Search

CN-121980821-A - Sea ice thickness partition inversion method of three-layer dynamic roughness based on wind cloud satellite

CN121980821ACN 121980821 ACN121980821 ACN 121980821ACN-121980821-A

Abstract

The invention discloses a sea ice thickness partition inversion method based on three-layer dynamic roughness of a wind cloud satellite, which comprises the steps of firstly obtaining GNSS-R remote sensing observation data, sea ice reference product data and sea environment parameter data, and obtaining a matching sample set through correlation matching; the method comprises the steps of obtaining electromagnetic physical parameters of sea ice and sea water and parameters related to propagation attenuation in the ice through calculation, constructing a dynamic roughness reflection model of an air-sea ice-sea water three-layer medium structure, dividing specular reflection points in a matched sample set into a normal area sea ice sample and a rough area sea ice sample, carrying out physical inversion on the normal area sea ice sample by adopting the dynamic roughness reflection model to obtain sea ice thickness of the normal area, constructing a second-stage machine learning model for carrying out regression inversion on the rough area sea ice sample to obtain sea ice thickness of the rough area, and finally carrying out fusion processing on sea ice thickness inversion results of the normal area and the rough area and outputting the sea ice thickness inversion results.

Inventors

  • Xie Tingyao
  • SU MINGKUN
  • LI ZHAO
  • HE XIAOXING
  • HU SHUNQIANG
  • SHANG JUNNA
  • WENG QI
  • ZHOU YUWEI
  • ZHANG YUYANG

Assignees

  • 杭州电子科技大学

Dates

Publication Date
20260505
Application Date
20260403

Claims (14)

  1. 1. A sea ice thickness partition inversion method based on three-layer dynamic roughness of a wind cloud satellite is characterized by comprising the following steps: Step 1, acquiring GNSS-R remote sensing observation data, sea ice reference product data and sea environment parameter data, and performing association matching on specular reflection points of the GNSS-R remote sensing observation data, the sea ice reference product data and the sea environment parameter data by adopting a space-time nearest neighbor matching method to obtain a matching sample set; Step 2, calculating the reflectivity of a specular reflection point based on GNSS-R remote sensing observation data, and calculating electromagnetic physical parameters of sea ice and sea water and propagation attenuation related parameters in ice by combining sea environment parameter data and sea ice reference product data; Step 3, constructing a dynamic roughness reflection model of the air-sea ice-sea water three-layer medium structure, introducing a dynamic roughness factor into the model as a modulation parameter, wherein the dynamic roughness factor changes along with space, time and environmental conditions and is used for describing the influence of sea ice surface scattering difference on GNSS-R reflection signals; Step 4, mapping the dynamic roughness factors in a logarithmic domain, constructing a first-stage machine learning model, carrying out regression prediction on the dynamic roughness factors in the logarithmic domain, reversely transforming a prediction result into a linear domain, and carrying out statistic deviation correction to obtain a correction estimated value of the dynamic roughness factors; Step 5, taking the product of the correction estimated value of the dynamic roughness factor and the modular square of the effective circular polarization reflection coefficient of the air-sea ice interface as a discrimination threshold, and dividing the specular reflection points in the matched sample set into a normal sea ice sample and a rough sea ice sample; step 6, carrying out physical inversion on the sea ice sample in the normal area by adopting the dynamic roughness reflection model of the three-layer medium structure to obtain the sea ice thickness in the normal area; and 7, carrying out fusion treatment on sea ice thickness inversion results of the normal area and the rough area to form continuous and consistent sea ice thickness inversion results and outputting the continuous and consistent sea ice thickness inversion results.
  2. 2. The sea ice thickness partition inversion method based on three-layer dynamic roughness of a wind cloud satellite according to claim 1, wherein the GNSS-R remote sensing observation data in the step 1 is GNSS-R first-order product observation data loaded by a FY-3E satellite GNOS-II, at least comprises longitude and latitude of a specular reflection point, three-dimensional coordinates of the specular reflection point, three-dimensional coordinates of a transmitter, three-dimensional coordinates of a receiver, DDM power peak, noise power, BRCS factors, an incidence angle of the specular reflection point and L-band signal frequency, the sea ice reference product data is SMOS sea ice product data, and at least comprises sea ice thickness, sea ice surface temperature and sea ice salinity, and the sea environment parameter data is sea water temperature and sea water salinity data provided by CMEMS.
  3. 3. The sea ice thickness partition inversion method based on three-layer dynamic roughness of wind cloud satellite according to claim 1, wherein the reflectivity in the step 2 is calculated based on a GNSS-R bistatic radar equation, and the calculation process is as follows: Calculating the net reflected power, wherein the net reflected power is 10 multiplied by 10, and the logarithm of the difference value between the DDM peak power count value and the noise power count value is taken as the base, and then subtracting the power conversion factor; Calculating a geometric diffusion factor and a decibel geometric diffusion factor, wherein the geometric diffusion factor is the product of dividing the square of the sum of the three-dimensional space distance from a transmitter to a specular reflection point and the three-dimensional space distance from a receiver to the specular reflection point by 4 times of the circumference ratio, the square of the three-dimensional space distance from the transmitter to the specular reflection point and the square of the three-dimensional space distance from the receiver to the specular reflection point, and the decibel geometric diffusion factor is the logarithm of 10 times the base geometric diffusion factor; calculating the reflectivity of the dB dimension, namely adding the geometric diffusion factor of the dB dimension to the net reflected power, and subtracting the BRCS factor; the reflectivity of the linear domain is calculated as 10 decibel dimensional reflectivity divided by the power of 10.
  4. 4. The sea ice thickness partition inversion method based on three-layer dynamic roughness of a wind cloud satellite according to claim 1, wherein the electromagnetic physical parameters in the step 2 at least comprise a sea water complex permittivity, a sea ice complex permittivity, an air-sea ice interface effective circular polarization reflection coefficient and a sea ice-sea water interface effective circular polarization reflection coefficient, wherein the sea water complex permittivity is calculated by using a Klein-Swift sea water dielectric model, and the sea ice complex permittivity is calculated by using a Vant sea ice dielectric model.
  5. 5. The sea ice thickness partition inversion method based on three-layer dynamic roughness of wind cloud satellites according to claim 4, wherein the effective circular polarization reflection coefficient is calculated by the following steps: The vertical polarization Fresnel reflection coefficient is the square root of the difference value of the relative dielectric constant and the sine value of the incident angle, which is multiplied by the cosine value of the incident angle, subtracted by the square root of the difference value of the relative dielectric constant and the sine value of the incident angle, and then divided by the cosine value of the relative dielectric constant and the sine value of the incident angle, and added by the square root of the difference value of the relative dielectric constant and the sine value of the incident angle, and the horizontal polarization Fresnel reflection coefficient is the square root of the difference value of the cosine value of the incident angle, which is subtracted by the cosine value of the incident angle, and added by the square root of the difference value of the relative dielectric constant and the sine value of the incident angle; Calculating the effective circular polarization reflection coefficient of the interface, wherein the effective circular polarization reflection coefficient is the difference between the vertical polarization Fresnel reflection coefficient and the horizontal polarization Fresnel reflection coefficient multiplied by one half; calculating the effective circular polarization reflection coefficient of the air-sea ice interface, namely taking the relative dielectric constant of the air-sea ice interface and the incident angle of the air-sea ice interface at the specular reflection point as inputs, and obtaining the effective circular polarization reflection coefficient according to a calculation method of the effective circular polarization reflection coefficient of the interface; Calculating the effective circular polarization reflection coefficient of the sea ice-sea water interface, namely taking the relative dielectric constant of the sea ice-sea water interface and the corresponding incident angle of the sea ice-sea water interface as inputs, and obtaining the sea ice-sea water interface according to the effective circular polarization reflection coefficient calculation method of the interface.
  6. 6. The sea ice thickness zoning inversion method based on three-layer dynamic roughness of wind cloud satellite of claim 5, wherein the relative dielectric constant of the air-sea ice interface is the ratio of sea ice complex dielectric constant to air dielectric constant, and the relative dielectric constant of the sea ice-sea water interface is the ratio of sea water complex dielectric constant to sea ice complex dielectric constant; The incidence angle of the sea ice-sea water interface is an intra-ice incidence angle, and is obtained by dividing the sine value of the incidence angle of the air-sea ice interface at the specular reflection point by the square root of the sea ice complex permittivity.
  7. 7. The sea ice thickness partition inversion method based on three-layer dynamic roughness of a wind cloud satellite according to claim 6, wherein in the step 3, in a dynamic roughness reflection model of the three-layer medium structure, the power reflectivity of a specular reflection point is a product of a dynamic roughness factor and a sum of a first power and a second integrated power, wherein the first power is the power reflectivity of an air-sea ice interface and is a modular square of an effective circular polarization reflection coefficient of the air-sea ice interface, and the second integrated power is a product of subtracting a square of a difference of the power reflectivity of the air-sea ice interface, the power reflectivity of the sea ice-sea water interface and an in-ice propagation attenuation parameter from 1 and is divided by a product of subtracting the power reflectivity of the air-sea ice interface, the power reflectivity of the sea ice-sea water interface and the in-ice propagation attenuation parameter from 1.
  8. 8. The sea ice thickness zoning inversion method based on three-layer dynamic roughness of wind cloud satellite of claim 7, wherein the power reflectivity of sea ice-sea water interface is the modular square of the sea ice-sea water interface effective circular polarization reflection coefficient, and the propagation attenuation parameter in ice is the product power of negative 4 times of the attenuation coefficient in ice of natural exponential function and sea ice thickness.
  9. 9. The sea ice thickness partition inversion method based on three-layer dynamic roughness of wind cloud satellites according to claim 1, wherein in the step 4, the first stage machine learning model is XGBoost models, input features of the XGBoost models in the first stage at least comprise reflectivity, specular reflection point incidence angle, effective circular polarization reflection coefficient of an air-sea ice interface and effective circular polarization reflection coefficient of a sea ice-sea water interface, training of a XGBoost model aims at minimizing a logarithmic domain mean square error, training is carried out in a monthly grouping mode, 70% of samples in each group are used for training, 30% of samples are used for testing, and training parameters of the XGBoost models in the first stage are the maximum iteration round number of 500, the learning rate of 0.05 and the maximum depth of a single tree of 6.
  10. 10. The sea ice thickness zoning inversion method of three-layer dynamic roughness based on wind cloud satellite as claimed in claim 9, wherein the process of statistical deviation correction is: substituting the sea ice thickness in sea ice reference product data into a dynamic roughness reflection model of a three-layer medium structure by using the sea ice thickness as reference thickness, and reversely deducing a target value of a dynamic roughness factor; Calculating residual errors of the predicted value and the target value of the dynamic roughness factor in the logarithmic domain, wherein the variance of the residual errors is the sum of squares of residual errors of all samples of the training set divided by the total number of samples of the training set; And carrying out natural exponential transformation on the dynamic roughness factor predicted value of the number domain, and multiplying the dynamic roughness factor predicted value by the power of half residual variance of a natural exponential function to obtain a correction estimated value of the dynamic roughness factor.
  11. 11. The sea ice thickness partition inversion method based on three-layer dynamic roughness of a wind cloud satellite according to claim 1, wherein in the step 5, the specific rule of sea ice sample partition discrimination is that when the reflectivity of a specular reflection point is lower than the discrimination threshold, a sea ice sample in a rough area is determined, and when the reflectivity of the specular reflection point is higher than the discrimination threshold, a sea ice sample in a normal area is determined.
  12. 12. The sea ice thickness partition inversion method based on three-layer dynamic roughness of a wind cloud satellite according to claim 1 is characterized in that in the step 6, sea ice thickness is a quotient of negative 1 times and 4 times of internal attenuation coefficient of ice, and then a natural logarithm operation result is multiplied, the true number of the natural logarithm is a quotient of a first difference term and a comprehensive sum term, wherein the first difference term is a result of difference operation between the quotient of the reflectivity and a correction estimated value of a dynamic roughness factor and the power reflectivity of an air-sea ice interface, the comprehensive sum term is a sum of a first product term and a second product term, the first product term is a product of the square of the difference of the power reflectivity of the 1 and the power reflectivity of the air-sea ice interface and the power reflectivity of the sea ice-sea ice interface, and the second product term is a product of the power reflectivity of the air-sea ice interface, the power reflectivity of the sea ice-sea ice interface and the first difference term.
  13. 13. The method according to claim 1, wherein in the step 6, the second-stage machine learning model is XGBoost models, input features of the XGBoost models in the second stage at least comprise correction estimation values of dynamic roughness factors, reflectivity, specular reflection point incidence angle, effective circular polarization reflection coefficient of an air-sea ice interface and effective circular polarization reflection coefficient of a sea ice-sea water interface, and training of the XGBoost models in the second stage aims at minimizing mean square error between sea ice reference thickness and predicted sea ice thickness.
  14. 14. The sea ice thickness partition inversion method based on three layers of dynamic roughness of wind and cloud satellites according to claim 13, wherein training parameters of XGBoost models in the second stage are 300 maximum iteration rounds, 0.05 learning rate and 5 maximum depths of single trees.

Description

Sea ice thickness partition inversion method of three-layer dynamic roughness based on wind cloud satellite Technical Field The invention relates to the technical field of navigation satellite remote sensing inversion, in particular to a sea ice thickness partition inversion method of three-layer dynamic roughness based on a wind-cloud satellite. Background Sea ice thickness is a key thermodynamic parameter in a polar climate system, and has important significance for global climate change research, ocean energy balance and polar shipping. GNSS-R (Global Navigation SATELLITE SYSTEM Reflectometry, global navigation satellite reflectometry) is an emerging technology for bistatic radar remote sensing, using navigation satellite signals, such as GPS (Global Positioning System ), BDS (BeiDou Navigation SATELLITE SYSTEM, the Beidou navigation satellite system) realizes inversion of geophysical parameters by the reflection echo of the sea surface, and has the advantages of high space-time resolution, low cost and all-weather observation. The existing GNSS-R sea ice thickness inversion method mainly comprises two methods, wherein the first method adopts a sea ice-sea water two-layer reflection model for inversion, and the method is characterized in that the reflection process is attributed to propagation attenuation of an ice layer and sea ice-sea water interface reflection by neglecting the scattering effect of an air-sea ice interface, so that the method has certain usability under the condition of thin sea ice with smooth surface. However, this model is limited in that the sea ice thickness information is rapidly saturated in echoes when the sea ice becomes thick and the attenuation in ice is significantly enhanced, and that the air-sea ice interface and its complex scattering are not negligible once the sea ice surface is rough or there is snow coverage, resulting in a sea ice-sea water two-layer model that cannot accurately measure thick sea ice. The second method is to invert by adopting an air-sea ice-sea water three-layer reflection model, and the method establishes multiple coherent reflections and interference of an air-sea ice interface and a sea ice-sea water interface, is suitable for thickness inversion of thick sea ice, but Bao Haibing inversion effect is inferior to that of a sea ice-sea water two-layer reflection model. In the second method, the thickness range of the invertible sea ice is expanded theoretically, but only part of samples meet the stable solving requirement of a model in actual observation due to the dependence on the coherent reflection condition with higher quality, so that the space coverage range of the remotely-sensed inverted sea ice thickness is limited. For example, when the method is used for inverting sea ice with complex situations such as rough surface or snow coverage, the problem that coherent components attenuate or even disappear is difficult to avoid, and unstable inversion results are generated. Therefore, a method for dynamically estimating the sea ice surface roughness with low requirements on reflection conditions is needed to break through the bottleneck of the existing sea ice thickness inversion method in the application range, inversion range and inversion precision. Disclosure of Invention The invention aims to provide a sea ice thickness zoning inversion method of three-layer dynamic roughness based on a wind cloud satellite, and aims to solve the problems that inversion saturation is easy to occur under the condition of thick sea ice and the stability of a model is insufficient under the condition of rough surface or snow coverage in the existing GNSS-R sea ice thickness inversion method. In order to achieve the above purpose, the technical scheme adopted by the invention is as follows: a sea ice thickness partition inversion method of three-layer dynamic roughness based on a wind cloud satellite comprises the following steps: Step 1, acquiring GNSS-R remote sensing observation data, sea ice reference product data and sea environment parameter data, and performing association matching on specular reflection points of the GNSS-R remote sensing observation data, the sea ice reference product data and the sea environment parameter data by adopting a space-time nearest neighbor matching method to obtain a matching sample set; Step 2, calculating the reflectivity of a specular reflection point based on GNSS-R remote sensing observation data, and calculating electromagnetic physical parameters of sea ice and sea water and propagation attenuation related parameters in ice by combining sea environment parameter data and sea ice reference product data; Step 3, constructing a dynamic roughness reflection model of the air-sea ice-sea water three-layer medium structure, introducing a dynamic roughness factor into the model as a modulation parameter, wherein the dynamic roughness factor changes along with space, time and environmental conditions and is used for descri