1. Precipitation Classification and Analysis from AMSU
Ralf Bennartz
University of Kansas
Lawrence, KS
Anke Thoss, Adam Dybbroe, Daniel B. Michelson
Swedish Meteorological and Hydrological Institute
Norrköping, Sweden
Work performed within EUMETSAT Nowcasting SAF

2. Outline Introduction Method overview and statistical significance
AMSU
AVHRR
combining AMSU and AVHRR
Case Studies
Outlook

3. Demands on precipitation estimates
Climate research
High absolute accuracy needed to detect climate signals
Global observations
Temporal and spatial averages
Long term observation
Monitoring of extreme events
Nowcasting/short range forecasting
Timely product generation
Best possible spatial resolution
Absolute accuracy not of major importance (three to four intensity classes are sufficient)
1.6% increase in global precipitation in the last 140 years
from rain gages that cover about 30% of land surfaces globally
Obvious need for remote sensing for both applications

4. The data set
Eight month of NOAA-15 AMSU-A/B and AVHRR (April 99-November 99)
Eight month of NOAA-16 AMSU-A/B and AVHRR (February 01- September 01)
Co-located BALTEX-radar Data Centre radar data for the entire Baltic region, 25 radars, gauge adjusted

5. AMSU-A/B AVHRR conically scanning microwave radiometer
spectral range GHz, channels used: 23 GHz, 89GHz, 150GHz
3.3 degree resolution AMSU-A
1.1 degree resolution AMSU-B
AVHRR
channels used: 0.6 m, 3.7 m, 11 m and 12 m
1km resolution at nadir ( degree )

6. Observation geometry 273 K isothermal typically at 2-3 km
Altitude of radar beam (elevation 0.5°):
@100km distance: 2.2 km
@200km distance: 5.2 km
273 K isothermal typically at 2-3 km

7. Four classes of precipitation intensity from
co-located radar data
Rain rate
Class 1: Precipitation-free mm/h
Class 2: Risk for precipitation mm/h
Class 3: Light/moderate precipitation mm/h
Class 4: Intensive precipitation mm/h

8. Passive microwave precipitation signal
Most directly linked to surface precipitation
Over cold (water) surfaces only
Works over both land and water surfaces
More indirect

9. The scattering index
Predict brightness temperature T* in absence of scattering from low frequencies (functional relation is found by inverse radiative transfer modelling or global brightness temperature statistics)
Take observed high frequency brightness temperature and subtract T*
Has been found to be a linear measure for precipitation intensity

10. AMSU-A water or coast, AMSU-B land:
SI= T89-T150-CORR
AMSU-A land (and AMSU-B land):
SI= T23-T150-CORR
AMSU-B water:
correction factor CORR corrects for scan position effects and statistical
offset for non scattering situations (for SI water adjusted dynamically).

11. Probability of different classes over sea

12. AVHRR algorithm: calculate a Precipitation index PI for day or night
11µm Tb most important
day: additional information from R0.6 µm/R3.7 µm:
night: additional information from 11 µm-12 µm:
PIday=a*Tsurf-b*T11+c*ln(R0.6/R3.8)
PInight=a*Tsurf-b*T11+c*(T11-T12)

13. + high spatial resolution + convective cells, even small ones,
AVHRR
+ high spatial resolution
+ convective cells, even small ones,
can be well identified
- no strong coupling between
spectral signature and rain
- area of potential rain overestimated
generally low likelihood
- intensity and likelihood not really
decoupled
AMSU
- low spatial resolution
- small convective cells
sometimes missed
+ stronger coupling between
rain and scattering signature
+ rain areas better delineated
+ more independent intensity and
likelihood information
- sometimes spurious light rain
- not applicable over snow and ice

14. Combining AVHRR and AMSU
AVHRR mainly used for QC of AMSU:
run cloud type analysis
for AVHRR pixels containing a potentially raining
cloud type compute precipitation likelihood
if total precipitation likelihood from AVHRR > 5%*
take replace precipitation estimate with AMSU estimate
(if available)
over snow and sea ice use AVHRR only
*thresholidng with a 5% likelihood from AVHRR has the
effect that about 5% of the rain according to (imperfect)
radar estimates are missed.

15. NOAA15 overpass 13 September 2000, 06:43 UTC
RGB AVHRR ch3,4,5
PC product RGB: red: very light
green:light/moderate
blue:intense
Radar composite

16. NOAA15 overpass 13 September 2000, 05:48 UTC
RGB AVHRR ch3,4,5
PC product RGB: red: very light
green:light/moderate
blue:intense
Radar composite

17. NOAA15 overpass 13 September 2000, 06:43 UTC
RGB AVHRR ch3,4,5
PC product RGB: red: very light
green:light/moderate
blue:intense
Radar composite
different projection!

18. NOAA15 overpass 13 September 2000, 05:48 UTC
RGB AVHRR ch3,4,5
PC product RGB: red: very light
green:light/moderate
blue:intense
Radar composite
different projection!

19. Conclusions
Empirical approach to detect precipitation and to classify intensity for use in nowcasting
User is provided with the likelihood of four different classes of precipitation intensity
150 GHz gives more detail especially over land surfaces
AVHRR good for QC of AMSU data

20. Outlook Extend to different climates (work in progress for Spain)
Better incorporation of AMSU and AVHRR
Extend method to combination of MODIS and AMSU/HSB on AQUA