A method and system for correcting model precipitation forecasts based on weather situation field classification

Abstract

The present invention discloses a model precipitation forecast correction method and system based on weather situation field classification. The method divides samples into flood season and non-flood season based on the location of the target area, and classifies the weather situation field types of the samples based on the forecast values ​​of the samples. Then, model precipitation forecast intensity correction models and model precipitation forecast location correction models for different weather situation field types under flood season and non-flood season are constructed respectively. Finally, based on the season and weather situation field corresponding to the real-time model precipitation forecast data, the corresponding model precipitation forecast intensity correction model and model precipitation forecast location correction model are selected to perform model precipitation forecast correction, thereby obtaining a more accurate model precipitation forecast correction result.

Description

Technical Field

[0001] The present invention belongs to the technical field of precipitation forecasting, and in particular relates to a method and system for correcting model precipitation forecasts based on weather situation field classification. Background Art

[0002] Quantitative precipitation forecasting (QPF) is considered one of the most challenging aspects of numerical model forecasting due to the uncertainty in its spatiotemporal distribution and intensity variations. In recent years, with the improvement of model spatial resolution, the enhanced data assimilation capabilities of initial and boundary fields, and the refinement of subgrid physical process parameterization schemes, the ability of QPF has been continuously enhanced. However, systematic errors in the spatiotemporal distribution and intensity of heavy precipitation forecasts based on different dynamical frameworks have also become increasingly apparent. Therefore, it is necessary to post-process and correct model precipitation forecast results.

[0003] Currently, corrections to model precipitation forecast products primarily occur in two ways: 1. Subjective correction. Forecasters first collect statistical data on weather patterns and meteorological field characteristics from historical heavy rainfall events, and then develop a correction strategy for the model precipitation forecast product. This method relies primarily on the forecaster's subjective experience and lacks a unified standard for identifying statistical weather patterns and meteorological field characteristics. 2. Objective correction. Statistical methods are used to fit the multi-factor forecast data output by the model with actual observations, constructing a forecast model for the relevant factors and generating objective precipitation forecasts. For example, the model output statistics method (MOS) constructs a regression equation based on specific forecast factors and precipitation at a station or grid point. However, this method can only perform simple bias corrections on precipitation forecasts and cannot correct for clear or rainy weather patterns. Heavy precipitation level forecasts based on the "batch method" (also known as the "component method") use the physical connection between the precipitation forecast and related indicators to establish corresponding forecast equations to predict the likelihood of heavy precipitation. Compared to the MOS method, this method breaks away from its simple reliance on regression analysis and possesses clear physical meaning. However, it is primarily used for qualitative assessment of heavy precipitation events and is difficult to quantitatively correct. In recent years, target score threshold correction methods, such as the frequency matching method (FM) and the optimal TS score correction method (OTS), have been gradually promoted and applied due to their advantages of enabling quantitative correction, requiring minimal computational resources, and operating stably. The frequency matching method (FM) corrects the model-based precipitation forecast curve using the frequency distribution curve of observed precipitation at various levels, thereby making the predicted precipitation distribution more consistent with the actual results and reducing forecast errors. The optimal TS score correction method (OTS) uses historical precipitation observations and model precipitation forecasts to construct a training model. It adjusts the model precipitation forecast to achieve the optimal TS score, thereby correcting for precipitation forecast bias. However, this type of method relies heavily on actual and model historical sample data for heavy rainfall forecasts. When the historical samples are too few, correction anomalies are prone to occur. Moreover, this method mainly corrects for precipitation intensity deviations, but does not adequately correct for precipitation area deviations. Summary of the Invention

[0004] In order to solve the problems existing in the prior art, the present invention proposes a precipitation forecast correction method and system based on weather situation field classification.

[0005] The technical solutions of the present invention are as follows:

[0006] A method for correcting model precipitation forecasts based on weather situation field classification includes:

[0007] Acquire historical model precipitation forecast data for a target area and historical actual precipitation data that matches the time of the historical model precipitation forecast data, and form a plurality of samples having observation values ​​corresponding to the historical actual precipitation data and forecast values ​​corresponding to the historical model precipitation forecast data;

[0008] Based on the location of the target area, the samples are divided into flood season and non-flood season, and based on the forecast values ​​of the samples, the weather situation field types of the samples are classified into weak precipitation type, typhoon-controlled type controlled by typhoon circulation, subtropical high-controlled type controlled by subtropical high pressure, trough-front type, trough-back type and trough-bottom type which are not controlled by typhoon circulation and subtropical high pressure and are divided by the different relative positions of the corresponding precipitation area and the upper-altitude shortwave trough;

[0009] Constructing model precipitation forecast intensity correction models and model precipitation forecast area correction models corresponding to different weather situation field types in flood season and non-flood season respectively, and using the observed values ​​of the samples as the model output values ​​and the predicted values ​​of the samples as the model input values, the model precipitation forecast intensity correction models and the model precipitation forecast area correction models are trained;

[0010] Based on the season and weather situation field corresponding to the real-time model precipitation forecast data, the corresponding trained model precipitation forecast intensity correction model and model precipitation forecast location correction model are selected to perform model precipitation forecast correction.

[0011] Furthermore, the classification method of the weather situation field types into the weak precipitation type, the typhoon-controlled type and the subtropical high-controlled type includes:

[0012] For a sample at a certain moment, the target area is gridded at a set resolution. The model precipitation forecast data in the sample is interpolated to the meteorological stations in the target area. The number of stations in each grid whose model precipitation forecast is greater than or equal to the first precipitation threshold is counted. If the number of stations corresponding to each grid is 0, the weather situation field type of the sample at that moment is weak precipitation type.

[0013] If the number of stations is not all 0, determine whether the typhoon center position in the typhoon message corresponding to the moment is located in the target area. If so, the weather situation field type of the sample at this moment is typhoon-controlled; otherwise, based on the number of stations corresponding to the target area grid, screen out the precipitation center grid of the target area, calculate the average longitude and latitude of all stations in the precipitation center grid whose model precipitation forecast is greater than or equal to the first precipitation threshold, and use the average longitude and latitude as the precipitation center position of the model precipitation forecast at this moment, and calculate the average potential height forecast field of the sample at this moment with the subtropical high characteristic line and the subtropical high peripheral characteristic line; if the precipitation center position of the sample is within the range of the subtropical high characteristic line, the weather situation field type of the sample at this moment is subtropical high-controlled.

[0014] Furthermore, the specific steps of screening out the precipitation center grid of the target area based on the number of stations corresponding to the grids in the target area include: dividing the target area into precipitation areas, and for each precipitation area, obtaining the number of heavy precipitation adjacent grids corresponding to each grid therein, wherein the heavy precipitation adjacent grids are grids corresponding to grids adjacent to each grid, wherein the number of stations whose model precipitation forecasts are greater than or equal to a first precipitation threshold is greater than 0; using the grid with the highest number of heavy precipitation adjacent grids as the precipitation center grid of each precipitation area; using the precipitation center grid with the highest number of heavy precipitation adjacent grids in different precipitation areas as the precipitation center grid of the target area; and in the case where there are multiple grids with the highest and consistent numbers of heavy precipitation adjacent grids, calculating the total precipitation forecast in each grid, and taking the grid with the largest total precipitation forecast as the precipitation center grid;

[0015] The specific method for dividing the target area into precipitation zones includes: obtaining all grids in the target area whose corresponding station numbers are not 0, calculating the number of interval grids between any two grids in all the obtained grids, and dividing two grids whose interval grid number is less than or equal to 1 into the same precipitation zone.

[0016] Furthermore, the classification method of the weather situation field types into trough front type, trough back type and trough bottom type includes:

[0017] For samples whose weather situation field is not classified as weak precipitation type, typhoon controlled type controlled by typhoon circulation, or subtropical high controlled type controlled by subtropical high pressure, the latitudinal deviation of the mean geopotential height of the sample at that moment is calculated and compared with the latitudinal deviation threshold of the mean geopotential height. If the latitudinal deviation of the mean geopotential height is less than the latitudinal deviation threshold of the mean geopotential height, the weather situation field type of the sample at that moment is the trough front type; otherwise, the latitude difference between the first intersection point of the subtropical high outer characteristic line in the mean geopotential height forecast field of the sample and the first judgment meridian and the second intersection point of the second judgment meridian is determined and compared with the latitude difference threshold. If the latitude difference is greater than the latitude difference threshold, the weather situation field type of the sample at that moment is the trough back type; otherwise, the weather situation field type of the sample at that moment is the trough bottom type.

[0018] Furthermore, the formula for the latitudinal deviation of the mean geopotential height is as follows:

[0019]

[0020] Where, is the latitudinal deviation of the average potential height corresponding to longitude lon_i and latitude lat_j, lon_i is longitude, lon_i∈[lon_i_min,lon_i_max], lat_j is latitude, lat_j∈[lat_j_min,lat_j_max]; lon_i_min is the first judgment longitude; lon_i_max is the second judgment longitude; lat_j_min is the first judgment latitude; lat_j_max is the second judgment latitude;

[0021] is the average geopotential height corresponding to longitude lon_i and latitude lat_j;

[0022] is the average value of the mean geopotential height corresponding to latitude lat_j at different longitudes.

[0023] Furthermore, the specific method for constructing and training the model for correcting the precipitation forecast intensity of different weather situation fields during the flood season and the non-flood season includes:

[0024] The model precipitation forecast correction models of typhoon-controlled type, subtropical high-controlled type, trough-front type, trough-back type and trough-bottom type in the flood season, and the model precipitation forecast correction models of subtropical high-controlled type, trough-front type, trough-back type and trough-bottom type in the non-flood season are respectively constructed; the subtropical high-controlled type, trough-front type and trough-back type in the flood season and non-flood season adopt the scaling coefficient correction model; the typhoon-controlled type and trough-bottom type in the flood season adopt the optimal TB score correction model; the trough-bottom type in the non-flood season adopts the scaling coefficient correction model;

[0025] The corresponding model precipitation forecast correction model is trained by taking the observed precipitation of samples of typhoon-controlled type, subtropical high-controlled type, trough-front type, trough-back type and trough-bottom type during the flood season as the model output value and taking the precipitation of the predicted value of the samples as the model input value. The corresponding model precipitation forecast correction model is trained by taking the observed precipitation of samples of subtropical high-controlled type, trough-front type, trough-back type and trough-bottom type during the non-flood season as the model output value and taking the precipitation of the predicted value of the samples as the model input value.

[0026] Furthermore, the expression of the scaling coefficient correction model is:

[0027] y=xF

[0028] Where y is the corrected value of the model precipitation forecast, i.e., the output of the scaling factor correction model; x is the original value of the model precipitation forecast, i.e., the input of the scaling factor correction model; F is the scaling factor;

[0029] The expression of the optimal TB score correction model is:

[0030]

[0031] Where, O k is the actual precipitation level of the kth level; O k+1 is the actual precipitation level of the k+1th level; E1 is the threshold of the predicted precipitation level of the first level; E k is the k-th level forecast precipitation threshold; E k+1 is the k+1th level forecast precipitation level threshold; E n is the forecast precipitation level threshold of level n; k∈[1,n-1], n is the total number of precipitation levels;

[0032] The optimal TB score correction model is trained by adjusting the k-th level forecast precipitation level threshold E k The TB score of the actual precipitation level of level k and above is maximized. The calculation formula of the TB score is as follows:

[0033] TB=TS-B

[0034]

[0035]

[0036] Where TS is the TS score of the model forecast for precipitation above a certain magnitude, 0≤TS≤1; B is the BIAS deviation of the model forecast for precipitation above a certain magnitude, B≥0; NA is the station with a correct forecast; NB is the station with an empty forecast; and NC is the station with a missed forecast.

[0037] Furthermore, the specific method for constructing and training the model for correcting the precipitation forecast area of ​​different weather situation fields in the flood season and the non-flood season includes:

[0038] The model precipitation forecast area correction models for the weather situation field types of trough front, trough back and trough bottom in flood season and non-flood season are constructed respectively. The latitude difference between the center of the model precipitation position corresponding to the sample forecast value and the actual precipitation position corresponding to the observation value is used as the model output value, and the latitude difference between the model precipitation forecast position corresponding to the sample forecast value and the outer characteristic line of the subtropical high is used as the model input value to train the corresponding model precipitation forecast correction model.

[0039] Furthermore, the model precipitation forecast area correction model adopts a univariate linear regression model, which is expressed as follows:

[0040] Y=a×X+b

[0041] where X is the latitude difference between the model precipitation center and the outer characteristic line of the subtropical high; Y is the latitude difference between the model precipitation center and the actual precipitation center; a is the slope of the univariate linear regression model; and b is the intercept of the univariate linear regression model.

[0042] Furthermore, the specific steps of obtaining historical model precipitation forecast data of the target area and historical actual precipitation data that matches the time of the historical model precipitation forecast data, and forming a plurality of samples having observation values ​​corresponding to the historical actual precipitation data and forecast values ​​corresponding to the historical model precipitation forecast data include:

[0043] A model precipitation forecast classification database is constructed, and the historical model precipitation forecast data of the target area and the historical actual precipitation data that matches the time of the historical model precipitation forecast data are stored in the database. Each sample in the database includes the observation value of the corresponding historical actual precipitation data and the forecast value of the corresponding historical model precipitation forecast data.

[0044] A model precipitation forecast correction system based on weather situation field classification includes a sample acquisition module, a sample division module, a model construction module and a correction module;

[0045] The sample acquisition module is used to acquire historical model precipitation forecast data of the target area and historical actual precipitation data that matches the time of the historical model precipitation forecast data, and form a plurality of samples having observation values ​​corresponding to the historical actual precipitation data and forecast values ​​corresponding to the historical model precipitation forecast data;

[0046] The sample classification module is used to classify the samples into flood season and non-flood season based on the location of the target area, and classify the samples into weak precipitation type, typhoon-controlled type controlled by typhoon circulation, subtropical high-controlled type controlled by subtropical high pressure, trough-front type, trough-back type and trough-bottom type that are not controlled by typhoon circulation and subtropical high pressure and have different latitudinal deviations of mean geopotential height based on the forecast values ​​of the samples;

[0047] The model construction module is used to respectively construct a model precipitation forecast intensity correction model and a model precipitation forecast area correction model corresponding to different weather situation field types in the flood season and the non-flood season, and train the model precipitation forecast intensity correction model and the model precipitation forecast area correction model using the observed value of the sample as the model output value and the predicted value of the sample as the model output value;

[0048] The correction module is used to select the corresponding trained model precipitation forecast intensity correction model and model precipitation forecast area correction model to perform model precipitation forecast correction based on the season and weather situation field corresponding to the real-time model precipitation forecast data.

[0049] An electronic device comprises a memory and a processor, wherein the memory stores a computer program, and the processor is configured to call and run the computer program stored in the memory to execute any of the above methods.

[0050] A computer-readable storage medium stores a computer program, wherein the computer program implements the steps of any of the above methods when executed by a processor.

[0051] Compared with the prior art, the present invention has the following beneficial effects:

[0052] The present invention proposes a model precipitation forecast correction method and system based on weather situation field classification. The method divides samples into flood season and non-flood season based on the location of the target area, and classifies the weather situation field types of the samples based on the forecast values ​​of the samples. Then, model precipitation forecast intensity correction models and model precipitation forecast location correction models for different weather situation field types under flood season and non-flood season are constructed respectively. Finally, based on the season and weather situation field corresponding to the real-time model precipitation forecast data, the corresponding model precipitation forecast intensity correction model and model precipitation forecast location correction model are selected to perform model precipitation forecast correction, thereby obtaining a more accurate model precipitation forecast correction result.

[0053] In classifying the weather situation field types of samples according to the method of the present invention, the method first divides the samples into weak precipitation type and "heavy precipitation type", and then divides the samples belonging to the "heavy precipitation type" into typhoon-controlled type, subtropical high-controlled type, trough-front type, trough-back type and trough-bottom type in turn. Since the samples of weak precipitation type are eliminated in advance, this design is equivalent to eliminating noise interference in the samples, so the subsequent calculation of the average potential height forecast field is more accurate, thereby making the classification of the weather situation field of the present invention more accurate.

[0054] The method of the present invention adopts the optimal TB score correction model to correct the model precipitation forecast. Compared with the traditional correction method based on TS score, the present invention takes into account the forecast BIAS deviation amplitude, reduces the impact of false alarms and missed reports on the forecast precipitation level threshold of the traditional method, and makes the precipitation level deviation smaller and closer to the observation results.

[0055] The input of the model precipitation forecast fall area correction model in the method of the present invention is the latitude difference between the model precipitation center position and the outer characteristic line of the subtropical high, and the output is the latitude difference between the model precipitation center position and the actual precipitation center position. This design introduces the outer characteristic line of the subtropical high to make the fall area correction easier to implement, avoiding the problem that the existing method directly uses the model precipitation center position as input and the actual precipitation center position as output, which makes the model difficult to converge.

[0056] The method of the present invention adopts a model precipitation forecast area correction model to correct the precipitation area. Compared with the existing method of subjectively correcting the precipitation area based on manual experience, the model precipitation forecast area correction model adopted by the method of the present invention is based on historical data and constructed using data fitting ideas, and is therefore more objective and scientific. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 Flowchart of the method for correcting model precipitation forecast based on weather situation field classification in an embodiment;

[0058] Figure 2 An example diagram of the precipitation center grid selection in the embodiment;

[0059] Figure 3 The figure is a scatter plot of the latitude difference between the model precipitation center and the 584 line (abscissa) and the latitude difference between the model rainstorm center and the actual center (ordinate) at the model start time 08:00 in the embodiment;

[0060] Figure 4 The figure is a scatter plot of the latitude difference between the model precipitation center and the 584 line (abscissa) and the latitude difference between the model rainstorm center and the actual center (ordinate) at the model start time 08:00 and 20:00 in the embodiment;

[0061] Figure 5 The figure is a scatter plot of the latitude difference between the model precipitation center and the 584 line (abscissa) and the latitude difference between the model rainstorm center and the actual center (ordinate) at the model start time 20:00 in the embodiment;

[0062] Figure 6 The figure is a scatter plot of the latitude difference between the model precipitation center and the 584 line (abscissa) and the latitude difference between the model rainstorm center and the actual center (ordinate) at the model start time 20:00 in the embodiment;

[0063] Figure 7 Schematic diagram of solving the scaling factor F in the embodiment;

[0064] Figure 8 Schematic diagram for comparing test results in the examples. DETAILED DESCRIPTION

[0065] The present invention is further illustrated below with reference to the accompanying drawings and specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention. After reading the present invention, various equivalent modifications of the present invention by those skilled in the art fall within the scope defined by the claims attached to this application.

[0066] Example 1:

[0067] The present invention is a method for correcting precipitation forecast based on weather situation field classification, such as Figure 1 As shown, the specific steps are as follows:

[0068] S1. Obtain historical model precipitation forecast data for a target area and historical actual precipitation data that matches the time of the historical model precipitation forecast data, and form a plurality of samples having observation values ​​corresponding to the historical actual precipitation data and forecast values ​​corresponding to the historical model precipitation forecast data;

[0069] S2. Based on the location of the target area, the samples are divided into flood season and non-flood season. Based on the forecast values ​​of the samples, several samples are classified into the following weather situation types: weak precipitation type, typhoon-controlled type controlled by typhoon circulation, subtropical high-controlled type controlled by subtropical high pressure, trough-front type, trough-back type, and trough-bottom type, which are not controlled by typhoon circulation and subtropical high pressure and are divided according to the different relative positions of the corresponding precipitation area and the upper-altitude shortwave trough;

[0070] S3. Constructing model precipitation forecast intensity correction models and model precipitation forecast area correction models corresponding to different weather situation field types in flood season and non-flood season, respectively, and using the observed values ​​of the samples as the model output values ​​and the predicted values ​​of the samples as the model input values, to train the model precipitation forecast intensity correction models and the model precipitation forecast area correction models;

[0071] S4. Based on the season corresponding to the real-time model precipitation forecast data, the samples are judged as being in the flood season or the non-flood season. Based on the weather situation field corresponding to the real-time model precipitation forecast data, the trained corresponding model precipitation forecast intensity correction model and model precipitation forecast fall area correction model are selected to perform model precipitation forecast correction.

[0072] Example 2:

[0073] This embodiment is further designed on the basis of the first embodiment in that the classification method of the weather situation field type in this embodiment includes:

[0074] For samples at a certain moment, the target area is gridded with a set resolution, and the model precipitation forecast data in the sample is interpolated to the meteorological station in the target area. The meteorological station can be a "national station", and the number of stations in each grid whose model precipitation forecast is greater than or equal to the first precipitation threshold is counted; if the number of stations corresponding to each grid is 0, then the weather situation field type of the sample at that moment is weak precipitation type; the above resolution can be 0.5°×0.5° or other resolutions. The above first precipitation threshold can be 3mm. If the number of stations is not all 0, it is a "heavy precipitation type", which includes typhoon control type, subtropical high control type, trough front type, trough back type and trough bottom type. The specific division method is as follows:

[0075] If the number of stations is not all 0, determine whether the typhoon center position in the typhoon message corresponding to the moment is located in the target area. If so, the weather situation field type of the sample at that moment is typhoon-controlled; otherwise, based on the number of stations corresponding to the target area grid, screen out the precipitation center grid of the target area, calculate the average longitude and latitude of all stations in the precipitation center grid whose model precipitation forecast is greater than or equal to the first precipitation threshold, and use the average longitude and latitude as the precipitation center position of the model precipitation forecast at that moment, and calculate the average geopotential height forecast field with the subtropical high characteristic line and the subtropical high peripheral characteristic line of the sample at that moment; if the precipitation center position of the sample is within the range of the subtropical high characteristic line, then the weather situation field type of the sample at that moment is subtropical high-controlled.

[0076] Furthermore, the above-mentioned subtropical high characteristic line can be the 588dagpm line, and the typhoon message can be the typhoon message issued by the Central Meteorological Observatory.

[0077] Example 3:

[0078] This embodiment is further designed on the basis of the second embodiment in that, in this embodiment, the specific steps of selecting the precipitation center grid of the target area based on the number of stations corresponding to the grids in the target area include: dividing the target area into precipitation areas, and for each precipitation area, obtaining the number of heavy precipitation adjacent grids corresponding to each grid therein, wherein the heavy precipitation adjacent grids are grids corresponding to grids adjacent to each grid where the number of stations having model precipitation forecasts greater than or equal to a first precipitation threshold is greater than 0; using the grid with the highest number of heavy precipitation adjacent grids as the precipitation center grid of each precipitation area; using the precipitation center grid with the highest number of heavy precipitation adjacent grids in different precipitation areas as the precipitation center grid of the target area; if there are multiple grids with the highest and consistent numbers of heavy precipitation adjacent grids, calculating the total precipitation forecast in each grid, and taking the grid with the largest total precipitation forecast as the precipitation center grid;

[0079] The specific method for dividing the target area into precipitation zones includes: obtaining all grids in the target area whose corresponding station numbers are not 0, calculating the number of grids between any two grids in all the obtained grids, and dividing two grids whose grid number is less than or equal to 1 into the same precipitation zone;

[0080] The following is an example to illustrate the above process. Figure 2 As shown in the figure, there are three precipitation areas in this example, namely the yellow area, the green area and the orange area. There are 8 grids adjacent to heavy precipitation in the yellow area and the green area. Therefore, the total amount of precipitation forecast in these grids should be calculated separately, and the grid with the largest total amount of precipitation forecast is taken as the precipitation center grid.

[0081] Example 4:

[0082] This embodiment is further designed on the basis of the second embodiment in that the classification method of the weather situation field type in this embodiment into the trough front type, the trough back type and the trough bottom type includes:

[0083] For samples whose weather situation field is not classified as a weak precipitation type, a typhoon-controlled type (controlled by typhoon circulation), or a subtropical high-controlled type (controlled by subtropical high pressure), the latitudinal deviation of the mean geopotential height at that moment is calculated and compared with the latitudinal deviation threshold of the mean geopotential height. If the latitudinal deviation of the mean geopotential height is less than the latitudinal deviation threshold of the mean geopotential height, the weather situation field type of the sample at that moment is a trough-front type. Otherwise, the latitude difference between the first intersection of the subtropical high outer characteristic line in the sample's mean geopotential height forecast field with the first judgment meridian and the second intersection with the second judgment meridian is determined and compared with the latitude difference threshold. If the latitude difference is greater than the latitude difference threshold, the weather situation field type of the sample at that moment is a trough-back type. Otherwise, the weather situation field type of the sample at that moment is a trough-base type. The latitudinal deviation threshold of the mean geopotential height is 0, and the latitude difference threshold is 0.5°.

[0084] Embodiment 5:

[0085] This embodiment is further designed on the basis of the fourth embodiment in that the formula for the latitudinal deviation of the mean geopotential height is as follows:

[0086]

[0087] Where, is the latitudinal deviation of the average potential height corresponding to longitude lon_i and latitude lat_j, lon_i is longitude, lon_i∈[lon_i_min,lon_i_max], lat_j is latitude, lat_j∈[lat_j_min,lat_j_max]; lon_i_min is the first judgment longitude; lon_i_max is the second judgment longitude; lat_j_min is the first judgment latitude; lat_j_max is the second judgment latitude;

[0088] is the average geopotential height corresponding to longitude lon_i and latitude lat_j;

[0089] is the average value of the mean geopotential height corresponding to latitude lat_j at different longitudes.

[0090] Example 6:

[0091] This embodiment is further designed based on the first embodiment in that the specific method of constructing and training the model for the precipitation forecast intensity correction of different weather situation field types in the flood season and the non-flood season in this embodiment includes:

[0092] Correction models for model precipitation forecasts of typhoon-controlled, subtropical high-controlled, trough-front, trough-back, and trough-bottom types during the flood season, and for subtropical high-controlled, trough-front, trough-back, and trough-bottom types during the non-flood season, were constructed. Scaling coefficient correction models were used for subtropical high-controlled, trough-front, and trough-back types during the flood season and non-flood season. The optimal TB score correction model was used for typhoon-controlled and trough-bottom types during the flood season. The scaling coefficient correction model was used for trough-bottom types during the non-flood season. No corrections were made for weak precipitation types during the flood season and non-flood season, and for typhoon types during the non-flood season. The correction models for the flood season and non-flood season are shown in Tables 2 and 3, respectively.

[0093] Table 2

[0094]

[0095] Table 3

[0096]

[0097]

[0098] The test levels in Tables 2 and 3 can be adjusted according to the timeliness of precipitation forecasts.

[0099] The corresponding model precipitation forecast correction model is trained by taking the observed precipitation of samples of typhoon-controlled type, subtropical high-controlled type, trough-front type, trough-back type and trough-bottom type during the flood season as the model output value and taking the precipitation of the predicted value of the samples as the model input value. The corresponding model precipitation forecast correction model is trained by taking the observed precipitation of samples of subtropical high-controlled type, trough-front type, trough-back type and trough-bottom type during the non-flood season as the model output value and taking the precipitation of the predicted value of the samples as the model input value.

[0100] Embodiment seven:

[0101] This embodiment is further designed based on the sixth embodiment in that the expression of the scaling coefficient correction model in this embodiment is:

[0102] y=xF

[0103] Where y is the correction value of the model precipitation forecast, i.e., the output of the scaling factor correction model; x is the original value of the model precipitation forecast, i.e., the input of the scaling factor correction model; F is the scaling factor. The goal of this design is to minimize the BIAS deviation of the model precipitation forecast above a certain level by adjusting the scaling factor F during the sample training period. For details, see Figure 7 .

[0104] The expression of the optimal TB score correction model is:

[0105]

[0106] Where, O k is the actual precipitation level of the kth level; O k+1 is the actual precipitation level of the k+1th level; E1 is the threshold of the predicted precipitation level of the first level; E k is the k-th level forecast precipitation threshold; E k+1 is the k+1th level forecast precipitation level threshold; E n is the forecast precipitation level threshold of level n; k∈[1,n-1], n is the total number of precipitation levels;

[0107] The optimal TB score correction model is trained by adjusting the k-th level forecast precipitation level threshold E k The TB score of the actual precipitation level of level k and above is maximized. The TB score is used to evaluate the model's forecasting ability for precipitation above a certain level. The specific calculation formula is as follows:

[0108] TB=TS-B

[0109]

[0110]

[0111] Where TS is the model's TS score for precipitation above a certain magnitude, with a value of 0 ≤ TS ≤ 1. A higher score indicates a higher percentage of accurate model predictions. B is the model's BIAS bias for precipitation above a certain magnitude, with a value of B ≥ 0. A smaller bias indicates a closer match between the model's predicted precipitation range and the actual precipitation. NA is the number of stations correctly predicted; NB is the number of stations with missed predictions; and NC is the number of stations with missed predictions. As the formula shows, a higher TB score indicates a higher percentage of accurate model precipitation forecasts, a closer match between the model's predicted precipitation range and the actual precipitation, and a stronger overall forecast capability.

[0112] Compared with the OTS algorithm, the method of the present invention can maximize the number of accurate model precipitation forecasts NA while controlling its precipitation forecast range to a certain extent, thereby avoiding the rapid growth of the number of false alarms NB or the number of missed alarms NC that may be caused in the process.

[0113] Embodiment 8:

[0114] This embodiment is further designed on the basis of the first embodiment in that the specific method for constructing and training the model for correcting the precipitation forecast area under different weather situation types in the flood season and the non-flood season includes:

[0115] The model precipitation forecast area correction models for the weather situation field types of trough front, trough back and trough bottom in flood season and non-flood season are constructed respectively. The latitude difference between the center of the model precipitation position corresponding to the sample forecast value and the actual precipitation position corresponding to the observation value is used as the model output value, and the latitude difference between the model precipitation forecast position corresponding to the sample forecast value and the outer characteristic line of the subtropical high is used as the model input value to train the corresponding model precipitation forecast correction model.

[0116] Embodiment 9:

[0117] This embodiment is further designed on the basis of the first embodiment in that the model precipitation forecast area correction model in this embodiment adopts a univariate linear regression model, which is expressed as follows:

[0118] Y=a×X+b

[0119] Where X is the latitude difference between the model precipitation center and the outer characteristic line of the subtropical high; Y is the latitude difference between the model precipitation center and the actual precipitation center. If Y > 0, it means that the model precipitation center is located to the north of the actual precipitation center; otherwise, it means that it is located to the south; a is the slope of the univariate linear regression model; b is the intercept of the univariate linear regression model. Both the slope and intercept are constants.

[0120] Embodiment 10:

[0121] This embodiment is further designed based on the first embodiment. In this embodiment, the specific steps of obtaining historical model precipitation forecast data of the target area and historical actual precipitation data that matches the time of the historical model precipitation forecast data, and forming a plurality of samples having observation values ​​corresponding to the historical actual precipitation data and forecast values ​​corresponding to the historical model precipitation forecast data include:

[0122] A model precipitation forecast classification database is constructed, and the historical model precipitation forecast data of the target area and the historical actual precipitation data that matches the time of the historical model precipitation forecast data are stored in the database. Each sample in the database includes the observation value corresponding to the historical actual precipitation data and the forecast value corresponding to the historical model precipitation forecast data.

[0123] Existing methods can be used to classify samples into flood and non-flood seasons based on the location of the target area. For Anhui Province, this classification can be based on 30 years of climate data compiled for the province (1971-2000). Heavy rain can occur in all months of the year, but it (daily precipitation ≥ 50 mm) is primarily concentrated between May and August, with a particularly high incidence in June and July. Heavy rain (daily precipitation ≥ 100 mm) has a similar distribution pattern to that of heavy rain, with June and July also being the most frequent months. Winter heavy rain never occurs. Extremely heavy rain (daily precipitation ≥ 250 mm) occurs only in the summer (June-August), with June and July still being the most frequent. Looking at the frequency of heavy rain in Anhui Province by month, over 50% of heavy rainstorms occur in June-July, and over 80% occur in May-August. Therefore, June-August is considered the flood season, and the remaining months are considered the non-flood season. Other existing methods can also be used for division, such as the method recorded in "Basics and Practice of Anhui Weather Forecast Business".

[0124] Example 11:

[0125] The present invention provides a model precipitation forecast correction system based on weather situation field classification, comprising a sample acquisition module, a sample division module, a model construction module and a correction module;

[0126] A sample acquisition module is used to obtain historical model precipitation forecast data of the target area and historical actual precipitation data that matches the time of the historical model precipitation forecast data, and form a number of samples having observation values ​​corresponding to the historical actual precipitation data and forecast values ​​corresponding to the historical model precipitation forecast data;

[0127] The sample classification module is used to classify samples into flood season and non-flood season based on the location of the target area, and based on the forecast values ​​of several samples, classify the weather situation field types of the samples into weak precipitation type, typhoon-controlled type controlled by typhoon circulation, subtropical high-controlled type controlled by subtropical high pressure, trough-front type, trough-back type and trough-bottom type that are not controlled by typhoon circulation and subtropical high pressure and have different latitudinal deviations of mean geopotential height;

[0128] The model construction module is used to construct the model precipitation forecast intensity correction model and the model precipitation forecast area correction model corresponding to different weather situation field types in the flood season and the non-flood season, and train the model precipitation forecast intensity correction model and the model precipitation forecast area correction model using the sample observation value as the model output value and the sample forecast value as the model output value;

[0129] The correction module is used to select the corresponding trained model precipitation forecast intensity correction model and model precipitation forecast area correction model to correct the model precipitation forecast based on the season and weather situation field corresponding to the real-time model precipitation forecast data.

[0130] Example 12:

[0131] An electronic device includes a memory and a processor, wherein the memory stores a computer program, and the processor is used to call and run the computer program stored in the memory to execute the method described in any of the above embodiments.

[0132] A computer-readable storage medium stores a computer program, which implements the steps of the method described in any of the above embodiments when executed by a processor.

[0133] Application examples:

[0134] This example takes Anhui Province and its surrounding areas (114.7°-119.7°E, 29°-35°N) as the target area. The 3-hour precipitation forecast for forecast hours 015-036 in the ECMWF model for June-August 2024 in this target area is revised. The specific process is as follows:

[0135] First, based on the objective identification algorithm of precipitation weather situation field, the ECMWF model 3-h precipitation forecast classification database is constructed as follows:

[0136] Step 101, objectively determining heavy and light precipitation and the precipitation center, is as follows:

[0137] First, the target area was divided into a 10 (longitudinal) x 12 (latitudinal) grid using a 0.5° x 0.5° grid. The ECMWF model's 3-hour precipitation forecast was interpolated onto all national stations within the region. The number of stations within each grid with a model 3-hour precipitation forecast ≥ 3 mm was then calculated. If the number of stations with a model 3-hour precipitation forecast ≥ 3 mm across all grids was zero, the ECMWF model's 3-hour precipitation forecast for Anhui Province and its surrounding areas at that moment was considered "weak precipitation." Otherwise, a "heavy precipitation" forecast was considered. Next, the precipitation center grid and location for the region were obtained using the method of the present invention.

[0138] Step 102, objective identification of the precipitation weather situation field, is as follows:

[0139] Count all the times when "heavy rainfall" was forecasted from 015 to 036, and calculate the ECMWF model 500hPa average geopotential height forecast field And the average position of the precipitation center. Then, determine whether the precipitation process is controlled by typhoon circulation or subtropical high pressure. Taking Anhui Province and its surrounding areas (114.7°-119.7°E, 29°-35°N) as an example, the method for determining whether it is controlled by typhoon circulation is: According to the typhoon message issued by the Central Meteorological Observatory every day, if the 24-hour forecast typhoon center position is within the range of 113°-123°E, 25°-37°N, it is determined that the precipitation process belongs to "typhoon type". The method for determining whether it is controlled by subtropical high pressure: Based on the 500hPa average potential height field If the average position of the precipitation center is within the range of the 588dagpm line (hereinafter referred to as the 588 line) and is not affected by the typhoon circulation, it is determined to be "subtropical high-controlled type". Finally, for heavy precipitation processes that are not "typhoon-type" or "subtropical high-controlled type", based on the 500hPa average potential height field The weather situation field is divided into "trough front type", "trough back type" and "trough bottom type". The "trough front type" is determined by: the latitudinal deviation of the average geopotential height in the area (108°-120°E, 35°-38°N) and the latitudinal deviation of the mean geopotential height at (108°-115°E, 30°-33°N) or (108°-115°E, 33°-35°N) The formula for the latitudinal deviation of mean geopotential height is as follows:

[0140]

[0141] Where lon_i is the longitude (range: 90°-140°E), and lat_j is the latitude (range: 26°-38°N). is the mean potential height field Averaged over latitude lat_j between 90° and 140°E.

[0142] The method for determining the "trough-back type" is: the latitude difference between line 584 at 115°E and 120°E is greater than 0.5°.

[0143] The method for determining "groove bottom type" is: except for the above two categories, the rest are classified as groove bottom type.

[0144] Step 103: Construct the ECMWF model 3-hour precipitation forecast classification database, as follows:

[0145] Using the 3-hour multi-factor forecast data from 08:00 and 20:00 daily reported by the ECMWF global model from January 1, 2017, a 3-hour precipitation forecast classification database was constructed, including "typhoon type", "subtropical high controlled type", "trough front type", "trough back type", "trough bottom type" and "weak precipitation type".

[0146] Second, the ECMWF model precipitation forecast area has been revised based on different weather conditions, as follows:

[0147] Numerous previous studies have shown a correlation between the location of heavy rain during the Meiyu period in the Jianghuai region and the 500hPa geopotential height. Most heavy rain areas are concentrated between 582 and 586 dagpm, with the main rain belt located near the 584 line. This example uses the position of the 584 line in the ECMWF model's 500hPa geopotential height field as a reference to correct precipitation forecasts for heavy rain events under different weather conditions. The specific steps are as follows:

[0148] Step 201: Using the ECMWF model 24h cumulative precipitation forecast data (015-036) and the corresponding actual precipitation observation data from June to August 2011 to 2023, the method of determining whether there is a heavy precipitation forecast in the model for 3 hours is used to count all rainstorm processes in Anhui Province and its surrounding areas (24h cumulative precipitation forecast ≥ 50mm), and calculate the predicted value Lat_m and observed value Lat_o of the center latitude of the rainstorm, as well as the ECMWF model 500hPa average potential height field. The latitude forecast value of line 584 within the range of 115°-120°E is Lat_584.

[0149] Step 202, classify the rainstorm process according to different weather conditions and model start time, and analyze the relationship between the latitude forecast value Lat_m of the rainstorm center, the observed value Lat_o and the latitude forecast value Lat_584 of the 584th line. The results show that: for the "trough-front" and "trough-bottom" rainstorm processes reported by the model at 08:00, and the "trough-front", "trough-back" and "trough-bottom" rainstorm processes reported by the model at 20:00, there is a univariate linear regression relationship between the latitude difference between the model rainstorm center and the 584th line (Lat_m-Lat_584) and the latitude difference between the model rainstorm center and the actual center (Lat_m-Lat_o), and the 95% confidence test ( Figure 1-4 ), the specific formula is as follows:

[0150] Y=a×X+b

[0151] Where Y represents the latitude difference between the model heavy rain center and the actual center (Lat_m - Lat_o); Y>0 means that the model heavy rain center is biased north compared with the actual forecast, and Y<0 means that the forecast is biased south; X represents the latitude difference between the model heavy rain center and the 584 line (Lat_m - Lat_584); a and b are the slope and intercept of the linear regression equation (both constants), respectively.

[0152] Step 203: Based on the aforementioned regression relationship, the ECMWF model's 3-hour precipitation forecast areas for different forecast start times and weather conditions from June to August are corrected. Specifically, when the ECMWF model has a 3-hour heavy rainfall forecast within the 015-036 forecast period, the model's 24-hour cumulative precipitation forecast is first used to determine whether a rainstorm center exists. If so, Lat_m - Lat_584 is calculated. Lat_m - Lat_o is then calculated using the corresponding regression equation. Finally, the model's 3-hour precipitation forecast areas are shifted by Lat_m - Lat_o, thereby correcting the precipitation area.

[0153] Step 204: Using this method, the 24-hour precipitation areas of the ECMWF model for all rainstorm processes in Anhui Province from June to August from 2011 to 2023 are corrected and compared with the original model data. The results are as follows: Figures 3 to 6 As shown in Table 4. The results show that after correction by this method, the TS scores of the model for heavy rain (50 mm) and heavy rain (25 mm) and above are improved by 7.5% and 2.5% respectively compared with the original forecast, and the corresponding BIAS scores are still maintained around 1, which proves that this method has a certain effect on improving the model's heavy rain forecast.

[0154]

[0155]

[0156] 8 p.m. Revised Forecast Original forecast ts50 0.249 0.233 bias50 0.847 0.849 ts25 0.423 0.409 bias25 1.177 1.174

[0157] Third, the ECMWF model precipitation forecast intensity is revised based on different weather conditions, as follows:

[0158] In this example, according to the season (flood season: June to August, non-flood season: other months) and weather situation background field of the current ECMWF model 3-h precipitation forecast, the corresponding historical precipitation forecast samples are selected from the 3-h precipitation forecast classification database, and the intensity correction method of the present invention is used.

[0159] Since this example is a correction for the 3-h precipitation forecast, the precipitation level used for June-August is 10 mm / 3 h and for other months it is 5 mm / 3 h. Figure 7 A schematic diagram of the solution of the scaling factor F in the actual calculation is given. During the sample training process, if the BIAS deviation of y = 1.5x for precipitation above 10 mm reaches the minimum when the scaling factor F is 1.5, then y = 1.5x is considered the revised result of the model precipitation forecast.

[0160] Fourth, the results are tested as follows:

[0161] like Figure 8As shown in the figure, the revised product was compared with the original ECMWF precipitation forecast and other model precipitation forecasts. The results of the 24-hour cumulative precipitation forecast show that the revised product's TS score improves by an average of 10% and 19% compared to the original forecasts of various models, for both heavy rainfall (25mm) and torrential rainfall (50mm and above), respectively.

[0162] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by a technician familiar with this technical field within the technical scope disclosed by the present invention should be covered by the scope of protection of the present invention.

Claims

1. A method for correcting model precipitation forecasts based on weather situation field classification, characterized in that: include: Acquire historical model precipitation forecast data for a target area and historical actual precipitation data that matches the time of the historical model precipitation forecast data, and form a plurality of samples having observation values ​​corresponding to the historical actual precipitation data and forecast values ​​corresponding to the historical model precipitation forecast data; Based on the location of the target area, the samples are divided into flood season and non-flood season, and based on the forecast values ​​of the samples, the weather situation field types of the samples are classified into weak precipitation type, typhoon-controlled type controlled by typhoon circulation, subtropical high-controlled type controlled by subtropical high pressure, trough-front type, trough-back type and trough-bottom type which are not controlled by typhoon circulation and subtropical high pressure and are divided by the different relative positions of the corresponding precipitation area and the upper-altitude shortwave trough; Constructing model precipitation forecast intensity correction models and model precipitation forecast area correction models corresponding to different weather situation field types in flood season and non-flood season respectively, and using the observed values ​​of the samples as the model output values ​​and the predicted values ​​of the samples as the model input values, the model precipitation forecast intensity correction models and the model precipitation forecast area correction models are trained; Based on the season and weather situation field corresponding to the real-time model precipitation forecast data, the corresponding trained model precipitation forecast intensity correction model and model precipitation forecast location correction model are selected to perform model precipitation forecast correction.

2. The method for correcting precipitation forecast based on weather situation field classification according to claim 1, characterized in that: The classification methods of the weather situation field types, such as weak precipitation type, typhoon-controlled type and subtropical high-controlled type, include: For a sample at a certain moment, the target area is gridded at a set resolution. The model precipitation forecast data in the sample is interpolated to the meteorological stations in the target area. The number of stations in each grid whose model precipitation forecast is greater than or equal to the first precipitation threshold is counted. If the number of stations corresponding to each grid is 0, the weather situation field type of the sample at that moment is weak precipitation type. If the number of stations is not all 0, determine whether the typhoon center position in the typhoon message corresponding to the moment is located in the target area. If so, the weather situation field type of the sample at this moment is typhoon-controlled; otherwise, based on the number of stations corresponding to the target area grid, screen out the precipitation center grid of the target area, calculate the average longitude and latitude of all stations in the precipitation center grid whose model precipitation forecast is greater than or equal to the first precipitation threshold, and use the average longitude and latitude as the precipitation center position of the model precipitation forecast at this moment, and calculate the average potential height forecast field of the sample at this moment with the subtropical high characteristic line and the subtropical high peripheral characteristic line; if the precipitation center position of the sample is within the range of the subtropical high characteristic line, the weather situation field type of the sample at this moment is subtropical high-controlled.

3. The method for correcting precipitation forecast based on weather situation field classification according to claim 2, characterized in that: The specific steps of screening out the precipitation center grid of the target area based on the number of stations corresponding to the target area grids include: dividing the target area into precipitation areas, and for each precipitation area, obtaining the number of heavy precipitation adjacent grids corresponding to each grid therein, wherein the heavy precipitation adjacent grids are grids corresponding to grids adjacent to each grid, wherein the number of stations whose model precipitation forecasts are greater than or equal to a first precipitation threshold is greater than 0; using the grid with the highest number of heavy precipitation adjacent grids as the precipitation center grid of each precipitation area; using the precipitation center grid with the highest number of heavy precipitation adjacent grids in different precipitation areas as the precipitation center grid of the target area; and in the case where there are multiple grids with the highest and consistent numbers of heavy precipitation adjacent grids, calculating the total precipitation forecast in each grid, and taking the grid with the largest total precipitation forecast as the precipitation center grid; The specific method for dividing the target area into precipitation zones includes: obtaining all grids in the target area whose corresponding station numbers are not 0, calculating the number of interval grids between any two grids in all the obtained grids, and dividing two grids whose interval grid number is less than or equal to 1 into the same precipitation zone.

4. The method for correcting precipitation forecast based on weather situation field classification according to claim 2, characterized in that: The classification methods of the weather situation field types include: trough front type, trough back type and trough bottom type. For samples whose weather situation field is not classified as weak precipitation type, typhoon controlled type controlled by typhoon circulation, or subtropical high controlled type controlled by subtropical high pressure, the latitudinal deviation of the mean geopotential height of the sample at that moment is calculated and compared with the latitudinal deviation threshold of the mean geopotential height. If the latitudinal deviation of the mean geopotential height is less than the latitudinal deviation threshold of the mean geopotential height, the weather situation field type of the sample at that moment is the trough front type; otherwise, the latitude difference between the first intersection point of the subtropical high outer characteristic line in the mean geopotential height forecast field of the sample and the first judgment meridian and the second intersection point of the second judgment meridian is determined and compared with the latitude difference threshold. If the latitude difference is greater than the latitude difference threshold, the weather situation field type of the sample at that moment is the trough back type; otherwise, the weather situation field type of the sample at that moment is the trough bottom type.

5. The method for correcting precipitation forecast based on weather situation field classification according to claim 4, characterized in that: The formula for the latitudinal deviation of the mean geopotential height is as follows: Where, is the latitudinal deviation of the average potential height corresponding to longitude lon_i and latitude lat_j, lon_i is longitude, lon_i∈[lon_i_min,lon_i_max], lat_j is latitude, lat_j∈[lat_j_min,lat_j_max]; lon_i_min is the first judgment longitude; lon_i_max is the second judgment longitude; lat_j_min is the first judgment latitude; lat_j_max is the second judgment latitude; is the average geopotential height corresponding to longitude lon_i and latitude lat_j; is the average value of the mean geopotential height corresponding to latitude lat_j at different longitudes.

6. The method for correcting precipitation forecast based on weather situation field classification according to claim 1, characterized in that: The specific method for constructing and training the model for the correction of precipitation forecast intensity for different weather situation types during the flood season and the non-flood season includes: The model precipitation forecast correction models of typhoon-controlled type, subtropical high-controlled type, trough-front type, trough-back type and trough-bottom type in the flood season, and the model precipitation forecast correction models of subtropical high-controlled type, trough-front type, trough-back type and trough-bottom type in the non-flood season are respectively constructed; the subtropical high-controlled type, trough-front type and trough-back type in the flood season and non-flood season adopt the scaling coefficient correction model; the typhoon-controlled type and trough-bottom type in the flood season adopt the optimal TB score correction model; the trough-bottom type in the non-flood season adopts the scaling coefficient correction model; The corresponding model precipitation forecast correction model is trained by taking the observed precipitation of samples of typhoon-controlled type, subtropical high-controlled type, trough-front type, trough-back type and trough-bottom type during the flood season as the model output value and taking the precipitation of the predicted value of the samples as the model input value. The corresponding model precipitation forecast correction model is trained by taking the observed precipitation of samples of subtropical high-controlled type, trough-front type, trough-back type and trough-bottom type during the non-flood season as the model output value and taking the precipitation of the predicted value of the samples as the model input value.

7. The method for correcting precipitation forecast based on weather situation field classification according to claim 6, characterized in that: The expression of the scaling factor correction model is: y=xF Where y is the corrected value of the model precipitation forecast, i.e., the output of the scaling factor correction model; x is the original value of the model precipitation forecast, i.e., the input of the scaling factor correction model; F is the scaling factor; The expression of the optimal TB score correction model is: Where, O k is the actual precipitation level of the kth level; O k+1 is the actual precipitation level of the k+1th level; E1 is the threshold of the predicted precipitation level of the first level; E k is the k-th level forecast precipitation threshold; E k+1 is the k+1th level forecast precipitation level threshold; E n is the forecast precipitation level threshold of level n; n is the total number of precipitation levels; The optimal TB score correction model is trained by adjusting the k-th level forecast precipitation level threshold E k The TB score of the actual precipitation level of level k and above is maximized. The calculation formula of the TB score is as follows: TB=TS-B Where TS is the TS score of the model forecast for precipitation above a certain magnitude, 0≤TS≤1; B is the BIAS deviation of the model forecast for precipitation above a certain magnitude, B≥0; NA is the station with a correct forecast; NB is the station with an empty forecast; and NC is the station with a missed forecast.

8. The method for correcting precipitation forecast based on weather situation field classification according to claim 1, characterized in that: The specific method for constructing and training the model for correcting the precipitation forecast area under different weather situation field types in the flood season and the non-flood season includes: The model precipitation forecast area correction models for the weather situation field types of trough front, trough back and trough bottom in flood season and non-flood season are constructed respectively. The latitude difference between the center of the model precipitation position corresponding to the sample forecast value and the actual precipitation position corresponding to the observation value is used as the model output value, and the latitude difference between the model precipitation forecast position corresponding to the sample forecast value and the outer characteristic line of the subtropical high is used as the model input value to train the corresponding model precipitation forecast correction model.

9. The method for correcting precipitation forecast based on weather situation field classification according to claim 8, characterized in that: The model precipitation forecast area correction model adopts a univariate linear regression model, which is expressed as follows: Y=a×X+b Where X is the latitude difference between the model precipitation center and the outer characteristic line of the subtropical high; Y is the latitude difference between the model precipitation center and the actual precipitation center; a is the slope of the univariate linear regression model; and b is the intercept of the univariate linear regression model.

10. The method for correcting precipitation forecast based on weather situation field classification according to claim 1, characterized in that: The specific steps of obtaining historical model precipitation forecast data of a target area and historical actual precipitation data that matches the time of the historical model precipitation forecast data, and forming a plurality of samples having observation values ​​corresponding to the historical actual precipitation data and forecast values ​​corresponding to the historical model precipitation forecast data include: A model precipitation forecast classification database is constructed, and the historical model precipitation forecast data of the target area and the historical actual precipitation data that matches the time of the historical model precipitation forecast data are stored in the database. Each sample in the database includes the observation value of the corresponding historical actual precipitation data and the forecast value of the corresponding historical model precipitation forecast data.

11. A precipitation forecast correction system based on weather situation field classification, characterized in that: It includes sample acquisition module, sample division module, model construction module and correction module; The sample acquisition module is used to acquire historical model precipitation forecast data of the target area and historical actual precipitation data that matches the time of the historical model precipitation forecast data, and form a plurality of samples having observation values ​​corresponding to the historical actual precipitation data and forecast values ​​corresponding to the historical model precipitation forecast data; The sample classification module is used to classify the samples into flood season and non-flood season based on the location of the target area, and classify the samples into weak precipitation type, typhoon-controlled type controlled by typhoon circulation, subtropical high-controlled type controlled by subtropical high pressure, trough-front type, trough-back type and trough-bottom type that are not controlled by typhoon circulation and subtropical high pressure and have different latitudinal deviations of mean geopotential height based on the forecast values ​​of the samples; The model construction module is used to respectively construct a model precipitation forecast intensity correction model and a model precipitation forecast area correction model corresponding to different weather situation field types in the flood season and the non-flood season, and train the model precipitation forecast intensity correction model and the model precipitation forecast area correction model using the observed value of the sample as the model output value and the predicted value of the sample as the model output value; The correction module is used to select the corresponding trained model precipitation forecast intensity correction model and model precipitation forecast area correction model to perform model precipitation forecast correction based on the season and weather situation field corresponding to the real-time model precipitation forecast data.

Images