Ocean data assimilation methods, devices, media, and products

Abstract

This application discloses a method, apparatus, medium, and product for ocean data assimilation. The method includes: acquiring matching data of first and second type observation data within first and second assimilation time windows, where the first and second type observation data are observation data from different observation periods corresponding to the region simulated by a regional ocean model; mapping the first and second model average states corresponding to the regional ocean model within the first and second assimilation time windows to the three-dimensional spatial positions of the first and second type observation data, obtaining first and second model estimated data; determining first and second innovation vectors of the regional ocean model based on the difference between the first and second model estimated data and the corresponding matching data; determining the target analysis increment of the regional ocean model based on the first and second innovation vectors; and updating the target analysis increment to the regional ocean model using a relaxed incremental analysis update method. This application can improve the accuracy of ocean data assimilation.

Description

Technical Field

[0001] This application belongs to the field of marine data processing technology, and in particular relates to a marine data assimilation method, device, medium and product. Background Technology

[0002] Ocean data assimilation is used to fuse simulation results from physical ocean numerical models with real ocean observation data to generate temperature, salinity, and current analysis data that more closely reflect the actual ocean conditions.

[0003] Ocean data assimilation typically comprises three core components. The first is the physical ocean model, a mathematical model solved on a computer using partial differential equations of hydrodynamics, used to predict the temporal and spatial variations of seawater temperature, salinity, and current velocity. The second is observational data, primarily divided into satellite remote sensing observations and in-situ observations of the ocean interior. Satellite observations can cover a vast ocean surface, providing information such as sea surface temperature and sea level, but cannot directly capture the conditions below the surface. In-situ observations measure temperature and salinity at different depths using buoys or moorings deployed in the ocean, but the distribution of observation points is sparse and uneven. The third is the data assimilation method, used to optimally fuse model forecasts with observational data, thereby correcting the model's state.

[0004] Currently, when assimilating marine data, a single fixed-length time window is typically used to process all observation data uniformly. However, forcibly comparing and fusing data with inconsistent time scales within the same time window can lead to scale confusion and increased errors in the assimilation analysis results. Summary of the Invention

[0005] This application provides a method, apparatus, medium, and product for marine data assimilation, which can improve the accuracy of marine data assimilation.

[0006] On the one hand, embodiments of this application provide a marine data assimilation method, the method comprising: Acquire matching data of the first type of observation data within the first assimilation time window, and matching data of the second type of observation data within the second assimilation time window; the first type of observation data and the second type of observation data are observation data of different observation periods corresponding to the region simulated by the regional ocean model; The first model average state of the regional ocean model within the first assimilation time window is mapped to the three-dimensional spatial position of the first type of observation data to obtain the first model estimation data; and the second model average state of the regional ocean model within the second assimilation time window is mapped to the three-dimensional spatial position of the second type of observation data to obtain the second model estimation data. Based on the difference between the first model estimation data and the corresponding matching data, a first innovation vector of the regional ocean model is determined, and based on the difference between the second model estimation data and the corresponding matching data, a second innovation vector of the regional ocean model is determined. Based on the first and second information vectors, the target analysis increment of the regional ocean model is determined; The target analysis is incrementally updated to the regional ocean model using a relaxed incremental analysis update method.

[0007] On the other hand, embodiments of this application provide a marine data assimilation device, the device comprising: The acquisition module is used to acquire matching data of the first type of observation data within a first assimilation time window, and matching data of the second type of observation data within a second assimilation time window; the first type of observation data and the second type of observation data are observation data of different observation periods corresponding to the region simulated by the regional ocean model; The mapping module is used to map the average state of the first mode of the regional ocean model within the first assimilation time window to the three-dimensional spatial position of the first type of observation data to obtain the first mode estimation data, and to map the average state of the second mode of the regional ocean model within the second assimilation time window to the three-dimensional spatial position of the second type of observation data to obtain the second mode estimation data. The first determining module is used to determine a first innovation vector of the regional ocean model based on the difference between the first model estimation data and the corresponding matching data, and to determine a second innovation vector of the regional ocean model based on the difference between the second model estimation data and the corresponding matching data. The second determining module is used to determine the target analysis increment of the regional ocean model based on the first innovation vector and the second innovation vector; An assimilation module is used to update the target analysis incrementally to the regional ocean model in a relaxed manner using incremental analysis updates.

[0008] In another aspect, embodiments of this application provide an electronic device, the device comprising: a processor and a memory storing computer program instructions; When the processor executes the computer program instructions, it implements the ocean data assimilation method as described in one aspect.

[0009] In another aspect, embodiments of this application provide a computer-readable storage medium storing computer program instructions, which, when executed by a processor, implement the marine data assimilation method as described in one aspect.

[0010] In another aspect, embodiments of this application provide a computer program product in which instructions, when executed by a processor of an electronic device, cause the electronic device to perform the marine data assimilation method as described in one aspect.

[0011] The marine data assimilation method, device, medium, and product of this application acquire matching data of first and second types of observation data within first and second assimilation time windows, respectively. The average states of the first and second models of a regional marine model within the first and second assimilation time windows are mapped to the three-dimensional spatial positions of the corresponding first and second types of observation data, resulting in estimated data for the first and second models. This approach is compatible with the time scales of observation data from different observation periods, enabling effective fusion of multi-source observation data across time and space scales and reducing assimilation errors caused by time and space scale mismatches. Based on the differences between the estimated data of the first and second models and their corresponding matching data, first and second innovation vectors corresponding to the regional marine model are determined. Based on these first and second innovation vectors, the target analysis increment of the regional marine model is determined. The target analysis increment is updated to the regional marine model using a relaxed incremental analysis update method. This reduces the numerical integration shock caused by applying corrections all at once to the forecast state of the regional marine model, smoothly and gradually absorbing observation information, maintaining the geostrophic equilibrium and temperature-salinity field structure of the regional marine model, and improving the accuracy of marine data assimilation. Attached Figure Description

[0012] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a flowchart illustrating a marine data assimilation method provided in one embodiment of this application; Figure 2 This is an algorithmic framework diagram of a marine data assimilation method provided in one embodiment of this application; Figure 3 This is an interface diagram comparing the root mean square error time series for performance verification of the marine data assimilation method provided in one embodiment of this application; Figure 4 This is a schematic diagram of the structure of a marine data assimilation device provided in another embodiment of this application; Figure 5 This is a schematic diagram of the structure of an electronic device provided in another embodiment of this application. Detailed Implementation

[0014] The features and exemplary embodiments of various aspects of this application will be described in detail below. To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain this application and not to limit it. For those skilled in the art, this application can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this application by illustrating examples.

[0015] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes said element.

[0016] First, let me introduce the technical terms used in this application.

[0017] Regional ocean models are a class of ocean circulation models designed for finite sea areas. They perform high-resolution numerical simulations of seawater motion and its thermodynamic state by solving primitive hydrodynamic equations. Unlike global ocean models that cover the entire globe, regional ocean models limit the computational domain to a specific sea area. Horizontally, they discretize this area using orthogonal curved grids or unstructured grids, while vertically they employ variable-layered coordinates stretched according to topography. By covering the finite computational domain with high-resolution grids, regional ocean models can accurately characterize small- and medium-scale ocean dynamic processes such as tidal fluctuations, coastal current systems, mesoscale eddies, and fronts at horizontal resolutions on the order of hundreds of meters.

[0018] Ocean data assimilation is used to fuse simulation results from physical ocean numerical models with real ocean observation data to generate temperature, salinity, and current analysis data that more closely reflect the actual ocean conditions.

[0019] Ocean data assimilation typically comprises three core components. The first is the physical ocean model, a mathematical model solved on a computer using partial differential equations of hydrodynamics, used to predict the temporal and spatial variations of seawater temperature, salinity, and current velocity. The second is observational data, primarily divided into satellite remote sensing observations and in-situ observations of the ocean interior. Satellite observations can cover a vast ocean surface, providing information such as sea surface temperature and sea level, but cannot directly capture the conditions below the surface. In-situ observations measure temperature and salinity at different depths using buoys or moorings deployed in the ocean, but the distribution of observation points is sparse and uneven. The third is the data assimilation method, used to optimally fuse model forecasts with observational data, thereby correcting the model's state.

[0020] Currently, when assimilating marine data, a single fixed-length time window is typically used to process all observation data uniformly. However, forcibly comparing and fusing data with inconsistent time scales within the same time window can lead to scale confusion and increased errors in the assimilation analysis results.

[0021] To address the problems of the prior art, embodiments of this application provide a marine data assimilation method, apparatus, medium, and product. The marine data assimilation method provided in this application embodiment is described below.

[0022] Figure 1 This is a flowchart illustrating a marine data assimilation method provided in one embodiment of this application. Figure 1 As shown, the execution subject of the marine data assimilation method provided in this application embodiment is an electronic device, and the method includes steps 101 to 105.

[0023] Step 101: Obtain matching data of the first type of observation data within the first assimilation time window, and matching data of the second type of observation data within the second assimilation time window; the first type of observation data and the second type of observation data are observation data of different observation periods corresponding to the region simulated by the regional ocean model.

[0024] The first type of observational data can be the average state estimate of the three-dimensional temperature and salinity field of the ocean within an observation period, generated through objective analysis based on multi-source historical observations. The second type of observational data can be the actual observational data of the sea area simulated by the ocean model covering the region.

[0025] For example, the first type of observation data may be gridded three-dimensional temperature and salinity state estimation data of the sea area simulated by the ocean model covering the region, and the second type of observation data may be sea surface remote sensing observation data and in-situ temperature and salinity profile observation data of the sea area simulated by the ocean model covering the region.

[0026] Among them, sea surface remote sensing observation data includes satellite observation data of sea surface temperature and sea surface height, and in-situ temperature and salinity profile observation data includes vertical temperature and salinity distribution data obtained by platforms such as Argo buoys.

[0027] The first assimilation time window is used to match the observation period of the first type of observation data. The length of the first assimilation time window can be greater than or equal to the observation period of the first type of observation data.

[0028] The second assimilation time window is used to match the observation period of the second type of observation data. The length of the second assimilation time window can be greater than the observation period of the second type of observation data, for example, it can be an integer multiple of the observation period of the second type of observation data.

[0029] In one implementation, the baseline observation value, after time-scale alignment of the observed data, that matches the corresponding assimilation time window in the time dimension can be determined as the matching data of the observed data within the assimilation time window. Specifically, matching data of the first type of observed data within the first assimilation time window and matching data of the second type of observed data within the second assimilation time window can be obtained through time interpolation.

[0030] Step 102: Map the first model average state of the regional ocean model within the first assimilation time window to the three-dimensional spatial position of the first type of observation data to obtain the first model estimation data; and map the second model average state of the regional ocean model within the second assimilation time window to the three-dimensional spatial position of the second type of observation data to obtain the second model estimation data.

[0031] Model-averaged state refers to the time average of the predicted state of a regional ocean model within an assimilation time window.

[0032] Model estimation data refers to the physical quantity values ​​obtained after the mean state of a regional ocean model is transformed into the three-dimensional spatial location of the observation data through spatial mapping. It represents the estimation value of physical quantities at the observation location by the regional ocean model.

[0033] In one implementation, the forecast states of a regional ocean model within a first assimilation time window are averaged over time to obtain a first model-averaged state; the forecast states of the regional ocean model within a second assimilation time window are averaged over time to obtain a second model-averaged state. Then, a spatial mapping method is used to map the first model-averaged state from the three-dimensional grid of the regional ocean model to the grid locations of the first type of observation data to obtain the first model estimated data; the second model-averaged state is mapped from the three-dimensional grid of the regional ocean model to the observation station or profile locations of the second type of observation data to obtain the second model estimated data. The specific method of spatial mapping can be determined based on the spatial distribution characteristics of the observation data.

[0034] Step 103: Determine the first innovation vector of the regional ocean model based on the difference between the first model estimated data and the corresponding matching data, and determine the second innovation vector of the regional ocean model based on the difference between the second model estimated data and the corresponding matching data.

[0035] The first innovation vector is used to quantify the deviation of the regional ocean model's forecast state from the first type of observation data. It can be obtained by subtracting the values ​​of the matched data from the first model's estimated data and the first type of observation data at their corresponding spatial locations.

[0036] The second innovation vector is used to quantify the deviation of the regional ocean model's forecast state from the second type of observation data. It can be obtained by subtracting the values ​​of the matched data from the second model's estimated data and the second type of observation data at corresponding observation locations.

[0037] Step 104: Based on the first information vector and the second information vector, determine the target analysis increment of the regional ocean model.

[0038] The target analysis increment is used to correct deviations in the forecast status of regional ocean models from the observed data.

[0039] In one implementation, the forecast state of a regional ocean model can be updated based on a first and a second innovation vector. During the update process, a Kalman filter is used to map the first and second innovation vectors from the observation space back to the model state space, obtaining a first and a second analysis increment. This ensures that the first and second analysis increments reside in the same state space as the forecast state of the regional ocean model. The vector sum of the first and second analysis increments can then be determined as the target analysis increment for the regional ocean model.

[0040] Step 105: Update the target analysis incrementally to the regional ocean model in a relaxed manner using incremental analysis updates.

[0041] Nudging in Incremental Analysis Update (IAU) is an assimilation method that smoothly updates the model state. By employing nudging in IAU, the target analysis increment is evenly distributed across each time step of the regional ocean model's re-integration, and the corresponding share is introduced into the model's tendency term before each integration step. This allows for a gradual and smooth update of the model state, resulting in an assimilated three-dimensional ocean analysis field.

[0042] The method provided in this application acquires matching data of first and second type observation data within first and second assimilation time windows, respectively. It maps the average states of the first and second models of a regional ocean model within the first and second assimilation time windows to the three-dimensional spatial positions of the corresponding first and second type observation data, thus obtaining the estimated data of the first and second models. This method is compatible with the time scales of observation data from different observation periods, enabling effective fusion of multi-source observation data at both temporal and spatial scales, and reducing assimilation errors caused by time and spatial scale mismatches. Based on the differences between the estimated data of the first and second models and their corresponding matching data, it determines the first and second innovation vectors corresponding to the regional ocean model. Based on these first and second innovation vectors, it determines the target analysis increment of the regional ocean model. The target analysis increment is then updated to the regional ocean model using a relaxed incremental analysis update method. This reduces the numerical integration shock caused by applying corrections all at once to the forecast state of the regional ocean model, smoothly and gradually absorbing observation information, maintaining the geostrophic equilibrium and temperature-salinity field structure of the regional ocean model, and improving the accuracy of ocean data assimilation.

[0043] In some embodiments, the first assimilation time window is greater than or equal to the observation period of the first type of observation data. The process of obtaining matching data of the first type of observation data within the first assimilation time window in step 101 includes steps 201 to 203.

[0044] Step 201: Identify at least two periods of observation data in the first type of observation data where the observation period overlaps with the first assimilation time window.

[0045] An observation period refers to the time range covered by a single observation data period. For the first type of observation data, each data period represents the average state within an observation cycle, and each data period has its corresponding start and end times.

[0046] Since the first assimilation time window is greater than or equal to the observation period of the first type of observation data, it may cover the time range of two or more periods of observation data on the time axis. Based on the start and end times of the first assimilation time window and the start and end times of the corresponding observation period for each period of observation data, at least two periods of observation data whose observation periods overlap with the first assimilation time window can be selected from the first type of observation data.

[0047] Step 202: Determine the weighting coefficients of each period of the observation data in the at least two periods of observation data based on the length of time overlap between each period of observation data and the first assimilation time window.

[0048] Temporal overlap refers to the fact that the observation period of the first observation data and the first assimilation time window have a common coverage on the time axis.

[0049] In one implementation, for each selected period of observation data, the length of the time overlap between its observation period and the first assimilation time window can be calculated. Based on the length of the time overlap for each period, the weight coefficient corresponding to each period of observation data can be calculated. The weight coefficient can be positively correlated with the length of the time overlap, that is, the longer the time overlap, the larger the weight coefficient. Specifically, the normalized value of the time overlap length of each period of data can be used as the weight coefficient.

[0050] Step 203: Based on the weighting coefficients of the observation data in each period, perform a weighted average of the observation data in at least two periods to generate matching data for the first type of observation data within the first assimilation time window.

[0051] Since each observation in the first type of observation data represents the average state within its observation period, and the first assimilation time window may not completely coincide with the observation period of any single period product—for example, the first type of observation data is released every Wednesday, but the 7-day model forecast provided by the first assimilation time window may not start every Wednesday and forecast to the following Wednesday—when the start and end times of the first assimilation time window do not completely coincide with the observation period boundary of the first type of observation data, directly using a certain period of observation data as matching data will introduce time representativeness error.

[0052] Therefore, by weighting the observation data from at least two periods according to the length of time overlap, specifically, the observation data from each period can be weighted and averaged point by point in three-dimensional space. The weighted average result is used as the matching data for the first type of observation data within the first assimilation time window. This can generate an observation benchmark value that is precisely matched with the first assimilation time window in the time dimension. This not only allows the matching data to accurately reflect the reasonable proportion of the contribution of each period of observation data within the first assimilation time window, but also achieves precise alignment of the first type of observation data with the first assimilation window in terms of time scale. This ensures that subsequent comparisons with the model average state are based on the same time scale.

[0053] In some embodiments, the process of obtaining matching data of the second type of observation data within the second assimilation time window in step 101 includes steps 301 to 302.

[0054] Step 301: Determine at least two periods of observation data in the second type of observation data that fall within the second assimilation time window.

[0055] In one implementation, the observation period of the second type of observation data is shorter than the second assimilation time window. When the observation period of a period of observation data is completely contained within the second assimilation time window on the time axis, that is, the entire time span of the period of observation data is within the boundary range of the second assimilation time window, then it is determined that the period of observation data falls within the second assimilation time window.

[0056] For example, the second assimilation time window is 3 or 7 days, and the second type of observation data includes sea surface remote sensing observation data and in-situ temperature and salinity profile observation data. Among them, the sea surface remote sensing observation data includes sea surface temperature satellite observation data and sea surface height satellite observation data. Although satellite and in-situ observation data are released daily, their data can only represent a large-scale signal. It is necessary to regard the average state over 3 to 7 days as a relatively accurate observation of the true state and integrate it into the physical model.

[0057] Step 302: Average the observation data from at least two periods to obtain the matching data of the second type of observation data within the second assimilation time window.

[0058] In one implementation, the single-period observations of the second type of observation data may contain high-frequency noise components such as tidal fluctuations, intraday temperature variations, and random errors of the observation equipment. By averaging the observation data from multiple periods within the second assimilation time window—specifically, by performing an arithmetic mean on the observation data at the corresponding spatial locations for each period—and using this arithmetic mean as the matching data for the second type of observation data within the second assimilation time window, high-frequency fluctuations can be smoothed out, accurately reflecting the steady-state characteristics of the ocean state within the second assimilation time window. Furthermore, this achieves precise alignment between the second type of observation data and the second assimilation window on the same time scale, ensuring that subsequent comparisons with the model's average state are based on the same time scale.

[0059] In some embodiments, the three-dimensional spatial location of the regional ocean model is a first regular three-dimensional grid, and the three-dimensional spatial location of the first type of observation data is a second regular three-dimensional grid. Step 102 is a process of mapping the average state of the first model within the first assimilation time window of the regional ocean model to the three-dimensional spatial location of the first type of observation data to obtain the first model estimation data, including step 401.

[0060] Step 401: Using three-dimensional spatial interpolation, the physical quantity values ​​of the average state of the first mode at each grid point of the first regular three-dimensional grid are mapped to the grid point positions of the second regular three-dimensional grid to obtain the estimated data of the first mode.

[0061] The first-rule 3D grid is a structured 3D grid used in the spatial discretization of regional ocean models. The first-rule 3D grid can use orthogonal curvilinear coordinates or spherical coordinates in the horizontal direction, and variable layered coordinates stretched according to the terrain in the vertical direction. Each grid point of the first-rule 3D grid stores numerical values ​​of physical quantities such as temperature, salinity, and current velocity.

[0062] The second-rule 3D grid is a structured 3D grid used spatially for the first type of observation data. The second-rule 3D grid can be a latitude-longitude grid, arranged at fixed latitude-longitude intervals in the horizontal direction, and using standard depth layers or isodense surface coordinates in the vertical direction.

[0063] Three-dimensional spatial interpolation is a process of using known physical quantity values ​​at grid points in a three-dimensional mesh to mathematically deduce the physical quantity values ​​at a specified location in another three-dimensional mesh. For structured meshes, three-dimensional spatial interpolation methods include trilinear interpolation, cubic spline interpolation, and combinations of vertically independent interpolation and horizontal interpolation.

[0064] In one implementation, the physical quantity values ​​of the first mode's average state at each grid point of the first regular 3D grid can be used as the input data source, and the grid point positions of the second regular 3D grid can be used as the output target positions. For each grid point position of the second regular 3D grid, a 3D spatial interpolation method is used to calculate the interpolation result at the grid point position using the neighboring grid points and their physical quantity values ​​in the first regular 3D grid. After traversing all grid point positions of the second regular 3D grid, the interpolation results of all grid points are collected to form the first mode estimation data.

[0065] During the interpolation process, since the vertical coordinate systems of the first regular 3D grid and the second regular 3D grid may be different, depth linear interpolation or nonlinear interpolation can be performed in the vertical direction. In the horizontal direction, bilinear interpolation or bicubic interpolation is used according to the latitude and longitude distribution characteristics of the first regular 3D grid and the second regular 3D grid.

[0066] The method provided in this application uses three-dimensional spatial interpolation to map the physical quantity values ​​of the average state of the first mode at each grid point of the first regular three-dimensional grid to each grid point position of the second regular three-dimensional grid. This can maintain the spatial continuity and gradient structure of the mode's temperature and salt field, reduce spatial abrupt changes or information loss caused by grid transformation, and achieve accurate spatial transformation of the mode state from the first regular three-dimensional grid to the second regular three-dimensional grid.

[0067] In one embodiment, the three-dimensional spatial location of the regional ocean model is a first regular three-dimensional grid, and the second type of observation data includes sea surface remote sensing observation data and in-situ temperature and salinity profile observation data. The three-dimensional spatial location of the sea surface remote sensing observation data is a discrete point on the sea surface, and the three-dimensional spatial location of the in-situ temperature and salinity profile observation data is a discrete point on the profile. The second model estimation data includes model estimation values ​​at each discrete point on the sea surface and model estimation values ​​at each depth layer at each discrete point on the profile. Step 102 maps the second model average state of the regional ocean model within the second assimilation time window to the three-dimensional spatial location of the second type of observation data to obtain the second model estimation data. The processing steps include steps 501 to 503.

[0068] Sea surface discrete points are the spatial locations of sea surface remote sensing data. They are located on the ocean surface and are irregularly distributed. There is no fixed grid topology between the observation points, and each observation point is uniquely determined by its latitude and longitude coordinates.

[0069] Discrete points in a profile are the spatial locations of in-situ temperature, salinity, and thermal profile observation data. These discrete points are located within the ocean and are irregularly distributed. Each observation point is determined by its latitude and longitude coordinates and the depth values ​​of each depth layer. An observation profile contains multiple depth layers, and each depth layer corresponds to a three-dimensional spatial location.

[0070] Step 501: Using horizontal spatial interpolation, the physical quantity values ​​of the second model's average state located at each grid point on the sea surface in the first regular three-dimensional grid are mapped to each discrete point on the sea surface to obtain the model estimate value at each discrete point on the sea surface.

[0071] Since the spatial location of sea surface remote sensing data consists of discrete points on the sea surface, distributed only across the ocean surface, a horizontal spatial interpolation method can be used to map the average state of the second model to each discrete sea surface point. Specifically, for each discrete sea surface point, its set of neighboring grid points in the model's horizontal sea surface grid can be determined based on its latitude and longitude coordinates. The physical quantity values ​​at that point are then calculated using bilinear interpolation or cubic spline interpolation. After traversing all discrete sea surface points, the set of interpolation results for each point constitutes the model estimate for the discrete sea surface points.

[0072] Step 502: Using horizontal spatial interpolation, the physical quantity values ​​of the average state of the second mode located at each grid point inside the ocean in the first regular three-dimensional grid are mapped to the latitude and longitude positions of each discrete point of the profile to obtain the horizontal interpolation result.

[0073] Step 503: Vertical depth interpolation is used to map the horizontal interpolation results to each depth layer at each discrete point of the profile, so as to obtain the model estimate value of each depth layer at each discrete point of the profile.

[0074] Since the spatial locations of in-situ temperature and salinity profile observation data are discrete points distributed across multiple depth layers within the ocean, for each discrete point, its set of neighboring grid points in the model's horizontal grid can be determined first. Then, horizontal interpolation methods are used to calculate the physical quantity values ​​of the model at that latitude and longitude in each vertical layer, obtaining the horizontal interpolation results. Next, vertical linear or nonlinear interpolation is used to map the model's vertical layer values ​​to the observed depth layers, obtaining model estimates for each depth layer at that discrete point. Finally, the model estimates at each sea surface discrete point and the model estimates for each depth layer at each discrete point are combined to obtain the second model estimation data.

[0075] The method provided in this application, through horizontal spatial interpolation, maps the average state of the second model to each discrete point on the sea surface, and through horizontal spatial interpolation and vertical depth interpolation, maps the average state of the second model to each discrete point on the profile. This enables the mapped model estimation data and the matching data to correspond completely in spatial location, thereby improving the estimation accuracy of model physical quantities at each observation location.

[0076] In some embodiments, step 104, which determines the target analysis increment of the regional ocean model based on the first innovation vector and the second innovation vector, includes steps 601 to 602.

[0077] Step 601: Transform the first innovation vector and the second innovation vector to the three-dimensional spatial position of the regional ocean model to obtain the first analysis increment corresponding to the first innovation vector and the second analysis increment corresponding to the second innovation vector.

[0078] Since the first innovation vector is defined in the observation space of the first type of observation data and the second innovation vector is defined in the observation space of the second type of observation data, it is necessary to convert the first innovation vector and the second innovation vector to the three-dimensional spatial position of the regional ocean model.

[0079] In one implementation, the first and second innovation vectors can be mapped from their respective observation spaces back to the three-dimensional grid space of the regional ocean model using Kalman gain matrices. Specifically, the first analysis increment is obtained by multiplying the first Kalman gain matrix by the first innovation vector, and the second analysis increment is obtained by multiplying the second Kalman gain matrix by the second innovation vector. The dimensions of both the first and second analysis increments are equal to the total number of state variables in the regional ocean model. The first Kalman gain matrix distributes the difference values ​​at each observation location in the first innovation vector to each grid point of the three-dimensional grid of the regional ocean model according to the ensemble forecast error covariance structure. The second Kalman gain matrix distributes the difference values ​​at each observation location in the second innovation vector to each grid point of the three-dimensional grid of the regional ocean model according to the ensemble forecast error covariance structure.

[0080] The calculation of the Kalman gain matrix depends on the observation error of the first type of observation data and the transpose of the first observation operator, as well as the observation error of the second type of observation data and the transpose of the second observation operator. The transpose of the first observation operator back-projects information from the observation space of the first type of observation data to the model space, and the transpose of the second observation operator back-projects information from the observation space of the second type of observation data to the model space. The first observation operator is a spatial mapping operator constructed to map the model-averaged state of the regional ocean model to the three-dimensional spatial position of the first type of observation data, and the second observation operator is a spatial mapping operator constructed to map the model-averaged state of the regional ocean model to the three-dimensional spatial position of the second type of observation data. The observation error of the first type of observation data determines the reliability of the difference values ​​at each observation position in the first innovation vector; the smaller the observation error, the greater the contribution weight of the innovation at the corresponding position in the analysis increment. The observation error of the second type of observation data determines the contribution weight of the difference values ​​at each observation position in the second innovation vector. The observation error of the first type of observation data is a measure of uncertainty introduced during the generation of the first type of observation data.

[0081] Under most normal operating conditions, both forecasts and observations have errors. For regions or variables where the model forecast variance is small and the model is highly deterministic, the Kalman gain matrix decreases to suppress the correction magnitude, thus preventing noisy observations from compromising a good forecast. For observation points with small observation errors, the gain increases to amplify the correction magnitude, ensuring that high-precision observations can effectively correct the model.

[0082] In some embodiments, the first type of observation data is gridded three-dimensional temperature and salinity state estimation data generated by objective analysis and statistical fusion of multi-source observations. The observation errors of the first type of observation data include the errors inherent in each input observation data during the objective analysis process, the non-uniformity of the observation distribution, and the interpolation algorithm error. The observation errors of the second type of observation data include: instrument noise and inversion algorithm errors in sea surface remote sensing observation data, and sensor drift and positioning bias in in-situ temperature and salinity profile observation data.

[0083] In one implementation, the first innovation vector can be weighted element-wise using the observation error of the first type of observation data, and then the first innovation vector can be mapped from the observation space of the first type of observation data back to the three-dimensional spatial position of the regional ocean model according to the transpose of the first observation operator to obtain the first analysis increment.

[0084] First, the observation error of the second type of observation data is used to perform element-wise weighting on the second innovation vector. Then, according to the transpose of the second observation operator, the second innovation vector is mapped from the observation space of the second type of observation data back to the three-dimensional spatial position of the regional ocean model to obtain the second analysis increment.

[0085] The observation operator, denoted by H, is a mapping function from the model state space to the observation space. The specific form of the observation operator is determined by the spatial distribution characteristics of the observation data. For gridded 3D temperature and salinity state estimation data, the observation operator is a 3D spatial interpolation, mapping the physical quantity values ​​at model grid points to the locations of each grid point in the observation data grid, generating the model's estimate of the physical quantity at each observation grid point. For sea surface remote sensing observation data, the observation operator is a horizontal spatial interpolation, mapping the physical quantity values ​​at each grid point on the model sea surface to the latitude and longitude locations of each discrete point on the sea surface.

[0086] The transpose H of the observation operator TIt can backpropagate innovation information from the observation space to the model state space. For example, during forward mapping of the observation operator, the observation operator aggregates the physical quantity values ​​at each grid point of the regional ocean model into a model estimate at the observation location through spatial interpolation weights. During backward mapping of the transposed observation operator, the transposed observation operator, through the same spatial interpolation weight structure, backpropagates the innovation values ​​at the observation location to each model grid point that contributed to the estimate at that observation location, thus obtaining the analysis increment at each model grid point.

[0087] Step 602: The vector sum of the first and second analysis increments is determined as the target analysis increment.

[0088] The first and second analysis increment vectors can be added together in the vector space, that is, the values ​​at each corresponding grid point are added together to obtain the target analysis increment. The value of the target analysis increment at each grid point is equal to the algebraic sum of the first and second analysis increments at that grid point, representing the total correction that the state variable at that grid point needs to undergo under the joint constraints of the two types of observation data.

[0089] The method provided in this application embodiment determines the target analysis increment by the vector sum of the first analysis increment and the second analysis increment, and fuses the deviation information of different observation periods to achieve the collaborative fusion of observation information at multiple time scales, so that the target analysis increment simultaneously includes large-scale background constraints of long periods and high-frequency detail constraints of short periods.

[0090] In one embodiment, before obtaining the matching data of the first type of observation data within the first assimilation time window and the matching data of the second type of observation data within the second assimilation time window in step 101, the method further includes: The first assimilation time window is determined based on the data assimilation period of the regional ocean model and the observation period of the first type of observation data; The second assimilation time window is determined based on the data assimilation period of the regional ocean model and the observation period of the second type of observation data.

[0091] The data assimilation period refers to the length of time covered by a complete data assimilation cycle performed by a regional ocean model. The data assimilation period can be the same as the observation period for Type I observation data, such as seven days.

[0092] In one implementation, the data assimilation period can be an integer multiple of the observation period of the first type of observation data, and the length of the first assimilation time window can be equal to the data assimilation period. The length of the second assimilation time window can be an integer multiple of the observation period of the second type of observation data, and the boundary of the second assimilation time window is aligned with the boundary of the second type of observation data. The second assimilation time window is nested within the first assimilation time window.

[0093] Step 105, which involves incrementally updating the target analysis to the regional ocean model using a relaxed approach, includes: Step 701: Determine the number of integration steps for data assimilation based on the data assimilation period and integration step size of the regional ocean model.

[0094] The integration step size is the time interval between two adjacent integration moments during the numerical integration process of a regional ocean model. The integration step size can range from tens of seconds to hundreds of seconds.

[0095] The integration steps are the total number of steps required for a regional ocean model to complete numerical integration within one data assimilation period. The integration steps are equal to the total duration of the data assimilation period divided by the integration step size.

[0096] Step 702: Determine the assimilation increment share corresponding to each integration step based on the number of integration steps.

[0097] The assimilation increment share refers to the increment share at each integration step after the target analysis increment is evenly divided according to the number of integration steps. The assimilation increment share can remain constant during re-integration, being added to the model tendency term at each step of the model re-integration, such that the assimilation increment share can be equal to the target analysis increment divided by the number of integration steps. Using a nudging strategy to subdivide the assimilation increment into each integration step can avoid the disruption of ocean eddy structure and temperature-salinity relationships caused by abrupt changes in ocean temperature and salinity.

[0098] Step 703: Starting from the forecast status of the regional ocean model, perform numerical integration according to the number of integration steps. Before each integration step, add the assimilation increment share to the model tendency term. After integration through the number of integration steps, assimilate the target analysis increment to the regional ocean model to complete the data assimilation.

[0099] The forecast state refers to the state of the regional ocean model at the start of the current assimilation cycle before it has been updated by the current assimilation. The forecast state is obtained by integrating the analysis field generated in the previous assimilation cycle using the regional ocean model, and serves as the initial field for re-integration in the current assimilation cycle.

[0100] In each step of the re-integration, an assimilation increment share can be added to the model tendency term. The model tendency term is a forcing term that drives the change of model state variables over time. Before each integration step, the assimilation increment share is added to the tendency terms of the corresponding state variables such as temperature tendency and salinity tendency. After N integration steps, the total increment introduced is exactly equal to the target analysis increment. The updated forecast state is the analysis field generated in this assimilation cycle, and this analysis field is then used as the forecast state for the next assimilation cycle.

[0101] The method provided in this application reduces the numerical integration divergence and physical equilibrium disruption caused by applying the analysis increment to the model state all at once, by evenly distributing the target analysis increment and gradually adding it at each time step. It also reduces the spurious high-frequency oscillations caused by artificial discontinuities introduced into the model time integration path, so that the model state can continuously and smoothly transition to the updated state throughout the entire assimilation period, maintain the geostrophic balance relationship and the physical consistency of the temperature and salinity field, and enable the assimilated analysis field to comply with the dynamic equations of the regional ocean model.

[0102] Figure 2 This is an algorithmic framework diagram of a marine data assimilation method provided in one embodiment of this application, as shown below. Figure 2 As shown, in one embodiment, the regional ocean model employs a Regional Ocean Simulation System (ROMS). The state variables of the regional ocean model include a three-dimensional temperature field, a three-dimensional salinity field, and a three-dimensional current field. The three-dimensional spatial location of the regional ocean model is a first regular three-dimensional grid. The first regular three-dimensional grid uses orthogonal curvilinear coordinates in the horizontal direction and variable layered coordinates stretched with the terrain in the vertical direction.

[0103] The first type of observational data consists of gridded three-dimensional temperature, salinity, and temperature (THT) state estimates. This data is released weekly, with each period representing the average state over the past seven days. The three-dimensional spatial location of the first type of observational data is a second-regular three-dimensional grid, which is a latitude and longitude grid with a horizontal resolution of approximately one-third of a degree and a vertical resolution representing the standard depth layer. The first type of observational data includes three-dimensional temperature and three-dimensional salinity. This data can provide large-scale background field constraints to address the problem of sparse observations within the ocean interior.

[0104] The second category of observational data includes sea surface remote sensing data and in-situ temperature and salinity profile observation data. Sea surface remote sensing data specifically comprises satellite observations of sea surface temperature and sea surface height, released daily. In-situ temperature and salinity profile observation data specifically comprises Argo buoy temperature and salinity profile observation data, uploaded irregularly on a daily basis. The three-dimensional spatial location of sea surface remote sensing data is a discrete point on the sea surface. The three-dimensional spatial location of in-situ temperature and salinity profile observation data is a discrete point within the profile.

[0105] In a regional ocean model with a data assimilation period of 7 days, a first assimilation time window of 7 days, and a second assimilation time window of 3 days, when the boundary of the first assimilation time window is not perfectly aligned with the boundary of the first type of observation data release period, the first assimilation time window may overlap with both periods of observation data. For example, if the fourth day of the first assimilation time window, Wednesday, happens to coincide with the release date of the first type of observation data, then the first assimilation time window overlaps with both the products released on the previous Wednesday and the products released on the following Wednesday. Therefore, the length of the time overlap between each period of observation data and the first assimilation time window is calculated. The observation period of the previous period of observation data overlaps with the first assimilation time window of this embodiment by four days, and the observation period of the subsequent period of observation data overlaps with the first assimilation time window of this embodiment by three days. The weighting coefficients for each period of observation data are determined based on the length of the time overlap. The longer the time overlap, the larger the weighting coefficient. The weighting coefficients are then used to perform a weighted average of the two periods of observation data at each grid point to generate matching data for the first type of observation data within the first assimilation time window.

[0106] For the second type of observation data, all observation data falling within the second assimilation time window are identified. Since the second assimilation time window covers three days from Monday to Wednesday, the sea surface remote sensing observation data falling within this window includes daily observation data from Monday, Tuesday, and Wednesday. The in-situ temperature and salinity profile observation data falling within this window includes all uploaded Argo profile data during this period. Then, the arithmetic mean of each period's observation data falling within the window is calculated at each corresponding observation location to generate matching data for the second type of observation data within the second assimilation time window.

[0107] The regional ocean model has a forecast state at the start of the current assimilation period. Starting from this forecast state, the model generates a complete model forecast sequence covering a seven-day data assimilation period through a single continuous model integration. Model forecast data within the seven-day period corresponding to the first assimilation time window is extracted from the complete model forecast sequence and averaged over time to obtain the first model average state. Similarly, model forecast data within the three-day period corresponding to the second assimilation time window is extracted from the complete model forecast sequence and averaged over time to obtain the second model average state.

[0108] Since the average state of the first model is defined on the first regular 3D grid, and the spatial location of the first type of observation data is on the second regular 3D grid, 3D spatial interpolation is used to map the physical quantity values ​​of the average state of the first model at each grid point of the first regular 3D grid to the grid point locations of the second regular 3D grid, thus obtaining the estimated data of the first model. The interpolation operation uses bilinear interpolation in the horizontal direction and linear interpolation in the vertical direction.

[0109] Since the second type of observation data includes sea surface remote sensing observation data and in-situ temperature and salinity profile observation data, the spatial location of the sea surface remote sensing observation data is a discrete point on the sea surface, and the spatial location of the in-situ temperature and salinity profile observation data is a discrete point on the profile. Therefore, for the sea surface remote sensing observation data, horizontal spatial interpolation is used to map the physical quantity values ​​of the second model's average state located at each grid point on the sea surface in the first regular three-dimensional grid to the latitude and longitude positions of each discrete point on the sea surface, obtaining the model estimate values ​​at each discrete point on the sea surface. For the in-situ temperature and salinity profile observation data, horizontal spatial interpolation is first used to map the physical quantity values ​​of the second model's average state located at each grid point inside the ocean in the first regular three-dimensional grid to the latitude and longitude positions of each discrete point on the profile, obtaining the horizontal interpolation result. Then, vertical depth interpolation is used to map the horizontal interpolation result vertically to each depth layer of each discrete point on the profile, obtaining the model estimate values ​​for each depth layer at each discrete point on the profile. The model estimate values ​​at each discrete point on the sea surface and the model estimate values ​​for each depth layer at each discrete point on the profile are combined to form the second model estimation data.

[0110] By using the first and second assimilation time windows to match observation data with different time resolutions, the accuracy of the assimilation results can be improved without affecting the vortex structure and other parameters obtained by the model itself.

[0111] The first model estimation data and the matching data of the first type of observation data within the first assimilation time window are subtracted one by one at their corresponding spatial locations. The differences at each spatial location constitute the first innovation vector. The first innovation vector is used to quantify the degree of deviation of the regional ocean model from the first type of observation data on the time scale of the first assimilation time window.

[0112] The second model estimation data and the matching data of the second type of observation data within the second assimilation time window are subtracted one by one at the corresponding observation locations. The differences at each observation location constitute the second innovation vector. The second innovation vector is used to quantify the degree of deviation of the regional ocean model from the second type of observation data on the time scale of the second assimilation time window.

[0113] A first innovation vector is input into a Kalman filter with a first gain matrix. The first gain matrix is ​​multiplied by the first innovation vector, mapping the first innovation vector from the observation space of the first type of observation data back to the three-dimensional grid space of the regional ocean model, thus obtaining a first analysis increment. Similarly, a second innovation vector is input into a Kalman filter with a second gain matrix. The second gain matrix is ​​multiplied by the second innovation vector, mapping the second innovation vector from the observation space of the second type of observation data back to the three-dimensional grid space of the regional ocean model, thus obtaining a second analysis increment. The first Kalman gain matrix is ​​determined based on the observation error of the first type of observation data and the transpose of the first observation operator. The first observation operator is a three-dimensional space interpolation operator, and the transpose of the first observation operator back-projects information from the observation space of the first type of observation data to the model space of the regional ocean model.

[0114] The second Kalman gain matrix is ​​determined based on the observation errors of the second type of observation data and the transpose of the second observation operator. The second observation operator includes a horizontal spatial interpolation operator and a vertical depth interpolation operator. The transpose of the second observation operator back-projects the information in the observation space of the second type of observation data to the model space of the regional ocean model. The observation errors of the first type of observation data are used to determine the contribution weight of the difference values ​​at each observation location in the first innovation vector to the analysis increment. The observation errors of the second type of observation data are used to determine the contribution weight of the difference values ​​at each observation location in the second innovation vector to the analysis increment.

[0115] The difference between the updated forecast state and the original forecast state is defined as the target analysis increment. The target analysis increment integrates the large-scale background constraint information from the first type of observation data and the high-frequency detail constraint information from the second type of observation data.

[0116] Specifically, the updated forecast state is calculated using the following formula: ,in, Indicates the updated forecast status; This indicates the forecast status before the update; This represents the gain matrix of the Kalman filter; This refers to the matching data of the observed data within the assimilation time window, such as the matching data of the first type of observed data within the first assimilation time window, and the matching data of the second type of observed data within the second assimilation time window; H This represents the observation operator, i.e., the mapping relationship between the model space and the observation space; This indicates the model-averaged state of a regional ocean model within an assimilation time window, such as the first model-averaged state of a regional ocean model within a first assimilation time window, and the second model-averaged state of a regional ocean model within a second assimilation time window.

[0117] The incremental calculation can be performed using the following formula: .

[0118] The Kalman gain matrix is ​​expressed as: ,in, This refers to the observation error, specifically the observation error of the first type of observation data and the observation error of the second type of observation data.

[0119] After obtaining the target analysis increment, the number of integration steps for data assimilation is determined based on the data assimilation period and integration step size of the regional ocean model. For example, if the assimilation period is seven days (604,800 seconds) and the integration step size is sixty seconds, the number of integration steps is 10,080. The assimilation increment share corresponding to each integration step is determined based on the number of integration steps. Dividing the target analysis increment by 10,080 yields the assimilation increment share for each step. .

[0120] Then, the integration is restarted starting from the forecast state of the regional ocean model, and at each step of the reintegration, an assimilation increment is added to the temperature and salinity trend terms of the regional ocean model. The assimilation increment added at each step is of the same size. After 10,080 integration steps, the total increment introduced is equal to the target analysis increment.

[0121] After re-integration, the state of the regional ocean model is smoothly updated to the assimilated state, which is the three-dimensional ocean analysis field generated in this data assimilation cycle. The three-dimensional ocean analysis field includes the assimilated and updated three-dimensional temperature field, three-dimensional salinity field, and three-dimensional current field. It fits the weekly average large-scale observation background of the first type of observation data and also incorporates the daily high-frequency observation details of the second type of observation data. Moreover, the update process is completed step by step in each step of the model re-integration, thus maintaining the geostrophic balance relationship and the physical consistency of the temperature and salinity field.

[0122] Figure 3 This is an interface diagram comparing the root mean square error time series for performance verification of the marine data assimilation method provided in one embodiment of this application. Figure 3 As shown, the different colored lines represent the changes in the deviation (RMSE) between different models or estimated states and actual observation data over time in the sea area simulated by the regional ocean model from June 26, 2011 to June 5, 2012. The actual observation data can be Argo profile data.

[0123] exist Figure 3In the diagram, the gray line represents the model simulation results when the model runs freely without introducing observation corrections; the yellow line represents the first type of observation data, such as the 3D ocean observation data ARMOR3D based on multi-source observation statistical fusion; the blue line represents the short-term forecast results starting from the analysis field within the assimilation period; the green line represents the average state of the first model within the first assimilation time window; the red line represents the average state of the second model within the second assimilation time window; the purple line represents the forecast state sequence after assimilating the first type of observation data, the second type of observation data, and the forecast state using the ocean data assimilation method provided in this application embodiment; and the black line represents the 3D gridded ocean state sequence generated by the Global Ocean Reanalysis System (GLORYS).

[0124] like Figure 3 As shown, (a) represents the root mean square error of temperature relative to Argo for different models or estimated states, and (b) represents the root mean square error of salinity relative to Argo for different models or estimated states. (c) is a vertical profile of linear depth versus root mean square error of temperature, (d) is a vertical profile of linear depth versus root mean square error of salinity, (e) is a vertical profile of logarithmic depth versus root mean square error of temperature, and (f) is a vertical profile of logarithmic depth versus root mean square error of salinity.

[0125] The root mean square error (MSE) quantifies the degree of deviation between different models or estimated states and the actual ocean state. A smaller MSE indicates that the assimilation result is closer to the actual state, and the assimilation effect is better. After assimilating the data using the method provided in this application, the MSE for temperature is 0.2643, and the MSE for salinity is 0.0297. This demonstrates that the data assimilation method provided in this application can effectively eliminate model bias under the constraints of long-period and short-period observations.

[0126] Figure 4 This is a schematic diagram of the structure of a marine data assimilation device provided in another embodiment of this application. Figure 4 As shown, the ocean data assimilation device 40 includes: The acquisition module 41 is used to acquire matching data of the first type of observation data within the first assimilation time window, and matching data of the second type of observation data within the second assimilation time window; the first type of observation data and the second type of observation data are observation data of different observation periods corresponding to the region simulated by the regional ocean model. The mapping module 42 is used to map the average state of the first model of the regional ocean model within the first assimilation time window to the three-dimensional spatial position of the first type of observation data to obtain the first model estimation data, and to map the average state of the second model of the regional ocean model within the second assimilation time window to the three-dimensional spatial position of the second type of observation data to obtain the second model estimation data. The first determining module 43 is used to determine the first innovation vector of the regional ocean model based on the difference between the first model estimated data and the corresponding matching data, and to determine the second innovation vector of the regional ocean model based on the difference between the second model estimated data and the corresponding matching data. The second determination module 44 is used to determine the target analysis increment of the regional ocean model based on the first and second innovation vectors. Assimilation module 45 is used to incrementally update the target analysis to the regional ocean model in a relaxed manner.

[0127] In some embodiments, the first assimilation time window is greater than or equal to the observation period of the first type of observation data, and the acquisition module 41 is specifically used for: Identify at least two observation periods in the first type of observation data that overlap with the first assimilation time window; The weighting coefficients of each period of the observation data in the at least two periods of observation data are determined based on the length of time overlap between each period of observation data and the first assimilation time window. Based on the weighting coefficients of the observation data in each period, a weighted average is calculated on the observation data of at least two periods to generate matching data for the first type of observation data within the first assimilation time window.

[0128] In some embodiments, the acquisition module 41 is specifically used for: Identify at least two periods of observation data within the second assimilation time window from the second type of observation data; The average of at least two periods of observation data is used to obtain the matching data of the second type of observation data within the second assimilation time window.

[0129] In some embodiments, the three-dimensional spatial location of the regional ocean model is a first regular three-dimensional grid, and the three-dimensional spatial location of the first type of observation data is a second regular three-dimensional grid. The mapping module 42 is specifically used for: By using three-dimensional spatial interpolation, the physical quantity values ​​of the average state of the first mode at each grid point of the first regular three-dimensional grid are mapped to the grid point positions of the second regular three-dimensional grid to obtain the estimated data of the first mode.

[0130] In some embodiments, the three-dimensional spatial location of the regional ocean model is a first regular three-dimensional grid, the second type of observation data includes sea surface remote sensing observation data and in-situ temperature and salinity profile observation data, the three-dimensional spatial location of the sea surface remote sensing observation data is a discrete point on the sea surface, the three-dimensional spatial location of the in-situ temperature and salinity profile observation data is a discrete point on the profile, the second model estimation data includes model estimates at each discrete point on the sea surface and model estimates at each depth layer at each discrete point on the profile, and the mapping module 42 is specifically used for: By using horizontal spatial interpolation, the physical quantity values ​​of the average state of the second model located at each grid point on the sea surface in the first regular three-dimensional grid are mapped to each discrete point on the sea surface to obtain the model estimate value at each discrete point on the sea surface. Horizontal spatial interpolation is used to map the physical quantity values ​​of the second model's average state at each grid point inside the ocean in the first regular three-dimensional grid to the latitude and longitude positions of each discrete point in the profile, thus obtaining the horizontal interpolation result. Vertical depth interpolation is used to map the horizontal interpolation results to each depth layer at each discrete point of the profile, thereby obtaining the model estimate value of each depth layer at each discrete point of the profile.

[0131] In some embodiments, the second determining module 44 is specifically used for: The first and second innovation vectors are transformed to their three-dimensional spatial positions in the regional ocean model to obtain the first analysis increment corresponding to the first innovation vector and the second analysis increment corresponding to the second innovation vector. The vector sum of the first and second analysis increments is determined as the target analysis increment.

[0132] In some embodiments, the marine data assimilation device further includes: The third determination module is used to determine the first assimilation time window based on the data assimilation period of the regional ocean model and the observation period of the first type of observation data; and to determine the second assimilation time window based on the data assimilation period of the regional ocean model and the observation period of the second type of observation data; the assimilation module 45 is specifically used for: The number of integration steps for data assimilation is determined based on the data assimilation period and integration step size of the regional ocean model. Based on the number of integration steps, determine the assimilation increment share corresponding to each integration step; Starting from the forecast status of the regional ocean model, numerical integration is performed according to the number of integration steps. Before each integration step, an assimilation increment is added to the model tendency term. After integration through the number of integration steps, the target analysis increment is assimilated into the regional ocean model, thus completing the data assimilation.

[0133] The marine data assimilation device provided in this application embodiment can execute the marine data assimilation method provided in any of the above embodiments, has the same principle and can achieve the same technical effect, and will not be described in detail here.

[0134] Figure 5 This is a schematic diagram of the structure of an electronic device provided in another embodiment of this application. For example... Figure 5 As shown, the electronic device may include a processor 51 and a memory 52 storing computer program instructions.

[0135] Specifically, the processor 51 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0136] Memory 52 may include mass storage for data or instructions. For example, and not limitingly, memory 52 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 52 may include removable or non-removable (or fixed) media. Where appropriate, memory 52 may be internal or external to the integrated gateway disaster recovery device. In a particular embodiment, memory 52 is non-volatile solid-state memory.

[0137] Memory may include read-only memory (ROM), random access memory (RAM), disk storage media devices, optical storage media devices, flash memory devices, and electrical, optical, or other physical / tangible memory storage devices. Therefore, typically, memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the methods according to one aspect of this disclosure.

[0138] The processor 51 implements any of the ocean data assimilation methods described above by reading and executing computer program instructions stored in the memory 52.

[0139] In one example, the electronic device may also include a communication interface 53 and a bus 54. Wherein, for example... Figure 5 As shown, the processor 51, memory 52, and communication interface 53 are connected through bus 54 and complete communication with each other.

[0140] Communication interface 53 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application.

[0141] Bus 54 includes hardware, software, or both, that couples components of an electronic device together. For example, and not limitingly, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 54 may include one or more buses. Although specific buses are described and illustrated in embodiments of this application, this application contemplates any suitable bus or interconnect.

[0142] Furthermore, in conjunction with the marine data assimilation methods in the above embodiments, this application embodiment can provide a computer-readable storage medium for implementation. This computer-readable storage medium stores computer program instructions; when executed by a processor, these computer program instructions implement any of the marine data assimilation methods in the above embodiments.

[0143] This application also provides a computer program product, including a computer program that, when executed by a processor, implements any of the marine data assimilation methods described in the above embodiments.

[0144] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.

[0145] The functional blocks shown in the above-described block diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. The programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0146] It should also be noted that the exemplary embodiments mentioned in this application describe methods or systems based on a series of steps or apparatus. However, this application is not limited to the order of the above steps; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0147] Flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure describe aspects of this disclosure. It should be understood that each block in the flowcharts and / or block diagrams, and combinations of blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to create a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowcharts and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by special-purpose hardware performing the specified functions or actions, or can be implemented by a combination of special-purpose hardware and computer instructions.

[0148] The above description is merely a specific implementation of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application.

Claims

1. A method for assimilating marine data, characterized in that, include: Acquire matching data for the first type of observation data within the first assimilation time window, and matching data for the second type of observation data within the second assimilation time window; The first type of observation data and the second type of observation data are observation data from different observation periods corresponding to the regions simulated by the regional ocean model; The first model average state of the regional ocean model within the first assimilation time window is mapped to the three-dimensional spatial position of the first type of observation data to obtain the first model estimation data; and the second model average state of the regional ocean model within the second assimilation time window is mapped to the three-dimensional spatial position of the second type of observation data to obtain the second model estimation data. Based on the difference between the first model estimation data and the corresponding matching data, a first innovation vector of the regional ocean model is determined, and based on the difference between the second model estimation data and the corresponding matching data, a second innovation vector of the regional ocean model is determined. Based on the first and second information vectors, the target analysis increment of the regional ocean model is determined; The target analysis is incrementally updated to the regional ocean model using a relaxed incremental analysis update method.

2. The marine data assimilation method according to claim 1, characterized in that, The first assimilation time window is greater than or equal to the observation period of the first type of observation data. Acquiring matching data for the first type of observation data within the first assimilation time window includes: Identify at least two periods of observation data in the first type of observation data where the observation period overlaps with the first assimilation time window; The weighting coefficients of each period of the at least two periods of observation data are determined based on the length of time overlap between each period of observation data and the first assimilation time window. Based on the weighting coefficients of the observation data in each period, a weighted average is performed on the observation data of at least two periods to generate matching data for the first type of observation data within the first assimilation time window.

3. The marine data assimilation method according to claim 1, characterized in that, Obtaining matching data for the second type of observation data within the second assimilation time window includes: Identify at least two periods of observation data in the second type of observation data that fall within the second assimilation time window; The average of the at least two periods of observation data is used to obtain the matching data of the second type of observation data within the second assimilation time window.

4. The marine data assimilation method according to claim 1, characterized in that, The three-dimensional spatial location of the regional ocean model is a first regular three-dimensional grid, and the three-dimensional spatial location of the first type of observation data is a second regular three-dimensional grid. Mapping the average state of the first model within the first assimilation time window to the three-dimensional spatial location of the first type of observation data to obtain the first model estimation data includes: By using three-dimensional spatial interpolation, the physical quantity values ​​of the average state of the first mode at each grid point of the first regular three-dimensional grid are mapped to the grid point positions of the second regular three-dimensional grid to obtain the estimated data of the first mode.

5. The marine data assimilation method according to claim 1, characterized in that, The three-dimensional spatial location of the regional ocean model is a first regular three-dimensional grid. The second type of observation data includes sea surface remote sensing observation data and in-situ temperature and salinity profile observation data. The three-dimensional spatial location of the sea surface remote sensing observation data is a discrete point on the sea surface, and the three-dimensional spatial location of the in-situ temperature and salinity profile observation data is a discrete point on the profile. The second model estimation data includes model estimates at each of the sea surface discrete points and model estimates at each depth layer at each of the discrete points on the profile. The step of mapping the second model average state of the regional ocean model within the second assimilation time window to the three-dimensional spatial location of the second type of observation data to obtain the second model estimation data includes: By using horizontal spatial interpolation, the physical quantity values ​​of the average state of the second model located at each grid point on the sea surface in the first regular three-dimensional grid are mapped to each of the sea surface discrete points to obtain the model estimate values ​​at each of the sea surface discrete points. By using horizontal spatial interpolation, the physical quantity values ​​of the average state of the second mode located at each grid point inside the ocean in the first regular three-dimensional grid are mapped to the latitude and longitude positions of each discrete point of the profile to obtain the horizontal interpolation result. Vertical depth interpolation is used to map the horizontal interpolation results to each depth layer of each discrete point of the profile, thereby obtaining the mode estimation value of each depth layer at each discrete point of the profile.

6. The marine data assimilation method according to claim 1, characterized in that, The step of determining the target analysis increment of the regional ocean model based on the first and second innovation vectors includes: The first innovation vector and the second innovation vector are transformed to their three-dimensional spatial positions in the regional ocean model to obtain the first analysis increment corresponding to the first innovation vector and the second analysis increment corresponding to the second innovation vector; The vector sum of the first analysis increment and the second analysis increment is determined as the target analysis increment.

7. The marine data assimilation method according to any one of claims 1-6, characterized in that, Before acquiring the matching data of the first type of observation data within the first assimilation time window and the matching data of the second type of observation data within the second assimilation time window, the method further includes: The first assimilation time window is determined based on the data assimilation period of the regional ocean model and the observation period of the first type of observation data; The second assimilation time window is determined based on the data assimilation period of the regional ocean model and the observation period of the second type of observation data; The step of incrementally updating the target analysis to the regional ocean model using a relaxed approach includes: The number of integration steps for data assimilation is determined based on the data assimilation period and integration step size of the regional ocean model. Based on the number of integration steps, determine the assimilation increment share corresponding to each integration step; Starting from the forecast status of the regional ocean model, numerical integration is performed according to the number of integration steps. Before each integration step, the assimilation increment is added to the model tendency term. After integration for the specified number of integration steps, the target analysis increment is assimilated into the regional ocean model, thus completing data assimilation.

8. An electronic device, characterized in that, The device includes: a processor and a memory storing computer program instructions; When the processor executes the computer program instructions, it implements the marine data assimilation method as described in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer program instructions that, when executed by a processor, implement the marine data assimilation method as described in any one of claims 1-7.

10. A computer program product, characterized in that, When the instructions in the computer program product are executed by the processor of the electronic device, the electronic device causes the electronic device to perform the marine data assimilation method as described in any one of claims 1-7.

Images