Taiwan CWB MOS System (Invited Presentation)

The CWB MOS system was established at 1997, and officially operated since 2000 (MOS 2000) till date. However, the MOS guidance performance is restricted by the NWP forecast potential, which is dominated by the extraordinarily geographic location and topographic distribution of Taiwan area. Geographically Taiwan is located in the flank of eastern Asia continent and bordering North-western Pacific ocean-basin; climatically it locates in the Eastern subtropical Monsoon region. Since the special seasonality for 2-month's period, CWB MOS System is designed to base on each month's MOS equations and its model training period is 2-month data in a year, which includes previous and following half-month to count for the seasonal shifting. Inside Taiwan area, around 385km from north to south and 143km for the maximum width in the west-east direction, the steep central mountain penetrates through north to south, and the highest mountain peak, Yushan, is 3,952 meters. Because of the complex topography, the river flows are rapid and turbulent during raining seasons. Serious disaster, flash flood and mud-slide etc., becomes as the inevitable result when there is a severe storm occurred or Typhoon invaded. Due to the complex characters of seasonality and topography, the CWB MOS equation is generated by each station individually, not based on climatic regional station group. Under both monitoring and observing purposes, a dense CWB observatory network is built for more than 20-year. There are around 500 observing stations and 4 meteorological radars, and also meteorological satellite operate routinely. The observatory net almost fully covers Taiwan area and provides sufficient records to develop MOS System. The Taiwan CWB MOS System provides both regional NWP model for short range and global NWP model for medium range forecast guidance. Although the regional NWP model simulation can provide fine grid information to generate MOS equation, both regional and global NWP model show that the topographic affect obviously reduces the MOS model's explanation potential, from coastal to mountain regions, seriously. For the temperature related MOS guidance, it only shows slight bias difference from coastal to mountain region. But for precipitation related guidance, it might provide the wrong location to precipitate, since the NWP convergence simulated failure. Under this circumstance, the CWB MOS wind direction guidance is not recommended to refer for forecaster, because NWP forecast wind direction in mountain area might be totally reversed with station's observation. To gain more capability of MOS model explanation, the NWP model persistence is utilized to develop the CWB MOS model by including certain numbers of neighboring leading and after projections in the target projection MOS equation. It does increase the forecast equation's explanation, but also it might take a higher risk of forecast stability for the MOS equation. The CWB MOS model's maintenance is issued by the forecast performance of the same NWP model's paralleling PP(Perfect Prog.) forecast better than the MOS model's. There are two options provided, one is to update dataset and adjust predictors' weighting, the other is to regenerate the MOS model. Before 2009, the CWB MOS System is focus on developing CWB's regional(LFS and NFS) and global(GFS) NWP models. We experienced to develope Tmax, Tmin, PoP, categorical cloud amount, wind speed, wind direction and categorical QPF station guidance. Since 2009, under the FIFOW(Fine Information of Formosa Weather) project, both station and gridded(2.5km*2.5km) MOS guidance were created and more NWP models' guidance were included. For the regional (NFS and WRF) models, the MOS operations are performed on 00Z, 06Z, 12Z and 18Z every day, and MOS products are provided by every 3-hour interval till 84-hour leading forecast projection for each initial operation. The medium range MOS guidance are based on NCEP, EC and JMA GFS models, 2 initial(00Z and 12Z) operations per day, and products are provided by every 12-hour interval till 192 leading forecast hours for each initial operation. The CWB FIFOW MOS System works as the forecast foundation to allow forecaster providing more meticulous weather forecast services of weather element's evolution from major cities to local villages within Taiwan area. The future challenges of the CWB FIFOW MOS System will focus on increasing the forecast accuracy and also providing probabilistic products. Developing ensemble MOS scheme will be the most important fundamental task. Then base on the ensemble scheme, the application of multi-model ensemble MOS will be utilized to benefit the CWB forecast services. Since recent NWP model's improvement and the needs from social or business activities, we also begin to study the potential of MOS works on the 2nd week to one month forecast development. Taiwan area experienced severe weather disaster frequently. Although it may be warming by the MOS probabilistic guidance, however it is difficult to predict its occurrence, and not to mention to provide the quantitative intensity. In according to the regional CWB-NFS MOS categorical QPF developing experiences, when the MOS model successfully decide the correct rainfall category, then employ the indicated categorical rainfall interval's historical data setup the QPF MOS equation directly, it gave higher potential to approach the observed quantitative precipitation. Since Taiwan's topographic affect cause the regional NWP model provides limited information to make a right choose for MOS QPF category. We are planning an online operating process, the first procedure is to utilize the ANALOG methods to analyze and select the similar cases from the previous reanalysis global data and historical Taiwan's observed data, which in according with the most recent local and global observed records. The purpose is to select the cases in the same cluster, as to narrow the categorical interval. The second procedure is to form the quantitative MOS equations directly by the assigned reforecast NWP data with those selected cases in the first procedure, and then generate the MOS forecast guidance. Also those selected similar initial cases may suggest the possible following weather evolution in the probabilistic sense.