Symbolic isotherm extraction and semantic character generation method

Abstract

The invention discloses a symbolic isotherm extraction and semantic character generation method. The method comprises the following steps: step P1, identifying valid points, isolated points and invalid points from an original data matrix A and marking the valid points, the isolated points and the invalid points on corresponding positions of an adjoint matrix B; p2, interpolating the adjoint matrix B and modifying the adjoint matrix B at a corresponding position of the original data matrix A; step P3, smoothing the interpolated data matrix A1, and modifying a mark value at a corresponding position of the adjoint matrix B1; step P4, eliminating internal points in an isothermal surface area of the smoothed adjoint matrix B2; and step P5, storing effective points in the accompanying matrix B3 after fine screening into a warehouse. Addressing and connecting in the adjoint matrix B3 through the step P6 and the step P7 to realize isotherm connection; step P8, finding out a first isotherm which traverses a 5-kilometer buffer area of the Chinese east-west boundary line as a symbolic isotherm; and step P9, generating semantic information. The problem that an existing symbolic isotherm extraction method is difficult to meet meteorological service application requirements can be solved.

Description

Technical Field

[0001] This invention relates to the field of meteorological grid data processing technology. Specifically, it is a method for extracting landmark isotherms and generating semantic text. Background Technology

[0002] In meteorological services and related industry applications, landmark isotherms specifically refer to complete and representative isotherms extending from west to east within the Chinese region, typically such as the 0℃ isotherm, -5℃ isotherm, 5℃ isotherm, and 10℃ isotherm. The location of landmark isotherms provides crucial reference for agricultural production planning, transportation scheduling, and public life arrangements in my country. Around the time of seasonal transitions, forecasters at the National Meteorological Center conduct real-time weather reviews and future weather trend analyses based on landmark isotherms. Therefore, accurately extracting, fully plotting, and automatically generating relevant semantic text for landmark isotherms has become a core requirement for ensuring the efficiency of meteorological consultations and improving service quality.

[0003] However, the processing and application of landmark isotherms in meteorological operational consultation scenarios currently faces numerous unresolved issues. On one hand, different operational systems and statistical methods result in inconsistent standards for landmark isotherm labeling. During the extraction and plotting of isotherms from real-time grid data, segmentation and convergence problems are common—due to differences in extraction methods, the generated map segments are uneven and disorganized, severely impacting forecasters' efficiency in interpreting and analyzing maps. On the other hand, the generation of semantic text related to landmark isotherms remains primarily manual. Forecasters or consultation support personnel must manually interpret maps to understand the isotherm locations and characteristics before writing the text. This method is not only inefficient but also susceptible to subjective factors, making it difficult to guarantee the accuracy and standardization of the textual descriptions, significantly increasing the workload of consultation service support personnel.

[0004] Currently, existing isotherm extraction technologies are mainly divided into two categories: numerical calculation-based methods and image processing-based methods. However, neither can meet the specific needs of meteorological service support for processing iconic isotherms. Among them, numerical calculation-based methods interpolate discrete temperature data points to obtain a continuous raster dataset, and then extract points with the same temperature value to connect them into lines. Although this method can handle irregularly distributed data and has high extraction accuracy, the interpolation process is prone to introducing errors in sparse data or regions with drastic temperature changes, and the extraction efficiency is significantly limited by the size of the raster dataset. Image processing-based methods target carriers such as meteorological fax images, and extract isotherms through techniques such as interference removal and traversal recognition. Although this method is simple to implement and highly efficient, it is extremely dependent on image quality. Noise and interference information will seriously affect the extraction effect, and it is difficult to adapt to complex isotherm shapes.

[0005] The two methods mentioned above are designed for discrete point data and image products, respectively. However, they fall short of the application requirements for landmark isotherms in meteorological operations in terms of extraction accuracy, plotting completeness, and semantic generation adaptability. They cannot meet the needs of morning weather consultations and the refined meteorological data services required by various industries. Therefore, developing a method that can adapt to grid data of different resolutions (including real-time, forecast, and reanalysis grid data) and achieve accurate extraction, complete plotting, and automatic semantic text generation of landmark isotherms has become a key issue urgently needing breakthroughs in the field of meteorological service technology. Summary of the Invention

[0006] Therefore, the technical problem to be solved by the present invention is to provide a method for extracting landmark isotherms and generating semantic text, so as to solve the problem that the existing landmark isotherm extraction methods are difficult to meet the application needs of meteorological business in terms of extraction accuracy, plotting completeness and semantic generation adaptability, so as to meet the requirements of morning weather consultation and the refined meteorological data services of various industries.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0008] A method for extracting iconic isotherms and generating semantic text includes the following steps:

[0009] Step P1, coarse screening: Store the temperature values ​​obtained from parsing the meteorological grid data into the original data matrix A; screen and identify valid points, isolated points and invalid points from the original data matrix A and mark their corresponding positions in the adjoint matrix B; the dimensions of the adjoint matrix B are completely consistent with those of the original data matrix A;

[0010] Step P2, Interpolation: Perform eight-neighbor interpolation calculations around the valid points of the adjoint matrix B using the interpolation operator to obtain the adjoint matrix B1; Traverse the adjoint matrix B1 and modify the temperature values ​​of the valid points at the corresponding positions in the original data matrix A to the temperature values ​​of the landmark isotherms to obtain the data matrix A1.

[0011] Step P3, Smoothing: Perform eight-neighborhood smoothing calculation around each valid point in data matrix A1 using the smoothing operator to obtain data matrix A2; traverse data matrix A2 and modify the label values ​​of the corresponding positions of valid points in adjoint matrix B1 to obtain adjoint matrix B2.

[0012] Step P4, Fine Screening: Remove points inside the isothermal surface region in the adjoint matrix B2; that is: traverse the adjoint matrix B2. If all points within the eight-neighbor range of any point M in the adjoint matrix B2 are valid points, then any point M is determined to be a point inside the isothermal surface region; mark the points inside the isothermal surface region as invalid points and remove them to obtain the adjoint matrix B3; points inside the isothermal surface region will not participate in the subsequent landmark isotherm addressing calculation; the purpose of fine screening is to remove points inside the target region, reduce the number of valid points calculated, and improve the efficiency and accuracy of subsequent landmark isotherm addressing.

[0013] Step P5, Storage: Batch store the valid points filtered from the accompaniment matrix B3 into the spatial database, sort them by latitude, and add GIST indexes to the point type fields.

[0014] Step P6, Addressing: In the adjoint matrix B3, the starting point is found by searching the spatial database, and the next valid point is found according to the direction of movement and the counterclockwise eight-neighbor addressing method to realize the isotherm connection.

[0015] Step P7, Connection: If the target value is not found during addressing, the isotherm connection is achieved by performing a neighborhood search through the spatial database;

[0016] Step P8, Identification: Sort the extracted isotherms according to the generation order from low latitude to high latitude, and find the first isotherm that runs through a 5-kilometer buffer zone across the east-west border of China, which is the landmark isotherm.

[0017] Step P9, Semantic Generation: Overlay the extracted landmark isotherms with geospatial data to find the provinces covered by the line segment and generate semantic information.

[0018] The method of this invention can quickly generate iconic isotherms and semantic text within China, making it easier for chief forecasters and users to understand the temperature distribution in grid data.

[0019] The above-mentioned method for extracting landmark isotherms and generating semantic text includes the following steps in step P1:

[0020] Step P101: Parse the meteorological grid data to obtain temperature values, and store the temperature values ​​in a two-dimensional array A as the original data matrix A; define a two-dimensional array B with the same dimensions as the two-dimensional array A as the adjoint matrix B; synchronously acquire the grid spatial resolution, data time, observation element code, data starting longitude, data starting latitude, number of observation element grid rows, and number of observation element grid columns of the meteorological grid data.

[0021] Step P102: Define the interval values ​​of the landmark isotherms as the target value range; the target value range is the target value ± 0.05℃; traverse the original data matrix A. If the temperature value at any point M in the original data matrix A is within the target value range, and at least one of the temperature values ​​at any eight neighboring locations of any point M is within the target value range, then mark the corresponding position of any point M in the adjoint matrix B as a valid point and assign a valid value; if the temperature value at any point M in the original data matrix A is within the target value range, and none of the temperature values ​​at any eight neighboring locations of any point M are within the target value range, then mark the corresponding position of any point M in the adjoint matrix B as an isolated point and assign an invalid value; if the temperature value at any point M in the original data matrix A is not within the target value range, or the temperature value at any point M in the original data matrix A is NaN or 999999, then mark the corresponding position in the adjoint matrix B as an invalid point and assign an invalid value.

[0022] The eight neighborhoods are the grid points in the eight directions of east, west, south, north, southeast, northeast, southwest, and northwest of any given point.

[0023] The above-mentioned method for extracting landmark isotherms and generating semantic text, in step P2, aims to increase the density of effective points by performing interpolation calculations around the effective points using interpolation operators. Specifically, this includes the following steps:

[0024] Step P201: Set the interpolation operator: Select any grid points in the eight neighborhoods around the point to be interpolated. If at least two grid points in the eight neighborhoods have temperature values ​​within the target value range, then modify the point to be interpolated as a valid point and assign a valid value.

[0025] Step P202: Define a temporary matrix C with the same dimensions as the adjoint matrix B. Traverse the adjoint matrix B, use the interpolation operator to perform interpolation calculations on each point to be interpolated in the adjoint matrix B, and write the interpolation calculation results into the corresponding positions in the temporary matrix C.

[0026] Step P203: After the interpolation calculation of all interpolation points in the adjoint matrix B is completed, assign the value of the temporary matrix C to the adjoint matrix B.

[0027] Step P204: Repeat steps P202 and P203 to perform interpolation iteration 8-12 times (preferably 10 times) to obtain the adjoint matrix B1;

[0028] Step P205: Traverse the adjoint matrix B1. If the position of the interpolation point in the adjoint matrix B1 is a valid value, then modify the corresponding position of the interpolation point in the original matrix A to the temperature value of the landmark isotherm, and obtain the data matrix A1.

[0029] The above-mentioned method for extracting landmark isotherms and generating semantic text includes the following steps in step P3:

[0030] Step P301: Define the smoothing operator: Define the average temperature value of any point and the eight neighboring grid points around the point as the smoothing operator;

[0031] Step P302: Define a temporary matrix D with dimensions that are exactly the same as those of the data matrix A1. Traverse the data matrix A1, use the smoothing operator to perform smoothing calculations on each point in the data matrix A1, and write the smoothing calculation results into the corresponding positions of the temporary matrix D.

[0032] Step P303: After the smoothing calculation of all points in the data matrix A1 is completed, assign the value of the temporary matrix D to the data matrix A1;

[0033] Step P304: Repeat steps P302 and P303 for smoothing iteration 25-35 times (preferably 30 times) to obtain data matrix A2;

[0034] Step P305: Traverse the data matrix A2. According to the temperature value of each point in the data matrix A2, re-mark the valid points, isolated points and invalid points in the adjoint matrix B1 according to the method in step P102 to obtain the adjoint matrix B2.

[0035] This invention effectively avoids data pollution by constructing a temporary matrix for interpolation and smoothing. By controlling the number of interpolation iterations to within 8-12 times and the number of smoothing iterations to within 25-35 times, the accuracy of isotherm extraction location can be significantly improved and the deviation of isotherm extraction location can be reduced.

[0036] The above-mentioned method for extracting landmark isotherms and generating semantic text includes the following steps in step P5:

[0037] Step P501, Data Storage Class Construction and Object Assignment: The variables of the data storage class include index number (ID), row number (i), column number (j), longitude (lon), latitude (lat), and point type (point); assign the parameters corresponding to each point in the accompanying matrix B3 to the objects generated by the data class storage, and store the objects sequentially in the array list A of the spatial database; the index number of the point in the accompanying matrix B3 in the spatial database is:

[0038] (1);

[0039] In equation (1), ID is the index number, i is the row number, n is the column number, and j is the column number;

[0040] Step P502, Border Area Data Filtering: Traverse array list A, filter and remove objects whose geographical location is not within the border area of ​​China, and obtain array list B;

[0041] Step P503, Spatial Database Storage and Index Creation: Batch store array list B into the spatial database; create a GIST spatial index for the "Point Type" spatial field in the spatial database that represents point location attributes;

[0042] Step P504, Border Buffer Retrieval and Latitude Sorting: Based on the retrieval function of the spatial database, perform spatial inclusion relationship calculation on the points stored in step P503 and the 5-kilometer buffer range of China's western border; extract the retrieved target points and sort them in order of latitude from low to high.

[0043] The addressing step in step P6 of the above-mentioned method for extracting landmark isotherms and generating semantic text includes the following steps:

[0044] Step P601, Determining the starting point and initial addressing direction: Take the points sorted in step P504 as the starting point, define the rightward movement from each starting point as the starting direction of the landmark isotherm addressing, and define the first point to the right of the movement direction as the starting point of the eight-neighbor addressing.

[0045] Step P602: Calculate the direction of travel for any point other than the starting point of isotherm addressing. The direction of travel is determined by the azimuth angle between the latitude and longitude of the current point p2 and the previous point p1.

[0046] Step P603: Following the direction of travel calculated in step P602, starting from the first neighboring point to the right of the current point p2, sequentially judge the eight neighboring points around the current point counterclockwise according to the adjoint matrix B3. The first valid point O found is the next target point for isotherm addressing, and the valid point O is marked as p2, while the original p2 is updated to p1. Record the valid points found in turn and store them in the array list S to complete the point cloud data retention for single-step addressing.

[0047] Step P604 involves iteratively executing steps P602 and P603, continuously updating the direction of travel, retrieving valid points in the neighborhood, updating point coordinate labels, and storing point cloud data to achieve automated extension addressing of the isotherm until the complete trajectory retrieval and point cloud set construction of the entire isotherm are completed.

[0048] In the above method for extracting landmark isotherms and generating semantic text, in step P602, assuming the latitude and longitude coordinates of the current point p2 are p2(lon2, lat2) and the latitude and longitude coordinates of the previous point p1 are p1(lon1, lat1), the method for calculating the direction of travel is as follows:

[0049] Step P6021: Calculate the latitude and longitude differences between the current point p2 and the previous point p1; the calculation method is as follows:

[0050] Latitude difference: A positive latitude difference indicates that the current point p2 is north of the previous point p1, while a negative latitude difference indicates that the current point p2 is south of the previous point p1; Longitude difference: A positive longitude difference indicates that the current point p2 is east of the previous point p1, and a negative longitude difference indicates that the current point p2 is west of the previous point p1. If Δlat=0 ​​and Δlon=0, then the current point p2 and the previous point p1 are directly determined to be "the same position", and the algorithm is terminated.

[0051] Step P6022: Calculate the initial azimuth angle; the azimuth angle is the angle from point p1 due north, rotated clockwise to the line connecting p1p2. Using Δlat and Δlon as the opposite side (vertical axis) and adjacent side (horizontal axis) of a right triangle, calculate the initial azimuth angle θdeg0:

[0052] , (Note: arctan2(y,x) prioritizes quadrant issues, where y=Δlon and x=Δlat to ensure the angles correspond to clockwise directions.)

[0053] Step P6023, Azimuth Correction; For initial azimuth angles less than 0, correction is performed, and the corrected azimuth angle is... Ensure that the azimuth angle range is 0° to 360°.

[0054] Step P6024: Based on the interval of the corrected azimuth angle θdeg, determine the direction of the current point p2 relative to the previous point p1, i.e., the direction of travel. The rules for determining the direction of travel are as follows (in Table 1, "≈0" means "approximately 0", that is, if the degree of Δlon or Δlat is within the range of 0°±0.1°, then the value is considered "≈0"):

[0055]

[0056] In the aforementioned method for extracting landmark isotherms and generating semantic text, in step P7, if none of the eight neighboring points around the current point in step P603 are valid points, then a neighborhood search is performed using a spatial database to connect the isotherms. During the connection process, distance and downward angle need to be considered. The connection method includes the following steps:

[0057] Step P701: Let the current point be p0(lon0,lat0). Use the spatial database to retrieve the neighboring points of the current point p0. Sort the retrieved neighboring points from nearest to farthest from the current point p0 to obtain an ordered neighboring point sequence (p1, p2, ..., pn). The latitude and longitude coordinates corresponding to each neighboring point in the ordered neighboring point sequence are p1(lon1,lat1), p2(lon2,lat2), ..., pn(lonn,latn). The formula for calculating the spatial distance of each neighboring point from the current point p0 is as follows:

[0058] (2);

[0059] (3);

[0060] ……………

[0061] (4);

[0062] In equations (2) to (4), p0p1 represents the spatial distance from the nearest point pn to the current point p0, p0p2 represents the spatial distance from point p0 to point p2, and p0pn represents the spatial distance from point p0 to point pn. and These are the latitude and longitude coordinates of the nearest point pn. and These are the latitude and longitude coordinates of the current point p0;

[0063] Step P702: Select a referencing point from the ordered nearest neighbor sequence using longitude, latitude, and spatial distance, and store the found referencing points in array list S. The referencing point determination principle is: select the point in the ordered nearest neighbor sequence whose longitude increases and whose latitude decreases compared to the current point p0, and which is closest to the current point p0, as the referencing point. That is, the referencing point has a longitude that increases compared to the current point p0 (addressing from west to east) and a latitude that decreases (addressing from south to south), and is as close to the current point p0 as possible. If there is no point that meets the referencing point determination principle, then select the point in the ordered nearest neighbor sequence whose longitude increases and which is closest to the current point p0 as the referencing point.

[0064] Step P703: Repeat steps P602 to P702 to achieve automatic extension addressing of isotherms; when steps P602 to P702 cannot find the next point, or when a 5-kilometer buffer zone of the eastern border of China is encountered, the addressing of the first isotherm ends.

[0065] Step P704: For the target points retrieved in step P504, repeat steps P6 and P7 in sequence until all isotherm addressing is completed.

[0066] In the above-mentioned method for extracting landmark isotherms and generating semantic texts, in step P702, the specific method for selecting and determining the connection point is as follows:

[0067] Sequentially determine whether each point in the ordered adjacent point sequence satisfies the connection point determination principle in sequence, that is: if lon1>lon0 and lat1<lat0, then the adjacent point p1 is directly selected as the connection point; otherwise, continue to judge the next adjacent point according to the sorting; if lonn>lon0, latn<lat0, and , then the adjacent point pn is selected; if none of the adjacent points in the ordered adjacent point sequence satisfy the connection point determination principle, then select the adjacent point with a longitude greater than lon0 and the closest distance to the current point p0 as the connection point; otherwise, the addressing of the entire isotherm ends; the specific judgment process is as follows:

[0068] (1) If lon1>lon0 and lat1<lat0, then the adjacent point p1 is directly selected, that is, the point with a larger longitude, a smaller latitude and the closest distance relative to the current point is selected; otherwise, make the following judgment;

[0069] (2) If lon2>lon0, lat2<lat0, and , then the adjacent point p2 is selected; otherwise, make the following judgment; here 0.5 is an empirical value;

[0070] (3) If lon3>lon0, lat3<lat0, and , then the adjacent point p3 is selected; otherwise, make the following judgment;

[0071] (4) If lon4>lon0, lat4<lat0, and , then the adjacent point p4 is selected; otherwise, make the following judgment;

[0072] …………

[0073] (5) If lonn>lon0, latn<lat0, and , then the adjacent point pn is selected; otherwise, make the following judgment;

[0074] (6) Select the point with a longitude greater than lon0 and the closest distance to the current point p0; otherwise, the addressing of this isotherm ends.

[0075] The above-mentioned method for extracting landmark isotherms and generating semantic texts, step P9 includes the following steps:

[0076] Step P901: Overlay and calculate the extracted landmark isotherm with meteorological and geographical spatial data through the spatial database, and determine the provinces covered by the landmark isotherm line segment through the topological relationship;

[0077] Step P902: Arrange the found provinces in spatial order from west to east, and combine them with the time information of meteorological grid data to form semantic text.

[0078] The technical solution of the present invention achieves the following beneficial technical effects:

[0079] 1. This invention provides a method for extracting and generating semantic text for landmark isotherms, enabling high-precision extraction and standardized drawing of these isotherms. Addressing issues such as segmented convergence, uneven line segments, and confusion in traditional isotherm extraction, this invention employs an eight-neighbor self-driven addressing and spatial connection method. By traversing grid data along eight neighborhoods and automatically addressing the found grid points, the isotherm direction is determined by comparing temperature values ​​of adjacent grid points. Then, segments are connected along the latitude direction, effectively solving the fragmentation problem in isotherm extraction for split-type cold regions, east-west continuous cold regions, and nested complex cold regions. The landmark isotherms extracted using this invention exhibit strong continuity and accurate direction, precisely locating landmark isotherms within China. This provides forecasters with clear and standardized visualization support for map analysis, significantly improving the practicality of isotherm data.

[0080] 2. The method for extracting landmark isotherms and generating semantic text in this invention enables automated and highly accurate generation of semantic information for landmark isotherms. This invention breaks through the inefficient traditional model that relies on manual map reading and writing of semantic descriptions. By integrating GIS (Geographic Information System) and natural language processing technology, it associates the extracted isotherm location information with geographic data such as administrative divisions and latitude / longitude coordinates. Aligning with the expression standards of the Central Meteorological Observatory's weather bulletins, it achieves automatic generation of semantic text through template matching. This not only significantly reduces the workload of consultation and support personnel and improves work efficiency, but also avoids errors that may occur with manual descriptions, ensuring the accuracy and consistency of semantic information and providing core data support for the automated generation of morning consultation PPTs.

[0081] 3. The method for extracting landmark isotherms and generating semantic text in this invention possesses excellent scalability and scenario adaptability: Taking the 0℃ isotherm as a starting point, this invention achieves rapid expansion support for multiple landmark isotherms such as -5℃, 5℃, and 10℃ through modular design; simultaneously, after performance testing, it can be smoothly migrated to 6.25km, 5km, and 1km meteorological grid data services, adapting to data processing needs with higher precision. This scalability enables the method for extracting landmark isotherms and generating semantic text in this invention to cover a wider range of meteorological business scenarios, providing comprehensive support for meteorological analysis and consultation decision-making at different temperature levels, significantly enhancing the application value and reusability of the technology.

[0082] 4. The method for extracting landmark isotherms and generating semantic text in this invention can improve the automation and intelligence level of meteorological data processing: This invention automatically accesses grid data from a meteorological big data cloud platform, combines data cleaning processes to ensure data quality, and then uses algorithms to automate the entire process of isotherm extraction, drawing, and semantic text generation, deeply integrating artificial intelligence, big data technology, and meteorological operations. This not only reduces manual intervention and lowers the risk of human operation, but also improves the efficiency and intelligence level of meteorological data processing, providing key technical support for the automation of meteorological consultation and forecasting decisions, and promoting the digital transformation of meteorological operations. Attached Figure Description

[0083] Figure 1 A schematic diagram of the process for extracting iconic isotherms and generating semantic text in an embodiment of the present invention;

[0084] Figure 2 A schematic diagram of the local distribution of effective points from -0.05℃ to 0.05℃ in an embodiment of the present invention;

[0085] Figure 3 A schematic diagram of the interpolation operator structure in an embodiment of the present invention;

[0086] Figure 4 Gradient effect diagrams of 3rd, 6th, and 10th interpolation in embodiments of the present invention;

[0087] Figure 5 Images showing the smoothing effects of 3, 10, and 30 gradient transitions in this invention embodiment;

[0088] Figure 6 Schematic diagram of internal points in an embodiment of the present invention;

[0089] Figure 7 Eight-neighbor addressing schematic diagram in an embodiment of the present invention;

[0090] Figure 8 A schematic diagram of automatic extension of isotherm addressing in an embodiment of the present invention;

[0091] Figure 9 A schematic diagram of the key isotherms extracted in this embodiment of the invention;

[0092] Figure 10 A schematic diagram comparing the accuracy in embodiments of the present invention. Detailed Implementation

[0093] 1. Materials and Methods

[0094] 1.1 Research Data

[0095] The core meteorological grid data used in this embodiment comes from the 1-kilometer resolution Asian regional temperature condition analysis product released by the National Meteorological Information Center. This product is built on a multi-source observation data fusion and advanced numerical processing technology system, and has extremely high authority and reliability. Its data fusion system comprehensively integrates near-surface measured data from automatic ground meteorological observation stations, atmospheric vertical structure detection data from weather radar, and large-scale remote sensing observation data from the Fengyun series meteorological satellites, among other diverse data sources.

[0096] This embodiment uses the 1-kilometer resolution Asian regional temperature real-time analysis product released by the National Meteorological Information Center as meteorological grid data to illustrate the method of extracting the landmark isotherms and generating semantic text of the present invention. However, those skilled in the art can use meteorological grid data from other sources to implement the method.

[0097] 1.2 Research Methods

[0098] Taking the extraction of the 0℃ isotherm as an example, meteorological grid data with a spatial resolution of 1 km at 6:00 AM on October 29, 2025, was processed. After grid data filtering, interpolation, smoothing, and neighborhood addressing, the southernmost 0℃ vector isotherm in China was obtained. This isotherm was then overlaid with meteorological geospatial zoning to generate isotherm-related semantic textual information. The file name of this data is: ARTInterim_HRCLDAS_RT_CHN_0P01_HOR-MNT_24h-2025102906.GRB2.

[0099] like Figure 1 As shown, the method for extracting landmark isotherms and generating semantic text in this embodiment specifically includes the following steps:

[0100] Step P1, Coarse Screening: Read the meteorological grid data and screen to identify valid points, isolated points, and invalid points; specifically including the following steps:

[0101] Step P101: Analyze the meteorological grid data and store the temperature values ​​in a two-dimensional array A, called the original data matrix A. Define a two-dimensional array B with the same dimensions as array A, denoted as the adjoint matrix B. Simultaneously acquire the core parameters of the meteorological grid data: grid spatial resolution of 0.01°, data time of 2025102906, observation element code of Minimum_temperature_height_above_ground, data starting longitude of 70°, data starting latitude of 0°; the observation element grid has 6001 rows in the latitude direction and 7001 columns in the longitude direction.

[0102] Step P102: Mark valid points. Taking the extraction of the 0℃ landmark isotherm as an example, the range of the landmark isotherm is defined as -0.05℃ to 0.05℃ (hereinafter referred to as the target value range).

[0103] Traverse the original data matrix A[i][j] (i∈[1,6000], j∈[1,7000]). If the temperature value at any point M is within the target value domain, and at least one of the temperature values ​​in any eight neighboring locations of point M is within the target value domain, then that point is identified as a valid point, and the corresponding position of point M in the adjoint matrix B is marked as t (i.e., true). The distribution of locally valid points identified in this embodiment is shown in [see figure]. Figure 2 ,from Figure 2 As can be seen, in most areas, the effective points form continuous stripes, but in some areas there are obvious breaks and sparse areas, so interpolation processing is required in the later stages.

[0104] If the temperature value at any point M is within the target range, and the temperature values ​​of the eight neighboring regions (east, west, south, north, southeast, northeast, southwest, and northwest) of any point M are all outside the target range, i.e., not within the range of -0.05℃ to 0.05℃, then point M is determined to be an isolated point, and the corresponding position of any point M in the adjoint matrix B is marked as f (i.e., false).

[0105] If the temperature value at any point M is not within the target range, or if the temperature value at any point M is NaN or 999999, then point M is determined to be an invalid point, and the corresponding position of point M in the adjoint matrix B is marked as f (i.e., false).

[0106] Step P2, Interpolation: Interpolation calculations are performed around the valid points using interpolation operators. The adjoint matrix B is traversed, and eight-neighbor interpolation calculations are performed around the valid points. The corresponding positions in the original data matrix A are then modified to reflect the target values. The purpose is to increase the density of valid points. The specific steps are as follows:

[0107] Step P201: Set the interpolation operator, defining the following rules: For any point to be interpolated, select the grid points in its eight neighboring areas; if at least two grid points within the eight neighboring areas have temperature values ​​in the range of -0.05℃ to 0.05℃, then the point to be interpolated is determined as a valid value and assigned the value t. For example... Figure 3 The diagram shown is a structural schematic of the interpolation operator, which clearly illustrates the one-to-one mapping relationship between row numbers, column numbers, and eight neighboring grid points.

[0108] Step P202: Define a temporary matrix C with the same dimensions as the adjoint matrix B. Traverse the adjoint matrix B[i][j] (i∈[1,6000], j∈[1,7000]). Perform interpolation calculations on each point in the adjoint matrix B using the interpolation operator in step P201. Write the interpolation calculation results into the corresponding positions in the temporary matrix C.

[0109] Step P203: After the interpolation calculation of all points in the adjoint matrix B is completed, assign the value of the temporary matrix C to the adjoint matrix B;

[0110] Step P204: Repeat steps P202 and P203 for 10 interpolation iterations to obtain the adjoint matrix B1; the gradation effect of 3, 6, and 10 interpolation iterations in this embodiment is shown in [the image / description]. Figure 4 ,from Figure 4 As can be seen, with the increase of the number of interpolation iterations, the boundary of the orange area gradually becomes continuous and smooth, changing from the initial fragmented and jagged shape.

[0111] Step P205: Traverse the adjoint matrix B1. If the position of the adjoint matrix B[i][j] is a valid value t, then modify the corresponding position of the interpolation point in the original data matrix A[i][j] to 0, that is, the temperature value of the 0℃ isotherm, and obtain the data matrix A1.

[0112] Step P3, Smoothing: Noise is removed using a smoothing operator. The original data matrix A is traversed, and the interpolated data matrix A1 is modified. Eight-neighbor smoothing is calculated for each point in data matrix A1. Comparing this to the smoothed data matrix, the points are re-marked in the adjoint matrix B1. The aim is to merge adjacent target regions as much as possible, while simultaneously erasing tiny gaps and noise in the target regions through smoothing. Specifically, the steps include:

[0113] Step P301: Set the smoothing operator. The calculation rule of the smoothing operator is: select the temperature value of any point to be processed and the eight neighboring points around the point, calculate the arithmetic mean of the temperature values ​​of the above nine points, and the arithmetic mean is the output value of the smoothing operator.

[0114] Step 302: Define a temporary matrix D with dimensions that are exactly the same as the size of the data matrix A1; iterate through all elements of the data matrix A1, perform smoothing calculations on the points corresponding to each element in the data matrix A1 according to the smoothing operator set in step P301, and write the calculation results into the corresponding positions of the temporary matrix D.

[0115] Step 303: After all the smoothing calculations for all points in data matrix A1 are completed, assign all element values ​​of temporary matrix D to data matrix A1 to complete the numerical update of data matrix A1.

[0116] Step 304: Repeat steps P302 and P303 for smoothing iteration 30 times to obtain data matrix A2; the gradual change effect of smoothing iteration 3, 10, and 30 times in this embodiment is shown in [the image / description]. Figure 5 ,from Figure 5 As can be seen, with the increase of the smoothing iteration process, the originally isolated and broken orange patches gradually connect, forming a more natural dendritic and strip-like distribution, and the coupling with the blue water system / terrain texture in the figure is higher.

[0117] Step P305: Traverse the data matrix A2. Based on the temperature values ​​of each point in the data matrix A2, re-mark the types of valid points, isolated points, and invalid points corresponding to each point in the adjoint matrix B1 to obtain the adjoint matrix B2.

[0118] Step P4, Fine Screening: Remove points inside the isothermal surface. For the adjoint matrix B2, remove points inside the target region to reduce the number of valid calculation points, thereby improving the efficiency and accuracy of subsequent isotherm extraction. The specific method is: traverse the adjoint matrix B2, and for any point within B2, determine whether all points within its eight neighborhoods are valid values. If this condition is met, the point is determined to be inside the isothermal surface region. See [link to relevant documentation]. Figure 6 In the adjoint matrix B2, the corresponding position of this point is marked as an invalid value and removed. This point will not be involved in the addressing calculation of the landmark isotherm in the future. The adjoint matrix after fine screening is denoted as adjoint matrix B3.

[0119] Step P5, Storage: Spatial retrieval will then be used to connect isotherms. For the adjoint matrix B3, the filtered valid points are stored in batches into the spatial database, and a GIST index is added to the point-type fields; specifically, the following steps are included:

[0120] Step P501, Data Storage Class Construction and Object Assignment: Construct a data storage class containing the following core variables: index number (ID), row number (i), column number (j), longitude (lon), latitude (lat), and point type (point); based on the adjoint matrix B3 filtered in step P4, assign the corresponding parameters of each point in adjoint matrix B3 to the objects generated by the data storage class, and store all objects sequentially in array list A; the method for calculating the index number ID of each point stored in the spatial database is as follows:

[0121] (1);

[0122] In equation (1), n ​​is the number of columns; assuming that point p1 in the adjoint matrix B3 is in the 10th row and 20th column, i.e., p1(10,20), then the ID number of p1 stored in the spatial database is:

[0123] ;

[0124] Step P502, Border Area Data Filtering: Traverse the array list A obtained in step P501, filter and remove objects in the list whose geographical location is not within the border area of ​​China, and obtain array list B;

[0125] Step P503, Spatial Database Storage and Index Creation: Batch store the filtered object array (i.e., array list B) from step P502 into the spatial database; create a GIST spatial index for the "point type" spatial field that represents point attributes in the spatial database to improve the efficiency of subsequent spatial retrieval.

[0126] Step P504, Border Buffer Retrieval and Latitude Sorting: Based on the retrieval function of the spatial database, perform spatial inclusion relationship calculation on the points stored in step P503 and the 5-kilometer buffer range of China's western border; extract the retrieved target points and sort them in order of latitude from low to high.

[0127] Step P6, Addressing: Find the next point by using the direction of travel and a counterclockwise eight-neighbor addressing method. In the adjoint matrix B3, calculate the direction of travel using the two points before and after, and find the next point within the eight-neighbor range of the direction of travel. Specifically, this includes the following steps:

[0128] Step P601, Determining the starting point and initial addressing direction: Take the points sorted by latitude and longitude in step P504 as the starting points for addressing; for each starting point p1(i,j), set the direction to its right as the starting direction for addressing the 0℃ landmark isotherm, and define the first point to the right of this starting direction, i.e., p2(i,j-1), as the starting point for eight-neighbor addressing;

[0129] Step P602, Definition of the Calculation Rule for Travel Direction: For any point to be addressed other than the isotherm addressing starting point, its travel direction is determined according to the following rule: Based on the latitude and longitude coordinates of the current point p2 (latitude and longitude: lon2, lat2) and the previous addressing point p1 (latitude and longitude: lon1, lat1), the travel direction of the point to be addressed is determined by calculating the azimuth angle between the two points. Assuming p1(10,20) and p2(15,15), the specific algorithm is as follows:

[0130] Latitude difference: Longitude difference: The radian value corresponding to θrad=arctan2(5,-5) is 2.356, which is equivalent to 135° in azimuth. 135° falls within the range of 112.5°-157.5°, and is therefore determined to be "Southeast (SE)", which is consistent with the simplified logic (Δlat<0, Δlon>0).

[0131] Step P603: Based on the travel direction solved in step P602, take the first neighboring point to the right of the current point p2 (lon2, lat2) in the travel direction as the starting point of the eight-neighborhood search. Relying on the adjoint matrix B3, the validity of each of the eight neighboring points around the current point is determined in counterclockwise order. The first valid point retrieved is determined as the next target point for isotherm addressing, and the label is updated - the valid point is reassigned to p2, and the original p2 is updated to p1. At the same time, the valid points obtained in this addressing are recorded and stored in the array list S in sequence, completing the point cloud data retention of single-step addressing.

[0132] Assuming the direction of travel calculated in step P602 is southeast, the first addressing point on the right side of the eight-neighborhood addressing based on the current point p2(i,j) in the direction of travel should be... That is, the eight-neighbor addressing of the current point p2 starts from the southwest point, see Figure 7 Assuming p2(15,15), the southwest point should be (14,14).

[0133] Step P604 involves iteratively executing steps P602 and P603. Through continuous updates to the travel direction, retrieval of valid neighboring points, updating point coordinate labels, and storing point cloud data, automated extension addressing of the isotherm is achieved until the complete trajectory retrieval and point cloud set construction of the entire isotherm are completed. (See...) Figure 8 .

[0134] Step P7, Targeting: If none of the eight neighboring points of the current point in step P603 are valid points, i.e., the eight-neighbor addressing did not find the target value, then the spatial database ( Figure 1 The system (hereinafter referred to as "the space library") performs neighborhood searches to achieve isotherm continuation. During the continuation process, distance and downward angle need to be considered. Isotherm addressing stops at a 5-kilometer buffer zone along the eastern border of China. The specific steps include:

[0135] Step P701, Neighbor Point Retrieval and Sorting: Set the current address position as p0, retrieve the neighboring positions of the current position p0 through the spatial database; sort all the retrieved neighboring positions from near to far according to their spatial distance from the current position p0, and obtain an ordered neighboring point sequence (p1, p2, p3, p4, p5).

[0136] The latitude and longitude coordinates of each point in this embodiment are as follows: The coordinates of the current point p0 are (111.05, 36.21); after sorting by distance, the coordinates of each neighboring point in the ordered neighboring point sequence are p1 (111.04, 36.21), p2 (111.05, 36.22), p3 (111.06, 36.21), p4 (111.06, 36.22), and p5 (111.03, 36.21).

[0137] Step P702,接引点判定:

[0138] Based on the current point p0(111.05, 36.21) and the ordered adjacent point sequence (p1(111.04, 36.21), p2(111.05, 36.22), p3(111.06, 36.21), p4(111.06, 36.22), p5(111.03, 36.21)) determined in step P701, determine the isotherm addressing接引点 according to the core principle of "longitude increasing, latitude decreasing, and distance being close", and the specific process is as follows:

[0139] ;

[0140] ;

[0141] ;

[0142] ;

[0143] ;

[0144] Among them, p0p1 represents the distance between adjacent point p1 and the current point p0, p0p2 represents the distance between adjacent point p2 and the current point p0, and similarly p0p5 represents the distance between adjacent point p5 and the current point p0; the接引点 is judged successively as follows:

[0145] (1) Judge adjacent point p1: The longitude lon1 of adjacent point p1 = 111.04 < lon0 = 111.05 (does not meet lon1 > lon0), and the latitude lat1 = 36.21 = lat0 = 36.21 (does not meet lat1 < lat0), so adjacent point p1 is not directly selected and enters the judgment of the next point;

[0146] (2) Judge adjacent point p2: The longitude lon2 of adjacent point p2 = 111.05 = lon0 = 111.05 (does not meet lon2 > lon0), and the latitude lat2 = 36.22 > lat0 = 36.21 (does not meet lat2 < lat0), so adjacent point p2 is not selected and enters the judgment of the next point;

[0147] (3) Judge adjacent point p3: The longitude lon3 of adjacent point p3 = 111.06 > lon0 = 111.05 (meets lon3 > lon0), but the latitude lat3 = 36.21 = lat0 = 36.21 (does not meet lat3 < lat0), calculate the proportion of the distance difference , still because the latitude condition is not met, enter the judgment of the next point;

[0148] (4)Determine the adjacent point p4: The longitude of the adjacent point p4, lon4 = 111.06 > lon0 = 111.05 (satisfying lon4 > lon0), but the latitude lat4 = 36.22 > lat0 = 36.21 (not satisfying lat4 < lat0). Therefore, the adjacent point p4 is not selected and proceed to the determination of the next point;

[0149] (5)Determine the adjacent point p5: The longitude of the adjacent point p5, lon5 = 111.03 < lon0 = 111.05 (not satisfying lon5 > lon0), and the latitude lat5 = 36.21 = lat0 = 36.21 (not satisfying lat5 < lat0). Therefore, the adjacent point p5 is not selected;

[0150] (6)Final fallback determination: Check one by one the longitudes. Only the longitudes of the adjacent point p3 and the adjacent point p4 are greater than the longitude of the current point p0. Among them, the adjacent point p3 is the closest to the current point p0 (p0p3 = 0.01), and lon3 > lon0. Therefore, determine that the adjacent point p3 is the接引点 (reception point);

[0151] Store the determined reception point p3 into the array list S in step P603; Since the longitude of the reception point p3 is greater than the longitude of the current point p0, the addressing end condition is not triggered;

[0152] Step P703, Repeat steps P602 to P702 to achieve automatic extension addressing of the isotherm; When no next point can be found in steps P602 to P702, or when encountering the 5 - kilometer buffer zone of the eastern national border of China, the addressing of the first isotherm ends;

[0153] Step P704, For the points retrieved in step P504, repeat steps P6 and P7 in sequence until the addressing of all isotherms ends.

[0154] Step P8, Identification: For the extracted isotherms, in the order of generation, the first one that traverses the 5 - kilometer buffer zone of the east - west national border of China is the标志性等温线 (landmark isotherm), see Figure 9 ; The specific identification method is:

[0155] After completing the extraction and sequence generation of all isotherms based on the foregoing steps, conduct the landmark isotherm identification operation on all extraction results in the order of generation of the isotherms from low latitude to high latitude;

[0156] In this embodiment, through the processes of grid data processing, address point screening, connection point determination, and isotherm tracing in steps P1 to P7, a total of 5 isotherms were extracted and labeled as L1, L2, L3, L4, and L5 in the order of generation. The spatial distribution range of each isotherm was verified one by one: the coverage areas of isotherms L1 to L4 did not completely penetrate the 5-kilometer buffer zone of China's east and west borders. Only the spatial direction of isotherm L5 started from the 5-kilometer buffer zone of China's west border, extended through multiple inland areas, and finally reached the 5-kilometer buffer zone of China's east border, thus achieving traverse coverage of the 5-kilometer buffer zone of China's east and west borders.

[0157] Based on this, it is determined that the first isotherm L5, which runs through a 5-kilometer buffer zone across the east-west border of China and is sorted in the order of generation, is the landmark isotherm extracted in this study.

[0158] Step P9, Semantic Generation: Overlay the landmark isotherm with meteorological geospatial data to identify the provinces covered by the line segment, ultimately generating semantic information. This includes the following steps:

[0159] Step P901: Spatial overlay and province matching. The overlay analysis function of the spatial database is invoked to perform spatial overlay calculations on the vector data of the landmark isotherm L5 and the meteorological geospatial data of the provincial-level administrative regions across the country. The provincial-level administrative regions covered by the isotherm segment are determined by topological relationships. The provinces covered by the isotherm segment are Tibet Autonomous Region, Sichuan Province, Gansu Province, Shanxi Province, and Liaoning Province, in that order.

[0160] Step P902: Semantic text organization and generation. According to the spatial order from west to east, the provinces obtained by the above matching are sorted and combined with the time information of the meteorological grid data (2025102906) to form standardized semantic text: From 6:00 on October 28, 2025 to 6:00 on October 29, the 0℃ minimum temperature isotherm in the past 24 hours is located in southeastern Tibet, western and central Sichuan, southeastern and central Gansu, central and southeastern Shanxi, and southwestern Liaoning.

[0161] 2. Results and Analysis

[0162] 2.1 Experimental Design

[0163] To verify the effectiveness, accuracy, and efficiency of the landmark isotherm extraction and semantic text generation method in this embodiment, a control group and an experimental group were set up for comparative testing. The specific experimental design is as follows:

[0164] (1) Group setup: The control group used the traditional manual interpretation service, while the experimental group used the extraction and semantic generation service based on innovative meteorological grid data addressing in this embodiment. The control group consisted of 4 senior meteorological forecasters with more than 5 years of experience in morning consultation duty, who completed the task by manually interpreting, plotting, and writing semantic descriptions; the experimental group used a series of technologies proposed in this embodiment, such as coarse screening, interpolation, smoothing, fine screening, and neighborhood addressing, to achieve fully automated processing.

[0165] (2) Test environment: CPU is Intel Core i7-12700H, memory is 32GB, hard disk is 1TB SSD, operating system is Windows 10 Professional Edition.

[0166] (3) Test samples: 1 km real-time temperature grid data of China during the key period of autumn-winter-spring transition (October to April of the following year) were selected, with a total of 10 independent samples covering the distribution scenarios of isotherms under different weather conditions. The test was carried out focusing on four types of landmark isotherms: 0℃, -5℃, 5℃ and 10℃.

[0167] (4) Evaluation indicators and subjects: The core evaluation indicators include the accuracy of isotherm extraction, the completeness of plotting, the accuracy of semantic generation, the processing time of a single sample, and the integration adaptability; an evaluation group composed of 5 meteorological forecasters will make a comprehensive judgment on the test results of the two groups.

[0168] 2.2 Comparative Test Results

[0169] The tests revealed significant differences between the control group and the experimental group in all core evaluation indicators. The specific results are as follows:

[0170] (1) Processing time per sample: The average processing time for the control group to complete isotherm extraction, complete plotting, semantic text writing and PPT production was 20 minutes; the average processing time for the experimental group to complete the same operation was less than 2 minutes, and the processing efficiency was improved by more than 10 times.

[0171] (2) Integration and adaptability: The control group adopted a manual operation mode, which could not adapt to the automatic generation process of morning meeting PPT and required manual completion of PPT production and integration; the experimental group was fully automated and could be directly and seamlessly connected with the automatic generation process of morning meeting PPT, with excellent integration and adaptability.

[0172] (3) Core quality indicators: The experimental group was significantly better than the control group in terms of isotherm extraction accuracy, plotting completeness (line segment uniformity, positional accuracy) and semantic generation accuracy. The comparison results are shown in the figure. Figure 10 ,from Figure 10As can be seen, the isotherm boundary refinement and complexity extracted by the method of the experimental group are significantly better than those of the control group. Furthermore, the experimental group can effectively avoid problems such as positional deviation, uneven distribution of line segments, and semantic description deviation in the traditional manual method.

[0173] 3. Conclusions and Discussion

[0174] The method for extracting landmark isotherms and generating semantic text in this embodiment is simple, efficient, and highly automated. Using this method, users do not need to perform complex manual interpretation and operations; they only need to import meteorological grid data for the corresponding time period to automatically and accurately extract, fully plot, and generate semantic text for landmark isotherms such as 0℃, -5℃, 5℃, and 10℃. This greatly reduces the operational threshold and learning cost for users of this type of meteorological service, and is particularly suitable for business scenarios with stringent timeliness requirements, such as morning meteorological consultations.

[0175] On the technical implementation side, this embodiment constructs a fully automated processing chain of "coarse screening-interpolation-smoothing-fine screening-neighborhood addressing" to perform hierarchical and refined processing of meteorological grid data, effectively solving the technical defects of inaccurate isotherm extraction location and uneven line segment distribution in traditional manual methods. Simultaneously, by abandoning the traditional manual operation mode and adopting automated processing to replace manual interpretation, plotting, and semantic writing, it avoids the efficiency bottleneck caused by manual intervention in real-time business processing. Furthermore, it eliminates the need for additional manual PPT creation and integration, and can seamlessly integrate with the automatic PPT generation process for morning consultations, significantly improving the integration and adaptability of business processes. In addition, addressing the core requirements of isotherm extraction and semantic generation, this embodiment innovatively designs neighborhood addressing technology and semantic generation algorithms, which significantly reduce subjective errors and improve the consistency and reliability of results compared to traditional manual judgment and text writing. The landmark isotherm extraction and semantic text generation method proposed in this embodiment has been applied and tested in the meteorological morning consultation business system. Practice has proven that the test application effect is good and can stably support the isotherm-related service needs in daily consultations.

[0176] This invention is applicable to operational scenarios involving isotherm extraction, plotting, and semantic description generation from meteorological grid data. This includes automated processing of characteristic isotherms under different weather conditions and rapid generation of meteorological consultation materials. It is particularly suitable for real-time meteorological operational scenarios with high requirements for processing timeliness and result accuracy. This invention is not applicable to isotherm extraction needs from non-meteorological grid data sources, or to semantic generation scenarios involving ultra-large-scale, irregular, discrete temperature data.

[0177] It should be noted that the shortcomings of existing traditional manual interpretation services mentioned in this invention, such as inaccurate isotherm extraction locations, uneven line segments, low processing efficiency, and poor integration adaptability, are all results derived by the inventors after extensive meteorological operational practice and careful research. Therefore, the discovery process of the aforementioned technical defects, as well as the solutions proposed in this invention, such as the "fully automated processing link" and "neighborhood addressing technology," are all important contributions made by the inventors during the development of this invention.

Claims

1. A method for extracting iconic isotherms and generating semantic text, characterized in that, Includes the following steps: Step P1, coarse screening: Store the temperature values ​​obtained from parsing the meteorological grid data into the original data matrix A; screen and identify valid points, isolated points and invalid points from the original data matrix A and mark their corresponding positions in the adjoint matrix B; the dimensions of the adjoint matrix B are completely consistent with those of the original data matrix A; Step P2, Interpolation: Perform eight-neighbor interpolation around the valid points of the adjoint matrix B to obtain the adjoint matrix B1; traverse the adjoint matrix B1 and modify the temperature values ​​of the valid points at the corresponding positions in the original data matrix A to the temperature values ​​of the landmark isotherms to obtain the data matrix A1. Step P3, Smoothing: Perform eight-neighbor smoothing calculation around the valid points in data matrix A1 to obtain data matrix A2; traverse data matrix A2 and modify the label values ​​of the corresponding positions of valid points in adjoint matrix B1 to obtain adjoint matrix B2. Step P4, fine screening: Traverse the adjoint matrix B2. If all points within the eight-neighborhood of any point M in the adjoint matrix B2 are valid points, then any point M is determined to be a point inside the isothermal surface region. The points inside the isothermal surface region are marked as invalid points and removed to obtain the adjoint matrix B3. Step P5, Storage: Store the valid points filtered out from the adjoint matrix B3 into the spatial database and sort them by latitude. Create a GIST index on the point type field. Step P6, Addressing: In the adjoint matrix B3, the starting point is found by searching the spatial database, and the next valid point is found according to the direction of movement and the counterclockwise eight-neighbor addressing method to realize the isotherm connection. Step P7, Connection: If the target value is not found during addressing, the isotherm is connected by performing a neighborhood search in the spatial database. Step P8, Identification: Sort the extracted isotherms according to the generation order from low latitude to high latitude, and find the first isotherm that runs through a 5-kilometer buffer zone across the east-west border of China, which is the landmark isotherm. Step P9, Semantic Generation: The extracted landmark isotherms are overlaid with geospatial data to generate semantic information.

2. The method for extracting landmark isotherms and generating semantic text according to claim 1, characterized in that, Step P1 includes the following steps: Step P101: Parse the meteorological grid data to obtain temperature values, and store the temperature values ​​in a two-dimensional array A as the original data matrix A; define a two-dimensional array B with the same dimensions as the two-dimensional array A as the adjoint matrix B; synchronously acquire the grid spatial resolution, data time, observation element code, data starting longitude, data starting latitude, number of observation element grid rows, and number of observation element grid columns of the meteorological grid data. Step P102: Define the interval values ​​of the landmark isotherm as the target range; the range of the target range is the target value ± 0.05℃; Traverse the original data matrix A. If the temperature value at any point M in the original data matrix A is within the target value range, and at least one of the temperature values ​​at any eight neighboring locations of point M is within the target value range, then mark the corresponding position of point M in the adjoint matrix B as a valid point and assign a valid value. If the temperature value at any point M in the original data matrix A is within the target value range, and none of the temperature values ​​at any eight neighboring locations of point M are within the target value range, then mark the corresponding position of point M in the adjoint matrix B as an isolated point and assign an invalid value. If the temperature value at any point M in the original data matrix A is not within the target value range, or the temperature value at any point M in the original data matrix A is NaN or 999999, then mark the corresponding position in the adjoint matrix B as an invalid point and assign an invalid value. The eight neighborhoods are the grid points in the eight directions of east, west, south, north, southeast, northeast, southwest, and northwest of any given point.

3. The method for extracting landmark isotherms and generating semantic text according to claim 2, characterized in that, Step P2 includes the following steps: Step P201: Set the interpolation operator: Select any grid points in the eight neighborhoods around the point to be interpolated. If at least two grid points in the eight neighborhoods have temperature values ​​within the target value range, then modify the point to be interpolated as a valid point and assign a valid value. Step P202: Define a temporary matrix C with the same dimensions as the adjoint matrix B. Traverse the adjoint matrix B, use the interpolation operator to perform interpolation calculations on each point to be interpolated in the adjoint matrix B, and write the interpolation calculation results into the corresponding positions in the temporary matrix C. Step P203: After the interpolation calculation of all interpolation points in the adjoint matrix B is completed, assign the value of the temporary matrix C to the adjoint matrix B. Step P204: Repeat steps P202 and P203 for interpolation iteration 8-12 times to obtain the adjoint matrix B1; Step P205: Traverse the adjoint matrix B1. If the position of the interpolation point in the adjoint matrix B1 is a valid value, then modify the corresponding position of the interpolation point in the original matrix A to the temperature value of the landmark isotherm, and obtain the data matrix A1.

4. The method for extracting landmark isotherms and generating semantic text according to claim 3, characterized in that, Step P3 includes the following steps: Step P301: Define the smoothing operator: Define the average temperature value of any point and the eight neighboring grid points around the point as the smoothing operator; Step P302: Define a temporary matrix D with dimensions that are exactly the same as those of the data matrix A1. Traverse the data matrix A1, use the smoothing operator to perform smoothing calculations on each point in the data matrix A1, and write the smoothing calculation results into the corresponding positions of the temporary matrix D. Step P303: After the smoothing calculation of all points in the data matrix A1 is completed, assign the value of the temporary matrix D to the data matrix A1; Step P304: Repeat steps P302 and P303 for smoothing iteration 25-35 times to obtain data matrix A2; Step P305: Traverse the data matrix A2. According to the temperature value of each point in the data matrix A2, re-mark the valid points, isolated points and invalid points in the adjoint matrix B1 according to the method in step P102 to obtain the adjoint matrix B2.

5. The method for extracting landmark isotherms and generating semantic text according to claim 4, characterized in that, Step P5 includes the following steps: Step P501, Data Storage Class Construction and Object Assignment: The variables of the data storage class include index number, row number, column number, longitude, latitude, and point type; assign the parameters corresponding to each point in the accompanying matrix B3 to the objects generated by the data class storage, and store the objects sequentially in the array list A of the spatial database; the index number of the point in the accompanying matrix B3 in the spatial database is: ID = i*n + j (1); In equation (1), ID is the index number, i is the row number, n is the column number, and j is the column number; Step P502, Border Area Data Filtering: Traverse array list A, filter and remove objects whose geographical location is not within the border area of ​​China, and obtain array list B; Step P503, Spatial Database Storage and Index Creation: Batch store array list B to the spatial database; Create a GIST spatial index on the "point type" spatial field that represents point location attributes in the spatial database; Step P504, Border Buffer Retrieval and Latitude Sorting: Based on the retrieval function of the spatial database, perform spatial inclusion relationship calculation on the points stored in step P503 and the 5-kilometer buffer range of China's western border; extract the retrieved target points and sort them in order of latitude from low to high.

6. The method for extracting iconic isotherms and generating semantic text according to claim 5, characterized in that, The addressing in step P6 includes the following steps: Step P601, Determining the starting point and initial addressing direction: Take the points sorted in step P504 as the starting point, define the rightward movement from each starting point as the starting direction of the landmark isotherm addressing, and define the first point to the right of the movement direction as the starting point of the eight-neighbor addressing. Step P602: Calculate the direction of travel for any point other than the starting point of isotherm addressing. The direction of travel is determined by the azimuth angle between the latitude and longitude of the current point p2 and the previous point p1. Step P603: Following the direction of travel calculated in step P602, starting from the first neighboring point to the right of the current point p2, sequentially judge the eight neighboring points around the current point counterclockwise according to the adjoint matrix B3. The first valid point O found is the next target point for isotherm addressing, and the valid point O is marked as p2, while the original p2 is updated to p1. Record the valid points found in turn and store them in the array list S to complete the point cloud data retention for single-step addressing. Step P604 involves iteratively executing steps P602 and P603, continuously updating the direction of travel, retrieving valid points in the neighborhood, updating point coordinate labels, and storing point cloud data to achieve automated extension addressing of the isotherm until the complete trajectory retrieval and point cloud set construction of the entire isotherm are completed.

7. The method for extracting landmark isotherms and generating semantic text according to claim 6, characterized in that, In step P602, assuming the latitude and longitude coordinates of the current point p2 are p2(lon2, lat2) and the latitude and longitude coordinates of the previous point p1 are p1(lon1, lat1), the method for calculating the direction of travel is as follows: Step P6021: Calculate the latitude and longitude differences between the current point p2 and the previous point p1. The calculation method is as follows: Latitude difference: A positive latitude difference indicates that the current point p2 is north of the previous point p1, while a negative latitude difference indicates that the current point p2 is south of the previous point p1; Longitude difference: A positive longitude difference indicates that the current point p2 is east of the previous point p1, and a negative longitude difference indicates that the current point p2 is west of the previous point p1. If Δlat=0 ​​and Δlon=0, then the current point p2 and the previous point p1 are directly determined to be "the same position", and the algorithm is terminated. Step P6022: Calculate the initial azimuth angle; the azimuth angle is the angle from point p1 due north, rotated clockwise to the line connecting p1p2. Using Δlat and Δlon as the vertical and horizontal axes of the right triangle, calculate the initial azimuth angle θdeg0: θdeg0=θrad×180 / π, θrad=arctan2(Δlon,Δlat); Step P6023, Azimuth correction; For initial azimuth angles less than 0, correction is performed, and the corrected azimuth angle θdeg = θdeg0 + 360°; Step P6024: Based on the interval of the corrected azimuth angle θdeg, determine the direction of the current point p2 relative to the previous point p1, i.e., the direction of travel. The rules for determining the direction of travel are as follows: (1) When the azimuth angle θdeg is in the range of 337.5°~360° or 0°~22.5°, the direction of the current point p2 relative to the previous point p1 is north; at this time, Δlat>0 and Δlon≈0; (2) When the azimuth angle θdeg is in the range of 22.5° to 67.5°, the direction of the current point p2 relative to the previous point p1 is northeast; at this time, Δlat>0 and Δlon>0; (3) When the azimuth angle θdeg is in the range of 67.5° to 112.5°, the direction of the current point p2 relative to the previous point p1 is east; at this time, Δlon>0 and Δlat≈0; (4) When the azimuth angle θdeg is in the range of 112.5° to 157.5°, the direction of the current point p2 relative to the previous point p1 is southeast; at this time, Δlat<0 and Δlon>0; (5) When the azimuth angle θdeg is in the range of 157.5° to 202.5°, the direction of the current point p2 relative to the previous point p1 is south; at this time, Δlat<0 and Δlon≈0; (6) When the azimuth angle θdeg is in the range of 202.5° to 247.5°, the direction of the current point p2 relative to the previous point p1 is southwest; at this time, Δlat<0 and Δlon<0; (7) When the azimuth angle θdeg is in the range of 247.5° to 292.5°, the direction of the current point p2 relative to the previous point p1 is west; at this time, Δlon<0 and Δlat≈0; (8) When the azimuth angle θdeg is in the range of 292.5° to 337.5°, the direction of the current point p2 relative to the previous point p1 is northwest; at this time, Δlat>0 and Δlon<0.

8. The method for extracting landmark isotherms and generating semantic text according to claim 7, characterized in that, In step P7, if none of the eight neighboring points around the current point in step P603 are valid points, then a neighborhood search is performed using the spatial database to connect the isotherms. The connection method includes the following steps: Step P701: Let the current point be p0(lon0,lat0). Use the spatial database to retrieve the neighboring points of the current point p0. Sort the retrieved neighboring points from nearest to farthest from the current point p0 to obtain an ordered neighboring point sequence (p1, p2, ..., pn). The latitude and longitude coordinates corresponding to each neighboring point in the ordered neighboring point sequence are p1(lon1,lat1), p2(lon2,lat2), ..., pn(lonn,latn). The formula for calculating the spatial distance of each neighboring point from the current point p0 is as follows: （4）； In equation (4), p0pn represents the spatial distance from the nearest point pn to the current point p0; and These are the latitude and longitude coordinates of the nearest point pn. and These are the latitude and longitude coordinates of the current point p0; Step P702: Select a connecting point from the ordered nearest neighbor sequence using longitude, latitude, and spatial distance, and store the found connecting points in array list S; The principle for determining the reference point is as follows: select the point in the ordered nearest neighbor sequence whose longitude increases and whose latitude decreases compared to the current point p0, and which is closest to the current point p0, as the reference point; if there is no point that meets the reference point determination principle, then select the point in the ordered nearest neighbor sequence whose longitude increases and which is closest to the current point p0 as the reference point. Step P703: Repeat steps P602 to P702 to achieve automatic extension addressing of isotherms; when steps P602 to P702 cannot find the next point, or when a 5-kilometer buffer zone of the eastern border of China is encountered, the addressing of the first isotherm ends. Step P704: For the target points retrieved in step P504, repeat steps P6 and P7 in sequence until all isotherm addressing is completed.

9. The method for extracting landmark isotherms and generating semantic text according to claim 8, characterized in that, In step P702, the specific method for selecting and determining the connection point is as follows: Judging in sequence according to the sequence order from the ordered adjacent point sequence whether it meets the access point determination principle, that is: if lon1>lon0 and lat1<lat0, then the adjacent point p1 is directly selected as the access point, otherwise continue to judge the next adjacent point according to the sorting; if lonn>lon0, latn<lat0, and , then the adjacent point pn is selected; if all adjacent points in the ordered adjacent point sequence do not meet the access point determination principle, then select the adjacent point with a longitude greater than lon0 and the closest distance to the current point p0 as the access point, otherwise the addressing of the entire isotherm ends.

10. The method for extracting landmark isotherms and generating semantic text according to any one of claims 1-9, characterized in that, Step P9 includes the following steps: Step P901: Overlay the extracted landmark isotherms with meteorological geospatial data using a spatial database, and determine the provinces covered by the landmark isotherm segments through topological relationships. Step P902: Arrange the found provinces in spatial order from west to east, and combine them with the time information of meteorological grid data to form semantic text.

Images