Nowcasting Conveciton

Download Report

Transcript Nowcasting Conveciton

Nowcasting Convection
Fusing 0-6 hour observation- and
model-based probability forecasts
Collaborators: Cindy Mueller, Steve Weygandt, Jim Wilson, David Ahijevych,
Dan Megenhardt
WWRP Symposium on Nowcasting and Very Short Range Forecasting
Toulouse, France, 7 September 2005
Gap in Forecast Skill
1.0
Accuracy of Rainfall Nowcasts
>1 mm/h
CSI
.8
.6
GRID MESH 20 km
Jun-Oct 2002
.4
.2
NWP
0
Courtesy of
Shingo Yamada
JMA
1
2
3
4
5
6
Forecast Length, hours
Others have also quantified this in various ways (e.g., Golding
2000, and many others at this conference.
Outline
I) Probabilistic Forecasting
-> National Convective Weather Forecast
(NCWF) – obs-based See Mueller et al. poster 5.21!
-> RUC Convective Probability Forecast
(RCPF) – model-based just saw Weygandt talk!
II) Methodology
III) Results/Verification
IV) Future Work
Operational NCWF
– 0-2 hr probability forecasts
– Includes extrapolation, growth and dissipation
– Available on Experimental ADDS (http://weather.aero/convection)
Probabilistic Forecasts Systems
National Convective Weather Forecast
RUC Convective Prob. Forecast
1
2h
k
. P
P
Probabilities based on:
– Spatial coverage of MergedGrowth (MG)
– Elliptical filter with time-dependent size
(1 hr,
2 hr, 3-6 hr :
60 km, 120 km, 180 km)
– MG(VIL,ltng) thresholded for convection
Probabilities based on:
– Spatial coverage of convective precip
predicted by the RUC-20 model
– Square filter of 180 km
– Precipitation rate threshold for
convection (1-2 mm/hr)
Summary of Strengths and
Weaknesses of NCWF and RCPF
• Area-coverage
– RUC : overestimates coverage and likelihoods
too high
• Initiation
– RUC : good (large-scale instability and frontal)
– NCWF : not handled
• Motion
– RUC : improves with lead time
– NCWF : degrades with lead time
• Dissipation
– NCWF and RUC: similar skill
Methodology
Schematic of Methodology
RCPF
Calibration
NCWF
Climatological
Dissipation
Summing
Interpolation to
4km Grid
WSR-88D
Climatology
Coverage Maps
WSR-88D
Merged
Probabilistic
Product
Validation
Methodology
Calibration of RUC Probabilities using June 2005 Validation Data
Obs Coverage = 5%
Obs Coverage = 10%
RUC Prob Levels, p
50%
60%
30%
25%
Obs Coverage = 20%
40%
75%
40%
75%
50%
2, 4, 6 hr fcsts for each p
OBS COV
0%
5%
10%
20%
NCWF
0%
5%
10%
20%
RUC .
0-35%
35%
55%
70%
60%
50%
Methodology
Original 6hr RUC Prob Fcst
Fcst time: 2000 UTC
Valid: 0200 UTC
After Scaling
Fcst time: 2000 UTC
•Remove excessive coverage values
•Shrink Area Coverage without decreasing POD
Valid: 0200 UTC
WSI - Validation
WSI @ Forecast Time
Methodology
Trending dissipation using Climo
“…a combination of rainfall
statistics containing propagation
information with NWP
predictions may offer significant
improvement in warm rain
prediction.” – Davis et al. (2003)
June 2005 Area Coverage Diurnal Composite
t0
t1
Area Coverage > 40 dBZ
F(t1)/F(t0) in 6 hrs
Use diurnal climo of fractional
change in WSR-88D freq of
convection to incorporate
dissipation.
WSR-88 D climo from 6 warm
seasons frequency of echo > 40
dBZ (Knievel et al. 2004)
Freq40+dBZ(06 UTC) / F40+dBZ(00 UTC)
Methodology
Masking of RUC Probabilities using Climo
Nationally: Convective area
shrinking between 19 and 4
UTC.
Regionally:
• Diurnal Cycle in SE US
• Propagation evident across
Great Plains
*Note: Moving gray box indicates 6 hr period
over which fractional change is calculated.
Methodology
Original 6hr RUC Prob Fcst
Fcst time: 2000 UTC
Valid: 0200 UTC
After
Climo
Masking
After
Scaling
Fcst time: 2000 UTC
•Remove excessive coverage values
•Shrink Area Coverage without decreasing POD
Valid: 0200 UTC
WSI - Validation
At Forecast Time
• Apply climo trending by multiplying with RCFP
• Reduces RCFP in areas/time where convection is not climatologically preferred
(SE at night)
Case Study
1-6 hr Probability Forecasts
Extrapolation
Fcst Valid Times: 1600 – 2200 UTC by 60 min
Radar Data 1400- 1600 UTC by 30 min
RUC Convective Probabilities
Fcst Valid Times: 1600 – 2200 UTC by 120 min
WSI - Validation
WSI @ Forecast Time
1-6 hr Merged Probability Forecast
WSI: 1400-1600 UTC by 30 min; Fcst: 1600 – 2200 UTC by 60 min
WSI - Validation
WSI @ Forecast Time
Statistical Evaluation
Validation Period : 01-14 Aug 2005
NCWF
Merged
Comparing NCWF05-05,
RCPF50-05, Merged05-05
RUC
Validation Region
Thank You!
Methodology for Merging
• Calibrate RUC Probabilities to be comparable with
NCWF values
- Validation of RCFP and NCWF from June 05 data
• Mask RUC Probs. using climatological tendencies
observed with National network of WSR-88Ds
-(Carbone et al. 2002, Davis et al. 2003)
-Create climatological diurnal cycle of fractional
changes in the coverage of convection.
• Interpolate RCFP to NCWF 4 km grid
• Add NCWF and RUC Probabilities
• Apply smoothing filter
Box diagram
12 Aug 2005 Case Study
WSI National Mosaic