A method for diagnosing errors of heavy rain belt based on target morphological parameters

Abstract

The application discloses a strong precipitation rain belt error diagnosis method based on target morphological parameters and relates to the technical field of meteorological prediction. The method first acquires observed rain belt data and predicted rain belt data of a strong precipitation process, constructs a target morphological parameter system containing spatial topology, intensity gradient and dynamic evolution parameters, and extracts corresponding parameter sets from the two types of data; through an attention-weighted parameter matching algorithm, multi-dimensional error contribution degrees are calculated and dominant error types are located; based on the dominant error types and the contribution degrees, layered diagnosis results covering error source tracing, influence range quantification and mode optimization direction guidance are generated. The method breaks through the limitations of error diagnosis, realizes comprehensive characterization of rain belt morphological characteristics, accurately locates error sources, provides a scientific basis for numerical prediction model optimization, and effectively improves the prediction accuracy of strong precipitation rain belts.

Description

Technical Field

[0001] This invention relates to the field of meteorological forecasting technology, specifically to a method for diagnosing errors in heavy precipitation rainbands based on target morphological parameters. Background Technology

[0002] Accurate forecasting of heavy precipitation belts is a core component of meteorological disaster prevention and mitigation, directly impacting the effectiveness of controlling secondary risks such as floods and geological disasters. With the increasing frequency of extreme weather events, the movement paths, coverage, and intensity changes of heavy precipitation belts pose significant threats to urban operations, agricultural production, and the safety of people's lives and property. Therefore, improving the accuracy of heavy precipitation forecasts has become an urgent need in the meteorological field. Numerical forecasting models, as the core tool for heavy precipitation forecasting, are susceptible to factors such as initial field errors and physical process parameterization schemes, leading to discrepancies between forecasts and actual observations. Efficient error diagnosis techniques are urgently needed to support forecast optimization.

[0003] Current methods for diagnosing heavy precipitation rainbands primarily focus on point-to-point numerical deviation correction or overall error compensation based on statistical models, lacking targeted analysis of the rainband's structural characteristics. These traditional methods often overlook the morphological attributes of heavy precipitation rainbands as continuous weather systems, such as aspect ratio, axial angle, and area, making it difficult to accurately pinpoint core error sources like rainband morphological distortion and positional shifts. With the increasing sophistication of meteorological observation data, error diagnosis models relying solely on numerical statistics can no longer meet the demands of high-precision forecasting. There is an urgent need to construct a more targeted error diagnosis system from the perspective of rainband target morphology.

[0004] Chinese patent (CN111538935B) discloses a refined precipitation fusion method based on topographic features and multi-source model products. This technology integrates multiple model forecast data, calculates station weight coefficients by combining topographic features, and achieves gridded integration, solving the problem of large systematic biases in traditional single-model forecasts and improving the refinement of precipitation forecasts. However, this technology still has significant shortcomings. Its core focuses on the weighted fusion and post-correction of multi-model data, without constructing a dedicated error diagnosis mechanism. It cannot clearly define the mapping relationship between rainband morphology parameters and error sources, and the application of topographic factors is limited to weight allocation, without considering the errors caused by rainband morphology distortion under the influence of topography.

[0005] To address the aforementioned technical deficiencies, Chinese patent (CN119106949B) discloses an extended-range continuous precipitation distribution probability forecasting method that integrates fine topographic features. This technology extracts multi-dimensional topographic feature factors and constructs a hybrid deep learning model, which to some extent solves the problems of single application of topographic features and ambiguous error sources in the aforementioned patent. However, this technology still has limitations. Its core objective is to improve the performance of precipitation probability forecasts, but it does not design a dedicated error diagnosis system for the unique morphology of heavy precipitation rainbands, nor does it involve quantitative correlation analysis between rainband morphology parameters and forecast errors. This makes it difficult to achieve accurate error location and targeted correction, and it cannot meet the actual needs of fine-grained control of heavy precipitation rainband forecast errors.

[0006] Therefore, there is an urgent need for error diagnosis technology for the target morphology of heavy precipitation rain belts, so as to achieve accurate positioning and quantitative analysis of errors, provide core technical support for targeted optimization of heavy precipitation rain belt forecasts, and thus improve the accuracy and timeliness of meteorological disaster prevention and mitigation. Summary of the Invention

[0007] To address the aforementioned technical problems, this application discloses a method for diagnosing errors in heavy precipitation rainbands based on target morphological parameters, specifically including:

[0008] Acquire observed rainband data and forecasted rainband data corresponding to the heavy precipitation process. The observed rainband data includes multi-source observed fused precipitation data, and the forecasted rainband data includes precipitation forecast data output by numerical models.

[0009] Based on the multi-dimensional characteristics of rainband morphology, a target morphology parameter system is constructed. The target morphology parameter system includes spatial topology parameters, intensity gradient parameters, and dynamic evolution parameters. The spatial topology parameters are used to characterize the geometric structure correlation of the rainband, the intensity gradient parameters are used to quantify the spatial variation rate of precipitation intensity within the rainband, and the dynamic evolution parameters are used to characterize the morphological change features of the rainband in the time series.

[0010] Each parameter in the target morphological parameter system is extracted from the observed rainband data and the predicted rainband data respectively to obtain the observed morphological parameter set and the predicted morphological parameter set;

[0011] The multidimensional error contribution between the observed morphological parameter set and the predicted morphological parameter set is calculated using an attention-weighted parameter matching algorithm to identify the dominant error type.

[0012] Based on the dominant error type and multidimensional error contribution, a hierarchical diagnostic result for rainband forecast error is generated. The hierarchical diagnostic result includes error source tracing, error impact range quantification, and model optimization direction guidance.

[0013] Preferably, the spatial topological parameters include rainband topological connectivity, morphological fractal dimension, and centroid offset vector, and the rainband topological connectivity is calculated by the following formula: ,in, Rainband pixels With pixels The connectivity matrix is ​​denoted by 1 if the elements are connected, and 0 otherwise. For connectivity weights, This represents the total number of effective pixels in the rainband. This represents the maximum weight.

[0014] Preferably, the morphological fractal dimension is calculated using the box counting method, with the following formula: ,in, Let's say it's the side length of the box. The minimum number of boxes required to cover the entire rainband region was obtained by performing multi-scale box coverage statistics on the binary mask image of the rainband.

[0015] Preferably, the intensity gradient parameters include radial gradient entropy and tangential gradient magnitude, and the formula for calculating the radial gradient entropy is: ,in, For the first The magnitude of the precipitation intensity gradient in the radial direction, Divide the number of items in the radial direction. For the first The probability density of gradients in each direction.

[0016] Preferably, the magnitude of the tangential gradient is calculated using Gaussian kernel convolution and the directional derivative, as shown in the formula: ,in, Rainband pixels The precipitation intensity value, The tangential direction angle, The standard deviation is Two-dimensional Gaussian kernel, This represents the convolution operation.

[0017] Preferably, the dynamic evolution parameters include the morphological transition rate and the shape similarity decay coefficient, and the morphological transition rate is calculated as follows:

[0018]

[0019] in, This represents the change in the topological connectivity of the rainband at adjacent time points. This represents the change in the fractal dimension of the morphology. This represents the change in radial gradient entropy. For time intervals.

[0020] Preferably, the extraction of target morphology parameters from observed rainband data and predicted rainband data includes:

[0021] The observed rainband data and the predicted rainband data are preprocessed, including resolution normalization, noise removal and rainband region segmentation.

[0022] Based on the preprocessed dataset, spatial topological parameters are extracted using topological analysis algorithms, intensity gradient parameters are extracted using gradient operators and entropy calculations, and dynamic evolution parameters are extracted using time series difference analysis.

[0023] The extracted parameters are standardized to obtain a standardized set of observed morphological parameters. With forecast morphological parameter set ,in This represents the total number of parameters.

[0024] Preferably, the attention-weighted parameter matching algorithm includes:

[0025] Calculate the dimension-by-dimensional error between observed and predicted morphological parameters. Construct the error vector ;

[0026] The importance weights of the parameters are calculated using a self-attention mechanism. The formula is: ,in, , These represent the mean and standard deviation of the observed morphological parameter set, respectively. , These represent the mean and standard deviation of the forecast morphological parameter set, respectively. For the first The correlation score of each parameter;

[0027] Calculate the multidimensional error contribution based on the error vector and importance weights. And the dominant error type is identified by threshold filtering.

[0028] Preferably, the dominant error types include topological deviation type, intensity distribution deviation type, and dynamic evolution deviation type, and the determination criterion for the topological deviation type is as follows: ,in This is the set of indices corresponding to spatial topology parameters; the determination criterion for the intensity distribution deviation type is... ,in This is the set of indices corresponding to the intensity gradient parameters; the determination criterion for the dynamic evolution deviation type is... ,in This is the set of indices corresponding to the dynamic evolution parameters.

[0029] Preferably, the stratified diagnostic results for generating rainband forecast errors include:

[0030] Based on the dominant error type, the corresponding numerical model error sources are traced, including initial field error, physical parameterization scheme deviation, and dynamic frame discretization error.

[0031] The range of error influence is quantified by the error contribution, and the area of ​​error influence is calculated. ,in This represents the total area of ​​the rainband;

[0032] Based on the sources and scope of error, the generation mode optimization direction is guided, including the adjustment of the initial field assimilation strategy, the correction of physical parameterization scheme parameters, and the adaptation of dynamic frame resolution.

[0033] Compared with the prior art, the technical solution of this application has the following technical effects:

[0034] This invention overcomes the limitations of traditional error diagnosis, which only focuses on numerical deviations, by constructing a multi-dimensional target morphological parameter system that encompasses spatial topology, intensity gradient, and dynamic evolution. It achieves a comprehensive characterization of the geometric structure, intensity distribution, and temporal evolution characteristics of heavy precipitation rainbands. The synergistic effect of multi-dimensional parameters can accurately capture fine-grained differences in rainband morphology, providing richer feature support for error diagnosis. This upgrades error identification from numerical comparison to morphological feature matching, significantly improving the comprehensiveness and accuracy of error diagnosis.

[0035] The attention-weighted parameter matching algorithm of this invention dynamically allocates the importance weight of parameters through a self-attention mechanism, which solves the drawback of traditional methods that treat all parameters equally. It can adaptively focus on the core parameters that have a significant impact on the accuracy of rainband forecasts. By quantifying the contribution of multidimensional errors, the algorithm can accurately locate the dominant error type, avoid the fuzzy determination of error sources, make error diagnosis more targeted, and provide a clear target direction for subsequent model optimization.

[0036] This invention ensures the comparability and reliability of observed and predicted morphological parameters through a series of refined parameter extraction processes, including standardized processing, topological analysis, and gradient calculation. The parameter extraction process fully considers the spatial correlation and temporal dynamics of rainband data, effectively filtering out the influence of noise interference and scale differences, so that the extracted parameters can truly reflect the essential morphological characteristics of the rainband, laying a solid foundation for error calculation and source tracing.

[0037] The hierarchical diagnostic results of this invention not only enable precise tracing of error sources, but also provide specific and actionable directions for numerical model optimization by quantifying the scope of error impact. This diagnostic model directly links error diagnosis with model improvement, breaking through the problem of the disconnect between traditional diagnostic methods and practical applications. It can guide researchers to adjust initial field assimilation strategies, correct physical parameterization schemes, or adapt to dynamic framework resolution in a targeted manner, thereby helping to improve the overall quality of heavy precipitation rainband forecasts.

[0038] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the preferred embodiments of this application are described in detail below with reference to the accompanying drawings.

[0039] The above and other objects, advantages and features of this application will become more apparent to those skilled in the art from the following detailed description of specific embodiments in conjunction with the accompanying drawings. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In all drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0041] Based on the description of the figures and their corresponding technical content in the document, the titles of the figures are as follows:

[0042] Figure 1 A schematic diagram of the overall process of the error diagnosis method for heavy precipitation rainbands based on target morphological parameters;

[0043] Figure 2 Schematic diagram illustrating the composition of the target morphological parameter system for heavy precipitation rainbands and the function of each parameter;

[0044] Figure 3 Schematic diagram of the process for preprocessing heavy precipitation rainband observation and forecast data and extracting target morphological parameters;

[0045] Figure 4 A schematic diagram illustrating the process of calculating the contribution of rainband error and locating its type using an attention-weighted parameter matching algorithm.

[0046] Figure 5 Schematic diagram showing the spatial distribution of the error contribution of heavy precipitation rainbands and the corresponding rainfall levels in the regions;

[0047] Figure 6 A schematic diagram comparing the characteristic distribution of extracted and observed values ​​of target morphological parameters from 100 sets of heavy precipitation samples;

[0048] Figure 7 Schematic diagram of the quantitative results of multidimensional error contribution, dominant error type and impact range of heavy precipitation rainband;

[0049] Figure 8 Schematic diagram of the time series variation of the rate of change of heavy precipitation rainband morphology and the decay coefficient of shape similarity;

[0050] Figure 9 A comparative diagram of the accuracy, computational efficiency, and F1 score of error diagnosis for heavy precipitation rainbands under different parameter combinations;

[0051] Figure 10 : A schematic diagram showing the relationship between the adjustment range of heavy precipitation rainband model parameters and the corresponding error reduction rate. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. In the following description, specific details such as specific configurations and components are provided merely to help fully understand the embodiments of this application. Therefore, those skilled in the art should understand that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this application. In addition, for clarity and brevity, descriptions of known functions and structures are omitted in the embodiments.

[0053] It should be understood that the phrase "an embodiment" or "this embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "an embodiment" or "this embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments.

[0054] Furthermore, reference numerals and / or letters may be repeated in different examples within this application. Such repetition is for the purpose of simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or settings discussed.

[0055] In this article, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, B exists alone, and A and B exist simultaneously. The term " / and" in this article describes another type of relationship between related objects, indicating that two relationships can exist. For example, A / and B can mean: A exists alone, and A and B exist alone. In addition, the character " / " in this article generally indicates that the related objects before and after it are in an "or" relationship.

[0056] In this article, the term "at least one" is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, "at least one of A and B" can mean: A exists alone, A and B exist simultaneously, or B exists alone.

[0057] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion.

[0058] Example 1

[0059] This embodiment mainly describes a method for diagnosing errors in heavy precipitation rainbands based on target morphological parameters, such as... Figure 1 As shown, it specifically includes:

[0060] Acquire observed rainband data and forecasted rainband data corresponding to the heavy precipitation process. The observed rainband data includes multi-source observed fused precipitation data, and the forecasted rainband data includes precipitation forecast data output by numerical models.

[0061] Based on the multi-dimensional characteristics of rainband morphology, a target morphology parameter system is constructed. The target morphology parameter system includes spatial topology parameters, intensity gradient parameters, and dynamic evolution parameters. The spatial topology parameters are used to characterize the geometric structure correlation of the rainband, the intensity gradient parameters are used to quantify the spatial variation rate of precipitation intensity within the rainband, and the dynamic evolution parameters are used to characterize the morphological change features of the rainband in the time series.

[0062] Each parameter in the target morphological parameter system is extracted from the observed rainband data and the predicted rainband data respectively to obtain the observed morphological parameter set and the predicted morphological parameter set;

[0063] The multidimensional error contribution between the observed morphological parameter set and the predicted morphological parameter set is calculated using an attention-weighted parameter matching algorithm to identify the dominant error type.

[0064] Based on the dominant error type and multidimensional error contribution, a hierarchical diagnostic result for rainband forecast error is generated. The hierarchical diagnostic result includes error source tracing, error impact range quantification, and model optimization direction guidance.

[0065] Furthermore, we obtained the observed rainband data and forecast rainband data corresponding to the heavy precipitation process, specifically:

[0066] For the observed rainband data, the data sources for the multi-source observation and fusion precipitation data include hourly precipitation observation data from ground automatic weather stations, precipitation data retrieved from Doppler weather radar reflectivity factors, and precipitation data retrieved from geostationary satellite infrared and microwave channels. After acquisition, a three-level quality control process is first performed: Level 1 quality control removes outliers exceeding physical thresholds (e.g., hourly rainfall intensity > 200 mm); Level 2 quality control removes abrupt noise through temporal continuity checks; and Level 3 quality control removes isolated outliers through spatial consistency checks. Subsequently, a multi-source data fusion algorithm based on Bayesian estimation is used, assigning weights according to the error covariance matrix of each data source to generate unified grid precipitation data with a spatial resolution of 0.1° × 0.1° and a temporal resolution of 3 hours. The data format is NetCDF, and the coordinate system adopts the WGS84 geographic coordinate system.

[0067] For forecast rainband data, the precipitation forecast data output by the numerical model comes from global or regional numerical weather prediction models, including 12-hour, 24-hour, and 36-hour lead-time forecasts for corresponding heavy precipitation events output after model integration. During acquisition, a forecast field that spatially and temporally matches the observed rainband data is extracted. The forecast data is interpolated to a standard grid of 0.1°×0.1° using bilinear interpolation to ensure consistent spatial resolution. Temporally, lead-time forecast data is extracted at 3-hour intervals to achieve temporal alignment with the observed data. The data format is uniformly converted to NetCDF to ensure compatibility with the observed data.

[0068] The processed observation and forecast data are checked for spatiotemporal consistency. Temporally, the integrity of the timestamps is checked, and data with missing time nodes are removed. Spatially, the effective coverage of the data is checked to ensure that there are no data gaps in the core area of ​​the rainband. At the same time, the rationality of the numerical range of precipitation intensity is checked to avoid numerical deviations caused by unit conversion errors or interpolation anomalies. Finally, a complete and consistent observation rainband dataset and forecast rainband dataset are formed.

[0069] Furthermore, such as Figure 2 As shown, a target morphological parameter system is constructed through spatial structure, intensity distribution, and temporal evolution characteristics, specifically as follows:

[0070] Spatial topology parameters: focusing on the correlation and integrity of the rainband's geometric structure, including three core parameters:

[0071] Rainband topological connectivity: Used to quantify the connectivity of rainband regions, the calculation formula is as follows: ,in, for Order connectivity matrix, The total number of effective pixels in the rainband (i.e., the number of pixels with precipitation intensity ≥ the heavy precipitation threshold, where the heavy precipitation threshold is set according to different regional climate characteristics as hourly rainfall intensity ≥ 20 mm or ≥ 15 mm). With pixels When 8-neighbor connectivity is satisfied (i.e., adjacent in horizontal, vertical, and diagonal directions), ,otherwise ; The connectivity weights are calculated using a distance decay function: , For pixels and Euclidean distance between them (unit: km) The characteristic distance (fixed at 10km) ensures that short-range connectivity contributes more; This is the theoretical maximum value of the sum of the elements of the connectivity matrix (i.e., the sum of the elements when all pixels are fully connected). The maximum value of the connectivity weight (when hour, ),final The value ranges from 0 to 1, and the closer it is to 1, the better the connectivity of the rain belt.

[0072] Morphological fractal dimension: used to characterize the complexity of rainband morphology, calculated using box counting, with the formula as follows. In practice, the binary mask image of the rainband (rainband areas are 1, non-rainband areas are 0) is placed in a Cartesian coordinate system, and a series of box side lengths are set in a geometric progression. The initial side length is the maximum side length of the bounding rectangle of the rainband, and subsequent side lengths decrease by a scaling factor of 0.5 until the side length equals the side length of a single pixel; for each side length Calculate the minimum number of boxes required to cover the entire rainband area. (That is, the total number of boxes that contain at least one rainband pixel); for and Perform a linear fit, and the absolute value of the slope of the fitted line is the morphological fractal dimension. The value ranges from 1 to 2, and the closer it is to 2, the more complex the rainband morphology.

[0073] Centroid offset vector: Used to characterize the spatial positional deviation of the rainband, it consists of the coordinate difference between the observed rainband centroid and the predicted rainband centroid. The centroid coordinates are calculated using a weighted average. ,in For pixels Geographic coordinates (longitude and latitude). The precipitation intensity value for this pixel; the centroid offset vector is... , To observe the coordinates of the rainband's centroid, To predict the centroid coordinates of the rainband, the magnitude of the vector represents the position offset distance (unit: km), and the direction represents the offset azimuth.

[0074] Intensity gradient parameter: Focuses on the spatial variation characteristics of precipitation intensity within the rainband, including two core parameters:

[0075] Radial gradient entropy: used to quantify the uniformity of the distribution of rainband intensity gradient in different radial directions, calculated as follows: ,in Establish a polar coordinate system with the centroid of the rainband as the origin, and divide the 360° circle evenly into... Each radial direction (each direction spaced 45° apart); for each radial direction The first-order difference of precipitation intensity is calculated along this direction to obtain the gradient magnitude in that direction. , This represents the number of rainband pixels in that direction. For the first Precipitation intensity per pixel; For the first The probability density of a gradient in a given direction is obtained by the ratio of the magnitude of the gradient in that direction to the sum of the magnitudes of the gradients in all directions. The value range is 0- The closer to 0, the more uniform the intensity gradient distribution in all directions; the closer to 0, the more uniform the intensity gradient distribution in all directions. This indicates a more uneven distribution.

[0076] Tangential gradient magnitude: Used to quantify the rate of change of rainband intensity in the tangential direction, calculated through Gaussian kernel convolution and directional derivative, as shown in the formula. ,in, Rainband pixels The precipitation intensity value; The tangential direction angle satisfies the following condition with the radial direction angle: ( (This refers to the radial direction angle, with values ​​of 0°, 45°, ..., 315°). , The precipitation intensity is respectively at , The first-order partial derivative in the direction is calculated using the Sobel operator; The standard deviation is A two-dimensional Gaussian kernel that satisfies , The side length is adaptively set to 3 pixels based on the data resolution to smooth the gradient calculation results and reduce noise interference. This represents a two-dimensional convolution operation, implemented by traversing the rainband region through a sliding window, ultimately... The larger the value, the more drastic the change in tangential intensity at that pixel.

[0077] Dynamic evolution parameters: Focusing on the changes in rainband morphology over time, including two core parameters:

[0078] The rate of morphological change, used to quantify how quickly the rainband morphology changes, is calculated using the following formula: .in, Adjacent time points and The change in the topological connectivity of the rainband ranges from -1 to 1. This is the change in the fractal dimension of the morphology, with a value range of -1 to 1; This represents the change in radial gradient entropy, with a value range of - - ; The time interval is fixed at 3 hours (consistent with the data time resolution). The unit is (dimensionless h⁻¹), and the larger the value, the faster the rainband shape changes.

[0079] Shape similarity decay coefficient: used to quantify the change in similarity of rainband morphology over time, calculated as follows: .in, For a moment morphological parameter vector ( For a moment (mean of tangential gradient magnitude) This is the shape parameter vector at the initial moment (the moment when the heavy precipitation process begins); It is the dot product of two vectors; Let be the magnitudes of the two vectors, respectively. The value range is -1 to 1, with values ​​closer to 1 indicating a higher time. The more similar the shape of the rainband is to the initial shape, the closer it is to -1, indicating a greater difference in shape.

[0080] Furthermore, the target morphological parameters are extracted through preprocessing, parameter calculation, and standardization, specifically as follows:

[0081] Resolution normalization was achieved by using a bilinear interpolation algorithm to interpolate both observed and predicted rainband data to a standard grid of 0.1°×0.1°. During the interpolation process, the geographic coordinate system (WGS84) was kept consistent to avoid parameter calculation deviations caused by resolution differences. For the portion of the predicted data that exceeds the time range of the observed data, it was removed, and only the data in the spatiotemporally overlapping areas were retained for parameter extraction.

[0082] Noise Removal: A combined denoising strategy of median filtering and morphological filtering is adopted. Isolated point noise is removed by median filtering with a 3×3 window, and small connected noise is removed by morphological opening operation (erosion followed by dilation). The structural elements of erosion and dilation are both 3×3 square structural elements to ensure that the main shape of the rainband is not destroyed.

[0083] Rainband region segmentation: An adaptive thresholding method is used to determine the heavy precipitation threshold. Based on the precipitation intensity histogram of rainband data, the optimal threshold is calculated using the Otsu algorithm. The intensity of precipitation The area was identified as a rain belt region. The region is identified as a non-rainband region, and a rainband binary mask image is generated for subsequent calculation of topological and gradient parameters.

[0084] Parameter calculation, such as Figure 3 As shown, it specifically includes:

[0085] Spatial topology parameter calculation: Based on the rainband binary mask image, a connectivity matrix is ​​constructed using a neighborhood traversal algorithm. Calculate the connectivity weights using the distance decay function. Substituting into the formula, we obtain the topological connectivity of the rainband. Different side lengths were obtained through multi-scale box cover statistics. Corresponding number of boxes The morphological fractal dimension was obtained through linear fitting. The centroid coordinates of the rainband are calculated by weighted averaging, and then the centroid offset vector is obtained.

[0086] Intensity gradient parameter calculation: Based on the preprocessed precipitation intensity data, the precipitation is divided into 8 radial directions with the centroid of the rainband as the origin, and the gradient magnitude in each direction is calculated. With probability density Substituting into the formula yields the radial gradient entropy. The first-order partial derivative of precipitation intensity is calculated using the Sobel operator, and the tangential gradient magnitude is obtained by combining Gaussian kernel convolution with the directional derivative. And calculate the mean value of the rainband region. .

[0087] Dynamic evolution parameter calculation: The spatial topological parameters and intensity gradient parameters on the time series are differencinged to obtain... , , Substituting into the formula, we obtain the morphological transition rate. Construct morphological parameter vectors at each time step. With the initial time vector The shape similarity attenuation coefficient is calculated by dot product and modulus. .

[0088] Standardization: To eliminate the influence of differences in the units and numerical ranges of different parameters on subsequent error calculations, the Z-score standardization method is used to standardize all extracted parameters. The formula is as follows: .in, These are the original parameter values. This is the mean of the parameter across all samples. This is the sample standard deviation of the parameter. The sample size (i.e., the time series length of the rainband data); after standardization, the set of observed morphological parameters is obtained. With forecast morphological parameter set ,in The total number of parameters (spatial topology parameters, intensity gradient parameters, and dynamic evolution parameters) has a mean of 0 and a standard deviation of 1.

[0089] Furthermore, attention-weighted parameter matching algorithms, such as Figure 4 As shown, specifically:

[0090] Dimensional error calculation is performed by analyzing the standardized set of observed morphological parameters. With forecast morphological parameter set Calculate the absolute error of each parameter. ( ),in For the first set of observed morphological parameters The standardized values ​​of each parameter For the first set of forecast morphological parameters The standardized values ​​of each parameter; all dimension-wise errors constitute the error vector. The dimension of the error vector is the same as the total number of parameters.

[0091] Parameter importance weight calculation: The importance weight of each parameter is dynamically calculated based on a self-attention mechanism. The core is to quantify the contribution of each parameter to rainband morphology matching. The formula is as follows: ,in For the first The relevance score of each parameter. In specific calculations, , The mean and standard deviation of the observed morphological parameter set (after standardization) are respectively. , ); , These are the mean and standard deviation (after standardization) of the forecast morphological parameter set. , Therefore, the relevance score can be simplified to The value ranges from -1 to 1. The closer it is to 1, the stronger the correlation between the observed and predicted parameters, and the greater its contribution to morphological matching. The Softmax function is used to... Normalization is performed to obtain the parameter importance weights. The value range is 0-1, and satisfies A larger weight indicates that the parameter has a more significant impact on the error.

[0092] Multidimensional error contribution calculation and dominant error type identification:

[0093] Multidimensional error contribution calculation: Combining parameter importance weights and dimension-wise errors, the multidimensional error contribution of each parameter is calculated. ,in This is the maximum value in the error vector, used to normalize the error contribution to the 0-1 range, ensuring that the contributions of different parameters are comparable; The larger the value, the greater the contribution of the error corresponding to that parameter to the overall error.

[0094] Dominant error type localization: The index set is divided according to the dimension to which the parameter belongs; the index set corresponding to the spatial topology parameter. (Corresponding to topological connectivity, morphological fractal dimension, and centroid offset vector), the set of indices corresponding to the intensity gradient parameters. (Corresponding to radial gradient entropy and tangential gradient magnitude), the set of indices corresponding to the dynamic evolution parameters. (Corresponding to morphological change rate and shape similarity decay coefficient); calculate the sum of the error contributions of each dimension parameter respectively. , , The threshold for judgment is set to 0.5. If and , Then the dominant error type is determined to be topological deviation type; if and , If so, the dominant error type is determined to be intensity distribution deviation type; if and , If the error is less than 0.5, it is determined to be a dynamic evolution deviation type; if none of them exceed 0.5, it is determined to be a comprehensive deviation type.

[0095] Furthermore, the generation of tiered diagnostic results involves a logical chain of error localization, range quantification, and optimization guidance, achieving a closed loop from error identification to pattern improvement. The specific process is as follows:

[0096] Error source tracing: Based on the dominant error type, combined with the operating mechanism and physical processes of the numerical model, the root cause of the error is accurately traced.

[0097] Topological deviation type: The main sources of error include initial field error and dynamic frame discretization error. Initial field error stems from insufficient assimilation of observation data, resulting in discrepancies between the spatial distribution of the initial water vapor field and wind field and the actual situation, which in turn affects the simulation of the geometric structure of the rainband; dynamic frame discretization error stems from the truncation error of the spatial discretization scheme (such as finite difference, finite volume method) used by the model, which leads to distortion in the simulation of the connectivity and morphological complexity of the rainband.

[0098] Intensity distribution deviation type: The main source of error is the deviation of the physical parameterization scheme. In the micro-physical parameterization scheme, the falling velocity and terminal velocity of precipitation particles are not accurately parameterized, resulting in deviations in the simulation of the vertical distribution of precipitation intensity; in the cumulus convection parameterization scheme, the calculation deviations of convection triggering conditions, convective heating rate and water vapor condensation rate result in the intensity gradient distribution inside the rainband not matching reality.

[0099] Dynamic evolution deviation type: The main sources of error are deviations in the time integration scheme and unreasonable settings of physical process timescale parameters. Excessive numerical dissipation in the time integration scheme leads to a slower evolution rate of the rainband morphology; the settings of physical process timescale parameters (such as the boundary layer turbulent mixing timescale and convection adjustment timescale) do not match the actual atmospheric processes, resulting in distortion of the simulated rainband morphology change rate and similarity attenuation coefficient.

[0100] Comprehensive deviation type: The error source is the superposition of the above two or three types of errors. It is necessary to combine the proportion of the error contribution of each dimension and trace the corresponding error root cause.

[0101] Quantifying the scope of error impact: Quantifying the degree of error impact from both spatial range and magnitude of intensity.

[0102] Spatial extent quantification: achieved through area calculation based on the impact of errors, using the following formula: ,in The total area of ​​the rainband This represents the total number of effective pixels in the rainband. The area of ​​a single pixel (0.1° × 0.1°, approximately 11.1 km²). The maximum value among the multidimensional error contributions is used to characterize the severity of the error. This indicates the area of ​​the rainband affected by the error; the larger the area, the wider the spatial range affected by the error.

[0103] Intensity magnitude quantization: achieved through error intensity index calculation, the formula is as follows: ,in For the first The mean of the original parameters is used to restore the standardized error contribution to the scale of the original parameters; The larger the value, the greater the magnitude of the error and the more significant its impact on the accuracy of precipitation forecasts.

[0104] Pattern optimization guidance: Based on the sources and scope of error, generate targeted and actionable pattern optimization suggestions.

[0105] To address initial field errors: optimize the initial field assimilation strategy, increase the assimilation weight of high-resolution observation data (such as radar and satellite data), and adopt advanced assimilation algorithms such as ensemble Kalman filtering to improve the spatial resolution and accuracy of the initial field; increase the types of assimilation variables, supplementing precipitation-related variables such as water vapor mixing ratio and cloud condensation nucleus concentration in addition to conventional temperature, air pressure, and wind field.

[0106] To address the discretization error of the dynamic frame: optimize the spatial resolution of the model by using nested grid technology in areas with frequent heavy precipitation to improve the spatial resolution of local areas; improve the spatial discretization scheme by using high-order precision numerical methods (such as the WENO scheme) to reduce the impact of truncation error on the simulation of rainband geometry.

[0107] To address deviations in the physical parameterization scheme: the microphysical parameterization scheme was revised, and the spectral distribution parameters and falling velocity parameters of precipitation particles were calibrated based on observational data; the cumulus convection parameterization scheme was optimized, the critical threshold for convection triggering (such as the CAPE threshold) was adjusted, and the calculation scheme for convective heating and water vapor transport was improved to make the intensity gradient distribution inside the rainband more realistic.

[0108] To address the bias in the time integration scheme: the model's time step was adjusted, with a smaller time step used during periods of heavy precipitation to reduce numerical dissipation; the time integration scheme was improved by adopting a semi-implicit, semi-Lagrange scheme to balance computational efficiency and numerical accuracy; and the physical process timescale parameters were calibrated by obtaining timescale values ​​that match the actual atmospheric processes based on statistical analysis of observational data.

[0109] This implementation details how to conduct error diagnosis of heavy precipitation rainbands through target morphological parameters, accurately capture core deviation characteristics such as rainband morphology, range, and direction, quantify the magnitude and distribution patterns of errors, significantly improve the pertinence and accuracy of rainband error identification, provide a scientific quantitative basis for the correction of heavy precipitation forecasts, and effectively improve the overall accuracy of heavy precipitation rainband forecasts.

[0110] Based on Example 1, this example details the effectiveness and practicality of the heavy precipitation rainband error diagnosis method based on target morphological parameters proposed in this application. The experiment uses the CMORPH-CN multi-source fusion precipitation dataset released by the National Meteorological Information Center as the observation data source. This dataset integrates precipitation observation data from ground automatic weather stations, Doppler radar, and geostationary meteorological satellites, with a spatial resolution of 0.1°×0.1° and a temporal resolution of 3 hours. The 24-hour precipitation forecast product from the China Meteorological Administration's GRAPES_GFS global numerical weather prediction model is used as the forecast data source. The model's horizontal resolution is 0.25°×0.25°, and after bilinear interpolation and resampling to 0.1°×0.1°, it is spatiotemporally matched with the observation data. A total of 100 independent heavy precipitation rainband samples from Yunnan Province were selected. All samples met the heavy precipitation standard of hourly rainfall intensity ≥20 mm. The samples cover heavy precipitation processes in different regions and seasons of Yunnan, ensuring the universality of the experimental results. Figure 5 As shown, Figure 5 (a) By drawing a spatial distribution map of error contribution and using hierarchical coloring, the location and shape of high-value error regions are visually displayed. Figure 5 (a) and (b) are schematic diagrams of the rainband and rainfall, respectively. The experiment used morphological parameter extraction accuracy, error diagnosis accuracy, and the effectiveness of pattern optimization guidance as core evaluation indicators. All data were calculated five times and the average was taken. The calculation software was Python 3.9 combined with libraries such as GDAL, NumPy, and Matplotlib to ensure efficient and accurate data processing.

[0111] The experiment first verified the extraction accuracy of the target morphological parameter system. The extraction results of three types of parameters—spatial topology, intensity gradient, and dynamic evolution—were statistically analyzed from 100 sets of samples. The mean, standard deviation, absolute error (AE), and relative error (RE) of the parameters were calculated, and the results are shown in Table 1. In Table 1, TC represents the rainband topological connectivity, FD represents the fractal dimension, COV represents the centroid offset vector (unit: km), RGE represents the radial gradient entropy, and TGM represents the tangential gradient magnitude (unit: mm·h). -1 ·km -1 MCR stands for Morphological Change Rate (unit: h). -1SSAC stands for Shape Similarity Attenuation Coefficient. As shown in Table 1, the relative errors of all parameters in the 100 samples are consistently controlled within 8%. Specifically, the average relative errors of topological connectivity and morphological fractal dimension are only 2.3% and 3.1%, respectively. The average absolute error of the centroid offset vector is 4.2 km, with a maximum absolute error not exceeding 8.5 km. The relative errors of radial gradient entropy and tangential gradient magnitude are 6.5% and 8.0%, respectively, all within acceptable ranges. This indicates that the parameter extraction process of this method has extremely high accuracy and stability, and can truly reflect the morphological characteristics of the rainband.

[0112] Table 1. Statistics on the accuracy of target morphological parameter extraction from 100 sets of samples.

[0113]

[0114] To visually represent the distribution characteristics and extraction results of morphological parameters, a visualization of morphological parameter features is generated based on the statistical results of 100 sets of samples, such as... Figure 6 As shown, a standardized coordinate system is used (the horizontal axis represents the parameter dimension index, and the vertical axis represents the standardized parameter values; the standardization method is Z-score). Subplot (a) shows the distribution characteristics of topological connectivity (TC) and morphological fractal dimension (FD). The blue bars represent the mean distribution of the observed parameters, the red line represents the mean distribution of the extracted parameters, and the error bars represent the standard deviation range of the parameters. It can be seen that the mean overlap of the two reaches 98.2%, and the error bars completely overlap, indicating that the extracted values ​​are highly consistent with the observed values. Subplot (b) shows the distribution characteristics of centroid offset vector (COV) and radial gradient entropy (RGE). The green scatter points are the extracted values ​​of 100 samples, and the black solid line is the observed value. The fitting curves show that the distribution error of the scatter points on both sides of the fitting curves is less than ±0.1. Subplot (c) shows the distribution characteristics of tangential gradient magnitude (TGM) and morphological change rate (MCR). The orange bar chart represents the frequency distribution of the extracted parameters, and the gray shading represents the frequency distribution range of the observed parameters. The frequency peaks of the two completely overlap, and the overlap of the distribution ranges reaches 95.6%. Subplot (d) shows the temporal distribution characteristics of shape similarity decay coefficient (SSAC). The purple curve represents the average temporal change of 100 samples, and the light blue shading represents the confidence interval of ±1 standard deviation. The curve shows a smooth downward trend, and the confidence interval width does not exceed 0.08, indicating that the parameter extraction results have good stability.

[0115] Error diagnosis accuracy was validated by comparing the diagnostic results of the attention-weighted parameter matching algorithm with the manually labeled results. Based on the comparative analysis of observational and forecast data, the dominant error types of 100 samples were labeled, and the diagnostic results of the 100 samples were statistically analyzed. The accuracy, precision, recall, and F1 score of different dominant error types were calculated, and the results are shown in Table 2. In Table 2, TSB represents Topological Structure Bias, IDB represents Intensity Distribution Bias, DEB represents Dynamic Evolution Bias, and CB represents Comprehensive Bias. Data shows that among 100 samples, the accuracy rates for topological structure deviation and intensity distribution deviation reached 92.5% and 90.8%, respectively, with precision and recall both exceeding 90%, and F1 scores of 0.925 and 0.908, respectively. The accuracy and F1 score for dynamic evolution deviation also reached 87.5% and 0.875, respectively. The overall deviation type had relatively lower indicators but still remained above 80%, with an overall average accuracy of 89.7% and a Kappa coefficient of 0.86. This indicates that the error diagnosis algorithm of this method has a high degree of consistency with the results of manual annotation and can accurately locate the dominant error type.

[0116] Table 2 Diagnostic performance indicators for different dominant error types in 100 sample groups

[0117]

[0118] To demonstrate the calculation results of error contribution and the spatial distribution characteristics of dominant error types, a representative set of samples from 100 sets of samples was selected to generate a visualization of error diagnosis results, such as... Figure 7As shown, a two-dimensional coordinate system is used (the horizontal axis represents the pixel index of the sample, and the vertical axis represents the parameter value). Subplot (a) shows the spatial distribution of multidimensional error contribution. The broken lines of different colors represent the error contribution of three types of parameters: spatial topology, intensity gradient, and dynamic evolution. The peak error contribution of the intensity gradient parameter reaches 0.68, the peak error contribution of the spatial topology parameter is 0.52, and the peak error contribution of the dynamic evolution parameter is 0.45. Subplot (b) shows the spatial distribution of dominant error types. Different colored blocks are used to distinguish the dominant error types (blue represents TSB, red represents IDB, yellow represents DEB, and green represents CB). The area ratio of the colored blocks is completely consistent with the sample number ratio in Table 2. Subplot (c) shows the quantification result of the error influence range. The blue curve is the cumulative error ratio, the horizontal axis is the error contribution threshold, and the vertical axis is the cumulative influence area ratio. When the threshold is 0.5, the cumulative influence area ratio is 45.7%, which is completely consistent with the theoretical influence range calculated from the maximum error contribution in Table 1, verifying the accuracy of the error influence range quantification method.

[0119] The effectiveness of the dynamic evolution parameters was verified through time series analysis of 100 sets of samples. Sample data from eight consecutive time periods (every 3 hours) were selected, and the mean values ​​of morphological change rate and shape similarity decay coefficient were extracted to generate the time series of dynamic evolution parameters, such as... Figure 8 As shown, the horizontal axis represents the time step (8 steps in total, corresponding to 24 hours), and the vertical axis represents the standardized parameter values. Subplot (a) is the time series of morphological change rate (MCR). The blue line represents the mean of 100 samples, and the error bars represent the standard deviation range. It can be seen that the rate rapidly increases from 0.08h⁻¹ to 0.21h in steps 1-3 (the precipitation development stage). -1 The step length remains stable in steps 4-5 (precipitation maturity stage), and gradually decreases to 0.06h in steps 6-8 (precipitation decay stage). -1 The width of the error bar is always less than 0.03, indicating that the difference between samples is minimal. Subplot (b) is the time series of shape similarity decay coefficient (SSAC). The red line is the mean of 100 samples, and the light blue shading is the 95% confidence interval. The coefficient decreases continuously from 0.92 at the initial step size to 0.45 at the 8th step size. The confidence interval width does not exceed 0.05, which is completely synchronized with the life cycle of the heavy precipitation rainband from formation, development, maturity to decay. This shows that the dynamic evolution parameter can accurately characterize the temporal change characteristics of the rainband morphology.

[0120] To verify the impact of parameter combinations on error diagnosis results, three sets of comparative experiments were designed: Experiment 1 used only spatial topology parameters, Experiment 2 used only intensity gradient parameters, and Experiment 3 used the complete target morphology parameter system. Each experiment used 100 samples for testing. Diagnostic accuracy, computational efficiency, and F1 score were statistically analyzed to generate a performance comparison of parameter combinations, such as... Figure 9 As shown, subplot (a) compares diagnostic accuracy. Experiment 3 achieved an accuracy of 89.7%, significantly higher than Experiment 1's 76.3% and Experiment 2's 78.5%, with improvements of 13.4% and 11.2%, respectively. Subplot (b) compares computational efficiency, with the vertical axis representing single-sample processing time (in seconds). Experiment 1's processing time was 0.8 s / sample, Experiment 2's was 0.9 s / sample, and Experiment 3's was 1.0 s / sample, only 0.1-0.2 s more than the first two groups, indicating that computational efficiency remained high. Subplot (c) compares F1 scores. Experiment 3 achieved an F1 score of 0.883, an improvement of 0.133 and 0.113 compared to Experiments 1 and 2, respectively, demonstrating that a complete target morphological parameter system can significantly improve error diagnosis performance, and the increase in computational overhead is within an acceptable range.

[0121] To verify the effectiveness of the model optimization guidance, based on the diagnostic results of this method, the physical parameterization scheme of the GRAPES_GFS model was adjusted: for intensity distribution deviation errors, the falling velocity parameters of precipitation particles in the microphysical parameterization scheme were corrected; for topology deviation errors, the assimilation weight of the water vapor field in the initial field was increased. The adjusted model output data was compared with the original data, and the parameter error changes of 100 samples were statistically analyzed to generate a correlation between parameter adjustment and error changes, such as... Figure 10 As shown, the horizontal axis represents the parameter adjustment range (0-20%), and the vertical axis represents the error reduction rate (%). Subplot (a) shows the relationship between topological parameter adjustment and error change. The blue curve represents the error reduction rate of topological connectivity. When the adjustment range is 15%, the error reduction rate reaches its maximum value of 62%, and then tends to stabilize. Subplot (b) shows the relationship between intensity parameter adjustment and error change. The red curve represents the error reduction rate of radial gradient entropy. When the adjustment range is 12%, the error reduction rate reaches its maximum value of 58%, which is completely consistent with the experimental results, indicating that the parameter adjustment direction is accurate and the error reduction effect is significant.

[0122] In summary, experimental verification using 100 sets of heavy precipitation samples based on the CMORPH-CN observation dataset and the GRAPES_GFS forecast dataset demonstrates that the proposed method for diagnosing heavy precipitation rainband errors based on target morphological parameters exhibits excellent performance in terms of parameter extraction accuracy, error diagnosis accuracy, and the effectiveness of model optimization guidance. All core indicators meet the requirements for practical application and can provide scientific and reliable technical support for the forecast error diagnosis of heavy precipitation rainbands.

[0123] The above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. For those skilled in the art, the present invention can have various modifications and variations. Any changes, modifications, substitutions, integrations, and parameter changes made to these embodiments within the spirit and principles of the present invention, without departing from the principles and spirit of the present invention, through conventional substitutions or to achieve the same function, fall within the scope of protection of the present invention.

Claims

1. A method for diagnosing errors in heavy precipitation rainbands based on target morphological parameters, characterized in that, include: Acquire observed rainband data and forecasted rainband data corresponding to the heavy precipitation process. The observed rainband data includes multi-source observed fused precipitation data, and the forecasted rainband data includes precipitation forecast data output by numerical models. Based on the multi-dimensional characteristics of rainband morphology, a target morphology parameter system is constructed. The target morphology parameter system includes spatial topology parameters, intensity gradient parameters, and dynamic evolution parameters. The spatial topology parameters are used to characterize the geometric structure correlation of the rainband, the intensity gradient parameters are used to quantify the spatial variation rate of precipitation intensity within the rainband, and the dynamic evolution parameters are used to characterize the morphological change features of the rainband in the time series. Each parameter in the target morphological parameter system is extracted from the observed rainband data and the predicted rainband data respectively to obtain the observed morphological parameter set and the predicted morphological parameter set; The multidimensional error contribution between the observed morphological parameter set and the predicted morphological parameter set is calculated using an attention-weighted parameter matching algorithm to identify the dominant error type. Based on the dominant error type and multidimensional error contribution, a hierarchical diagnostic result for rainband forecast error is generated. The hierarchical diagnostic result includes error source tracing, error impact range quantification, and model optimization direction guidance.

2. The method according to claim 1, characterized in that, The spatial topological parameters include rainband topological connectivity, morphological fractal dimension, and centroid offset vector. The rainband topological connectivity is calculated using the following formula: ,in, Rainband pixels With pixels The connectivity matrix is ​​denoted by 1 if the elements are connected, and 0 otherwise. For connectivity weights, This represents the total number of effective pixels in the rainband. This represents the maximum weight.

3. The method according to claim 2, characterized in that, The morphological fractal dimension is calculated using the box counting method, with the following formula: ,in, Let's say it's the side length of the box. The minimum number of boxes required to cover the entire rainband region was obtained by performing multi-scale box coverage statistics on the binary mask image of the rainband.

4. The method according to claim 3, characterized in that, The intensity gradient parameters include radial gradient entropy and tangential gradient magnitude. The formula for calculating the radial gradient entropy is: ,in, For the first The magnitude of the precipitation intensity gradient in the radial direction, Divide the number of items in the radial direction. For the first The probability density of gradients in each direction.

5. The method according to claim 4, characterized in that, The magnitude of the tangential gradient is calculated using Gaussian kernel convolution and the directional derivative, as follows: ,in, Rainband pixels The precipitation intensity value, The tangential direction angle, The standard deviation is Two-dimensional Gaussian kernel, This represents the convolution operation.

6. The method according to claim 5, characterized in that, The dynamic evolution parameters include the morphological transition rate and the shape similarity decay coefficient. The formula for calculating the morphological transition rate is: in, This represents the change in the topological connectivity of the rainband at adjacent time points. This represents the change in the fractal dimension of the morphology. This represents the change in radial gradient entropy. For time intervals.

7. The method according to claim 6, characterized in that, The extraction of target morphology parameters from observed rainband data and predicted rainband data includes: The observed rainband data and the predicted rainband data are preprocessed, including resolution normalization, noise removal and rainband region segmentation. Based on the preprocessed dataset, spatial topological parameters are extracted using topological analysis algorithms, intensity gradient parameters are extracted using gradient operators and entropy calculations, and dynamic evolution parameters are extracted using time series difference analysis. The extracted parameters are standardized to obtain a standardized set of observed morphological parameters. With forecast morphological parameter set ,in This represents the total number of parameters.

8. The method according to claim 7, characterized in that, The attention-weighted parameter matching algorithm includes: Calculate the dimension-by-dimensional error between observed and predicted morphological parameters. Construct the error vector ; The importance weights of the parameters are calculated using a self-attention mechanism. The formula is: ,in, , These represent the mean and standard deviation of the observed morphological parameter set, respectively. , These represent the mean and standard deviation of the forecast morphological parameter set, respectively. For the first The correlation score of each parameter; Calculate the multidimensional error contribution based on the error vector and importance weights. And the dominant error type is identified by threshold filtering.

9. The method according to claim 8, characterized in that, The dominant error types include topological deviation type, intensity distribution deviation type, and dynamic evolution deviation type. The determination criterion for the topological deviation type is as follows: ,in This is the set of indices corresponding to spatial topology parameters; the determination criterion for the intensity distribution deviation type is... ,in This is the set of indices corresponding to the intensity gradient parameters; the determination criterion for the dynamic evolution deviation type is... ,in This is the set of indices corresponding to the dynamic evolution parameters.

10. The method according to claim 9, characterized in that, The stratified diagnostic results of the generated rainband forecast error include: Based on the dominant error type, the corresponding numerical model error sources are traced, including initial field error, physical parameterization scheme deviation, and dynamic frame discretization error. The range of error influence is quantified by the error contribution, and the area of ​​error influence is calculated. ,in The total area of ​​the rainband; Based on the sources and scope of error, the generation mode optimization direction is guided, including the adjustment of the initial field assimilation strategy, the correction of physical parameterization scheme parameters, and the adaptation of dynamic frame resolution.

Images