Levenberg-Marquardt was selected as the optimal algorithm based on the accuracy results.

[0087] The results of global sensitivity analysis show that the parameters c, c', h, B0 and others have low sensitivity. When establishing the inversion model, dimensionality reduction is performed and they are set to fixed values ​​to reduce the computational complexity of the inversion model.

[0088] Step 5: Extract the band reflectivity and incident angle parameters of the Sentinal-2 satellite in western Jilin Province from the ESA official website (https: / / earthexplorer.usgs.gov / ) Figure 2 ), the single scattering albedo ω is obtained according to the inversion model in step 4, and the DHRF is calculated according to ω:

[0089]

[0090] The bi-hemispherical reflectivity, r0, is the ratio of light scattered from the surface in all directions to the incident uncollimated light:

[0091]

[0092] From formulas (3), (7) and (8), we can know that

[0093]

[0094] Step 6: Calculate the soil specular reflectance sr:

[0095] sr=BRF-DHRF (11)

[0096] Where BRF is obtained by spectrometer measurement, and DHRF is obtained by step five;

[0097] Step 7: Based on the measured data SR and soil salinity, establish an empirical relationship model between the two; based on the relationship between SR and soil salinity, calculate the soil salinity from the SR obtained in step 6. The result is as follows: Figure 3 shown.

[0098] Step 8. Based on the relationship in step 7, the Sentinel-2 satellite data B-band and G-band image data are input to obtain the salt content of saline soil in western Jilin Province. The results are as follows: Figures 4-5 shown.

Claims

1. Soil salt content estimation method based on Hapke model, characterized by The soil salinity estimation method based on the Hapke model is as follows: Step 1: For soils with different salt contents, obtain the reflectivity at multiple zenith angles, the reflectivity at multiple azimuth angles, and the bidirectional reflectivity factor (BRF); Step 2: Obtain soil surface roughness through field measurements or public data sets to calculate the average slope angle Where a(θ) is a function of the inclination angle θ, and A and B are constants; Step 3: Based on the average slope angle Establish the Hapke extension model; Among them, i′ and e′ represent the average slope Corrected incident angle and exit angle; is the roughness correction function, obtained by formula 4; ω is the single scattering albedo; Step 4: Implement ω inversion and model optimization based on Hapke extended model; The bidirectional reflectivity factor BRF measured in step 1 and the reflectivity at the zenith angle, the reflectivity at the azimuth angle and the average slope angle are used to calculate the reflectivity. The parameters are taken as input and (ω, B0, h, b, c, b′, c′) are taken as output. After the initial values ​​of the parameters are given, the optimal solution of each parameter is obtained by using the nonlinear least squares fitting algorithm based on the Hapke extended model in step 3; the error function is: Where n is the number of multi-angle spectrum observation angle combinations, r k is the viewing angle (θ s , θ0, φ) of multi-angle spectrum observations, r is the simulated value of the model; the algorithm for finding the minimum value can choose Levenberg-Marquardt, Newton method or gradient method, and the determination coefficient R 2 , mean square error MSE represents model accuracy: In the formula, y i represents the true value, f i represents the predicted value, represents the mean, n represents the sample size; the coefficient of determination R 2 The larger the value, the smaller the mean square error (MSE) value, and the higher the model accuracy. The accuracy results will be compared and the optimal algorithm will be selected. Finally, depending on the sensitivity of the parameters (ω, B0, h, b, c, b′, c′, ξ, rs), some insensitive parameters are discarded to simplify the model; Step 5: Obtain the single scattering albedo ω according to the inversion model of step 4, and calculate the directional hemispherical reflectivity DHRF according to ω; The bi-hemispherical reflectivity, r0, is the ratio of light scattered from the surface in all directions to the incident uncollimated light: From formulas (3), (8) and (9), we can know that Step 6: Calculate the soil specular reflectance sr; sr=BRF-DHRF (11) Where BRF is obtained by spectrometer measurement, and DHRF is obtained by step five; Step 7: Based on the measured data SR and soil salinity, establish an empirical relationship model between the two; Step 8: Calculate the soil salt content using the SR obtained in step 6 according to the relationship between the SR in step 7 and the soil salt content; Step 9: Based on any satellite spectrum and surface roughness, repeat steps 4 to 8 to estimate the soil salt content.

Images