Method for inverting soil water content through active and passive remote sensing

Abstract

The invention discloses a method for carrying out soil water content inversion through active and passive remote sensing. The method comprises the steps that optical image data, radar data and actual soil detection data are preprocessed; obtaining a backscattering coefficient, a vegetation index, surface roughness and an incident angle by calculating the preprocessed data; extreme learning machine model input parameters are obtained through water cloud model processing; inputting parameters through the extreme learning machine model to construct and train the extreme learning machine model; an output result obtained through the extreme learning machine model is the soil water content; and running the extreme learning machine model through a Google Earth Engine remote sensing calculation platform to obtain the soil water content. According to the method, simulation is carried out by means of a Google Earth Engine remote sensing computing platform during data inversion and verification. By combining the GEE cloud platform with the inversion model, multi-source data can be effectively integrated, the operation efficiency is greatly improved, and the complexity of data collection and calculation processes in the inversion model is reduced.

Description

Technical Field

[0001] The present application relates to the technical field of soil detection, and in particular to a method for soil moisture inversion using active and passive remote sensing. Background Art

[0002] On the Loess Plateau, soil moisture is the direct source of water required for vegetation growth, and plays a very important role in agricultural water management, crop yield estimation, drought monitoring and precision irrigation. Therefore, it is of great significance to quickly and accurately monitor the spatial changes of soil moisture.

[0003] The traditional method of measuring soil moisture by probe method or weight method has high accuracy, but it cannot meet the needs of large-scale soil moisture monitoring.

[0004] Nowadays, the emergence of remote sensing technology has provided a very important technical means for research in this area. Remote sensing technology has the characteristics of wide coverage and strong timeliness, and is currently the main means of large-scale soil moisture monitoring. Optical remote sensing can provide data with high spectral resolution and high spatial resolution. Soil moisture monitoring technology based on optical remote sensing is mainly suitable for exposed surfaces. In vegetation coverage areas, it is impossible to directly obtain soil spectral information, resulting in limited soil moisture monitoring capabilities. Thermal infrared remote sensing can estimate soil moisture based on changes in surface temperature. In low vegetation coverage areas, the thermal inertia method is the most commonly used inversion method, because soil thermal inertia increases with the increase of soil moisture content, resulting in a decrease in the day and night temperature difference on the surface; in medium and high vegetation coverage areas, the temperature vegetation drought index method is a common method for combining optical and thermal infrared remote sensing for soil moisture inversion or soil drought degree monitoring. Optical and thermal infrared remote sensing are easily affected by atmospheric conditions such as clouds and fog, and the continuity of soil moisture monitoring is difficult to guarantee. Synthetic aperture radar has strong atmospheric penetration ability and has the ability to observe the earth all day long under various meteorological conditions. In addition, the SAR echo intensity is highly sensitive to soil moisture changes. In vegetation-covered areas, due to the strong scattering of microwaves by the vegetation layer, the sensitivity of the SAR backscatter coefficient to soil moisture changes decreases, reducing the inversion accuracy of soil moisture. Summary of the invention

[0005] The embodiment of the present application provides a method for inverting soil moisture content by active and passive remote sensing, which is used to solve the problem that the soil moisture monitoring capability of the soil moisture monitoring technology based on optical remote sensing in the prior art is limited, optical and thermal infrared remote sensing are easily affected by atmospheric conditions such as clouds and fog, and the continuity of soil moisture monitoring is difficult to ensure. In vegetation-covered areas, due to the strong scattering effect of the vegetation layer on microwaves, the sensitivity of the SAR backscattering coefficient to soil moisture changes decreases, thereby reducing the inversion accuracy of soil moisture.

[0006] On the one hand, an embodiment of the present application provides a method for soil moisture inversion using active and passive remote sensing, comprising:

[0007] Preprocessing the optical image data, the radar data and the actual soil detection data to obtain preprocessed data of the optical image data, the radar data and the actual soil detection data;

[0008] Obtaining a backscattering coefficient, a vegetation index, a surface roughness and an incident angle by calculating the preprocessed data;

[0009] Processing the backscattering coefficient, vegetation index, surface roughness and incident angle through a water cloud model to obtain input parameters of an extreme learning machine model;

[0010] Constructing and training an extreme learning machine model by inputting parameters of the extreme learning machine model;

[0011] The output result obtained by the extreme learning machine model input parameters and the extreme learning machine model is soil moisture content;

[0012] The soil moisture content is obtained by running the extreme learning machine model through the Google Earth Engine remote sensing computing platform.

[0013] In a possible implementation, the preprocessing of optical image data is performed by fusing a color standardization transformation method, a phase recovery method, a color space transformation method, a principal component transformation method, and a super-resolution Bayesian method; the preprocessing of radar data is performed by correcting the radar data using ENVI; and the preprocessing of actual soil detection data is performed by setting the measurement value of each sampling point to the average of the measurement values ​​of the sampling point and three neighboring sampling points within a range of 1 m around it.

[0014] In a possible implementation manner, the backscatter coefficient, surface roughness and incident angle are obtained by calculating preprocessed data of the radar data, and the vegetation index is obtained by calculating preprocessed data of the optical image data.

[0015] In a possible implementation, the vegetation index is a leaf area index and a normalized vegetation index extracted from the optical image data by ENVI.

[0016] In a possible implementation, the water cloud model performs inversion calculation through the backscattering coefficient, vegetation index, surface roughness and incident angle to obtain soil moisture inversion data with vegetation effects removed.

[0017] In a possible implementation, after obtaining the output result of soil moisture through the extreme learning machine model input parameters and the extreme learning machine model, the method further includes performing an accuracy test on the soil moisture content through a root mean square error (RMSE).

[0018] In a possible implementation, the root mean square error RMSE is obtained by calculating the inversion value and the measured value.

[0019] In a possible implementation, the soil moisture content obtained by running the extreme learning machine model through the Google Earth Engine remote sensing computing platform is obtained by accelerating the calculation rate of the extreme learning machine model through the Google Earth Engine remote sensing computing platform.

[0020] The method of soil moisture inversion by active and passive remote sensing in this application has the following advantages:

[0021] (1) Eliminate the influence of vegetation on inversion by combining optical images with radar data.

[0022] (2) The Google Earth Engine remote sensing computing platform is used for simulation during data inversion and verification. Combining the GEE cloud platform with the inversion model can effectively integrate multi-source data, greatly improve computing efficiency, and reduce the complexity of data collection and calculation processes in the inversion model. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0024] Figure 1 A schematic diagram of a method for soil moisture inversion using active and passive remote sensing provided in an embodiment of the present application;

[0025] Figure 2 A schematic diagram of field sampling of a method for soil moisture inversion using active and passive remote sensing provided in an embodiment of the present application. DETAILED DESCRIPTION

[0026] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0027] Figure 1 A schematic diagram of a method for inverting soil moisture content by active and passive remote sensing provided in an embodiment of the present application. An embodiment of the present application provides a method for inverting soil moisture content by active and passive remote sensing, comprising:

[0028] Preprocessing the optical image data, the radar data and the actual soil detection data to obtain preprocessed data of the optical image data, the radar data and the actual soil detection data;

[0029] Obtaining a backscattering coefficient, a vegetation index, a surface roughness and an incident angle by calculating the preprocessed data;

[0030] Processing the backscattering coefficient, vegetation index, surface roughness and incident angle through a water cloud model to obtain input parameters of an extreme learning machine model;

[0031] Constructing and training an extreme learning machine model by inputting parameters of the extreme learning machine model;

[0032] The output result obtained by the extreme learning machine model input parameters and the extreme learning machine model is soil moisture content;

[0033] The soil moisture content is obtained by running the extreme learning machine model through the Google Earth Engine remote sensing computing platform.

[0034] Exemplarily, the optical image data is first preprocessed. The embodiment uses Landsat8-OLI images of the hilly and gully areas of the Loess Plateau. Five fusion methods, including color normalization transformation (Brovey) method, phase recovery (Gram-Schmidt) method, chromaticity space transformation (HSV) method, principal component transformation (PCA) method and super-resolution Bayesian (Pansharp) method, are used to carry out fusion experiments on the images of the experimental area. First, the five fusion results are visually evaluated, and then five indicators, including standard deviation, information entropy, average gradient, correlation coefficient and distortion degree, are constructed to quantitatively evaluate the fusion results.

[0035] Then the radar data is preprocessed. The radar data is Sentinel-1B data. During the data preprocessing process, ENVI is used to perform correction preprocessing on the radar data.

[0036] Finally, the actual soil test data is preprocessed. The field experiment work involved in this embodiment is mainly divided into the layout of sampling points, the collection of soil volume water content and soil sample sampling. The quality of field work is related to the accuracy of indoor analysis. Therefore, the field measured data as first-hand data lays the foundation for this study.

[0037] (1) Field sampling point layout plan

[0038] The sampling point of this embodiment is located in the Yangjuangou small watershed in Liqu Town, Yan'an City. There is a soil moisture TDR device for long-term monitoring. A portable GPS receiver is used to locate the sampling point and record its longitude, latitude, altitude and other data. The spacing between each sampling point is 30m. Figure 2 shown.

[0039] (2) Soil volumetric moisture content measurement and soil sampling

[0040] The soil volumetric moisture content was measured using a WET instrument to measure the average soil volumetric moisture content on the surface. The measurement value of each sampling point is the average of the measurement values ​​of the point and the three points within 1m around it. At the same time, some sampling points were sampled with aluminum boxes and ring cutters, and then the aluminum box soil samples and ring cutter soil samples were brought into the laboratory for testing.

[0041] The backscattering coefficient, vegetation index, surface roughness and incident angle are obtained by calculating the preprocessed data.

[0042] (1) Radar data backscatter coefficient extraction is first calculated through radar images. The radar equation is used to describe the relationship between the echo signal received by the radar antenna and the radar system parameters and target scattering characteristics. The radar antenna transmits a spherical wave to the target object, and the target object then reflects the same spherical wave to the radar. The radar image backscatter value is extracted using NEST and ENVI software. The value is obtained by converting the DN value of each grid point on the remote sensing image:

[0043]

[0044] Where: σ is the backscatter coefficient; A is the DN value on the remote sensing image; θ is the radar incident angle; K is the absolute calibration factor.

[0045] Then the backscattering coefficient σ is corrected according to the incident angle, and the calculation formula is as follows:

[0046]

[0047] Where: γ is the backscattering coefficient corrected by the incident angle.

[0048] (2) Extraction of vegetation index and leaf area index

[0049] In this study, ENVI software is needed to extract leaf area index LAI and normalized vegetation index NDVI from Landsat images. In vegetation-covered areas, the amount of water in the soil affects the growth of surface vegetation. Therefore, the distribution of soil moisture can be indirectly inverted through the vegetation index. Generally speaking, the more soil moisture there is, the larger the vegetation index is. Especially in arid and semi-arid areas, soil moisture plays a decisive and controlling role in vegetation growth, and there is a positive correlation between the two. Therefore, the vegetation index method can be used to indirectly estimate soil moisture. This article uses the normalized vegetation index NDVI to reflect the growth of vegetation, which is defined as the following formula:

[0050] NDVI=(ρ nir -ρ red ) / (ρ nir +ρ red ) (3)

[0051] Among them, ρ nir and ρ red are the reflectances of the near-infrared and visible red bands, respectively.

[0052] Leaf area index LAI is an important indicator reflecting the plant coverage. Its calculation formula is:

[0053]

[0054] Where n is the total number of leaves of the Jth plant; m is the number of plants measured; ρ is the planting density; L ij is the leaf length of each leaf, B ij The maximum leaf width.

[0055] The backscattering coefficient, vegetation index, surface roughness and incident angle are processed by a water cloud model to obtain the input parameters of the extreme learning machine model, that is, the scattering coefficient, vegetation index, surface roughness and incident angle parameters are input into the water cloud model for inversion.

[0056] (1) Calculation of soil backscatter coefficient

[0057] In the study of soil moisture inversion, a variety of models have been proposed, but most models require the input of multiple measured data, and the models need to be recalculated for different regions, which is not suitable for soil moisture inversion in a wide area. This embodiment selects a water cloud model suitable for soil moisture inversion in the study area.

[0058]

[0059] σ 0 (θ) is the total radar backscatter coefficient under the vegetation cover surface; σ 0 vegis the direct backscattering coefficient of the vegetation layer; σ 0 soil is the direct backscattering coefficient of the soil; γ 2 (θ) is the double-layer attenuation factor of the crop layer, and is a function related to θ. The backscatter value of the radar image is extracted using ENVI software. Among them, the value is obtained by converting the DN value of each grid point on the remote sensing image, as shown in formula (1)(2).

[0060] (2) Simulation of surface roughness By improving the integral equation model IEM, an advanced integral equation AIEM is obtained, which is expressed as follows:

[0061]

[0062]

[0063] Where pq represents the polarization mode; k is the wave number; S is the root mean square height of the ground surface; θ is the incident angle of the radar wave; k z =cosθ,k x =ksinθ;W n (-2k x, 0) is the surface roughness; n-order power spectrum; F pq and f pq It is a function of the Fresnel reflection coefficient and is related to various soil parameters including soil moisture. The calculation model based on the root mean square height and correlation length parameters can better reflect the relationship between the backscattering coefficient and the surface roughness in the simulation with higher accuracy.

[0064] (3) Determine model input parameters

[0065] The input samples of the model are selected as NDVI, soil backscatter coefficient, incident angle and surface roughness.

[0066] Finally, the entire model was set up on the Google Earth Engine remote sensing computing platform for simulation. Based on the GEE cloud platform and the inversion model driven by active and passive collaborative remote sensing data, a platform framework for soil moisture estimation in the Loess Plateau was built. This can effectively integrate multi-source data, greatly improve computing efficiency, and reduce the complexity of data collection and calculation processes in the inversion model.

[0067] The extreme learning machine model (ELM model) is characterized by the fact that the input weights are randomly or artificially given, and the output weights need to be calculated by solving a set of equations during the entire calculation process. No iterative calculation is required, so that the learning process can quickly reach the optimal solution while meeting the experimental accuracy.

[0068] The expression of extreme learning machine is:

[0069] y=Hβ(8)

[0070] H=g(ω i ,x,b)(9)

[0071] Where: g(ω i ,x,b) is the activation function; β is the output layer weight: ω i is the input layer weight, x is the input sample, b is the bias value, and y is the output sample. When using the EI,M model for training, h is randomly or artificially given, and H is actually a known quantity. From formula (10), it can be seen that the model is converted into a problem of solving the output weight, which greatly simplifies the setting process of the learning parameters of the feedback neural network model.

[0072] β^=H T y(10)

[0073] When using the extreme learning machine model for inversion, the sample data is first divided into two parts, two-thirds of the samples are used to build the model, and the remaining one-third is used to verify the inversion accuracy. Model mat1ab implementation.

[0074] In a possible embodiment, the preprocessing of optical image data is performed by fusing a color standardization transformation method, a phase recovery method, a color space transformation method, a principal component transformation method, and a super-resolution Bayesian method; the preprocessing of radar data is performed by using ENVI to calibrate the radar data; the preprocessing of actual soil detection data is performed by setting the measurement value of each sampling point to the average value of the measurement values ​​of the sampling point and three neighboring sampling points within a range of 1 m around the sampling point; the backscatter coefficient, surface roughness, and incident angle are calculated and obtained through the preprocessing data of the radar data; and the vegetation index is calculated through the optical The vegetation index is obtained by calculating the preprocessing data of the optical image data through ENVI, the vegetation index is the leaf area index and the normalized vegetation index extracted from the optical image data through ENVI, the water cloud model performs inversion calculation through the backscattering coefficient, the vegetation index, the surface roughness and the incident angle to obtain the soil moisture inversion data without the influence of vegetation, the input parameters of the extreme learning machine model and the output result of the extreme learning machine model are the soil moisture content, and the accuracy of the soil moisture content is also tested through the root mean square error RMSE, the root mean square error RMSE is obtained by calculating the inversion value and the measured value, and the soil moisture content obtained by running the extreme learning machine model through the Google Earth Engine remote sensing computing platform is the calculation rate of the extreme learning machine model accelerated by the Google EarthEngine remote sensing computing platform.

[0075] Exemplarily, after the extreme learning machine model input parameters and the extreme learning machine model obtain an output result of soil moisture content, it is also necessary to perform an accuracy test on the soil moisture content through a root mean square error RMSE.

[0076] Considering the influence of vegetation on soil moisture inversion, remote sensing datasets and ground observation datasets are used to implement the inversion experiment and verification of soil moisture under vegetation cover. The accuracy test can be evaluated based on the root mean square error (RMSE) between the measured value and the inverted value of soil relative humidity. The specific calculation formula is:

[0077]

[0078] Where: S i and M i are the measured value and inverted value of soil relative humidity respectively; i is the sample number; n is the number of samples.

[0079] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0080] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A method for soil moisture inversion using active and passive remote sensing, characterized in that: include: Preprocessing the optical image data, the radar data and the actual soil detection data to obtain preprocessed data of the optical image data, the radar data and the actual soil detection data; Obtaining a backscattering coefficient, a vegetation index, a surface roughness and an incident angle by calculating the preprocessed data; Processing the backscattering coefficient, vegetation index, surface roughness and incident angle through a water cloud model to obtain input parameters of an extreme learning machine model; Constructing and training an extreme learning machine model by inputting parameters of the extreme learning machine model; The output result obtained by the extreme learning machine model input parameters and the extreme learning machine model is soil moisture content; The soil moisture content is obtained by running the extreme learning machine model through the Google Earth Engine remote sensing computing platform.

2. The method for soil moisture inversion by active and passive remote sensing according to claim 1 is characterized in that: The preprocessing of the optical image data is performed by fusing a color standardization transformation method, a phase recovery method, a chromaticity space transformation method, a principal component transformation method and a super-resolution Bayesian method. The preprocessing of the radar data is performed by using ENVI to calibrate the radar data. The preprocessing of the actual soil detection data is performed by setting the measurement value of each sampling point to the average of the measurement values ​​of the sampling point and three adjacent sampling points within a range of 1m around it.

3. The method for soil moisture inversion by active and passive remote sensing according to claim 1 is characterized in that: The backscatter coefficient, surface roughness and incident angle are obtained by calculating the preprocessed data of the radar data, and the vegetation index is obtained by calculating the preprocessed data of the optical image data.

4. The method for soil moisture inversion by active and passive remote sensing according to claim 3 is characterized in that: The vegetation index is the leaf area index and the normalized difference vegetation index extracted from the optical image data by ENVI.

5. The method for soil moisture inversion by active and passive remote sensing according to claim 1, characterized in that: The water cloud model performs inversion calculation through the backscattering coefficient, vegetation index, surface roughness and incident angle to obtain soil moisture inversion data with vegetation influence removed.

6. The method for soil moisture inversion by active and passive remote sensing according to claim 1, characterized in that: After obtaining the output result of soil moisture as the result of inputting parameters of the extreme learning machine model and the extreme learning machine model, the method also includes performing an accuracy test on the soil moisture content through a root mean square error (RMSE).

7. The method for soil moisture inversion by active and passive remote sensing according to claim 6 is characterized in that: The root mean square error RMSE is obtained by calculating the inversion value and the measured value.

8. The method for soil moisture inversion by active and passive remote sensing according to claim 7 is characterized in that: The soil moisture content obtained by running the extreme learning machine model through the Google Earth Engine remote sensing computing platform is a calculation rate of the extreme learning machine model accelerated by the Google Earth Engine remote sensing computing platform.

Images