GNSS-R frozen soil freezing and thawing classification method based on machine learning model

Abstract

The invention provides a GNSS-R frozen soil freezing and thawing classification method based on a machine learning model. The method comprises the following steps of performing quality control on satellite-borne GNSS-R data; satellite-borne GNSS-R data observation quantity extraction and surface reflectance calculation are carried out; carrying out gridding on the satellite-borne GNSS-R surface reflectance data; preprocessing the SMAP data; constructing an inversion model; and satellite-borne GNSS-R frozen soil freeze thawing inversion and precision evaluation are carried out. The method has the beneficial effects that the GNSS-R frozen soil freeze-thaw inversion algorithm based on the random forest model is introduced, so that the space-time coverage range, the inversion precision and the reliability of frozen soil freeze-thaw state data are remarkably improved; data support is provided for key fields of climate change research, cold region engineering planning, ecological environment management, hydrogeological modeling and the like, and decision scientificity and practice effectiveness are further enhanced.

Description

Technical Field

[0001] The present invention belongs to the field of satellite-borne GNSS-R applications, and in particular relates to a GNSS-R frozen soil freeze-thaw classification method based on a machine learning model. Background Art

[0002] The freeze-thaw state of permafrost has an important impact on the global climate system, hydrological cycle and ecosystem balance, and its dynamic change characteristics are crucial to environmental science research, agricultural planning, disaster warning and other fields. However, traditional monitoring methods are often limited by small coverage, low temporal resolution and poor environmental adaptability, making it difficult to meet large-scale, high-precision monitoring needs.

[0003] Spaceborne GNSS-R technology, with its low cost, low power consumption, high revisit rate and all-weather working ability, provides a new solution for remote sensing monitoring of frozen soil thawing status. This technology can indirectly obtain the physical characteristics of the surface by receiving and analyzing the GNSS signals reflected from the surface. The complex process of frozen soil thawing is also affected by many factors, such as soil type, groundwater level, temperature fluctuations and vegetation type, which makes it extremely challenging to directly extract freeze-thaw status information from GNSS-R observation data. Summary of the invention

[0004] In view of this, the present invention aims to propose a GNSS-R frozen soil freeze-thaw classification method based on a machine learning model to solve the problem of high-precision real-time monitoring of the frozen soil freeze-thaw state in the process of inverting the frozen soil freeze-thaw state by satellite-borne GNSS-R technology.

[0005] To achieve the above object, the technical solution of the present invention is achieved as follows: A GNSS-R frozen soil freeze-thaw classification method based on a machine learning model includes the following steps: S1. Perform quality control on onboard GNSS-R data; S2, extraction of satellite-borne GNSS-R data observations and calculation of surface reflectivity; S3, gridding the satellite-borne GNSS-R surface reflectivity data; S4, SMAP data preprocessing; S5, inversion model construction; S6. Spaceborne GNSS-R frozen soil freeze-thaw inversion and accuracy assessment; In step S2, the onboard GNSS-R data observations are extracted and the surface reflectivity is calculated, including: S21, extract the latitude and longitude coordinates of the satellite-borne GNSS-R mirror point and other observation values ​​used to calculate the surface reflectivity; S22. Calculate the surface reflectivity by combining the power expression of the coherent component and the power expression of the incoherent scattered component.

[0006] Furthermore, in step S1, the quality control of the spaceborne GNSS-R data includes: S11, converting the quality label data into binary, and determining the data quality according to the bit value, wherein the label bits used include black body load delay Doppler pattern mode, radio frequency interference and test mode; S12. Perform secondary screening on the data according to empirical criteria, and eliminate data with an incident angle greater than 65°, a signal-to-noise ratio less than 0, or an antenna gain less than 0.

[0007] Further, in step S22, the power expression of the coherent component is: (1) In the formula, is the coherent scattering component power; the subscript r represents the right-hand circularly polarized antenna of the GNSS transmitter, and l represents the left-hand circularly polarized antenna of the GNSS-R receiver; is the signal wavelength; is the peak power of the transmitted signal; , are the gains of the transmitting antenna and the receiving antenna respectively; , are the distances between the transmitter, receiver and the mirror reflection point respectively; represents the surface reflectivity; The power expression of the incoherent scattered component is: (2) In the formula, is the incoherent scattering component power; is the bistatic radar scattering coefficient, in m 2 ; make = , combine formula (1) and formula (2) to get the surface reflectivity writing Function, such as formula (3): (3) In the formula, represents the surface reflectivity; , are the distances between the transmitter, receiver and the mirror reflection point respectively; is the bistatic radar scattering coefficient, in m 2 .

[0008] Furthermore, in step S3, the spaceborne GNSS-R surface reflectivity data is gridded, including: The EASE-Grid of SMAP data is used to grid the surface reflectivity of the space-borne GNSS-R, and the surface reflectivity of the sampling points falling within each grid is averaged.

[0009] Furthermore, in step S4, SMAP data preprocessing includes: S41. Extract soil moisture estimation, quality label, vegetation optical depth, land cover type, and grid coordinates of ascending and descending regions based on SMAP L3 soil moisture products as auxiliary input features for the classification model. S42. Extract and process frozen soil freeze-thaw data based on SMAP L3 frozen soil freeze-thaw products as the output characteristics of the model and reference data for accuracy evaluation; S43, merging the ascending track and the descending track to represent the result of one day, and ignoring the null value of the data of three consecutive days to average them; S44. Filter SMAP data using quality tags.

[0010] Furthermore, in step S5, the inversion model is constructed, including: S51, spaceborne GNSS-R pseudo-random sampling points are temporally and spatially matched with SMAP; S52. Construct a data set and randomly divide all daily observation data into two groups, one group is a training set accounting for 70% of the total data, and the other group is a test set accounting for 30% of the total data; S53, model training, the hyperparameters of the random forest classifier were tuned using the 10-fold cross-validation method.

[0011] Furthermore, in step S6, the spaceborne GNSS-R frozen soil freeze-thaw inversion and accuracy assessment includes: The input features are substituted into the inversion model to obtain the spaceborne GNSS-R soil moisture inversion results, and the quality of the results is evaluated from three perspectives: model accuracy, temporal variability, and spatial variability.

[0012] Compared with the prior art, the GNSS-R frozen soil freeze-thaw classification method based on a machine learning model described in the present invention has the following advantages: The present invention aims to solve the problem of high-precision real-time monitoring of frozen soil freeze-thaw status inversion using satellite-borne GNSS-R technology, which is directly related to the scientificity and timeliness of decision-making in the fields of cold region engineering, frozen soil ecology, climate change research, and disaster risk assessment. The present invention introduces a GNSS-R frozen soil freeze-thaw inversion algorithm based on a random forest model, which significantly improves the temporal and spatial coverage, inversion accuracy, and reliability of frozen soil freeze-thaw status data, provides data support for key areas such as climate change research, cold region engineering planning, ecological environment management, and hydrogeological modeling, and further enhances the scientificity of decision-making and practical effectiveness. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The accompanying drawings constituting a part of the present invention are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the accompanying drawings: Figure 1 A technical roadmap for the embodiments of the present invention; Figure 2 The GNSS-R frozen soil freeze-thaw distribution map described in the embodiment of the present invention; Figure 3 This is a time series diagram of the accuracy of frozen soil freeze-thaw classification using the satellite-borne GNSS-R according to an embodiment of the present invention. DETAILED DESCRIPTION

[0014] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0015] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and the like are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, features defined as "first", "second", and the like may explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, "multiple" means two or more.

[0016] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood by specific circumstances.

[0017] The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0018] like Figures 1 to 3 As shown, a GNSS-R frozen soil freeze-thaw classification method based on a machine learning model includes the following steps: S1. Perform quality control on onboard GNSS-R data; S2, extraction of satellite-borne GNSS-R data observations and calculation of surface reflectivity; S3, gridding the satellite-borne GNSS-R surface reflectivity data; S4, SMAP (Soil Moisture Active Passive) data preprocessing; S5, inversion model construction; S6. Spaceborne GNSS-R frozen soil freeze-thaw inversion and accuracy assessment.

[0019] The present invention introduces a random forest classifier, and uses its powerful ability to handle nonlinear relationships, excellent generalization performance, and high robustness to construct a multi-parameter fusion GNSS-R frozen soil freeze-thaw inversion model. The random forest model can comprehensively consider the impact of various surface parameters on the reflection signal, effectively distinguish between frozen soil and non-frozen soil states, and improve the accuracy of inversion. In order to verify and improve the performance of the model, the present invention uses the SMAP (Soil Moisture ActivePassive) L3 level frozen soil freeze-thaw data product as a reference and verification data set. SMAP data, with its global coverage and strictly verified frozen soil information, provides a reliable standard for model training and verification. The present invention includes the following specific steps: Quality control of spaceborne GNSS-R data Quality control adopts two methods: quality label and empirical criteria. First, the quality label data is converted into binary, and the data quality is determined according to the bit value. The label bits used mainly include black body load delay Doppler pattern mode, radio frequency interference, test mode, etc.; secondly, the data is screened again according to empirical criteria. The variables and thresholds used mainly include: 1) the incident angle is greater than 65°, 2) the signal-to-noise ratio is less than 0, and 3) the antenna gain is less than 0. Data that meets the above conditions will be screened out.

[0020] Extraction of observations from spaceborne GNSS-R data and calculation of surface reflectivity The latitude and longitude coordinates of the satellite-borne GNSS-R mirror point and other observations used to calculate the surface reflectance are extracted to calculate the surface reflectivity. Assuming that coherent scattering mainly occurs on land, the coherent component is expressed in the following form: (1) In the formula, is the coherent scattering component power; the subscript r represents the right-hand circularly polarized antenna of the GNSS transmitter, and l represents the left-hand circularly polarized antenna of the GNSS-R receiver; is the signal wavelength; is the peak power of the transmitted signal; , are the gains of the transmitting antenna and the receiving antenna respectively; , are the distances between the transmitter, receiver and the mirror reflection point respectively; Represents the surface reflectivity.

[0021] The power expression of the incoherent scattered component is: (2) In the formula, is the incoherent scattering component power; Bistatic Radar Cross Section (BRCS), in m 2 .

[0022] make = , combine formula (1) and formula (2) to get the surface reflectivity writing function, as shown in formula (3).

[0023] (3) In the formula, represents the surface reflectivity; , are the distances between the transmitter, receiver and the mirror reflection point respectively; Bistatic Radar Cross Section (BRCS), in m 2 .

[0024] In the spaceborne GNSS-R technology, the core significance of calculating the surface reflectivity by combining coherent scattering components and incoherent scattering components is to more comprehensively capture the complex physical characteristics of surface scattering, thereby improving the accuracy and applicability of remote sensing data. Surface reflection signals usually contain both coherent scattering from smooth areas (such as specular reflection, which has phase consistency and high correlation) and incoherent scattering from rough areas (such as diffuse reflection, which exhibits random diffusion). Simply relying on one of the components will cause the model to fail or deviate in a dynamic environment. In land scenes, the reflection of moist smooth soil is mainly coherent, while vegetation cover or dry surface is dominated by incoherence. By combining the two components, the contribution of different surface conditions can be adaptively balanced to avoid systematic errors caused by a single assumption. In addition, this joint processing can suppress noise interference and enhance data robustness through signal statistical separation or power allocation methods. Therefore, the simultaneous coherent and incoherent scattering is not only conducive to the requirements of physical mechanisms, but also a key technical approach to achieve accurate detection of GNSS-R in highly dynamic and heterogeneous surface environments, providing a more reliable solution for remote sensing applications in complex scenes such as oceans, land and polar regions.

[0025] Gridding of spaceborne GNSS-R surface reflectivity data GNSS-R data are pseudo-random sampling points. In order to spatially match them with SMAP to construct a dataset, the GNSS-R surface reflectivity is gridded using the EASE-Grid of the SMAP data. The surface reflectivity of the sampling points falling within each grid is averaged to represent the surface reflectivity of the grid.

[0026] SMAP data preprocessing Based on the SMAP L3 soil moisture product, soil moisture estimation, quality labels, vegetation optical thickness, land cover type, and grid-related coordinates of ascending and descending regions were extracted as auxiliary input features of the classification model; based on the SMAL L3 frozen soil freeze-thaw product, frozen soil freeze-thaw data were extracted and processed as output features of the model and reference data for accuracy evaluation. The ascending and descending orbits were merged to represent the results of one day. In order to obtain global coverage data, the data of three consecutive days were averaged ignoring null values. In addition, the quality labels were used to filter the SMAP data.

[0027] Inversion model construction All daily observation data were randomly divided into two groups, forming a training set and a test set, accounting for 70% and 30% of the total data, respectively. The 10-fold cross-validation method was used to tune the hyperparameters of the random forest classifier.

[0028] Spaceborne GNSS-R frozen soil freeze-thaw inversion and accuracy assessment The input features are substituted into the model to obtain the spaceborne GNSS-R soil moisture inversion results, and the quality of the results is evaluated from three perspectives: model accuracy, temporal variability, and spatial variability. Figure 2 This is the distribution map of frozen soil freeze-thaw results from spaceborne GNSS-R, which has good consistency with the spatial distribution of reference data; Figure 3 This is a time series diagram of the accuracy of frozen soil freeze-thaw classification using satellite-borne GNSS-R. It can be seen from the figure that the daily classification accuracy is higher than 0.95, and the average accuracy can reach 0.99, which verifies the feasibility and effectiveness of satellite-borne GNSS-R inversion of frozen soil freeze-thaw.

[0029] Advantages of the present invention: The present invention aims to solve the problem of high-precision real-time monitoring of frozen soil freeze-thaw status inversion using satellite-borne GNSS-R technology, which is directly related to the scientificity and timeliness of decision-making in the fields of cold region engineering, frozen soil ecology, climate change research, and disaster risk assessment. The present invention introduces a GNSS-R frozen soil freeze-thaw inversion algorithm based on a random forest model, which significantly improves the temporal and spatial coverage, inversion accuracy, and reliability of frozen soil freeze-thaw status data, provides data support for key areas such as climate change research, cold region engineering planning, ecological environment management, and hydrogeological modeling, and further enhances the scientificity of decision-making and practical effectiveness.

[0030] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A GNSS-R frozen soil freeze-thaw classification method based on a machine learning model, characterized by: The following steps are involved: S1. Perform quality control on onboard GNSS-R data; S2, extraction of satellite-borne GNSS-R data observations and calculation of surface reflectivity; S3, gridding the satellite-borne GNSS-R surface reflectivity data; S4, SMAP data preprocessing; S5, inversion model construction; S6. Spaceborne GNSS-R frozen soil freeze-thaw inversion and accuracy assessment; In step S2, the onboard GNSS-R data observations are extracted and the surface reflectivity is calculated, including: S21, extract the latitude and longitude coordinates of the satellite-borne GNSS-R mirror point and other observation values ​​used to calculate the surface reflectivity; S22. Calculate the surface reflectivity by combining the power expression of the coherent component and the power expression of the incoherent scattered component.

2. According to claim 1, a GNSS-R frozen soil freeze-thaw classification method based on a machine learning model is characterized in that: In step S1, the quality control of the spaceborne GNSS-R data is performed, including: S11, converting the quality label data into binary, and determining the data quality according to the bit value, wherein the label bits used include black body load delay Doppler pattern mode, radio frequency interference and test mode; S12. Perform secondary screening on the data according to empirical criteria, and eliminate data with an incident angle greater than 65°, a signal-to-noise ratio less than 0, or an antenna gain less than 0.

3. The GNSS-R frozen soil freeze-thaw classification method based on a machine learning model according to claim 1, characterized in that: In step S22, the power expression of the coherent component is: (1) In the formula, is the coherent scattering component power; the subscript r represents the right-hand circularly polarized antenna of the GNSS transmitter, and l represents the left-hand circularly polarized antenna of the GNSS-R receiver; is the signal wavelength; is the peak power of the transmitted signal; , are the gains of the transmitting antenna and the receiving antenna respectively; , are the distances between the transmitter, receiver and the mirror reflection point respectively; represents the surface reflectivity; The power expression of the incoherent scattered component is: (2) In the formula, is the incoherent scattering component power; is the bistatic radar scattering coefficient, in m 2 ; make = , combine formula (1) and formula (2) to get the surface reflectivity writing Function, such as formula (3): (3) In the formula, represents the surface reflectivity; , are the distances between the transmitter, receiver and the mirror reflection point respectively; is the bistatic radar scattering coefficient, in m 2 .

4. The GNSS-R frozen soil freeze-thaw classification method based on a machine learning model according to claim 1, characterized in that: In step S3, the spaceborne GNSS-R surface reflectivity data is gridded, including: The EASE-Grid of SMAP data is used to grid the surface reflectivity of the space-borne GNSS-R, and the surface reflectivity of the sampling points falling within each grid is averaged.

5. The GNSS-R frozen soil freeze-thaw classification method based on a machine learning model according to claim 1, characterized in that: In step S4, SMAP data preprocessing includes: S41. Extract soil moisture estimation, quality label, vegetation optical depth, land cover type, and grid coordinates of ascending and descending regions based on SMAP L3 soil moisture products as auxiliary input features for the classification model. S42. Extract and process frozen soil freeze-thaw data based on SMAP L3 frozen soil freeze-thaw products as the output characteristics of the model and reference data for accuracy evaluation; S43, merging the ascending track and the descending track to represent the result of one day, and ignoring the null value of the data of three consecutive days to average them; S44. Filter SMAP data using quality tags.

6. The GNSS-R frozen soil freeze-thaw classification method based on a machine learning model according to claim 1, characterized in that: In step S5, the inversion model is constructed, including: S51, spaceborne GNSS-R pseudo-random sampling points are temporally and spatially matched with SMAP; S52. Construct a data set and randomly divide all daily observation data into two groups, one group is a training set accounting for 70% of the total data, and the other group is a test set accounting for 30% of the total data; S53, model training, the hyperparameters of the random forest classifier were tuned using the 10-fold cross-validation method.

7. The GNSS-R frozen soil freeze-thaw classification method based on a machine learning model according to claim 1, characterized in that: In step S6, the spaceborne GNSS-R frozen soil freeze-thaw inversion and accuracy assessment includes: The input features are substituted into the inversion model to obtain the spaceborne GNSS-R soil moisture inversion results, and the quality of the results is evaluated from three perspectives: model accuracy, temporal variability, and spatial variability.

Images