Numerical Weather Prediction Activities at Republic of Korea - Operational Viewpoint

1. Early period

The numerical weather prediction (NWP) plays an increasingly important role on daily weather forecasting. Most of the advanced meteorological centres nowdays operate various NWP models ranging from storm-scale to ultra-long scale.

Research and development in NWP at Korea goes back to early 1980's. Few scientists developed limited area models (LAMs) based on quasi-geostrophic dynamics. The multi-level baroclinic model was mainly used for understanding of large-scale dynamics. The models, however, could not be used for routine operation, since data processing facilities and associated pre/post processing softwares were not available at that time.

After few years passed, a 5-layer LAM and associated pre/post processing package were transferred to Korea Meteorological Administration (KMA) from Irish Met. Office. The model was based on primitive equations, and had been in operation with Cyber 932 mainframe computer. In early 1990's, when Cray-2S supercomputer was open to public sectors, more sophistigated model such as MM4 was in operation with batch mode at Cray-2S. The Cyber 932 was used as a front-end machine to control input/output data stream. The lateral boundary condition was provided by global spectral model of the Japan Meteorological Agency (JMA), and transmitted via Seoul-Tokyo point-to-point circuit. By that time KMA established an NWP division, and for the first time the NWP guidance were produced at routine basis. Since then various NWP models (mostly from the United States) have been investigated for the numerical simulation of atmospheric phenomena. The general circulation models were widely used for climate simulation at some universities around that time.

A high performance computer VPX 220/10 was installed at KMA in 1995. It has peak performance of 1.25 Gflops with 1 Gbyte memory and 45 Gbyte hard disk. A global spectral model and typhoon spectral model were transferred from JMA. Both models were implemented at VPX 220/10, and have been running in operation.

2. Present status

A global spectral model of T106L21 have been running in operational basis, since the observational data coverage had extended to the global domain with the cooperation of regional telecommunication hub, Japan. Along with the 4-dimensional data assimilation having 6 hour updating cycle, the model works as an integrator for the global data assimilation and prediction system (GDAPS), producing 120 hour forecasts for the large-scale atmospheric variables. The extension of model projection up to 10 days is in progress. The moisture analysis is corrected in the GDAPS with the input of cloud coverage at different vertical layers and cloud top temperature, derived from GMS-5 images.

The regional data assimilation and prediction system (RDAPS) also runs twice a day for 48 hour forecasts for northeast asian regions, with 6 hour assimilation cycle. The resolution is 40 km in x-y plane, and 23 levels in height.

The Korea Typhoon Model (KTM) is a limited-area spectral model with E70 truncation, and 8 levels in sigma coordinate. The first guess field and lateral boundary conditions are provided from GDAPS. The typhoon vortex are generated by empirical formulas. The asymmetric steering current in the ambient flow can be preserved during vogusing procedure. It produces a typhoon track forecasts for up to 60 hours when a typhoon approaches to Korean Peninsula beyond 25 degrees to North and 140 degrees from the East.

In addition to the existing typhoon model (KTM), two additional models have been implemented, and under test operation since June 1997; Geophysical Fluid Dynamics Korea (GFDK) model and Barotropic Adaptive Typhoon System (BATS). The GFDK has been adopted from the GFDL hurricane model. The BATS has been developed in KMA in cooperation with Department of Atmospheric Science at Kongju National University in Korea, utilizing the continuous dynamic grid adaptation technique. The Global spectral model also produces its own typhoon track forecasts since June 1997.

There are two types of applied models; diagnostic models for wave height and sea surface wind, and statistical models for max/min temperature and probability of precipitation. Sea surface wind at 10 m, and associated ocean wave height are diagnosed in the wave height model using the predicted wind at lowest level of RDAPS, with the mixed layer PBL model. Perfect Prog Method (PPM) are applied for the statistical interpretation of local weather elements from the RDAPS forecasts, such as maximum/ minimum temperatures for 68 stations, and Probability of Precipitation (PoP) over 18 regions up to 48 hours.

For further details on the NWP activities at KMA, please refer to the WWW (World Weather Watch)/GDPS (Global Data Processing Sytem) NWP progress report.

3. Prospects for the future

a. Understanding of meso-scale characteristic at Korean Peninsula

The skill of short-range weather forecasts during rainy season, so called Changma, is the lowest in a year. This is mostly due to the limited predictability associated with the meso-scale convective system along the stationary front with weak westerly current. The heavy rainfall during the Changma season is related to the interaction between multiple scales ranging from cloud to planetary waves. So it is essential to understand the physical mechanism in order to improve the precipitation forecasts in the NWP model. A special field observation project called KORMEX (Korea Monsoon Experiment) begins early 1998 with the cooperation of KMA/METRI (Meteorological Research Institute) and universities nationwide. The scientific understanding gained from the field study will contribute to the improvement on the physical parameterization in the NWP model.

b. Real-time simulation

So far, asynoptic information from radar and dense automatic weather station network are more effective for very-short range weather forecasts than NWP products It is urgently needed to assimilate those data to better represent the current state of atmosphere, and to improve the forecast skill for 6-18 hours in advance. Various meso-scale models as ARPS, LAPS, MM5, MAS, and Regional Spectral Model of NCEP (National Centers for Environmental Prediction, USA) are under investigation to select the most appropriate one for the storm-scale assimilation and prediction. A research is going on to incorporate satellite radiance to the analysis of moisture field, as used at JMA and Burea of Meteorology Australia. Four dimensional data assimilation with adjoint technique will be one of major area for development in the future.

c. Economy versus accuracy

It is found that reduced Gaussian-grid and semi-Lagrangian scheme provides savings in cpu time in global spectral model without significant degradation in the quality of forecasts. The extra cpu time can be allocated to more detailed physics calculations in the model. The research to adopt the aformentioned technique as used at ECMWF or BMRC (Bureau of Meteolorogical Research Centre, Australia) is currently going on.

d. Global versus storm-scale

The use of several models for different scales requires extensive man power for the research and development. If the computer resource is enough to operate a fine-mesh model for a global domain, it is desirable to developed a unified model for a small national centre. The typhoon model and LAMs can be unified into storm-scale model with coarse-grained global model. The resolution of global model itself can be downsized, so that even storm-scale circulation can be simulated. Ultimately all models can be unified into single global model, including GCMs. For the time being, however, GCM, global model, and storm-scale model can co-exist until computer technology and assimilation technique are advanced accordingly.

e. Upgrade of supercomputer

KMA is planing to procure new supercomputer by 1999, when it moves to new and intelligent building. It is expected that the new computer would have at least 100 times higher peak performance than the current one. The spacial and temporal resolution of both golbal and storm-scale models can be enhanced by factor of two. In addition, ensemble forecasts for medium-range will be available with low resolution global model. The research on long-range numerical prediction would be activated for practical purpose.

Acknowledgments: The author sincerely appreciate the comments and suggestions from Mr. Soon-Kab Chung, Director, and other staff of Numerical Weather Prediction Division, and Dr. Jae-Ho Oh, Director of Forecasting Research Laboratory at Meteorological Research Institute.