The Performances and Improvements of Taiwan Central Weather Bureau Limited Area Model

1. Introduction

The 2nd generation of Limited Forecast System (LFS) of Taiwan Central Weather Bureau (CWB) has been operating since 1994, which has two runs daily (0000 UTC and 1200 UTC). LFS is a nested second order finite differencing model which has coarse grid (grid points: 161x121, grid size: 60 km, 48 h forecasts), and fine grid (91x91, 20 km, 36 h forecasts), using split-explicit temporal integration scheme and Kuo (1974) cumulus scheme, Harshvardhan et al. (1987) radiation scheme, TKE E- planetary boundary layer parameterization scheme. In this article, for the limitation of the paper length, we only discuss the forecasts using LFS with the fine-grid.

2. The Performances of CWB LFS

We have studied the forecasts of LFS in detail during the Mei-Yu season (there exists a heavy rainfall season from early May to late June in Taiwan area) this year. Wherever the direction of weather system is moving (to sea or still on land), the preliminary results have shown that they have good performance for weather systems larger than one hundred kilometer when the initial conditions of LFS involve almost all the weather systems. That is, the dynamics and physics of LFS could simulate all the weather systems. But if there exist some unknown mechanisms in a weather system, whice are not caught in the initial conditions, LFS would have not enough information to such weather system eventually, especially on the scale below one hundred kilometer. In this article we use a heavy rainfall case during the Mei-Yu season to describe the performance and improvements of LFS.

Fig. 1a shows the 36 h forecast of precipitation (12 h accumulated), sea level pressure and 1000 hPa wind field of June 20, 1991 with fine-grid LFS. Fig 1b depicts the IR imagery at the same date as in Fig 1a. Comparing Fig. 1a and 1b, we note that LFS is pretty good in forecasting the detailed structure of the Mei-Yu front from Taiwan area to Japan. However, LFS does not produce precipitations from the cloud cluster from the South China Sea to Philippines due to the shortage of observation data related to convection in that region. From Fig 1a, it is also shown that LFS forecasts only a little precipitation on Taiwan island while the satellite picture shows quite intensive precipitation.

(a)

(b)

Fig. 1. (a) 36 h forecast of precipitation (12 h accumulated, shading), sea level pressure and 1000 hPa wind field of June 20, 1991 with fine-grid LFS and (b) the IR imagery at the same time.

3. Recent Improvements of CWB LFS

Recently we have done some tests and are planning to do more tests to improve the performances of LFS, which include the following items:

1. Advection schemes: We have upgraded the operational 2nd order finite difference scheme to the 4th order finite difference scheme, including both mass and momentum field. The results show that the precipitation forecasts using the 4th order scheme are more organized and produce larger precipitation area than using the 2nd order scheme. Beside we also have implemented the Smolarkiewicz positive definite moisture advection scheme (Smolarkiewicz 1983) in LFS, not only to horizontal direction but also to vertical direction. The preliminary results show that the Smolarkiewicz scheme has more accuracy and less phase error than the 2nd and 4th order schemes. And the most prominent feature of the positive definite Smolarkiewicz scheme is that it does not require artificial setting for the negative moisture and thus produces better moisture budget than other finite difference schemes.

2. Infrared satellite image data: For the spin-up time problem of LFS, we have used the IR data to enhance the moisture field of initial condition, especially in the place of deep convection activities. The results has shown that more precipitation is produced at 12 h forecast and precipitation pattern is in better agreement on the ocean surface. But without any theoritical bases, it shows pretty serious change in the moisture budget and the LFS may be wetter than before.

3. Special Sensor Microwave Image (SSM/I): We have used SSM/I data plus the reverse Kuo scheme (Krishnamurti 1988, 1991). And the assimilation process could recover the mass, momentum and moisture fields of real phenomena in the region of deep convection. For example, the convective-scale observation data are often rare for tropical cyclones and it is hard to incorporate the exact information on the structure and strength of tropical cyclones, hence it deteriorates the ability of an NWP model to forecast the accurate track of tropical cyclones. By using the SSM/I data and the reverse Kuo scheme, it is possible to reconstruct the moisture, temperature and momentum fields of tropical cyclone, which would improve the accuracy of typhoon track forecast.

4. Topography of Taiwan: We have upgraded the topography of Taiwan island in fine-grid LFS, using 1-km resolution to reconstruct the detailed structure of the Taiwan terrain. The preliminary results have shown that it results in a great improvement to precipitation forecasts on Taiwan island.

5. Quantitative Precipitation Forecasts (QPF): In the Preprints of 11th NWP Conference, Norfolk, Virginia, 1996, there are several papers which discussed QPF, especially on the performance of the NCEP ETA model. On the Taiwan island, there are more than 400 automatic raingages, which cover almost whole Taiwan area. We have set a period to collect the raingage data and the LFS forecasts to proceed to the task of QPF, such as to calculate the threat score, equitable skill score, etc., to show the ability of LFS for QPF.

4. Conclusion

CWB is planning to develop a non-hydrostatic model (NHM), which has 5 km grid-size and covers Taiwan island only. The forecasts of NHM could involve meso-scale weather systems on the scale from one hundred kilometer to about twenty-five kilometer, which include most of meso- scale phenomena. It is of benefit to weather forecasts in Taiwan. But there exist some important and difficult factors; for example, the initial conditions of NHM may not catch the detailed charateristics of initial weather system, especially over the ocean around the Taiwan where the convective-scale observation data are seldom available. Also the mountains of Taiwan, whose heights often exceeds 3 kilometers, might make the simulations more complicated.

References

Harshvardhan, R. D., D. Randall, and T. Corsetti, 1987: A fast radiation parameterization for the atmospheric circulation model. J. Geohpys. Res., 92, 1009-1016.

Krishnamurti, T. N., H. S. Bedi, W. Heckley, and K. Ingles, 1988: Reduction of spin-up time for evaporation and precipitation in a spectral model. Mon. Wea. Rev., 116, 907-920.

____, J. Xue, H. S. Bedi, K. Ingles, and D. Oosterhof, 1991: Physical initialization for numerical weather prediction over the tropics. Tellus, 43A, 53-81.

Kuo, H. L., 1974: Further studies of the parameterization of the influence of cumulus convection on large scale flow. J. Atmos. Sci., 31, 1232-1240.

Smolarkiewicz, P. K., 1983: A simple positive definite advection scheme with small implicit diffusion. Mon. Wea. Rev., 111, 479-486.