Transcript Document

Validation and Analysis of
Precipitation Extremes in TMPA
G.J. Huffman1,2, R.F. Adler1,3, D.T. Bolvin1,2 , E.J. Nelkin1,2
1: NASA/GSFC
2: Science Systems and Applications, Inc.
3: Univ. of Maryland/ESSIC
Outline
1.
The TMPA
2.
Climate-Oriented Indices of “Extreme”
3.
40°N-S: Results and Issues
4.
Status
1. TMPA – Data Sources
1998
2000
2002
2004
2006
2008
TMI,PR
A diverse, growing set of input precip
estimates – various
- periods of record
- regions of coverage
- sensor-specific strengths and
limitations
Seek the longest, most detailed
record of “global“ precip
Combine the input estimates into a
“best” data set
SSM/I F13
SSM/I F14
SSM/I F15
SSMIS F16
SSMIS F17
AMSR-E
AMSU-B N15
AMSU-B N16
TRMM includes combinations as
standard products
- a joint mission of NASA and JAXA
- heritage in Global Precipitation
Climatology Project (GPCP)
- we know more about 1998’s
precip than we did in 1998!
AMSU-B N17
MHS N18
MHS MetOp
GPCP IR Histograms
CPC Merged IR
1. TMPA – Combinations
Both real-time and post-realtime, on a 3-hr 0.25° grid
Microwave precip:
- intercalibrate, combine
IR precip:
- calibrate with microwave
Combined microwave/IR:
- IR fills gaps in microwave
Sat-gauge (post-RT only):
- accumulate combined 3hr precip for the month
- weighted combination
with gauge analysis
- rescale 3-hr precip to sum
to the monthly sat-gauge
combination
Instantaneous
SSM/I
TRMM
AMSR
AMSU
Calibrate High-Quality
(HQ) Estimates to
“Best”
30-day HQ coefficients
Merge HQ Estimates
3-hourly merged HQ
Match IR and HQ,
generate coeffs
3hourly
IR Tb
30-day IR coefficients
Apply IR coefficients
Hourly HQ-calib IR
precip
Merge IR, merged HQ
estimates
3-hourly multi-satellite
(MS)
Monthly
gauges
Compute monthly
satellite-gauge
combination (SG)
Monthly SG
Rescale 3-hourly MS to
monthly SG
Rescaled 3-hourly MS
2. CLIMATE-ORIENTED INDICES OF “EXTREME”
CCl/CLIVAR/JCOMM Expert Team (ET) on Climate Change Detection and Indices
(ETCCDI)
- address the objective measurement and characterization of climate variability and
change
- provide international coordination and help organize collaboration on climate
change detection and indices relevant to climate change detection
- encourage the comparison of modeled data and observations
27 “core indices”
- 16 for temperature
- 11 for precipitation
- computed for (generally) multi-decade records for stations around the globe
• 1960-1990 base period
• posted at http://cccma.seos.uvic.ca/ETCCDI/data.shtml
• climatologies not posted
“Rainy Day” = 1 mm
2. CLIMATE-ORIENTED INDICES OF “EXTREME” - Selecting Comparisons
17. Rx1day,
18. Rx5day,
19. SDII,
20. R10mm,
21. R20mm,
22. Rnnmm,
23. CDD,
24. CDW,
25. R95pTOT,
26. R99pTOT,
27. PRCPTOT,
Monthly maximum 1-day precipitation
Monthly maximum consecutive 5-day precipitation
Simple precipitation intensity index
Annual count of days when PRCP  10mm
Annual count of days when PRCP  20mm
Annual count of days when PRCP  nnmm
Maximum length of dry spell
Maximum length of wet spell
Annual total PRCP when RR > 95p
Annual total PRCP when RR > 99p
Annual total precipitation
Choices made to
- represent dry and wet extremes
- be less sensitive to artifacts
• TMPA tends to under-represent light rain over land
• 99th percentile and maximum values easily contaminated
3. RESULTS – Quasi-global TMPA 1998-2007 “Climatology”
PRCP (mm/d)
R95p (mm/d)
The patterns resemble each other, but there are important differences
- flip-flop of PRCP and R95p in South America
- extra R95p maximum southwest of Mexico
3. RESULTS – Quasi-global TMPA 1998-2006 “Climatology” (cont.)
PRCP (mm/d)
CDDavg (days)
100
200
300
400
500+
The patterns are nearly inverses, but again with interesting differences
- PRCP gradient moving south in SPCZ not reflected in CDDavg
- blacked-out areas had rain events in less than half the years
3. RESULTS – Station Distribution
PRCP (mm/d)
Stations had to have all data in 1998-2003 (6 years) and be in the band 40° N-S
- coverage depends on contributing national organization
- black dot is 0.25° box with at least one station; red is 3x3 “halo”
- reasonable range of climate zones, but missing highest rain areas
3. RESULTS – PRCPTOT Comparison for 747 Stations Around Globe
- 1998-2003
- comparing single stations to
0.25° grid boxes
- note consistent overall
relationship, spread
- note wild values (mostly
high)
Why are some values so high?
- Identify all stations with
|Ts-Tg| / (Ts + Tg) > 0.4
- plot their yearly values in
red
- Most wild values are
suppressed
3. RESULTS – PRCPTOT Comparison for 747 Stations Around Globe (cont.)
Bhutan is a mix of good and
very bad difference
- The 4 stations along the
southern border have much
higher gauge (orange)
• foothills of Himalayas
• no GPCC analysis gauges
in high-rain band
• climatologies not posted
- 8 other stations are
reasonable (green)
3. RESULTS – Station Distribution, Take 2
PRCP (mm/d)
Stations have all 6 years of data in 1998-2003 and be in the band 40°N-S
- coverage depends on contributing national organization
- black dot is 0.25° box with at least one station; red is 3x3 “halo”
- reasonable range of climate zones, but missing highest rain areas
Stations with normalized 6-yr difference > 40% have a yellow “halo”
- mostly coastal, orographic, or dry
3. RESULTS – PRCPTOT Comparison for 747 Stations Around Globe (cont.)
Global PRCPTOT comparison
- tests using data from
Argentina, U.S. were good,
not like this
- gauge variables were
computed individually by
member countries
- incompatibility between my
software and some files
caused an inconsistency in
the years extracted between
gauge and satellite data
- more work underway
- meanwhile, want to reiterate
results in Argentina that are
representative of other
countries without the year
problem
3. RESULTS – PRCPTOT Comparison for 36 Stations in Argentina
- 1998-2003
- comparing single stations to
0.25° grid boxes
- note consistent overall
relationship, spread
3. RESULTS – PRCPTOT Comparison for 36 Stations in Argentina (cont.)
- interannual correlations at
individual stations generally
high
- stations with < 6 yr of data
tend to deviate from general
cloud of points
- the 3 stations with CC<0.3
have suspect time series
- remaining stations tend to
show better CC for higher
annual rainfall
3. RESULTS – R95pTOT Comparison for 36 Stations in Argentina
- 1998-2003
- comparing single stations to
grid boxes
- typically 2-4 days per year
- more spread due to less
sampling
- higher TMPA values at high
end, perhaps due to lack of
light precipitation
3. RESULTS – R95pTOT Comparison for 36 Stations in Argentina (cont.)
- lower correlations, as
expected
- stations with data problems
tend to have lower
correlations
- rest of stations still show a
trend towards higher
correlation for higher R95p
accumulations
3. RESULTS – Consistency between Indices for 36 Stations in Argentina
- scatter of R95pTOT against
PRCPTOT for stations and
TMPA
- high, similar interannual
correlations
- as before, higher R95pTOT
for TMPA
3. RESULTS – CDD Comparison for 36 Stations in Argentina
- 1998-2003
- comparing single stations to
grid boxes
- more spread due to less
sampling, isolated events
- TMPA tends to have longer
runs of dry days, likely due
to under-representing light
precipitation over land
3. RESULTS – CDD Comparison for 36 Stations in Argentina (cont.)
- lower correlations, as
expected
- stations with data problems
tend to have lower
correlations
- rest of stations show a trend
towards lower correlation for
higher CDD – fewer longer
runs can be sampled in a
year
- but, a whole population of
stations at the low end has
serious disagreement
3. RESULTS – Consistency between Indices for 36 Stations in Argentina
- scatter of CDD against
R95pTOT for stations and
TMPA
- larger spread at low rain
totals likely reflects
differences in seasonality is there a rainy season or
light precip year-round?
- as before, higher R95pTOT
for TMPA, so red points
tend to extend to right at all
values
4. CONCLUDING REMARKS
Satellite-based “high-resolution” precipitation datasets are being used to investigate
extreme events
We can use existing definitions of climate statistics to enhance communication
Diverse origins of data sets means analysis software has to be tested country-bycountry
In first two tests, TMPA’s PRCPTOT, R95pTOT, CDD compare relatively well to
gauge data in this study
- level of interannual correlation at a location depends on relative range of
interannual variation
- TMPA-based R95pTOT and CDD both tend to be high
- reflects under-representation of light precip in the TMPA over land
- Think we can use large-scale variable (PRCPTOT) as marker for extremes
quality
[email protected]
http://precip.gsfc.nasa.gov
http://trmm.gsfc.nasa.gov
3. RESULTS – PRCPTOT Comparison for 747 Stations Around Globe (cont.)
- bad results don’t match
good results in testing with
Argentina, U.S.
- gauge data computed
individually by member
countries
• plotted results have China
off by 1 year (red)
3. RESULTS – PRCPTOT Comparison for 747 Stations Around Globe (cont.)
- bad results don’t match
good results in testing with
Argentina, U.S.
- gauge data computed
individually by member
countries
• plotted results have China
off by 1 year (red)
- Temporarily dropping bad
match-ups gives a much
cleaner picture
1. TMPA - Implementation
Composites of individual
overpasses (i.e.,
“instantaneous”)
0.25°, 3-hourly
- spatial scale > typical
satellite pixel size
- resolve diurnal cycle
Intercalibrate microwave
sensors with “TRMM Best”
- TRMM Combined Inst.
best vs. atolls over ocean
in V.6
- TMI only choice in RT
- apply histogram matching
(no constraint on pattern)
Non-trivial differences due to
- time of observation
- sensor characteristics
2. RESULTS – 3-hourly Statistics
Rain rate histograms at 0.25°, 3-hr
look good compared to KWAJ radar
1999-2004
At the same time, the skill is still
modest in most cases at 0.5°,
daily against TAO/TRITON
gauges 1998-2004
- tendency for better skill at
high rates is fairly typical of
such estimates
2. RESULTS – Monthly Statistics (cont.)
Monthly comparisons for V.6 3B43 at 0.5°
- None of the gauges have wind adjustment
- W. Pac. buoys and atolls are roughly comparable (note averages); we believe
the buoys have much higher wind loss
- KWAJ radar calibrated by gauge, and % bias is comparable to atolls
- V.6 uses gauges, with wind correction, so the positive bias for MELB land is
reasonable
- RMS % difference lower for radar comparisons (which are area-average)
- RMS % differences are higher for lighter rain
Avg.
Number of
Observed
Gauge Source comparisons Precipitation
Atolls
1572
7.27
All Buoys
1021
3.87
W. Pac. Buoys 323
7.26
E. Pac. Buoys 698
2.32
KWAJ
504
5.40
MELB land
441
3.26
Bias
(3B43-obs.)
-1.21 (-16.6%)
-0.29 (-7.5%)
-0.16 (-2.2%)
-0.35 (-15.2%)
-0.99 (-18.4%)
+0.23 (+7.1%)
RMS
Difference
3.10 (42.4%)
1.95 (50.4%)
2.34 (32.2%)
1.74 (75.0%)
1.94 (36.0%)
1.17 (35.4%)