Title: Ingen bildrubrik
1The Precipitating Clouds Product of the
Nowcasting SAF Anke Thoss, Ralf Bennartz, Adam
Dybbroe University of Wisconsin, USA
2- Outline
- Introduction
- Method overview
- AVHRR
- AMSU
- combining AMSU and AVHRR
- algorithm performance
- Case Studies
- Summary and outlook
3- GoalAlgorithm for Nowcasting applications
- fast
- absolute accuracy not of primary importance
- applicable over land and sea, day and night
- use satellite data directly received at weather
service NOAA /(EPS) IR-VIS-MW MSG
IR-VIS - considerable uncertainties in both VIS/IR
- as well as scattering based MW precipitation
retrieval - ? likelihood estimates in intensity classes more
appropriate
4Four classes of precipitation intensity from
co-located radar data Rain rate Class
0 Precipitation-free 0.0 - 0.1 mm/h Class 1
Very light precipitation 0.1 - 0.5 mm/h Class
2 Light/moderate precipitation 0.5 - 5.0
mm/h Class 3 Intensive precipitation 5.0 -
... mm/h
5The data set
- Eight months of NOAA-16 AMSU-A/B (Feb -Aug 2001,
867 overpaths) for AMSU algorithm development - 12 months (June 99 - May 00) for AVHRR algorithm
development. - Co-located BALTEX-radar Data Centre radar data
for the entire Baltic region, up to 30 radars,
gauge adjusted
6- AMSU-A/B
- cross track scanning microwave radiometer
- spectral range 23-190 GHz, channels used23 GHz,
89GHz, 150GHz - 3.3 degree resolution AMSU-A (23-89GHz)
- 1.1 degree resolution AMSU-B (89-190GHz)
- AVHRR
- channels used 0.6 ?m, 1.6 ?m 3.7 ?m, 11 ?m and
12 ?m - 1km resolution at nadir ( 0.054 degree )
7- AVHRR Algorithm development
- Based on Cloud type output
- Correlation of spectral features with
precipitation investigated -
- Special attention to cloud microphysics
(day/night algorithms)
- Precipitation Index PI constructed as linear
combination of spectral features
- Algorithms cloud type specific
8Which Cloud types are potentially raining? all
cloudfree types P(rain) lt 2.6
9Correlation of Spectral features with rain
Correlation with class, all potentially raining
cloudtypes T11 -0.24 Tsurf - T11
0.26 T11-T11 -0.16 R0.6 0.18 R3.7
-0.18 ln(R0.6/R3.7) 0.26 R0.6/R0.6
0.42 3.7?m day algorithm, all
0.35 1.6?m day algorithm, all 0.44 night
algorithm, all 0.30
10Precipitation Index
Example 3.7 day algorithm, all cloud
types PI350.644(Tsurf-T11)5.99(ln(R0.7/R3.7))
-3.93(T11-T12) Example 1.6 day algorithm, all
cloud types PI 65 -15abs(4.45-R0.6
/R1.6)0.495R0.6-0.915(T11-T12)
0Tsurf0T11
11Probability distribution, all raining Cloudtypes
1.6 Day algorithm
Night algorithm
3.7 Day algorithm
12NOAA-15 overpass 27 May 2000 1722 UT
AVHRR Cloud type
AVHRR-RGB CH 3,4,5
AVHRR-RGB CH 1,2,4
13NOAA-15 overpass 27 May 2000 1722 UT
- AVHRR -night
- Precipitation classification RGB
- blue intensive
- green moderate
- red light
AVHRR-RGB CH 3,4,5
BALTRAD radar composite
- AVHRR -day/night
- Precipitation classification RGB
- blue intensive
- green moderate
- red light
14Passive microwave precipitation signal
- Most directly linked to surface precipitation
- Over cold (water) surfaces only
- Works over both land and water surfaces
- More indirect
15The 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 T and subtract the observed high
frequency brightness temperature
Has been found to be a linear measure for
precipitation intensity
16For our algorithm
CORR corrects for scan position effects and
statistical offset for non scattering situations
AMSU-A water or coast, AMSU-B land SI
T89-CORR -T150 AMSU-A land (and AMSU-B
land) SI T23-CORR -T150 AMSU-B water SI
T89-CORR -T150
for SI water CORR is adjusted dynamically
17AMSU-A
Linear dependence of MW Tbs on land fraction
within FOV ? coastal estimates can be
computed as a linear combination of land and sea
estimate according to land fraction SIcoast
(1-Nland)SIsea NlandSIland
T23
Nland
? important to properly convolve a high
resolution LSM to the AMSU FOV
? important to properly convolve AMSU-B to AMSU-A
? for algorithm development convolve radar to
AMSU-B!
18150 GHz versus 89 GHz scattering index over land,
Results from NOAA15 (23GHz as low frequency
channel)
19Results hardclustering NOAA16
AMSU SEA c0 c1
c2 c3 radar c0 70 27 3
0 radar c1 15 53 31 1
radar c2 4 24 55 17 radar c3
3 6 27 64
AMSU Land c0
c1 c2 c3 radar c0 69 26
4 1 radar c1 16 49 24 11
radar c2 4 31 33 32 radar c3
10 5 14 71
20NOAA-15 overpass 27 May 2000 1722 UT
- Precipitation classification RGB
- blue intensive
- green moderate
- red lighPt
AVHRR-RGB CH 3,4,5
AMSU-RGB 89,150,1837 GHz
BALTRAD radar composite
21AVHRR AMSU
high spatial resolution - low
spatial resolution
convective cells, even small - small
convective cells ones can be well identified
sometimes missed
- no strong coupling between stronger
coupling between spectral signature and rain
rain and scattering signature
- area of potential rain rain areas
better delineated overestimated
?generally low likelihood
- intensity and likelihood not more
independent intensity really decoupled
and likelihood information
- sometimes spurious light rain
- not applicable over
snow and ice
22Combining 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 gt
5, replace precipitation estimate with AMSU
estimate (if available)
? over snow and sea ice use AVHRR only (to be
implemented )
thresholidng with a 5 likelihood from AVHRR has
the effect that about 2.5 of the rain according
to radar estimates for potentially raining clouds
are missed.
23NOAA-15 overpass 27 May 2000 1722 UT
- AMSU/AVHRR
- Precipitation classification RGB
- blue intensive
- green moderate
- red light
AVHRR-RGB CH 3,4,5
BALTRAD radar composite
- AVHRR -day/night
- Precipitation classification RGB
- blue intensive
- green moderate
- red light
24Algorithm performance
Different algorithms - different
characteristics to compare
different algorithms ? hardclustering performed
with monthly varying, algorithm dependent
thresholds. If P(rr) ? threshold, assign to
rain class with greatest likelihood,
otherwise assign to no-rain.
Thresholds selected according to average monthly
likelihood per class
Total rain thresholds usedmonth 1
2 3 4 5 6 7 8
9 10 11 12 day3.7/ night 30 30 40
40 50 50 50 40 30 30 30
30 night 30 30 30 30 30 30
30 30 30 30 30 30 AMSU 30
40 40 50 50 50 50 40 50 50
40 40
25Flexible clustering, potentially raining cloud
types one year data set
AVHRR day/night
c0 c1 c2 c3 radar c0 57
2 35 6 radar c1 36 2
51 11 radar c2 18 2 59
21 radar c3 8 0 44 48
AVHRR night
c0 c1 c2 c3 radar c0 71
1 28 0 radar c1 46 3
51 0 radar c2 28 2 69
1 radar c3 13 4 77 6
AMSU only Coastal
c0 c1 c2 c3 radar c0 70
26 4 0 radar c1 24 42
29 5 radar c2 9 26
44 21 radar c3 5 8 26 61
AMSU/AVHRR
c0 c1 c2 c3 radar c0 70
19 8 3 radar c1 46 36
13 5 radar c2 27 38 23
12 radar c3 10 24 29 37
All year (120 scenes), every 30th pixel
26Flexible clustering correctly identified class0
(no rain)
27Flexible clustering class2 (0.5-5mm/h
classified as rain)
28Flexible clustering class3 (gt5mm/h classified
as rain)
29Flexible clustering, potentially raining cloud
types April-May 2001
AVHRR day 1.6, rainthresh 20
c0 c1 c2 c3 radar c0 74
15 11 0 radar c1 45 34
21 0 radar c2 27 42
31 0 radar c3 16 43 41
0
AVHRR night, rainthresh 30
c0 c1 c2 c3 radar c0 63
1 36 0 radar c1 54 1
45 0 radar c2 39 1
60 0 radar c3 21 0 78
1
SCORES
RAIN THRESH POD FAR HK HSS night
30 0.53 0.63 0.16 0.11 day 1.6
20 0.63 0.50 0.38 0.35
AMSU/AVHRR 20 0.65 0.55 0.33
0.29 AMSU/AVHRR 50 0.52 0.52 0.29
0.27 AMSU sea 0.89 0.83 0.47
0.17 AMSU land 0.88 0.75 0.57
0.27
AMSU/AVHRR, rainthresh 20
c0 c1 c2 c3 radar c0 68
24 6 2 radar c1 43 42
11 4 radar c2 25 42 22
11 radar c3 15 32 29 24
every 10th Pixel
30Total rain likelihood 10 20 30 40 50 60-80
90-100
NOAA16, 2001-04-05, 1130UTC upper total
precipitation likelihood, leftIR,
middleVIS,rightAMSU lower left AMSU
likelihood RGB Redintensive green
light/moderate bluevery light lower right BRDC
radar composite
31Total rain likelihood 10 20 30 40 50 60-80
90-100
NOAA16, 2001-04-23, 1145UTC upper total
precipitation likelihood, leftIR,
middleVIS,rightAMSU lower left AMSU
likelihood RGB Redintensive green
light/moderate bluevery light lower right BRDC
radar composite
32Total rain likelihood 10 20 30 40 50 60-80
90-100
NOAA16, 2001-05-19, 1045UTC upper total
precipitation likelihood, leftIR,
middleVIS,rightAMSU lower left AMSU
likelihood RGB Redintensive green
light/moderate bluevery light lower right BRDC
radar composite
33NOAA16, 2001-05-21, 14000UTC
10 20 30 40 50 60-80 90-100
upper total precipitation likelihood, leftIR,
middleVIS,rightAMSU lower left AMSU
likelihood RGB Redintensive green
light/moderate bluevery light lower right INM
radar composite
34Algorithm Performance - Summary
? all algorithms miss a lot of precipitation
events in winter, but AMSU Alg. was recently
improved on this point
? in summer generally acceptable performance, but
area extend of precipitation overestimated.
? AVHRR 3.7 day algorithm can delineate moderate
to strong precipitation,but assigns too many no
rain cases high precipitation likelihood in
summer ? AVHRR 1.6 day algorithm can delineate
precipitation areas quite well, but can not
delineate intensity. Seasonal and angular
dependence needs to be investigated.
? AMSU highest potential to delineate intensity
classes. Underestimates intensity when estimates
are translated to pixel level (Scale!)
35Outlook
? Check stability of 1.6um algorithm
? Develop combined 1.6, 3.9 ?m algorithm for
MSG ? Refine coupling of VIS/IR/MW ? Calibrate
MSG estimates with MW estimates?