Ocean internal wave prediction method based on satellite remote sensing data

Abstract

The invention discloses an ocean internal wave prediction method based on satellite remote sensing data, and the method comprises the steps: firstly obtaining a multi-source satellite remote sensing image of a target sea area, and building a multi-source satellite remote sensing internal wave crest line data set; and then constructing an internal wave numerical model, performing internal wave numerical simulation calculation by adopting the internal wave numerical model in a computational domain range of each two-dimensional section, further obtaining an internal wave generation statistical data set, finally constructing an internal wave generation statistical model, and obtaining an initial internal wave crest line. On the basis, matching an internal wave crest line of the remote sensing image with an initial internal wave crest line to obtain corresponding initial crest line information, an initial arrival time field and generation moments of N two-dimensional sections at a generation source of the Lusong strait, and finally constructing an internal wave initial arrival time error statistical model. And correcting the initial arrival time field to obtain a corrected arrival time field, thereby constructing an internal wave propagation model and realizing prediction of an internal wave crest line.

Description

Technical Field

[0001] The present invention belongs to the technical field of internal wave prediction, and in particular relates to a method for predicting ocean internal waves based on satellite remote sensing data. Background Art

[0002] Internal ocean waves are small- to medium-scale fluctuations that occur in stably density-stratified seawater and are widely distributed across the globe. Their generation mechanism stems primarily from the interaction between barotropic currents and seafloor topography, which stimulates the formation of internal waves in stably stratified seawater. The maximum amplitude of internal ocean waves occurs within the ocean interior, with a frequency between the buoyancy frequency and the inertial frequency. Their restoring force is primarily reduced gravity. However, because the reduced gravity generated by density differences in stratified seawater is only one thousandth of gravity, the amplitude of internal ocean waves is much greater than that of ocean waves, but their impact on sea surface fluctuations is almost negligible. Furthermore, internal ocean waves have a significant impact on vertical mixing in the upper ocean, the transport of marine nutrients, marine engineering, and underwater navigation. Therefore, research into prediction technologies for internal ocean waves is of great scientific value and practical significance.

[0003] The development of internal wave prediction technology has always been an important topic in the field of marine science. Among existing ocean internal wave prediction technologies, Jackson (2009) established an internal wave propagation model based on an empirical formula. This model establishes a functional relationship between the internal wave propagation velocity and the seawater depth to obtain the internal wave arrival time field, thereby achieving internal wave propagation prediction. However, during the internal wave propagation process, because only the single influencing factor of water depth is considered, other factors such as seawater stratification, internal wave amplitude, and background flow field are ignored. As a result, the arrival time error of the model prediction gradually accumulates with the increase of propagation time. Zhang and Li (2022) proposed a machine learning-based internal wave prediction method. By training a large number of remote sensing data pairs, this method can effectively obtain the internal wave propagation velocity field and achieve internal wave prediction. However, there is a significant technical flaw: this method must rely on remote sensing data to provide the initial crest line information of the internal wave, which may bring many limitations in practical applications. Gong et al. (2023) proposed an ocean internal wave prediction method based on the three-dimensional MITgcm numerical model. However, due to the high complexity of numerical simulation calculations and large consumption of computing resources, this method has the technical defect of low computational efficiency in long-term internal wave prediction applications, making it difficult to achieve commercial operation. Summary of the Invention

[0004] The present invention aims to provide a method for predicting ocean internal waves based on satellite remote sensing data.

[0005] In order to achieve the above objectives, the present invention adopts the following technical solutions.

[0006] A method for predicting ocean internal waves based on satellite remote sensing data, specifically comprising steps (1) to (6).

[0007] (1) Input the internal wave generation date into the A-wave generation statistical model, and output the A-wave on the first to Nth two-dimensional sections, respectively, from the generation source Luzon Strait East Ridge The time difference between reaching the western boundary of the corresponding spawning area after the moment is generated , and the corresponding maximum eastward tidal current , the maximum eastward tidal current corresponds to the tidal range , and solve them separately and The average value of and .

[0008] The internal wave generation date is input into the B-wave generation statistical model, and the output B-wave is generated on the first to Nth two-dimensional sections, respectively, from the generation source Luzon Strait West Ridge The time difference between reaching the western boundary of the corresponding spawning area after the moment is generated . And get the corresponding maximum value of eastward tidal current , the maximum eastward tidal current corresponds to the tidal range , the maximum moment of eastward tidal current The previous weak peak value of the positive pressure tidal current , and the weak wave peak corresponds to the tidal range , and solve them separately , , and The average value of , , and .

[0009] (2) Based on the time difference of the A wave reaching the western boundary of the first to N two-dimensional section generation area It can accurately calculate the average propagation velocity of the internal wave in the first to N two-dimensional sections within the A wave generation area, and use the average propagation velocity of the internal wave to determine the time difference between the A wave and the Nth two-dimensional section. After propagation, it reaches the position of the first to N two-dimensional sections, connects the positions of the first to N two-dimensional sections in sequence to obtain the initial peak line of wave A, and determines the formation time of the initial peak line of wave A .

[0010] Based on the time difference of B waves reaching the western boundary of the area generated from the first to N 2D sections It can accurately calculate the average propagation velocity of the internal wave in the first to N two-dimensional sections within the B-wave generation area, and use the average propagation velocity of the internal wave to determine the time difference between the B wave and the given time difference. After propagation, it reaches the position of the first to N two-dimensional sections, connects the positions of the first to N two-dimensional sections in sequence to obtain the initial peak line of the B wave, and determines the formation time of the initial peak line of the B wave .

[0011] (3) Using the internal wave propagation velocity-water depth function relationship, calculate the longitude and latitude coordinates of the i-th row and j-th column in the target sea area The propagation speed on , the propagation speed corresponding to all longitude and latitude coordinate points in the target sea area The set of is the velocity field in the internal wave propagation area. The internal wave propagation velocity-water depth function relationship is: .

[0012] in, Indicates the internal wave at the latitude and longitude coordinates The propagation speed at Indicates the longitude and latitude coordinates of a point Water depth at =2.971 m / s; B1=0.003; B2=1390.758 m.

[0013] (4) For the initial crest line of wave A determined in step (2), propagate it in the target sea area with the velocity field calculated in step (3), and use the fast marching algorithm to solve the Eikonal equation to obtain the coordinate point of the initial crest line of wave A in row i and column j in the target sea area. The earliest time .

[0014] For the initial crest line of the B wave determined in step (2), propagate in the target sea area with the velocity field calculated in step (3), and use the fast marching algorithm to solve the Eikonal equation to obtain the coordinate point of the initial crest line of the B wave at the i-th row and j-th column in the target sea area. The earliest time .

[0015] At the same time, obtain the coordinate point of row i and column j in the target sea area The corresponding maximum buoyancy frequency , jump depth , seabed topography height .

[0016] (5) Input A wave arrival time error statistical model, output , coordinate point Corrected arrival time The corrected arrival time set of all coordinate points in the target sea area obtained through this step is the corrected arrival time field of the initial crest line of wave A.

[0017] Will Input B wave arrival time error statistical model, output , coordinate point Corrected arrival time The corrected arrival time set of all coordinate points in the target sea area obtained through this step is the corrected arrival time field of the initial crest line of the B wave.

[0018] (6) Based on the corrected arrival time field of the initial wave crest line of wave A and the corrected arrival time field of the initial wave crest line of wave B, the arrival positions of the initial wave crest line of wave A and the initial wave crest line of wave B in step (2) at a given propagation time, or the arrival time at a given position, are calculated respectively, thus realizing the prediction of the internal wave crest line.

[0019] The internal waves are A waves and B waves, and the longitude and latitude range of the target sea area is 114°E to 123°E and 19°N to 23°N.

[0020] Specifically, the method for constructing the A-wave generation statistical model and the B-wave generation statistical model includes S1 to S4.

[0021] S1. Acquire multi-source satellite remote sensing images of the target sea area and establish a multi-source satellite remote sensing internal wave crest line dataset.

[0022] S2. In the target sea area, according to the spatial distribution characteristics of the internal wave crest lines in the remote sensing image and the location of the buoy observation point in the target sea area, N two-dimensional sections are selected, N ≥ 3, and each two-dimensional section represents a two-dimensional numerical simulation calculation domain. On each two-dimensional section, the two-dimensional MITgcm numerical mode is used to construct an internal wave numerical model for internal wave numerical simulation calculation.

[0023] S3. Based on the results of internal wave numerical simulation, determine the generation location, generation time and generation range of the internal wave, and use the maximum time of the eastward current in the barotropic current time series diagram at the generation source Indicates the generation time of the internal wave at the generation source.

[0024] S4. Constructing an internal wave generation statistical model, specifically including steps S401 to S403.

[0025] S401. Within the calculation domain of each two-dimensional section, the internal wave numerical model constructed in step S2 is used to perform internal wave numerical simulation calculations for no less than two years.

[0026] S402. Based on the numerical simulation results, for wave A, in the first two-dimensional section, based on the Hovmöller diagram within a certain time period of the two-dimensional section, and using the Hovmöller diagram, extract the maximum value of the eastward tidal current on the east ridge 、 Corresponding tidal range And A wave The time difference between the moment spawn and the western edge of the spawn area At the same time, according to the configuration of the first two-dimensional cross-section numerical model, the stratification parameters within the same time period are extracted, including the maximum value of the buoyancy frequency , jump depth and the average height of the seabed topography in the source area Similarly, extract the corresponding second 2D section in this time period , the third two-dimensional section corresponds to ...and the Nth 2D section corresponding to , the data in the same time period form the following matrix Through this step, multiple groups of matrices of different time periods can be obtained within the simulation time range of step S401. .

[0027]

[0028] For B waves, the Hovmöller diagram within a certain time period is used in the first two-dimensional section, and the maximum value of the eastward tidal current on the west ridge is extracted using the Hovmöller diagram. 、 Corresponding tidal range 、B wave The time difference between the moment spawn and the western edge of the spawn area , weak wave crest of the East Ridge positive pressure tidal current 、 Corresponding tidal range At the same time, according to the configuration of the first two-dimensional cross-section numerical model, the stratification parameters within the same time period are extracted, including the maximum value of the buoyancy frequency , jump depth and the average height of the seabed topography in the source area Similarly, extract the corresponding second 2D section within this time period , the third two-dimensional section corresponds to ...and the Nth 2D section corresponding to , the data in the same time period form the following matrix Through this step, multiple groups of matrices of different time periods can be obtained within the simulation time range of step S401. .

[0029] .

[0030] S403, multiple groups of matrices in different time periods Construct the A-wave generation statistical data set, input the A-wave generation statistical data set into the MLP feedforward artificial neural network, and construct the A-wave generation statistical model. The input of the A-wave generation statistical model is , the output is .

[0031] Multiple matrices of different time periods Construct a B-wave generation statistical data set, input the B-wave generation statistical data set into the MLP feedforward artificial neural network, and construct a B-wave generation statistical model. The B-wave generation statistical model input is , the output is .

[0032] Specifically, the method for constructing the A-wave arrival time error statistical model or the B-wave arrival time error statistical model includes steps S6-S7.

[0033] S6. First, based on the generation date of a certain internal wave peak line in the remote sensing image, the A wave generation statistical model and the B wave generation statistical model, k initial A wave peak lines and k initial B wave peak lines are determined; then, based on the internal wave propagation model of the initial arrival time field, k initial predicted A wave internal wave peak lines and k initial predicted B wave internal wave peak lines corresponding to the k initial A wave peak lines and the k initial B wave peak lines are determined; finally, the internal wave peak line and the k initial predicted A wave internal wave peak lines in the remote sensing image are calculated respectively. The average distance between the initial predicted internal wave crest line and k B-wave initial predicted internal wave crest lines is determined. The initial crest line corresponding to the initial predicted internal wave crest line with the smallest average distance is determined as the initial crest line corresponding to the internal wave crest line in the remote sensing image. The initial crest line information, the initial arrival time field, and the generation time of N two-dimensional sections at the Luzon Strait generation source are retained. The initial crest line information includes the initial crest line position, the generation time, and the internal wave type (A wave or B wave). The internal wave type is also the type of the internal wave crest line in the remote sensing image. S6 includes steps S601 to S605.

[0034] S601. Determine k possible generation dates of a certain internal wave peak line in the remote sensing image based on the acquisition time of the remote sensing image. Apply the A-wave generation statistical model and the B-wave generation statistical model to each generation date to obtain the time difference when the internal wave reaches the western boundary of the first to N two-dimensional section generation area, and then determine the average propagation velocity of the internal wave in the first to N two-dimensional sections within the generation area. Using the corresponding average propagation velocity of the internal wave, under the condition of a given time difference, determine the position where the internal wave reaches the first to N two-dimensional sections after propagating through the corresponding given time difference. Connect the positions where the internal wave reaches the first to N two-dimensional sections in sequence to obtain the initial peak line, and finally obtain k initial A-wave peak lines and k initial B-wave peak lines corresponding to the k generation dates.

[0035] S602, using the internal wave propagation speed-water depth function relationship, calculate the longitude and latitude coordinates of the i-th row and j-th column in the target sea area The propagation speed on , the propagation speed corresponding to all longitude and latitude coordinate points in the target sea area The collection of is the velocity field in the internal wave propagation area.

[0036] S603: For the k initial wave crest lines of wave A and wave B determined in step S601, propagate them in the target sea area using the velocity field calculated in step S602, and use the fast marching algorithm to solve the Eikonal equation to obtain the coordinate point of row i and column j of each initial wave crest line of wave A arriving at the target sea area. The earliest time And the initial crest line B reaches the coordinate point in row i and column j in the target sea area The earliest time The earliest arrival time set of all coordinate points within the target sea area is the initial arrival time field of the initial wave crest line. Based on the initial arrival time field, the A-wave propagation model and the B-wave propagation model are established, and the arrival position of the corresponding initial wave crest line at the time of remote sensing observation is calculated, resulting in k initial predicted internal wave crest lines for A wave and k initial predicted internal wave crest lines for B wave.

[0037] S604. Calculate the distance between any initial predicted internal wave peak line and the internal wave peak line in the remote sensing image, select and retain only the initial predicted internal wave peak line with the smallest average distance, and the initial peak line corresponding to the initial predicted internal wave peak line with the smallest average distance is the initial peak line corresponding to the internal wave peak line in the remote sensing image.

[0038] S605. Repeat steps S601 to S604 until the initial crest line corresponding to the internal wave crest line in all collected remote sensing images is obtained. After completion, it is necessary to retain and record the initial crest line information corresponding to the internal wave crest line in the remote sensing image, the initial arrival time field, the maximum eastward tidal current of N two-dimensional sections at the source of the Luzon Strait, the tidal range corresponding to the maximum eastward tidal current, the peak value of the barotropic tidal current before the maximum eastward tidal current, and the average value of the tidal range corresponding to the peak value of the barotropic tidal current. The initial crest line information includes the initial crest line position, the formation time, and the internal wave type. The internal wave type is also the type of the internal wave crest line in the remote sensing image.

[0039] If the internal wave crest line in the remote sensing image determined in step S604 is an A wave, the generation time of the first to N two-dimensional sections of the internal wave crest line in the remote sensing image at the generation source of the Luzon Strait is extracted. , the corresponding maximum value of the eastward tidal current , tidal range , and solve and The average value of and .

[0040] If the internal wave crest line in the remote sensing image determined in step S604 is a B wave, the generation time of the internal wave crest line in the remote sensing image at the generation source is extracted. , the corresponding maximum value of the eastward tidal current , tidal range , the maximum moment of eastward tidal current The previous weak peak of the barotropic tidal current , and the corresponding , and solve , , and The average value of , , and .

[0041] S7. Constructing a statistical model of the internal wave initial arrival time error

[0042] S7 specifically includes steps S701 to S703.

[0043] S701: From step S1, it can be obtained that the internal wave crest line in the remote sensing image is composed of several spatial points. Assume that the longitude and latitude coordinates of a spatial point on the internal wave crest line in the remote sensing image are , determine the longitude and latitude coordinates of the remote sensing image corresponding to the initial arrival time field The corresponding initial arrival time value )or ). Then calculate the longitude and latitude coordinates of the point Initial arrival time error )or ). , .

[0044] Where, is the acquisition time of the internal wave peak line in the remote sensing image, The initial peak line corresponding to the internal wave peak line in the remote sensing image is formed at the same time. The corresponding maximum value of the stratification parameter buoyancy frequency , jump depth and topographic parameters water depth .

[0045] S702: If the internal wave crest line in the remote sensing image is determined to be an A wave in step S6, then the corresponding internal wave crest line in the remote sensing image is Corresponding one-dimensional array Corresponding to the internal wave peak line in the remote sensing image and Combine to get a one-dimensional array Multiple spatial points in an A-wave crest line in a remote sensing image form a set of multidimensional arrays. Multiple A-wave crest lines can be used to obtain multiple sets of multidimensional arrays, which are the statistical data sets of the A-wave initial arrival time error.

[0046] If the internal wave crest line in the remote sensing image is determined to be a B wave in step S6, each spatial point of the internal wave crest line in the remote sensing image is positioned at The corresponding one-dimensional array Corresponding to the internal wave peak line in the remote sensing image , , and Combine to get a one-dimensional array Multiple spatial points in a B-wave crest line in a remote sensing image form a set of multidimensional arrays, and multiple B-wave crest lines generate multiple sets of multidimensional arrays, which are the statistical data sets of the B-wave initial arrival time errors.

[0047] S703, input the A wave initial arrival time error statistical data set into the MLP feedforward artificial neural network, and use the initial arrival time error The distribution law of A wave arrival time error is the learning goal, and the statistical model of A wave arrival time error is constructed. When applied, for a certain longitude and latitude coordinate point, the output of the statistical model of A wave arrival time error corresponds to the predicted initial arrival time error. .

[0048] The statistical data set of the initial arrival time error of the B wave is input into the MLP feedforward artificial neural network to obtain the initial arrival time error. The distribution law of the B-wave arrival time error is the learning goal, and a statistical model of the B-wave arrival time error is constructed. When applied, for a certain longitude and latitude coordinate point, the B-wave arrival time error statistical model output corresponds to the predicted initial arrival time error .

[0049] The present invention first acquires multi-source satellite remote sensing images of the target sea area and establishes a multi-source satellite remote sensing internal wave crest line dataset. A numerical internal wave model is then constructed to determine the internal wave generation location, generation time, and generation range. Within the computational domain of each two-dimensional section, the internal wave numerical model is used to perform internal wave numerical simulation calculations for no less than two years, obtaining a potential temperature data set (hereinafter referred to as potential temperature field data) and horizontal and vertical flow field data for N two-dimensional sections over the two-year period. This data set then provides an internal wave generation statistical data set, ultimately constructing an internal wave generation statistical model and obtaining an initial internal wave crest line. On this basis, the internal wave crest line of the remote sensing image is matched with the initial internal wave crest line to obtain corresponding initial crest line information and an initial arrival time field, as well as the generation time of the N two-dimensional sections at the generation source in the Luzon Strait. Finally, a statistical model of the internal wave initial arrival time error is constructed. This model is used to correct the initial arrival time field to obtain a corrected arrival time field, which is then used to construct an internal wave propagation model and achieve prediction of the internal wave crest line.

[0050] Compared with the prior art, the present invention has the following advantages: (1) The internal wave generation statistical data set covers the key parameters that affect the internal wave generation, including the barotropic tidal current parameters ( ), stratification parameters ( 、 ) and terrain parameters (H), ensuring the accuracy of the output results of the established generation statistical model; (2) The generation time of the internal wave is represented by the maximum time of the eastward tide in the positive pressure tide time series diagram in the generation statistical model, which simplifies the determination of the generation time in the process of ocean internal wave prediction; (3) The establishment of the internal wave generation statistical model can replace the computationally complex numerical model. Under the condition that the generation date of the internal wave is known, the initial crest line position and formation time of the internal wave can be quickly and accurately obtained, thereby solving the problem that the initial crest line input of the propagation model depends on remote sensing data; (4) In the existing propagation model, only the influence of seawater depth on the propagation speed of internal waves is considered, while the influence of other factors is ignored, which leads to the initial arrival time error accumulating with the increase of propagation distance and propagation time. The establishment of the initial arrival time error statistical model aims to learn the distribution law of the initial arrival time error, and incorporates the factors affecting the internal wave propagation speed into the input feature space of the model, including terrain parameters (H), stratification parameters ( 、 ), a barotropic tidal current parameter that characterizes the influence of internal wave amplitude , characterizing the geographical location parameters including background flow field and mesoscale eddy influence , the initial arrival time value parameter that characterizes the effect of propagation time , so that the propagation model not only follows the physical laws of wavefront propagation, but also significantly improves the accuracy of prediction through data-driven methods; (5) The internal wave prediction method of the present invention can automatically call the generation statistical model to form the initial peak line of the internal wave when the internal wave generation date at the generation source is given, and the initial peak line is input to the propagation model to realize the prediction of the arrival time or arrival position of the internal wave peak line. This method has high computational efficiency and is convenient for business operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is a schematic diagram of the internal wave generation moment involved in the ocean internal wave prediction process of the present invention, the initial wave crest line predicted by the generation statistical model, the internal wave crest line predicted by the A-wave propagation model of the initial arrival time field (red curve), the internal wave crest line predicted by the A-wave propagation model of the corrected arrival time field (green curve), and the remote sensing observation internal wave crest line (blue curve).

[0052] Figure 2 Comparison and verification of the internal wave crest lines in the numerical model and remote sensing images. (a) is an Envisat / ASAR remote sensing image acquired at 14:10:22 (UTC) on August 4, 2009. (b) is a time series plot of the isothermal distribution at the intersection of the second two-dimensional section and the internal wave crest line A1 in the remote sensing image. (c) is a time series plot of the isothermal distribution at the intersection of the second two-dimensional section and the internal wave crest line B1 in the remote sensing image. (d) is a time series plot of the isothermal distribution at the intersection of the second two-dimensional section and the internal wave crest line A2 in the remote sensing image. White triangles indicate the locations of internal waves in the numerical simulation results, and the white dashed line indicates the time of remote sensing image acquisition.

[0053] Figure 3 Hovmöller diagram (a) drawn from the potential temperature data of the second two-dimensional section at the jump depth, time series of the east-west barotropic current on the east ridge (b), and the corresponding topographic profile of the second two-dimensional section (c).

[0054] Figure 4 Hovmöller diagram (a) drawn from the potential temperature data of the second two-dimensional section at the depth of the jump layer, time series diagram of the east-west barotropic current on the west ridge (b), and corresponding topographic profile of the second two-dimensional section (c).

[0055] Figure 5 is the horizontal-vertical flow field distribution diagram, u represents the horizontal flow velocity, and w represents the vertical flow velocity.

[0056] Figure 6 This is a scatter plot of the test set results of the statistical model of the initial arrival time error of the A wave.

[0057] Figure 7 This is a scatter plot of the test set results of the statistical model of the B-wave initial arrival time error.

[0058] Figure 8 This is a comparison chart of the model prediction results and remote sensing images described in the present invention.

[0059] Figure 9 The figure is a flow chart for predicting the internal wave on a given generation date based on the A-wave generation statistical model and the A-wave arrival time error statistical model of the present invention.

[0060] Figure 10 The flowchart is for predicting the internal wave on a given generation date based on the B-wave generation statistical model and the B-wave arrival time error statistical model of the present invention. DETAILED DESCRIPTION

[0061] The present invention will be further described below through specific examples.

[0062] Example 1

[0063] The ocean internal wave prediction method based on satellite remote sensing data involved in this embodiment specifically includes steps S1 to S8.

[0064] S1. Acquire multi-source satellite remote sensing images of the target sea area and establish a multi-source satellite remote sensing internal wave crest line dataset. The satellite remote sensing internal wave crest line dataset includes the latitude and longitude coordinates of the spatial points constituting each internal wave crest line and the acquisition time of each internal wave crest line. S1 specifically includes steps S101 to S102.

[0065] S101. The longitude and latitude range of the target sea area in this embodiment is 114°E to 123°E and 19°N to 23°N. In the target sea area, the interaction between the positive pressure current and the submarine ridge (the east and west ridges of the Luzon Strait) will generate disturbances, forming two different types of internal waves (baroclinic motion), A waves and B waves. The internal waves described in the present invention are A waves and B waves. According to the data from the buoy observation, the amplitude of the A wave is relatively large and the waveform is relatively narrow. It usually appears as a significant leading wave followed by several solitons; the amplitude of the B wave is relatively small and the waveform is relatively wide. It is usually a single waveform. Ocean internal waves are a small and medium-scale fluctuation phenomenon that occurs in stable density stratified seawater. During their movement and propagation, they will induce the convergence / divergence effect of the sea surface current field, thereby modulating the sea surface roughness. This modulation effect makes the ocean internal waves appear as characteristic stripes of alternating light and dark on remote sensing images. This embodiment collects 200 multi-source satellite remote sensing images of the target sea area containing internal wave characteristic fringes. The multi-source refers to data coming from multiple different independent or complementary satellite platforms, sensors or observation systems, such as at least two of ERS-1 / 2SAR, Envisat / ASAR, Radarsat-2 / SAR and GF-3 / SAR.

[0066] S102: Input the multi-source satellite remote sensing image into the remote sensing image processing platform (Envi). After geo-correction and radiometric correction, the crest lines of the internal wave characteristic fringes are extracted using image enhancement and edge detection modules. The latitude and longitude coordinates of the spatial points on the internal wave crest lines, as well as the acquisition time, are recorded to form a satellite remote sensing internal wave crest line dataset for the target sea area. The satellite remote sensing internal wave crest line dataset includes the latitude and longitude coordinates of the spatial points constituting each internal wave crest line and the acquisition time of the internal wave crest line. The geo-correction enables a one-to-one correspondence between pixels in the remote sensing image and latitude and longitude coordinates.

[0067] S2. Constructing an internal wave numerical model. S2 specifically includes steps S201 to S202.

[0068] S201. Construct an internal wave numerical model: In the target sea area, according to the spatial distribution characteristics of the internal wave crest line in the remote sensing image and the location of the buoy observation point in the target sea area, select N two-dimensional sections, N ≥ 3, and each two-dimensional section represents a two-dimensional numerical simulation calculation domain. On each two-dimensional section, use the two-dimensional MITgcm numerical mode to construct an internal wave numerical model to perform internal wave numerical simulation calculations, and output the potential temperature field data and horizontal and vertical flow field data of a certain period of time in the calculation domain.

[0069] This embodiment specifically selects three two-dimensional sections. For ease of explanation, the three two-dimensional sections are respectively the first two-dimensional section, the second two-dimensional section and the third two-dimensional section. Each two-dimensional section represents a two-dimensional numerical simulation calculation domain, and the calculation domain range corresponding to the three two-dimensional sections is 116°E to 122.3°E. On each two-dimensional section, a two-dimensional MITgcm numerical model is used to construct a numerical model for internal wave numerical simulation calculations. Specifically: the eastern boundary of the numerical model adopts a fixed boundary condition, and the positive pressure current is transmitted into the two-dimensional numerical simulation calculation domain through the eastern boundary, and the potential temperature field data and horizontal and vertical flow field data are output. The western boundary of the numerical model adopts a free boundary condition. The numerical model adopts a uniform ocean stratification condition. In order to prevent the reflection of internal waves, a 50-kilometer-wide sponge layer is set at the east and west boundaries of the numerical model to absorb the baroclinic energy.

[0070] S202, verify the internal wave numerical model: select some satellite remote sensing internal wave data with temporal and spatial representativeness to verify the internal wave numerical model, specifically: on each two-dimensional section, use the potential temperature field data obtained in step S201 to draw an isotherm distribution map (such as Figure 2 ), and compared and verified with the internal wave crest line observed by remote sensing, confirming that the numerical model constructed in step S201 has high simulation accuracy.

[0071] Taking the second 2D cross-section as an example, this example validated the numerical model for A and B waves. Statistical results show that in deepwater areas east of 118°E, the root mean square error of the arrival times of A and B waves is relatively small, at 0.77 hours. However, in continental slope waters west of 118°E, this error increases to 1.18 hours due to factors such as topography. These results demonstrate good consistency between the numerical model and remote sensing observation data, particularly demonstrating high simulation accuracy in the deepwater areas east of 118°E.

[0072] S3. Determine the generation position, generation time and generation range of the internal wave. S3 specifically includes steps S301 to S302.

[0073] S301. Use the potential temperature data at each 2D section corresponding to the thermocline depth (the thermocline depth refers to the depth with the largest vertical temperature gradient in the ocean) to draw a Hovmöller diagram, track the generation and propagation of internal waves, and accurately identify their generation location and time at the source. At the same time, obtain the positive pressure tidal current time series diagram at the generating source, and determine the maximum time of the eastward tidal current in the positive pressure tidal current time series diagram at the generating source through comparative analysis. Can be used to represent the generation time of internal waves at the generation source .

[0074] For example, Figure 3 (a) is the Hovmöller diagram drawn from the potential temperature data of the second two-dimensional cross-section jump depth. Figure 3 (b) is the time series diagram of the east-west positive pressure current on the east ridge. Figure 3 (c) in the figure is the second two-dimensional topographic profile. Figure 4 (a) is the Hovmöller diagram drawn from the potential temperature data of the second two-dimensional cross-section jump depth. Figure 4 (b) is the time series diagram of the east-west positive pressure current on the west ridge. Figure 4 (c) in the figure is the second two-dimensional cross-sectional topographic profile. It can be seen from the barotropic tidal current time series diagram that the generating source presents obvious mixed characteristics of semi-diurnal tidal components and diurnal tidal components. The two adjacent peaks of the semi-diurnal tidal component overlap with the peak and trough of the diurnal tidal component respectively, making the barotropic tidal current present a pattern of alternating strong and weak double peaks based on the semi-diurnal tidal cycle, that is, the peak of the semi-diurnal tidal component and the peak of the diurnal tidal component superimpose to form a strong peak (corresponding to the maximum value of the eastward tidal current), and the peak of the semi-diurnal tidal component and the trough of the diurnal tidal component superimpose to form a weak peak. Determine the maximum value of the eastward tidal current based on the barotropic tidal current time series diagram , when the eastward current reaches its maximum value , Corresponding tidal range ( for The difference between the previous positive pressure tidal trough and the maximum value of the eastward tidal current The previous weak peak value of the positive pressure tidal current , Corresponding tidal range ( for the difference between the trough of the previous barotropic tidal current and the trough of the previous barotropic tidal current).

[0075] Specifically, from Figure 3 (a) in the figure can determine that the generation position of the A wave on the second 2D section on August 4, 2009 is , located on the east ridge, generated at ,from Figure 3 (b) in the figure can determine that the maximum value of the eastward tidal current on the eastern ridge of the Luzon Strait is , correspond , which corresponds to the maximum time of eastward tidal current .

[0076] from Figure 4 (a) in the figure can determine that the generating position of the B wave on the second 2D section on August 4, 2009 is , located on the west ridge, generated at ,from Figure 4 (b) in the figure can determine that the maximum value of the eastward tidal current on the west ridge is , correspond , which corresponds to the maximum time of eastward tidal current .at the same time, The weak peak value of the previous East Ridge barotropic tidal current ,as well as Corresponding tidal range It can be determined that Figure 3 As shown in (b) in .

[0077] The statistical results show that: and The average time deviation was 5.8±9 minutes. and The average time deviation is 7.2±8 minutes. The tidal cycle at the source is mainly a semi-diurnal tide of 12.48 hours. For both A and B waves, the time difference between the generation time and the time when the eastward tidal current reaches its maximum value does not exceed 2% of the tidal cycle. Therefore, it is believed that the generation time of wave A is the time when the eastward tidal current reaches its maximum value on the eastern ridge. The B wave is generated at the time of the maximum value of the eastward current on the west ridge. , so as to determine the internal wave generation time more accurately and quickly.

[0078] S302: Based on the generation location determined in step S301 and the fact that it is difficult for remote sensing images to capture obvious internal wave crest lines in the area east of 120.5°E, the generation area of ​​the internal wave can be determined.

[0079] In this embodiment, the three 2D cross-sections show the A-wave generation regions at (120.5°E, 121.9°E), (120.5°E, 121.8°E), and (120.5°E, 121.8°E). The B-wave generation regions are at (120°E, 120.85°E), (120°E, 120.8°E), and (120°E, 120.8°E).

[0080] S4: Constructing an internal wave generation statistical model. S4 specifically includes steps S401 to S403.

[0081] S401. Within the calculation domain of each two-dimensional section, the internal wave numerical model constructed in step S201 is used to perform internal wave numerical simulation calculations for no less than two years to obtain a set of potential temperature data (hereinafter referred to as potential temperature field data) and horizontal and vertical flow field data for N two-dimensional sections within two years.

[0082] S402. In order to ensure the accuracy of the output results of the established generation statistical model, key parameters that affect the generation of internal waves are used in its establishment process, including barotropic tidal flow parameters, stratification parameters and terrain parameters.

[0083] The horizontal and vertical flow field data obtained in step S201 are used to draw a horizontal-vertical flow field distribution diagram ( Figure 5 ), through analysis, it is determined that the generation of B waves is not only related to the maximum value of the eastward tidal current on the west ridge, but also to the strengthening effect of the weak wave peak in the first half of the diurnal tidal cycle on the east ridge. Figure 5 The horizontal-vertical flow distribution diagram shows that both the West Ridge and the East Ridge contribute to the generation of the B wave. At t = 14 hours, when the barotropic current reached a weak peak, a significant downwelling flow was generated east of the East Ridge. This wave gradually propagated to the east of the West Ridge at t = 24 hours, reinforcing the downwelling flow generated east of the West Ridge at the time of the maximum eastward current. This, in turn, led to the formation of a distinct internal wave structure at t = 31 hours. Specifically, the horizontal flow field exhibited opposite directions between the upper and lower layers, while the vertical flow field showed alternating strong downwelling and upwelling. The A wave, on the other hand, was associated only with the maximum eastward current at the East Ridge. Therefore, the barotropic current parameters used in the statistical model for A wave generation were the maximum eastward current at the East Ridge and its corresponding tidal range, while the barotropic current parameters used in the statistical model for B wave generation were the maximum eastward current at the West Ridge and its corresponding tidal range, and the weak peak of the barotropic current at the East Ridge and its corresponding tidal range.

[0084] Specifically, for wave A, in the first two-dimensional section, the Hovmöller diagram within a certain time period of the two-dimensional section is obtained based on the numerical simulation results in step S401, and the maximum value of the eastward tidal current on the east ridge is extracted using the Hovmöller diagram. 、 Corresponding tidal range And A wave The time difference between the moment spawn and the western edge of the spawn area At the same time, according to the configuration of the first two-dimensional cross-section numerical model, the stratification parameters within the same time period are extracted, including the maximum value of the buoyancy frequency , jump depth and the average height of the seabed topography in the source area Similarly, the terrain parameters including the second two-dimensional section in this time period are extracted. , the third two-dimensional section corresponds to ...and the Nth 2D section corresponding to , the data in the same time period form the following matrix Through this step, multiple groups of matrices of different time periods can be obtained within the simulation time range of step S401. .

[0085] .

[0086] Specifically, for B waves, in the first two-dimensional section, the Hovmöller diagram within a certain time period of the two-dimensional section is obtained based on the numerical simulation results in step S401, and the maximum value of the eastward tidal current on the west ridge is extracted using the Hovmöller diagram. 、 Corresponding tidal range 、B wave The time difference between the moment spawn and the western edge of the spawn area , weak wave crest of the East Ridge positive pressure tidal current 、 Corresponding tidal range At the same time, according to the configuration of the first two-dimensional cross-section numerical model, the stratification parameters within the same time period are extracted, including the maximum value of the buoyancy frequency , jump depth and the average height of the seabed topography in the source area Similarly, extract the terrain parameters corresponding to the second two-dimensional section within this time period. , the third two-dimensional section corresponds to ...and the Nth 2D section corresponding to , the data in the same time period form the following matrix Through this step, multiple groups of matrices of different time periods can be obtained within the simulation time range of step S401. .

[0087] .

[0088] S403: The matrices of the multiple groups of different time periods Construct the A-wave generation statistical data set, input the A-wave generation statistical data set into the MLP feedforward artificial neural network, and construct the A-wave generation statistical model. The input of the A-wave generation statistical model is , the output is .

[0089] The matrices of the multiple groups of different time periods Construct a B-wave generation statistical data set, input the B-wave generation statistical data set into the MLP feedforward artificial neural network, and construct a B-wave generation statistical model. The B-wave generation statistical model input is , the output is .

[0090] Specifically, the A-wave and B-wave generated statistical data sets were divided into training, validation, and test sets at a ratio of 60%, 20%, and 20%, respectively. The training set was used for model weight adjustment, the validation set was used to monitor model overfitting and select the optimal network structure, and the test set was used to ultimately evaluate the model's generalization ability. The root mean square error of the test set was used as an objective measure of the performance of the generated statistical model.

[0091] This example uses the aforementioned method in conjunction with an MLP feedforward artificial neural network to establish statistical models for the generation of A and B waves, respectively. The results show that the root mean square error (RMS) of the A-wave generation statistical model test set is 0.23 hours, and the RMS error of the B-wave generation statistical model test set is 0.24 hours. Furthermore, the time difference between the A-wave's source and the western boundary of the generation region on different two-dimensional sections ranges from 12 hours to 16.8 hours, while the time difference between the B-wave's source and the western boundary of the generation region on different two-dimensional sections ranges from 7.2 hours to 12 hours.

[0092] According to the root mean square error results of the test set, it is shown that the generative statistical model can replace the MITgcm numerical model and quickly and accurately calculate the A-wave on the first to N two-dimensional sections of the east ridge. After the moment is generated, the time difference to the western boundary of the corresponding generation area is , and the first to N two-dimensional sections of B wave on the west ridge are respectively After the moment is generated, the time difference to the western boundary of the corresponding generation area is .

[0093] S5. Determine the location and formation time of the initial internal wave crest line: Based on the time difference between the internal wave reaching the western boundary of the first to N two-dimensional section generation area, the average internal wave propagation velocity for the first to N two-dimensional sections within the generation area can be accurately calculated. Using this average internal wave propagation velocity, given a given time difference, the location at which the internal wave reaches the first to N two-dimensional sections after propagating for the given time difference is determined. The locations at which the internal wave reaches the first to N two-dimensional sections are sequentially connected to obtain the initial crest line, and the formation time of the initial crest line is determined.

[0094] like Figure 1 For A waves, the time difference between reaching the western boundary of the region from the first to the Nth two-dimensional sections is used. Calculate the average propagation velocity of the first to N two-dimensional sections within the generated area . use , at a given time difference When the calculation is Arrival at the position of the first to N two-dimensional sections (X 1A , Y 1A )、(X 2A , Y 2A )、(X 3A , Y 3A )……(X NA , Y NA ), They represent the time difference between the moment when the A wave is generated in the first two-dimensional section and the moment when the second to N two-dimensional sections are generated, such as, is the time difference between the time when the A wave is generated in the first two-dimensional section and the time when the second two-dimensional section is generated. '-' indicates a phase lag, and '+' indicates a phase advance. 1A , Y 1A )、(X 2A , Y 2A )、(X 3A , Y 3A )……(X NA , Y NA ) are connected in sequence as the initial crest line position of the internal wave propagation model, and the formation time of the initial crest line is Theoretically, wave A reaches (X 1A , Y 1A )、(X 2A , Y 2A )、(X 3A , Y 3A )……(X NA , Y NA ), given that (X 1A , Y 1A )、(X 2A , Y 2A)、(X 3A , Y 3A )……(X NA , Y NA ) must be within the A wave generation area, so The minimum value within the range (12 hours to 16.8 hours) should be taken.

[0095] For B waves, the time difference between the arrival of the B wave at the western boundary of the first to N 2D sections is used to generate the area. Calculate the average propagation velocity of the first to N two-dimensional sections within the generated area . use , at a given time difference When Arrival at the position of the first to N two-dimensional sections (X 1B , Y 1B )、(X 2B , Y 2A )、(X 3B , Y 3B )……(X NB , Y NB ), They represent the time difference between the moment when the B wave is generated in the first two-dimensional section and the moment when the second to N two-dimensional sections are generated, such as is the time difference between the time when the A wave is generated in the first two-dimensional section and the time when the second two-dimensional section is generated. '-' indicates a phase lag, and '+' indicates a phase advance. 1B , Y 1B )、(X 2B , Y 2A )、(X 3B , Y 3B )……(X NB , Y NB ) are connected in sequence as the initial crest line position of the internal wave propagation model. The formation time of the initial crest line is Theoretically, the B wave reaches (X 1B , Y 1B )、(X 2B , Y 2A )、(X 3B , Y 3B )……(X NB , Y NB ), given that (X 1B , Y 1B )、(X 2B , Y 2A )、(X 3B , Y 3B )……(X NB , Y NB ) must be within the B wave generation area, so The minimum value within the range (7.2 hours to 12 hours) should be taken.

[0096] Specifically, the A wave is in the first two-dimensional section at Generate at the source at any time, then 1A The average velocity propagates along the first two-dimensional section toward the western boundary of the generating area. Then propagate to (X 1A , Y 1A ), similarly, the internal wave is in the second two-dimensional section Generate at the source at any time, then 2A The average velocity propagates along the second two-dimensional section toward the western boundary of the generating area. Then propagate to (X 2A , Y 2A ), the internal wave in the third two-dimensional section is Generate at the source at any time, then 3A The average velocity propagates along the third two-dimensional section toward the western boundary of the generating area. Then propagate to (X 3A , Y 3A ). And according to the result of step S4, The minimum value within the range should be taken, i.e. 12 hours. The minimum value within the range, i.e. 7.2 hours, should also be taken.

[0097] Based on the internal wave generation statistical model, the initial crest line position and formation time of the internal wave can be obtained under the condition that the generation date of the internal wave is known. Specifically, the A wave generation statistical model is used as an example to illustrate that by inputting the generation date of the internal wave at the generation source (X year X month X day), the A wave generation statistical model automatically calls N two-dimensional sections (taking three two-dimensional sections as an example) of the generation date to generate the time series data of the positive pressure tidal current at the ridge east of the source, and the seawater stratification parameter data. , topographic parameter data of the average height of the East Ridge The maximum time of the eastward current on the east ridge of each two-dimensional section on the day of the generation is automatically calculated based on the time series data of the positive pressure current on the east ridge. , as the generation time of wave A within the generation date, extract the maximum value of the eastward tidal current corresponding to each two-dimensional section and tidal range , forming a matrix. The matrix is ​​input into the A-wave generation statistical model to obtain the time difference between the A-wave generation and the western boundary of the generation area in different two-dimensional sections. Then, step S5 is performed according to the given time difference. Get the initial peak line position and the initial peak line formation time.

[0098] S6. First, based on the possible k generation dates of a certain internal wave peak line in the remote sensing image, the A wave generation statistical model and the B wave generation statistical model, k initial A wave peak lines and k initial B wave peak lines are determined; then, based on the internal wave propagation model of the initial arrival time field, k initial predicted A wave internal wave peak lines and k initial predicted B wave internal wave peak lines corresponding to the k initial A wave peak lines and the k initial B wave peak lines are determined; finally, the correlation between the internal wave peak line in the remote sensing image and the k initial predicted A wave internal wave peak lines is calculated respectively. The average distance between the B-wave crest line and k initial predicted internal wave crest lines of the B wave is determined. The initial crest line corresponding to the initial predicted internal wave crest line with the smallest average distance is determined as the initial crest line corresponding to the internal wave crest line in the remote sensing image. The initial crest line information, the initial arrival time field, and the generation time of N two-dimensional sections at the source of the Luzon Strait are retained. The initial crest line information includes the initial crest line position, the formation time, and the internal wave type (A wave or B wave). The internal wave type is also the type of the internal wave crest line in the remote sensing image. S6 specifically includes steps S601 to S605.

[0099] S601. Determine k possible generation dates of a certain internal wave peak line in the remote sensing image based on the acquisition time of the remote sensing image. Apply the A-wave generation statistical model and the B-wave generation statistical model to each generation date, and perform step S5 to obtain k initial A-wave peak lines and k initial B-wave peak lines corresponding to the k generation dates. Retain the initial peak line information and the corresponding generation date. The initial peak line information includes the initial peak line position, formation time, and internal wave type (A wave or B wave).

[0100] After forming in the generation region, the internal wave continues to propagate within the propagation region and is captured by remote sensing images, forming the internal wave crest line. Since the generation time of this internal wave crest line in the remote sensing image is unknown, it is assumed that it may have generated between 1 and k days before the remote sensing observation time. This allows us to determine k possible generation dates for the internal wave at the generation source: X year, X month, and X day. By applying the A-wave generation statistical model and the B-wave generation statistical model, respectively, combined with step S5, we obtain k possible initial A-wave crest lines and k possible initial B-wave crest lines. For example, after the internal wave is generated in the Luzon Strait generation area, it will gradually propagate westward until it dissipates. The entire evolution process takes about 4 days. Assuming that the satellite remote sensing image is collected on July 6, 2024, and k is 4, then the possible generation dates of the internal wave are July 2, 2024, July 3, 2024, July 4, 2024, and July 5, 2024. Each possible generation date corresponds to two initial wave crest lines, one A wave initial crest line and one B wave initial crest line.

[0101] S602, using the internal wave propagation velocity-water depth function relationship proposed by Jackson (2009), calculate the longitude and latitude coordinates of the i-th row and j-th column in the target sea area The propagation speed on , The value is the latitude range of the target sea area. The value is the longitude range of the target sea area. and as well as and The interval between them can be determined as needed, such as 1°, (1 / 60)°. The propagation speed corresponding to all longitude and latitude coordinate points in the target sea area The collection of is the velocity field in the internal wave propagation area.

[0102] The internal wave propagation velocity-water depth function relationship is: .

[0103] in, Indicates the internal wave at the latitude and longitude coordinates The propagation speed at Indicates the longitude and latitude coordinates of a point Water depth at =2.971 m / s; B1=0.003; B2=1390.758 m.

[0104] S603: For the k initial wave crest lines of A wave determined in step S601, propagate them in the target sea area using the velocity field calculated in step S602, and use the fast marching algorithm to solve the Eikonal equation to obtain the coordinate point of the i-th row and j-th column of each initial wave crest line of A wave arriving at the target sea area. The earliest time The earliest arrival time set for all coordinate points within the target sea area is the initial arrival time field of the initial A-wave crest line. Based on the initial arrival time field of the initial A-wave crest line, the arrival position of the corresponding initial crest line at the time of remote sensing observation is calculated, i.e., the initial predicted internal wave crest line of the A-wave. The initial predicted internal wave crest line of the A-wave is represented by the contour line of the time difference between the remote sensing observation time and the initial crest line formation time determined in step S601. This step yields k initial predicted internal wave crest lines of the A-wave.

[0105] For the k initial B-wave crest lines determined in step S601, propagate them in the target sea area using the velocity field calculated in step S602, and use the fast marching algorithm to solve the Eikonal equation to obtain the coordinate point of the i-th row and j-th column of each initial B-wave crest line in the target sea area. The earliest time The earliest arrival time set for all coordinate points within the target sea area is the initial arrival time field of the B-wave initial crest line. Based on the initial arrival time field of the B-wave initial crest line, the arrival position of the corresponding initial crest line at the time of remote sensing observation is calculated, namely the B-wave initial predicted internal wave crest line. The B-wave initial predicted internal wave crest line is represented by the contour line of the time difference between the remote sensing observation time and the initial crest line formation time determined in step S601. This step yields k B-wave initial predicted internal wave crest lines.

[0106] The fast marching algorithm solves the eikonal equation as: .

[0107] S604. Calculate the distance between any A-wave initial predicted internal wave peak line and any B-wave initial predicted internal wave peak line and the internal wave peak line in the remote sensing image (the actual internal wave peak line). Select and retain only the initial predicted internal wave peak line with the smallest average distance. The initial peak line corresponding to the initial predicted internal wave peak line with the smallest average distance is the initial peak line corresponding to the internal wave peak line in the remote sensing image. Retain the information of the initial peak line, the initial arrival time field, and the generation time of N two-dimensional sections at the generation source in the Luzon Strait. The initial peak line information includes the initial peak line position, the formation time, and the internal wave type (A wave or B wave). The internal wave type is also the type of the internal wave peak line in the remote sensing image.

[0108] For example, if the satellite remote sensing image is acquired at 12:00:00 on July 6, 2024, and k is 4, then the possible internal wave generation dates are July 2, 2024, July 3, 2024, July 4, 2024, and July 5, 2024. Taking July 5, 2024 as an example, assuming that the initial peak line of the A wave corresponding to July 5, 2024, is formed at 13:00:00 on July 5, 2024, and the generation time difference between 12:00:00 on July 6, 2024 and 13:00:00 on July 5, 2024, is 25 hours. In the corresponding time field, connect the longitude and latitude coordinates with arrival times of 25 hours to obtain a contour line with a time difference of 25 hours, which is the initial predicted internal wave peak line. Calculate the distance between the initial predicted internal wave peak line and the internal wave peak line in the satellite remote sensing image (the actual internal wave peak line). According to the above method, the distances between the initial predicted internal wave peak lines corresponding to the prediction generation dates of July 2, 2024, July 3, 2024, and July 4, 2024 and the internal wave peak lines in the satellite remote sensing image are calculated in sequence, and only the initial predicted internal wave peak lines with the smallest average distance are selected and retained.

[0109] S605. Repeat steps S601 to S604 until the initial crest lines corresponding to the internal wave crest lines in all collected remote sensing images are obtained. Upon completion, the initial crest line information corresponding to the internal wave crest lines in the remote sensing images, the initial arrival time field, and the generation time of N two-dimensional sections at the source of the Luzon Strait must be retained and recorded. The maximum eastward tidal current, the tidal range corresponding to the maximum eastward tidal current, the peak value of the barotropic tidal current before the maximum eastward tidal current, and the average value of the tidal range corresponding to the peak value of the barotropic tidal current are determined based on the generation time of each two-dimensional section.

[0110] If the internal wave crest line in the remote sensing image determined in step S604 is an A wave, the generation time of the internal wave crest line in the remote sensing image obtained in step S605 at the first to N two-dimensional sections of the Luzon Strait generation source is extracted. , the corresponding maximum value of the eastward tidal current , tidal range , and solve them separately and The average value of and .

[0111] If the internal wave crest line in the remote sensing image determined in step S604 is a B wave, the generation time of the internal wave crest line in the remote sensing image at the generation source is extracted. , the corresponding maximum value of the eastward tidal current , tidal range , the maximum moment of eastward tidal current The previous weak peak value of the positive pressure tidal current , and the corresponding , and solve them separately , , and The average value of , , and .

[0112] Based on the above explanation, it can be understood that the velocity field and initial arrival time field can be used to predict the internal wave crest line formed during the remote sensing observation time. However, due to the large error in the initial arrival time field, the prediction accuracy is low. Therefore, this application introduces an initial arrival time error statistical model to correct the initial arrival time field to improve the accuracy of the prediction.

[0113] S7. Constructing a statistical model of the internal wave initial arrival time error, specifically including steps S701 to S703.

[0114] S701: From step S1, it can be obtained that the internal wave crest line in the remote sensing image is composed of several spatial points. Assume that the longitude and latitude coordinates of a spatial point on the internal wave crest line in the remote sensing image are , according to the initial arrival time field corresponding to the internal wave peak line in the remote sensing image, determine the longitude and latitude coordinates of the point The corresponding initial arrival time value )or ). Then calculate the latitude and longitude coordinates of the point according to the following formula Initial arrival time error )or ). , .

[0115] Where, is the acquisition time of the internal wave peak line in the remote sensing image, The initial peak line corresponding to the internal wave peak line in the remote sensing image is formed at the same time. The corresponding maximum value of the stratification parameter buoyancy frequency , jump depth and topographic parameters water depth .

[0116] S702: If the internal wave crest line in the remote sensing image is determined to be an A wave in step S6, then the corresponding spatial point position of each internal wave crest line in the remote sensing image is Corresponding one-dimensional array Corresponding to the internal wave peak line in the remote sensing image and Combine to get a one-dimensional array Multiple spatial points in an A-wave crest line in a remote sensing image form a set of multidimensional arrays. Multiple A-wave crest lines can be used to obtain multiple sets of multidimensional arrays, which are the statistical data sets of the A-wave initial arrival time error.

[0117] If step S6 determines that the internal wave crest line in the remote sensing image is a B wave, each spatial point of the internal wave crest line in the remote sensing image is positioned at The corresponding one-dimensional array Corresponding to the internal wave peak line in the remote sensing image , , and Combine to get a one-dimensional array Multiple spatial points in a B-wave crest line in a remote sensing image form a set of multidimensional arrays, and multiple B-wave crest lines generate multiple sets of multidimensional arrays, which are the statistical data sets of the B-wave initial arrival time errors.

[0118] S703, input the A wave initial arrival time error statistical data set into the MLP feedforward artificial neural network, and use the initial arrival time error The distribution law of A wave is the learning target, and the statistical model of the arrival time error of wave A is constructed. During the construction process, the input of the statistical model of the initial arrival error of wave A is multiple groups. , the output is the corresponding When applied, for a certain latitude and longitude coordinate point, the A wave arrival time error statistical model output corresponds to the predicted initial arrival time error .

[0119] The statistical data set of the initial arrival time error of the B wave is input into the MLP feedforward artificial neural network to obtain the initial arrival time error. The distribution law of is the learning goal, and the B-wave arrival time error statistical model is constructed. During the construction process, the B-wave initial arrival error statistical model input is multiple groups , the output is the corresponding When applied, for a certain latitude and longitude coordinate point, the B-wave arrival time error statistical model output corresponds to the predicted initial arrival time error .

[0120] Specifically, the statistical data sets of the initial arrival time errors of wave A and wave B are divided into training set, validation set and test set at a ratio of 60%, 20% and 20% respectively. The training set is used for model weight adjustment, the validation set is used for model overfitting monitoring and screening the optimal network structure, and the test set is used for the final evaluation of the generalization ability of the model. The root mean square error of the test set is used as an objective measure of the performance of the initial arrival time error model.

[0121] This embodiment uses the above method in combination with an MLP feedforward artificial neural network to establish statistical models of the initial arrival time errors of the A wave and the B wave respectively. Figure 6 and Figure 7 These are scatter plots based on the test set results of the statistical model of the initial arrival time error of waves A and B, respectively. The horizontal axis represents the latitude coordinate point. The true initial arrival time error or , the vertical axis represents the longitude and latitude coordinates of the point Predicted initial arrival time error or According to statistics, the root mean square error of the A-wave initial arrival time error statistical model test set is 1.80 hours, and the root mean square error of the B-wave initial arrival time error statistical model test set is 1.96 hours.

[0122] S8. Constructing an internal wave propagation model based on the modified arrival time field. Specifically, the internal wave propagation model includes: for the longitude and latitude coordinate point in the i-th row and j-th column in the target sea area, , calculate the initial arrival time value corresponding to a certain initial peak line according to step S602 and step S603 and Based on the statistical model of the initial arrival time error of wave A and wave B constructed in step S7, the coordinate point can be output. The predicted initial arrival time error and , and thus obtain the corrected arrival time and , The set of corrected arrival times for all coordinate points within the target sea area is the corrected arrival time field for the initial crest line of the internal wave. Based on this, the A-wave propagation model and the B-wave propagation model can be constructed based on the corrected arrival time field. Based on the corrected arrival time field of the internal wave propagation model, the internal wave crest line can be predicted. Specifically, given the propagation time, the arrival position of the internal wave crest line can be predicted by drawing propagation time contours. Alternatively, given the arrival position of the internal wave crest line, the arrival time of the internal wave crest line can be predicted by accessing the corrected arrival time value corresponding to the arrival position.

[0123] For the test set divided in step S703, the distribution of the initial arrival time error and its statistical model prediction output value is as follows: Figure 6 As shown in the figure. For the test data set, statistics show that in the initial prediction, the root mean square error (RMS) of the A-wave propagation model based on the initial arrival time field is 2.31 hours, and the root mean square error of the B-wave propagation model based on the initial arrival time field is 2.81 hours. After the initial arrival error statistical model learns the distribution of the initial arrival error, the RMS error of the A-wave propagation model based on the revised arrival time field is reduced to 1.80 hours, and the RMS error of the B-wave propagation model based on the revised arrival time field is reduced to 1.96 hours. The former improves by 22%, and the latter improves by 30%. Taking both internal wave types into consideration, the RMS error of the internal wave propagation model based on the initial arrival time field is 2.48 hours in the initial prediction. After the initial arrival error statistical model learns the distribution of the initial arrival error, the RMS error of the internal wave propagation model based on the revised arrival time field is reduced to 1.85 hours, an overall improvement of 25%.

[0124] Figure 8 This is a remote sensing image acquired by the Envisat / ASAR satellite at 02:05:04 (UTC) on March 9, 2004. The black curve represents the prediction result of the model of the present invention at the time of remote sensing observation. Figure 8As shown, the root mean square errors between the predicted positions of spatial points on the A3, A4, and B2 crest lines are 3.40 km, 5.75 km, and 4.12 km, respectively. Comparative analysis of the predicted results with the internal wave crest lines in remote sensing images shows that the invention described in this application can accurately predict the arrival positions of internal wave crest lines, and the geometric morphological features and distribution spacing characteristics of the internal wave crest lines are highly consistent with those in remote sensing images.

[0125] In summary, the A-wave generation statistical model, the B-wave generation statistical model, the A-wave arrival time error statistical model, and the B-wave arrival time error statistical model are constructed to complete the prediction of the internal wave on a given generation date. The specific process is steps (1) to (6).

[0126] like Figure 9 and 10 As shown in the figure, (1) the internal wave generation date is input into the A wave generation statistical model, and the output A wave is generated on the first to N two-dimensional sections, respectively, from the generation source Luzon Strait East Ridge The time difference between reaching the western boundary of the corresponding spawning area after the moment is generated , and the corresponding maximum eastward tidal current , the maximum eastward tidal current corresponds to the tidal range , and solve them separately and The average value of and .

[0127] The internal wave generation date is input into the B-wave generation statistical model, and the output B-wave is generated on the first to Nth two-dimensional sections, respectively, from the generation source Luzon Strait West Ridge The time difference between reaching the western boundary of the corresponding spawning area after the moment is generated . And get the corresponding maximum value of eastward tidal current , the maximum eastward tidal current corresponds to the tidal range , the maximum moment of eastward tidal current The previous weak peak value of the positive pressure tidal current , and the weak wave peak corresponds to the tidal range , and solve them separately , , and The average value of , , and .

[0128] (2) Based on the time difference of the A wave reaching the western boundary of the first to N two-dimensional section generation area It can accurately calculate the average propagation velocity of the internal wave in the first to N two-dimensional sections within the A wave generation area, and use the average propagation velocity of the internal wave to determine the time difference between the A wave and the Nth two-dimensional section. After propagation, it reaches the position of the first to N two-dimensional sections, connects the positions of the first to N two-dimensional sections in sequence to obtain the initial peak line of wave A, and determines the formation time of the initial peak line of wave A .

[0129] Based on the time difference of B waves reaching the western boundary of the area generated from the first to N 2D sections It can accurately calculate the average propagation velocity of the internal wave in the first to N two-dimensional sections within the B-wave generation area, and use the average propagation velocity of the internal wave to determine the time difference between the B wave and the given time difference. After propagation, it reaches the position of the first to N two-dimensional sections, connects the positions of the first to N two-dimensional sections in sequence to obtain the initial peak line of the B wave, and determines the formation time of the initial peak line of the B wave .

[0130] (3) Using the internal wave propagation velocity-water depth function relationship, calculate the longitude and latitude coordinates of the i-th row and j-th column in the target sea area The propagation speed on , the propagation speed corresponding to all longitude and latitude coordinate points in the target sea area The set of is the velocity field in the internal wave propagation area. The internal wave propagation velocity-water depth function relationship is: .

[0131] in, Indicates the internal wave at the latitude and longitude coordinates The propagation speed at Indicates the longitude and latitude coordinates of a point Water depth at =2.971 m / s; B1=0.003; B2=1390.758 m.

[0132] (4) For the initial crest line of wave A determined in step (2), propagate it in the target sea area with the velocity field calculated in step (3), and use the fast marching algorithm to solve the Eikonal equation to obtain the coordinate point of the initial crest line of wave A in row i and column j in the target sea area. The earliest time .

[0133] For the initial crest line of the B wave determined in step (2), propagate in the target sea area with the velocity field calculated in step (3), and use the fast marching algorithm to solve the Eikonal equation to obtain the coordinate point of the initial crest line of the B wave at the i-th row and j-th column in the target sea area. The earliest time .

[0134] At the same time, get the coordinate point of row i and column j in the target domain The corresponding maximum buoyancy frequency , jump depth , seabed topography height .

[0135] (5) Input A wave arrival time error statistical model output , coordinate point Corrected arrival time The corrected arrival time set of all coordinate points in the target sea area obtained through this step is the corrected arrival time field of the initial crest line of wave A.

[0136] Will Input B-wave arrival time error statistical model output , coordinate point Corrected arrival time The corrected arrival time set of all coordinate points in the target sea area obtained through this step is the corrected arrival time field of the initial crest line of the B wave.

[0137] (6) Based on the corrected arrival time field of the initial wave crest line of wave A and the corrected arrival time field of the initial wave crest line of wave B, the arrival positions of the initial wave crest line of wave A and the initial wave crest line of wave B in step (2) at a given propagation time, or the arrival time at a given position, are calculated respectively, thus realizing the prediction of the internal wave crest line.

Claims

1. A method for predicting ocean internal waves based on satellite remote sensing data, characterized in that: The specific steps include: (1) Input the internal wave generation date into the A-wave generation statistical model, and output the A-wave on the first to Nth two-dimensional sections, respectively, from the generation source Luzon Strait East Ridge The time difference between reaching the western boundary of the corresponding spawning area after the moment is generated , and the corresponding maximum eastward tidal current , the maximum eastward tidal current corresponds to the tidal range , and solve them separately and The average value of and ; The internal wave generation date is input into the B-wave generation statistical model, and the output B-wave is generated on the first to Nth two-dimensional sections, respectively, from the generation source Luzon Strait West Ridge The time difference between reaching the western boundary of the corresponding spawning area after the moment is generated , and obtain the corresponding maximum value of the eastward tidal current , the maximum eastward tidal current corresponds to the tidal range , the maximum moment of eastward tidal current The previous weak peak value of the positive pressure tidal current , and the weak wave peak corresponds to the tidal range , and solve them separately , , and The average value of , , and ; (2) Based on the time difference of the A wave reaching the western boundary of the first to N two-dimensional section generation area It can accurately calculate the average propagation velocity of the internal wave in the first to N two-dimensional sections within the A wave generation area, and use the average propagation velocity of the internal wave to determine the time difference between the A wave and the Nth two-dimensional section. After propagation, it reaches the position of the first to N two-dimensional sections, connects the positions of the first to N two-dimensional sections in sequence to obtain the initial peak line of wave A, and determines the formation time of the initial peak line of wave A ; Based on the time difference of B waves reaching the western boundary of the area generated from the first to N 2D sections It can accurately calculate the average propagation velocity of the internal wave in the first to N two-dimensional sections within the B-wave generation area, and use the average propagation velocity of the internal wave to determine the time difference between the B wave and the given time difference. After propagation, it reaches the position of the first to N two-dimensional sections, connects the positions of the first to N two-dimensional sections in sequence to obtain the initial peak line of the B wave, and determines the formation time of the initial peak line of the B wave ; (3) Using the internal wave propagation velocity-water depth function relationship, calculate the longitude and latitude coordinates of the i-th row and j-th column in the target sea area The propagation speed on , the propagation speed corresponding to all longitude and latitude coordinate points in the target sea area The set of is the velocity field in the internal wave propagation area, and the relationship between internal wave propagation velocity and water depth function is: ; in, Indicates the internal wave at the latitude and longitude coordinates The propagation speed at Indicates the longitude and latitude coordinates of a point Water depth at =2.971 m / s; B1=0.003; B2=1390.758 m; (4) For the initial crest line of wave A determined in step (2), propagate it in the target sea area with the velocity field calculated in step (3), and use the fast marching algorithm to solve the Eikonal equation to obtain the coordinate point of the initial crest line of wave A in row i and column j in the target sea area. The earliest time ; For the initial crest line of the B wave determined in step (2), propagate in the target sea area with the velocity field calculated in step (3), and use the fast marching algorithm to solve the Eikonal equation to obtain the coordinate point of the initial crest line of the B wave at the i-th row and j-th column in the target sea area. The earliest time ; At the same time, obtain the coordinate point of row i and column j in the target sea area The corresponding maximum buoyancy frequency , jump depth , seabed topography height ; (5) Input A wave arrival time error statistical model, output , coordinate point Corrected arrival time ,The corrected arrival time set of all coordinate points in the target sea area obtained through this step is the corrected arrival time field of the initial crest line of wave A; Will Input B wave arrival time error statistical model, output , coordinate point Corrected arrival time ,The corrected arrival time set of all coordinate points in the target sea area obtained through this step is the corrected arrival time field of the initial crest line of the B wave; (6) Based on the corrected arrival time field of the initial wave crest line of wave A and the corrected arrival time field of the initial wave crest line of wave B, the arrival positions of the initial wave crest line of wave A and the initial wave crest line of wave B in step (2) at a given propagation time, or the arrival time at a given position, are calculated, that is, the prediction of the internal wave crest line is achieved; The internal waves are A waves and / or B waves, and the target sea area has a longitude and latitude range of 114°E to 123°E and 19°N to 23°N.

2. The method for predicting ocean internal waves based on satellite remote sensing data according to claim 1, characterized in that: The construction method of the A-wave generation statistical model and the B-wave generation statistical model includes S1 to S4; S1. Acquire multi-source satellite remote sensing images of the target sea area and establish a multi-source satellite remote sensing internal wave crest line dataset; S2. In the target sea area, based on the spatial distribution characteristics of the internal wave crest lines in the remote sensing image and the location of the buoy observation point in the target sea area, select N two-dimensional sections (N ≥ 3), each of which represents a two-dimensional numerical simulation domain. On each two-dimensional section, use the two-dimensional MITgcm numerical model to construct an internal wave numerical model for internal wave numerical simulation calculation; S3. Based on the results of internal wave numerical simulation, determine the generation location, generation time and generation range of the internal wave, and use the maximum time of the eastward current in the barotropic current time series diagram at the generation source represents the generation time of the internal wave at the generation source; S4, constructing an internal wave generation statistical model, specifically including steps S401 to S403; S401, within the calculation domain of each two-dimensional section, perform internal wave numerical simulation calculations for no less than two years using the internal wave numerical model constructed in step S2; S402. Based on the numerical simulation results, for wave A, in the first two-dimensional section, based on the Hovmöller diagram within a certain time period of the two-dimensional section, and using the Hovmöller diagram, extract the maximum value of the eastward tidal current on the east ridge 、 Corresponding tidal range And A wave The time difference between the moment spawn and the western edge of the spawn area At the same time, according to the configuration of the first two-dimensional cross-section numerical model, the stratification parameters within the same time period are extracted, including the maximum value of the buoyancy frequency , jump depth and the average height of the seabed topography in the source area Similarly, extract the corresponding second 2D section in this time period , the third two-dimensional section corresponds to ...and the Nth 2D section corresponding to , the data in the same time period form the following matrix Through this step, multiple groups of matrices of different time periods can be obtained within the simulation time range of step S401. ; ； For B waves, the Hovmöller diagram within a certain time period is used in the first two-dimensional section, and the maximum value of the eastward tidal current on the west ridge is extracted using the Hovmöller diagram. 、 Corresponding tidal range 、B wave The time difference between the moment spawn and the western edge of the spawn area , weak wave crest of the East Ridge positive pressure tidal current 、 Corresponding tidal range At the same time, according to the configuration of the first two-dimensional cross-section numerical model, the stratification parameters within the same time period are extracted, including the maximum value of the buoyancy frequency , jump depth and the average height of the seabed topography in the source area Similarly, extract the second 2D section corresponding to the time period , the third two-dimensional section corresponds to ...and the Nth 2D section corresponding to , the data in the same time period form the following matrix Through this step, multiple groups of matrices of different time periods can be obtained within the simulation time range of step S401. ; ； S403, multiple groups of matrices in different time periods Construct the A-wave generation statistical data set, input the A-wave generation statistical data set into the MLP feedforward artificial neural network, and construct the A-wave generation statistical model. The input of the A-wave generation statistical model is , the output is ; Multiple matrices of different time periods Construct a B-wave generation statistical data set, input the B-wave generation statistical data set into the MLP feedforward artificial neural network, and construct a B-wave generation statistical model. The B-wave generation statistical model input is , the output is .

3. The method for predicting ocean internal waves based on satellite remote sensing data according to claim 1, characterized in that: The method for constructing the A-wave arrival time error statistical model or the B-wave arrival time error statistical model includes steps S6-S7. S6. First, based on the generation date of a certain internal wave peak line in the remote sensing image, the A wave generation statistical model, and the B wave generation statistical model, k initial A wave peak lines and k initial B wave peak lines are determined; Then, based on the internal wave propagation model of the initial arrival time field, k A-wave initial predicted internal wave peak lines and k B-wave initial predicted internal wave peak lines corresponding to the k A-wave initial wave peak lines and the k B-wave initial predicted internal wave peak lines are determined; finally, the average distances between the internal wave peak line and the k A-wave initial predicted internal wave peak lines and the k B-wave initial predicted internal wave peak lines in the remote sensing image are calculated respectively, and the initial peak line corresponding to the initial predicted internal wave peak line with the smallest average distance is determined to be the initial peak line corresponding to the internal wave peak line in the remote sensing image, and the initial peak line information, the initial arrival time field and the generation time of N two-dimensional sections at the generation source of the Luzon Strait are retained. The initial peak line information includes the initial peak line position, the formation time and the internal wave type, and the internal wave type is also the type of the internal wave peak line in the remote sensing image. S6 includes steps S601 to S605; S601. Determine k possible generation dates of a certain internal wave peak line in the remote sensing image based on the acquisition time of the remote sensing image. For each generation date, apply the A-wave generation statistical model and the B-wave generation statistical model to obtain the time difference when the internal wave reaches the western boundary of the generation area of ​​the first to N two-dimensional sections. Then determine the average propagation velocity of the internal wave in the first to N two-dimensional sections within the generation area. Using the corresponding average propagation velocity of the internal wave, determine the position at which the internal wave reaches the first to N two-dimensional sections after propagating through the corresponding given time difference, and sequentially connect the positions at which the internal wave reaches the first to N two-dimensional sections to obtain an initial peak line. Ultimately, obtain k initial A-wave peak lines and k initial B-wave peak lines corresponding to the k generation dates. S602, using the internal wave propagation speed-water depth function relationship, calculate the longitude and latitude coordinates of the i-th row and j-th column in the target sea area The propagation speed on , the propagation speed corresponding to all longitude and latitude coordinate points in the target sea area The collection of is the velocity field in the internal wave propagation area; S603: For the k initial wave crest lines of wave A and wave B determined in step S601, propagate them in the target sea area using the velocity field calculated in step S602, and use the fast marching algorithm to solve the Eikonal equation to obtain the coordinate point of row i and column j of each initial wave crest line of wave A arriving at the target sea area. The earliest time And the initial crest line B reaches the coordinate point in row i and column j in the target sea area The earliest time , the earliest arrival time set of all coordinate points in the target sea area is the initial arrival time field of the initial wave crest line. Based on the initial arrival time field, the A-wave propagation model and the B-wave propagation model are established, and the arrival position of the corresponding initial wave crest line at the time of remote sensing observation is calculated, and k initial predicted internal wave crest lines of A wave and k initial predicted internal wave crest lines of B wave are obtained; S604, calculating the distance between any initial predicted internal wave peak line and the internal wave peak line in the remote sensing image, selecting and retaining only the initial predicted internal wave peak line with the smallest average distance, and the initial peak line corresponding to the initial predicted internal wave peak line with the smallest average distance is the initial peak line corresponding to the internal wave peak line in the remote sensing image; S605. Repeat steps S601 to S604 until the initial crest line corresponding to the internal wave crest line in all collected remote sensing images is obtained. After completion, it is necessary to retain and record the initial crest line information corresponding to the internal wave crest line in the remote sensing image, the initial arrival time field, and the maximum eastward tidal current of N two-dimensional sections at the generation source of the Luzon Strait, the tidal range corresponding to the maximum eastward tidal current, the peak value of the barotropic tidal current before the maximum eastward tidal current, and the average value of the tidal range corresponding to the peak value of the barotropic tidal current. The initial crest line information includes the initial crest line position, the formation time, and the internal wave type. The internal wave type is also the type of the internal wave crest line in the remote sensing image. If the internal wave crest line in the remote sensing image determined in step S604 is an A wave, the generation time of the first to N two-dimensional sections of the internal wave crest line in the remote sensing image at the generation source of the Luzon Strait is extracted. , the corresponding maximum value of the eastward tidal current , tidal range , and solve and The average value of and ; If the internal wave crest line in the remote sensing image determined in step S604 is a B wave, the generation time of the internal wave crest line in the remote sensing image at the generation source is extracted. , the corresponding maximum value of the eastward tidal current , tidal range , the maximum moment of eastward tidal current The previous weak peak of the barotropic tidal current , and the corresponding , and solve , , and The average value of , , and ; S7, constructing a statistical model of internal wave initial arrival time error, S7 specifically includes steps S701 to S703; S701: From step S1, it can be obtained that the internal wave crest line in the remote sensing image is composed of several spatial points. Assume that the longitude and latitude coordinates of a spatial point on the internal wave crest line in the remote sensing image are , determine the longitude and latitude coordinates of the remote sensing image corresponding to the initial arrival time field The corresponding initial arrival time value )or ), and then calculate the longitude and latitude coordinates of the point Initial arrival time error )or ), , ; Where, is the acquisition time of the internal wave peak line in the remote sensing image, The initial peak line corresponding to the internal wave peak line in the remote sensing image is formed at the same time, and the longitude and latitude coordinates are determined. The corresponding maximum value of the stratification parameter buoyancy frequency , jump depth and topographic parameters water depth ; S702: If the internal wave crest line in the remote sensing image is determined to be an A wave in step S6, then the corresponding internal wave crest line in the remote sensing image is Corresponding one-dimensional array Corresponding to the internal wave peak line in the remote sensing image and Combine to get a one-dimensional array , multiple spatial points in a wave A crest line in the remote sensing image form a set of multidimensional arrays, and multiple wave A crest lines obtain multiple sets of multidimensional arrays, which are the statistical data sets of the initial arrival time error of the wave A; If the internal wave crest line in the remote sensing image is determined to be a B wave in step S6, each spatial point of the internal wave crest line in the remote sensing image is positioned at The corresponding one-dimensional array Corresponding to the internal wave peak line in the remote sensing image , , and Combine to get a one-dimensional array , multiple spatial points in a B-wave crest line in the remote sensing image form a set of multidimensional arrays, and multiple B-wave crest lines obtain multiple sets of multidimensional arrays, which are the statistical data sets of the B-wave initial arrival time error; S703, input the A wave initial arrival time error statistical data set into the MLP feedforward artificial neural network, and use the initial arrival time error The distribution law of A wave arrival time error is the learning goal, and the statistical model of A wave arrival time error is constructed. When applied, for a certain longitude and latitude coordinate point, the output of the statistical model of A wave arrival time error corresponds to the predicted initial arrival time error. ; The statistical data set of the initial arrival time error of the B wave is input into the MLP feedforward artificial neural network to obtain the initial arrival time error. The distribution law of the B-wave arrival time error is the learning goal, and a statistical model of the B-wave arrival time error is constructed. When applied, for a certain longitude and latitude coordinate point, the B-wave arrival time error statistical model output corresponds to the predicted initial arrival time error .

Images