Object-based set probability matching average method

Abstract

The present invention discloses an object-based ensemble probability matching average method, comprising the following steps: identifying all precipitation monomers within the calculation range from the data set of the ensemble mode, and merging adjacent precipitation monomers to form a number of precipitation objects; obtaining the minimum ensemble forecast area covering similar precipitation objects; in the minimum ensemble forecast area of ​​each precipitation object, using the PM mean method to generate the ensemble mean field of the precipitation object; merging all calculated precipitation objects and the remaining regional OPM ensemble mean fields to obtain the full-region OPM ensemble mean field. The present invention solves the problem of incorrect matching caused by the inability of the existing PM method to adapt to the scale and distance of the precipitation system, and will further enhance the distribution and intensity forecast capabilities of heavy precipitation.

Description

Technical Field

[0001] The present invention belongs to the field of atmospheric science research, relates to meteorological data processing, and in particular to an object-based ensemble probability matching mean (OPM) method, which is used to enhance the distribution forecasting capability of ensemble mean precipitation for heavy precipitation. Background Art

[0002] Heavy precipitation forecasting has always been a difficult problem, and high-resolution numerical weather forecasting is the main reference for current operational forecasting. However, there is great uncertainty in the forecast of heavy precipitation areas, and the reference value of precipitation forecasting using a single deterministic numerical model is limited. As early as 1995, the literature 1 [Vislocky, RL, and JM Fritsch, 1995: Improved model output statistics forecasts through model consensus. Bull. Amer. Meteor. Soc., 76, 1157–1164.] pointed out that the optimal precipitation forecast requires an ensemble forecast of numerical models, rather than relying solely on a single optimal product.

[0003] In this regard, reference 2 [Speer, MS, and LM Leslie, 1997: An example of the utility of ensemble rainfall forecasting. Austr. Meteor. Mag., 46, 75–78.] and reference 3 [Du, J., SL Mullen, and F. Sanders, 1997: Short-range ensemble forecasting of quantitative precipitation. Mon. Wea. Rev., 125, 2427–2459.] both adopted a simple arithmetic average of the precipitation forecasts of the ensemble model. For precipitation of general intensity, the ensemble average forecast is better than the forecast of a single model in terms of precipitation distribution, but the simple ensemble average is not suitable for heavy precipitation, mainly because the heavy precipitation areas predicted by different members vary greatly, and the simple ensemble average will cause a serious underestimate of the precipitation intensity.

[0004] To address this problem, the Probability Matching Average Method (Probability Matching Average Method) was proposed in Reference 4 [Rosenfeld, D., DB Wolff, and D. Atlas, 1993: General probability matched relations between radar reflectivity and rainrate. J. Appl. Meteor., 32, 50–72, https: / / doi.org / 10.1175 / 1520-0450(1993)032,0050:GPMRBR.2.0.CO;2.] and Reference 5 [Anagnostou, EN, AJ Negri, and RF Adler, 1999: A satellite infrared technique for diurnal rainfall variability studies. J. Geophys. Res., 104, 31 477–31 488, https: / / doi.org / 10.1029 / 1999JD900157.]. Matching (PM) is a method that effectively improves the above defects in ensemble average forecast by combining data with better spatial representation and data with more accurate frequency distribution. Reference 6 [Ebert EE. Ability of a Poor Man's Ensemble to Predict the Probability and Distribution of Precipitation [J]. Monthly Weather Review, 2000, 129 (10): 2461. DOI: 10.1175 / 1520-0493 (2001) 1292.0. CO; 2.] first applied this technology to ensemble forecasting, generating quantitative precipitation forecast products based on ensemble forecasts and significantly improving the forecast skill score. Therefore, the PM averaging method is widely used in ensemble average precipitation forecasting.

[0005] However, the PM mean method is not perfect. In the literature 7 [Surcel, M., I. Zawadzki, and MKYau, 2014: On the filtering properties of ensemble averaging for storm-scale precipitation forecasts. Mon. Wea. Rev., 142, 1093–1105, https: / / doi.org / 10.1175 / MWR-D-13-00134.1.], it is pointed out that although the PM mean method can improve the forecast skill score, it ignores the possibility of multiple precipitation systems within the calculation range. For a large area, the PM mean method systematically redistributes precipitation. Even the PM mean of neighboring points may come from precipitation systems of different scales or precipitation in different climatic environments, resulting in the inability to accurately reflect the true distribution characteristics of precipitation when applied locally. For example, if there are both type A and type B precipitation systems in the calculation area, the PM mean method may allocate the predicted type A precipitation to the type B system, resulting in the phenomenon that the precipitation system with strong intensity becomes stronger.

[0006] To overcome this limitation, some scholars have proposed the Localized Probability-Matching (LPM) method. For example, in reference 8 [Clark, AJ, 2017: Generation of ensemble meanprecipitation forecasts from convection-allowing ensembles. Wea. Forecasting, 32, 1569–1583, https: / / doi.org / 10.1175 / WAF-D-16-0199.1.], this method is used to apply the PM mean to a specific impact area centered on a single point or a small rectangular area, which to a certain extent avoids the problem of mismatch in the intensity of precipitation systems in different geographical regions. However, the rectangular regular grid design (fixed side length) of the local method is actually unable to accurately separate precipitation systems of different categories, especially when there are heavy precipitation systems of different scales in the region and the distances between different precipitation systems are very close. The local PM method cannot effectively solve such complex situations.

[0007] In summary, the traditional PM mean method takes the ensemble mean precipitation value as the benchmark, sorts from large to small, and then rematches the precipitation values ​​of the ensemble members, effectively avoiding the problem of small average values ​​caused by differences in heavy precipitation areas between different members, but it is only suitable for small areas. The local PM method can avoid the mismatch of precipitation values ​​in geographical areas with different climatic states to a certain extent, but it still cannot solve the problem of coexistence of multiple precipitation systems. For example, when typhoons, fronts, and short-term heavy precipitation in the afternoon exist at the same time, in theory, the PM ensemble mean precipitation field of different systems should be classified and calculated, but because the PM method cannot effectively distinguish the system, the typhoon has a large range and strong intensity, and the precipitation generated by the front and short-term heavy precipitation is often mistakenly allocated to the typhoon system, resulting in a larger range and stronger intensity of typhoon heavy precipitation. On the contrary, the range and intensity of heavy precipitation in the other two types of precipitation systems will be weakened to a certain extent. In general, when there are multiple heavy precipitation systems in a region, and the intensities of different heavy precipitation systems are not much different, the distances are close, and they cover different scales, both the traditional PM method and the local PM method will cause precipitation mismatch problems, and the strong will always be strong. Summary of the invention

[0008] Purpose of the invention: The purpose of the present invention is to address the deficiencies in the prior art and provide an object-based probability matching average method (OPM for short), which first identifies precipitation objects and adapts to the scale of precipitation objects, then uses the PM method to reallocate precipitation intensity, thereby increasing the applicability of the PM method, and improves the adaptability of the calculation scale of the PM ensemble average precipitation field through object identification, thereby improving the forecast performance of the ensemble average precipitation product for the intensity and range of heavy precipitation.

[0009] To achieve the above object, the present invention proposes an object-based set probability matching averaging method, comprising the following steps:

[0010] S100, identifying all precipitation monomers within a calculation range from the data set of the ensemble model, and merging adjacent precipitation monomers to form a plurality of precipitation objects;

[0011] S200, obtaining a minimum ensemble forecast area covering each precipitation object;

[0012] S300, in the minimum ensemble forecast area of ​​each precipitation object, using the PM mean method to generate the ensemble mean field of the precipitation object;

[0013] S400: Merge all calculated precipitation objects and the ensemble mean fields of the remaining areas to obtain the OPM ensemble mean field of the entire area.

[0014] A further preferred technical solution of the present invention is that the step S100 identifies all precipitation monomers within the calculation range from the data set of the collective model and merges adjacent precipitation monomers, including:

[0015] S110, identifying each precipitation cell within the calculation range from the precipitation forecast of the ensemble model according to the object-based MODE inspection method;

[0016] S120: According to a proximity algorithm, adjacent precipitation monomers are merged into a plurality of precipitation objects.

[0017] Preferably, in step S110, according to the object-based MODE inspection method, each precipitation monomer within the calculation range is identified from the precipitation forecast of the ensemble model, and the specific method is:

[0018] S111, obtaining a precipitation forecast data set of an ensemble model;

[0019] S112. For each set member, a disk-shaped convolution kernel with a radius of R is used to convolve and smooth the forecast field. The specific formula is:

[0020] (1);

[0021] (2);

[0022] in, It is the original precipitation forecast field. is the filter function, is the result field after convolution, The target point is The coordinates in yes The grid point coordinates used in yes Corresponding to The grid point coordinates of

[0023] Filter function It is a circular filter, which is determined by the influence radius R and height H. R and H satisfy the relationship H =1;

[0024] S113, use the Unicom domain extraction algorithm to identify the precipitation monomers in the forecast field, and set the precipitation threshold Create a mask field , the formula is:

[0025] (3);

[0026] In this mask field, connected areas with a value of 1 are identified as the same precipitation cell.

[0027] As an advantage, it also includes the steps of:

[0028] S114. Further set a minimum area for identifying a precipitation cell. If the area of ​​a precipitation cell in the forecast field is smaller than the minimum area, delete the precipitation cell.

[0029] Preferably, in step S120, adjacent precipitation monomers are merged into a plurality of precipitation objects according to a proximity algorithm, and the specific method is as follows:

[0030] S121, calculate the proximity between each precipitation monomer according to the proximity judgment method, and the calculation formula is:

[0031] (4);

[0032] In the formula, and are the total number of pixels of monomer A and monomer B, and is the number of pixels where monomer A and monomer B are in each other's adjacent areas;

[0033] S122, selecting the maximum value therein, and if the maximum value is greater than a preset proximity threshold, the corresponding two monomers are merged into one precipitation monomer;

[0034] S123, repeating step S121 and step S122 until the proximity between all pairs of monomers is less than the preset threshold, completing the merging of adjacent precipitation monomers to form a number of different precipitation objects.

[0035] S124. Repeat steps S112 to S123 to complete the precipitation object identification of all set members.

[0036] Preferably, the step S200 of obtaining the minimum set forecast area covering each precipitation object is specifically performed as follows:

[0037] The spatial ranges of similar precipitation objects predicted by all ensemble members are superimposed to form the minimum ensemble forecast area for each precipitation object. For a single precipitation object, the spatial ranges of similar precipitation objects of each ensemble member are expressed as Indicates that represents different set members, and The value range is 1 to ; The minimum ensemble forecast area A for a single precipitation object is expressed as:

[0038] (5);

[0039] is the number of members in the set.

[0040] Preferably, in step S300, in the minimum ensemble forecast area of ​​each precipitation object, the PM mean method is used to generate the ensemble mean field of the precipitation object, and the specific method is:

[0041] S310, sorting the forecast values ​​of each grid point in the minimum ensemble forecast area of ​​the precipitation object by size, and recording their positions:

[0042] (6);

[0043] Where m is the number of set members, n is the number of grid points, The original precipitation forecast fields in the minimum ensemble forecast area of ​​all members, arranged from large to small;

[0044] S320. According to the sorting result, a median is selected for every number of set members. The formula is:

[0045] (7);

[0046] where j = 1,…,n;

[0047] Composition of the median field ;

[0048] S330. Redistribute the ensemble median field to the ensemble arithmetic mean field according to the probability matching principle, that is, the ensemble arithmetic mean field is arranged from large to small, but the values ​​therein are replaced from large to small by the corresponding ensemble median field from large to small, and then obtain the OPM ensemble mean field of a single precipitation object according to the position information of the ensemble arithmetic mean field.

[0049] S340, repeating steps S310 to S330 to complete the OPM ensemble mean field calculation for all identified precipitation objects.

[0050] Preferably, in addition to merging the OPM ensemble mean fields of all identified precipitation objects in step S400, the PM mean method is continued to be used to generate the precipitation ensemble mean field for points that are not outside the identified precipitation object area. Finally, all calculated precipitation objects and the ensemble mean fields of the remaining area are merged to obtain the OPM ensemble mean field of the entire area.

[0051] Beneficial effects: The present invention provides an object-based ensemble probability matching average (OPM) method to solve the mismatching problem caused by the inability of existing PM methods to adapt to the scale and distance of precipitation systems, which will further improve the ability to forecast the distribution and intensity of heavy precipitation.

[0052] Specifically, based on the "frequency matching method", the present invention first objectively identifies and merges precipitation monomers formed by weather systems of different scales, then calculates the minimum overlapping area covering each precipitation object of all members, and finally calculates the OPM ensemble mean field for each precipitation object, thereby achieving the optimization of the ensemble mean precipitation field when there are multiple heavy precipitation objects in the calculation area. It not only ensures the intensity accuracy of the ensemble average of different precipitation objects, but also fully considers the differences in precipitation properties between different precipitation objects, avoids mismatching, and generates a more refined and accurate precipitation ensemble mean field, which will further improve the forecast of ensemble precipitation distribution and intensity.

[0053] The significant advantage of the present invention is that the OPM ensemble average method expands the scope of application of the PM method. When multiple precipitation objects can be identified in the calculation area, the OPM ensemble precipitation ensures that the properties of the precipitation object remain unchanged, while effectively improving the forecast accuracy of the intensity and range of the center of heavy precipitation. This improvement is of great significance for improving the accuracy and reliability of the ensemble precipitation product for heavy precipitation forecasts. Compared with the traditional PM method, this method has obvious advantages when there are multiple precipitation objects of different scales in the calculation area. The OPM method can not only improve the accuracy of the forecast of the intensity and range of heavy precipitation, but also retain the precipitation properties of the original system, which will improve the level of detailed forecasting of heavy precipitation by the ensemble forecast product. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a flowchart of the implementation of the object-based set probability matching average method in an embodiment of the present invention;

[0055] Figure 2 A diagram comparing the ensemble mean fields and their differences calculated by different ensemble probability matching methods in an embodiment of the present invention;

[0056] Figure 3 is a comparison diagram of precipitation object areas calculated by different set probability matching methods and observations in the embodiment;

[0057] Figure 4 is an objective score diagram of frontal precipitation FSS in the region by different set probability matching methods in the embodiment;

[0058] Figure 5 It is an objective score diagram of typhoon precipitation FSS in the region using different set probability matching methods in the embodiment. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings in the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments, and they should not be understood as limitations on the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. In the description of the present invention, it should be understood that the terms used are only for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0060] Combine the following Figure 1-Figure 5 The object-based set probability matching averaging method provided by the present invention is described.

[0061] Embodiment: This embodiment provides an object-based set probability matching average method, such as Figure 1 As shown, first identify all precipitation objects within the calculation range. Second, superimpose the ranges of all member similar precipitation objects and calculate PM precipitation separately. For the convenience of display, only the identified frontal and typhoon precipitation cases are selected here. Finally, synthesize each precipitation object and the remaining area map to form the OPM ensemble mean field of the entire area.

[0062] Specifically, the steps include:

[0063] S100, identifying all precipitation monomers within a calculation range from the data set of the ensemble model, and merging adjacent precipitation monomers to form a plurality of precipitation objects;

[0064] S200, obtaining a minimum ensemble forecast area covering each precipitation object;

[0065] S300, in the minimum ensemble forecast area of ​​each precipitation object, using the PM mean method to generate the ensemble mean field of the precipitation object;

[0066] S400: Merge all calculated precipitation objects and the ensemble mean fields of the remaining areas to obtain the OPM ensemble mean field of the entire area.

[0067] Taking the 24-hour cumulative precipitation forecast as an example, the TIGGE (THORPEX Interactive GrandGlobal Ensemble) data set is used, and 20 ensemble members of the European Center for Medium-Range Weather Forecasts (ECMWF) global ensemble forecast model starting at 12UTC are selected, with a spatial resolution of 0.5° × 0.5°. The steps of this embodiment are described in more detail:

[0068] In step S100, it includes:

[0069] S110, according to the object-based MODE inspection method, identifying each precipitation monomer within the calculation range from the precipitation forecast of the ensemble model; the specific method is:

[0070] S111, obtaining a precipitation forecast data set of an ensemble model;

[0071] S112. For each set member, a disk-shaped convolution kernel with a radius of R is used to convolve and smooth the forecast field. The specific formula is:

[0072] (1);

[0073] (2);

[0074] in, It is the original precipitation forecast field. is the filter function, is the result field after convolution, The target point is The coordinates in yes The grid point coordinates used in yes Corresponding to The grid point coordinates of

[0075] Filter function It is a circular filter, which is determined by the influence radius R and height H. R and H satisfy the relationship H =1;

[0076] S113, use the Unicom domain extraction algorithm to identify the precipitation monomers in the forecast field, and set the precipitation threshold Create a mask field , the formula is:

[0077] (3);

[0078] In this mask field, connected areas with a value of 1 are identified as the same precipitation cell;

[0079] S114. In order to ensure the accuracy of the data, a minimum area for identifying a precipitation cell is further set. In this embodiment, 10 grid points are selected as a standard. If the area of ​​a precipitation cell in the forecast field is smaller than the minimum area, the precipitation cell is deleted.

[0080] S120, according to the proximity algorithm, adjacent precipitation monomers are merged into a plurality of precipitation objects, the specific method is as follows:

[0081] S121, calculate the proximity between each precipitation monomer according to the proximity judgment method, and the calculation formula is:

[0082] (4);

[0083] In the formula, and are the total number of pixels of monomer A and monomer B, and is the number of pixels where monomer A and monomer B are in each other's adjacent areas;

[0084] S122, selecting the maximum value therein, and if the maximum value is greater than a preset proximity threshold, the corresponding two monomers are merged into one precipitation monomer;

[0085] S123, repeating step S121 and step S122 until the proximity between all pairs of monomers is less than the preset threshold, completing the merging of adjacent precipitation monomers to form a number of different precipitation objects.

[0086] S124. Repeat steps S112, S113, S114, S121, S122 and S123 to complete the precipitation object identification of all set members.

[0087] In step S200:

[0088] The spatial ranges of similar precipitation objects predicted by all ensemble members are superimposed to form the minimum ensemble forecast area for each precipitation object. For a single precipitation object, the spatial ranges of similar precipitation objects of each ensemble member are expressed as Indicates that represents different set members, and The value range is 1 to ; The minimum ensemble forecast area A for a single precipitation object is expressed as:

[0089] (5);

[0090] is the number of members in the set.

[0091] In step S300:

[0092] S310, sorting the forecast values ​​of each grid point in the minimum ensemble forecast area of ​​the precipitation object by size, and recording their positions:

[0093] (6);

[0094] Where m is the number of set members, n is the number of grid points, The original precipitation forecast fields in the minimum ensemble forecast area of ​​all members, arranged from large to small;

[0095] S320. According to the sorting result, a median is selected for every number of set members. The formula is:

[0096] (7);

[0097] where j = 1,…,n;

[0098] Composition of the median field ;

[0099] S330. Redistribute the ensemble median field to the ensemble arithmetic mean field according to the probability matching principle, that is, the ensemble arithmetic mean field is arranged from large to small, but the values ​​therein are replaced from large to small by the corresponding ensemble median field from large to small, and then obtain the OPM ensemble mean field of a single precipitation object according to the position information of the ensemble arithmetic mean field.

[0100] S340, repeating steps S310 to S330 to complete the OPM ensemble mean field calculation for all identified precipitation objects.

[0101] In step S400, in addition to merging the OPM ensemble mean fields of all identified precipitation objects, the PM mean method is continued to be used to generate the precipitation ensemble mean field for points that are not in the identified precipitation object area, thereby forming the OPM ensemble mean field of the entire area.

[0102] In order to comprehensively evaluate the effect of the object-based ensemble probability matching average method of the present invention, this embodiment uses the TIGGE (THORPEX Interactive Grand Global Ensemble) data set, selects the European Center for Medium-Term Weather Forecasts (ECMWF) global ensemble forecast model, a total of 20 ensemble members, and comprehensively conducts subjective and objective tests to evaluate the improvement effect of the OPM method. In the division of precipitation intensity, we refer to the classification standards of the Central Meteorological Observatory for 24-hour precipitation and set five thresholds, namely 10 mm, 25 mm, 50 mm, 100 mm and 250 mm.

[0103] like Figure 2 As shown, the example time displayed is the spatial distribution map of the 24-hour cumulative precipitation from 08:00 on July 25 to 08:00 on July 26, 2021 (unit: mm / 24h): Figure 2 (a) is the gridded fused precipitation product (OBS); Figure 2 (b) is the PM ensemble mean field product; Figure 2 (c) is the OPM ensemble mean field product, and the precipitation forecast deviation map (unit: mm / 24h) is shown in Figure 2 (d) PM–OBS comparison diagram, and Figure 2 As shown in the OPM–PM comparison diagram shown in (e).

[0104] Compared with the traditional PM method, the OPM method calculates similar precipitation objects, but the PM method has more heavy precipitation points in terms of the number of heavy precipitation centers. The PM method significantly overestimates the precipitation on the land side of the typhoon. This is mainly because the typhoon has a larger precipitation range and stronger intensity than the front and other weather systems that cause precipitation, resulting in the front ( Figure 2 Middle (d) and Figure 2 The red box 1 in (e) and other weather systems with heavy precipitation ( Figure 2 Middle (d) and Figure 2 The red box 3 in (e) was reallocated to the typhoon system ( Figure 2 Middle (d) and Figure 2 The OPM method avoids this situation well, and the precipitation probability matching is limited to the same precipitation object, thus well preserving the properties of different precipitation objects. The problem of overestimation of typhoon precipitation range has been significantly improved. Figure 3 As shown in the figure, the precipitation areas obtained by the two methods are objectively compared. Compared with the PM method, the OPM method significantly reduces the overestimation of the typhoon and frontal heavy precipitation areas. In general, the ranges of multiple heavy precipitation centers shown by the OPM method are closer to observations.

[0105] like Figure 4 and Figure 5 As shown in the figure, after the frontal and typhoon precipitation objects in the region are identified and separated, the FSS scores are calculated separately to test the improvement effect of OPM on the identified different precipitation objects. FSS evaluates the forecasting skill by comparing the difference in the probability of precipitation occurrence between the forecast and the observation within the neighborhood range. The closer the score is to 1, the better the forecasting performance. In this embodiment, multiple neighborhood radii (1, 2, 3, 4 and 5 grid points) are calculated to examine the forecasting effect at different spatial scales. The results show that whether it is frontal or typhoon precipitation, the OPM ensemble precipitation averages most thresholds in the full neighborhood spatial scale. The performance is better than the traditional PM method, which is mainly due to the fact that OPM avoids the mismatch problem between different precipitation objects in the traditional PM method. In general, the OPM method first identifies the precipitation object, adaptively adjusts the calculation range, and maximizes the guarantee that the ensemble precipitation comes from similar precipitation objects in the forecast, and improves the ensemble precipitation product for precipitation distribution, especially the forecast of the range and intensity of heavy precipitation.

[0106] 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 object-based set probability matching average method, characterized in that: The steps include: S100, identifying all precipitation monomers within the calculation range from the data set of the ensemble model, and merging adjacent precipitation monomers to form a plurality of precipitation objects; including: S110, according to the object-based MODE inspection method, identifying each precipitation monomer within the calculation range from the precipitation forecast of the ensemble model; the specific method is: S111, obtaining a precipitation forecast data set of an ensemble model; S112. For each set member, a disk-shaped convolution kernel with a radius of R is used to convolve and smooth the forecast field. The specific formula is: C(x,y)=∑ u,v φ(u,v)f(x-u,y-v) (1); Where f is the original precipitation forecast field, φ is the filter function, C is the result field after convolution, (x, y) is the coordinate of the target point in f, (u, v) is the grid point coordinate used in φ, and (xu, yv) is the grid point coordinate corresponding to φ(u, v) in f(x, y); The filter function φ is a circular filter, which is determined by the influence radius R and height H. R and H satisfy the relationship πR 2 H = 1; S113, using the Unicom domain extraction algorithm to identify the precipitation monomer in the forecast field, and creating a mask field M by setting the precipitation threshold T, the formula is: In this mask field, connected areas with a value of 1 are identified as the same precipitation cell; S120, merging adjacent precipitation monomers into a plurality of precipitation objects according to a proximity algorithm; S200, obtaining a minimum ensemble forecast area covering each precipitation object; S300, in the minimum ensemble forecast area of ​​each precipitation object, using the PM mean method to generate the ensemble mean field of the precipitation object; S400: Merge all calculated precipitation objects and the ensemble mean fields of the remaining areas to obtain the OPM ensemble mean field of the entire area.

2. The object-based set probability matching average method according to claim 1, characterized in that: Also includes the steps: S114. Further set a minimum area for identifying a precipitation cell. If the area of ​​a precipitation cell in the forecast field is smaller than the minimum area, delete the precipitation cell.

3. The object-based set probability matching average method according to claim 2, characterized in that: In step S120, adjacent precipitation monomers are merged into a plurality of precipitation objects according to the proximity algorithm. The specific method is as follows: S121, calculate the proximity between each precipitation monomer according to the proximity judgment method, and the calculation formula is: Where N A and N B are the total number of pixels of monomer A and monomer B, n A and n B is the number of pixels where monomer A and monomer B are in each other's adjacent areas; S122, selecting the maximum value therein, and if the maximum value is greater than a preset proximity threshold, the corresponding two monomers are merged into one precipitation monomer; S123, repeating step S121 and step S122 until the proximity between all two monomers is less than the preset threshold, completing the merging of adjacent precipitation monomers to form a number of different precipitation objects; S124. Repeat steps S112 to S123 to complete the precipitation object identification of all set members.

4. The object-based set probability matching average method according to claim 1, characterized in that: The specific method of obtaining the minimum set forecast area covering each precipitation object in step S200 is as follows: The spatial ranges of similar precipitation objects predicted by all ensemble members are superimposed to form the minimum ensemble forecast area for each precipitation object. For a single precipitation object, the spatial range of similar precipitation objects of each ensemble member is expressed as A i Indicated by, where i represents different ensemble members, and the value range of i is 1 to m; the minimum ensemble forecast area A for a single precipitation object is expressed as: A=A1∪A2∪A3…∪A m (5); m is the number of members in the set.

5. The object-based set probability matching average method according to claim 1, characterized in that: In step S300, in the minimum ensemble forecast area of ​​each precipitation object, the PM mean method is used to generate the ensemble mean field of the precipitation object. The specific method is: S310, sorting the forecast values ​​of each grid point in the minimum ensemble forecast area of ​​the precipitation object by size, and recording their positions: f′=(f N ,f2,…,f R ,f RUN ,f RU2 ,…,f 2R ,…,f (DVN)R ,f (DVN)RUN ,…,f DR ) (6); Where m is the number of ensemble members, n is the number of grid points, and f is the original precipitation forecast field in the minimum ensemble forecast area of ​​all members, arranged from large to small; S320. According to the sorting result, a median is selected for every number of set members. The formula is: f′ j，PM =Median(f (j-1)m+1 ,f (j-1)m+1 ,…,f (j-1)m+m ) (7); Where, j = 1,…,n; The median field f′ of the composition set PM =(f′ 1,PM ,…,f′ n,PM ); S330, reallocating the ensemble median field to the ensemble arithmetic mean field according to the probability matching principle, that is, the ensemble arithmetic mean field is arranged from large to small, but the values ​​therein are replaced from large to small by the corresponding ensemble median field from large to small, and then obtaining the OPM ensemble mean field of a single precipitation object according to the position information of the ensemble arithmetic mean field; S340, repeating steps S310 to S330 to complete the OPM ensemble mean field calculation for all identified precipitation objects.

6. The object-based set probability matching average method according to claim 1, characterized in that: In step S400, in addition to merging the OPM ensemble mean fields of all identified precipitation objects, the PM mean method is continued to be used to generate the precipitation ensemble mean field for points that are not outside the identified precipitation object area. Finally, all calculated precipitation objects and the ensemble mean fields of the remaining area are merged to obtain the OPM ensemble mean field of the entire area.

Images