Improved method for spatial structure forecast of disastrous gale based on multi-model integration
By improving the weight calculation of the multi-mode integrated model and combining the spatial structural characteristics of catastrophic strong winds, the accuracy and reliability problems of catastrophic strong wind forecasts in the existing technology are solved, and a more efficient catastrophic strong wind space structure forecast is achieved.
Patent Information
- Application Number
- CN202410957391.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-17
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-07-17
AI Technical Summary
The prior art has high underreport rate, low accuracy and high altitude rate in the forecast of catastrophic strong winds, and the multi-mode integration method fails to fully utilize the spatial structure characteristics between modes, resulting in uncertainty and deviation of forecast results.
By analyzing the spatial structure characteristics of catastrophic strong winds, the weight calculation method of the multi-mode integrated model is improved, convolution processing and threshold control are used to identify catastrophic strong wind objects, calculate their area, center of mass position, axis angle and aspect ratio, and use the maximum Similar median scores or object-based risk scoring methods evaluate spatial similarity and build multi-modal integrated forecasting models to improve forecast accuracy.
It improves the accuracy of the numerical forecasting of the spatial structure of catastrophic high winds and the reliability of the forecast results, and is better than the single-mode forecasting and equal-weight multi-mode integration methods, especially in the forecasting of catastrophic high winds with complex spatial characteristics.
Smart Images

Figure CN118501986B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of atmospheric science research, and in particular to an improved method for forecasting the spatial structure of disastrous gale winds based on multi-mode integration. Background Art
[0002] Catastrophic gale winds above level 5 are characterized by localized and sudden weather conditions, and are highly destructive and disastrous. Therefore, accurate forecasting and warning of catastrophic gale winds are crucial to the safety of people's lives and property, as well as industries such as agriculture, transportation, electricity, and communications. It is also one of the key points and difficulties in current meteorological disaster prevention and mitigation.
[0003] In recent years, the research on disastrous gale at home and abroad has mainly focused on case analysis and cause analysis. The operational forecast of disastrous gale faces the current situation of high reporting rate, high missed reporting rate and low accuracy. In addition, the current domestic classified severe convective weather monitoring and early warning are mainly based on subjective conceptual model forecasting technology based on radar observation. It is difficult to make subjective judgments within a limited time. Therefore, it is urgent to develop automatic and objective disastrous gale forecasting and early warning technology.
[0004] Numerical forecasting is an important support for current weather forecasting and one of the main technologies for forecasting disastrous gale winds. However, numerical models are subject to many constraints such as initial fields and physical processes, and numerical forecast results usually have forecast deviations. In addition, various numerical models differ in many aspects such as dynamic frameworks, initial conditions, and parameterization schemes, resulting in certain differences in the forecast capabilities of different models in time and space. Statistical post-processing of numerical forecast results, such as bias correction for a single numerical model or ensemble averaging of multiple numerical forecast results, can generally eliminate forecast deviations and quantify forecast uncertainties, which is an effective means to further improve the level of numerical forecasting.
[0005] As early as the 1970s, studies have found that a specific linear combination of two independent prediction results can reduce the forecast error. Since then, methods such as simple ensemble average and bias-eliminated ensemble average have been widely used to reduce forecast bias and improve forecast accuracy. However, such equal-weighted multi-model integration methods usually assume that the model with poor forecasting ability is equivalent to or equivalent to the optimal model after eliminating the bias, so it is impossible to fully and reasonably utilize the effective forecast information of different models. For this superensemble model, the weight of each model is determined by multivariate regression least squares method, and then the weighted average of multiple models after eliminating the forecast bias is performed. Many studies have shown that the forecasting skills of unequal-weighted multi-model integration are generally better than those of equal-weighted multi-model integration and the optimal single model. However, there are few studies on the forecast of disastrous gale and the application of multi-model integration technology to improve the forecasting skills of the spatial structure characteristics of disastrous gale.
[0006] The weight assigned to each model in the multi-model integration model reflects the forecast performance of the model during the training period and has an important impact on the integrated forecast results. The existing multi-model integration technology mainly determines the weight allocation of multiple models participating in the integration by constructing the "point-to-point" error minimization in the spatial field as the objective function, which has strong statistical significance, but ignores the similarity of the spatial structural characteristics of meteorological elements in the observation field and the forecast field. Especially for localized disastrous gale, it is necessary to focus on the changes in the spatial characteristics of the location, shape, area, distribution, etc. of the gale area. The uncertainty of these spatial characteristics is more physically meaningful. Therefore, it is necessary to develop a multi-model integrated forecasting method that is suitable for disastrous gale. Summary of the invention
[0007] Purpose of the invention: The purpose of the present invention is to address the deficiencies of the prior art and to provide an improved method for forecasting the spatial structure of disastrous gale winds based on multi-model integration. The purpose is to develop a multi-model integration model adapted to disastrous gale winds by analyzing the spatial structure of disastrous gale winds and improving the calculation method of the weights of the multi-model integration model, thereby improving the numerical forecasting skills of the spatial structure of disastrous gale winds.
[0008] Technical solution: The improved method for forecasting the spatial structure of disastrous gale winds based on multi-mode integration of the present invention comprises the following steps:
[0009] S1. Obtain historical observation data and model forecast data in the forecast area, and identify the disastrous gale objects in the observation field and model forecast field respectively;
[0010] S2, respectively calculating the attributes of the disastrous gale objects in the identified observation field and model forecast field;
[0011] S3. Match the disastrous gale objects in the observation field with the disastrous gale objects in the model forecast field according to the calculated attributes of the disastrous gale objects, and evaluate the overall spatial similarity of all disastrous gale objects in the observation field and the model forecast field;
[0012] S4. Using the spatial similarity of all disastrous gale objects in the observation field and the model forecast field as weight, a multi-model integrated forecast model is constructed to form forecast products for disastrous gale forecasts.
[0013] A further preferred technical solution of the present invention is that the method for identifying the disastrous gale objects in the observation field and the model forecast field in step S1 is:
[0014] S11. The original 10 m wind speed field is spatially convolved with a convolution radius R to obtain a more continuous wind speed field. The convolution algorithm is:
[0015]
[0016] in, is the original wind speed field, is the convolution field, and are the grid coordinates;
[0017] S12, convolution field Perform threshold control to obtain the mask field , that is, only care about the wind speed intensity in the convolution field greater than or equal to the threshold For the area, the threshold control method is as follows:
[0018]
[0019] S13, assign the grid points in the continuous area of the mask field M=1 to the values of the corresponding grid points in the original 10 m wind speed field to obtain the reconstructed field , the reconstructed field not only retains the most original wind speed information of each object, but also identifies the disastrous high wind objects that reach the wind speed threshold and are worthy of attention, namely:
[0020] F(x,y)=M(x,y)f(x,y).
[0021] Preferably, the attributes of the disastrous gale object in step S2 at least include the area, centroid position, axis angle and aspect ratio of the disastrous gale object;
[0022] The area of a disastrous gale object is the number of grid points it covers;
[0023] The centroid position of the severe wind object The calculation method is:
[0024]
[0025] in, Indicates the identified disastrous wind object. represents the grid coordinates, Indicates the area of the disastrous gale object;
[0026] The calculation method of the axis angle of the disastrous gale object is: the line connecting the centroid position and the origin is taken as the main axis, and the counterclockwise angle between the main axis and the x-axis represents the axis angle of the disastrous gale object;
[0027] The method for calculating the aspect ratio of a disastrous gale object is: draw a minimum rectangle that can completely surround the disastrous gale object, with its major axis parallel to the main axis. The ratio of the major axis to the minor axis of the rectangle is the aspect ratio of the enclosed disastrous gale object.
[0028] Preferably, in step S3, the disastrous wind objects in the observation field and the disastrous wind objects in the model forecast field are matched according to the calculated attributes of the disastrous wind objects, and the specific method is:
[0029] Compare the centroid distances between the disastrous high wind objects in the observation field and the disastrous high wind objects in the model forecast field. When the centroid distance between the two objects is less than the average size of the two objects, the two objects are considered a pair of matching objects, that is:
[0030]
[0031] in, and are the observation object area of disastrous gale and the area of model forecast object, respectively, and D represents the centroid distance between the two objects.
[0032] Preferably, the overall spatial similarity of all disastrous gale objects in the evaluation observation field and the model forecast field described in step S3 is evaluated by using the maximum similarity median score or the object-based risk score method to evaluate the overall spatial similarity between the two;
[0033] Among them, the overall spatial similarity calculated by the maximum similarity median scoring method is: the median of the maximum total similarity between the disastrous gale objects in the observation field and the disastrous gale objects in the model forecast field;
[0034] The overall spatial similarity calculated by the object-based risk scoring method is: the inverse of the area of all matched objects and unmatched objects multiplied by the total similarity of the corresponding matched objects.
[0035] Preferably, the maximum similarity median score or the total similarity calculation method used in the object-based risk scoring method is:
[0036]
[0037] in, and are the confidence factor and weight of the i-th attribute respectively, is the similarity factor of the i-th attribute.
[0038] Preferably, step S4 uses the spatial similarity of all disastrous gale objects in the observation field and the model forecast field as weight to construct a multi-model integrated forecast model, and the specific calculation formula is:
[0039]
[0040] in, It represents the multi-model integrated forecast result obtained by weighted averaging multiple member models participating in the integration with the overall spatial similarity score as the weight. Indicates the total number of patterns involved in the integration, represents the forecast value of the ith model, Represents the weight of the mode, calculated as:
[0041]
[0042] in, represents the overall spatial similarity score of the ith pattern on the lth day of the training period, and L represents the total length of the training period.
[0043] Beneficial effects: The present invention improves the weight design of the multi-mode integrated model based on the spatial characteristics of disastrous gale. The method of the present invention can reflect the spatial structural characteristics of disastrous gale in the multi-mode integrated model. By analyzing the spatial structural characteristics of disastrous gale and improving the calculation method of the weight of the multi-mode integrated model, a multi-mode integrated model adapted to disastrous gale is developed, thereby improving the numerical forecasting skills of the spatial structure of disastrous gale. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] Figure 1 A flowchart of the improved method for spatial structure forecast of severe gale based on multi-model integration;
[0045] Figure 2 It is a graph showing the variation of MMI and OTS scores of multi-model integration and multiple single models with forecast time in the embodiment. DETAILED DESCRIPTION
[0046] The technical solution of the present invention is described in detail below with reference to the accompanying drawings, but the protection scope of the present invention is not limited to the embodiments.
[0047] Embodiment: A method for improving the spatial structure forecast of disastrous gale based on multi-mode integration comprises the following steps:
[0048] S1. Obtain historical observation data and model forecast data in the forecast area, and identify the disastrous gale objects in the observation field and model forecast field respectively;
[0049] S11. The original 10 m wind speed field is spatially convolved with a convolution radius R to obtain a more continuous wind speed field. The convolution algorithm is:
[0050]
[0051] in, is the original wind speed field, is the convolution field, and are the grid coordinates;
[0052] S12, convolution field Perform threshold control to obtain the mask field , that is, only care about the wind speed intensity in the convolution field greater than or equal to the threshold For the area, the threshold control method is as follows:
[0053]
[0054] S13, assign the grid points in the continuous area of the mask field M=1 to the values of the corresponding grid points in the original 10 m wind speed field to obtain the reconstructed field , the reconstructed field not only retains the most original wind speed information of each object, but also identifies the disastrous high wind objects that reach the wind speed threshold and are worthy of attention, namely:
[0055] F(x,y)=M(x,y)f(x,y).
[0056] S2. Calculate the attributes of the disastrous gale objects in the identified observation field and model forecast field respectively; including the area, centroid position, axis angle and aspect ratio of the disastrous gale objects; the object attributes can be used to match the target objects in the observation field and forecast field and calculate the spatial similarity of the two objects based on the attribute differences. The calculation method of each attribute is as follows:
[0057] The area of a disastrous gale object is the number of grid points it covers;
[0058] The centroid position of the severe wind object The calculation method is:
[0059]
[0060] in, Indicates the identified disastrous wind object. represents the grid coordinates, Indicates the area of the disastrous gale object;
[0061] The calculation method of the axis angle of the disastrous gale object is: the line connecting the centroid position and the origin is taken as the main axis, and the counterclockwise angle between the main axis and the x-axis represents the axis angle of the disastrous gale object;
[0062] The method for calculating the aspect ratio of a disastrous gale object is: draw a minimum rectangle that can completely surround the disastrous gale object, with its major axis parallel to the main axis. The ratio of the major axis to the minor axis of the rectangle is the aspect ratio of the enclosed disastrous gale object.
[0063] S3. Match the disastrous gale objects in the observation field with the disastrous gale objects in the model forecast field according to the calculated attributes of the disastrous gale objects, and evaluate the overall spatial similarity of all disastrous gale objects in the observation field and the model forecast field;
[0064] The specific method for matching the disastrous gale objects in the observation field with the disastrous gale objects in the model forecast field is as follows:
[0065] Compare the centroid distances between the disastrous high wind objects in the observation field and the disastrous high wind objects in the model forecast field. When the centroid distance between the two objects is less than the average size (square root of the area) of the two objects, the two objects are considered as a pair of matching objects, that is:
[0066]
[0067] in, and are the observation object area of disastrous gale and the area of model forecast object, respectively, and D represents the centroid distance between the two objects.
[0068] To evaluate the overall spatial similarity of all disastrous high wind objects in the observation field and the model forecast field, the Median of Maximum Interest (MMI) and the Object-based Threat Score (OTS) can be used to evaluate the overall spatial similarity of all disastrous high wind objects in the observation field and the model forecast field.
[0069] Among them, the overall spatial similarity calculated by the maximum similarity median scoring method is: the median of the maximum total similarity between the disastrous gale objects in the observation field and the disastrous gale objects in the model forecast field;
[0070] The overall spatial similarity calculated by the object-based risk scoring method is: the inverse of the area of all matched objects and unmatched objects multiplied by the total similarity of the corresponding matched objects.
[0071] Both scores are calculated based on the total similarity between different disastrous wind objects. The total similarity calculation method is:
[0072]
[0073] in, and are the confidence factor and weight of the i-th attribute respectively, is the similarity factor of the i-th attribute.
[0074] S4. Using the spatial similarity of all disastrous gale objects in the observation field and the model forecast field as weight, a multi-model integrated forecast model is constructed to form forecast products for disastrous gale forecasts.
[0075] The specific operation steps of the multi-model integrated forecasting model are as follows:
[0076]
[0077] in, It represents the multi-model integrated forecast result obtained by weighted averaging multiple member models participating in the integration with the overall spatial similarity score as the weight. Indicates the total number of patterns involved in the integration, represents the forecast value of the ith model, Represents the weight of the mode, calculated as:
[0078]
[0079] in, represents the overall spatial similarity score of the ith pattern on the lth day of the training period, and L represents the total length of the training period.
[0080] In this embodiment, the OTS score is used as the indicator for calculating the weight because the OTS score takes into account the area of the disastrous gale object and the number of matching object pairs, which is more comprehensive than the MMI score. In this embodiment, the China Meteorological Administration, the European Center for Medium-Range Weather Forecasts, the National Center for Environmental Prediction of the United States and the British Meteorological Office, which have relatively high numerical forecasting skills, are selected as member models participating in the multi-model integration.
[0081] S5. Perform spatial verification on the multi-model ensemble forecast results and compare them with the forecast results of multiple single models involved in the ensemble.
[0082] The forecast results of the multi-model integration model are spatially tested using the MMI and OTS evaluation indicators, and their test results are compared with those of multiple single models involved in the integration.
[0083] like Figure 2 As shown in the figure, the MMI and OTS scores of the multi-model integrated forecast results with the spatial characteristics of the disastrous wind object as weights and the four single-model forecast results involved in the integration are given as the forecast time changes. It can be seen that under the level 5, 6 and 7 gale thresholds, the MMI and OTS scores of the multi-model integrated forecast are better than all single models, which shows the advantages of the multi-model integrated forecast model with the spatial characteristics of the disastrous wind object as weights.
[0084] As described above, although the present invention has been shown and described with reference to specific preferred embodiments, it should not be construed as limiting the present invention itself. Various changes in form and details may be made without departing from the spirit and scope of the present invention as defined in the appended claims.
Claims
1. An improved method for forecasting the spatial structure of disastrous gale winds based on multi-model integration, characterized in that: The steps include: S1. Obtain historical observation data and model forecast data in the forecast area, and identify the disastrous gale objects in the observation field and model forecast field respectively; the specific method is: S11. The original 10 m wind speed field is spatially convolved with a convolution radius R to obtain a more continuous wind speed field. The convolution algorithm is: ; in, is the original wind speed field, is the convolution field, and are the grid coordinates; S12, convolution field Perform threshold control to obtain the mask field , that is, only care about the wind speed intensity in the convolution field greater than or equal to the threshold For the area, the threshold control method is as follows: ; S13, the mask field The grid points in the continuous area are assigned the values of the corresponding grid points in the original 10 m wind speed field to obtain the reconstructed field. , the reconstructed field not only retains the most original wind speed information of each object, but also identifies the disastrous high wind objects that reach the wind speed threshold and are worthy of attention, namely: ; S2, respectively calculating the attributes of the disastrous gale objects in the identified observation field and model forecast field; S3. Match the disastrous gale objects in the observation field with the disastrous gale objects in the model forecast field according to the calculated attributes of the disastrous gale objects, and evaluate the overall spatial similarity of all disastrous gale objects in the observation field and the model forecast field; S4. Using the spatial similarity of all disastrous gale objects in the observation field and the model forecast field as weight, a multi-model integrated forecast model is constructed to form forecast products for disastrous gale forecasts.
2. The improved method for forecasting the spatial structure of disastrous gale based on multi-mode integration according to claim 1 is characterized in that: The attributes of the disastrous gale object in step S2 at least include the area, centroid position, axis angle and aspect ratio of the disastrous gale object; The area of a disastrous gale object is the number of grid points it covers; The centroid position of the severe wind object The calculation method is: ; in, Indicates the identified disastrous wind object. represents the grid coordinates, Indicates the area of the disastrous gale object; The calculation method of the axis angle of the disastrous gale object is: the line connecting the centroid position and the origin is taken as the main axis, and the counterclockwise angle between the main axis and the x-axis represents the axis angle of the disastrous gale object; The method for calculating the aspect ratio of a disastrous gale object is: draw a minimum rectangle that can completely surround the disastrous gale object, with its major axis parallel to the main axis. The ratio of the major axis to the minor axis of the rectangle is the aspect ratio of the enclosed disastrous gale object.
3. The improved method for forecasting spatial structure of disastrous gale based on multi-mode integration according to claim 2 is characterized in that: According to the calculated attributes of the disastrous gale objects in step S3, the disastrous gale objects in the observation field and the disastrous gale objects in the model forecast field are matched, and the specific method is as follows: Compare the centroid distances between the disastrous high wind objects in the observation field and the disastrous high wind objects in the model forecast field. When the centroid distance between the two objects is less than the average size of the two objects, the two objects are considered a pair of matching objects, that is: ; in, and are the observation object area of disastrous gale and the area of model forecast object, respectively, and D represents the centroid distance between the two objects.
4. The improved method for forecasting spatial structure of disastrous gale based on multi-mode integration according to claim 3 is characterized in that: The overall spatial similarity of all disastrous gale objects in the evaluation observation field and the model forecast field described in step S3 is evaluated by using the maximum similarity median score or the object-based risk score method to evaluate the overall spatial similarity between the two; Among them, the overall spatial similarity calculated by the maximum similarity median scoring method is: the median of the maximum total similarity between the disastrous gale objects in the observation field and the disastrous gale objects in the model forecast field; The overall spatial similarity calculated by the object-based risk scoring method is: the inverse of the area of all matched objects and unmatched objects multiplied by the total similarity of the corresponding matched objects.
5. The improved method for forecasting spatial structure of disastrous gale based on multi-mode integration according to claim 4 is characterized in that: The maximum similarity median score or the total similarity calculation used in the object-based risk scoring method is: ; in, and are the confidence factor and weight of the i-th attribute respectively, is the similarity factor of the i-th attribute.
6. The improved method for forecasting spatial structure of disastrous gale based on multi-mode integration according to claim 4 is characterized in that: Step S4 uses the spatial similarity of all disastrous gale objects in the observation field and the model forecast field as weight to construct a multi-model integrated forecast model. The specific calculation formula is: ; in, It represents the multi-model integrated forecast result obtained by weighted averaging multiple member models participating in the integration with the overall spatial similarity score as the weight. Indicates the total number of patterns involved in the integration, represents the forecast value of the ith model, Represents the weight of the mode, calculated as: ; in, represents the overall spatial similarity score of the ith pattern on the lth day of the training period, and L represents the total length of the training period.
