A weight coefficient-based hourly time-lag set precipitation forecast correction method

Abstract

The present application relates to the technical field of precipitation forecast, in particular to a kind of hourly time lag ensemble precipitation forecast correction method based on weight coefficient, it is based on the time lag ensemble forecast method of Taylor test result and the forecast-observation probability matching method based on GAMMA cumulative probability analysis, and the fine gridding display of precipitation correction product is realized, aims at reducing precipitation forecast error, improve hourly precipitation forecast and 24 hours precipitation forecast accuracy, provide scientific reference for hourly precipitation, especially short-time heavy precipitation forecast;It includes the following steps: using new type of ensemble weight coefficient Wi based on Taylor test result is developed;Using the time lag ensemble forecast method based on new type of weight coefficient, the forecast member of different time starting is collected;Using the forecast-observation probability matching method based on GAMMA cumulative probability analysis, corrects hourly precipitation change and 24 hours precipitation order of magnitude;Precipitation forecast test;Its beneficial effects are: the present application improves hourly, 24 hours sunny and heavy rain forecast capability.

Description

Technical Field

[0001] This invention relates to the field of precipitation forecasting technology, and in particular to a method for correcting hourly time-lag ensemble precipitation forecasts based on weighted coefficients. Background Technology

[0002] In existing technologies, high-resolution numerical models have achieved hourly quantitative precipitation forecasts. However, due to model errors, correction studies based on model products have become crucial for improving the accuracy of refined precipitation forecasts. Within the short-term nowcast lead time, model forecast errors are mainly affected by uncertainties in the initial analysis field. Ensemble forecasting techniques can effectively reduce forecast uncertainties and improve model forecast skill. Among these, the time-lag ensemble method has attracted attention because it can better utilize rapidly updated mesoscale forecast products. The advantage of this method is that for forecasts with different start times at the same time point, it objectively increases the proportion of forecast members with better initial forecast performance and reduces the proportion of forecast members with poorer forecast performance. This not only improves the errors caused by uncertainties in the initial analysis but also provides objectivity, saves forecasters' time costs, and improves forecast efficiency.

[0003] Scholars such as Fu Na et al. (2013) and Tang Wenyuan et al. (2019) have applied the time-lag ensemble method and achieved certain results. However, the existing time-lag methods, including ensemble averaging and methods that construct unequal weights based on precipitation TS scores, cannot fully reflect the simulation ability of the ensemble model in the early stage of the event. Therefore, how to verify the simulation effect of the model in the early stage and construct the optimal ensemble weight coefficients have become the main problems of the time-lag ensemble method. Summary of the Invention

[0004] The purpose of this invention is to provide an hourly time-lag ensemble precipitation forecast correction method based on weighted coefficients. It is based on the time-lag ensemble forecast method with Taylor test results and the forecast-observation probability matching method based on GAMMA cumulative probability analysis, and realizes fine grid display. It aims to reduce precipitation forecast errors, improve the accuracy of hourly precipitation forecasts and 24-hour precipitation forecasts, and provide a scientific reference for hourly precipitation forecasts, especially short-term heavy precipitation forecasts.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A method for correcting hourly time-lag ensemble precipitation forecasts based on weighted coefficients, characterized by the following steps:

[0007] Step 1: Develop a new set weight coefficient Wi based on Taylor's test results;

[0008] Step 2: Use a time lag ensemble forecasting method based on novel weighting coefficients to assemble forecast members that started forecasting at different times;

[0009] Step 3: Correct hourly precipitation variations and magnitudes using a forecast-observation probability matching method based on GAMMA cumulative probability analysis;

[0010] Step 4: Verify the precipitation forecast.

[0011] Preferably, in step 1, the weighting coefficient Wi is mainly developed based on the results of the Taylor test.

[0012] Specifically, the Taylor plot selects the spatial correlation coefficient (R) between the simulated field and the observed field, and the ratio of the standard deviations (σ) between the simulated field (y field) and the observed field (x field). y / σ x The three variables—root mean square error (E') after removing systematic errors—are used to comprehensively examine the differences between the simulated field and the observed field.

[0013] In the Taylor diagram, according to the triangle cosine theorem, we can obtain the following relationship:

[0014] E' 2 / σ x 2 =1+σ y 2 / σ x 2 -2·σ y / σ x ·R

[0015] Where E' 2 / σ x 2 E' represents the distance between the simulated field (Mod.) and the observed field (Obs.). 2 / σ x 2 The smaller the value, the closer the simulated field (Mod.) is to the observed field (Obs.); therefore, the variable E' / σ x It can be used to evaluate the difference between the ensemble's earlier forecasts and the actual field, E' / σ x The smaller the value, the larger the given weight coefficient should be, and vice versa;

[0016] Based on the above principle, construct the variable IA=f(E' / σ) x )=σ x / E', thus ensuring that IA and σ x / E' exhibits an inverse correlation, i.e., E' / σ x Smaller set members have higher weights, so the weight coefficients Wi for each set's forecast members (i=1,2,...n are the forecast members of each set) are:

[0017] .

[0018] Preferably, in step 2, the time lag ensemble forecasting method comprehensively considers the model forecast results of the same time point with different start times and different forecast lead times, and obtains the precipitation correction forecast value by ensemble the forecast members through different weighting coefficients.

[0019] Specifically, the core of the time-lag ensemble forecasting method is to construct ensemble members based on a rapidly updating assimilation system. Each cycle of update will generate a high-frequency forecast field and contribute new ensemble members. This process does not consume additional computer resources, making it an economical and practical ensemble forecasting scheme.

[0020] Given the weighting coefficients W of forecast members at different start times i (i=1,2,3...n are the forecast members of each ensemble), then the time-lag ensemble precipitation forecast P value is,

[0021] .

[0022] In step 3, the above-mentioned time-lag ensemble forecast products can effectively improve the spatial error and root mean square error between forecast and actual conditions. However, due to the effect of ensemble smoothing, they may not be able to improve the precipitation level forecast. Therefore, based on this new type of time-lag ensemble forecast, the forecast-observation probability matching method is used to correct the precipitation level.

[0023] The forecast-observation probability matching method assumes that for a certain precipitation threshold Ti, its precipitation probability (or cumulative precipitation probability) Pr(T'i) is the same as the observed probability Po(Ti), and the actual precipitation Ti is the frequency matching correction value of the model forecast precipitation T'i.

[0024] Specifically, since the cumulative precipitation probability distribution is non-normal, numerous studies both domestically and internationally have shown that the Gamma distribution is more suitable for fitting the cumulative precipitation probability distribution curve. The Gamma cumulative probability distribution function is as follows:

[0025]

[0026] In the formula, α>0, β>0, where α is the shape parameter and β is the scale parameter. For precipitation. α and β are obtained by maximum likelihood estimation; sample mean. variance s 2 The relationship with parameters α and β is as follows: .

[0027] In step 4, according to the China Meteorological Administration's precipitation forecast assessment requirements and the "Intelligent Forecasting Technology Method Competition Verification Scheme," the main focus is on verifying 1-hour and 24-hour precipitation, selecting the following precipitation verification statistics:

[0028] ① Rain / sunshine forecast accuracy: PC = (NA + ND) / (NA + NB + NC + ND) × 100%

[0029] In the formula, NA represents the number of stations with correct precipitation forecasts, NB represents the number of stations with no forecasts, NC represents the number of stations with missed forecasts, and ND represents the number of stations with correct no precipitation forecasts. The verification parameters in this invention include the 1-hour precipitation forecast accuracy rate and the 24-hour precipitation forecast accuracy rate.

[0030] ② Heavy precipitation testing includes TS score and BIAS score.

[0031] TS score: TS = NA / (NA + NB + NC)

[0032] BIAS score: BIAS = (NA + NB) / (NA + NC)

[0033] In the formula, The number of stations with accurate precipitation forecasts, Empty station count, This represents the number of stations that were missed in reporting. The number of stations with no heavy precipitation;

[0034] The TS score is used to verify the effectiveness of precipitation level forecasts; the BIAS score is used to verify the area deviation between the forecast and the actual situation. BIAS=1 indicates that the two areas are equal, >1 indicates that the forecast area is too large (wet deviation), and <1 indicates that the forecast area is too small (dry deviation).

[0035] The testing parameters in this invention include the TS score and BIAS score for 1-hour heavy precipitation, and the TS score and BIAS score for 24-hour torrential and heavy precipitation. The testing parameter in this invention is the average relative error of 1-hour precipitation. The precipitation thresholds are shown in the table below:

[0036] Table 1 Precipitation Thresholds

[0037]

[0038] ③ Average relative error of precipitation:

[0039] In the formula To observe precipitation, The corresponding forecast precipitation is calculated as 0 when both the actual and forecast values ​​are 0.

[0040] Compared with the prior art, the beneficial effects of the present invention are:

[0041] 1. This invention constructs the weight coefficients of time-lag ensemble forecast members based on Taylor plot test results, which objectively reflects the difference between forecasts and actual conditions, fundamentally improving model forecast errors. This is more reasonable than the unequal weight coefficients defined solely based on TS scores.

[0042] 2. This invention applies the forecast-observation probability matching method based on GAMMA cumulative probability function analysis to the correction of 24-hour and hourly precipitation. It corrects both the 24-hour precipitation magnitude and the hourly precipitation variation, with good correction effect, which improves the accuracy of both 24-hour and hourly precipitation forecasts after correction.

[0043] 3. This invention improves the accuracy of hourly and 24-hour weather forecasts.

[0044] 4. This invention improves the forecasting techniques for 24-hour heavy rain (50-100mm / 24h) and torrential precipitation (≥50mm / 24h).

[0045] 5. This invention improves the forecasting skills for short-duration heavy precipitation (≥50mm / h) and reduces the average relative error of 1-hour precipitation. Attached Figure Description

[0046] Figure 1 This is a schematic diagram of the time lag set method of the present invention;

[0047] Figure 2 This is a Taylor illustration in the present invention, where Obs. represents the observation field location and Mod. represents the simulated field location;

[0048] Figure 3 This is a schematic diagram of the forecast-observation frequency matching correction process in this invention;

[0049] Figure 4 This is a flowchart of the implementation scheme in Embodiment 1 of the present invention;

[0050] Figure 5 This is the accuracy chart of sunny / rainy weather data from June 1st to August 31st, 2022, in Embodiment 2 of the present invention.

[0051] Figure 6 The TS score (a) and BIAS score (b) for the 24-hour heavy rain forecast from June 1 to August 31, 2022, and the TS score (c) and BIAS score (d) for the 24-hour heavy precipitation in Embodiment 2 of the present invention.

[0052] Figure 7 The data in Embodiment 2 of this invention are the hourly weather accuracy (a) and average relative error (b) for the period from June 1 to August 31, 2022.

[0053] Figure 8The TS score (a) and BIAS score (b) for 1 hour of heavy precipitation from June 1 to August 31, 2022 in Embodiment 2 of the present invention.

[0054] Figure 9 The following are the precipitation data from 08:00 on August 18 to 08:00 on August 19, 2022 in Embodiment 2 of the present invention: (a), the forecast starting at the next hour (starting at 02:00 on the 18th), (b), and the ensemble probability matching fusion forecast (c).

[0055] Figure 10 The TS score (a) and BIAS score (b) for the 24-hour rainstorm from 08:00 on August 18 to 08:00 on August 19, 2022, in Embodiment 2 of the present invention, and the TS score (c) and BIAS score (d) for the 24-hour heavy precipitation.

[0056] Figure 11 The hourly overall forecast for August 18, 2022, from 08:00 to 08:00 on August 19, 2022, in Embodiment 2 of the present invention is verified as follows: 1-hour accuracy of clear / rainy weather (a), 1-hour average relative error (b), 1-hour heavy precipitation TS score (c), and 1-hour heavy precipitation BIAS score (d).

[0057] Figure 12 This is a diagram illustrating the MICAPS forecast file format of the operational forecasting system in Embodiment 3 of the present invention;

[0058] Figure 13 This is the distribution map of the ensemble probability matching fusion forecast and the gridded display map of the operational forecasting system MICAPS in Embodiment 3 of the present invention. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0060] Example 1

[0061] Implementation plan, such as Figure 4 As shown,

[0062] (1) Perform Taylor test on the forecast members of the forecast ensemble in the early stage of the forecast.

[0063] For forecasts at any given time each day, a period preceding the forecast is selected as the training period. The forecast sequences for each day within this training period are compiled, and Taylor tests are performed on all ensemble forecast members. The Taylor test parameter E' / σ for each ensemble forecast member is calculated. x .

[0064] (2) Using a novel weighting coefficient based on Taylor test results to construct time lag ensemble forecasts

[0065] Based on the Taylor test results during the training period, the construct variable IA = f(E' / σ) for each ensemble forecast member is calculated. x )=σ x / E', and then obtain the new weight coefficients Wi of each ensemble forecast member, and use the time lag ensemble forecast method to calculate the hourly time lag precipitation ensemble forecast for any time of day and the corresponding time of day during the training period.

[0066] (3) A correction model is established using the forecast-observation probability matching method based on GAMMA cumulative probability function analysis.

[0067] A 24-hour forecast period is selected (generally 08:00 on the current day to 08:00 on the next day or 20:00 on the current day to 20:00 on the next day). The corresponding 24-hour precipitation ensemble forecast products are accumulated to obtain the corresponding 24-hour precipitation ensemble forecast products within the training period. A correction model is established and the 24-hour ensemble forecast is corrected using the forecast-observation probability matching method based on GAMMA cumulative probability function analysis. Finally, the TS scores of each precipitation level (light rain, moderate rain, heavy rain, rainstorm, heavy rainstorm and above) of the 24-hour precipitation ensemble forecast and the 24-hour ensemble probability matching forecast during the training period are tested to verify the correction effect of the 24-hour forecast during the training period.

[0068] (4) Time-lag precipitation ensemble forecasts were corrected using the GAMMA cumulative probability matching correction model.

[0069] For any 24-hour forecast period, the following corrections are made:

[0070] ① Based on the TS score verification effect of the 24-hour precipitation ensemble and ensemble probability matching forecast (corrected forecast) during the early training period, and given the order of magnitude improvement in the TS score of the 24-hour ensemble corrected forecast during the training period, the correction model is used to correct the precipitation ensemble forecast for the 24-hour forecast period, thus obtaining the 24-hour precipitation ensemble probability matching fusion forecast.

[0071] ② For this 24-hour forecast period, the hourly precipitation ensemble forecast within the forecast period is directly corrected using a correction model, and based on P i / (p1+p2...+p 24 (P) i For any given time, the hourly weights are reset for the precipitation forecast (i = 1, 2, 3, ..., 24 hours). Then, using the 24-hour precipitation ensemble probability matching fusion forecast obtained from ①, the weights are redistributed according to the corrected hourly weights to obtain the hourly precipitation ensemble probability matching fusion forecast, and a gridded product with a 3km resolution is generated.

[0072] (5) Validate the optimization model

[0073] Typical cases were selected, and the training period was continuously changed for forecasting and testing. The optimal training period was determined, and the forecasting model was optimized.

[0074] Example 2 Specific Implementation

[0075] 1. Specific information:

[0076] (1) Original forecast data:

[0077] The specific model product used in this invention is the CMA-BJ (obtained by Beijing Urban Meteorological Institute) hourly precipitation forecast product with a spatial resolution of 0.03°x0.03° (3km), which is reported daily at 02:00, 05:00, 08:00, 11:00, 14:00, 17:00, 20:00 and 23:00.

[0078] Considering the 4-6 hour lag in forecast lead time and arrival time for each CMA-BJ model (i.e., the arrival time is generally 4-6 hours later than the lead time, and forecast products from the first 4-6 hours are unusable), and following the principle of maximizing the selection of ensemble forecast members, the ensemble forecast members for the 08:00-08:00 forecast include 7 members that started at 02:00 on the same day and 23:00, 20:00, 17:00, 14:00, 11:00, and 08:00 on the previous day; the ensemble forecast members for the 20:00-20:00 forecast include 7 members that started at 14:00, 11:00, 08:00, 05:00, 02:00 on the same day and 23:00 and 20:00 on the previous day. Furthermore, since the forecast effectiveness covered by forecasts at different lead times varies, for any forecast time, all forecast members with forecast results within that time are included in the ensemble. The forecast lead time and corresponding forecast time for each forecast member are shown in Table 2.

[0079] Table 2. Forecast lead time and corresponding forecast time for each CMA-BJ forecast time.

[0080]

[0081] (2) Correcting forecast products

[0082] The newly developed time-lag ensemble probability matching correction product is an hourly precipitation ensemble probability matching forecast product with a 24-hour lead time, reported daily from 08:00 and 20:00.

[0083] (3) Inspect the comparison products

[0084] Based on the forecast lead times of each ensemble member (Table 2), the nearest forecast members that can cover the entire forecast period from 08:00 to 08:00 and from 20:00 to 20:00 are identified as the members that started forecasting at 02:00 and 11:00 on the same day, respectively. Therefore, the forecast lead time of CMA-BJ from 02:00 for forecasts 7 to 30 is used as the forecast verification control product for the 08:00 ensemble forecast, and the forecast lead time of the forecast from 11:00 for forecasts 10 to 33 is used as the forecast verification control product for the 20:00 to 20:00 ensemble forecast, so as to verify the correction effect of the probability matching fusion forecast.

[0085] 2. Specific operating steps

[0086] For the 0-24 hour forecast lead time reported daily from 08:00 and 20:00 in 2022, modeling and correction were carried out according to the following steps.

[0087] (1) Select typical precipitation cases in 2021 to test and determine the training period before the best forecast.

[0088] The training period was selected from 7 to 30 days before the forecast. Modeling correction was carried out according to the steps (1) to (4) in the implementation plan. The set probability matching fusion forecast of typical cases was obtained and the forecast was verified. The training period length was determined by the improvement of the TS score of heavy precipitation. Finally, the 14 days before the forecast was determined as the sliding training period for establishing the correction model.

[0089] (2) Model test and verification during the flood season of 2022, from June 1 to August 31.

[0090] The 14 days prior to the daily ensemble forecast were selected as the training period for modeling. Rolling modeling correction was carried out according to steps (1)-(4) in the implementation plan. Hourly precipitation ensemble probability matching fusion forecasts for 08:00-08:00 and 20:00-20:00 were generated and processed into 3km gridded products, which were then connected to the MICAPS business display platform. The 2022 flood season trial forecast products were tested as a whole, and typical heavy precipitation cases were selected for individual case testing to test the model correction effect.

[0091] 3. Forecast effect display

[0092] 3.1 Overall Statistical Test of Forecasts from June 1 to August 31, 2022 (Flood Season)

[0093] 3.1.1 Verification of 24-hour precipitation forecast

[0094] By sequentially summing the hourly precipitation forecasts and actual precipitation data for each day from June 1st to August 31st, 2022 (08:08-08:00 and 20:00-20:00), 24-hour precipitation forecasts and actual precipitation data for each day from 08:08-08:00 to 20:00 are obtained. The daily precipitation forecast and actual precipitation sequences are then arranged sequentially, and the overall samples for 08:08-08:00 and 20:00 from June 1st to August 31st are tested. The results are as follows:

[0095] (1) Verification of weather forecast

[0096] The accuracy of weather forecasts from June 1st to August 31st was tested using near-time start forecasts (08:00-08:00 start; 20:00-20:00 start) and ensemble probability matching fusion forecasts (corrected forecasts). The results are as follows: Figure 5 As shown, the ensemble probability matching fusion forecasts from 08:08 and 20:00 show a positive skill improvement in the accuracy of weather forecasts, with the accuracy of weather forecasts being slightly higher than the results of forecasts from closer times.

[0097] (2) Evaluation of Rainstorm and Heavy Precipitation Forecast Scoring

[0098] The TS score and BIAS score of the 24-hour heavy rainfall forecast were calculated to test the correction ability of the ensemble probability matching forecast for heavy rainfall and torrential precipitation. The overall test was performed on the period from June 1st to August 31st, 2022, and the results are as follows: Figure 6 As shown, the 24-hour heavy rainfall forecast and 24-hour strong precipitation forecast obtained from ensemble probability matching are significantly higher than those obtained from forecasts originating closer to the current time, demonstrating a positive skill improvement. Figure 6 a, c), the BIAS score is also significantly closer to 1 than the result reported at the nearest time. Figure 6 (b, d) This indicates that the magnitude and location of the rainstorm forecasts obtained through ensemble probability matching are closer to the actual situation, demonstrating a good correction effect. Overall, the corrected ensemble probability matching forecasts effectively improve the TS score of 24-hour rainstorm and heavy precipitation forecasts, reduce area bias, and exhibit significant positive correction techniques.

[0099] 3.1.2 Verification of hourly precipitation forecasts

[0100] The hourly precipitation forecasts and actual precipitation data for June 1st to August 31st, 2022, from 08:00 to 08:00 and 20:00 to 20:00 daily, are arranged sequentially. The hourly forecasts for June to August are then verified, and the verification results are as follows:

[0101] (1) Accuracy of weather forecast and average relative error

[0102] Figure 7The results of the verification of the accuracy and average relative error of hourly precipitation forecasts from June 1 to August 31, 2022, show that the accuracy of the 1-hour precipitation forecast using ensemble probability matching fusion forecasts is slightly better than that of forecasts starting closer to the current time. Figure 7 a) The 1-hour average relative error decreased slightly ( Figure 7 (b) This indicates that ensemble probability matching fusion has a positive correction technique improvement for hourly weather forecasts and error correction.

[0103] (2) Scoring and verification of short-term heavy precipitation forecast

[0104] The hourly heavy precipitation forecast verification results for June 1st to August 31st, 2022 are as follows: Figure 8 As shown, the 1-hour heavy precipitation TS score of the ensemble probability matching fusion forecast ( Figure 8 a) and BIAS score ( Figure 8 b) are all significantly higher than the results reported at the nearest time, and the BIAS score is closer to 1 ( Figure 8 (b) This indicates that the revised forecast has a good correction effect on the magnitude and location of short-term heavy precipitation, and has significantly improved the forecasting capability for short-term heavy precipitation.

[0105] 3.2 Typical Heavy Rainfall Case Study - Heavy Rainfall Process from Daytime to Nighttime on August 18, 2022

[0106] 3.2.1 Actual situation and forecast of this process

[0107] From 08:00 on August 18 to 08:00 on August 19, 2022, most areas of Tianjin experienced heavy to torrential rain, with some areas experiencing extremely heavy rain. The main areas experiencing heavy to extremely heavy rain were concentrated in the urban area and the four suburban districts, Baodi, Ninghe, and Binhai New Area. Figure 9 a). Based on the forecast data, both the near-term forecast (starting at 11:00 on the 5th) (before correction) and the ensemble probability matching fusion forecast (after correction) were able to predict the overall heavy rain event. However, there was a clear oversight regarding the heavy to torrential rain that occurred in Jizhou in the north. Figure 9 (b, c). Comparative analysis of the two forecasts shows that the ensemble probability matching fusion forecast is more similar to the actual distribution, accurately predicting the areas of heavy to torrential rain in the central and eastern regions. It improves the underreporting of heavy rain in the central and southern regions (urban area, Xiqing, Jinnan, and central Binhai New Area) and the false alarms of heavy rain in the northwest (Wuqing) in near-term forecasts. Figure 9 b, c).

[0108] 3.2.2 Statistical Test for Forecasting

[0109] Analysis of the 24-hour forecast verification scores shows that the TS scores for ensemble probability matching fusion forecasts of 24-hour heavy rainfall and 24-hour torrential precipitation exhibit a highly significant positive skill improvement compared to the results from closer time-to-time forecasts, with TS scores reaching 0.542 and 0.674 respectively, representing increases of 129.4% and 116%. Figure 10 a, c); The BIAS score for 24-hour heavy rainfall and torrential precipitation improved from less than 1 in the immediate forecast to greater than 1, but the deviation of the BIAS score for 24-hour heavy rainfall from 1 did not change significantly. Figure 10 b), the deviation of the BIAS score for 24-hour heavy precipitation from 1 is slightly larger ( Figure 10 (d) This indicates that the area deviation of the ensemble probability matching fusion forecast for 24-hour heavy rainfall has not changed significantly, while the area deviation of the forecast for torrential rainfall exceeding 100 mm has increased slightly. Overall, the ensemble probability matching fusion forecast can effectively improve the forecasting skills for 24-hour heavy rainfall and torrential rainfall during this heavy precipitation event, and the correction effect for 24-hour heavy rainfall forecast is better.

[0110] Analyzing the overall test results of the hourly forecasts, it can be seen that the accuracy of the 1-hour weather forecast in the corrected ensemble probability matching fusion forecast has slightly increased, the 1-hour average relative error has slightly decreased, and the hourly corrected forecasts are closer to the actual conditions. Figure 11 (a, b) Analyzing the short-term heavy precipitation forecast, the TS score for 1-hour heavy precipitation increased, showing a positive trend, while the BIAS score for 1-hour heavy precipitation did not change much, with BIAS around 1, and the area deviation from the actual situation was very small.

[0111] Overall, for this event, the ensemble probability matching forecast had a certain correction effect on both the 24-hour and hourly forecasts. The correction effect was more significant for the magnitude of the 24-hour heavy rain and precipitation forecasts (TS scores increased significantly), but the area deviation for heavy rainstorms increased. For the hourly forecasts, both the clear / rain forecast and the short-term heavy precipitation forecast scores had a certain positive skill correction effect.

[0112] Example 3

[0113] The revised forecast products have been processed into 0.3° x 0.3° (3km) grid forecast products and saved in a format that can be called by the MICAPS operational forecast system. Each file contains a single forecast lead time revised forecast product.

[0114] File name such as Figure 12 As shown, the file name is 2022060108.01, where 2022060108 represents the reporting start time, that is, 08:00 on June 1, 2022, and .01 represents the forecast period, that is, the forecast result one hour past 09:00 on June 1, 2022.

[0115] The file opening style is as follows: Figure 13 As shown, the left image is a precipitation distribution map, and the right image is a display of the precipitation forecast map for that moment from the MICAPS operational forecasting system.

[0116] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for correcting hourly time-lag ensemble precipitation forecasts based on weighted coefficients, characterized in that: Includes the following steps: Step 1: Develop set weighting coefficients Wi based on Taylor test results; the weighting coefficients Wi are mainly developed using Taylor test results; the Taylor plot selects the spatial correlation coefficient R between the simulated field and the observed field, and the ratio σ of the standard deviations of the simulated field y-field and the observed field x-field. y / σ x The three variables, namely the root mean square error E' after removing systematic errors, are used to comprehensively examine the differences between the simulated field and the observed field. In the Taylor diagram, according to the triangle cosine theorem, we can obtain the following relationship: ， Among them, E' 2 / σ x 2 E' represents the distance between the simulated field Mod. and the observed field Obs. 2 / σ x 2 The smaller the value, the closer the simulated field Mod. is to the observed field Obs.; therefore, the variable E' / σ x It can be used to evaluate the difference between the ensemble's earlier forecasts and the actual field, E' / σ x The smaller the value, the larger the given weight coefficient should be, and vice versa; Based on the above principle, construct the variable IA=f(E' / σ) x )=σ x / E', thus ensuring that IA and σ x / E' exhibits an inverse correlation, i.e., E' / σ x Smaller ensemble members have higher weights. Thus, the weight coefficients Wi for each ensemble's forecast members, where i = 1, 2, ..., n, are the forecast members of each ensemble. ； Step 2: Use a time lag ensemble forecasting method based on weighted coefficients to aggregate forecast members that started at different times; the time lag ensemble forecasting method comprehensively considers the model forecast results of the same time point with different start times and different forecast lead times, and aggregates the forecast members through different weighted coefficients to obtain the ensemble precipitation forecast value; Given the weight coefficients Wi of forecast members at different start times, where i = 1, 2, 3... n are the forecast members of each ensemble, the time-lag ensemble precipitation forecast P value is: ； Step 3: Correct the total 24-hour precipitation and hourly precipitation variation and magnitude using the forecast-observation probability matching method based on GAMMA cumulative probability analysis; the forecast-observation probability matching method is to assume that for a certain precipitation threshold Ti, its precipitation probability or cumulative precipitation probability Pr(T'i) is the same as the observed probability Po(Ti), and the actual precipitation Ti is the frequency matching correction value of the model forecast precipitation T'i; Step 4: Verify the precipitation forecast.

2. The hourly time-lag ensemble precipitation forecast correction method based on weighted coefficients according to claim 1, characterized in that: In step 3, since the cumulative probability distribution of precipitation is non-normal, the Gamma cumulative probability distribution function is as follows: ； In the formula, α>0, β>0, where α is the shape parameter and β is the scale parameter. Precipitation; α and β are obtained by maximum likelihood estimation; sample mean. Variance s 2 The relationship with parameters α and β is as follows: .

3. The hourly time-lag ensemble precipitation forecast correction method based on weighted coefficients according to claim 1, characterized in that: In step 4, the precipitation over 1 hour and 24 hours is mainly tested, and the following precipitation test statistics are selected: ① Rain / sunshine forecast accuracy: PC = (NA + ND) / (NA + NB + NC + ND) × 100% In the formula, NA represents the number of stations with correct precipitation forecasts, NB represents the number of stations with no forecasts, NC represents the number of stations with missed forecasts, and ND represents the number of stations with correct no precipitation forecasts. The verification parameters include the accuracy rate of 1-hour and 24-hour weather forecasts. ② Heavy precipitation testing includes TS score and BIAS score. TS score: TS = NA / (NA + NB + NC) BIAS score: BIAS = (NA + NB) / (NA + NC) In the formula, The number of stations with accurate precipitation forecasts, Empty station count, This represents the number of stations that were missed in reporting. The number of stations with correct forecasts of no heavy precipitation; The TS score is used to verify the effectiveness of precipitation level forecasts; the BIAS score is used to verify the area deviation between the forecast and the actual situation. BIAS=1 indicates that the two areas are equal, >1 indicates that the forecast area is too large, and <1 indicates that the forecast area is too small. ③ Average relative error of precipitation: ; In the formula, To observe precipitation, The corresponding forecast precipitation is used. When both the actual and forecast values ​​are 0, the relative error is 0. In this invention, the verification parameter is the average relative error of 1-hour precipitation.

Images