Multi-source data fusion method and device based on improved multi-grid variational assimilation

Abstract

The embodiment of the invention provides a multi-source data fusion method and device based on improved multi-grid variational assimilation. The method is applied to the technical field of data processing, and comprises the following steps: performing quality control on multi-source observation data according to a meteorological standard to obtain a standardized data set; constructing a multiple analysis grid according to the standardized data set to obtain a grid structure; separating data features through transformation and filtering processing by using the grid structure to obtain multi-scale feature data; constructing an observation operator based on the multi-scale feature data to form an observation operator system; performing variational assimilation processing on the grid data according to the observation operator system to obtain an assimilation analysis field; and according to the assimilation analysis field, carrying out inter-grid information exchange processing to obtain a fusion result. In this way, high-efficiency and high-precision data fusion can be realized through the design of a multi-grid structure and the separation processing of multi-scale features.

Description

Technical Field

[0001] The present disclosure relates to the field of data processing, and in particular, to a multi-source data fusion method and device based on improved multi-grid variational assimilation. Background Art

[0002] With the development of meteorological observation technologies, the means of obtaining multi-source observation data have been continuously enriched, including various observation platforms such as surface meteorological stations, weather radars, and meteorological satellites. These observation systems can provide meteorological data with different spatio-temporal resolutions and different observation elements, providing important data support for weather forecasting and climate research. Existing data assimilation methods mainly include optimal interpolation method, three-dimensional variational assimilation, four-dimensional variational assimilation, etc., and these methods have been widely used in meteorological data assimilation. Among them, the variational assimilation method constructs a cost function to find the optimal balance between observation information and background fields, realizing the effective utilization of observation data. At the same time, the multi-grid technology has shown unique advantages in improving computational efficiency and dealing with multi-scale problems, and has achieved successful applications in fields such as atmospheric science and computational fluid dynamics.

[0003] However, existing data assimilation methods still have some technical difficulties in dealing with multi-source and multi-scale meteorological data. First of all, the data obtained by different observation systems have different spatio-temporal resolutions and error characteristics, and how to effectively fuse these heterogeneous data is a challenge. Secondly, traditional single-grid assimilation methods are difficult to simultaneously meet the description requirements of large-scale weather systems and local weather phenomena, and it is often difficult to achieve a good balance between efficiency and accuracy. In addition, existing methods have low computational efficiency when dealing with large-scale data. Especially in high-resolution numerical prediction systems, the assimilation window time is long and the computational resources consumed are large, making it difficult to meet the requirements of operational applications. Summary of the Invention

[0004] The present disclosure provides a multi-source data fusion method and device based on improved multi-grid variational assimilation. Aiming at the problems of strong heterogeneity of data from different observation systems and low computational efficiency in the prior art, through the design of a multi-grid structure and the separation processing of multi-scale features, high-efficiency and high-precision data fusion is achieved.

[0005] According to the first aspect of the present disclosure, a multi-source data fusion method based on improved multi-grid variational assimilation is provided, including:

[0006] Performing quality control on multi-source observation data according to meteorological standards to obtain a standardized data set;

[0007] Constructing a multi-analysis grid according to the standardized data set to obtain a grid structure;

[0008] Using the grid structure, separating data features through transformation and filtering processing to obtain multi-scale feature data;

[0009] Construct an observation operator based on the multi-scale feature data to form an observation operator system;

[0010] Perform variational assimilation processing on the grid data according to the observation operator system to obtain an assimilation analysis field;

[0011] Perform information exchange processing between grids based on the assimilation analysis field to obtain a fusion result.

[0012] According to a second aspect of the present disclosure, there is provided a multi-source data fusion device based on improved multi-grid variational assimilation, including:

[0013] A control module for performing quality control on multi-source observation data according to meteorological standards to obtain a standardized data set;

[0014] An analysis module for constructing a multi-analysis grid based on the standardized data set to obtain a grid structure;

[0015] A transformation module for using the grid structure to separate data features through transformation and filtering processing to obtain multi-scale feature data;

[0016] A construction module for constructing an observation operator based on the multi-scale feature data to form an observation operator system;

[0017] An assimilation module for performing variational assimilation processing on the grid data according to the observation operator system to obtain an assimilation analysis field;

[0018] An exchange module for performing information exchange processing between grids based on the assimilation analysis field to obtain a fusion result.

[0019] The present disclosure ensures the reliability and consistency of the input data by performing quality control on multi-source observation data according to meteorological standards, laying a good foundation for subsequent processing. The design of constructing multiple analysis grids based on the standardized data set enables effective description of weather systems at different scales. The coarse grid is used to capture the characteristics of large-scale weather systems, and the fine grid is used to depict local weather phenomena, improving the accuracy of data analysis. The method of separating data features through transformation and filtering processing using the grid structure effectively realizes the identification and extraction of multi-scale features, enabling accurate expression of the characteristics of weather systems at different scales. The technical means of constructing an observation operator system based on multi-scale feature data realizes the accurate mapping from the observation space to the model space, improving the utilization efficiency of observation information in the assimilation process. The method of performing variational assimilation processing on grid data according to the observation operator system ensures the optimality of the assimilation result through optimized solution, making the analysis field conform to physical laws and fully utilize observation information. The design of performing information exchange processing between grids based on the assimilation analysis field realizes the effective transmission of information between the coarse and fine grids, ensuring the coordination and unity of information at different scales. The organic combination of these technical features ultimately realizes the efficient fusion of multi-source data, improves the quality of the initial field of numerical weather prediction, and provides strong support for improving the prediction accuracy. At the same time, the design strategy of the multiple grids significantly reduces the computational complexity and improves the assimilation efficiency.

[0020] It should be understood that the content described in the summary of the invention section is not intended to limit the key or important features of the embodiments of the present disclosure, nor to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In combination with the accompanying drawings and with reference to the following detailed description, the above and other features, advantages, and aspects of the embodiments of the present disclosure will become more apparent. The drawings are used to better understand the solution and do not constitute a limitation to the present disclosure. In the drawings, the same or similar reference numerals represent the same or similar elements, where:

[0022] Figure 1 FIG. shows a flowchart of a multi-source data fusion method based on improved multiple grid variational assimilation according to an embodiment of the present disclosure;

[0023] Figure 2 FIG. shows a block diagram of a multi-source data fusion device based on improved multiple grid variational assimilation according to an embodiment of the present disclosure. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present disclosure clearer, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present disclosure. Apparently, the described embodiments are only a part rather than all of the embodiments of the present disclosure. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present disclosure without creative efforts shall fall within the scope of protection of the present disclosure.

[0025] In addition, the term "and / or" in this document is merely a description of the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B may represent: the individual existence of A, the simultaneous existence of A and B, and the individual existence of B. In addition, the character " / " in this document generally represents an "or" relationship between the associated objects before and after.

[0026] Figure 1 The flowchart of a multi-source data fusion method 100 based on improved multi-grid variational assimilation in the embodiments of the present disclosure is shown, as Figure 1 shown, the method 100 includes:

[0027] S110: Perform quality control on multi-source observation data according to meteorological standards to obtain a standardized data set;

[0028] Optionally, collect ground conventional observation data through a ground meteorological station. The ground conventional observation data includes temperature, humidity, air pressure, precipitation, and wind speed; collect radar observation data through a weather radar. The radar observation data includes radar reflectivity; perform quality control on the ground conventional observation data according to the value range test, continuity test, and consistency test in the ground standard for quality control of meteorological observation data; perform quality control on the radar observation data according to the data rationality test, time continuity test, and spatial consistency test in the weather radar standard for quality control of meteorological observation data; perform format conversion on the quality-controlled ground conventional observation data and radar observation data to generate standardized data in a common binary format; collect global forecast products, and combine the global forecast products with the standardized data in the common binary format to form the standardized data set.

[0029] Among them, ground conventional observation data are mainly collected through sensor devices distributed at each observation station. The data elements collected include temperature, humidity, air pressure, precipitation, and wind speed. Temperature data are collected through temperature sensors to record the temperature changes in the atmosphere; humidity data are collected through humidity sensors to reflect the water vapor content in the atmosphere; air pressure data are collected through air pressure sensors to characterize the changes in atmospheric pressure; precipitation data are collected through rain gauges to record the changes in precipitation; wind speed data are collected through anemometers to measure the air flow velocity. Weather radar observation data mainly include radar reflectivity information. Radar reflectivity is calculated by the radar emitting electromagnetic waves and receiving the returned echo signals, which reflects the distribution and intensity characteristics of targets such as water vapor and precipitation in the atmosphere. The radar system emits electromagnetic waves of different wavelengths, receives the intensity and phase information of the returned signals, and obtains the reflectivity data after signal processing. The quality control of ground conventional observation data mainly includes three aspects of inspection: numerical range inspection, continuity inspection, and consistency inspection. Numerical range inspection mainly checks whether the observed data are within a reasonable physical range. For example, the temperature data are between -80°C and 60°C, the relative humidity is between 0% and 100%, and the air pressure is between 500 hPa and 1100 hPa. Continuity inspection mainly checks the time continuity of the observed data and judges whether the changes in data at adjacent times conform to physical laws. For example, the temperature will not change violently in a short period of time. Consistency inspection is to check whether the physical relationships between different elements are coordinated. For example, the relationships among temperature, humidity, and dew point temperature need to satisfy physical laws.

[0030] For the quality control of radar observation data, it includes data rationality inspection, time continuity inspection, and spatial consistency inspection. Data rationality inspection mainly checks whether the radar reflectivity data are within a physically reasonable range, generally between -30 dBZ and 75 dBZ. Time continuity inspection mainly checks whether the changes in reflectivity data at adjacent scan times are reasonable and excludes abnormal values caused by equipment failures or interference. Spatial consistency inspection is to check the rationality of the spatial distribution of radar observation data, including the continuity inspection of adjacent radial and azimuth data. The data after quality control need to be subjected to format conversion to uniformly convert the observation data from different sources into a common binary format. The common binary format is a standardized data storage format, which is convenient for the unified processing and analysis of different types of data. Information such as the spatio-temporal resolution and physical units of the data needs to be unified during the conversion process. At the same time, global forecast products also need to be collected as background field data. Global forecast products are grid forecast data generated based on numerical weather prediction models, which contain forecast information on multiple meteorological elements such as temperature, humidity, and air pressure. The quality-controlled observation data are combined with global forecast products to form a complete standardized data set, providing a data basis for subsequent multi-grid analysis and data assimilation.

[0031] During the data processing, there are often problems such as missing observations and outliers in the original observed data. These problematic data can be effectively identified and processed through quality control. Data rationality tests can identify outliers that significantly exceed the physical range, continuity tests can detect mutation points in the time series, and consistency tests can find contradictory values between different elements. The quality control process is actually a process of data screening and correction, ensuring the reliability and consistency of the data through a series of test criteria. During the data format conversion process, the spatio-temporal resolution differences of different data sources need to be considered. Ground observation data are usually discrete site observations, while radar data are continuous spatial scan data. Different-resolution data are unified into the same spatio-temporal framework through methods such as interpolation to ensure the comparability and consistency of the data. At the same time, the physical units of the data also need to be unified, for example, converting temperature to degrees Celsius uniformly and pressure to hectopascals uniformly.

[0032] For example, during a data processing, first, quality control is performed on the temperature data of a ground observation station. Through numerical range tests, it is found that the reported temperature value at a certain site is 85°C, which significantly exceeds the reasonable range, and it is marked as an outlier. Through continuity tests, it is found that the temperatures at two consecutive observation times at this site are 25°C and 26°C respectively, while an abnormal value of 85°C suddenly appears at the intermediate time, further confirming that this is an incorrect data that needs to be removed. Similar problems also exist in radar data. For example, an abnormal value of -50 dBZ appears in the reflectivity data at a certain azimuth angle. Through spatial consistency tests, it is found that the surrounding reflectivity values are all around 20 dBZ, indicating that this abnormal value may be caused by equipment failure or interference.

[0033] S120: Construct a multi-analysis grid based on the standardized data set to obtain a grid structure;

[0034] Optionally, determine the spatial range of the study area based on the standardized data set, divide the study area into equal-sized basic units along the X and Y directions to obtain a basic grid layout; perform hierarchical processing on the basic grid layout through a spatial configuration method to divide the basic grid layout into a coarse grid layer and a fine grid layer; set the grids of the coarse grid layer using the grid spacing ratio relationship, and set the grid spacing of the coarse grid as an integer multiple of the fine grid spacing; stack and combine the coarse grid layer and the fine grid layer according to the grid spacing ratio relationship to form a nested grid distribution structure; set the grid node parameters according to the nested grid distribution structure, and assign values to the grid node parameters according to the spatial distribution law; organize the nested grid distribution structure as a whole based on the grid node parameters to generate the grid structure.

[0035] Among them, the spatial range of the study area is determined based on the obtained standardized data set, which includes ground-based conventional observation data, radar observation data, and global forecast product data that have completed quality control. The determination of the spatial range needs to consider the spatial distribution characteristics of the observation data. By analyzing the longitude and latitude distribution of the observation stations and the coverage range of the radar observations, the longitude and latitude boundaries of the entire study area are determined. Subsequently, the determined study area is divided into basic units of equal size along the X and Y directions to form a basic grid layout. The division of the basic grid layout uses the equidistant grid division method, that is, the same grid spacing is used in the X and Y directions. The selection of the grid spacing needs to comprehensively consider the spatial resolution characteristics of the observation data. For the ground observation network, the basic grid spacing is usually set to 1 / 2 to 1 / 3 of the average spacing of the observation stations to ensure that the grid can fully express the observation information. For example, when the average spacing of the observation stations is 30 km, the basic grid spacing can be set to 10 km, such a setting can ensure that each grid cell contains at least the information of one observation station.

[0036] When performing hierarchical processing on the basic grid layout through the spatial configuration method, the idea of multiple grids is adopted, and the basic grid layout is divided into a coarse grid layer and a fine grid layer. The core of the hierarchical processing lies in determining the appropriate hierarchical relationship so that the coarse grid can describe the characteristics of large-scale weather systems, while the fine grid can depict local weather characteristics. When dividing the grid levels, the spatial variation characteristics of meteorological elements need to be considered. The scale of the coarse grid should be adapted to the characteristic scale of the large-scale weather system, and the scale of the fine grid needs to match the characteristic scale of the medium and small-scale weather systems. When setting the grids for the coarse grid layer using the grid spacing ratio relationship, the grid spacing of the coarse grid is set to an integer multiple of the fine grid spacing, and this multiple is usually between 2 and 5. This setting is based on two considerations: First, the integer multiple relationship facilitates the information transfer and scale conversion between grids; second, the range of 2 to 5 times can achieve a better balance between computational efficiency and accuracy. When the fine grid spacing is 10 km, the coarse grid spacing can be set to 20 km or 30 km, such a setting enables the coarse grid to effectively capture the characteristics of weather systems at a scale of 50 - 150 km. The coarse grid layer and the fine grid layer are superimposed and combined according to the grid spacing ratio relationship to form a nested grid distribution structure. The superimposition and combination process needs to ensure the spatial correspondence relationship of the grid nodes, and the nodes of the fine grid must be spatially consistent with the nodes of the coarse grid. For example, when the coarse grid spacing is 2 times the fine grid spacing, each coarse grid node coincides with the corresponding fine grid node, and this coincidence relationship facilitates subsequent data exchange.

[0037] When setting grid node parameters according to the nested grid distribution structure, the spatial distribution characteristics of the observed data need to be considered. The grid node parameters mainly include the spatial position coordinates of the nodes, the grid resolution information, and the topological relationships between the nodes. The assignment of grid node parameters follows the spatial distribution law to ensure that the grid structure can accurately reflect the spatial distribution characteristics of the observed data. For example, when setting the grid node parameters of the temperature field, the spatial continuity characteristics of the temperature field need to be considered, and an appropriate difference scheme is used to describe the relationships between the nodes. When organizing the nested grid distribution structure as a whole based on the grid node parameters, a complete grid index system needs to be established to provide a basis for subsequent data assimilation and information fusion. The grid index system includes information such as grid numbers, grid level relationships, and connection relationships between grids. Through this organization method, a multi-grid structure is formed that can describe both large-scale weather systems and local weather characteristics.

[0038] For example, assume that a multi-grid analysis of the temperature field in a certain area is required. First, the research area range is determined to be 300 km × 300 km according to the distribution of ground observation stations. Considering the spatial variation characteristics of the temperature field, the basic grid spacing is set to 10 km, forming a basic grid layout of 30 × 30. Subsequently, during the hierarchical processing, the coarse grid spacing is set to 20 km, forming a coarse grid layer of 15 × 15 to describe the large-scale temperature field distribution. The fine grid maintains a 10 km spacing to depict the local temperature field details. Through this grid setting, both the large-scale change trend and the local temperature gradient of the temperature field can be reasonably expressed. The assignment of grid node parameters is based on the actual observed data. The discrete site observations are mapped to the regular grid through interpolation methods, and the corresponding relationships between the coarse and fine grids are established. During the data processing process, each grid node stores the observed or analyzed values of meteorological elements. For example, the temperature value of a certain fine grid node is obtained by performing distance-weighted averaging on the temperature observation values of the surrounding observation stations, while the temperature value of the corresponding coarse grid node is obtained by performing regional averaging on the values of the fine grid nodes within its coverage area. This multi-grid structure not only improves the calculation efficiency but also ensures the effective fusion of information at different scales.

[0039] S130: Using the grid structure, separating data features through transformation and filtering processing to obtain multi-scale feature data;

[0040] Optionally, in some embodiments, the meteorological data in the grid structure is subjected to two-dimensional discrete cosine transform to obtain a spectral coefficient field; a smooth transition filter function is set for the spectral coefficient field, and the response value of the smooth transition filter function is assigned within the wavelength range; the smooth transition filter function is used to screen the wavelength of the spectral coefficient field to separate the large-scale wavelength component and the small-scale wavelength component; the large-scale wavelength component is subjected to inverse transform processing to obtain a large-scale feature quantity; the small-scale wavelength component is subjected to inverse transform processing to obtain a small-scale feature quantity; the large-scale feature quantity and the small-scale feature quantity are combined to form the multi-scale feature data.

[0041] Among them, using the grid structure for data feature separation is a key link in multi-source data fusion, and effective extraction of multi-scale features needs to be achieved through mathematical processing means such as transformation and filtering. First, the meteorological data in the grid structure needs to be processed by two-dimensional discrete cosine transform (2D-DCT). Two-dimensional discrete cosine transform is a mathematical tool that converts spatial domain data to frequency domain. Its core idea is to represent two-dimensional data as a linear combination of cosine functions with different frequencies. For a two-dimensional data matrix f(n, m) of N×M, its two-dimensional discrete cosine transform can be expressed as:

[0042]

[0043] where: F(u, v) is the transformed frequency domain coefficient, also known as the spectral coefficient field; f(n, m) is the original spatial domain data; α(u) and α(v) are normalization coefficients; N and M are the number of rows and columns of the data matrix respectively; u and v are frequency domain variables, and their value ranges are 0≤u≤N - 1 and 0≤v≤M - 1 respectively.

[0044] Converting the original meteorological data into a spectral coefficient field enables the manifestation of information at different scales in the frequency domain. In the spectral coefficient field, low-frequency components correspond to large-scale features, and high-frequency components correspond to small-scale features.

[0045] A smooth transition filter function needs to be set for the obtained spectral coefficient field. The design of this function adopts the form based on the coS 2 function to ensure smooth transition between different wavelength ranges and avoid Gibbs phenomenon caused by mutations. The mathematical expression of the filter function is:

[0046]

[0047] where: H(λ) is the response value of the filter function; λ is the wavelength; λ 1 is the lower cut-off wavelength; λ 2 is the upper cut-off wavelength.

[0048] For mesoscale weather systems, the wavelength range is generally between 100 km and 1500 km. Therefore, λ 1 needs to be set around 100 km, and λ 2 needs to be set around 1500 km. Such settings can effectively isolate the weather features of meso - and small - scales while retaining the characteristics of large - scale weather systems.

[0049] Apply the designed smooth - transition filtering function to the spectral coefficient field to separate the large - scale and small - scale components through wavelength screening. The wavelength screening process is actually multiplying the filtering function by the spectral coefficients to obtain frequency components of different scales. The large - scale wavelength components are composed of the low - frequency part, that is, the part with wavelengths greater than λ 2 ; the small - scale wavelength components are composed of the high - frequency part, that is, the part with wavelengths less than λ 1 ; and in the transition band between λ 1 and λ 2 , it is distributed according to the weight of the cos 2 function. Perform an inverse transformation on the separated large - scale wavelength components, that is, perform a two - dimensional inverse discrete cosine transform (IDCT) to obtain the large - scale characteristic quantities in the spatial domain. The mathematical expression of the inverse transformation is:

[0050]

[0051] Similarly, perform the same inverse transformation on the small - scale wavelength components to obtain the small - scale characteristic quantities in the spatial domain. In this way, the original meteorological field is decomposed into two parts: a large - scale background field and a small - scale perturbation field. Finally, combine the obtained large - scale and small - scale characteristic quantities to form complete multi - scale characteristic data. The combination process needs to consider the relative importance between the two - scale characteristic quantities and ensure the reasonable fusion of information at different scales through appropriate weight configuration.

[0052] For example: Suppose there is now a temperature field data of 300 km×300 km with a grid spacing of 10 km, forming a 30×30 data matrix. First, perform a two - dimensional discrete cosine transform on the temperature field data to obtain the spectral coefficients in the frequency domain. Then design a filtering function, set λ 1 to 100 km, and λ 2Set it to 300 km. After filtering, the low-frequency component reflecting the large-scale temperature distribution trend and the high-frequency component reflecting the local temperature change characteristics are obtained. Through inverse transformation, the large-scale temperature field and the small-scale temperature perturbation field are obtained respectively. In this embodiment, the temperature observation data of a region is first converted into a spectral coefficient matrix, and the temperature field characteristics of different scales are separated by filtering. The large-scale temperature field usually shows the overall temperature gradient and the large-scale temperature distribution characteristics, while the small-scale temperature field reflects the local temperature anomalies and detailed characteristics. This separation enables the subsequent data assimilation process to process the temperature field information of different scales separately, improving the accuracy of the assimilation result. Finally, by recombining the large-scale temperature field and the small-scale temperature field, a complete temperature field that retains both the overall temperature distribution characteristics and includes local detailed information is obtained.

[0053] S140: Construct an observation operator based on the multi-scale feature data to form an observation operator system;

[0054] Optionally, in some embodiments, perform spatial continuity analysis on the large-scale feature data in the multi-scale feature data to construct a large-scale spatial mapping function; perform local feature analysis on the small-scale feature data in the multi-scale feature data to construct a small-scale spatial mapping function; set large-scale observation error parameters according to the large-scale spatial mapping function to obtain a large-scale observation operator; set small-scale observation error parameters according to the small-scale spatial mapping function to obtain a small-scale observation operator; combine the large-scale observation operator and the small-scale observation operator according to the scale distribution; generate the observation operator system based on the combination relationship between the large-scale observation operator and the small-scale observation operator.

[0055] Among them, constructing an observation operator system based on multi-scale feature data is a key link in realizing data assimilation. It is necessary to process the large-scale and small-scale feature data separately and establish corresponding observation operators. When performing spatial continuity analysis on the large-scale feature data in the multi-scale feature data, first evaluate the change characteristics of the data in space, including the gradient distribution, continuity degree, and spatial correlation of the data. The spatial continuity analysis uses the variogram method to quantify the spatial correlation by calculating the data differences at different spatial distances. Based on the results of the spatial continuity analysis, a large-scale spatial mapping function is constructed, which describes the transformation relationship from the observation space to the model space. When performing local feature analysis on the small-scale feature data, focus on the local change characteristics and small-scale structures of the data. The local feature analysis includes local anomaly detection, local correlation analysis, and local scale feature extraction. Through these analyses, a small-scale spatial mapping function is constructed, which can accurately describe the corresponding relationship between the observations and the model variables at the local scale.

[0056] When setting large-scale observation error parameters according to the large-scale spatial mapping function, it is necessary to consider the sources of uncertainty in the observation data at the large scale. Large-scale observation errors mainly include instrument systematic errors, representativeness errors, and spatial interpolation errors. By analyzing these error sources and setting appropriate error parameters, a large-scale observation operator can be obtained. The role of the large-scale observation operator is to map the model variables to the large-scale observation space, making the model output comparable to the observation data. When setting small-scale observation error parameters according to the small-scale spatial mapping function, it is necessary to focus on the observation uncertainty at the local scale. Small-scale observation errors mainly come from the random errors of observations, local representativeness errors, and the uncertainty of small-scale physical processes. By evaluating these error characteristics and setting the corresponding error parameters, a small-scale observation operator is constructed. The small-scale observation operator is responsible for processing the observation information at the local scale to ensure that the model can accurately describe small-scale weather phenomena.

[0057] When combining the large-scale observation operator and the small-scale observation operator according to the scale distribution, it is necessary to consider the interaction between the two scales. In the combination process, a weight function is used to adjust the contributions of the observation operators at different scales, and the weights are set based on the quality, reliability, and physical meaning of the observation data. Through this combination method, a complete observation operator system is formed, which can not only process the observation information of large-scale weather systems but also effectively utilize the small-scale local observation data. Taking the temperature field observation as an example: First, conduct a spatial continuity analysis of the large-scale temperature field characteristics to evaluate the spatial correlation of the temperature field. By calculating the temperature correlation coefficients between different stations, the spatial correlation scale of the temperature field is determined. Based on these analysis results, a spatial mapping function of the large-scale temperature field is constructed, which describes the corresponding relationship between the observed temperature and the model temperature. At the same time, by analyzing the systematic errors and representativeness errors of the temperature observations, the error parameters of the large-scale temperature observations are set.

[0058] For the characteristics of the small-scale temperature field, focus on analyzing the characteristics of local temperature changes, such as temperature anomalies caused by the urban heat island effect, terrain influence, etc. Construct a spatial mapping function for the small-scale temperature field, which can describe the relationship between local temperature anomalies and the model temperature field. By evaluating the random error and representativeness error of local temperature observations, set the error parameters for small-scale temperature observations. When combining the large-scale and small-scale temperature observation operators, set the weights according to the quality and reliability of the observation data to form a complete temperature field observation operator system. Such an observation operator system is not only applicable to the temperature field, but also applicable to other meteorological elements such as the humidity field, wind field, etc. The key lies in accurately evaluating the observation characteristics and error characteristics at different scales, constructing a suitable spatial mapping function, setting reasonable error parameters, and finally forming an observation operator system that can effectively process multi-source observation data. The construction of the observation operator system directly affects the effect of data assimilation, so special attention needs to be paid to the reasonable setting of observation errors and the coordinated cooperation of observation operators at different scales.

[0059] S150: Perform variational assimilation processing on the grid data according to the observation operator system to obtain an assimilation analysis field;

[0060] Optionally, in some embodiments, set the error parameters for the observation operator system, construct a background error covariance matrix and an observation error covariance matrix; establish a variational assimilation cost function based on the background error covariance matrix and the observation error covariance matrix; substitute the observation operator system into the variational assimilation cost function to establish a variational assimilation solution equation; perform numerical iterative calculation on the variational assimilation solution equation to obtain the optimal solution of the model variables; perform a mapping transformation on the optimal solution and the observation operator system to obtain the assimilation result in the observation space; organize and sort the data based on the assimilation result to generate the assimilation analysis field.

[0061] Among them, error parameters are set for the observation operator system, and the background error covariance matrix and the observation error covariance matrix are constructed respectively. The background error covariance matrix describes the statistical characteristics of the background field error, including error variance and spatial correlation. The evaluation of the background error needs to consider the spatio-temporal distribution characteristics of the model prediction error, and determine the statistical characteristics of the background error through the comparative analysis of historical forecasts and observations. The observation error covariance matrix describes the error characteristics of the observation data, including the uncertainties from multiple sources such as instrument error and representativeness error. The evaluation of the observation error needs to combine the performance indicators of the observation system and the actual observation conditions. Based on the background error covariance matrix and the observation error covariance matrix, a variational assimilation cost function is established. The cost function consists of two parts: the first part measures the difference between the assimilation analysis field and the background field, and the second part measures the difference between the assimilation analysis field and the observed values. These two parts of the difference are weighted by the background error covariance matrix and the observation error covariance matrix respectively. The design goal of the cost function is to find the best balance between the background field information and the observation information, so that the assimilation result not only conforms to the physical laws but also can make full use of the observation information.

[0062] The observation operator system is substituted into the variational assimilation cost function to establish a variational assimilation solution equation. The role of the observation operator system is to map the variables in the model space to the observation space, making the model variables comparable with the observed values. The variational assimilation solution equation is essentially an optimization problem that needs to find the model state variables that minimize the cost function. Numerical iterative calculations are performed on the variational assimilation solution equation, and optimization algorithms such as the conjugate gradient method or the quasi-Newton method are used to find the minimum value of the cost function. During the iterative calculation process, the cost function value and its gradient need to be calculated simultaneously. By continuously adjusting the model variables, the cost function is gradually reduced until convergence. The convergence criteria for the iterative process include the relative change value of the cost function and the relative change value of the gradient. When these indicators are less than the preset threshold, the iteration is considered to converge.

[0063] The obtained optimal solution is subjected to a mapping transformation with the observation operator system to obtain the assimilation result in the observation space. The purpose of this step is to verify the rationality of the assimilation result. By mapping the optimal solution to the observation space, it can be directly compared with the observed values to evaluate the assimilation effect. The mapping transformation process needs to use the space mapping function defined in the observation operator system. Based on the assimilation result, data organization and collation are carried out to generate an assimilation analysis field. Data organization and collation include tasks such as format conversion, quality inspection, and data archiving. The assimilation analysis field needs to meet the input requirements of the numerical weather prediction model, so corresponding format conversion and data collation are required.

[0064] Taking temperature field assimilation as an example: First, analyze the statistical characteristics of the background field temperature forecast error based on historical data to construct the background error covariance matrix. At the same time, establish the observation error covariance matrix based on the performance indicators of the temperature observation system. The established variational assimilation cost function reflects the deviation between the analyzed value of the temperature field and the background field and the observed value. Through numerical iterative calculations, find the state that minimizes the temperature field cost function, which not only maintains the spatial continuity of the temperature distribution in the background field but also reasonably absorbs the observation information. Finally, normalize the assimilated temperature field to generate a temperature field analysis product that meets the requirements of the numerical prediction model. The key to the variational assimilation process lies in accurately evaluating the error characteristics, reasonably setting the weights of the cost function, and adopting an efficient numerical optimization algorithm. In this way, the background field and observation information can be effectively fused to obtain a physically reasonable and highly accurate assimilated analysis field. The quality of the assimilated analysis field directly affects the subsequent numerical prediction effect, so special attention needs to be paid to each technical link in the assimilation process to ensure the accuracy and reliability of the assimilation results.

[0065] S160: Perform information exchange processing between grids based on the assimilated analysis field to obtain a fusion result;

[0066] Optionally, in some embodiments, convert the coarse grid data in the assimilated analysis field to the fine grid coordinate system through interpolation mapping to obtain initial fine grid data; perform local assimilation calculations on the initial fine grid data to generate a fine grid assimilation field; extract large-scale information from the fine grid assimilation field to form coarse grid correction data; superimpose the coarse grid correction data on the coarse grid data in the assimilated analysis field to obtain updated coarse grid data; perform boundary constraints on the fine grid assimilation field based on the updated coarse grid data to generate grid fusion data; perform data integration processing on the grid fusion data to form the fusion result.

[0067] Among them, the coarse grid data in the assimilation analysis field is transformed into the fine grid coordinate system through interpolation mapping. The bilinear interpolation method is adopted in the interpolation mapping process to ensure the smooth transition of data in space. The continuity characteristics of physical quantities need to be considered in the interpolation process. For scalar fields such as temperature and air pressure, distance-weighted interpolation is used; for vector fields such as wind fields, interpolation of both direction and magnitude needs to be considered. When performing local assimilation calculations on the initial fine grid data, the influence of local observation information is mainly considered. Local optimal interpolation method is used for local assimilation, and the observation data within a certain range around the observation site are weighted and averaged. The setting of weights is based on the distance from the observation point to the grid point and the quality of the observation data. The local balance relationship of physical quantities needs to be maintained during the local assimilation calculation process to ensure the physical rationality of the assimilation results. When extracting large-scale information from the fine grid assimilation field, a spatial filtering method is used for scale separation. The moving average or regional average method is used in the filtering process, and the size of the average area is comparable to the coarse grid scale. The extracted large-scale information reflects the influence of the fine grid assimilation results on the coarse grid scale, and this information is used to correct the coarse grid data. When superimposing the corrected coarse grid data with the coarse grid data in the assimilation analysis field, reasonable weight coefficients need to be set. The setting of weights considers two factors: one is the reliability of the coarse grid data, and the other is the representativeness of the fine grid assimilation results. The weighted average method is used in the superimposing process to ensure that the corrected coarse grid data not only maintains the large-scale characteristics but also contains the information feedback from the fine grid.

[0068] When applying boundary constraints to the fine grid assimilation field based on the updated coarse grid data, a nested boundary processing technique is adopted. The boundary constraints ensure the coordination and consistency of the fine grid data with the coarse grid data in the boundary region. The relaxation zone processing method is used in the constraint process, and a transition zone is set near the boundary to make the fine grid data gradually transition to the coarse grid data. When performing data integration processing on the grid fusion data, the consistency and integrity of the data need to be ensured. The integration processing includes the unification of data formats, the processing of missing values, and the balance check of physical quantities. The final formed fusion result needs to meet the input requirements of the numerical prediction model.

[0069] For example: First, the coarse grid temperature field with a resolution of 30 km is converted to a fine grid with a resolution of 10 km through bilinear interpolation to obtain the initial fine grid temperature field. Then, local assimilation is performed using the temperature observation data within the 10 km grid range to generate a fine grid temperature field containing local characteristics. Large-scale temperature information is extracted from the fine grid temperature field through area averaging of 30 km × 30 km to correct the coarse grid temperature field. The corrected coarse grid temperature field affects the fine grid temperature field through boundary constraints, and a transition zone with a width of 5 grid points is set in the boundary area between the two grids to achieve a smooth transition of the temperature field. Finally, the integrated temperature field data is processed to ensure the spatial continuity and physical rationality of the temperature field. Through this information exchange mechanism between grids, the organic integration of coarse and fine grid data is achieved. The key lies in reasonably setting the processing parameters of each link, such as the selection of interpolation methods, the determination of weight coefficients, the width of the boundary transition zone, etc. This multi-scale fusion method is not only applicable to temperature fields, but also to other meteorological elements such as pressure fields and wind fields. The advantage of the method is that it can simultaneously maintain the overall characteristics of large-scale weather systems and the detailed characteristics of local weather phenomena, improving the accuracy and reliability of data fusion.

[0070] It should be noted that for the foregoing method embodiments, for the sake of simple description, they are all expressed as a series of action combinations. However, those skilled in the art should know that the present disclosure is not limited by the described action sequence, because according to the present disclosure, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily essential to the present disclosure.

[0071] The above is the introduction of method embodiments. The following further illustrates the solution of the present disclosure through device embodiments.

[0072] Figure 2 The block diagram of a multi-source data fusion device 200 based on improved multi-grid variational assimilation according to an embodiment of the present disclosure is shown. As Figure 2 shown, the device 200 includes:

[0073] A control module 210, configured to perform quality control on multi-source observation data according to meteorological standards to obtain a standardized data set;

[0074] An analysis module 220, configured to construct a multi-analysis grid based on the standardized data set to obtain a grid structure;

[0075] A transformation module 230, configured to use the grid structure to separate data features through transformation and filtering processes to obtain multi-scale feature data;

[0076] A construction module 240, configured to construct an observation operator based on the multi-scale feature data to form an observation operator system;

[0077] An assimilation module 250, configured to perform variational assimilation processing on grid data according to the observation operator system to obtain an assimilation analysis field;

[0078] An exchange module 260, configured to perform inter-grid information exchange processing based on the assimilation analysis field to obtain a fusion result.

[0079] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the described modules can refer to the corresponding processes in the foregoing method embodiments, and will not be elaborated herein.

[0080] The above specific implementation manners do not constitute a limitation to the protection scope of the present disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principle of the present disclosure shall be included within the protection scope of the present disclosure.

Claims

1. A multi-source data fusion method based on improved multi-grid variational assimilation, characterized in that: include: The multi-source observation data are quality controlled according to meteorological standards to obtain a standardized data set; constructing a multiple analysis grid according to the standardized data set to obtain a grid structure; Using the grid structure, data features are separated by transformation and filtering to obtain multi-scale feature data; constructing an observation operator based on the multi-scale feature data to form an observation operator system; Performing variational assimilation processing on the grid data according to the observation operator system to obtain an assimilated analysis field; The information exchange process between grids is performed according to the assimilated analysis field to obtain a fusion result.

2. The method according to claim 1, characterized in that The multi-source observation data is quality controlled according to meteorological standards to obtain a standardized data set, including: Collecting conventional ground observation data through ground meteorological stations, wherein the conventional ground observation data includes temperature, humidity, air pressure, precipitation, and wind speed; Collecting radar observation data by weather radar, wherein the radar observation data includes radar reflectivity; Conducting quality control on the conventional ground observation data according to the numerical range test, continuity test and consistency test in the ground standard for quality control of meteorological observation data; The radar observation data is quality controlled according to the data rationality test, time continuity test and spatial consistency test in the weather radar standard for meteorological observation data quality control; Convert the format of the ground conventional observation data and radar observation data after quality control to generate standardized data in universal binary format; A global forecast product is collected, and the global forecast product is combined with the standardized data in the universal binary format to form the standardized data set.

3. The method according to claim 1, characterized in that The step of constructing a multiple analysis grid according to the standardized data set to obtain a grid structure comprises: Determine the spatial scope of the study area based on the standardized data set, divide the study area into basic units of equal size along the X and Y directions, and obtain a basic grid layout; hierarchically processing the basic grid layout by a spatial configuration method, dividing the basic grid layout into a coarse grid layer and a fine grid layer; The coarse grid layer is grid-set using a grid spacing ratio relationship, and the grid spacing of the coarse grid is set to an integer multiple of the fine grid spacing; The coarse grid layer and the fine grid layer are superimposed and combined according to the grid spacing ratio to form a nested grid distribution structure; Setting grid node parameters according to the nested grid distribution structure, and assigning values ​​to the grid node parameters according to a spatial distribution rule; The nested grid distribution structure is organized as a whole based on the grid node parameters to generate the grid structure.

4. The method according to claim 1, characterized in that: The method of utilizing the grid structure to separate data features through transformation and filtering to obtain multi-scale feature data includes: Performing a two-dimensional discrete cosine transform on the meteorological data in the grid structure to obtain a spectral coefficient field; Setting a smooth transition filter function for the spectral coefficient field, and assigning a response value of the smooth transition filter function within a wavelength range; Using the smooth transition filter function to perform wavelength screening on the spectrum coefficient field to separate large-scale wavelength components and small-scale wavelength components; Performing inverse transformation processing on the large-scale wavelength component to obtain a large-scale characteristic quantity; Performing inverse transformation processing on the small-scale wavelength component to obtain a small-scale characteristic quantity; The large-scale feature quantity and the small-scale feature quantity are combined to form the multi-scale feature data.

5. The method according to claim 1, characterized in that The constructing of an observation operator based on the multi-scale feature data to form an observation operator system includes: Performing spatial continuity analysis on the large-scale feature data in the multi-scale feature data to construct a large-scale spatial mapping function; Performing local feature analysis on small-scale feature data in the multi-scale feature data to construct a small-scale spatial mapping function; Setting a large-scale observation error parameter according to the large-scale spatial mapping function to obtain a large-scale observation operator; Setting a small-scale observation error parameter according to the small-scale space mapping function to obtain a small-scale observation operator; Combining the large-scale observation operator and the small-scale observation operator according to scale distribution; The observation operator system is generated based on the combination relationship between the large-scale observation operator and the small-scale observation operator.

6. The method according to claim 1, characterized in that The step of performing variational assimilation processing on the grid data according to the observation operator system to obtain an assimilated analysis field includes: Setting error parameters for the observation operator system, and constructing a background error covariance matrix and an observation error covariance matrix; Establishing a variational assimilation cost function based on the background error covariance matrix and the observation error covariance matrix; Substituting the observation operator system into the variational assimilation cost function to establish a variational assimilation solution equation; Performing numerical iteration calculation on the variational assimilation solution equation to obtain the optimal solution of the model variables; Mapping and transforming the optimal solution and the observation operator system to obtain an assimilation result of the observation space; The data is organized and collated based on the assimilation result to generate the assimilation analysis field.

7. The method according to claim 1, characterized in that The inter-grid information exchange processing is performed based on the assimilated analysis field to obtain a fusion result, including: The coarse grid data in the assimilation analysis field is converted to a fine grid coordinate system through interpolation mapping to obtain initial fine grid data; Performing local assimilation calculation on the initial fine grid data to generate a fine grid assimilation field; Extracting large-scale information from the fine-grid assimilation field to form coarse-grid correction data; Superimposing the coarse grid correction data with the coarse grid data in the assimilation analysis field to obtain updated coarse grid data; Perform boundary constraints on the fine grid assimilation field based on the updated coarse grid data to generate grid fusion data; The grid fusion data is subjected to data integration processing to form the fusion result.

8. A multi-source data fusion device based on improved multi-grid variational assimilation, characterized in that: include: A control module is used to perform quality control on multi-source observation data according to meteorological standards to obtain a standardized data set; An analysis module, used for constructing a multiple analysis grid according to the standardized data set to obtain a grid structure; A transformation module, used to utilize the grid structure to separate data features through transformation and filtering to obtain multi-scale feature data; A construction module, used to construct an observation operator based on the multi-scale feature data to form an observation operator system; An assimilation module, used for performing variational assimilation processing on the grid data according to the observation operator system to obtain an assimilated analysis field; The exchange module is used to perform information exchange processing between grids based on the assimilation analysis field to obtain a fusion result.

Images