A method and system for retrieving SLA based on along-track data of an airborne altimeter

Abstract

The present invention discloses a method and system for inverting SLA based on along-track data of an airborne altimeter, belonging to the field of ocean observation technology. For the along-track data of the airborne altimeter, the present invention first conducts statistical analysis; filters the along-track data, and performs quadratic interpolation fitting based on the data interval of the airborne altimeter: interpolates the along-track MSS according to the sampling interval of the altimeter data; processes the along-track tidal data in the same way to obtain tidal data; obtains the corresponding SLA data; processes the AVISO and SWOT data in the same interpolation way; and finally realizes the inversion of SLA data based on the adaptive feature model. The present invention realizes the inversion of high-resolution SLA data from the aliased signals of the airborne synthetic aperture radar altimeter, with an accuracy reaching the centimeter level, verifies the working mechanism of the new system payload, obtains high-precision along-track SLA data, and lays a solid data and algorithm foundation for the development of subsequent satellite payloads.

Description

Technical Field

[0001] The present invention belongs to the technical field of ocean observation, and more particularly, to a method and system for inverting Sea Level Anomaly (SLA) based on airborne altimeter along-track data. Background Art

[0002] Sea Level Anomaly (SLA) is an important reference index for climate change. It represents the change of the instantaneous sea level relative to the Mean Sea Surface (MSS) and is widely used in the study of ocean dynamic phenomena. Studying the inversion mechanism and variation law of SLA can provide a reference for human life and ocean climate research.

[0003] Spaceborne or airborne altimeter data can be used to infer various parameters, including sea surface height, sea surface wind speed, significant wave height, and the topography of land, sea ice, and ice sheets. Spaceborne along-track altimeter data can ensure the acquisition of high-precision along-track data due to the high stability of the spaceborne platform and the high-precision calibration and verification field constructed. Currently, for airborne along-track SLA inversion, most are based on optimizing or improving the echo model to track the altimeter echo for along-track SLA data inversion. The tracking process is divided into two stages: height tracking and retracking. Among them, retracking mostly uses the least squares operator to obtain high-precision elevation parameters, and there are upper limits to the estimation accuracy of elevation parameters and the noise suppression performance of the algorithm, which easily causes overfitting of the elevation parameter estimation results and has limited adaptability to complex elevation changes. Summary of the Invention

[0004] The purpose of the present invention is to provide a method and system for inverting SLA based on airborne altimeter along-track data to address the signal error terms such as airborne platform jitter and sea waves in the airborne altimeter along-track data, and to realize the inversion of high-resolution SLA data from the along-track aliased signals to make up for the deficiencies of the existing technology.

[0005] To achieve the above object, the present invention is realized through the following technical solutions:

[0006] A method for inverting SLA based on airborne altimeter along-track data, comprising the following steps:

[0007] S1: Obtain the measured airborne altimeter along-track data and auxiliary data AVISO and SWOT;

[0008] S2: Filter the airborne altimeter along-track data: For the airborne altimeter along-track data A1, first perform statistical analysis on the sea wave signal and the oscillation frequency of the flight platform; first process the A1 data using a low-pass filter to obtain A2 data; perform spectral analysis on the flight altitude change data during the flight and perform high-pass filtering on the A2 data to obtain A3;

[0009] S3: Perform quadratic interpolation fitting on MSS, tides, AVISO, and SWOT based on the data interval of the airborne altimeter: According to the sampling interval of the altimeter data A3, interpolate the along-track MSS and obtain the corresponding MSS data B1; process the along-track tide data in the same way to obtain the tide data C1; obtain the corresponding SLA data D1 from A3 - B1 - C1; process the AVISO and SWOT data in the same interpolation method to obtain E1 and F1;

[0010] S4: Realize the inversion of SLA data based on the adaptive feature model: According to the obtained SLA data D1, build a deep learning framework; among them, D1 is used as the feature data, that is, the input X variable of the model; E1 and F1 are used as the label data Y(E) and Y(F), that is, the output result variables expected by the model; train the data with labels so that the neural network can learn the mapping relationship from input to output, and optimize the model according to the existing label data to make it more accurately output SLA(E) and SLA(F);

[0011] S5: Invert SLA according to the optimized adaptive feature model.

[0012] Furthermore, in S2, the calculation basis of the sampling filtering threshold is as follows:

[0013] According to the sampling theorem, the sampling frequency must be greater than twice the maximum frequency of the signal itself to restore the signal; regard the frequency of the airborne altimeter observation data as the sampling of SSH, and the sampling frequency is 1 - 2 hz;

[0014] Low-pass filtering, assuming the sampling frequency is 1000 HZ and the maximum frequency of the signal itself is 500 hz, to filter out the frequency part above 400 hz, that is, the cut-off frequency is 400 hz, then WN = 2 * 400 / 1000 = 0.8;

[0015] High-pass filtering, assuming the sampling frequency is 1000 HZ and the maximum frequency of the signal itself is 500 hz, to filter out the frequency part below 100 hz, that is, the cut-off frequency is 100 hz, then WN = 2 * 100 / 1000 = 0.2.

[0016] Furthermore, the adaptive feature model consists of 1 LSTM(256) containing 256 neurons, 2 LSTM(128) containing 128 neurons, 2 fully connected layers containing 64 neurons, and 1 fully connected layer containing 1 neuron.

[0017] Furthermore, the specific process of the adaptive feature model inverting high-precision along-track SLA data is as follows:

[0018] D1 is used as the feature data, that is, the input X variable of the model; E1 and F1 are used as the label data Y(E) , Y (F) , that is, the output result variable expected by the model, constructs the training sample θ based on X and Y:

[0019] X train , X test = split(X) (1)

[0020] Y train , Y test = split(Y) (2)

[0021] θ = (X train , Y train ) (3)

[0022] Train θ so that the neural network can learn the mapping relationship from input to output. The tanh function maps the input value to a state value in the range of -1 to 1:

[0023]

[0024] The input gate mainly includes two activation functions, σ and tanh, whose function is to update the memory cell state, and its structure can be expressed as:

[0025] i t = σ(W i · [h t-1 , x t + b i ) (5)

[0026]

[0027] In the formula, the acceptance weight i of the information is determined by the σ function t , and the candidate state is calculated through the tanh function

[0028] The activation value f of the forget gate t is calculated from the current input x t and the previous hidden state h t-1 through a fully connected layer with a σ activation function:

[0029] f t = σ(W f · [h t-1 , x t + b f ) (7)

[0030] In the formula, W and b represent the weight matrix and bias respectively; f t is a vector with a value range of 0 to 1, where the values inside the vector represent the cell state C t-1Whether the information in it is retained; a value of 0 means discard, and a value of 1 means retain;

[0031] Update the cell state C by combining the results of the forget gate and the input gate t :

[0032]

[0033] In the formula, f t *C t-1 determines the information forgotten in C t-1 , while determines what information in is added to the new memory cell C t ;

[0034] The output gate is mainly composed of o obtained by combining the short-term memory and the current input information t and h output by the long-term memory t in two parts, and its expression is:

[0035] o t = σ(W o ·[h t-1 , x t +b o ) (9)

[0036] h t = o t *tanh(C t ) (10)

[0037] In the formula, through the hidden layer o t , the value of the final hidden state h t is obtained, that is, the expected output result SLA (E) 、SLA (F) .

[0038] A system for inverting SLA based on along-track data of an airborne altimeter, comprising:

[0039] A data acquisition module to obtain the measured along-track data of the airborne altimeter and the auxiliary data AVISO, SWOT;

[0040] A data filtering and processing module for filtering the along-track data of the airborne altimeter;

[0041] A data quadratic interpolation and fitting module for performing quadratic interpolation and fitting on MSS, tides, AVISO, SWOT based on the data interval of the airborne altimeter;

[0042] A model optimization module for optimizing the adaptive feature model;

[0043] The inversion module realizes the inversion of SLA data based on the adaptive feature model.

[0044] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0045] The present invention proposes a method combining band-pass filtering and adaptive feature AI analysis to realize the inversion of high-resolution SLA data from the aliased signals of an airborne synthetic aperture radar altimeter, with an accuracy reaching the centimeter level, verifying the working mechanism of a new type of payload, obtaining high-precision along-track SLA data, and laying a solid data and algorithm foundation for the development of subsequent satellite payloads. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a schematic diagram of the unmanned aerial vehicle crossing from A (cold vortex) to B (warm vortex) on April 12 in Embodiment 1.

[0047] Figure 2 It is a schematic diagram of the SWOT orbit on April 22 in Embodiment 1.

[0048] Figure 3 It is a mapping relationship diagram from input to output.

[0049] Figure 4 It is a processing flow chart of the adaptive feature analysis model.

[0050] Figure 5 It is a basic architecture diagram of the adaptive feature analysis model.

[0051] Figure 6 It is a schematic diagram showing that the change trend of the SLA curve is opposite to the trend from the cold vortex to the warm vortex.

[0052] Figure 7 It is a schematic diagram showing that the change of the SLA signal is the same as the change trend from the cold vortex to the warm vortex.

[0053] Figure 8 It is a schematic diagram of obtaining the along-track SLA result by performing inversion calculation on the SLA.

[0054] Figure 9 It is a result diagram of SLA inversion for the overlapping part of the airborne altimeter and the SWOT track off the coast of Sanya. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0055] The following further describes and explains the technical solutions of the present invention in conjunction with the embodiments and the accompanying drawings.

[0056] Embodiment 1

[0057] A method for inverting SLA based on the along-track data of an airborne altimeter includes the following steps:

[0058] 1. Data set

[0059] 1) Measured along-track data of the airborne altimeter

[0060] In April 2023, the verification of the working mechanism and system mechanism of the new system payload was carried out in the northern South China Sea area, and relevant airborne altimeter along-track observation data was accumulated synchronously, such as Figure 1 shown.

[0061] 2) Auxiliary data

[0062] AVISO data E1 at 0.25 degrees along the track and SWOT transit data F1 during the test period, such as Figure 2 shown.

[0063] 2. Data processing logic

[0064] 1) Filter the airborne altimeter along-track data

[0065] For the airborne altimeter along-track data A1, first conduct statistical analysis on the sea wave signals (mainly wind waves and swells) and the oscillation frequency of the flight platform; first use low-pass filtering to process the A1 data of the along-track data to obtain A2 data; conduct spectral analysis on the flight altitude change data during flight and perform high-pass filtering on the A2 data to obtain A3;

[0066] 2) Perform quadratic interpolation fitting on MSS, tides, AVISO, and SWOT based on the data intervals of the airborne altimeter

[0067] According to the sampling interval of the altimeter data A3, interpolate the MSS along the track and obtain the corresponding MSS data B1; process the tidal data along the track in the same way to obtain the tidal data C1; subtract A3 - B1 - C1 to obtain the corresponding SLA data D1; process the AVISO and SWOT data in the same interpolation method to obtain E1 and F1;

[0068] 3) Realize the inversion of SLA data based on adaptive feature AI analysis

[0069] According to the obtained SLA data D1, build a deep learning framework; among them, D1 is used as the feature data, that is, the input X variable of the model; E1 and F1 are used as the label data Y(E) and Y(F), that is, the output result variables expected by the model. Train the data with labels so that the neural network can learn the mapping relationship from input to output, and optimize the model according to the existing label data to make it more accurately output SLA(E) and SLA(F).

[0070] The data processing flow chart, as Figure 4 shown, includes the following steps:

[0071] 1) Filter the along-track observation data of the airborne altimeter to remove the influence of sea waves (wind waves and swells) and the vibration of the flight platform, and obtain SSH1;

[0072] 2) Conduct frequency analysis on the flight altitude change signal of the data SSH1, and remove the flight altitude change factor from SSH1 through high-pass filtering to obtain SSH2;

[0073] 3) Based on the sampling interval of SSH2, perform quadratic interpolation fitting on the along-track MSS, tide, and AVISO data; subtract the MSS and tide data from SSH2 to obtain SLA data;

[0074] 4) Input the SLA data and AVISO into the adaptive feature analysis model to output high-precision SLA along-track data.

[0075] 3. Compare and analyze the inverted SLA(E) and SLA(F) with AVISO and SWOT, and calculate the RMSE value.

[0076] Compare and analyze this data with the 0.25-degree AVISO along the track using the adaptive feature analysis method to obtain the corresponding SLA(E) data, with an RMSE of 0.44 cm;

[0077] Compare and analyze this data with the SWOT data along the track using the same adaptive feature analysis method to obtain the corresponding SLA(F) data, with an RMSE of 1.12 cm.

[0078] Among them,

[0079] 1. Realize the analysis and processing of the airborne altimeter along-track data based on the filtering method

[0080] Analyze the sea waves (wind waves and swells) and the vibration frequency of the aircraft platform in the South China Sea sea trial area to obtain that the sea waves during April have a statistical period of 5s - 9S and a frequency of 0.1 - 0.2HZ;

[0081] The oscillation frequency of the aircraft platform is above 20HZ;

[0082] By using the principle of low-pass filtering, it is necessary to filter out the signal data above 0.2HZ, and the low-pass filtering threshold is 2 * 0.2hz / 1 = 0.4hz;

[0083] There is a signal of ascent and descent during the flight. Through frequency analysis, the along-track frequency of the aircraft on April 12 is 0.08HZ, so the high-pass filtering threshold of the aircraft is 0.16hz (2 * 0.08hz / 1 = 0.16hz).

[0084] Table 1 Processing table of airborne altimeter along-track data

[0085]

[0086]

[0087] The basis for calculating the sampling filtering threshold is as follows:

[0088] According to the sampling theorem, the sampling frequency must be greater than twice the maximum frequency of the signal itself in order to restore the signal.

[0089] Therefore, the frequency of the airborne altimeter observation data can be regarded as the sampling of SSH, and the sampling frequency is 1-2 hz. In this article, 1 hz is used as the basis.

[0090] Low-pass filtering. Assuming the sampling frequency is 1000 HZ and the maximum frequency of the signal itself is 500 hz, the frequency part above 400 hz needs to be filtered out, that is, the cut-off frequency is 400 hz, then WN = 2 * 400 / 1000 = 0.8;

[0091] High-pass filtering. Assuming the sampling frequency is 1000 HZ and the maximum frequency of the signal itself is 500 hz, the frequency part below 100 hz needs to be filtered out, that is, the cut-off frequency is 100 hz, then WN = 2 * 100 / 1000 = 0.2.

[0092] 2. Based on the analysis of the adaptive feature model, invert the high-precision along-track SLA data

[0093] The adaptive feature analysis model is as Figure 5 shown. This model consists of 1 LSTM(256) containing 256 neurons, 2 LSTM(128) containing 128 neurons, 2 fully connected layers containing 64 neurons, and 1 fully connected layer containing 1 neuron.

[0094] Table 2 Adaptive Feature Analysis Model Parameter Table

[0095]

[0096] Based on AVISO data, SWOT data and the filtered SLA data, the high-precision along-track SLA data is inverted through the adaptive feature model, and the accuracy reaches the centimeter level. Through the analysis of the inverted data, it can be seen that the airborne observation and the spaceborne observation have good consistency, and the airborne altimeter observation can obtain more small-scale change signals.

[0097] D1 is used as the feature data, that is, the input X variable of the model; E1 and F1 are used as the label data Y (E) 、Y (F) , that is, the output result variables expected by the model. Based on X and Y, the training sample θ is constructed:

[0098] Xtrain , X test = split(X) (1)

[0099] Y train , Y test = split(Y) (2)

[0100] θ = (X train , Y train ) (3)

[0101] Train θ so that the neural network can learn the mapping relationship from input to output. The tanh function maps the input value to a state value in the range of -1 to 1:

[0102]

[0103] The input gate mainly includes two activation functions, σ and tanh, whose function is to update the state of the memory cell, and its structure can be expressed as:

[0104] i t = σ(W i · [h t-1 , x t + b i ) (5)

[0105]

[0106] In the formula, the acceptance weight i of the information is determined by the σ function t , and the candidate state is calculated by the tanh function The activation value f of the forget gate t is calculated from the current input x t and the previous hidden state h t-1 through a fully connected layer with a σ activation function:

[0107] f t = σ(W f · [h t-1 , x t + b f ) (7)

[0108] In the formula, W and b represent the weight matrix and bias respectively. f t is a vector with a value range of 0 to 1, where the values within the vector indicate whether the information in the cell state C t-1 is retained. A value of 0 means discard, and a value of 1 means retain.

[0109] Combine the results of the forget gate and the input gate to update the cell state C t :

[0110]

[0111] In the formula, f t *C t-1 determines the information forgotten in C t-1 , while determines the addition of the information in to the new memory cell C t .

[0112] The output gate is mainly composed of o obtained by combining short-term memory and current input information t and h output from long-term memory t , as Figure 3 shown, and its expression is:

[0113] o t = σ(W o ·[h t-1 , x t + b o ) (9)

[0114] h t = o t * tanh(C t ) (10)

[0115] In the formula, through the hidden layer o t , the value of the final hidden state h t is obtained, that is, the expected output result SLA (E) , SLA (F) .

[0116] The data processing results are as follows:

[0117] 1) When obtaining the sea surface height anomaly (SLA) by subtracting the mean sea surface (MSS) from the sea surface height (SSH) parsed from the airborne altimeter signal (where the sea surface height data has removed the sea wave and flight platform oscillation signals through low-pass filtering), the change trend of its SLA curve is opposite to the trend from the cold eddy to the warm eddy, as Figure 6 shown.

[0118] 2) After removing the aircraft platform height change signal contained in the SSH in "Result Step 1)" through high-pass filtering and then calculating the SLA, the change of the SLA signal is the same as the change trend from the cold eddy to the warm eddy, but the accuracy gap is huge, SLA (meter level), AVISO (centimeter level), as Figure 7 shown.

[0119] 3) Based on the adaptive feature AI method proposed in the present invention, the SLA calculated in "Result Step 2)" is inverted to calculate the along-track SLA data, as Figure 8 shown.

[0120] The SLA inversion results of the aircraft along the track on April 12, compared with 0.25-degree AVISO, with an RMSE of 0.44 cm, achieving centimeter-level high-precision data inversion.

[0121] Based on the same processing logic, on April 22, the SLA inversion was performed on the overlapping part of the airborne altimeter and the SWOT track in the offshore area of Sanya, as Figure 9 shown.

[0122] On April 22, the SLA inversion of the in-transit SWOT data and the airborne altimeter data had an RMSE of 1.12 cm, and the airborne altimeter data could obtain more small-scale change signals.

[0123] The above results illustrate that the method and system provided by the present invention can achieve high-precision inversion of high-resolution SLA data from the aliased signals of airborne synthetic aperture radar altimeters.

[0124] Finally, although this specification is described according to the embodiments, not every embodiment only contains an independent technical solution. This narrative way of the specification is only for clarity. Those skilled in the art should regard the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for retrieving SLA based on along-track data of an airborne altimeter, characterized in that, The method includes the following steps: S1: Obtain the measured along-track data of the airborne altimeter and the auxiliary data AVISO and SWOT; S2: Perform filtering processing on the along-track data of the airborne altimeter: For the along-track data A1 of the airborne altimeter, first perform statistical analysis on the sea wave signal and the oscillation frequency of the flight platform; first process the along-track data A1 using low-pass filtering to obtain the processed data A2; perform spectral analysis on the flight altitude change data during flight, and perform high-pass filtering on the data A2 to obtain the altimeter data A3; S3: Perform quadratic interpolation fitting on MSS, tides, AVISO, and SWOT based on the data interval of the airborne altimeter: According to the sampling interval of the altimeter data A3, interpolate the MSS along the track and obtain the corresponding MSS data B1; process the tidal data along the track in the same way to obtain the tidal data C1; obtain the corresponding SLA data D1 by (A3 - B1 - C1); also use the interpolation method to process the AVISO and SWOT data respectively to obtain the processed data E1 and F1; where SLA is the sea surface height anomaly and MSS is the mean sea level; S4: Implement SLA data inversion based on the adaptive feature model: According to the obtained SLA data D1, build a deep learning framework; where D1 is used as feature data, that is, the input X variable of the model; E1 and F1 are used as label data Y (E) , Y (F) , that is, the output result variable expected by the model; train the data with labels so that the neural network can learn the mapping relationship from input to output, and optimize the model according to the existing label data to make it more accurately output SLA (E) , SLA (F) ; S5: Invert SLA according to the optimized adaptive feature model.

2. The method for inverting SLA based on along-track data of an airborne altimeter according to claim 1, wherein In the above S2, filtering processing is adopted, and the calculation basis is as follows: According to the sampling theorem, the sampling frequency must be greater than twice the maximum frequency of the signal itself to restore the signal; consider the observation data frequency of the airborne altimeter as the sampling of SSH, and the sampling frequency is 1 hz - 2 hz; For low-pass filtering, assume the sampling frequency is 1000 HZ, the maximum frequency of the signal itself is 500 hz, and the frequency part above 400 hz needs to be filtered out, that is, the cut-off frequency is 400 hz, then WN = 2 * 400 / 1000 = 0.8; For high-pass filtering, assume the sampling frequency is 1000 HZ, the maximum frequency of the signal itself is 500 hz, and the frequency part below 100 hz needs to be filtered out, that is, the cut-off frequency is 100 hz, then WN = 2 * 100 / 1000 = 0.

2.

3. The method for inverting SLA based on along-track data of an airborne altimeter according to claim 1, wherein The adaptive feature model is composed of 1 LSTM containing 256 neurons, 2 LSTMs containing 128 neurons, 2 fully connected layers containing 64 neurons, and 1 fully connected layer containing 1 neuron.

4. The method for inverting SLA based on along-track data of an airborne altimeter according to claim 1, wherein The specific process of the adaptive feature model for inverting high-precision along-track SLA data is as follows: D1 is used as feature data, that is, the input X variable of the model; E1 and F1 are used as label data Y (E) , Y (F) , that is, the output result variable expected by the model. Based on X and Y, a training sample θ is constructed: X train ,X test = split(X) (1) Y train ,Y test = split(Y) (2) θ = (X train , Y train ) (3) Train θ so that the neural network can learn the mapping relationship from input to output, and the tanh function maps the input value to the state value in the range of -1 to 1: The input gate includes two activation functions, σ and tanh, and its function is to update the memory cell state, and its structure is expressed as: i t = σ(W i· [h t-1 , x t + h i ) (5) wherein, the acceptance weight i of information is determined by the σ function t , and the candidate state is calculated by the tanh function Activation value f of the forget gate t Calculated from the current input x t and the previous hidden state h t-1 through a fully connected layer with the σ activation function as follows: f t = σ(W f· [h t-1 , x t + b f ) (7) where \(W\) and \(b\) represent the weight matrix and bias respectively; \(f\) t is a vector with values ranging from 0 to 1, where the values within the vector indicate whether the information in cell state \(C\) t-1 is retained; a value of 0 means discarded, and a value of 1 means retained; Update the cell state C by combining the results of the forget gate and the input gate t : where f t *C t-1 determines the forgotten information in C t-1 , and determines to add the information in to the new memory cell C t ; The output gate is mainly composed of o obtained by combining short-term memory and current input information t and h output from long-term memory t and consists of two parts, and its expression is: o t = σ(W o· [h t-1 ,x t +b o ) (9) h t = o t *tanh(C t ) (10) In the formula, through the hidden layer o t , the final hidden state h t is obtained, that is, the expected output result SLA (E) , SLA (F) .

5. The system for inverting SLA based on the along-track data of an airborne altimeter according to claim 1, characterized in that The system includes: A data acquisition module to obtain the measured along-track data of the airborne altimeter and the auxiliary data AVISO and SWOT; A data filtering processing module to perform filtering processing on the along-track data of the airborne altimeter; A data quadratic interpolation fitting module to perform quadratic interpolation fitting on MSS, tides, AVISO, and SWOT based on the data interval of the airborne altimeter; A model optimization module to optimize the adaptive feature model; An inversion module that implements SLA data inversion based on an adaptive feature model.

Images