1. On the Value of Radar-Derived Rainfall Assimilation on High-Resolution QPF
Daniel Leuenberger1, Christian Keil2 and George Craig2
1MeteoSwiss, Zurich, Switzerland
2DLR, Oberpfaffenhofen, Germany
COSMO GM 2008, Cracow
In my presentation I will demonstrate that radar-derived rainfall can help to improve short-range QPF out to 12h and I will present a potentially interesting new source of humidity information for NWP.

2. Introduction Convective-scale assimilation of radar rainfall data
Latent Heat Nudging (LHN)
Results of a 7 month test suite at MeteoSwiss
What determines the impact of LHN on QPF?

3. MeteoSwiss Model Setup
ECMWF IFS
COSMO-7
6.6km, 60 levels
Param. deep convection
Assimilation of conv. obs.
COSMO-2
2.2km, 60 levels
Explicit deep convection
Assimilation of conv. obs. and radar rainfall
Radar
~600 km

4. Setup of Experiments 2.2km assimilation cycle with/without LHN
Forecasts out to +12h, initialized at 00 and 12 UTC
11. June 2007 – 15. January 2008 (346 forecasts)

5. Examples of Improvement
LHN
NOLHN
Radar
0-6h Precipitation forecast ( ) Verifying Radar
6-12h Precipitation Forecast ( ) Verifying Radar
LHN
NOLHN
Radar

6. Verification against Radar
346 Forecasts, 11. June January 2008, hourly sums

7. Verification against Radar (Summer)
9 Forecasts, 11. June July 2007, hourly sums

8. Verification of other Variables
RMS of 74 12UTC Forecasts (Reference: ca. 60 Swiss Sfc. Stations)
Surface Pressure
10m Wind speed
deg 64 62 60 66 58
Wind direction
Time UTC
335
2.25
330
2.20
m/s
2.15 Pa
325
320
2.10
NOLHN
LHN
315
2.05
Time UTC
Time UTC

9. Verification of other Variables
RMS of 74 12UTC Forecasts (Reference: ca. 60 Swiss Sfc. Stations)
Cloud cover
2m Dewp. Temperature
2m Temperature 34
3.0
2.6
NOLHN
LHN 32
2.8
2.4 % 30 K
2.6 K
2.2 28
2.4
2.0 26
2.2
1.8
Time UTC
Time UTC
Time UTC

10. What determines the impact of LHN?
Use high-resolution NWP ensemble (2.8km mesh size)
Driven by regional COSMO-LEPS ensemble
10 members with LHN, 10 members without
Different mesoscale environment in each member
3 differently forced convection cases
forced frontal
non-forced frontal
airmass
31.July 2006
28. June 2006
12. July 2006

11. Example: Airmass convection
Timelines of observed and simulated area-averaged surface rainfall
NOLHN
0.8
0.6
0.4
1.0
0.2
0.0 06 09 12 15 18 21 00
Radar
NWP
Ensemble
Time UTC
LHN mm
Assimilation
Forecast 06 09 12 15 18 21 00
Time UTC

12. Definition of Time Scales
LHN impact factor
FLHN
time
0.5 1
tLHN
LHN time scale tLHN
Convective time scale
Done et al. (QJ 2006)

13. Stratification of Simulations
Results suggest 2 different regimes:
equilibrium situation:
short tc
precipitation only redistributed
short-lived impact of LHN
forced frontal, non-forced frontal airmass
100
tLHN [h] 10
non-equilibrium situation:
long tc
LHN triggers convection
long lasting impact of LHN 1
0.1 1 10
100
tc [h]

14. Findings LHN improves high-resolution NWP forecasts
QPF improvement in the first 3-12h (dependent on score and rainfall intensity)
Other variables slightly improved, particularly in summer
More realistic rainfall input for soil moisture
Impact on QPF dependent on
Precipitation forcing (equilibrium vs. non-equilibrium)
Life time of precipitation system
Mesoscale environment of convection (e.g. stability)
Extent of NWP model domain and radar data coverage

15. What Next at MeteoSwiss?
Introduce quality control
Based on statistics of radar data, can handle e.g.
visibility
non-rain echoes
foreign radar data (unknown specifications)
=> quality weight for each pixel
Add foreign radar data
France, Germany, Italy, Austria

16. Thank you for your attention