SREF PROB CONSISTENCY

1

• SREF PROB CONSISTENCY
• CHECKING TOOL
• Binbin Zhou, Jun Du, Jeff McQueen, Geoff
  DiMego
• 
  EMC/NCEP/NOAA
• Presented on
  3/8/2005 EMC Predictability Meeting

2

• Background Motivation
• - SREF Web displays images in time sequence
• - Users can only see probabilitys time
  evolution for ONE cycle
• - One cycle run has less confidence some time
• - Many users strongly suggested us to provide
  MULTIPLE cycles
• - We added a tool to check previous 4 or 5
  cycle runs, call Consistency Check

57 hr
45hr
33hr
21hr
9hr
Verified time
3

• Benefits
• - Users can focus on a fixed location to see
  Prob from various historical cycle runs
• - Improve the confidence to the PROB
  forecast since different cycles use
• different initial conditions
• - Different initial fcst time/initial
  conditions -gt more diverse
• Configuration
• Dec15 Dec16
  Dec17
  Dec18
• 21z 00 03 06 09z 12 15 18 21z 00 03 06 09z
  12 15 18 21z 00 03 06 09z 12 15
  18 21z 00 03 06 09z 12 15
• 21z Run 09 12 15 18 21 24 27 30 33
  36 39 42 45 48 51 54 57 60 63
• 09z Run 00 03 06 09 12 15 18 21
  24 27 30 33 36 39 42 45 48 51
  54 57 60 63
• 21z Run 00 03
  06 09 12 15 18 21 24 27 30 33 36
  39 42 45 48 51 54 57 60 63
• 
  09z Run 00 03 06 09 12 15 18
  21 24 27 30 33 36 39 42 45 48
  51 54
• 
  21z Run 00 03
  06 09 12 15 18 21 24 27 30
  33 36 39 42
• 
  
  09z Run 00 03 06 09 12 15
  18 21 24 27 30
• 
  
• 
  Verified-Time Dec17 00z 03z
  06z 09z 12z 15z 18z 21z 00z 03z 06z 09z 12z
  15z

4
SREF web interface for consistency checking Left
above Probability page. Right above
Consistency checking page Usage Click Click
here button on left page, the right page will
pop up automatically, then select a threshold
and verified time from right page
5

• General Behavior of probability
• - Case 1Closer to verified time, more and
  more members capture an event
• Case2 Closer to verified time, less and
  less members capture an event
• Case3 As close to verified time,
  probability is not steadily increase or decrease
• But oscillate with the time
• - In case 1, forecasters have more and more
  confidence to this forecast
• In case 2, forecasters have less
  confidence or have confidence of not occurring
• In case 3, the confidence depends on
  overall probability gain or loss

6

• Verification
• Verification with Low Level Wind Shear
  (LLWS)
• LLWS Wind vector change between surface
  and 2000feet
• Severe LLWS is defined as LLWS gt 20
  knots/2000feet
• Hazardous to airplane take off/landing
• SREF LLWS product has
• - Mean and spread
• - Prob of LLWS gt 20 knots/2000feet
• Verified Data EDAS Analysis data
• GRID-to-GRID (We are developing G2G
  verification package)
• Verification time Nov.10 Dec. 21, 2004,
  severe LLWS freq 14.7 on average
• 5 cycle forecasts are verified at a specific
  time
• 51hr
  P1
• 39hr
  P2
• 27hr
  P3

7
Left Conditional probabilities of different
cycles before severe LLWS event happenedRight
Conditional probabilities of different cycles
when severe LLWS event not happenCycle1 51 hr,
Cycle2 39hr, Cycle3 27hr, Cycle4 15hr, Cycle5
03hre.g. For severe LLWS event happen case, if
cycle1s 51hr fcst prob has 30(pink), the later
cycles fcst will increase as closer to the
event. Cycle5 (most recent cycle)s 3hr fcst prob
will reach to 50
8

• Probability gain and loss of cycles
• Def PROB Gain if dPP2 - P1 gt0
• Loss if dPP2 - P1 lt 0
• P1 is prob in previous cycle, P2 is prob in
  current cycle
• Total Gain or Loss over all previous cycles
• If Total is gain (gt0), confident, otherwise
  less confident
• Advantage rule out the forecast probability
  Bias by subtraction (is it true?)
  

9
(No Transcript)
10
Prob from different cycles forecast shown above
snow prob in DC increase steadily But NBC TV
Morning Forecast for Feb 3 (Thu) for DC saying
that Feb 1(Tue) light snow on 3th, Feb
2(Wed) light snow on 3th, Feb 3(Thu) morning
clear sky today Feb 2(Wed) both Operational ETA
and GFS forecasts no snow tomorrow Feb 3(Thu)
actual weather Snow from noon to 300 pm,
obvious accumulation on the ground,
11

• Conclusion
• Probability of multi-cycles are put together
• - Review PROB from different cycles at a
  specific area on fixed
• verified time
• - Improve the forecast confidence from
  probability consistency, which can
• be quantified with PROB gain /loss
  magnitude
• - PROB gain and loss over different cycles
  are bias free? (need to discuss)
• - Has been built into a tool and added to
  current SREF web page