CN117272843B - GNSS-R sea surface wind speed inversion method and system based on random forest - Google Patents
GNSS-R sea surface wind speed inversion method and system based on random forest Download PDFInfo
- Publication number
- CN117272843B CN117272843B CN202311557686.XA CN202311557686A CN117272843B CN 117272843 B CN117272843 B CN 117272843B CN 202311557686 A CN202311557686 A CN 202311557686A CN 117272843 B CN117272843 B CN 117272843B
- Authority
- CN
- China
- Prior art keywords
- wind speed
- ddm
- nbrcs
- les
- sea surface
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 64
- 238000007637 random forest analysis Methods 0.000 title claims abstract description 50
- 238000012549 training Methods 0.000 claims abstract description 28
- 238000007781 pre-processing Methods 0.000 claims abstract description 24
- 238000012937 correction Methods 0.000 claims description 24
- 230000000694 effects Effects 0.000 claims description 9
- 230000008569 process Effects 0.000 claims description 9
- 238000010276 construction Methods 0.000 claims description 7
- 238000011160 research Methods 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 41
- 238000003066 decision tree Methods 0.000 description 17
- 230000008859 change Effects 0.000 description 5
- 238000010801 machine learning Methods 0.000 description 5
- 238000000638 solvent extraction Methods 0.000 description 4
- 230000003746 surface roughness Effects 0.000 description 4
- 238000012952 Resampling Methods 0.000 description 3
- 238000013507 mapping Methods 0.000 description 3
- 238000005259 measurement Methods 0.000 description 3
- 238000004422 calculation algorithm Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000004927 fusion Effects 0.000 description 2
- 230000033001 locomotion Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000003908 quality control method Methods 0.000 description 2
- 230000011218 segmentation Effects 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- CHBOSHOWERDCMH-UHFFFAOYSA-N 1-chloro-2,2-bis(4-chlorophenyl)ethane Chemical class C=1C=C(Cl)C=CC=1C(CCl)C1=CC=C(Cl)C=C1 CHBOSHOWERDCMH-UHFFFAOYSA-N 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000002592 echocardiography Methods 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000000149 penetrating effect Effects 0.000 description 1
- 230000010287 polarization Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000002310 reflectometry Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 230000000630 rising effect Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S19/00—Satellite radio beacon positioning systems; Determining position, velocity or attitude using signals transmitted by such systems
- G01S19/01—Satellite radio beacon positioning systems transmitting time-stamped messages, e.g. GPS [Global Positioning System], GLONASS [Global Orbiting Navigation Satellite System] or GALILEO
- G01S19/13—Receivers
- G01S19/14—Receivers specially adapted for specific applications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
- G01S7/41—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/10—Pre-processing; Data cleansing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/01—Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2111/00—Details relating to CAD techniques
- G06F2111/08—Probabilistic or stochastic CAD
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2113/00—Details relating to the application field
- G06F2113/08—Fluids
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Medical Informatics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Computer Networks & Wireless Communication (AREA)
- Computational Linguistics (AREA)
- Geometry (AREA)
- Computer Hardware Design (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention discloses a GNSS-R sea surface wind speed inversion method and system based on random forests, which belong to the field of atmospheric science research, and the method comprises the following steps: obtaining L1 level data and reference wind speed of a CYGNSS, and respectively preprocessing the L1 level data and the reference wind speed to obtain a training sample set, wherein the L1 level data comprises: DDM front slope LES, normalized double-base radar cross section NBRCS and DDM signal-to-noise ratio SNR; constructing an exponential function model by using the training sample set; fitting the DDM front slope LES, the normalized bistatic radar scattering cross section NBRCS and the reference wind speed by using an exponential function model to obtain a wind speed inversion coarse result; and (3) inputting the wind speed inversion coarse result and the DDM signal-to-noise ratio SNR into a random forest together for equal weight combination to obtain a final wind speed inversion result, and finishing sea surface wind speed inversion. The invention realizes high-precision sea surface wind speed inversion.
Description
Technical Field
The invention belongs to the field of atmospheric science research, and particularly relates to a GNSS-R sea surface wind speed inversion method and system based on random forests.
Background
The global navigation satellite system (Global Navigation Satellite System, GNSS) mainly consists of the united states GPS, russian GLONASS, european Galileo (Galileo) and the chinese beidou satellite navigation system (BDS), providing global positioning, navigation and time service. The L-band signal used by the GNSS satellite has stronger penetrating power and is less influenced by cloud, fog, rain and snow, and can be used as a signal source for free earth observation. GNSS signals are subjected to the earth's surface during reflection, producing variations in the strength, time delay, frequency, etc. of the signals. Sea surface wind field is one of important factors influencing sea surface roughness, and the sea surface roughness is different under different wind speed conditions. The difference in roughness causes a change in reflectivity, which results in a change in reflected signal power, frequency, and ultimately captured by the receiver. The GNSS-R wind speed inversion is realized by using the relation of wind field- > roughness- > reflectivity- > signal characteristics.
Current GNSS-R wind speed inversion can be divided into three categories:
(1) Wind speed inversion method based on waveform matching
Waveform matching is one of the common methods of GNSS-R wind speed inversion. The method is realized by firstly obtaining a series of DDMs in a mode of field measurement or numerical simulation and recording corresponding wind speed values to construct a 'DDM-wind speed' reference library for matching. And then, performing difference comparison (matching) on the DDM of the unknown wind speed and each DDM in the reference library, and taking the wind speed corresponding to the reference DDM with the smallest difference as the inversion wind speed.
(2) Wind speed inversion method based on geographic model function
The wind speed inversion method based on the empirical function mainly performs wind speed inversion by establishing a relation between GNSS-R observables and wind speed mapping, and is one of the common methods for GNSS-R wind speed inversion. The empirical function used herein is a geographic model function (Geography Model Function, GMF) comprising the form of an exponential function, a polynomial function, a 2D look-up table (LUT), and the like.
(3) Wind speed inversion method based on machine learning (deep learning)
The machine learning model has a stronger nonlinear mapping capability than the empirical function model. The method takes DDM data and derivative features thereof as main input, and deep features are mined through models such as random forests, BP neural networks, convolutional neural networks and the like, so that a mapping relation between input and wind speeds is established.
In the existing GNSS-R wind speed inversion method, a DDM matching library needs to be built in advance in the waveform matching method, and because wind speed has randomness in actual measurement, the time cost, equipment cost and personnel cost required for building the matching library covering the full wind speed section are high; the DDM under each wind speed is calculated by adopting a simulation mode, a reflected signal model is required to be constructed, model errors are unavoidable, and the actual measurement effect is difficult to achieve, so that the construction difficulty of a matching library is high. The geographic model function method has a simple structure, less data volume is needed to construct, the model has strong interpretability, but the inversion accuracy is influenced by the used characteristics, and a reasonable characteristic set needs to be constructed. Meanwhile, the method is difficult to fully mine the relation between the features, and the difficulty of further improving the precision is high. Machine learning (deep learning) methods mainly rely on neural network models to mine deep links of features, but training requires a large amount of data and the models have poor interpretability.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a GNSS-R sea surface wind speed inversion method and a system based on random forests, which are used for comprehensively describing the sea surface roughness condition of a reflection area by using LES, NBRCS and SNR as input features; and fitting LES, NBRCS and wind speed by using an exponential function model to obtain a wind speed inversion coarse result, then inputting the two coarse results and SNR together into a random forest to perform equal weight combination, introducing the description of signal power through the SNR, reducing the interference of signal power change on wind speed inversion, obtaining a final wind speed inversion result, and realizing high-precision sea surface wind speed inversion.
In order to achieve the above object, the present invention provides the following solutions:
GNSS-R sea surface wind speed inversion method based on random forest comprises the following steps:
acquiring L1 level data and a reference wind speed of a CYGNSS, and respectively preprocessing the L1 level data and the reference wind speed to obtain a training sample set, wherein the L1 level data comprises: DDM front slope LES, normalized double-base radar cross section NBRCS and DDM signal-to-noise ratio SNR;
constructing an exponential function model by utilizing the training sample set;
fitting the DDM front slope LES, the normalized bistatic radar cross section NBRCS and the reference wind speed by using the exponential function model to obtain a wind speed inversion coarse result;
and inputting the wind speed inversion coarse result and the DDM signal-to-noise ratio SNR into a random forest together for equal weight combination to obtain a final wind speed inversion result, and finishing sea surface wind speed inversion.
Preferably, the method for preprocessing the L1 level data includes:
and carrying out angle correction on two characteristics of a normalized bistatic radar cross section NBRCS and a DDM front slope LES in an L1 level data product of the CYGNSS to obtain L2 level correction data.
Preferably, the method for preprocessing the reference wind speed comprises the following steps:
extracting longitude and latitude and observation time of an observation point from CYGNSS data;
and matching the corresponding observation point reference wind speed from wind speed assimilation data at a preset height above the sea surface issued by the ECMWF according to the principle that the longitude and latitude of the observation point are nearest to the observation time.
Preferably, the exponential function model includes: a LES index model and an NBRCS index model;
wherein the LES exponential function model is in the form of:；
the NBRCS exponential function model is in the form of:
；
wherein,for wind speed at 10m above sea surface, < >>And->To remove the LES and NBRCS values after the incidence angle effect, +.>Is a natural constant, and has a value of 2.71828%>The undetermined coefficients are determined by a nonlinear least squares method.
Preferably, the method for inputting the wind speed inversion coarse result and the DDM signal-to-noise ratio SNR into a random forest together for equal weight combination to obtain a final wind speed inversion result comprises the following steps:
。
the invention also provides a GNSS-R sea surface wind speed inversion system based on the random forest, which comprises: the device comprises a preprocessing module, a construction module, a fitting module and a combination module;
the preprocessing module is used for acquiring L1 level data and reference wind speed of a CYGNSS, and respectively preprocessing the L1 level data and the reference wind speed to obtain a training sample set, wherein the L1 level data comprises: DDM front slope LES, normalized double-base radar cross section NBRCS and DDM signal-to-noise ratio SNR;
the construction module is used for constructing an exponential function model by utilizing the training sample set;
the fitting module is used for respectively fitting the DDM front slope LES, the normalized bistatic radar scattering cross section NBRCS and the reference wind speed by using the exponential function model to obtain a wind speed inversion coarse result;
the combination module is used for inputting the wind speed inversion coarse result and the DDM signal-to-noise ratio SNR into a random forest together for equal weight combination to obtain a final wind speed inversion result, and finishing sea surface wind speed inversion.
Preferably, the preprocessing of the L1 level data includes:
and carrying out angle correction on two characteristics of a normalized bistatic radar cross section NBRCS and a DDM front slope LES in an L1 level data product of the CYGNSS to obtain L2 level correction data.
Preferably, the process of preprocessing the reference wind speed includes:
extracting longitude and latitude and observation time of an observation point from CYGNSS data;
and matching the corresponding observation point reference wind speed from wind speed assimilation data at a preset height above the sea surface issued by the ECMWF according to the principle that the longitude and latitude of the observation point are nearest to the observation time.
Preferably, the exponential function model includes: a LES index model and an NBRCS index model;
wherein the LES exponential function model is in the form of:；
the NBRCS exponential function model is in the form of:
；
wherein,for wind speed at 10m above sea surface, < >>And->To remove the LES and NBRCS values after the incidence angle effect, +.>Is a natural constant, and has a value of 2.71828%>The undetermined coefficients are determined by a nonlinear least squares method.
Preferably, the process of inputting the wind speed inversion coarse result and the DDM signal-to-noise ratio SNR together into a random forest to perform equal weight combination to obtain a final wind speed inversion result includes:
。
compared with the prior art, the invention has the beneficial effects that:
the invention provides a random forest model-based dual-model fusion GNSS-R sea surface wind speed inversion method and system, wherein the model combines the advantages of a geographic model function and a random forest model in a machine learning method, an exponential model is constructed by using LES and NBRCS, a wind speed inversion coarse result is obtained, then the two wind speed inversion coarse results and a signal-to-noise ratio are combined with each other by using random forest in equal weight, the description of signal power is introduced through the signal-to-noise ratio, the interference of signal power change on wind speed inversion is reduced, the characteristic relation among LES, NBRCS and SNR is established with LESs data quantity, and the sea surface wind speed inversion with high precision is realized.
Drawings
In order to more clearly illustrate the technical solutions of the present invention, the drawings that are needed in the embodiments are briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings can be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic flow chart of a GNSS-R sea surface wind speed inversion method based on random forests in an embodiment of the invention;
FIG. 2 is a flow chart of constructing a combined model based on random forests in an embodiment of the invention;
FIG. 3 is a schematic diagram showing the distribution of the reference wind speed of the sample in the embodiment of the present invention;
FIG. 4 is a diagram of an exponential model fitting scenario in an embodiment of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
In order that the above-recited objects, features and advantages of the present invention will become more readily apparent, a more particular description of the invention will be rendered by reference to the appended drawings and appended detailed description.
Global navigation satellite system reflection signal technology (GNSS-R): the L-band signal emitted by the navigation satellite is influenced by the reflecting surface in the aspects of waveform, amplitude, phase, frequency, polarization characteristics and the like to change after being reflected by the earth, and the novel remote sensing technology for determining the surface roughness and the surface characteristics by utilizing the reflected signal observation and the reflecting surface attribute in combination with the position of a receiver antenna and medium information.
CYGNSS: all Cyclone Global Navigation Satellite System is an earth system science finder task of NASA consisting of a constellation of eight microminiature planets, the observation station providing observation data of an orbital tilt of about 35 ° from the equator, with an average revisitation period of 7 hours, which can measure the sea surface in the range 38 ° N-38 ° S, which includes the tropical cyclone formation and the contiguous latitudinal zones of motion. The CYGNSS constellation may use GPS direct and reflected signals to measure ocean surface wind.
Delay-doppler correlation power plot (Delay Doppler Map, DDM): DDM is an important data source for GNSS-R wind speed inversion. The reflected signal is essentially a series of multipath signals reflected from areas surrounding the specular reflection point, which have different delay and doppler values due to the different propagation paths and movement of the receiver platform. Giving a Doppler correction frequency, multiplying a direct signal and a reflected signal at different delay points, adding and calculating to obtain delay related power at the Doppler frequency, changing the Doppler correction frequency, and repeating the process to obtain related power distribution at different delay points at different Doppler correction frequencies, thereby forming the DDM.
Leading edge slope (leading edge slope, LES): the fixed Doppler value is the Doppler frequency at the specular reflection point, the direct signal and the reflected signal are multiplied and added at different time delay points to obtain a one-dimensional time delay related power curve, and the front slope is the tangential slope of each point on the curve when the one-dimensional time delay related power curve is in the waveform rising stage.
Bistatic radar reflection section (bistatic radar cross section, BRCS): BRCS is a physical quantity that measures the intensity of echoes produced by a target under bistatic radar waves. It is the imaginary area of the target, expressed by the projected area of an equivalent reflector that is uniform in all directions, and has the same echo power as the defined target in the unit solid angle of the receiving direction.
Normalized bistatic radar reflection section (normalized bistatic radar cross section, NBRCS): the NBRCS is obtained by performing two-stage calibration of 1A and 1B on DDM, the 1A-stage calibration converts an original instrument receiving value into a DDM value of receiving power, and then a non-standardized BRCS and an effective scattering area are output through the 1B-stage calibration, wherein the NBRCS is obtained by dividing the BRCS by the effective scattering area, and represents the signal reflection capacity of a reflection target unit area.
Random forest regression (random forest): the random forest regression algorithm is a combination algorithm of decision tree regression, combining many regression decision trees together to reduce the risk of overfitting. A plurality of decision tree models are trained in parallel by random forests, and the prediction results of each decision tree are combined, so that the over-fitting problem can be reduced.
Example 1
As shown in fig. 1, the GNSS-R sea surface wind speed inversion method based on random forests includes the following steps:
obtaining L1 level data and reference wind speed of a CYGNSS, and respectively preprocessing the L1 level data and the reference wind speed to obtain a training sample set, wherein the L1 level data comprises: DDM front slope LES, normalized double-base radar cross section NBRCS and DDM signal-to-noise ratio SNR;
constructing an exponential function model by using the training sample set;
fitting the DDM front slope LES, the normalized bistatic radar scattering cross section NBRCS and a reference wind speed respectively by using an exponential function model to obtain a wind speed inversion coarse result;
and (3) inputting the wind speed inversion coarse result and the DDM signal-to-noise ratio SNR into a random forest together for equal weight combination to obtain a final wind speed inversion result, and finishing sea surface wind speed inversion.
In this embodiment, the invention performs GNSS-R wind speed inversion including three parts: (1) data preprocessing; (2) And (3) constructing and solving an exponential function model, and constructing a combined model based on the random forest. According to the method, first, angle correction is carried out on two characteristics of a Normalized Bistatic Radar Cross Section (NBRCS) and a DDM Leading Edge Slope (LES) in an L1 level data product of a CYGNSS, so that L2 level correction data are obtained. And then taking assimilation wind speed data from European mid-term weather forecast centers (European Centre for Medium-Range Weather Forecasts, ECMWF) as reference wind speeds, and respectively fitting NBRCS, LES and the reference wind speeds by using an exponential function model to obtain undetermined coefficients in the exponential function model. And finally, combining the two exponential models with the DDM signal-to-noise ratio (SNR) in the L1 level data product of the CYGNSS, and fusing by using a random forest regression model to obtain a final wind speed inversion model.
In this embodiment, the pretreatment of the features: the method is characterized in that two characteristics of a Normalized Bistatic Radar Cross Section (NBRCS) and a DDM front slope (LES) in an L1 level data product of CYGNSS are subjected to angle correction, so that in order to remove the influence of a signal incident angle, firstly, characteristic values are subjected to empirical correction, namely L2A correction in the processing of the CYGNSS product, and the expression formula is as follows:
（1）
wherein the method comprises the steps ofTwo features representing LES and NBRCS extracted from CYGNSS L1 DDM product; />To remove the correction value after the influence of the angle of incidence, +.>Is an angle of incidence empirical correction function:
（2）
wherein the angle of incidenceProvided by CYGNSS L1 product.
Pretreatment of reference wind speed: the longitude and latitude and the observation time of the observation point extracted from the CYGNSS data are matched with the reference wind speed of the point from the wind speed assimilation data of 10m above the sea surface issued by the European middle weather forecast center (ECMWF) according to the principle that the longitude and latitude are nearest to the time.
And (3) data quality control:
data screening is performed by signal-to-noise ratio SNR with distance gain correction (range corrected gain, RCG). Wherein the signal-to-noise ratio SNR is provided by a CYGNSS product, the distance gain correction RCG is calculated by the antenna gain at a specular reflection point (provided by the CYGNSS product), the distance from a receiver to the specular reflection point and the distance from a navigation satellite to the specular reflection point, the effect of the distance correction RCG is to reflect the actual gain condition of the signal at the position, and the larger the RCG is, the expression data quality is approximately suitable for wind speed inversion. The specific formula is as follows:
（3）
wherein,antenna gain in the direction of the specular reflection point, +.>For the receiver to specular reflection point distance, +.>For navigation satellite to specular reflection point distance, +.>For scaling the coefficients the objective is to scale the RCG to the order of 1-100.
The threshold value when picking data is: SNR > =4 and RCG > =10. The method aims to delete data with large noise and insufficient gain and avoid interference with wind speed inversion.
In this embodiment, the exponential function model includes: a LES index model and an NBRCS index model;
wherein the LES exponential function model is in the form of:（4）；
the NBRCS exponential function model is in the form of:
（5）；
wherein,for wind speed at 10m above sea surface, < >>And->To remove the LES and NBRCS values after the incidence angle effect, +.>Is a natural constant, and has a value of 2.71828%>The undetermined coefficients are determined by a nonlinear least squares method.
The model undetermined coefficient optimization method is Gaussian Newton method, and LES is taken as an example: firstly, constructing an optimized objective function:
（6）
i.e. minimizing the two norms of the difference between the output value of the exponential function model and the wind speed 10m above the sea surface, and the parameters to be optimized are. Then using an iterative method:
1. given an initial value
2. For the kth iteration, calculate the delta。
（7）
Wherein:
（8）
（9）
（10）
3. if it isLess than a given threshold, such as 1e-6, then stop. No->And returning to the step 2.
An exponential function model of LES and NBRCS to wind speed 10m above the sea surface was established using gauss newton method, respectively. The NBRCS exponential model determines coefficients in the same way as LES.
In the embodiment, the invention is based on a random forest method, and a combined model of wind speed inversion is constructed by using three characteristics of LES wind speed, NBRCS wind speed and SNR signal to noise ratio so as to improve the wind speed inversion precision. The model is as follows:
)（11）
the invention adopts the combined modelIs constructed based on a random forest method. Firstly, three characteristics of LES wind speed, NBRCS wind speed and SNR signal to noise ratio and a reference wind speed form a total training sample set. And then resampling the total training sample for designated times by adopting a Bagging resampling method to obtain a secondary training sample. And selecting segmentation features for each secondary sample, calculating Gini coefficients to obtain an optimal segmentation mode, and completing the establishment of a decision tree until a termination condition is reached. And finally, taking an arithmetic average value of the wind speed prediction result of each decision tree to obtain a final wind speed prediction result of the model. A flow chart for modeling is shown in fig. 2.
The main steps of constructing the combined model are as follows:
constructing a general training sample: bringing LES and NBRCS into the index model constructed in (2) to obtain respective prediction results of wind speedAnd->. DDM signal-to-noise ratio (SNR) is extracted from L1 level data product of CYGNSS, and the DDM signal-to-noise ratio (SNR) and the CYGNSS form a sample required for training, and the total training sampleSThe specific form of (2) is as follows:
（12）
（13）
a Bagging resampling method is adopted to construct a secondary training sample: for a given training sampleSIs provided to be common inMEach sample, each round of training samplesSIs extracted by a sampling mode with a put backMSecondary, co-proceedingnA wheel, obtaining a secondary training sample,nfor a given random forest size, herenThe training sets are independent.
Establishing a decision tree, and using the decision tree，/>And the data are continuously divided into two parts by the three characteristics of SNR, and the two parts of data are divided into two new nodes after being divided into two parts, so that the wind speeds of the two parts of divided data are as similar as possible. The index for evaluating the similarity is Gini (base coefficient), the smaller the base coefficient is, the higher the similarity of the nodes is, and the detailed calculation mode is referred to in (a). And (3) continuing to divide the new nodes obtained by dividing into two parts until the stopping condition is met, wherein the specific condition refers to (b).
(a) The Gini coefficient is calculated by: assuming training samplesSThenSThe method for calculating the coefficient of the kunity is as follows:
（15）
wherein the method comprises the steps ofIs sample belonging to->Probability of class, in the present method +.>Belonging to the two classification problems, let the probability of the sample belonging to the first class be +.>The probability of belonging to the second class is +.>Gini is at this time:
（16）
the features used in the method all belong to continuous values toCharacterized by example, the sample to be calculated Gini is calculated according toSequentially arranging from small to large, and taking +.>Is used as a dividing threshold, the Gini coefficient divided by the threshold is calculated, and the dividing threshold which can minimize the Gini coefficient is selected as +.>Optimal division threshold of features. Then pair->And->Repeating the above operationComparing the Gini obtained by dividing the three features, and selecting the dividing feature which can enable the Gini to be minimum and a threshold value as the dividing basis of the node. The higher the Gini value, the better the partitioning effect.
(b) Termination condition for decision tree partitioning: after division the sample space is split into two subsetsAnd->So Gini, which divides the nodes, is:
（17）
when all samples in the node belong to the same category, the Gini value of the node is 0, and the division is completed at this time, and the node stops splitting. Otherwise, continuing to calculate the node Gini value to be split, and selecting a splitting mode with the minimum Gini value for splitting. Stopping growth when one of two conditions is met: (1) The number of samples in the node belongs to the same category or is 1, (2) the number of decision tree layers reaches a set threshold value.
Calculating a final prediction result: the end result is a number average of the predicted results for each decision tree. The final output of the combined model is thus in the form of:
（14）
the random forest regression model can be completed with the aid of an open source machine learning library, for example, a random forest regression model is trained by a random forest regression function in a common Scikit-learn, and main parameters required to be set are as follows:
number of decision trees: generally, the number of decision trees is too small, the under fitting is easy, the number of decision trees is too large, the calculated amount is too large, and after the number of decision trees reaches a certain number, the model lifting obtained by increasing the number of decision trees is small, so that a moderate numerical value is generally selected. Default is 200.
Decision tree maximum depth: in general, this value may be taken into account when there is little data or little features, i.e. only one sample is divided into each branch. If the model sample size is large and the features are also large, the maximum depth is limited, and the specific value depends on the distribution of the data. The value is usually 10-100.
The internal node subdivides the minimum number of samples required: this value limits the condition for the sub-tree to continue partitioning, and if the number of samples of a node is less than the minimum number of samples, then no further partitioning will continue. Default is 2.
The leaf node least number of samples: this value limits the minimum number of samples for a leaf node, and if a leaf node is less than the number of samples, it will be pruned along with siblings, leaving only the parent node as it was. Default is 1.
The leaf node least number of samples: this value limits the minimum number of samples for a leaf node, and if a leaf node is less than the number of samples, it will be pruned along with siblings, leaving only the parent node as it was. Default is 1.
Example two
The invention also provides a GNSS-R sea surface wind speed inversion system based on the random forest, which comprises: the device comprises a preprocessing module, a construction module, a fitting module and a combination module;
the preprocessing module is used for acquiring L1 level data and reference wind speed of the CYGNSS, and respectively preprocessing the L1 level data and the reference wind speed to obtain a training sample set, wherein the L1 level data comprises: DDM front slope LES, normalized double-base radar cross section NBRCS and DDM signal-to-noise ratio SNR;
the construction module is used for constructing an exponential function model by utilizing the training sample set;
the fitting module is used for respectively fitting the DDM front slope LES, the normalized bistatic radar scattering cross section NBRCS and the reference wind speed by using an exponential function model to obtain a wind speed inversion coarse result;
the combination module is used for inputting the wind speed inversion coarse result and the DDM signal-to-noise ratio SNR into a random forest together for equal weight combination, obtaining a final wind speed inversion result and finishing sea surface wind speed inversion.
In this embodiment, the process of preprocessing the L1 level data includes:
and carrying out angle correction on two characteristics of a normalized bistatic radar cross section NBRCS and a DDM front slope LES in an L1 level data product of the CYGNSS to obtain L2 level correction data.
In this embodiment, the process of preprocessing the reference wind speed includes:
extracting longitude and latitude and observation time of an observation point from CYGNSS data;
and matching the corresponding observation point reference wind speed from wind speed assimilation data at a preset height above the sea surface issued by the ECMWF according to the principle that the longitude and latitude of the observation point are nearest to the observation time.
In this embodiment, the exponential function model includes: a LES index model and an NBRCS index model;
wherein, the LES exponential function model has the following form:；
the NBRCS exponential function model is in the form of:
；
wherein,for wind speed at 10m above sea surface, < >>And->To remove the LES and NBRCS values after the incidence angle effect, +.>Is a natural constant, and has a value of 2.71828%>For the coefficients to be determined,determined by a nonlinear least squares method.
In this embodiment, the process of inputting the wind speed inversion coarse result and the DDM signal-to-noise ratio SNR together into a random forest to perform equal weight combination, and obtaining a final wind speed inversion result includes:
。
the invention is based on a random forest method, and utilizes three characteristics of LES wind speed, NBRCS wind speed and SNR signal to noise ratio to construct a combined model of wind speed inversion so as to improve wind speed inversion accuracy. The random forest is used for combining the two exponential wind speed models, the response capability of the two features to wind speed is comprehensively utilized, and a wind speed inversion result with higher precision is obtained.
The invention introduces signal-to-noise ratio SNR as a supplementary feature in the random forest combination model for expressing signal power so as to reduce interference generated by signal power variation in the wind speed inversion process.
Example III
The verification stage of the invention uses data of 1 month 1 day to 7 days of 2020 provided by a CYGNSS satellite constellation for model training and testing, and the reference wind speed uses assimilation wind speed provided by ECMWF at the same time. The data quality control is realized by signal-to-noise ratio SNR and distance correction gain RCG, and data with SNR smaller than 4 or RCG smaller than 10 are removed, so that the interference of abnormal data is avoided, and 11295 data samples are generated. Wherein 90% of the data are randomly extracted for establishing an exponential model and the random forest regression model are used for fusion, and 10% of the data are used for accuracy verification. The distribution of the reference wind speeds of the samples is shown in fig. 3. The exponential model fitting case is shown in fig. 4.
The specific values of the index model are as follows:
the LES exponential function model is in the form of:
（4），
the specific values in the invention are shown in table 1:
TABLE 1
The NBRCS exponential function model is in the form of:
（5），
the specific values in the invention are shown in table 2:
TABLE 2
The pair of accuracies for the final model are shown in table 3:
TABLE 3 Table 3
The above embodiments are merely illustrative of the preferred embodiments of the present invention, and the scope of the present invention is not limited thereto, but various modifications and improvements made by those skilled in the art to which the present invention pertains are made without departing from the spirit of the present invention, and all modifications and improvements fall within the scope of the present invention as defined in the appended claims.
Claims (6)
1. The GNSS-R sea surface wind speed inversion method based on the random forest is characterized by comprising the following steps of:
acquiring L1 level data and a reference wind speed of a CYGNSS, and respectively preprocessing the L1 level data and the reference wind speed to obtain a training sample set, wherein the L1 level data comprises: DDM front slope LES, normalized double-base radar cross section NBRCS and DDM signal-to-noise ratio SNR;
constructing an exponential function model by utilizing the training sample set;
fitting the DDM front slope LES, the normalized bistatic radar cross section NBRCS and the reference wind speed by using the exponential function model to obtain a wind speed inversion coarse result;
inputting the wind speed inversion coarse result and the DDM signal-to-noise ratio SNR into a random forest together for equal weight combination to obtain a final wind speed inversion result, and finishing sea surface wind speed inversion;
the exponential function model includes: a LES exponential function model and an NBRCS exponential function model;
wherein the LES exponential function model is in the form of:；
the NBRCS exponential function model is in the form of:；
wherein,for wind speed at 10m above sea surface, < >>And->To remove the LES and NBRCS values after the incidence angle effect, +.>Is a natural constant, and has a value of 2.71828%>As a coefficient to be determined, determining by a nonlinear least square method;
the wind speed inversion coarse result and the DDM signal-to-noise ratio SNR are input into a random forest together for equal weight combination, and the method for obtaining the final wind speed inversion result comprises the following steps:
。
2. the random forest based GNSS-R sea surface wind speed inversion method of claim 1 wherein the method of preprocessing the L1 level data comprises:
and carrying out angle correction on two characteristics of a normalized bistatic radar cross section NBRCS and a DDM front slope LES in an L1 level data product of the CYGNSS to obtain L2 level correction data.
3. The random forest based GNSS-R sea surface wind speed inversion method of claim 1 wherein the method of preprocessing the reference wind speed comprises:
extracting longitude and latitude and observation time of an observation point from CYGNSS data;
and matching the corresponding observation point reference wind speed from wind speed assimilation data at a preset height above the sea surface issued by the ECMWF according to the principle that the longitude and latitude of the observation point are nearest to the observation time.
4. GNSS-R sea surface wind speed inversion system based on random forest, which is characterized by comprising: the device comprises a preprocessing module, a construction module, a fitting module and a combination module;
the preprocessing module is used for acquiring L1 level data and reference wind speed of a CYGNSS, and respectively preprocessing the L1 level data and the reference wind speed to obtain a training sample set, wherein the L1 level data comprises: DDM front slope LES, normalized double-base radar cross section NBRCS and DDM signal-to-noise ratio SNR;
the construction module is used for constructing an exponential function model by utilizing the training sample set;
the fitting module is used for respectively fitting the DDM front slope LES, the normalized bistatic radar scattering cross section NBRCS and the reference wind speed by using the exponential function model to obtain a wind speed inversion coarse result;
the combination module is used for inputting the wind speed inversion coarse result and the DDM signal-to-noise ratio SNR into a random forest together for equal weight combination to obtain a final wind speed inversion result, and finishing sea surface wind speed inversion;
the exponential function model includes: a LES exponential function model and an NBRCS exponential function model;
wherein the LES exponential function model is in the form of:；
the NBRCS exponential function model is in the form of:；
wherein,for wind speed at 10m above sea surface, < >>And->To remove the LES and NBRCS values after the incidence angle effect, +.>Is a natural constant, and has a value of 2.71828%>As a coefficient to be determined, determining by a nonlinear least square method;
inputting the wind speed inversion coarse result and the DDM signal-to-noise ratio SNR into a random forest together for equal weight combination, wherein the process of obtaining the final wind speed inversion result comprises the following steps:
。
5. the random forest based GNSS-R sea surface wind speed inversion system of claim 4 wherein preprocessing said L1 level data comprises:
and carrying out angle correction on two characteristics of a normalized bistatic radar cross section NBRCS and a DDM front slope LES in an L1 level data product of the CYGNSS to obtain L2 level correction data.
6. The random forest based GNSS-R sea surface wind speed inversion system of claim 4 wherein the process of preprocessing the reference wind speed comprises:
extracting longitude and latitude and observation time of an observation point from CYGNSS data;
and matching the corresponding observation point reference wind speed from wind speed assimilation data at a preset height above the sea surface issued by the ECMWF according to the principle that the longitude and latitude of the observation point are nearest to the observation time.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311557686.XA CN117272843B (en) | 2023-11-22 | 2023-11-22 | GNSS-R sea surface wind speed inversion method and system based on random forest |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311557686.XA CN117272843B (en) | 2023-11-22 | 2023-11-22 | GNSS-R sea surface wind speed inversion method and system based on random forest |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117272843A CN117272843A (en) | 2023-12-22 |
| CN117272843B true CN117272843B (en) | 2024-02-02 |
Family
ID=89218162
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311557686.XA Active CN117272843B (en) | 2023-11-22 | 2023-11-22 | GNSS-R sea surface wind speed inversion method and system based on random forest |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117272843B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117872424B (en) * | 2024-03-11 | 2024-05-10 | 中国石油大学(华东) | GNSS-R sea surface DDM data generation method and device |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112433233A (en) * | 2020-11-19 | 2021-03-02 | 武汉大学 | GNSS-R sea surface wind speed inversion method and system based on particle swarm optimization |
| WO2021218424A1 (en) * | 2020-04-30 | 2021-11-04 | 江苏科技大学 | Rbf neural network-based method for sea surface wind speed inversion from marine radar image |
| WO2022005619A2 (en) * | 2020-05-18 | 2022-01-06 | The Regents Of The University Of Colorado, A Body Corporate | Ocean surface wind direction retrieval from reflected radio signals on space-borne platforms |
| WO2022016884A1 (en) * | 2020-07-22 | 2022-01-27 | 江苏科技大学 | Method for extracting sea surface wind speed on basis of k-means clustering algorithm |
| CN114139566A (en) * | 2021-08-27 | 2022-03-04 | 中国空间技术研究院 | Method for improving accuracy of measurement based on machine learning weighted average fusion feature extraction |
| CN114647812A (en) * | 2022-03-17 | 2022-06-21 | 中国科学院国家空间科学中心 | GNSS-R sea wind inversion method based on multi-dimensional feature excavation neural network |
| CN114861537A (en) * | 2022-04-29 | 2022-08-05 | 武汉大学 | GNSS-R sea surface wind speed inversion method and system based on CNN multi-information fusion |
| CN115455801A (en) * | 2022-07-18 | 2022-12-09 | 北京信息科技大学 | Sea surface wind speed inversion method and device for HY-2B scanning microwave radiometer based on PSO-DNN |
| CN116148863A (en) * | 2022-12-20 | 2023-05-23 | 昆明理工大学 | Multi-mode deep learning method for satellite-borne GNSS-R sea surface wind speed inversion based on CNN |
| CN116539913A (en) * | 2023-05-04 | 2023-08-04 | 极诺星空(北京)科技有限公司 | Method and device for on-board real-time inversion of sea surface wind speed |
-
2023
- 2023-11-22 CN CN202311557686.XA patent/CN117272843B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021218424A1 (en) * | 2020-04-30 | 2021-11-04 | 江苏科技大学 | Rbf neural network-based method for sea surface wind speed inversion from marine radar image |
| WO2022005619A2 (en) * | 2020-05-18 | 2022-01-06 | The Regents Of The University Of Colorado, A Body Corporate | Ocean surface wind direction retrieval from reflected radio signals on space-borne platforms |
| WO2022016884A1 (en) * | 2020-07-22 | 2022-01-27 | 江苏科技大学 | Method for extracting sea surface wind speed on basis of k-means clustering algorithm |
| CN112433233A (en) * | 2020-11-19 | 2021-03-02 | 武汉大学 | GNSS-R sea surface wind speed inversion method and system based on particle swarm optimization |
| CN114139566A (en) * | 2021-08-27 | 2022-03-04 | 中国空间技术研究院 | Method for improving accuracy of measurement based on machine learning weighted average fusion feature extraction |
| CN114647812A (en) * | 2022-03-17 | 2022-06-21 | 中国科学院国家空间科学中心 | GNSS-R sea wind inversion method based on multi-dimensional feature excavation neural network |
| CN114861537A (en) * | 2022-04-29 | 2022-08-05 | 武汉大学 | GNSS-R sea surface wind speed inversion method and system based on CNN multi-information fusion |
| CN115455801A (en) * | 2022-07-18 | 2022-12-09 | 北京信息科技大学 | Sea surface wind speed inversion method and device for HY-2B scanning microwave radiometer based on PSO-DNN |
| CN116148863A (en) * | 2022-12-20 | 2023-05-23 | 昆明理工大学 | Multi-mode deep learning method for satellite-borne GNSS-R sea surface wind speed inversion based on CNN |
| CN116539913A (en) * | 2023-05-04 | 2023-08-04 | 极诺星空(北京)科技有限公司 | Method and device for on-board real-time inversion of sea surface wind speed |
Non-Patent Citations (3)
| Title |
|---|
| Milad Asgarimehr.A GNSS-R Geophysical Model Function：Machine Learning for Wind Speed Retrievals.《IEEE Geoscience and Remote Sensing Letters》.2019,全文. * |
| 布金伟.星载GNSS-R技术反演海面降雨强度及风速和浪高方法研究.《万方数据库》.2023,全文. * |
| 骆黎明 ; 白伟华 ; 孙越强 ; 夏俊明 ; .基于树模型机器学习方法的GNSS-R海面风速反演.空间科学学报.2020,(第04期),全文. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117272843A (en) | 2023-12-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110186823B (en) | Aerosol optical thickness inversion method | |
| CN117272843B (en) | GNSS-R sea surface wind speed inversion method and system based on random forest | |
| CN111639747B (en) | GNSS-R sea surface wind speed inversion method and system based on BP neural network | |
| von Albedyll et al. | Thermodynamic and dynamic contributions to seasonal Arctic sea ice thickness distributions from airborne observations | |
| CN111965608B (en) | Satellite-borne ocean laser radar detection capability assessment method based on chlorophyll concentration of water body | |
| WO2022005619A2 (en) | Ocean surface wind direction retrieval from reflected radio signals on space-borne platforms | |
| CN115293198A (en) | Method for improving GNSS-R height finding inversion accuracy based on multi-hidden-layer neural network | |
| Mooneyham et al. | SWRL Net: A spectral, residual deep learning model for improving short-term wave forecasts | |
| CN113468804A (en) | Underground pipeline identification method based on matrix bundle and deep neural network | |
| CN117348045A (en) | Optimization method and device for selecting reflected signals of multimode GNSS-R receiver | |
| CN117075149A (en) | DDM-based satellite-borne GNSS-R typhoon position estimation method and system | |
| Meucci et al. | Evaluation of Spectral Wave Models Physics as Applied to a 100‐Year Southern Hemisphere Extra‐Tropical Cyclone Sea State | |
| CN113093225B (en) | Wide-area and local-area fused high-precision ionospheric scintillation model establishment method | |
| TANG et al. | Application of LVQ neural network combined with the genetic algorithm in acoustic seafloor classification | |
| Cornford et al. | Improved neural network scatterometer forward models | |
| CN117872424B (en) | GNSS-R sea surface DDM data generation method and device | |
| CN110031847A (en) | The dynamic of wavelet transformation and support vector machines combination radar reflectivity is short to face Quantitative Precipitation estimating and measuring method | |
| CN113534085B (en) | Sea surface wind speed and effective wave height joint inversion method of interference imaging altimeter | |
| Bu et al. | Multi-Observable Wind Speed Retrieval Based on Spaceborne GNSS-R Delay Doppler Maps | |
| Lwin et al. | Leveraging Bayesian deep learning for spaceborne GNSS-R retrieval on global soil moisture | |
| Kim et al. | Application of data assimilation for spectral wave model in coastal regions of South Korea | |
| CN117332820A (en) | DEM correction method based on feature enhancement integrated learning framework | |
| CN116306802A (en) | Heterogeneous multi-mode deep learning model and GNSS-R height measurement method | |
| Borg et al. | Predicting sea surface wave and wind parameters from satellite radar images using machine learning | |
| Efecik | A Genetic Algorithm Approach to Estimate Drop Size Distribution Using GPM DPR |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |