中央氣象局數值模式產品介紹及 跨域應用經驗分享
洪景山
交通部中央氣象局
1897 1937 1996
2
「數值天氣預報」是天氣預報的基礎 過去是，現在是，未來也將是
什麼是「數值天氣預報」
Atmospheric motion, looks like the flow of the fluid
Weather forecast: To predict the motion of the flow
Shallow water equation
Hydraulic experiment Atmospheric observation
Numerical solution
溫度
V Velocity L L L
H H H
H H
H 氣壓
Equations to govern the atmospheric motion
Project to the earth coordinate
• Discretize the
governing equation to
the grid system
Numerical weather prediction (NWP)
The numerical solution of the fluid dynamic equations
Is it a real atmosphere ？
NWP need the knowledge among math, physics, computer science, statistics…
Sometimes miracle happened
Model forecast Radar observation
Almost reproduce the “Typhoon” in the MODEL
1326 mm 1362 mm
3-day accumulated rainfall（多采公司繪圖）
Typhoon Khanun related distant rainfall
However
Does the model performs
so Perfect all the time?
Model OBS
Wind speed prediction error
11 LST 14 LST 17 LST 20 LST 23 LST 02 LST 05 LST 08 LST
3 6 9 12 15 18 21 24
27 30 33 36 39 42 45 48
15-km resolution
5-km resolution
3 km
2 km
5 km
1 km
20161009 case
Observation
• The more details in the model, the better results.
• It means we need much more computer power
(A) 10 km resolution , (B) 5 km resolution , (C) 1.25 km resolution.
Resolution 1 2 8
Computing cost 1 8 512
• The uncertainty may be the NATURE of fluid system.
• The model process from, e.g. the mathematics, physical process, numeric, and initial condition could introduce the un-controllable uncertainties.
What can we do?
A single (deterministic) model behaves like a sniper rifle
Toward to the ensemble forecast
（系集預報）
How about a shotgun?
01 02 03 04 05
06 07 08 09 10
11 12 13 14 15
16 17 18 19 20
The most common statistics…
It is equivalent to a shotgun
01 02 03 04 05
06 07 08 09 10
11 12 13 14 15
16 17 18 19 20
However, this is not the rues
Terrain locking effect
What can we do to extract the useful model
TY QPF information from ensemble forecast?
Take the ensemble mean? Weighted over the members?
“Mean” is not always a good idea
31
Application of the probability products
32
因應對策：
• 引進大數據分析技術，提高負載/再生能 源預測準確度
•日前預測分析技術：
未來七天 (For Day-Ahead SCUC、SCED、
Ancillary Service)
•即時預測分析技術：
未 來 3~4 小 時 (For Real-Time SCED 、 Ancillary Service)
每5或15分鐘(For Real-Time Regulation &
Spinning Reserve)
台電調度處吳副處長
33
• ~25 km horizontal resolution
• Updated 4 times per day
• 6-Hourly output, extended to 14 days
34
00Z 06Z 12Z 18Z 00Z 06Z 12Z 18Z
6-hr output frequency to 14 days
15/3-km
• 15/3 km horizontal resolution
• Updated 4 times per day
• Hourly output, extended to 84 hour
Deterministic model system
36
00Z 06Z 12Z 18Z 00Z 06Z 12Z 18Z
hourly output frequency to 84-hr
15/3-km
Ensemble model system
• 15/3 km horizontal resolution
• Updated 4 times per day, 20 members per run
• Hourly output, extended to 84 hour
38
00Z 06Z 12Z 18Z 00Z 06Z 12Z 18Z
20 members 20 members 20 members 20 members
hourly output frequency to 84-hr
Convective scale nowcasting
• 10-km/2-km Domain
• Convective scale DA, assimilate the Taiwan local observations, especially for the radar oservation
• hourly updated (24 times per day)
• Hourly output, extended to 12-hr forecast
Realtime, hourly updated system extended to 24-hr forecast
Radar and all available observations
Applications
CWB NWP model output provides services in all aspects
Hydrology, Flooding, and mudslide Air quality prediction
Aviation services
Ocean current and wave
Search and Rescue Planning
Disaster prevention and decision making
…
41
N01
JAMMY
SEPIA
COPRA
ROBIN Model predict reflectivity
Air traffic control
43
待命區 即時回
波警示 17060404 17060405 17060406 17060407 17060408
N01 ● ● ● ●
JAMMY ● ● ● ● ●
SEPIA ● ● ● ●
COPRA ● ● ● ● ●
ROBIN ● ● ● ● ●
流域管制警示 (含過去三個預報報資訊)
●：最新預報達警示標準
●：前1報預報達警示標準
●：前第2報預報達警示標準
從過去的觀測 到未來的預報 降水
雲頂
空氣品質預報應用-PM 2.5
產品範例：多采公司提供
Applications
CWB NWP model output provides services in all aspects
Hydrology, Flooding, and mudslide Air quality prediction
Aviation services
Ocean current and wave
Search and Rescue Planning
Disaster prevention and decision making
…
RENEWABLE ENERGY --- Let’s work together
46
展 望
綠電的成功是營運，營運的效能是建立在預測的基礎 綠電的預測必須要有專責單位，以提供電力調度參考 之用
須有上位的政策支援
以防災/空污預報可以為例對比
對預測的誤差要有詮釋和論述的能力
量化預測的誤差（包括發電量與氣象預測）
分析誤差的來源，擬定改善的策略 預測多準才是準？
預測不確定性的存在是必然，需配合預測準確度，擬定因應策 略與風險管理作為
47
