CMA-BJ-EN method and system for convection-discriminating ensemble forecast

Abstract

The application discloses a CMA-BJ-EN convection-resolvable ensemble prediction method and system, which comprises the following steps: collecting observation data of a target area, performing global ensemble data preprocessing on the observation data to obtain an initial analysis field, and performing preprocessing and disturbance on the observation data to obtain disturbed observation data; performing ensemble data assimilation on the initial analysis field and the observation data based on three-dimensional variation, controlling a prediction initial value, a disturbed prediction initial value and obtaining side boundary conditions, performing integral prediction by adopting a convection-resolvable numerical mode, and introducing random physical process disturbance in the integral process to obtain a prediction result of an ensemble member; and performing seasonal post-processing on the prediction result to generate an ensemble prediction product.

Description

Technical Field

[0001] This invention relates to the field of resolvable ensemble forecasting technology, and more particularly to the CMA-BJ-EN resolvable ensemble forecasting method and system. Background Technology

[0002] Ensemble forecasting, by considering uncertainties in initial conditions, physical processes, and boundary conditions, provides probabilistic forecasts and assesses forecast errors, and has become a core technology in modern numerical weather prediction. With the improvement of computing power, convective-permissive ensemble forecasting systems based on 1–4 km resolution and without the need for cumulus convection parameterization schemes have gradually become an international research hotspot. They can explicitly distinguish small- and medium-scale hazardous weather events and improve the forecasting ability for deep convective systems and localized heavy precipitation.

[0003] However, existing regional ensemble forecasting systems have shortcomings in constructing initial perturbations. Currently, the mainstream methods fall into two categories: one is to directly downscale the global ensemble forecasting system to the regional model. While this method can guarantee a certain level of ensemble dispersion, it is prone to introducing large-scale errors due to the insufficient resolution of the global model; the other is to construct the perturbation field from the regional model itself, which often results in insufficient dispersion and large small-scale errors. Neither method can effectively improve the accuracy of the initial field while maintaining ensemble dispersion. Furthermore, model uncertainties caused by parameterization and truncation errors in model physics processes, as well as the significant impact of lateral boundary condition perturbations on smaller simulated areas, also severely restrict further improvements in ensemble forecasting techniques. In North China, the occurrence and development of extreme weather events are highly uncertain, and the foundation for probabilistic forecasting research in this region is relatively weak. Existing ensemble systems remain insufficient in providing refined local disaster early warning information.

[0004] Therefore, it is urgent to construct a convective-resolved ensemble forecasting method suitable for severe convective forecasting, effectively combining the advantages of global ensemble forecasting and regional data assimilation, and reasonably characterizing the uncertainties of initial values, models, and lateral boundaries. Summary of the Invention

[0005] The purpose of this invention is to provide a CMA-BJ-EN convection resolvable ensemble forecasting method.

[0006] To achieve the above objectives, the present invention is implemented according to the following technical solution: This invention includes the following steps: Observational data of the target area are collected, and global ensemble data preprocessing is performed on the observational data to obtain an initial analysis field. The observational data are then preprocessed and perturbed to obtain perturbed observational data. The observational data includes ground observation data, radiosonde data, ship data, buoy data, aircraft observations, data from nationwide densified automatic weather stations, and small-ball wind measurement data. The initial analysis field contains 21 ensemble members, including 1 control forecast member and 20 perturbed members. The initial analysis field and the observation data are assimilated using ensemble data based on three-dimensional variation. Initial values ​​for control forecasts and perturbation forecasts are obtained, and lateral boundary conditions are acquired. Integral forecasts are performed using a numerical model at a convection-resolvable scale, and random physical process perturbations are introduced during the integration process to obtain the forecast results for ensemble members. The horizontal resolution of the convection-resolvable scale is 1–4 km. The numerical model is a WRF model with 59 vertical layers, covering an area of ​​35.5°N–46.3°N and 105.2°E–122.4°E. The forecast results are subjected to seasonal post-processing to generate an ensemble forecast product; the forecast elements of the ensemble forecast product include precipitation, 10-meter wind, 2-meter temperature, cloud cover, convective effective potential energy (CAPE), convection suppression, K index, and so on.

[0007] Furthermore, the initial analysis field contains 21 ensemble members, including 1 control forecast member and 20 disturbance members.

[0008] Furthermore, the method for assimilating the relevant observation data includes: Dynamic downscaling of NCEP global ensemble forecast data was performed to obtain initial background fields for the CMA-BJ-EN model downscaling. Data was then assimilated with multiple sets of observations using a three-dimensional variational module to obtain initial analysis values. The NCEP global ensemble forecast data was GEFS.

[0009] Furthermore, the dynamic downscaling involves using the forecast field forwarded by the GEFS as the initial value and the subsequent forecast field of the GEFS as the lateral boundary, and obtaining the initial background field of the regional ensemble forecast through horizontal and vertical interpolation.

[0010] Further, the observation data is perturbed by: obtaining the observed values ​​and corresponding observation errors of each observed variable in the conventional observation data, generating multiple random numbers, each of which takes a value within a preset range; multiplying the observation error corresponding to each observed variable by the random number to obtain multiple observation perturbations; and superimposing the observed values ​​with each of the observation perturbations to obtain multiple sets of perturbed observation members.

[0011] Furthermore, the random number is generated by a random number generator and has a value range of [-2, 2]. The multiple sets of perturbation observation members are 20 sets, which together with the original observation constitute 21 sets of observation files, which are used for the assimilation of the 21 set members.

[0012] Furthermore, the lateral boundary conditions are obtained: the forecast fields of each ensemble member of the global ensemble forecast system are directly used as the lateral boundary conditions of the corresponding ensemble member of the regional ensemble forecast. Since the forecast fields of each ensemble member are inconsistent, lateral boundary disturbances are formed.

[0013] Furthermore, the stochastic physical process perturbation is the Stochastic Physical Process Tendency Perturbation Method (SPPT), which applies a stochastic force to the total tendency term after the calculation of all physical process parameterization schemes in the numerical model is completed.

[0014] Furthermore, the SPPT uses the same random type r(x,y,t) to perturb different forecast variables. The random type r(x,y,t) is a Gaussian random function with spatiotemporal correlation and a value range of [-1,1], which is determined by the time of loss of correlation, the space of loss of correlation, and the grid standard deviation.

[0015] Secondly, the CMA-BJ-EN convection-resolved ensemble forecasting system includes: Data collection module: The data collected includes the global ensemble forecast initial value field and forecast field at two times a day, as well as real-time observation data including conventional surface, radiosonde, small ball wind measurement, and aircraft reports; Initialization startup module: Starts twice a day, with the reporting start times being 20:00 and 08:00; uses the 12-hour forecast from the NCEP global ensemble forecast that is 12 hours ahead as the initial value, and uses the subsequent forecast field as the lateral boundary, and obtains the initial value and boundary of the regional ensemble forecast through horizontal and vertical interpolation; The ensemble data assimilation initial value generation module uses the global ensemble interpolated field as the background field, combines the observational disturbances formed by multiple sets of observational data, assimilates the observational data, and generates multiple sets of initial value analysis fields for the ensemble members through the variational assimilation module; the assimilated observational data includes conventional ground, radiosonde, small-sphere wind measurement, and aircraft reports. Mode integration module: Each set member is integrated using the WRF mode over 24 hours, and a random physical process perturbation method is introduced to perturb the mode; Ensemble forecast post-processing module: generates a series of ensemble forecast products, including stamp maps, precipitation probability maps, precipitation quantiles, and single-station box plots.

[0016] The beneficial effects of this invention are: This invention relates to a CMA-BJ-EN resolvable ensemble forecasting method and system for convection. Compared with existing technologies, this invention has the following technical advantages: This invention combines global ensemble forecast dynamic downscaling with regional three-dimensional variational ensemble normalization, introducing observational and stochastic physical process perturbations. While maintaining ensemble dispersion, it improves the accuracy of the initial field, achieving refined probabilistic forecasts at a 3km resolvable convection scale. It effectively reduces the root mean square error of 2m temperature by 9.32% and the root mean square error of 10m wind speed by 9.01%, and significantly improves precipitation forecasting skill: the TS scores for 0.1mm, 4mm, and 15mm precipitation thresholds are increased by 29.4%, 8.07%, and 25.47%, respectively. This allows for more accurate simulation of small- and medium-scale severe weather, providing more refined uncertainty information for local severe convective weather warnings. Attached Figure Description

[0017] Figure 1 This is a flowchart of the steps of the CMA-BJ-EN convection resolvable set prediction method of the present invention. Detailed Implementation

[0018] The present invention will be further described below through specific embodiments. The illustrative embodiments and descriptions herein are used to explain the present invention, but are not intended to limit the present invention.

[0019] The CMA-BJ-EN resolvable ensemble forecasting method and system of the present invention includes the following steps: like Figure 1 As shown, this embodiment includes the following steps: Observational data of the target area are collected, and global ensemble data preprocessing is performed on the observational data to obtain an initial analysis field. The observational data are then preprocessed and perturbed to obtain perturbed observational data. The observational data includes ground observation data, radiosonde data, ship data, buoy data, aircraft observations, data from nationwide densified automatic weather stations, and small-ball wind measurement data. The initial analysis field contains 21 ensemble members, including 1 control forecast member and 20 perturbed members. The initial analysis field and the observation data are assimilated using ensemble data based on three-dimensional variation. Initial values ​​for control forecasts and perturbation forecasts are obtained, and lateral boundary conditions are acquired. Integral forecasts are performed using a numerical model at a convection-resolvable scale, and random physical process perturbations are introduced during the integration process to obtain the forecast results for ensemble members. The horizontal resolution of the convection-resolvable scale is 1–4 km. The numerical model is a WRF model with 59 vertical layers, covering an area of ​​35.5°N–46.3°N and 105.2°E–122.4°E. The forecast results are subjected to seasonal post-processing to generate an ensemble forecast product; the forecast elements of the ensemble forecast product include precipitation, 10-meter wind, 2-meter temperature, cloud cover, convective effective potential energy (CAPE), convection suppression, K index, and so on.

[0020] In this embodiment, the initial analysis field contains 21 ensemble members, including 1 control forecast member and 20 disturbance members.

[0021] In this embodiment, the method for assimilating the relevant observation data includes: Dynamic downscaling of NCEP global ensemble forecast data was performed to obtain initial background fields for the CMA-BJ-EN model downscaling. Data was then assimilated with multiple sets of observations using a three-dimensional variational module to obtain initial analysis values. The NCEP global ensemble forecast data was GEFS.

[0022] In this embodiment, the dynamic downscaling involves using the forecast field initiated by the GEFS as the initial value and the subsequent forecast field of the GEFS as the lateral boundary, and obtaining the initial background field of the regional ensemble forecast through horizontal and vertical interpolation.

[0023] In this embodiment, the observation data is perturbed by: obtaining the observed values ​​and corresponding observation errors of each observed variable in the conventional observation data, generating multiple random numbers, each of which takes a value within a preset range; multiplying the observation error corresponding to each observed variable by the random number to obtain multiple observation perturbations; and superimposing the observed values ​​with each of the observation perturbations to obtain multiple sets of perturbed observation members.

[0024] In this embodiment, the random number is generated by a random number generator and has a value range of [-2, 2]. The multiple sets of perturbation observation members are 20 sets, which together with the original observation constitute 21 sets of observation files, which are used for the assimilation of the 21 set members.

[0025] In this embodiment, the lateral boundary conditions are obtained as follows: the forecast fields of each ensemble member of the global ensemble forecast system are directly used as the lateral boundary conditions of the corresponding ensemble member of the regional ensemble forecast. Since the forecast fields of each ensemble member are inconsistent, lateral boundary disturbances are formed.

[0026] In this embodiment, the random physical process perturbation is the random physical process tendency perturbation method SPPT, which applies random forcing to the total tendency term after the calculation of all physical process parameterization schemes in the numerical model is completed.

[0027] In this embodiment, the SPPT uses the same random type r(x,y,t) to perturb different forecast variables. The random type r(x,y,t) is a Gaussian random function with spatiotemporal correlation and a value range of [-1,1], which is determined by the time of loss of correlation, the space of loss of correlation, and the grid standard deviation.

[0028] Secondly, the CMA-BJ-EN convection-resolved ensemble forecasting system includes: Data collection module: The data collected includes the global ensemble forecast initial value field and forecast field at two times a day, as well as real-time observation data including conventional surface, radiosonde, small ball wind measurement, and aircraft reports; Initialization startup module: Starts twice a day, with the reporting start times being 20:00 and 08:00; uses the 12-hour forecast from the NCEP global ensemble forecast that is 12 hours ahead as the initial value, and uses the subsequent forecast field as the lateral boundary, and obtains the initial value and boundary of the regional ensemble forecast through horizontal and vertical interpolation; The ensemble data assimilation initial value generation module uses the global ensemble interpolated field as the background field, combines the observational disturbances formed by multiple sets of observational data, assimilates the observational data, and generates multiple sets of initial value analysis fields for the ensemble members through the variational assimilation module; the assimilated observational data includes conventional ground, radiosonde, small-sphere wind measurement, and aircraft reports. Mode integration module: Each set member is integrated using the WRF mode over 24 hours, and a random physical process perturbation method is introduced to perturb the mode; Ensemble forecast post-processing module: Generates a series of ensemble forecast products, including stamp maps, precipitation probability maps, precipitation quantiles, and single-station box plots; In the actual evaluation, in January 2022, the CMA-BJ-EN system, funded by the Development and Application Program of Ensemble Short-Term Numerical Prediction Systems in High-Latitude Regions, completed system testing. The CMA-BJ-EN system involves probabilistic forecast products for precipitation, 10m wind, 2m temperature, and cloud cover. The list of data required for observation is shown in Table 1. Table 1 List of Required Documents

[0029] The main product catalog of CMA-BJ-EN is shown in Table 2: Table 2. Main Product Catalog of CMA-BJ-EN

[0030] Table 3 lists the operation phases and core counts of the ensemble system starting at 10:20 AM. The peak resource consumption is during the model integration and simultaneous post-processing phases. Each ensemble member requires 360 cores, and the parallel computation of 21 members during the model integration phase requires 360 × 21 = 7560 cores. Considering that post-processing is performed concurrently, 10 × 21 = 210 cores are needed. Therefore, the peak phase requires a total of 7770 cores, approximately 8000 cores, which can ensure the timeliness of the ensemble forecast results. Table 3. CMA-BJ-EN system operation phase and core count starting at 10:20 AM

[0031] The system initiates reporting twice daily, at 10:20 AM and 10:20 PM, with a total uptime of approximately 1 hour and 30 minutes. This means that the system experiences two peak periods of computational resource usage each day. The operational status of the CMA-BJ-EN system is shown in Table 4. Table 4. Operational Status of CMA-BJ-EN System

[0032] A batch experiment lasting more than one month was conducted, with the NCEP Global Ensemble Forecasting System (NCEP-GEFS) used as a control experiment. The ensemble forecasting performance of the two batch experiments was compared and analyzed. The batch experiment period was from June 17 to July 25, 2021, with a forecast lead time of 24 hours. The RMSE score of near-surface elements and the Ts score of 6-hour cumulative precipitation were examined every 6 hours, and the average of TS and RMSE within 24 hours was used as the final verification index of the system. Table 5 shows the percentage improvement of CMA-BJ-EN TS scores and near-surface element RMSE scores for different precipitation thresholds compared to GEFS: Table 5 shows the percentage improvement of CMA-BJ-EN TS scores and near-surface element RMSE scores for different precipitation thresholds compared to GEFS.

[0033] Table 5 shows the 6-hour cumulative precipitation TS scores for CMA-BJ-EN and NCEP-GEFS with precipitation thresholds of 0.1 mm, 4 mm, and 14 mm, as well as the percentage improvement of CMA-BJ-EN compared to NCEP-GEFS, and the percentage improvement of near-surface element RMSE and CMA-BJ-EN compared to GEFS. As shown in Table 5, the TS scores of CMA-BJ-EN at precipitation thresholds of 0.1 mm, 4 mm, and 15 mm are 0.342, 0.278, and 0.166, respectively, while those of NCEP-GEFS are 0.264, 0.257, and 0.133, respectively. The improvement rates of CMA-BJ-EN's TS scores relative to GEFS at precipitation thresholds of 0.1 mm, 4 mm, and 15 mm are (0.342-0.26425) / 0.26425*100=29.4%, (0.278-0.25725) / 0.25725*100=8.07%, and (0.16625-0.1325) / 0.1325*100=25.47%, respectively. Table 5 shows that the RMSE of CMA-BJ-EN's T2M and WIND10M are 2.14K and 2.12m / s, respectively, while those of NCEP-GEFS are 2.36K and 2.33m / s, respectively. The improvement rates of CMA-BJ-EN's T2M and WIND10M compared to NCEP-GEFS are (2.36-2.14) / 2.36*100=9.32% and (2.33-2.12) / 2.33*100=9.01%, respectively. In summary, the CMA-BJ-EN system conducted a batch ensemble forecast experiment from June 17, 2021 to July 25, 2021. Compared with the near-surface and precipitation verification results of the NCEP global ensemble forecast system, the CMA-BJ-EN system can effectively reduce the RMSE of 2m temperature and 10m wind speed (actual greater than 5m / s), with improvement rates of 9.32% and 9.01%, respectively. At the same time, it can effectively improve the Ts score of precipitation forecasts, with improvement rates of 29.4%, 8.07%, and 25.47% for Ts scores of 0.1mm, 4mm, and 15mm precipitation thresholds, respectively.

[0034] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. The CMA-BJ-EN resolvable ensemble forecasting method for convection, characterized in that, Includes the following steps: Collect observation data of the target area, perform global ensemble data preprocessing on the observation data to obtain the initial analysis field, and perform preprocessing and perturbation on the observation data to obtain perturbed observation data; The initial analysis field and the observation data are assimilated based on the ensemble data of three-dimensional variation, the initial values ​​of the forecast and the initial values ​​of the perturbation forecast are controlled and the lateral boundary conditions are obtained. The numerical model with a convective resolvable scale is used for integral forecasting, and random physical process perturbation is introduced in the integration process to obtain the forecast results of the ensemble members. The forecast results are subjected to seasonal post-processing to generate ensemble forecast products.

2. The CMA-BJ-EN convection resolvable ensemble forecasting method according to claim 1, characterized in that, The initial analysis field contains 21 ensemble members, including 1 control forecast member and 20 disturbance members.

3. The CMA-BJ-EN convection resolvable ensemble forecasting method according to claim 1, characterized in that, Methods for assimilating the relevant observation data include: Dynamic downscaling of NCEP global ensemble forecast data was performed to obtain initial background fields for the CMA-BJ-EN model downscaling. Data was then assimilated with multiple sets of observations using a three-dimensional variational module to obtain initial analysis values. The NCEP global ensemble forecast data was GEFS.

4. The CMA-BJ-EN convection resolvable ensemble forecasting method according to claim 3, characterized in that, The dynamic downscaling involves using the forecast field forwarded by the GEFS as the initial value and the subsequent forecast field of the GEFS as the lateral boundary, and obtaining the initial background field of the regional ensemble forecast through horizontal and vertical interpolation.

5. The CMA-BJ-EN convection resolvable ensemble forecasting method according to claim 4, characterized in that, The observation data is perturbed by: obtaining the observed values ​​and corresponding observation errors of each observed variable in the conventional observation data, generating multiple random numbers, each of which takes values ​​within a preset range; multiplying the observation error corresponding to each observed variable by the random number to obtain multiple observation perturbations; and superimposing the observed values ​​with each of the observation perturbations to obtain multiple sets of perturbed observation members.

6. The CMA-BJ-EN convection resolvable ensemble forecasting method according to claim 5, characterized in that, The random number is generated by a random number generator and has a value range of [-2, 2]. The multiple sets of perturbation observation members are 20 sets, which together with the original observation constitute 21 sets of observation files, which are used for the assimilation of the 21 set members.

7. The CMA-BJ-EN convection resolvable ensemble forecasting method according to claim 1, characterized in that, Obtaining the lateral boundary conditions: The forecast fields of each ensemble member of the global ensemble forecast system are directly used as the lateral boundary conditions of the corresponding ensemble member of the regional ensemble forecast. Since the forecast fields of each ensemble member are inconsistent, lateral boundary disturbances are formed.

8. The CMA-BJ-EN convection resolvable ensemble forecasting method according to claim 1, characterized in that, The stochastic physical process perturbation is the Stochastic Physical Process Tendency Perturbation Method (SPPT), which applies a stochastic force to the total tendency term after the calculation of all physical process parameterization schemes in the numerical model is completed.

9. The CMA-BJ-EN convection resolvable ensemble forecasting method according to claim 9, characterized in that, The SPPT uses the same random type r(x,y,t) to perturb different forecast variables. The random type r(x,y,t) is a Gaussian random function with spatiotemporal correlation and a value range of [-1,1], which is determined by the time of loss of correlation, the space of loss of correlation, and the grid standard deviation.

10. The CMA-BJ-EN convection-resolved ensemble forecasting system, for performing the method according to any one of claims 1-9, characterized in that, include: Data collection module: The data collected includes the global ensemble forecast initial value field and forecast field at two times a day, as well as real-time observation data including conventional surface, radiosonde, small ball wind measurement, and aircraft reports; Initialization startup module: Starts twice a day, with the reporting start times being 20:00 and 08:00; uses the 12-hour forecast from the NCEP global ensemble forecast that is 12 hours ahead as the initial value, and uses the subsequent forecast field as the lateral boundary, and obtains the initial value and boundary of the regional ensemble forecast through horizontal and vertical interpolation; The ensemble data assimilation initial value generation module uses the global ensemble interpolated field as the background field, combines the observational disturbances formed by multiple sets of observational data, assimilates the observational data, and generates multiple sets of initial value analysis fields for the ensemble members through the variational assimilation module; the assimilated observational data includes conventional ground, radiosonde, small-sphere wind measurement, and aircraft reports. Mode integration module: Each set member is integrated using the WRF mode over 24 hours, and a random physical process perturbation method is introduced to perturb the mode; Ensemble forecast post-processing module: generates a series of ensemble forecast products, including stamp maps, precipitation probability maps, precipitation quantiles, and single-station box plots.

Images