Transcript Short-Range QPF for Flash Flood Prediction and Small

Short-Range QPF for Flash Flood
Prediction and Small Basin Forecasts
Prediction Forecasts
David Kitzmiller, Yu Zhang, Wanru Wu,
Shaorong Wu, Feng Ding
Office of Hydrologic Development
NOAA National Weather Service
Silver Spring, Maryland
2 June 2010
1
In this discussion:
 Recent performance of the High Resolution
Precipitation Nowcaster (HPN) algorithm in 0-1 hour
time frame
 Detection of precipitation at 25mm h-1 thresholds
 Verification at 16 km2 grid resolution (4x4 km)
 An approach to QPF in the 0-6-hour range
 Does blending of physical and extrapolation model precipitation
forecasts improve on either one, in the 0-6-hour time frame?
 HPN was targeted for FFMP application
 0-6h QPF targeted primarily for RFC use, but there are
potential applications to Site Specific
2
HPN Extrapolation Forecasts
in the 0-1 Hour Timeframe:



Based purely on extrapolation of radar echoes
Implemented in OB9.0, following implementation of HighResolution Precipitation Estimator (HPE)
Produces forecasts of:


Rainfall rate at 15, 30, 45, and 60 minutes
1-hour rainfall total

Forecasts are computed on 4-km grid mesh, output on
1-km grid mesh
 Can incorporate gauge/radar bias information from MPE
 See WDTB flash flood modules:
http://www.wdtb.noaa.gov/buildTraining/AWIPS_OB9/index.html
3
HPN verification study:
September-October 2009

HPN was run in offline mode over the conterminous
U.S., during development of 0-6h QPF algorithm

First two hours of the extrapolation forecast are from HPN
algorithm

Input from NMQ radar-only precipitation rate algorithm
 Forecasts verified relative to subsequent NMQ radaronly precipitation estimates
 30 study hours over 15 days, 15 Sep-31 October
 Verified detection of ≥12.5mm and ≥ 25mm amounts
 Documented performance relative to persistence
forecast (initial-time rain rates)
4
Example HPN Input/Forecast/Verification
Radar Rainrate
1845 UTC
24 Sep 2009
HPN Forecast
1845-1945 UTC
NMQ Estimate
1845-1945 UTC
5
HPN verification study: Detection of 4x4km rainfall
0.8
0.7
≥12.5mm
0.6
0.5
0.4
0.3
0.2
0.1
0
POD
FAR
HPN 12.5mm
1
0.9
CSI
Persistence 12.5mm
≥25mm
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
POD
FAR
HPN 25mm
CSI
Persistence 25mm
23.3 x 106 cases included in statistics
6
HPN verification study:
Forecast vs Radar-Estimated 4x4km rainfall
4-km forecasts and verification
25
Radar QPE verification
75th pct
20
Mean
15
25th pct
10
5
0
0
5
10
15
20
25
30
35
HPN 0-1h forecast, mm
Mean obs
25th percentile obs
75th percentile obs
Poly. (75th percentile obs)
Poly. (25th percentile obs)
Poly. (Mean obs)
22,000 grid boxes with precipitation forecasted, northeastern U.S.
7
HPN Verification Study:
Summary

HPN consistently improves on persistence
forecasts in terms of POD and FAR:
 40% more detections of 12.5- and 25-mm
amounts
 20% fewer false alarms
 HPN QPF has little bias overall (0.9 to 1.1)
 For HPN QPF > 10 mm: Expected (mean)
observation is about 0.67 of the forecast amount
 For HPN QPF > 10 mm: 25th percentile
observation is about 0.80 of the forecast amount
8
0-6 Hour QPF
From Radar Extrapolation and RUC forecasts



Original requests for development from ABRFC
Designed to use a statistically-weighted combination of
QPFs from radar extrapolation and from RUC2
Extrapolation/advection model for precipitation rate
fields:




Extrapolation based on recent radar echo motion for 0-2 hours
Motion vector field is morphed toward RUC2 700-500 hPa wind
field forecast for 3-6 hours
Radar precipitation rate input from NMQ radar-only
product (see succeeding NSSL presentation)
Model Output Statistics approach used to determine
optimum blend of extrapolation and RUC QPFs
9
Radar Precipitation Rates,1715 UTC, 16 May 2009
Radar-Observed Precipitation Rates, 1715 UTC 15 May 2009
From National Mosaic and Multisensor Quantitative Precipitation Estimation
system (NMQ)
Yellow: > 10mm 6-h-1
Red: > 25mm 6-h-1
Gray: > 38 mm 6-h-1
Blue: > 75 mm 6-h-1
10
Extrapolation forecasts of rate field, 1715-2315 UTC:
11
0-6h QPF Product Characteristics

Forecast products:


Probability of 6-hour precipitation ≥ 0.25, 2.5, 12.5,
25, 50, 75 mm
Precipitation amount forecast


Gridded forecasts, 4x4 km mesh length
Issue forecasts for periods 00-06, 06-12, 12-18,
18-00 UTC (cover entire day)
 Forecasts use input from the hour preceding
start of valid period
 RUC-Satellite-Lightning equations will be
applied in radar coverage gaps
 Forecasts disseminated before start of valid
period
12
Regression Equation for 0-6-h
Precip Amount: Southeastern US
Precipitation =
0.52 + 0.31 RADAR QPF(0-3h)
+ 0.24 RUC QPF (0-3h)
+ 0.26 RUC QPF (3-6h)
+ 0.17 RADAR QPF (3-6h)
given RADAR and/or RUC QPF > 0; forecasts and
predictors in mm, spatial area 4x4 km
Prediction equation based on 40,000 cases: Apr-Sep 2009,
Southeastern United States.
Mean observed precip = 1.9 mm; R2 = 0.14
13
Regression (RUC2+Radar) Forecasts:
Correlation to 6-H Rainfall, New England
(17,300 cases Apr-Sep 2009 – 18-00 UTC)
Reduction of Variance (R2)
30
25
20
15
10
5
0
Amount
RUC+Radar
P >2.5mm
RUC only
Radar only
P > 25mm
Operational
14
0-6h QPF Findings
 Explained
variance is small at this small
spatial scale. However skill increases as
accumulating area increases.
 Products combining RUC2 and extrapolation
QPF could match or improve on skill of
current operational guidance
 Radar and numerical prediction models are
clearly complementary for QPF in 0-6-hour
range
15
Ongoing Work – 0-6h QPF
 Collection
of new forecast and verification
data on a daily basis
 Aim for 3 years’ development data
 Creation of probability and amount equations
for cool and warm season, and subregions
of the conterminous U.S.
 Create disaggregation logic to get QPFs for
1-h subintervals in 6-h period
16
Questions? Suggestions?
Thanks to collaborators in NOAA National Severe Storms
Laboratory, Institute of Atmospheric Physics/Czech
Republic Academy of Sciences
17
Supplementary Slides
18
HPN verification study:
Detection of 8x8 km rainfall
FAR for individual events, 8km
POD for individual events, 8km
1
0.9
0.8
0.9
0.6
FAR for HPN
POD for HPN
0.7
0.5
0.4
0.8
0.7
0.3
0.2
0.6
0.1
0.5
0
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.5
0.6
POD, 25mm
All 12.5mm
All 25mm
0.8
0.9
1
FAR for initial rainrate
POD for initial rainrate
POD, 12.5mm
0.7
REF
FAR, 12.5mm
FAR, 25mm
All 12.5mm
All 25mm
11,100 grid boxes with precipitation observed or forecasted
REF
19