Preprint, 16th AMS Conf. Weather Analysis and Forecasting
Phoenix, Arizona, January 1998
Explicit Realtime Operational Prediction of Deep Convection over Korea During the 1997 Summer Monsoon Season
Kyung-Sup Shin, Soon-Kab Chung, Sun-Yong Lee,
Hee-Dong Yoo, Dong-I1 Lee
Korea Meteorological Administration
Ming Xue, Keith Brewster, Gene Bassett, Seon Ki Park, Kelvin Droegemeier
Center for Analysis and Prediction of Storms, University of Oklahoma
Korea can be affected by various types of severe weather; among them, flash floods of meso-[beta] scale or smaller are most difficult to forecast. Flash floods often occur during the season known as Changma (Meiyu over China and Beiyu over Japan) in the early summer. Monitoring 400 real-time automated weather stations (AWS) and five radars by the Korea Meteorological Administration (KMA) has not been sufficient to provide reliable warnings with a few hours lead time that the public demands. The current warning system, largely based on extrapolation, thus is not satisfactory and has its limitations in principle. Prediction based on a numerical model is most promising but issues such as computer power, initial data availability on the storm scale, balancing the initial fields, etc., make it a very challenging task.
In recent years, a model system for predicting rapidly growing, storm-scale hazardous weather has been developed at the Center of Analysis and Prediction of Storm (CAPS), University of Oklahoma. It is known as the Advanced Regional Prediction System (ARPS, Xue et al. 1995). It is the first system that is developed with primary emphasis on storm-scale weather prediction, and thus possesses several unique characteristics. It is also developed to run on massively parallel processors (MPP). The ARPS has been tested during the past several spring seasons in the central United States (Droegemeier et al. 1995, 1996; Xue et al. 1996), and shown promising skills in explicit storm prediction.
2. TAKE PROJECT
A four year project toward real-time operational prediction of flash floods by utilizing the ARPS has been instituted at the Korea Meteorological Administration (KMA). This project is known as the Test of ARPS in the Korean Environment (TAKE). The eventual goal of TAKE is routine operation of the ARPS for very short range (up to 12 hours) prediction of localized severe weather by the year 2000. To achieve this goal, we define the following objectives for the project:
1) Gain experience and understand technical issues related to storm-scale prediction.
2) Test the feasibility of the ARPS in predicting storm-scale events in the Korean environment;
3) Find and understand problems and limitations so as to improve the model, input data, and data analysis/assimilation system.
4) Recommend improvements to the observation system for the purpose of storm-scale prediction;
The year by year implementation plan is as follows:
TAKE-97: Test operation as feasibility study in summer, understand limitations and problems;
TAKE-98: Test operation with improved model capability and input data, apply observations from Doppler radars, understand limitations and problems;
TAKE-99: Test operation with improved model and basic inputs, establish computing environment for storm scale prediction;
TAKE-00: Start operation.
Following this plan, an initial operational test was carried out during the Changma season, from 1-20 July, 1997, and again during the period 2-10 August. A newly installed Cray T3E at the Korean System Engineering Research Institute (SERI) was used as the forecasting platform, while data analyses and post-processing were performed on their Cray C90. The T3E has 128 PE's and each PE has a peak performance of 900 Mflops. Operation of the 400 AWS's and 5 Doppler radars by KMA and the availability of the T3E provided a unique condition for such an experiment.
Working together with KMA scientists, CAPS scientists installed and tested the ARPS system on the SERI computer systems in the last two weeks of June, 1997. The tasks include: preparing geophysical data, including terrain, surface characteristics, and SST for the area, converting raw observations, in particular those from the 400 AWS stations, to a format required by the ARPS Data Analysis System (ADAS, Brewster 1996), linking ADAS and ARPS with the KMA operational 40 km regional prediction model RDAPS (Regional Data Assimilation and Prediction System, KMA 1996), adapting existing shell scripts to the local environment in order to automate the entire procedure of analysis, forecast and post-processing, and, finally, verifying the T3E forecast against that of the C90, and performing preliminary evaluation of a test forecast.
3. MODEL CONFIGURATION
Figure 1 shows the ARPS prediction domain as well as the ADAS analysis domain. The ARPS prediction grid size was 115x139x37 at 10-km horizontal resolution, and covered a physical domain of 1120x1360 km2. In an effort to include the influence of additional upper-soundings surrounding the prediction domain, the ADAS analysis was performed on the large grid shown in Fig. 1. This grid covers an area of 1640x188 km2. The terrain field and the surface characteristic fields were derived from 5-minute resolution data sets. In the ARPS domain, the highest mountains are located in North Korea and the highest grid value is about 2 km (Fig. 1).
Figure 1. The ARPS forecast domain (inner box) and the ADAS analysis domain (outer box) of TAKE-97. Terrain height is shown as shaded contours.
The ARPS predictions included essentially all physics available in the model. They include a two-layer soil model, surface-layer and PBL parameterizations, 1.5-order TKE-based subgrid-scale turbulence, a three-category ice microphysics and a comprehensive radiation package. No cumulus parameterization was used, however.
The ARPS was operated twice a day for 00 and 12 UTC initial states and was integrated for 12 hours for each run. The ADAS used the 12 hour forecast of RDAPS as the first guess field. The boundary conditions were obtained from the available RDAPS forecast output, at three hour intervals.
The 00 and 12 UTC surface and upper-air observations were the major sources of observation. There are about 20 upper-air stations and 100 regular synoptic surface stations in the prediction domain. In addition, data from 400 automated surface weather stations (AWS) operated by KMA were used in the initial analysis. This effort represents the first attempt to use the KMA AWS data in an objective analysis system for NWP purposes. Some buoy and ship reports were also used.
4. COMPUTATIONAL LOGISTICS
As mentioned earlier, the forecasting was performed on the SERI Cray T3E. The 12 hour prediction takes about 2 hours wallclock time using 64 processors. The data pre- and post-analyses were performed on a Cray C90. The entire procedure was automated using shell scripts.
5. PRODUCT GENERATION
Model output was produced every hour. Graphics were generated for a large number of fields, including MSLP, standard pressure level heights, temperature and winds, surface pressure, temperature and humidity, composite radar reflectivity, hourly precipitation. Time series of meteorological variables at station locations were also generated.
6. FORECAST VERIFICATION
During the test period there were six episodes of heavy rain. Fig. 2 shows an example of amounts during a 12 h period. It is apparent that the model shows a moisture spin-up problem in the early hours. However, the developing sequence of heavy convective rain shows agreement with observations. It is also found that predicted rainfall amounts seem to be overestimated in later times.
Figure 2. Comparison of 3-hour rainfall between model forecast (upper panel) and observation (lower panel) 00UTC July 1, 1997.
Table 1 summarizes the performance of the rainfall predictions by evaluating threat scores for each 3-hour period when rains were observed. The threat scores were calculated using a radius of 30-km from the observing site. The threat scores improve gradually as the model progresses in time, indicating the moisture spin-up problem for early times. For the heavy rain case, over 10 mm/3-h, it tends to remain around 0.25.
Despite poor initial conditions with respect to storm-scale predictions, the ARPS shows potential for predicting small scale, rapidly developing weather in general. It is suggested that utilization of available sources of information, such as Doppler radars, satellite and the use of asynoptic data from aircraft will enhance the performance of ARPS in the future. KMA and CAPS will continue the testing with better initial state, improved model and improved computing facilities until the year 2000.
Table 1. Threat Scores for ARPS 3-h Rainfall Prediction During TAKE-97
0-3h 3-6 h 6-9h 9-12h
0.25 0.31 0.39 0.46
0.07 0.26 0.27 0.25
7. FUTURE PLANS
As was outlined in Sec. 2, we will continue testing the ARPS during the monsoon seasons of earlier summer for the next three years. It is our hope that the model system be eventually used in operational prediction of explicit precipitation and convective events over Korea. At this time, we are performing experiments that use the ARPS 12- hour forecast as the initial guess. This leads to a continuous forecast cycle. For the 1998 experiment, we will include, among others, data from five Doppler radars operated by KMA.
This work was supported by the Korea Meteorological Administration, and by the Center for Analysis and Prediction of Storms under NSF grant ATM91-20009. This project would not have been possible with the Cray T3E time generously provided by the System Engineering Research Institute, Korea.
Brewster, K., 1996: Application of a Bratseth analysis scheme including Doppler radar data. Preprints, 15th Conference on Weather Analysis and Forecasting. Amer. Meteor. Soc., Norfolk, VA, 92-95.
Droegemeier, K. K., et al., 1996: Realtime numerical prediction of storm-scale weather during VORTEX '95. Part I: Goals and methodology. Preprints, 18th Conf. on Severe Local Storms. Amer. Meteor. Soc., San Francisco, CA, 6-10.
Droegemeier, K. K., et al., 1996: The 1996 CAPS spring operational forecasting period: Realtime storm-scale NWP, Part I: Goals and methodology. Preprint, 11th Conf. on Num. Wea. Pred. Amer. Meteor. Soc., Norfold, VA, 294-296.
Xue, M., K. et al., 1996: The 1996 CAPS spring operational forecasting period: Realtime storm-scale NWP, Part II: Operational summary and examples. Preprint, 11th Conf. on Num. Wea. Pred. Amer. Meteor. Soc., Norfolk, VA, 297-300.
Xue, M., K. K. Droegemeier, V. Wong, A. Shapiro, and K. Brewster, 1995: ARPS Version 4.0 User's Guide. Center for Analysis and Prediction of Storms, 380pp [Available from CAPS].