A large-scale assimilation method for oceanic dynamical downscaling

Abstract

The application discloses a large-scale assimilation method for ocean dynamic downscaling. The application is based on a to-be-corrected variable in a high-resolution model, selects reference data, uses the EOF expansion of a space-time matrix, obtains a large-scale spatial mode of the to-be-corrected variable of the high-resolution model, projects the to-be-corrected variable and the reference data to the large-scale mode to respectively obtain time coefficients, and realizes the approximation of the large-scale signal of the high-resolution model to the large-scale signal of the reference data based on the obtained time coefficients, thereby improving the simulation of the large-scale signal of the high-resolution model. The application has the advantages that the large-scale signal is obviously improved through the EOF approximation of the large-scale signal of the high-resolution model, the small-scale signal which is not limited can also become more accurate through the interaction of the large-scale and the small-scale in the nonlinear coupling, the decoupling problem of the high-resolution nested simulation of the ocean model is solved, and the simulation capability of the small-scale process in the high-resolution model is improved.

Description

Technical Field

[0001] This invention relates to a large-scale assimilation method for ocean dynamic downscaling, belonging to the field of marine scientific numerical simulation. Technical Background

[0002] In ocean numerical simulations, to perform high-resolution simulations of areas of interest, regional models or nested models are generally used. Currently, common regional or nested simulations typically employ open-boundary forcing to transfer large-scale signals from coarse-resolution large-region or global models into a high-resolution small-region model. The high-resolution small-region model is then used in conjunction with system dynamic processes to generate smaller-scale phenomena that cannot be characterized by the coarse-resolution large-region or global models. This method is known as ocean dynamic downscaling.

[0003] For high-resolution small-area simulation, the ideal state is to accurately transfer the large-scale signals characterized by the coarse-resolution large-area model or coarse-resolution global model into the high-resolution small-area model. That is, to make the large-scale signals in the high-resolution small-area model consistent with the large-scale signals in the coarse-resolution large-area model or coarse-resolution global model. Then, the high-resolution small-area model generates accurate small-scale signals by relying on the transferred large-scale signals and the interaction between large and small scales.

[0004] However, due to the numerical bottleneck of open boundary conditions and the effect of internal variability, the large-scale signals in high-resolution small-area models are not completely consistent with the large-scale signals in coarse-resolution large-area models or coarse-resolution global models, i.e., decoupling occurs. This decoupling problem causes the large-scale signals of high-resolution regional models to deviate from the accurate large-scale signals of low-resolution models, thus affecting the generation of small-scale signals in high-resolution models and hindering the generation of accurate small-scale signals from high-resolution models. Summary of the Invention

[0005] To address the decoupling problem in ocean dynamic downscaling simulations and improve the performance of high-resolution regional simulations, this invention aims to provide a large-scale assimilation method based on EOF approximation for ocean dynamic downscaling. This method improves the accuracy of large-scale signals for application in high-resolution regional models and enhances the simulation capability of small-scale processes.

[0006] This invention utilizes EOF approximation to approximate the large-scale signals of high-resolution model variables to the large-scale signals of corresponding variables in the reference data, thereby solving the decoupling problem in high-resolution nested simulations of ocean models. It accurately transmits large-scale signals into the high-resolution simulation region without changing the small-scale processes simulated by the high-resolution model, thus improving the simulation capability of small-scale processes in the high-resolution model.

[0007] The specific technical solution of the present invention is as follows:

[0008] (1) Variables based on the high-resolution small-area model to be corrected Select reference data X n×p Based on the EOF expansion of the spatiotemporal matrix, the spatiotemporal matrix X is obtained. n×p Principal component matrices and EOF space modes; where, a n×p The principal component matrix is ​​denoted by C, where each column represents a principal component. p×p This is the EOF spatial mode matrix, where each row represents an EOF spatial mode P. scs ;

[0009]

[0010] (2) Variables in high-resolution small-area models The spatiotemporal matrix is ​​decomposed into EOF scale to obtain its spatial mode matrix. The spatial scale of each spatial mode is calculated to determine the number of large-scale spatial modes.

[0011] (3) Based on the large-scale spatial mode determined in step (2), the large-scale spatial mode P based on the reference data obtained in step (1) scs ,Will ( for The value at time t can be decomposed into two parts. Where, α t It is Projected onto P scs The corresponding time coefficient, The remaining spatial modal synthesis fields are orthogonal to P. scs ,Right now

[0012] (4) Based on the large-scale spatial mode determined in step (2), the large-scale spatial mode P based on the reference data obtained in step (1) scs , reference data X n×p Also to P scs Projection, which corresponds to P scs The time coefficient is β t ;

[0013] (5) Using the time coefficient α obtained in steps (3) and (4) t and β t ,according to Will Revised to Where λ is the approximation coefficient, and its value ranges from 0 ≤ λ ≤ 1;

[0014] (6) Revised to back, The dynamic system of the model evolves to the next moment. The pattern integration process will be changed from the original Transform into Revised By correlating with other dynamic system variables, other dynamic system variables can be corrected, thereby obtaining more accurate high-resolution model results.

[0015] The variables mentioned in step (1) This can be measured by positive pressure flow velocity, sea surface height, temperature, or salinity.

[0016] In step (2), the large-scale spatial mode is defined as a spatial mode with a spatial scale of 250 km or more.

[0017] In step (5), when λ = 1, we have That is, the large-scale signals of the variables in the high-resolution model are completely replaced by the large-scale signals of the corresponding variables in the reference data; when λ = 0, That is, EOF approximation was not enabled;

[0018] Advantages of this invention:

[0019] This invention significantly improves the large-scale signal by performing EOF approximation on the high-resolution model, and the unrestricted small-scale signal can also be made more accurate through the interaction between the large and small scales via nonlinear coupling. This solves the decoupling problem in high-resolution nested simulations of ocean models, thereby improving the simulation capability of small-scale processes in high-resolution models. Detailed Implementation

[0020] To further understand the invention's content, features, and technical effects, the invention will be further described in detail below through specific embodiments.

[0021] Example 1:

[0022] The Hybrid Coordinate Ocean Model (HYCOM) was used to implement the correction of the sea surface height (SSH) of the HYCOM South China Sea model.

[0023] The specific implementation steps are as follows:

[0024] (1) In this embodiment, the sea surface height (SSH) of the HYCOM South China Sea model needs to be corrected. Therefore, the SSH data in the C-GLORS (CMCC Global Ocean Reanalysis System) reanalysis data is selected as the reference data. Based on EOF expansion, its principal component matrix and EOF spatial mode are obtained.

[0025] (2) EOF scale decomposition was performed on the SSH data to obtain the spatial mode matrix of the SSH data of the HYCOM South China Sea model, and the spatial scale of each spatial mode was calculated. The calculation results show that the spatial scale corresponding to the first six EOF modes is approximately 250km-700km, which corresponds to the spatial scale of large-scale signals of ocean processes. Therefore, the large-scale spatial modes are the first six EOF modes.

[0026] (3) Based on the first six EOF spatial modes P scs Seeking to obtain Projected onto P scs Time coefficient α t .

[0027] (4) Based on the first six EOF spatial modes P scs The reference data is projected onto P. scs Time coefficient β t .

[0028] (5) Using the time coefficient α obtained in steps (3) and (4) t and β t EOF approximation was performed on the first six EOF modes (large-scale signals) of the HYCOM South China Sea model's SSH. In order to correct the large-scale signals in the high-resolution model and maintain the matching of large and small scale processes in the high-resolution model, the approximation coefficient λ was 0.5.

[0029] (6) After the SSH is corrected, other dynamic system variables are corrected by relating the SSH to other dynamic system variables. Here, g represents gravitational acceleration, and montg1 represents the Montgomery potential of the first layer. It is the specific volume reference value, ηP′ b It is positive pressure.

[0030] To avoid randomness and ensure the complete effectiveness of the EOF approximation large-scale assimilation scheme, four sets of simulation experiments were conducted: four sets of high-resolution HYCOM South China Sea simulations without using the EOF approximation large-scale assimilation scheme, and four sets of high-resolution HYCOM South China Sea simulations using the EOF approximation large-scale assimilation scheme (Table 1). These four sets of experiments were identical except for the initial fields. The atmospheric forcing field was the 2008 data from the NOGAP real atmospheric forcing field, and the initial fields were from January 1st data of the 18th, 19th, 20th, and 21st years of the climatological South China Sea simulations.

[0031] Table 1. Experimental Design Schemes for Four Groups

[0032]

[0033]

[0034] Example results:

[0035] In this embodiment, the first six EOF modes (large-scale signals) of the HYCOM South China Sea model's SSH are approximated to the first six EOF modes (large-scale signals) of the C-GLORS data's SSH to obtain an optimized high-resolution South China Sea model. The large-scale signals of the first six EOF modes of the optimized South China Sea model can be approximated to the corresponding large-scale signals of the more accurate reference data, that is, the large-scale signals of the South China Sea simulation are improved (Table 2). Furthermore, the unrestricted small-scale signals can also be approximated to the small-scale signals of the more accurate reference data through the nonlinear coupling of large-scale and small-scale interactions (Table 3).

[0036] To facilitate comparison, a statistical variable σ was designed and calculated as follows: Here, PC_i (i = 1, 2, 3, 4) represents the PC time coefficient corresponding to each sample experiment in the four groups of experiments, and PC_C represents the PC time coefficient of the C-GLORS data. That is, σ represents the overall statistical difference between the PC time coefficient of each sample experiment in the HYCOM South China Sea simulation experiment and the PC time coefficient of C-GLORS.

[0037] Table 2 σ values ​​for the first 6 EOF modes of SSH

[0038]

[0039] Note: u indicates large-scale assimilation experiment without EOF; c indicates large-scale assimilation experiment with EOF.

[0040] Table 2 shows quantitatively that for the first six EOF modes of SSH, the PC time coefficients of each mode with EOF approximation large-scale assimilation experiments are closer to the PC time coefficients of C-GLORS data throughout the year and in each season than the PC time coefficients of each mode without EOF approximation large-scale assimilation experiments.

[0041] Table 3. σ for the 7th-10th EOF modes of SSH

[0042]

[0043] Note: u indicates large-scale assimilation experiment without EOF; c indicates large-scale assimilation experiment with EOF.

[0044] Table 3 shows quantitatively that for the PC time coefficients of the 7th and 8th EOF modes of SSH without any restrictions, the PC time coefficients of the EOF approximation large-scale assimilation experiment are closer to the PC time coefficients of the C-GLORS data than the PC time coefficients of the non-EOF approximation large-scale assimilation experiment throughout the entire simulation period.

[0045] Throughout the entire simulation period (2008), the largest discrepancy between the large-scale assimilation experiment without EOF approximation and the C-GLORS data occurred mainly in the northeastern part of the South China Sea near the Luzon Strait (approximately 16°N-22°N, 115°E-120°E). After performing large-scale assimilation with EOF approximation, the discrepancy between the simulation experiment and the C-GLORS data decreased significantly.

[0046] Throughout the entire simulation period (2008), after assimilation of the large-scale signal from SSH, the difference between the simulated experimental and unrestricted composite fields of C-GLORS data in the 7th and 8th composite fields was significantly reduced, especially in the southwestern waters of Taiwan.

[0047] The present invention has been described in detail above with general description and specific embodiments. However, some modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention are within the scope of protection claimed by the present invention.

Claims

1. A large-scale assimilation method for ocean dynamic downscaling, characterized in that, Includes the following steps: (1) Variables based on the high-resolution small-area model to be corrected Select reference data Based on the spatiotemporal matrix, an EOF expansion is performed to obtain the spatiotemporal matrix. The principal component matrix and EOF space modes; where, It is a principal component matrix, where each column represents a principal component; This is the EOF spatial mode matrix, where each row represents an EOF spatial mode. ; ； The variable For positive pressure flow velocity, sea surface height, temperature, or salinity; (2) Variables in high-resolution small-area models The spatiotemporal matrix is ​​decomposed by EOF to obtain its spatial mode matrix. The spatial scale of each spatial mode is calculated to determine the number of large-scale spatial modes. The large-scale spatial mode is defined as a spatial mode with a spatial scale of 250 km or more. (3) Based on the large-scale spatial modes determined in step (2), the large-scale spatial modes based on the reference data obtained in step (1) ,Will Decomposed into two parts, ;in, It is Projected to The corresponding time coefficient, ; The remaining spatial modal synthesis fields are orthogonal to ,Right now ; (4) Based on the large-scale spatial modes determined in step (2), the large-scale spatial modes based on the reference data obtained in step (1) Reference data also to Projection, its corresponding The time factor is ; (5) Use the time coefficients obtained in steps (3) and (4) and ,according to ,Will Revised to ;in, These are approximation coefficients, and their values ​​range from [value missing]. ; (6) Revised to back, The dynamic system of the model evolves to the next moment. The pattern integration process will be changed from the original Transform into ; Corrected By correlating with other dynamic system variables, other dynamic system variables can be corrected, thereby obtaining more accurate high-resolution model results.

2. The large-scale assimilation method for ocean dynamic downscaling as described in claim 1, characterized in that, In step (5), when Sometimes, That is, the large-scale signal of the high-resolution model variable is completely replaced by the large-scale signal of the corresponding variable in the reference data.

3. The large-scale assimilation method for ocean dynamic downscaling as described in claim 1, characterized in that, In step (5), when hour, This means that EOF approximation was not enabled.

Images