Evaluation of Model and Radar Precipitation

1
Evaluation of Model and Radar Precipitation

• Charles A. Lin
• McGill University

2
Team members

• S. Vasic, I. Zawadzki, O. Bousquet, D. Chaumont,
  L. Wen and A. Vincent

3
Objective

• To perform verification of model QPF using
  radar-retrieved precipitation with conventional
  statistical and scale decomposition analysis

4
Analysis domains

• Montreal area (McGill radar)
• Continental US (US radar network)
• Saguenay, Quebec region (Saguenay flood)
• Edmonton area (Carvel radar, Mackenzie GEWEX
  Study)

5
Models

• MC2 (Mesoscale Compressible Communty) model
• GEM (Global Environmental Multiscale) model
• GEM/HIMAP model
• NCEP Eta model

6
Comparison of precipitation from MC2 and McGill
radar

• MC2 6 km resolution 2 storm cases

7
Mean precipitation rate (mm/h), Oct. 14, 1995
MC2 (6-km)
200 km
McGill RADAR
Yu et al., 1998
8
Mean precipitation rate (mm/h), November 8, 1996
MC2 (6-km)
200 km
McGill RADAR
Yu et al., 1998
9
Analysis domain central and eastern continental
US (2160 x 2160 km)
10
Model precipitation

• Canadian GEM model
• NCEP Eta model
• Resolution ?x 24 km, 3-hour accumulated
  precipitation
• May 24-30, 2001

11
Radar precipitation

• US NexRad radar network
• Obtained from UCAR
• Rain-gauge corrected data

12
Diagnostic methods

• Conventional time series, POD,
• Scale decomposition
• Fourier analysis
• wavelet analysis

13
Conventional methods
14
3-hour accumulated precipitation (0600 UTC, May
25, 2001)
US Radar composite
GEM model
Eta model
2160 km
15
Time series of precipitation Maxima and
area-averaged
16
Scatter diagram area-averaged precipitation
17
Area fraction Pr gt (1, 5, 10) mm/3 hr / Pr gt
0.1 mm/3 hr
GEM and Eta models
Radar
18
Area fraction time-averaged statistics
19
Scale decomposition analysis
20
Power spectra Haar wavelet
?
?
wavenumber k 2 wavelength ? 1500 km
k 10 ? 310 km
21
Power law for precipitation

• E(k) k -ß (Harris et al., 1996)
• ß characterize precipitation fields
• Averaged radar spectra
• ßradar 2.2 for k 20 - 50
• (wavelength 154 - 62 km)
• Averaged GEM/Eta model spectra
• ßmodel 3.0

22
Model/radar coherence spectra
GEM/radar comparison
Eta/radar comparison
23
Scatter diagram power at different wavenumbers
(model/radar comparison)
Power falloff at k 32, about 4 ?x
Power falloff at k 24, about 5 ?x
24
Model/model comparison
25
Summary

• Conventional and scale decomposition analysis of
  model and radar precipitation
• Enhanced falloff of power of model precipitation
  at high wavenumbers compared to radar (ßradar
  2.2 lt ßmodel 3.0)
• Rapid drop in model/radar coherence as a function
  of wavenumber
• Significant loss in power of models at scales
  4-5 ?x

26
Saguenay flood study

• Coupled atmospheric/hydrologic model

27
Topography of Saguenay region
Hydrological basin
Precipitation
Flooding
Lin et al. (2002)
28
Two key components in coupled atmospheric/hydrolog
ic model
Precipitation Runoff - generation - routing
Result hydrograph
29
Application to Saguenay flood

• July 19-21, 1996 48-hour simulation
• Ha! Ha! River basin
• 1-km GIS database for soil, vegetation,
  topography
• Coupled model MC2/CLASS/GUH
• Router
• Hydrological module
• Atmospheric model

30
MC2/CLASS Simulation Domain
m
5-km domain
Montréal
800 x 800 km2 sub-domain
Lac Saint-Jean in Saguenay Region
10-km domain
31
Scatter plot 48-hour accumulated precipitation
(mm)
MC2 at 5, 10, 35 km resolution
Operational RFE model at 35 km resolution
32
10km
10km
Ha! Ha! River basin covered by 15 model grid
points (10 km 10 km)
33
Southern Ha!Ha! River basin covered by 6 model
grid points
Nearest rain gauge 7043713
?
34
Precipitation over Ha! Ha! River basin
Rain gauge 7043713
6 model grid points
35
Comparison of two reconstructed hydrographs at
outlet of Ha!Ha! Lake (July 19 to 21, 1996)
Reconstructed hydrograph from Lapointe et al
(1998)
36
Simulated runoff

• Total runoff
• Lin et al. (2002) 3.1107 m3
• Lapointe et al. (1998) 3.2107 m3
• Time of peak precipitation is about the same
• Significant difference in time of peak flow

37
Summary

• Proof-of-concept study coupled
• meteorological/hydrological models for
• flood forecasting
• Encouraging results from Saguenay study

38
Mackenzie GEWEX Study (MAGS)Comparison of
precipitation from GEM/HIMAP and Carvel radar
(near Edmonton)
39
(No Transcript)
40
Shift model 6-hour accumulated precipitation
patterns to maximize correlation with radar
41
Carvel radar
GEM/HIMAP model
Shifted GEM model
42
Contingency table (with threshold for rain)
Skill scores based on above measures POD
(probability of detection), FAR (false alarm
rate), CSI (critical success index), ETS
(equitable threat score)
43
Skill scores

• POD, FAR, ETS
• for
• HIMAP and HIMAP shifted

44
POD
FAR
ETS
45
Scale decomposition Power spectra
46
Summary

• GEM/HIMAP has some skill for 6-hour accumulated
  precipitation, when precipitation is strong
• Less skill for 3-hour accumulations

47
Combine nowcast and NWP model for precipitation
forecast?

• Nowcast (Lagrangian persistence) Good initial
  conditions, robust for short forecast lead times
• NWP models Resolves large scales

48
Theoretical limit of predictability
NWP model
Nowcast
Golding (1998) Austin et al. (1987)
49
Revisit earlier case
Include nowcast from Germann and Zawadzki (2002)
50
Mean skill scores for 9-hour forecast

• POD, FAR, CSI (categorical measures)
• CMAE (conditional mean absolute error)
• for GEM forecast, and unfiltered (NOWC) and
  filtered (NOFF) radar nowcasts

NOFF (near-optimal forecast filter) Turner et
al. (2003)
51
GEM forecast
Cross over point in time
Unfiltered radar nowcast
Filtered radar nowcast
Lin et al. (2003)
52
Summary

• Compared skill of radar nowcast and GEM model
  forecast
• Skill of nowcast decreases to that of model after
  about 6 hours of forecast lead time
• Model skill remains approximately constant

53
Ongoing work

• More cases in model/radar wavelet scale
  decomposition analysis, especially in Canada
• (with I. Zawadzki, McGill University)
• Blend statistically radar nowcast and NWP model
  forecast to yield optimum forecast
• (with I. Zawadzki)
• Implementation and verification of 2.5 km GEM/LAM
  regional model over Quebec(with L. Lefaivre, MSC)

54
Thank you!