CN-122020152-A - Method and device for establishing backscattering coefficient correction model

CN122020152ACN 122020152 ACN122020152 ACN 122020152ACN-122020152-A

Abstract

The application provides a method and a device for establishing a backscattering coefficient correction model, wherein the method comprises the steps of obtaining multisource data comprising observation values corresponding to backscattering coefficients, carrying out space-time matching on the multisource data, generating data units corresponding to satellite observation pixels, filtering the data units through preset rules to obtain a reserved data set, outputting simulation values through a physical model based on the reserved data set which does not comprise the observation values, constructing a training set based on the simulation values and the reserved data set, using the observation values as target variables of the training set, training a machine learning model by using the training set, and carrying out grid search optimization on preset super parameters in the machine learning model to obtain the final backscattering coefficient correction model. The application adopts a random forest algorithm to effectively learn and fit the complex and high-order nonlinear relation between the simulation value of the physical model and the satellite observation value.

Inventors

Wen Jinghan
Weng Fuzhong
XU JING
JIA TONG
HE LINGLI
CHEN LIJIA
HE ZHIWEI
ZHANG LEI
LI WENYU

Assignees

青岛海洋气象研究院

Dates

Publication Date: 20260512
Application Date: 20251215

Claims (10)

1. A method for establishing a backscatter coefficient correction model, comprising: Acquiring multi-source data comprising observed values corresponding to the backscattering coefficients; Performing space-time matching on the multi-source data to generate data units corresponding to all satellite observation pixels, and filtering the data units through preset rules to obtain a reserved data set; outputting a simulation value through a physical model based on a reserved data set which does not comprise an observation value, constructing a training set based on the simulation value and the reserved data set, and using the observation value as a target variable of the training set; And training the machine learning model by using the training set, and performing grid search optimization on preset super parameters in the machine learning model to obtain a final backscattering coefficient correction model.
2. The method according to claim 1, wherein the method further comprises: and outputting and storing a correction result by using the backscattering coefficient correction model.
3. The method according to claim 1, wherein the acquiring multi-source data including observations corresponding to backscatter coefficients specifically comprises: analyzing the normalized radar backscattering section value as an observation value, and analyzing an incident angle, an azimuth angle, longitude and latitude and a timestamp to obtain satellite-borne scatterometer parameters; analyzing sea surface wind speed, sea surface wind direction, sea surface temperature and sea surface salinity to obtain driving parameters; And taking the satellite-borne scatterometer parameters and the driving parameters together as multi-source data.
4. The method of claim 1, wherein the performing space-time matching on the multi-source data to generate the data unit corresponding to each satellite observation pixel specifically includes: For the current satellite observation pixel Calculating the current position and time corresponding parameter value by bilinear interpolation based on longitude, latitude and time stamp, and calculating the relative wind direction by using formula 1 : Equation 1; Wherein, the Represents the wind direction of the sea surface, Representing satellite observation azimuth angles; Generating data units corresponding to all satellite observation pixels , wherein, The value of the observation is represented by a value, Indicating the angle of incidence, Represents the sea surface wind speed, The sea surface temperature is indicated to be the sea surface temperature, Represents sea surface salinity.
5. The method according to claim 1, wherein filtering the data units by a preset rule to obtain a reserved data set specifically comprises: traversing each satellite observation pixel, and filtering and removing the data unit if any condition in a preset rule is met, wherein the preset rule comprises that sea ice density is Yu Haibing, the rainfall rate is higher than the rainfall rate threshold, and the sea surface wind speed is lower than the sea surface wind speed threshold.
6. The method of claim 4, wherein the outputting of the simulated values by the physical model based on the retained dataset excluding the observed values and the constructing of the training set based on the simulated values and the retained dataset specifically comprises: The reserved data set which does not comprise the observed value is passed through a pre-deployed pBRDF model, and is simulated by the pBRDF model according to the internal double-scale rough surface scattering model, so as to obtain a simulation value corresponding to the backscattering coefficient based on the physical model ; Constructing training set samples as , wherein, , 。
7. The method of claim 1, wherein training the machine learning model using the training set, and performing grid search optimization on preset hyper-parameters in the machine learning model, to obtain a final backscatter coefficient correction model specifically comprises: Training a random forest model by using the training set as a machine learning model; Grid search optimization is carried out on the minimum leaf node number serving as a preset super parameter, sampling samples and out-of-bag samples are divided in the training set, a random forest is built through the sampling samples in the grid search optimization process, and a prediction error is calculated through the out-of-bag samples by using a formula 2 : Equation 2; Wherein, the The number of samples outside the bag is indicated, The target variable is represented by a value of the target variable, Representing a predictive value of the sample outside the bag; And training after determining the optimal value of the preset super parameter to obtain a backscattering coefficient correction model.
8. A device for establishing a backscatter coefficient correction model, comprising: the multi-source loading module is used for acquiring multi-source data comprising observation values corresponding to the backscattering coefficients; The space-time matching module is used for performing space-time matching on the multi-source data to generate data units corresponding to each satellite observation pixel, and filtering the data units through preset rules to obtain a reserved data set; the physical simulation module is used for outputting a simulation value through a physical model based on a reserved data set which does not comprise an observation value, constructing a training set based on the simulation value and the reserved data set, and using the observation value as a target variable of the training set; and the model training module is used for training the machine learning model by using the training set, and carrying out grid search optimization on preset super parameters in the machine learning model to obtain a final backscattering coefficient correction model.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the computer program when executed by the processor implementing the steps of the method of creating a backscatter coefficient correction model according to any one of claims 1 to 7.
10. A computer-readable storage medium, wherein a program for realizing information transfer is stored on the computer-readable storage medium, and the program when executed by a processor realizes the steps of the method for creating the backscatter coefficient correction model according to any one of claims 1 to 7.

Description

Method and device for establishing backscattering coefficient correction model Technical Field The invention relates to the technical field of parameter correction, in particular to a method and a device for establishing a backscattering coefficient correction model. Background The satellite-borne microwave scatterometer is a key remote sensing sensor for acquiring parameters such as a global sea surface wind field. In the related art, a polarized bidirectional reflection distribution function (pBRDF) model is used as a physical model and can be used for simulating a microwave scattering process of the sea surface and simulating NRCS observed by an active sensor, however, a significant deviation exists between a simulation result of the model and an actual observation value of a satellite. The traditional deviation correction method is mostly dependent on empirical linear regression or a lookup table, is difficult to effectively process complex and nonlinear deviations formed by combined actions of model defects, instrument calibration and atmospheric transmission, has limited correction precision and generalization capability, has the publication number of CN119556247A, builds a calibration model dataset with evenly distributed azimuth angles and uses a least square method to fit out nonlinear deviations of a system for correction, and has natural non-uniform characteristics, so that the assumption of evenly distributed azimuth angles is easy to introduce systematic errors, and besides, the least square method is a linear or quasi-linear fitting tool essentially, and has insufficient capture and correction capability for complex and high-order nonlinear deviations generated by coupling actions of various factors such as physical model theory simplification, instrument calibration inaccuracy, electromagnetic wave atmospheric transmission attenuation, low wind speed model failure. By combining the analysis of the development status in the technical field, the prior art lacks a scheme of systematically correcting deviation of elements in the polarized bidirectional reflectance distribution function pBRDF based on a random forest machine learning algorithm and combining a physical model with a data driving method. Disclosure of Invention The invention aims to provide a method and a device for establishing a backscattering coefficient correction model, and aims to solve the problems in the prior art. According to a first aspect of an embodiment of the present invention, there is provided a method for establishing a backscatter coefficient correction model, including: Acquiring multi-source data comprising observed values corresponding to the backscattering coefficients; Performing space-time matching on the multi-source data to generate data units corresponding to all satellite observation pixels, and filtering the data units through preset rules to obtain a reserved data set; outputting a simulation value through a physical model based on a reserved data set which does not comprise an observation value, constructing a training set based on the simulation value and the reserved data set, and using the observation value as a target variable of the training set; and training a machine learning model by using a training set, and performing grid search optimization on preset super parameters in the machine learning model to obtain a final backscattering coefficient correction model. According to a second aspect of the embodiment of the present invention, there is provided an apparatus for establishing a backscatter coefficient correction model, including: the multi-source loading module is used for acquiring multi-source data comprising observation values corresponding to the backscattering coefficients; The space-time matching module is used for performing space-time matching on the multi-source data, generating data units corresponding to the satellite observation pixels, and filtering the data units through a preset rule to obtain a reserved data set; the physical simulation module is used for outputting simulation values through a physical model based on a reserved data set which does not comprise the observation values, constructing a training set based on the simulation values and the reserved data set, and using the observation values as target variables of the training set; The model training module is used for training the machine learning model by using the training set, and carrying out grid search optimization on preset super parameters in the machine learning model to obtain a final backscattering coefficient correction model. According to a third aspect of embodiments of the present invention there is provided an electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the computer program when executed by the processor implementing the steps of the method of building a backscatter coefficient correction model as provided in the fi