China Gridded Forecast System

China Gridded Forecast System

Xue Feng

（National Meteorological Center，China Meteorological Administration , Beijing 100081）

With development of modern numerical weather prediction (NWP) and massive computing technologies, the China Gridded Forecast System (CGFS) has been established for seamless forecasting in China since 2014. Now the system becomes a new-generation fundamental forecast operational system of the China Meteorological Administration (CMA). It can provide 1 -30 days gridded forecasts with the highest horizontal resolution of 0.05°×0.05° and over 20 predicted meteorological elements.

1. The Overview

The CGFS consists of Model Output Statistics, Forecast Blending & Merging, Verification sub-systems. All the operational and test data, analysis and forecast data, guided or revised data, are stored and processed centrally in National Weather Forecast Database (NWFD). NWFD use memory-accelerated techniques to improve the performance of forecast data accessing, processing, and persistence, it also synchronizes all data between national meteorological center and provinces bureaus remotely with China Integrated Meteorological Information Service System (CIMISS). while the data exchange service on the public cloud shares the products of meso-scale models run by provinces.

2. Model Output Statistics

Based on the GRAPES GFS, GRAPES-MESO, GRAPES-GEPS and other NWP models’ products, many statistics methods such as multiple linear regression, Bias correction, Kalman filter, RMOS are used in training a collective objective forecast model to get even better forecast skill. The forecast results will be store in NWFD and synchronized to the provincial bureaus through CIMISS to guide local forecasters subjective forecasting.

3. Blending & Merging

Blending & Merging is a set of event-triggered programs to blend different objective forecast results with subject forecast. With graphical tools, the forecaster can visualize and revise the forecast data, the event that modifications may causes subsequently and complicatedly data processing to keep all element forecast fields coordinated in space and time. Every time the latest hourly observation analysis field comes, the error between observation and forecast causes the forecast field in 24 hours to be revised to statistical regularity.

4. Verification

TS, ETS, Bias, accuracy rate and other traditional verification methods are used in gridded forecast verification. Besides, Neighborhood and MODE method are tried along with gridded observation analysis. The verification is near real-time routine to provide more immediate and rich verification information for forecasters. All verification data are online for analysis and evaluation to excavate the temporal and spatial distribution characteristics of the test error.

5. The Future

With the complexity of computing and the rapid growth of data, the methods and performance of the model post-processing will become sensitive and delicate. More big Data technology and tools such as private cloud, AI and GPU, will be applied to improve efficiency and increase elasticity. By 2020, the higher horizontal resolution, shorter update frequency, and more meteorological elements will be achieved in the gridded Forecast System.