Ocean current simulation method and system based on three-dimensional temperature-salinity value

Abstract

The invention relates to the technical field of data processing, and discloses an ocean current simulation method and system based on a three-dimensional temperature-salinity value. The method comprises the following steps: carrying out line calculation and adaptive mesh refinement processing on a three-dimensional temperature and salt field simulation model to obtain a model operation scheme; performing three-dimensional visualization processing on the three-dimensional temperature and salt field initial data set and the model operation scheme to obtain dynamic three-dimensional temperature and salt field visualization data, and analyzing the dynamic three-dimensional temperature and salt field visualization data to obtain three-dimensional structure data and evolution process data of the key ocean phenomenon; performing data assimilation processing on the multi-source real-time observation data, the three-dimensional structure data and the evolution process data to obtain a fusion data set; and performing ocean current prediction calculation on the fused data set to obtain a three-dimensional ocean current prediction field, and performing uncertainty quantitative analysis on the three-dimensional ocean current prediction field to obtain ocean current simulation probability distribution data. According to the invention, the accuracy and timeliness of ocean current simulation based on the three-dimensional temperature-salinity value are improved.

Description

Technical Field

[0001] The present application relates to the field of data processing, and in particular to a method and system for simulating ocean currents based on three-dimensional temperature-salinity values. Background Art

[0002] Existing ocean current simulation methods are mainly based on two-dimensional or quasi-three-dimensional models, which describe the ocean environment through simplified physical equations and parameterization schemes. These models usually adopt a fixed grid structure and are solved using conventional numerical calculation methods. In terms of data acquisition, they mainly rely on satellite remote sensing and fixed-site observations to obtain temperature and salinity information at the sea surface or limited depth. The visualization of simulation results mostly uses two-dimensional plane maps or simple three-dimensional surface rendering, which makes it difficult to fully display the complex ocean dynamic processes.

[0003] However, these traditional methods have limitations when dealing with complex three-dimensional ocean environments. First, simplified physical models are difficult to accurately describe the temperature-salinity stratification and mixing processes in the vertical direction. Second, fixed grid structures cannot effectively capture local-scale ocean phenomena. Furthermore, limited observational data cannot fully reflect the changes in the three-dimensional structure of the ocean. In addition, conventional numerical methods have low computational efficiency when dealing with large-scale three-dimensional problems and cannot meet the needs of real-time prediction. Finally, simple visualization methods cannot intuitively present the three-dimensional structure and dynamic evolution of ocean phenomena. Summary of the invention

[0004] The present application provides a method and system for simulating ocean currents based on three-dimensional temperature-salinity values, which are used to improve the accuracy and timeliness of ocean current simulation based on three-dimensional temperature-salinity values.

[0005] In a first aspect, the present application provides a method for simulating ocean currents based on three-dimensional temperature-salinity values, the method comprising: collecting three-dimensional temperature and salinity data of a target sea area to obtain original three-dimensional temperature-salinity data, and preprocessing the original three-dimensional temperature-salinity data to obtain a three-dimensional temperature-salinity field initial data set; based on the three-dimensional temperature-salinity field initial data set, improving a preset initial ocean model to obtain a three-dimensional temperature-salinity field simulation model; performing parallel calculation and adaptive grid refinement processing on the three-dimensional temperature-salinity field simulation model to obtain a model operation scheme; performing three-dimensional visualization processing on the three-dimensional temperature-salinity field initial data set and the model operation scheme to obtain dynamic three-dimensional temperature-salinity field visualization data, and analyzing the dynamic three-dimensional temperature-salinity field visualization data to obtain three-dimensional structural data and evolution process data of key ocean phenomena; performing data assimilation processing on multi-source real-time observation data collected in real time, the three-dimensional structural data and the evolution process data to obtain a fused data set; performing ocean current prediction calculation on the fused data set to obtain a three-dimensional ocean current prediction field, and performing uncertainty quantification analysis on the three-dimensional ocean current prediction field to obtain ocean current simulation probability distribution data.

[0006] In combination with the first aspect, in a first implementation of the first aspect of the present application, the three-dimensional temperature and salinity data of the target sea area are collected to obtain original three-dimensional temperature and salinity data, and the original three-dimensional temperature and salinity data are preprocessed to obtain an initial data set of a three-dimensional temperature and salinity field, including: multi-point and multi-level sampling of the target sea area to obtain discrete temperature and salinity measurement data; spatially reconstructing the discrete temperature and salinity measurement data through a three-dimensional scattered point interpolation algorithm to obtain an initial continuous temperature and salinity field; Gaussian filtering the initial continuous temperature and salinity field to obtain a smoothed temperature and salinity distribution; and performing Fourier transform on the smoothed temperature and salinity distribution. Spectral analysis is performed to obtain frequency domain features; band-pass filtering is performed on the frequency domain features to obtain temperature-salinity change information of a specific scale; the temperature-salinity change information of a specific scale is converted back to the spatial domain by inverse Fourier transform to obtain a multi-scale temperature-salinity field; outlier detection is performed on the multi-scale temperature-salinity field to obtain a temperature-salinity anomaly point set; the temperature-salinity anomaly point set is corrected by a local regression algorithm to obtain a corrected temperature-salinity field; vertical profile analysis is performed on the corrected temperature-salinity field to obtain thermocline and halocline information; the corrected temperature-salinity field is converted into a density field by a density calculation formula to obtain a three-dimensional temperature-salinity field initial data set.

[0007] In combination with the first aspect, in a second implementation of the first aspect of the present application, based on the three-dimensional temperature-salinity field initial data set, the preset initial ocean model is improved to obtain a three-dimensional temperature-salinity field simulation model, including: performing spatiotemporal correlation analysis on the three-dimensional temperature-salinity field initial data set to obtain temperature-salinity field change characteristics; performing multi-scale decomposition on the temperature-salinity field change characteristics through empirical mode decomposition to obtain an intrinsic mode function set; performing Hilbert transform on the intrinsic mode function set to obtain instantaneous frequency and amplitude information; based on the instantaneous frequency and amplitude information, constructing a temperature-salinity field dynamic characteristic matrix to obtain a nonlinear The invention discloses a method for realizing a three-dimensional temperature-salinity field simulation model, wherein the hybrid dynamics model is optimized by adaptive particle filtering and the hybrid dynamics model is obtained. The method comprises the following steps: performing singular spectrum analysis on the nonlinear characteristic representation to obtain the main change mode; fitting the main change mode by support vector regression to obtain the temperature-salinity field prediction function; performing variational discretization on the control equation of the preset initial ocean model to obtain a discrete model framework; embedding the temperature-salinity field prediction function into the discrete model framework to obtain a hybrid dynamics model; optimizing the parameters of the hybrid dynamics model by adaptive particle filtering to obtain optimized model parameters; calibrating the hybrid dynamics model based on the optimized model parameters to obtain a three-dimensional temperature-salinity field simulation model.

[0008] In combination with the first aspect, in a third implementation of the first aspect of the present application, the three-dimensional temperature-salt field simulation model is parallel calculated and adaptively refined to obtain a model operation scheme, including: performing spatial domain decomposition on the three-dimensional temperature-salt field simulation model to obtain sub-region data blocks; performing task assignment on the sub-region data blocks through a dynamic load balancing algorithm to obtain a parallel calculation task list; performing data dependency analysis on the parallel calculation task list to obtain a task execution graph; based on the task execution graph, scheduling computing resources through a critical path method to obtain a parallel execution scheme; monitoring gradient changes during the calculation process to obtain a grid refinement demand index; processing the grid refinement demand index through an adaptive multi-scale interpolation algorithm to obtain a dynamic grid structure; performing boundary condition matching on the dynamic grid structure to obtain a continuous grid layout; based on the continuous grid layout, discretizing the calculation equation through a spectral element method to obtain a high-order numerical format; performing stability analysis on the high-order numerical format to obtain a time step constraint; processing the time step constraint through an adaptive time stepping algorithm to obtain a model operation scheme.

[0009] In combination with the first aspect, in a fourth implementation of the first aspect of the present application, the three-dimensional temperature-salinity field initial data set and the model operation scheme are subjected to three-dimensional visualization processing to obtain dynamic three-dimensional temperature-salinity field visualization data, and the dynamic three-dimensional temperature-salinity field visualization data are analyzed to obtain three-dimensional structural data and evolution process data of key ocean phenomena, including: performing data interpolation on the three-dimensional temperature-salinity field initial data set to obtain a high-resolution temperature-salinity field grid; rendering the high-resolution temperature-salinity field grid through a volume rendering algorithm to obtain a three-dimensional temperature-salinity field volume image; performing color mapping on the three-dimensional temperature-salinity field volume image to obtain a color-enhanced visualization effect; based on the model operation scheme, the color-enhanced The visualization effect is used to generate a time-series animation to obtain dynamic three-dimensional temperature-salinity field visualization data; the dynamic three-dimensional temperature-salinity field visualization data is processed by an isosurface extraction algorithm to obtain an isosurface sequence of key ocean phenomena; the isosurface sequence is subjected to morphological analysis to obtain three-dimensional structural characteristic parameters; based on the three-dimensional structural characteristic parameters, the key ocean phenomena are classified by cluster analysis to obtain a phenomenon type division; a time series analysis is performed on the phenomenon type division to obtain an evolution process characteristic vector; the evolution process characteristic vector is reduced in dimension by principal component analysis to obtain a key evolution mode; the key evolution mode is subjected to spatiotemporal interpolation to obtain three-dimensional structural data and evolution process data of the key ocean phenomenon.

[0010] In combination with the first aspect, in the fifth implementation method of the first aspect of the present application, the multi-source real-time observation data, the three-dimensional structure data and the evolution process data collected in real time are subjected to data assimilation processing to obtain a fused data set, including: performing data quality control on the multi-source real-time observation data to obtain cleaned observation data; performing spatiotemporal interpolation on the cleaned observation data to obtain a gridded observation field; performing covariance analysis on the gridded observation field and the three-dimensional structure data to obtain a data correlation matrix; based on the data correlation matrix, assimilating the data through a collective Kalman filter algorithm to obtain preliminary fused data; performing a time series consistency check on the preliminary fused data and the evolution process data to obtain corrected fused data; performing constrained optimization on the corrected fused data through a variational analysis method to obtain physical consistency data; performing scale decomposition on the physical consistency data to obtain a multi-scale feature field; performing denoising on the multi-scale feature field through wavelet transform to obtain a smoothed feature field; performing spectral analysis on the smoothed feature field to obtain the main frequency components; based on the main frequency components, reconstructing the data through an adaptive spectrum synthesis method to obtain a fused data set.

[0011] In combination with the first aspect, in a sixth implementation of the first aspect of the present application, the fused data set is subjected to ocean current prediction calculation to obtain a three-dimensional ocean current prediction field, and the three-dimensional ocean current prediction field is subjected to uncertainty quantification analysis to obtain ocean current simulation probability distribution data, including: performing spatiotemporal decomposition on the fused data set to obtain multi-scale ocean dynamic characteristics; performing time series prediction on the multi-scale ocean dynamic characteristics through a long short-term memory network to obtain preliminary ocean current prediction results; performing physical constraint correction on the preliminary ocean current prediction results to obtain a physically consistent prediction field; and generating a multi-scale ocean current prediction field through a Monte Carlo method based on the physically consistent prediction field. The method comprises the following steps: forming a disturbance field of a group to obtain a set prediction member; performing statistical analysis on the set prediction member to obtain a prediction mean field and a variance field; performing probability distribution fitting on the prediction mean field and the variance field by a kernel density estimation method to obtain a preliminary probability distribution model; performing bias correction on the preliminary probability distribution model to obtain a revised probability distribution; based on the revised probability distribution, calculating a prediction confidence interval by a Bayesian inference method to obtain an uncertainty quantification result; performing spatial interpolation on the uncertainty quantification result to obtain a continuous uncertainty field; and grading the continuous uncertainty field by an adaptive threshold method to obtain ocean current simulation probability distribution data.

[0012] In a second aspect, the present application provides a current simulation system based on three-dimensional temperature-salinity values, and the current simulation system based on three-dimensional temperature-salinity values ​​comprises:

[0013] The acquisition module is used to collect three-dimensional temperature and salinity data of the target sea area to obtain original three-dimensional temperature and salinity data, and pre-process the original three-dimensional temperature and salinity data to obtain an initial data set of the three-dimensional temperature and salinity field;

[0014] An improvement module, used to improve a preset initial ocean model based on the three-dimensional temperature-salinity field initial data set to obtain a three-dimensional temperature-salinity field simulation model;

[0015] A processing module, used for performing parallel calculation and adaptive grid refinement processing on the three-dimensional temperature-salinity field simulation model to obtain a model operation scheme;

[0016] An analysis module is used to perform three-dimensional visualization processing on the three-dimensional temperature-salinity field initial data set and the model operation scheme to obtain dynamic three-dimensional temperature-salinity field visualization data, and to analyze the dynamic three-dimensional temperature-salinity field visualization data to obtain three-dimensional structure data and evolution process data of key ocean phenomena;

[0017] An assimilation module is used to perform data assimilation processing on the multi-source real-time observation data collected in real time, the three-dimensional structure data and the evolution process data to obtain a fused data set;

[0018] The prediction module is used to perform ocean current prediction calculation on the fused data set to obtain a three-dimensional ocean current prediction field, and to perform uncertainty quantification analysis on the three-dimensional ocean current prediction field to obtain ocean current simulation probability distribution data.

[0019] In the technical solution provided by the present application, the spatial resolution and accuracy of the initial data of the temperature and salinity field are significantly improved by multi-point multi-level sampling and three-dimensional scattered point interpolation of the target sea area, providing a more reliable input for subsequent simulation. The use of technologies such as Gaussian filtering and spectrum analysis effectively removes noise and outliers in the original data and improves the data quality. The application of multi-scale decomposition and local regression algorithms enables the model to better capture marine phenomena of different scales and enhances the precision of the simulation. Secondly, by improving the preset initial ocean model, especially introducing nonlinear feature representation and hybrid dynamics model, the model's ability to describe complex ocean processes is greatly improved. Adaptive particle filter optimization further improves the accuracy of model parameters, making the simulation results closer to the actual ocean conditions. Furthermore, the combination of parallel computing and adaptive grid refinement significantly improves the computational efficiency, making large-scale three-dimensional ocean current simulation possible. The application of dynamic load balancing and critical path method optimizes the utilization of computing resources and reduces computing time. The introduction of adaptive multi-scale interpolation and spectral element method not only improves numerical accuracy, but also ensures computational stability. In addition, the application of three-dimensional visualization processing technology, especially volume rendering algorithms and isosurface extraction, provides an intuitive and clear display of the three-dimensional structure and dynamic evolution of ocean phenomena, greatly enhancing the interpretability and practical value of the results. Through morphological analysis and clustering methods, the automatic identification and classification of key ocean phenomena are realized, providing a powerful tool for marine scientific research. At the same time, the introduction of multi-source real-time observation data and data assimilation processing significantly improve the timeliness and accuracy of simulation results. The combination of ensemble Kalman filtering and variational analysis methods ensures the physical consistency and reliability of the fused data. Finally, the application of long short-term memory networks in ocean current prediction calculations improves the model's ability to capture complex time series patterns and enhances the accuracy of predictions. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0021] Figure 1 A schematic diagram of an embodiment of a method for simulating ocean currents based on three-dimensional temperature-salinity values ​​in an embodiment of the present application;

[0022] Figure 2 This is a schematic diagram of an embodiment of an ocean current simulation system based on three-dimensional temperature and salinity values ​​in an embodiment of the present application. DETAILED DESCRIPTION

[0023] Embodiments of the present application provide a method and system for simulating ocean currents based on three-dimensional temperature-salinity values. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable in appropriate circumstances, so that the embodiments described here can be implemented in an order other than the content illustrated or described here. In addition, the terms "include" or "have" and any variations thereof are intended to cover non-exclusive inclusions, for example, the process, method, system, product or equipment comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or equipment.

[0024] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 , an embodiment of the ocean current simulation method based on three-dimensional temperature-salinity values ​​in the embodiment of the present application includes:

[0025] Step S101, collecting three-dimensional temperature and salinity data of the target sea area to obtain original three-dimensional temperature-salinity data, and preprocessing the original three-dimensional temperature-salinity data to obtain an initial data set of a three-dimensional temperature-salinity field;

[0026] Step S102: based on the three-dimensional temperature-salinity field initial data set, improving the preset initial ocean model to obtain a three-dimensional temperature-salinity field simulation model;

[0027] Step S103, performing parallel calculation and adaptive grid refinement processing on the three-dimensional temperature-salinity field simulation model to obtain a model operation scheme;

[0028] Step S104, performing three-dimensional visualization processing on the three-dimensional temperature-salinity field initial data set and the model operation scheme to obtain dynamic three-dimensional temperature-salinity field visualization data, and analyzing the dynamic three-dimensional temperature-salinity field visualization data to obtain three-dimensional structure data and evolution process data of key ocean phenomena;

[0029] Step S105, performing data assimilation processing on the multi-source real-time observation data, three-dimensional structure data and evolution process data collected in real time to obtain a fused data set;

[0030] Step S106: perform ocean current prediction calculation on the fused data set to obtain a three-dimensional ocean current prediction field, and perform uncertainty quantification analysis on the three-dimensional ocean current prediction field to obtain ocean current simulation probability distribution data.

[0031] It is understandable that the execution subject of the present application may be a current simulation system based on three-dimensional temperature-salinity values, or a terminal or a server, which is not specifically limited here. The present application embodiment is described by taking a server as the execution subject as an example.

[0032] Specifically, three-dimensional temperature and salinity data are collected in the target sea area, and discrete temperature and salinity measurement data are obtained through multi-point and multi-level sampling. The three-dimensional scattered point interpolation algorithm is used to spatially reconstruct these discrete data to obtain the initial continuous temperature and salinity field. The algorithm uses the data values ​​of known points to calculate the values ​​of unknown points through interpolation functions, thereby generating a complete three-dimensional temperature and salinity distribution. Then, the initial continuous temperature and salinity field is processed using Gaussian filtering to obtain a smoothed temperature and salinity distribution. Gaussian filtering reduces noise and subtle fluctuations in the data by applying a Gaussian function to each data point. The smoothed temperature and salinity distribution is Fourier transformed to convert spatial domain data into frequency domain features. Fourier transform decomposes complex spatial distributions into superpositions of simple periodic functions of different frequencies. Bandpass filtering is performed on the obtained frequency domain features to extract temperature and salinity change information of a specific scale. The bandpass filter only allows signals within a specific frequency range to pass, effectively removing high-frequency noise and low-frequency background. The filtered frequency domain information is converted back to the spatial domain through inverse Fourier transform to obtain a multi-scale temperature and salinity field. Outlier detection is performed on the multi-scale temperature-salinity field to identify a set of temperature-salinity anomaly points. Outlier detection is usually based on statistical methods, such as the standard deviation method or the interquartile range method, to find data points that deviate from the normal range. The outliers are corrected using a local regression algorithm to obtain a corrected temperature-salinity field. The local regression algorithm considers the local trend around the data points and smoothes the outliers. A vertical profile analysis is performed on the corrected temperature-salinity field to identify the thermocline and halocline information. Finally, the density calculation formula is used to convert the corrected temperature-salinity field into a density field to obtain the initial data set of the three-dimensional temperature-salinity field.

[0033] Based on the initial data set of the three-dimensional temperature-salinity field, the preset initial ocean model is improved. First, a spatiotemporal correlation analysis is performed to understand the changing characteristics of the temperature-salinity field. Empirical mode decomposition is used to decompose these characteristics at multiple scales to obtain the intrinsic mode function set. The intrinsic mode function set is Hilbert transformed to obtain the instantaneous frequency and amplitude information, and the dynamic characteristic matrix of the temperature-salinity field is constructed to form a nonlinear characteristic representation. The main change modes are extracted by singular spectrum analysis, and these modes are fitted using support vector regression to obtain the temperature-salinity field prediction function. The control equations of the preset initial ocean model are discretized by variational methods to establish a discrete model framework. The temperature-salinity field prediction function is embedded in this framework to form a hybrid dynamics model. The model parameters are optimized using adaptive particle filtering, and finally a three-dimensional temperature-salinity field simulation model is obtained.

[0034] Parallel calculation and adaptive mesh refinement are performed on the three-dimensional temperature-salinity field simulation model. First, the spatial domain is decomposed to divide the calculation area into sub-region data blocks. The dynamic load balancing algorithm is used to assign tasks to these data blocks to ensure efficient use of computing resources. The critical path method is used to schedule computing resources and formulate parallel execution plans. During the calculation process, the gradient changes are monitored to determine the mesh refinement requirements. The adaptive multi-scale interpolation algorithm is used to generate a dynamic mesh structure, and the boundary conditions are matched to obtain a continuous mesh layout. The spectral element method is used to discretize the calculation equations to obtain a high-order numerical format. Stability analysis is performed to determine the time step constraints, and these constraints are processed by the adaptive time stepping algorithm to finally obtain the model operation plan.

[0035] The initial data set and model operation scheme of the three-dimensional temperature-salinity field are processed for three-dimensional visualization. First, the data is interpolated to obtain a high-resolution temperature-salinity field grid. The grid data are rendered using a volume rendering algorithm to generate a three-dimensional temperature-salinity field volume image. The visualization effect is enhanced by color mapping, and a time series animation is generated based on the model operation scheme to obtain dynamic three-dimensional temperature-salinity field visualization data. The visualization data is processed using an isosurface extraction algorithm to obtain an isosurface sequence of key ocean phenomena. Morphological analysis is performed on these sequences to extract three-dimensional structural feature parameters. Key ocean phenomena are classified through cluster analysis, and time series analysis is performed to obtain the evolution process feature vector. Principal component analysis is used to reduce the dimension of the feature vector, extract key evolution patterns, and finally obtain the three-dimensional structure data and evolution process data of key ocean phenomena through spatiotemporal interpolation.

[0036] Data assimilation is performed on multi-source real-time observation data, three-dimensional structure data, and evolution process data collected in real time. First, quality control and spatiotemporal interpolation are performed on real-time observation data to obtain a gridded observation field. Data correlation is determined through covariance analysis, and the ensemble Kalman filter algorithm is used for data assimilation. The assimilation results are checked for temporal consistency and physical constraints are optimized to ensure the physical consistency of the data. The data is denoised through scale decomposition and wavelet transform to obtain a smoothed feature field. Spectral analysis is performed to extract the main frequency components, and finally the data is reconstructed through the adaptive spectrum synthesis method to obtain a fused data set.

[0037] The fused data set is used for ocean current prediction calculation. First, the data is decomposed in time and space to extract multi-scale ocean dynamic features. Long short-term memory networks are used to predict these features in time series and perform physical constraint correction. The Monte Carlo method is used to generate multiple sets of disturbance fields to form ensemble prediction members. These members are statistically analyzed to obtain the predicted mean field and variance field. The kernel density estimation method is used to fit the probability distribution of the prediction results and perform bias correction. The prediction confidence interval is calculated by the Bayesian inference method to quantify the uncertainty. Finally, the uncertainty results are spatially interpolated and graded to obtain the probability distribution data of ocean current simulation.

[0038] For example, when simulating the three-dimensional temperature-salinity value and ocean current in a certain sea area, multi-point and multi-level sampling is first performed. An underwater glider is used to collect temperature and salinity data at a depth interval of 50 meters in a horizontal grid of 10 km × 10 km within a depth range of 0-1000 meters, and a total of 10,000 discrete temperature-salinity measurement data points are obtained. The Kriging interpolation method is used to reconstruct these discrete data in three-dimensional space. Kriging interpolation takes into account spatial autocorrelation and estimates the value of unknown points based on the position and value of known points. Through this method, a 100×100×20 three-dimensional gridded temperature-salinity field is generated with a horizontal resolution of 1 km and a total of 20 layers in the vertical direction. Gaussian filtering is applied to the initial continuous temperature-salinity field, using a 3×3×3 filter kernel and a σ value of 1.5, which effectively reduces the high-frequency noise introduced by measurement errors and obtains a smoothed temperature-salinity distribution.

[0039] Then, the smoothed temperature-salinity distribution was subjected to a three-dimensional fast Fourier transform (3D FFT) to convert the spatial domain data into the frequency domain. In the frequency domain, a bandpass filter was designed to extract temperature-salinity variation information at a scale of 10-100 km through a wave number range of 0.01-0.1 cycles / km. The filtered frequency domain information was converted back to the spatial domain through a three-dimensional inverse Fourier transform (3DIFFT) to obtain a multi-scale temperature-salinity field. An outlier detection was performed on this field, and 37 temperature-salinity anomaly points were identified using the 3σ principle. These 37 anomaly points were corrected using LOESS (local weighted regression scatter point smoothing method). LOESS selected 100 nearest neighbor points around each anomaly point and performed local regression through polynomial fitting. The maximum correction amplitude reached 0.5℃ for temperature and 0.3psu for salinity. A vertical profile analysis was performed on the corrected temperature-salinity field. By calculating the vertical gradients of temperature and salinity, it was identified that the thermocline was located in the depth range of 30-40 meters and the halocline was in the depth range of 45-55 meters.

[0040] Finally, the temperature and salinity are converted into density using the UNESCO equation of state, resulting in an initial data set of 100×100×20 three-dimensional temperature-salinity-density field. This initial data set is then used for subsequent steps such as model improvement, parallel computing, and visualization. For example, in the model improvement stage, empirical mode decomposition is performed on this data set to extract five main intrinsic mode functions (IMFs), capturing different time scale characteristics from daily changes to seasonal changes. In parallel computing, this 100×100×20 data field is divided into 10×10×2 sub-region blocks and assigned to 100 computing nodes for simultaneous processing, which significantly improves the computing efficiency.

[0041] In the embodiment of the present application, by performing multi-point multi-level sampling and three-dimensional scattered point interpolation on the target sea area, the spatial resolution and accuracy of the initial data of the temperature and salinity field are significantly improved, providing a more reliable input for subsequent simulation. The use of techniques such as Gaussian filtering and spectrum analysis effectively removes noise and outliers in the original data and improves the data quality. The application of multi-scale decomposition and local regression algorithms enables the model to better capture marine phenomena of different scales and enhances the precision of the simulation. Secondly, by improving the preset initial ocean model, especially introducing nonlinear feature representation and hybrid dynamics model, the model's ability to describe complex ocean processes is greatly improved. Adaptive particle filter optimization further improves the accuracy of model parameters, making the simulation results closer to the actual ocean conditions. Furthermore, the combination of parallel computing and adaptive grid refinement significantly improves the computational efficiency, making large-scale three-dimensional ocean current simulation possible. The application of dynamic load balancing and critical path method optimizes the utilization of computing resources and reduces computing time. The introduction of adaptive multi-scale interpolation and spectral element method not only improves numerical accuracy, but also ensures computational stability. In addition, the application of three-dimensional visualization processing technology, especially volume rendering algorithms and isosurface extraction, provides an intuitive and clear display of the three-dimensional structure and dynamic evolution of ocean phenomena, greatly enhancing the interpretability and practical value of the results. Through morphological analysis and clustering methods, the automatic identification and classification of key ocean phenomena are realized, providing a powerful tool for marine scientific research. At the same time, the introduction of multi-source real-time observation data and data assimilation processing significantly improve the timeliness and accuracy of simulation results. The combination of ensemble Kalman filtering and variational analysis methods ensures the physical consistency and reliability of the fused data. Finally, the application of long short-term memory networks in ocean current prediction calculations improves the model's ability to capture complex time series patterns and enhances the accuracy of predictions.

[0042] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0043] (1) Carry out multi-point and multi-level sampling in the target sea area to obtain discrete temperature and salinity measurement data;

[0044] (2) The discrete temperature-salinity measurement data are spatially reconstructed through a three-dimensional scattered point interpolation algorithm to obtain the initial continuous temperature-salinity field;

[0045] (3) Perform Gaussian filtering on the initial continuous temperature-salinity field to obtain a smoothed temperature-salinity distribution;

[0046] (4) Perform spectrum analysis on the smoothed temperature-salinity distribution through Fourier transform to obtain frequency domain characteristics;

[0047] (5) Bandpass filtering is performed on the frequency domain features to obtain temperature-salinity change information at a specific scale;

[0048] (6) The temperature-salinity variation information of a specific scale is converted back to the spatial domain through inverse Fourier transform to obtain a multi-scale temperature-salinity field;

[0049] (7) Detect outliers on the multi-scale temperature-salinity field and obtain a set of temperature-salinity anomaly points;

[0050] (8) Correct the temperature-salinity anomaly point set through the local regression algorithm to obtain the corrected temperature-salinity field;

[0051] (9) Conduct vertical profile analysis on the corrected temperature-salinity field to obtain thermocline and halocline information;

[0052] (10) The corrected temperature-salinity field is converted into a density field through the density calculation formula to obtain the initial data set of the three-dimensional temperature-salinity field.

[0053] Specifically, multi-point and multi-level sampling is carried out in the target sea area to obtain discrete temperature and salinity measurement data. This process usually uses equipment such as underwater gliders, profiling buoys or ship-borne CTD (conductivity-temperature-depth) instruments to collect temperature and salinity data at different depths and horizontal positions. The sampling strategy usually adopts grid distribution to ensure the spatial representativeness of the data. After obtaining the discrete data, the data is spatially reconstructed using a three-dimensional scattered point interpolation algorithm to obtain the initial continuous temperature and salinity field. Commonly used interpolation methods include Kriging, inverse distance weighted method and spline interpolation. Taking Kriging as an example, this method takes into account spatial autocorrelation, describes the spatial variation characteristics of the data through a variation function, and then uses the data values ​​and location information of known points to estimate the values ​​of unknown points. This method can effectively handle irregularly distributed data points and provide error estimates for interpolation results.

[0054] The initial continuous temperature-salinity field is Gaussian filtered to obtain a smoothed temperature-salinity distribution. Gaussian filtering is a linear smoothing filter that uses a normal distribution (Gaussian function) as the filter kernel. During the filtering process, the new value of each data point is calculated by the weighted average of its surrounding points, and the weight is determined by the Gaussian function. This step effectively reduces the high-frequency noise in the data while retaining the main spatial structural characteristics. The smoothed temperature-salinity distribution is Fourier transformed to obtain frequency domain features. The three-dimensional fast Fourier transform (3DFFT) converts the temperature-salinity distribution in the spatial domain into a frequency domain representation, revealing the change characteristics of different spatial scales in the data. In the frequency domain, low-frequency components correspond to large-scale changes, and high-frequency components correspond to small-scale changes.

[0055] Bandpass filtering is performed on the frequency domain features to extract the temperature and salt change information at a specific scale. The bandpass filter allows signals within a specific frequency range to pass through while suppressing other frequencies. This step can separate the spatial scale features of interest, such as phenomena such as mesoscale vortices or internal waves. The temperature and salt change information at a specific scale is converted back to the spatial domain through inverse Fourier transform to obtain a multi-scale temperature and salt field. This process uses a three-dimensional inverse fast Fourier transform (3D IFFT) to reconvert the filtered frequency domain data into a spatial distribution. Outlier detection is performed on the multi-scale temperature and salt field to identify a set of temperature and salt anomaly points. Commonly used outlier detection methods include statistical threshold methods (such as the 3σ rule) and local outlier factor (LOF) methods. These methods identify points that deviate significantly from surrounding data based on statistical characteristics or spatial relationships.

[0056] The local regression algorithm is used to correct the set of temperature and salinity anomaly points to obtain the corrected temperature and salinity field. Local regression algorithms, such as LOESS (local weighted regression scatter point smoothing method), estimate the correct value by selecting a certain number of neighboring points around each anomaly point and performing local polynomial fitting. This method can effectively handle nonlinear relationships and is insensitive to outliers. The corrected temperature and salinity field is analyzed vertically to obtain thermocline and halocline information. This step calculates the vertical gradients of temperature and salinity to identify the depth intervals where the gradients change significantly. Thermocline and halocline are areas in the ocean where the physical properties change dramatically, and have an important impact on ocean dynamics.

[0057] Finally, the corrected temperature-salinity field is converted into a density field through the density calculation formula to obtain the initial data set of the three-dimensional temperature-salinity field. Commonly used density calculation formulas include the UNESCO seawater state equation, which takes into account the comprehensive effects of temperature, salinity and pressure on seawater density.

[0058] For example, in the data collection of the temperature and salinity field in a certain sea area, an underwater glider was used to collect data at a depth interval of 50 meters in a horizontal grid of 10 km × 10 km within a depth range of 0-1000 meters, and a total of 10,000 discrete temperature and salinity measurement points were obtained. The Kriging interpolation method was used to reconstruct these data in three-dimensional space, generating a 100×100×20 grid temperature and salinity field with a horizontal resolution of 1 km and 20 layers in the vertical direction. A 3×3×3 Gaussian filter kernel (σ=1.5) was applied to this field for smoothing. After Fourier transformation, a bandpass filter was designed to extract temperature and salinity changes at a scale of 10-100 km. Outlier detection identified 37 outliers, which were corrected using the LOESS method, with a maximum correction amplitude of 0.5℃ for temperature and 0.3psu for salinity. Vertical profile analysis found that the thermocline was located in the depth range of 30-40 meters. Finally, the temperature-salinity data were converted into density using the UNESCO seawater state equation to form an initial data set of three-dimensional temperature-salinity-density field of 100×100×20.

[0059] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0060] (1) Perform spatiotemporal correlation analysis on the initial data set of the three-dimensional temperature-salinity field to obtain the characteristics of the temperature-salinity field change;

[0061] (2) Through empirical mode decomposition, the temperature-salinity field variation characteristics are decomposed at multiple scales to obtain the intrinsic mode function set;

[0062] (3) Performing Hilbert transform on the eigenmode function set to obtain instantaneous frequency and amplitude information;

[0063] (4) Based on the instantaneous frequency and amplitude information, the dynamic characteristic matrix of the temperature-salinity field is constructed to obtain the nonlinear characteristic representation;

[0064] (5) Perform singular spectrum analysis on the nonlinear feature representation to obtain the main change mode;

[0065] (6) Fit the main change patterns through support vector regression to obtain the temperature-salinity field prediction function;

[0066] (7) performing variational discretization on the control equations of the preset initial ocean model to obtain a discrete model framework;

[0067] (8) The temperature-salinity field prediction function is embedded in the discrete model framework to obtain a hybrid dynamics model;

[0068] (9) Optimizing the parameters of the hybrid dynamics model through adaptive particle filtering to obtain optimized model parameters;

[0069] (10) Based on the optimized model parameters, the hybrid dynamics model is calibrated to obtain a three-dimensional temperature-salinity field simulation model.

[0070] Specifically, the temporal and spatial correlation analysis of the initial data set of the three-dimensional temperature-salinity field is first performed to obtain the characteristics of the temperature-salinity field change. This analysis process includes calculating the autocorrelation function of the time series and the cross-correlation function of the spatial distribution, thereby revealing the changing laws of the temperature-salinity field in the time and space dimensions. Then, the temperature-salinity field change characteristics are multi-scale decomposed by empirical mode decomposition (EMD) to obtain the intrinsic mode function (IMF) set. EMD is an adaptive signal processing method that can decompose complex signals into a finite number of intrinsic mode functions and a residual term, and each IMF represents an oscillation mode of different scales. The obtained intrinsic mode function set is Hilbert transformed to obtain instantaneous frequency and amplitude information. The Hilbert transform can extract the instantaneous properties of the signal by converting the real signal into an analytical signal. Based on these instantaneous frequency and amplitude information, the dynamic characteristic matrix of the temperature-salinity field is constructed to obtain a nonlinear feature representation. This process converts multi-dimensional spatiotemporal data into a characteristic matrix that can reflect the dynamic characteristics of the system.

[0071] Singular spectrum analysis (SSA) is performed on the nonlinear feature representation to obtain the main change patterns. SSA is a time series analysis method that can extract the main change patterns in the time series by constructing a trajectory matrix and performing singular value decomposition. These main change patterns are fitted by support vector regression (SVR) to obtain the temperature-salinity field prediction function. SVR is a machine learning method that achieves nonlinear regression by mapping data into a high-dimensional feature space and finding the optimal separation hyperplane. The control equations of the preset initial ocean model are discretized by variational method to obtain a discrete model framework. Variational discretization is a method of converting continuous partial differential equations into discrete forms, which ensures the properties of discrete models such as energy conservation. The temperature-salinity field prediction function obtained above is embedded in this discrete model framework to form a hybrid dynamics model. This step combines the data-driven statistical model with the physics-based dynamics model to improve the prediction ability and physical consistency of the model.

[0072] Finally, the hybrid dynamics model is optimized by adaptive particle filtering to obtain the optimized model parameters. Adaptive particle filtering is a sequential Monte Carlo method that can effectively handle the state estimation problem of nonlinear and non-Gaussian systems by continuously adjusting particle weights and resampling strategies. Based on the optimized model parameters, the hybrid dynamics model is calibrated to finally obtain a three-dimensional temperature-salinity field simulation model.

[0073] For example, in the simulation of the three-dimensional temperature-salinity field in a certain sea area, the spatiotemporal correlation analysis of the 100×100×20 grid temperature-salinity field data was first performed. By calculating the correlation coefficients at different time lags and spatial distances, it was found that the temperature field had significant spatial correlation within a range of 50 km, and showed obvious daily and seasonal variation characteristics on the time scale. The temperature time series was decomposed using EMD to obtain five IMF components, corresponding to the changes on the daily scale, weather scale, monthly scale, and seasonal scale. Each IMF was Hilbert transformed to extract the instantaneous frequency and amplitude, and a dynamic feature matrix of 100×100×20×5 was constructed. SSA was performed on this feature matrix, and the first three principal components were selected as the main variation mode, which cumulatively explained 85% of the variance. The three principal components were fitted using SVR, and the kernel function selected the RBF kernel. The optimal parameters C=10 and γ=0.1 were determined by cross-validation to obtain the temperature-salinity field prediction function. This prediction function was embedded in the discrete model framework based on the original Navier-Stokes equation to form a hybrid dynamics model. The model parameters were optimized using an adaptive particle filter with 100 particles, and the optimal parameter set was obtained after 50 iterations. The root mean square error of the final three-dimensional temperature-salinity field simulation model on the independent test set was 0.2℃ (temperature) and 0.1psu (salinity).

[0074] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0075] (1) Decompose the three-dimensional temperature-salinity field simulation model in the spatial domain to obtain sub-region data blocks;

[0076] (2) Allocate tasks to sub-region data blocks through a dynamic load balancing algorithm to obtain a parallel computing task list;

[0077] (3) Perform data dependency analysis on the parallel computing task list to obtain a task execution graph;

[0078] (4) Based on the task execution graph, the computing resources are scheduled using the critical path method to obtain a parallel execution plan;

[0079] (5) Monitor the gradient changes during the calculation process and obtain the grid refinement requirement index;

[0080] (6) The grid refinement requirement index is processed through an adaptive multi-scale interpolation algorithm to obtain a dynamic grid structure;

[0081] (7) matching boundary conditions of the dynamic grid structure to obtain a continuous grid layout;

[0082] (8) Based on the continuous grid layout, the computational equations are discretized by the spectral element method to obtain a high-order numerical format;

[0083] (9) Perform stability analysis on high-order numerical formats and obtain time step constraints;

[0084] (10) The time step constraint conditions are processed through an adaptive time stepping algorithm to obtain a model operation scheme.

[0085] Specifically, the three-dimensional temperature-salinity field simulation model is decomposed in the spatial domain to obtain sub-region data blocks. This process uses domain decomposition technology to divide the entire computational domain into several overlapping or non-overlapping sub-regions, each of which contains some grid points and corresponding temperature-salinity data. Then, the sub-region data blocks are assigned tasks through the dynamic load balancing algorithm to obtain a parallel computing task list. The dynamic load balancing algorithm takes into account the processing power and current load of each computing node, and dynamically adjusts the task allocation to maximize parallel efficiency. Data dependency analysis is performed on the parallel computing task list to obtain a task execution graph. Data dependency analysis identifies the front-and-back relationships and data exchange requirements between tasks, and generates a directed acyclic graph to represent the task execution order. Based on this task execution graph, the computing resources are scheduled through the critical path method to obtain a parallel execution plan. The critical path method finds the longest execution path in the task graph and prioritizes the tasks on the critical path to minimize the overall execution time.

[0086] During the calculation process, the gradient changes are monitored in real time to obtain the grid refinement demand indicators. These indicators reflect the areas where the temperature and salinity fields change rapidly in space, such as fronts or vortex edges. The grid refinement demand indicators are processed by the adaptive multi-scale interpolation algorithm to obtain a dynamic grid structure. The adaptive multi-scale interpolation algorithm dynamically adjusts the grid density according to the local gradient size, adding grid points in areas with drastic changes and maintaining a coarser grid in areas with gentle changes. The boundary conditions of the dynamic grid structure are matched to obtain a continuous grid layout. This step ensures the continuity of the grid and the continuity of physical quantities between sub-areas, avoiding non-physical oscillations in numerical calculations. Based on the continuous grid layout, the calculation equations are discretized by the spectral element method to obtain a high-order numerical format. The spectral element method combines the geometric flexibility of the finite element method and the high-precision characteristics of the spectral method, and can achieve high-precision numerical solutions on complex geometric shapes.

[0087] The stability analysis of the high-order numerical format is performed to obtain the time step constraint conditions. The stability analysis usually uses the von Neumann method or the matrix method to determine the stability conditions of the numerical format. Finally, the time step constraint conditions are processed by the adaptive time stepping algorithm to obtain the model operation scheme. The adaptive time stepping algorithm dynamically adjusts the time step according to the local flow field characteristics and numerical stability requirements, improving the calculation efficiency while ensuring the calculation accuracy.

[0088] For example, in the simulation of the three-dimensional temperature-salinity field in a certain sea area, the computational domain is 500km×500km×1000m, and the initial grid is 100×100×50. First, the computational domain is divided into 10×10×5 sub-region data blocks, each of which contains 10×10×10 grid points. Using the dynamic load balancing algorithm based on work stealing, the 500 sub-region tasks are assigned to 64 computing nodes. Through data flow analysis, a task dependency graph is generated, and 200 critical path tasks are identified. During the calculation process, it is monitored that there is a strong temperature gradient in the water depth range of 200-300m, and the temperature changes by more than 5℃ within 10m. Using the adaptive multi-scale interpolation algorithm, the grid is encrypted to 1m resolution in this area, while the resolution is maintained at 10m in other areas, and finally a dynamic grid structure containing about 1.5 million grid points is formed. The spectral element method is applied to this unstructured grid, and the 8th-order polynomial basis function is used for spatial discretization. Stability analysis shows that in the finest grid area, the maximum allowed time step of the explicit time marching method is 0.1s. The adaptive Runge-Kutta-Fehlberg method is used for time marching, using a 30s time step in most areas and automatically reducing it to 0.1s in fine grid areas. This adaptive strategy allows the simulation of the 10-day ocean current evolution process to be completed in about 6 hours on a 64-node cluster, which is about 3 times more efficient than the fixed time step and uniform grid method. The final model calculation scheme not only accurately captures the fine structure of the thermocline, but also successfully simulates the generation and evolution of mesoscale vortices.

[0089] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0090] (1) Perform data interpolation on the initial data set of the three-dimensional temperature-salinity field to obtain a high-resolution temperature-salinity field grid;

[0091] (2) Rendering the high-resolution temperature-salinity field grid through a volume rendering algorithm to obtain a three-dimensional temperature-salinity field volume image;

[0092] (3) Perform color mapping on the three-dimensional temperature-salinity field volume image to obtain a color-enhanced visualization effect;

[0093] (4) Based on the model operation scheme, a time-series animation is generated for the color-enhanced visualization effect to obtain dynamic three-dimensional temperature-salinity field visualization data;

[0094] (5) The dynamic three-dimensional temperature-salinity field visualization data is processed by isosurface extraction algorithm to obtain isosurface sequences of key ocean phenomena;

[0095] (6) Perform morphological analysis on the isosurface sequence to obtain three-dimensional structural characteristic parameters;

[0096] (7) Based on the three-dimensional structural characteristic parameters, key marine phenomena are classified through cluster analysis to obtain the phenomenon type classification;

[0097] (8) Conduct time series analysis on the phenomenon type classification to obtain the characteristic vector of the evolution process;

[0098] (9) Reduce the dimension of the evolution process feature vector through principal component analysis to obtain the key evolution pattern;

[0099] (10) Perform spatiotemporal interpolation on key evolution patterns to obtain three-dimensional structural data and evolution process data of key ocean phenomena.

[0100] Specifically, the ocean current simulation method based on three-dimensional temperature-salinity values ​​first interpolates the initial data set of the three-dimensional temperature-salinity field to obtain a high-resolution temperature-salinity field grid. This process uses a trilinear interpolation algorithm to expand the original low-resolution data to a finer grid to improve the spatial resolution. Then, the high-resolution temperature-salinity field grid is rendered using a volume rendering algorithm to obtain a three-dimensional temperature-salinity field volume image. The volume rendering algorithm uses a ray casting method to simulate the process of light passing through a three-dimensional data field, mapping the three-dimensional data onto a two-dimensional plane to generate a volume image with a sense of depth.

[0101] Color mapping is performed on the three-dimensional temperature-salinity field volume image to obtain a color-enhanced visualization effect. The color mapping process uses a predefined color scale to map the temperature and salinity values ​​to different colors. Usually, warm and cold tones are used to represent temperature changes, and saturation changes are used to represent salinity changes. Based on the model operation scheme, a time-series animation is generated for the color-enhanced visualization effect to obtain dynamic three-dimensional temperature-salinity field visualization data. This step interpolates the data at the intermediate moments and generates a series of image frames in chronological order to form a continuous animation effect.

[0102] The dynamic three-dimensional temperature-salinity field visualization data is processed by isosurface extraction algorithms to obtain isosurface sequences of key ocean phenomena. Isosurface extraction algorithms such as Marching Cubes search for continuous surfaces of specific values ​​in three-dimensional space to represent specific distributions of temperature or salinity. Morphological analysis is performed on the isosurface sequence to obtain three-dimensional structural characteristic parameters. Morphological analysis includes calculating geometric features such as area, volume, curvature, and topological features such as connectivity and the number of holes of the isosurface. Based on the three-dimensional structural characteristic parameters, key ocean phenomena are classified by cluster analysis to obtain the phenomenon type classification. Cluster analysis uses algorithms such as K-means or hierarchical clustering to classify ocean phenomena with similar characteristics. Time series analysis is performed on the phenomenon type classification to obtain the evolution process feature vector. Time series analysis methods such as the autoregressive moving average model (ARMA) are used to capture the changing patterns of ocean phenomena over time.

[0103] The key evolution patterns are obtained by reducing the dimensionality of the evolution process feature vectors through principal component analysis. Principal component analysis (PCA) projects high-dimensional data into low-dimensional space through linear transformation, retaining the main variation information. Finally, the key evolution patterns are interpolated in time and space to obtain the three-dimensional structure data and evolution process data of key ocean phenomena. The spatiotemporal interpolation combines spatial interpolation and time series prediction methods to generate continuous three-dimensional structure and evolution process data.

[0104] For example, in the simulation of the three-dimensional temperature and salt field of a certain sea area, the initial data set is a grid of 100×100×50. It is expanded to a high-resolution grid of 500×500×250 by trilinear interpolation. The ray casting method is used for volume rendering, and the opacity of each voxel is dynamically adjusted according to its gradient size to highlight the boundary structure of the temperature and salt field. The color mapping uses blue-white-red scale to represent temperature (0-30℃), and green saturation represents salinity (30-38psu).

[0105] Based on the model output at 6-hour intervals, a 30-day animation sequence containing 720 frames was generated. The 20°C isotherm and 35psu isosalinity surfaces were extracted using the MarchingCubes algorithm to form an isosurface sequence. The area, volume, and mean curvature of these isosurfaces were calculated to obtain the three-dimensional structural characteristic parameters. The ocean phenomena were divided into five categories: warm vortex, cold vortex, upwelling, downwelling, and front by K-means clustering (K=5).

[0106] ARMA (2,1) time series analysis was performed on each type of phenomenon to obtain the eigenvector of the evolution process. PCA was used to reduce the 50-dimensional eigenvector to 5 dimensions, retaining 95% of the variation information. Finally, the five principal components were interpolated in time and space using the Kriging interpolation method to obtain continuous three-dimensional structure and evolution process data. This process successfully identified a warm vortex structure with a diameter of 50km and a depth of 200m, and tracked its movement and evolution over 30 days, revealing its interaction mechanism with the surrounding cold water.

[0107] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0108] (1) Perform data quality control on multi-source real-time observation data to obtain cleaned observation data;

[0109] (2) Perform spatiotemporal interpolation on the cleaned observation data to obtain a gridded observation field;

[0110] (3) Perform covariance analysis on the gridded observation field and the three-dimensional structure data to obtain the data correlation matrix;

[0111] (4) Based on the data correlation matrix, the data is assimilated through the ensemble Kalman filter algorithm to obtain preliminary fusion data;

[0112] (5) Performing a time series consistency check on the preliminary fusion data and the evolution process data to obtain the corrected fusion data;

[0113] (6) Constrained optimization of the corrected fusion data is performed using a variational analysis method to obtain physically consistent data;

[0114] (7) Decomposing the physical consistency data to obtain a multi-scale feature field;

[0115] (8) De-noising the multi-scale feature field by wavelet transform to obtain a smoothed feature field;

[0116] (9) Performing spectral analysis on the smoothed characteristic field to obtain the main frequency components;

[0117] (10) Based on the main frequency components, the data is reconstructed through an adaptive spectrum synthesis method to obtain a fused data set.

[0118] Specifically, data quality control is performed on multi-source real-time observation data to obtain cleaned observation data. This process includes steps such as outlier detection, consistency check and instrument error correction. Statistical methods such as the 3σ criterion and spatial consistency test are used to identify and remove abnormal data points. Next, the cleaned observation data is interpolated in time and space to obtain a gridded observation field. The spatiotemporal interpolation uses the Kriging method, which takes into account the spatial correlation and temporal continuity of the data and interpolates the discrete observation point data onto a regular three-dimensional grid. Covariance analysis is performed on the gridded observation field and the three-dimensional structure data to obtain a data correlation matrix. Covariance analysis calculates the degree of correlation between different variables to form a symmetric matrix in which each element represents the covariance between two variables. Based on this data correlation matrix, the data is assimilated by the ensemble Kalman filter algorithm to obtain preliminary fused data. Ensemble Kalman filtering is a recursive data assimilation method that continuously updates the system state estimate by alternating between prediction steps and analysis steps.

[0119] The initial fusion data and the evolution process data are checked for time consistency to obtain the corrected fusion data. The time consistency check compares the consistency of the fusion data with the historical evolution trend, identifies and corrects abnormal time jumps. The corrected fusion data is constrained and optimized by variational analysis to obtain physically consistent data. The variational analysis method adjusts the data by minimizing the cost function so that it meets physical constraints such as mass conservation and energy balance. The physically consistent data is scaled to obtain a multi-scale feature field. Scale decomposition uses multi-resolution analysis methods, such as wavelet decomposition, to decompose the data into components of different spatial scales. The multi-scale feature field is denoised by wavelet transform to obtain a smoothed feature field. Wavelet denoising removes high-frequency noise by thresholding the wavelet coefficients while retaining the main features of the signal.

[0120] Perform spectral analysis on the smoothed feature field to obtain the main frequency components. Spectral analysis uses fast Fourier transform (FFT) to convert the time series into the frequency domain and identify the main periodic components. Based on these main frequency components, the data is reconstructed through adaptive spectrum synthesis to obtain the final fused data set. Adaptive spectrum synthesis dynamically adjusts the weights of different frequency components according to their importance to reconstruct a smoother and physically consistent data field.

[0121] For example, in the simulation of the three-dimensional temperature and salinity field in a certain sea area, the multi-source real-time observation data includes the temperature and salinity data of 50 buoy stations, 10 moored observation stations and 2 ship-borne CTD sections. During the data quality control process, 87 outliers were identified using the 3σ criterion, and 23 suspicious data points were eliminated through spatial consistency test. Ordinary Kriging method was used for spatiotemporal interpolation to interpolate discrete observation data onto a regular grid of 100×100×50, with a horizontal resolution of 1km and 50 layers in the vertical direction.

[0122] Covariance analysis generated a 100×100 correlation matrix reflecting the correlation between temperature and salinity at different depths and horizontal positions. Data assimilation was performed using an ensemble Kalman filter with 100 ensemble members, assimilating observations every 6 hours. The temporal consistency check compared the assimilation results with the evolution trend of the past 7 days and corrected the abnormal jumps at 15 time points. The variational analysis method used the conjugate gradient method to minimize the cost function, and iterated 50 times to reach convergence, ensuring that the data met thermodynamic equilibrium.

[0123] Scale decomposition uses discrete wavelet transform to decompose the data into four scale levels. Wavelet denoising uses a soft threshold method, and the threshold is set to three times the standard deviation of the noise, which effectively removes high-frequency noise. Spectral analysis identifies three main frequency components through FFT, corresponding to tidal cycles, inertial oscillations, and seasonal changes. Adaptive spectrum synthesis dynamically adjusts the reconstruction weights according to the energy distribution of these frequency components. The final fused data set retains the original observation information while showing more coherent and physically consistent ocean dynamic characteristics. This fused data set successfully captured a mesoscale vortex structure with a diameter of about 50 kilometers and a depth of 200 meters, and clearly shows the distribution of the surrounding temperature and salinity fronts.

[0124] In a specific embodiment, the process of executing step S106 may specifically include the following steps:

[0125] (1) Perform spatiotemporal decomposition of the fused dataset to obtain multi-scale ocean dynamic characteristics;

[0126] (2) Use long short-term memory networks to perform time series predictions on multi-scale ocean dynamics characteristics and obtain preliminary ocean current prediction results;

[0127] (3) Perform physical constraint correction on the preliminary ocean current prediction results to obtain a physically consistent prediction field;

[0128] (4) Based on the physical consistency prediction field, multiple sets of perturbation fields are generated by Monte Carlo method to obtain the set prediction members;

[0129] (5) Perform statistical analysis on the ensemble prediction members to obtain the prediction mean field and variance field;

[0130] (6) Perform probability distribution fitting on the predicted mean field and variance field by using the kernel density estimation method to obtain a preliminary probability distribution model;

[0131] (7) performing bias correction on the preliminary probability distribution model to obtain a revised probability distribution;

[0132] (8) Based on the modified probability distribution, the prediction confidence interval is calculated by Bayesian inference method to obtain the uncertainty quantification result;

[0133] (9) Perform spatial interpolation on the uncertainty quantification results to obtain a continuous uncertainty field;

[0134] (10) The continuous uncertainty field is classified by the adaptive threshold method to obtain the probability distribution data of ocean current simulation.

[0135] Specifically, the fused data set is decomposed in time and space to obtain multi-scale ocean dynamic characteristics. This process uses the empirical mode decomposition (EMD) method to decompose the three-dimensional temperature and salinity field into intrinsic mode functions (IMFs) of different time and space scales, each of which represents the ocean dynamic characteristics of a specific scale. Then, the multi-scale ocean dynamic characteristics are predicted in time series through the long short-term memory network (LSTM) to obtain preliminary ocean current prediction results. LSTM is a special recurrent neural network that can effectively capture long-term dependencies and is suitable for processing long-term dynamic processes in the ocean.

[0136] The preliminary ocean current forecast results are corrected by physical constraints to obtain a physically consistent forecast field. The physical constraint correction uses a variational method to adjust the forecast results by minimizing the cost function so that they satisfy physical laws such as mass conservation and energy balance. Based on the physically consistent forecast field, multiple sets of perturbation fields are generated by the Monte Carlo method to obtain ensemble forecast members. The Monte Carlo method generates multiple possible forecast scenarios by randomly perturbing the initial conditions and model parameters to reflect the uncertainty of the forecast.

[0137] Statistical analysis is performed on the ensemble prediction members to obtain the prediction mean field and variance field. Statistical analysis calculates the mean and variance of all ensemble members at each grid point, representing the most likely prediction result and the uncertainty of the prediction, respectively. The kernel density estimation method is used to fit the probability distribution of the prediction mean field and variance field to obtain a preliminary probability distribution model. Kernel density estimation is a non-parametric method that can flexibly fit complex probability distributions.

[0138] The preliminary probability distribution model is bias-corrected to obtain a revised probability distribution. Bias correction uses the quantile mapping method to correct the systematic bias of the model prediction through historical data. Based on the revised probability distribution, the prediction confidence interval is calculated by the Bayesian inference method to obtain the uncertainty quantification result. Bayesian inference combines prior knowledge and observed data to provide the probability distribution and confidence interval of the prediction results.

[0139] The uncertainty quantification results are spatially interpolated to obtain a continuous uncertainty field. The spatial interpolation uses the Kriging method, which takes into account the spatial correlation of uncertainty and generates a continuous field covering the entire prediction area. Finally, the continuous uncertainty field is classified by the adaptive threshold method to obtain the probability distribution data of ocean current simulation. The adaptive threshold method dynamically adjusts the classification criteria according to the statistical characteristics of the uncertainty field and provides intuitive uncertainty visualization results.

[0140] An example is used to illustrate this process: in the simulation of the three-dimensional temperature and salinity field in a certain sea area, the fused data set contains 100×100×50 grid points, covering a sea area of ​​500km×500km×1000m. The EMD method is used to decompose the data, and 5 IMFs and a residual term are obtained, corresponding to the changes at the daily scale, weather scale, monthly scale, and seasonal scale. A network structure containing 3 layers of LSTM and 128 neurons in each layer is used to perform time series prediction for each IMF, with a prediction period of 7 days and a time step of 6 hours.

[0141] The physical constraint correction uses the 4D-Var method, and the physically consistent prediction field is obtained after 50 iterations of minimization. 100 ensemble members are generated by the Monte Carlo method, and each member adds a random perturbation that follows a Gaussian distribution (mean 0, standard deviation 10% of the observation error) to the initial temperature and salinity field. Statistical analysis and calculations obtain the mean field and variance field of 100×100×50.

[0142] Kernel density estimation uses a Gaussian kernel, and the bandwidth is determined by cross-validation. Bias correction is based on the prediction-observation comparison of the past 30 days, using the quantile mapping method. Bayesian inference uses the Markov Chain Monte Carlo (MCMC) method, sampling using the Metropolis-Hastings algorithm, and 95% confidence intervals are obtained after running 10,000 iterations.

[0143] Ordinary Kriging is used for spatial interpolation, and the exponential model is used for the variogram model. The adaptive threshold method divides the uncertainty field into 5 levels, and the threshold is set to the mean plus or minus 1 and 2 standard deviations. The final ocean current simulation probability distribution data shows that on the 7th day of the forecast, the uncertainty of the surface current forecast in the central area of ​​the sea area (200km×200km) is the lowest, with a 95% confidence interval of ±0.1m / s, while the uncertainty is higher in the boundary area, and the 95% confidence interval is expanded to ±0.3m / s.

[0144] The above describes the ocean current simulation method based on three-dimensional temperature-salinity values ​​in the embodiment of the present application. The following describes the ocean current simulation system based on three-dimensional temperature-salinity values ​​in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of the ocean current simulation system based on three-dimensional temperature-salinity value includes:

[0145] The acquisition module 201 is used to acquire three-dimensional temperature and salinity data of the target sea area to obtain original three-dimensional temperature and salinity data, and pre-process the original three-dimensional temperature and salinity data to obtain an initial data set of a three-dimensional temperature and salinity field;

[0146] An improvement module 202 is used to improve a preset initial ocean model based on the three-dimensional temperature-salinity field initial data set to obtain a three-dimensional temperature-salinity field simulation model;

[0147] The processing module 203 is used to perform parallel calculation and adaptive grid refinement processing on the three-dimensional temperature-salinity field simulation model to obtain a model operation scheme;

[0148] The analysis module 204 is used to perform three-dimensional visualization processing on the three-dimensional temperature-salinity field initial data set and the model operation scheme to obtain dynamic three-dimensional temperature-salinity field visualization data, and analyze the dynamic three-dimensional temperature-salinity field visualization data to obtain three-dimensional structure data and evolution process data of key ocean phenomena;

[0149] Assimilation module 205, used for performing data assimilation processing on multi-source real-time observation data collected in real time, the three-dimensional structure data and the evolution process data to obtain a fused data set;

[0150] The prediction module 206 is used to perform ocean current prediction calculation on the fused data set to obtain a three-dimensional ocean current prediction field, and perform uncertainty quantification analysis on the three-dimensional ocean current prediction field to obtain ocean current simulation probability distribution data.

[0151] Through the collaboration of the above components, the spatial resolution and accuracy of the initial data of the temperature and salinity field are significantly improved by multi-point and multi-level sampling and three-dimensional scattered point interpolation of the target sea area, providing more reliable input for subsequent simulations. The use of technologies such as Gaussian filtering and spectral analysis effectively removes noise and outliers in the original data and improves data quality. The application of multi-scale decomposition and local regression algorithms enables the model to better capture ocean phenomena of different scales and enhances the precision of the simulation. Secondly, by improving the preset initial ocean model, especially introducing nonlinear feature representation and hybrid dynamics model, the model's ability to describe complex ocean processes is greatly improved. Adaptive particle filter optimization further improves the accuracy of model parameters, making the simulation results closer to the actual ocean conditions. Furthermore, the combination of parallel computing and adaptive grid refinement significantly improves the computational efficiency, making large-scale three-dimensional ocean current simulation possible. The application of dynamic load balancing and critical path method optimizes the utilization of computing resources and reduces computing time. The introduction of adaptive multi-scale interpolation and spectral element method not only improves numerical accuracy but also ensures computational stability. In addition, the application of three-dimensional visualization processing technology, especially volume rendering algorithms and isosurface extraction, provides an intuitive and clear display of the three-dimensional structure and dynamic evolution of ocean phenomena, greatly enhancing the interpretability and practical value of the results. Through morphological analysis and clustering methods, the automatic identification and classification of key ocean phenomena are realized, providing a powerful tool for marine scientific research. At the same time, the introduction of multi-source real-time observation data and data assimilation processing significantly improve the timeliness and accuracy of simulation results. The combination of ensemble Kalman filtering and variational analysis methods ensures the physical consistency and reliability of the fused data. Finally, the application of long short-term memory networks in ocean current prediction calculations improves the model's ability to capture complex time series patterns and enhances the accuracy of predictions.

[0152] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0153] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for simulating ocean currents based on three-dimensional temperature-salinity values, characterized in that: The ocean current simulation method based on three-dimensional temperature-salinity value comprises: Collecting three-dimensional temperature and salinity data of the target sea area to obtain original three-dimensional temperature and salinity data, and preprocessing the original three-dimensional temperature and salinity data to obtain an initial data set of a three-dimensional temperature and salinity field; Based on the three-dimensional temperature-salinity field initial data set, the preset initial ocean model is improved to obtain a three-dimensional temperature-salinity field simulation model; Performing parallel calculation and adaptive grid refinement processing on the three-dimensional temperature-salinity field simulation model to obtain a model operation scheme; Performing three-dimensional visualization processing on the three-dimensional temperature-salinity field initial data set and the model operation scheme to obtain dynamic three-dimensional temperature-salinity field visualization data, and analyzing the dynamic three-dimensional temperature-salinity field visualization data to obtain three-dimensional structure data and evolution process data of key ocean phenomena; Performing data assimilation processing on the multi-source real-time observation data collected in real time, the three-dimensional structure data and the evolution process data to obtain a fused data set; The fused data set is subjected to ocean current prediction calculation to obtain a three-dimensional ocean current prediction field, and the three-dimensional ocean current prediction field is subjected to uncertainty quantification analysis to obtain ocean current simulation probability distribution data.

2. The ocean current simulation method based on three-dimensional temperature-salinity value according to claim 1 is characterized in that: The three-dimensional temperature and salinity data of the target sea area are collected to obtain original three-dimensional temperature and salinity data, and the original three-dimensional temperature and salinity data are preprocessed to obtain an initial data set of a three-dimensional temperature and salinity field, including: Multi-point and multi-level sampling is carried out in the target sea area to obtain discrete temperature and salinity measurement data; The discrete temperature-salinity measurement data are spatially reconstructed by a three-dimensional scattered point interpolation algorithm to obtain an initial continuous temperature-salinity field; Performing Gaussian filtering on the initial continuous temperature-salinity field to obtain a smoothed temperature-salinity distribution; Performing spectrum analysis on the smoothed temperature-salinity distribution by Fourier transform to obtain frequency domain characteristics; Band-pass filtering is performed on the frequency domain features to obtain temperature-salinity change information of a specific scale; The temperature-salinity change information of the specific scale is converted back to the spatial domain by inverse Fourier transform to obtain a multi-scale temperature-salinity field; Performing outlier detection on the multi-scale temperature-salinity field to obtain a set of temperature-salinity anomaly points; The temperature-salinity anomaly point set is corrected by a local regression algorithm to obtain a corrected temperature-salinity field; Performing vertical profile analysis on the corrected temperature-salinity field to obtain thermocline and halocline information; The corrected temperature-salinity field is converted into a density field by a density calculation formula to obtain a three-dimensional temperature-salinity field initial data set.

3. The ocean current simulation method based on three-dimensional temperature-salinity value according to claim 1 is characterized in that: The method improves the preset initial ocean model based on the three-dimensional temperature-salinity field initial data set to obtain a three-dimensional temperature-salinity field simulation model, including: Performing a spatiotemporal correlation analysis on the three-dimensional temperature-salinity field initial data set to obtain the temperature-salinity field change characteristics; Performing multi-scale decomposition on the temperature-salinity field variation characteristics through empirical mode decomposition to obtain an intrinsic mode function set; Performing Hilbert transform on the intrinsic mode function set to obtain instantaneous frequency and amplitude information; Based on the instantaneous frequency and amplitude information, a dynamic characteristic matrix of the temperature-salinity field is constructed to obtain a nonlinear characteristic representation; Performing singular spectrum analysis on the nonlinear characteristic representation to obtain main change modes; Fitting the main variation pattern by support vector regression to obtain a temperature-salinity field prediction function; Performing variational discretization on the control equations of the preset initial ocean model to obtain a discrete model framework; The temperature-salinity field prediction function is embedded in the discrete model framework to obtain a hybrid dynamics model; Optimizing the parameters of the hybrid dynamics model by using an adaptive particle filter to obtain optimized model parameters; Based on the optimized model parameters, the hybrid dynamics model is calibrated to obtain a three-dimensional temperature-salinity field simulation model.

4. The ocean current simulation method based on three-dimensional temperature-salinity value according to claim 1 is characterized in that: The three-dimensional temperature-salt field simulation model is subjected to parallel calculation and adaptive grid refinement processing to obtain a model operation scheme, including: Decomposing the three-dimensional temperature-salinity field simulation model in the spatial domain to obtain sub-region data blocks; Allocate tasks to the sub-region data blocks through a dynamic load balancing algorithm to obtain a parallel computing task list; Performing data dependency analysis on the parallel computing task list to obtain a task execution graph; Based on the task execution graph, computing resources are scheduled by a critical path method to obtain a parallel execution plan; Monitor the gradient changes during the calculation process to obtain the grid refinement requirement index; The grid refinement requirement index is processed by an adaptive multi-scale interpolation algorithm to obtain a dynamic grid structure; Performing boundary condition matching on the dynamic grid structure to obtain a continuous grid layout; Based on the continuous grid layout, the computational equations are discretized by the spectral element method to obtain a high-order numerical format; Performing stability analysis on the high-order numerical format to obtain time step constraint conditions; The time step constraint condition is processed by an adaptive time stepping algorithm to obtain a model operation scheme.

5. The ocean current simulation method based on three-dimensional temperature-salinity value according to claim 1 is characterized in that: The three-dimensional visualization processing is performed on the three-dimensional temperature-salinity field initial data set and the model operation scheme to obtain dynamic three-dimensional temperature-salinity field visualization data, and the dynamic three-dimensional temperature-salinity field visualization data is analyzed to obtain three-dimensional structure data and evolution process data of key ocean phenomena, including: Performing data interpolation on the three-dimensional temperature-salinity field initial data set to obtain a high-resolution temperature-salinity field grid; Rendering the high-resolution temperature-salinity field grid by a volume rendering algorithm to obtain a three-dimensional temperature-salinity field volume image; Performing color mapping on the three-dimensional temperature-salinity field volume image to obtain a color-enhanced visualization effect; Based on the model operation scheme, a time-series animation is generated for the color-enhanced visualization effect to obtain dynamic three-dimensional temperature-salinity field visualization data; The dynamic three-dimensional temperature-salinity field visualization data is processed by an isosurface extraction algorithm to obtain an isosurface sequence of key ocean phenomena; Performing morphological analysis on the isosurface sequence to obtain three-dimensional structural characteristic parameters; Based on the three-dimensional structural characteristic parameters, key ocean phenomena are classified by cluster analysis to obtain phenomenon type classification; Performing time series analysis on the phenomenon type classification to obtain a characteristic vector of the evolution process; By principal component analysis, the eigenvector of the evolution process is reduced in dimension to obtain a key evolution pattern; The key evolution patterns are interpolated in time and space to obtain three-dimensional structural data and evolution process data of key ocean phenomena.

6. The ocean current simulation method based on three-dimensional temperature-salinity value according to claim 1 is characterized in that: The data assimilation processing is performed on the multi-source real-time observation data, the three-dimensional structure data and the evolution process data collected in real time to obtain a fused data set, including: Performing data quality control on the multi-source real-time observation data to obtain cleaned observation data; Performing spatiotemporal interpolation on the cleaned observation data to obtain a gridded observation field; Performing covariance analysis on the gridded observation field and the three-dimensional structure data to obtain a data correlation matrix; Based on the data correlation matrix, the data is assimilated by using an ensemble Kalman filter algorithm to obtain preliminary fusion data; Performing a time series consistency check on the preliminary fusion data and the evolution process data to obtain corrected fusion data; Performing constrained optimization on the corrected fusion data by using a variational analysis method to obtain physically consistent data; Decomposing the physical consistency data to obtain a multi-scale feature field; Performing denoising on the multi-scale feature field by wavelet transform to obtain a smoothed feature field; Performing spectrum analysis on the smoothed characteristic field to obtain main frequency components; Based on the main frequency components, data is reconstructed by an adaptive spectrum synthesis method to obtain a fused data set.

7. The ocean current simulation method based on three-dimensional temperature-salinity value according to claim 1 is characterized in that: The performing of ocean current prediction calculation on the fused data set to obtain a three-dimensional ocean current prediction field, and performing uncertainty quantification analysis on the three-dimensional ocean current prediction field to obtain ocean current simulation probability distribution data, including: Performing spatiotemporal decomposition on the fused data set to obtain multi-scale ocean dynamic characteristics; The multi-scale ocean dynamic characteristics are predicted in time series by using a long short-term memory network to obtain preliminary ocean current prediction results; Performing physical constraint correction on the preliminary ocean current prediction results to obtain a physically consistent prediction field; Based on the physical consistency prediction field, multiple groups of disturbance fields are generated by Monte Carlo method to obtain set prediction members; Performing statistical analysis on the set prediction members to obtain a prediction mean field and a variance field; Performing probability distribution fitting on the predicted mean field and variance field by a kernel density estimation method to obtain a preliminary probability distribution model; Performing bias correction on the preliminary probability distribution model to obtain a revised probability distribution; Based on the modified probability distribution, the prediction confidence interval is calculated by Bayesian inference method to obtain the uncertainty quantification result; Performing spatial interpolation on the uncertainty quantification result to obtain a continuous uncertainty field; The continuous uncertainty field is classified by an adaptive threshold method to obtain ocean current simulation probability distribution data.

8. A current simulation system based on three-dimensional temperature and salinity values, characterized in that: The ocean current simulation system based on three-dimensional temperature-salinity values ​​comprises: The acquisition module is used to collect three-dimensional temperature and salinity data of the target sea area to obtain original three-dimensional temperature and salinity data, and pre-process the original three-dimensional temperature and salinity data to obtain an initial data set of the three-dimensional temperature and salinity field; An improvement module, used to improve a preset initial ocean model based on the three-dimensional temperature-salinity field initial data set to obtain a three-dimensional temperature-salinity field simulation model; A processing module, used for performing parallel calculation and adaptive grid refinement processing on the three-dimensional temperature-salinity field simulation model to obtain a model operation scheme; An analysis module is used to perform three-dimensional visualization processing on the three-dimensional temperature-salinity field initial data set and the model operation scheme to obtain dynamic three-dimensional temperature-salinity field visualization data, and to analyze the dynamic three-dimensional temperature-salinity field visualization data to obtain three-dimensional structure data and evolution process data of key ocean phenomena; An assimilation module is used to perform data assimilation processing on the multi-source real-time observation data collected in real time, the three-dimensional structure data and the evolution process data to obtain a fused data set; The prediction module is used to perform ocean current prediction calculation on the fused data set to obtain a three-dimensional ocean current prediction field, and to perform uncertainty quantification analysis on the three-dimensional ocean current prediction field to obtain ocean current simulation probability distribution data.

Images