Tropical Rainfall Measuring Missi

Document Sample
Tropical Rainfall Measuring Missi Powered By Docstoc
					Q2 Workshop, University Of Oklahoma, Norman, OK, June 28-30, 2005

NASA Precipitation Measuring Missions and NMQ QPE Products Eyal Amitai
George Mason University (GMU), School of Computational Sciences (SCS), Center for Earth Observing and Space Research (CEOSR) & NASA Goddard Space Flight Center, Maryland, USA
 TRMM/GPM Validation Programs (NASA)  Evaluation of GV products  Evaluation of TRMM Satellite Products  Studies for GPM GV  VOLTAIRE: Multi-national project for validation of multisensors precipitation fields and numerical modeling (EC)

 Both NASA PMM GV & NOAA NMQ projects have much in common
We will benefit by sharing our experience (e.g., generating radar rainfall products- QC, Z-R; using gauge data, integrating sensors; verification; RTO)

 Does NASA/TRMM have any needs/requirements for NMQ products?

 What NASA/TRMM products may improve NMQ products?
 NMQ is a great project and TRMM/GPM scientists will use its products for different applications:  Verification of TRMM Satellite Observations  R area coverage  Diurnal cycle  Total R  PDF (R)

NASA “Requirements” for Validation Products
 A value without an uncertainty value is not much of a value TRMM era: Product generation and product comparison

GPM era:
Error structure characterization and uncertainties determination in nearreal-time, and understanding the processes that lead to these uncertainties  „10% error @ monthly 300km x 300km‟ (GPM Project Scientist)
Radar gauge adjustment fields  Radar calibration shifts <1 dB  QC, QC and QC [exp. AQC]  Super dense gauge network for verification

 Focus on instantaneous products [exp. PDF comparisons]

‘A value without an uncertainty value is not much of a value…’
Data assimilation and many hydrologic applications require satellite observations of precipitation. However, providing values of precipitation is not sufficient unless they are accompanied by the associated uncertainty estimates.

While this principle is well known, and that the main approach of quantifying satellite precipitation uncertainties generally requires establishment of reliable uncertainty estimates for the GV products, we must remember that much research remains to be done before a map of probable error can be estimated and presented alongside the GV radar rainfall map in real time, and yet this has to be considered an important scientific goal. Therefore, the GV uncertainties might be very large and in many cases even larger than the satellite uncertainties.

If GV uncertainty values > Satellite uncertainty values do we need GV?

If GV uncertainty values > TRMM uncertainty values do we need GV? YES (in some cases)!



The overlap zone of both uncertainties might bring us closer to the truth even if the satellite algorithm-based uncertainties are smaller than those of the reference products

Distribution of Error Estimates
70 0


60 0

Count (months)

50 0

Distribution of monthly GV radar (R)-gauge (G) accumulation differences based on 6,153 values, during 12/1997-06/2004 in Central FL

40 0

• 6,153 gauge-months --> 585,760 mm • R/G=1.008; r=0.95

30 0

20 0

10 0

• NMAD=S|Ri-Gi|/S Gi= 0.17
-0 .5 -0 .4 -0 .3 -0 .2 -0 .1 0 0.1 0.2 0.3 0.4 0.5



• (R-G)/G positive skewed --> (R-G)/(R+G); Mean=0.0; Std. Dev.=0.13 (0.09) • RD=|R-G|/G; Mean RD=0.20 (0.15); Median RD=0.15 (0.12) • Max freq of RD @ RD=0

70 0


60 0

50 0

• Tails are associated with low rain accumulations: For the 1,538 gauge-months associated with the highest rain accumulations (the top 25% G), see the values in red.

40 0

30 0

20 0

10 0

 Natural variability of rainfall and gauge instrumental error combined responsible for keeping the radar-gauge NMAD above 0.15 (validation issue)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3



 Inter-annual radar calibration shifts primary factor in obtaining higher monthly radar-gauge NMAD values (Kwaj).

Independent Gauge Networks for Evaluating GV Rainfall Products
Comparison of monthly radar rainfall product with independent gauges at Melbourne, Florida Site for August 1998. The radar estimates (TSP 3A-54, version 5) are based on WPMM Ze-R relations using 21 qc-ed gauges within 15-50 km from the radar.

• The average difference between the radar estimates over the independent gauges and the gauge accumulations (MAD) is only 8% • MAD of 8% might be explained by the natural variability of rain and gauge instrumental errors • The figure includes several gauges located within the same radar pixel of 2x2 km2. The difference in gauge accumulations within each group (marked by rectangles) is in the same order as the MAD, suggesting Radar accuracy may be higher, but a denser gauge network is required for verification


15 Independent Gauges (34-41 km)



R •adar / •auge = 1.05 G MAD= • -G | / G = 8%; Corr. Coef.=0.93 |R •
i i i

0 0 50 100 150 200

Gauge [mm]

The effect of the AQC algorithm* on the radar rainfall estimates
The ratio of the monthly radar estimates derived from the post-AQC dataset to the estimates derived from the pre-AQC dataset, for each month during 1998 at Melbourne..
The number above each column represents the percentage of gauges approved for Ze-R development by the AQC algorithm. Different panels represent different range internals from the radar.
*Algorithm to filter unreliable gauge and radar data upon comparison of the G-R merged

data, developed by the TRMM Validation program (Amitai, 2000)

• When using any gauge adjustment technique for radar rainfall estimation, independent QC of radar and gauge data alone is not sufficient. Proper QC of rain gauge data upon comparison with radar data is essential
• AQC increased rainfall by 13%.

Comparing TRMM PR-NEXRAD PDF of Rain Rate Central Florida 1998-2002 (105 overpasses)
Comparing TRMM PR-NEXRAD PDF of Rain Rate
0.15 PR (V5): 21,877 pixels; 6.2 mm/h GV (V5): 25,297 pixels ; 5.5 mm /h

Advantage of PDF comparisons  Free of large uncertainties associated with pixel by pixel comparisons  Free of satellite temporal sampling errors associated with the monthly products;  PDF of rain volume by R are less sensitive to instrument thresholds and with a direct hydrological significance

Relative rain volume

PR/G V=0.96


Which curve better represents the truth?
0 -10 -5 0 5 10 15 20 25

Rain rate [dBR]
Distribution of rain volume by R for the Melbourne, FL WSR-88D (GV V5) and TRMM PR V5 datasets based on 105 overpasses during 1998-2002 and co-located GV data of less than 100 km from Melbourne

Dense gauge networks required for better estimation of the true rain rate distribution at the scale of a radar pixel.

NMQ provides larger sample size
Amitai E. et al. 2005: Accuracy verification of spaceborne radar estimates of rain rate. The Royal Meteorological Society Atmospheric Science Letters (ASL), 6, 2-6.

• PR underestimates the rain by 4% compared to GV radar estimates, but also does not detect 4.5% of the rain. When PR detects rain, it compares well with GV estimates

1998 V5
0.15 PR (V5): 5,849 pixels ; 8.6 mm /h GV (V5): 6,537 pixels ; 6.7 m m/h; PR/GV=1.16 0.15

1998 V6
PR (V6): 5,819 pixels ; 6.7 mm /h GV (V5): 6,537 pixels ; 6.7 m m/h; PR/GV=0.89

Relative rain volume


Relative rain volume
-5 0 5 10 15 20 25




0 -10

0 -10








Rain rate [dBR]

Rain rate [dBR]

Distribution of rain volume by R for the1998 Melbourne, Florida, WSR-88D (GV) and TRMM PR V5 & V6 datasets. V6/V5 PR rain accumulation: 0.77

Which (PR) curve better represents the truth…

A framework for validation of spaceborne estimates of R
Amitai et al. 2005: Accuracy verification of spaceborne radar estimates of rain rate. The Royal Meteorological Society Atmospheric Science Letters (ASL), 6, 2-6.
Framework Features • Demonstrates how a hydrologic approach that uses statistical properties of the precipitation to estimate the uncertainties can be combined with a meteorological approach that uses physical properties of the rainfall • Based on comparing PDF of R from gauge, groundand space-based radar observations • Includes the use of PDF comparisons after rain type classification. This will allow for 1) better evaluation of the algorithms under different conditions (Physical validation); 2) extrapolation of the uncertainties to regions not covered by validation data sets, but characterized by the same rain types (Globalization) • Focuses on determining and reducing the uncertainties in the GV pdfs (PMM, super dense gauge networks)

NMQ allows to verify that uncertainties associate with a given rain type remain the same at different locations

Comparing GV-Satellite PDFs After Rain Type Classification

Determining/Reducing Uncertainties in GV PDFs
(Using Super Dense Gauge Networks)

Refining Classification Scheme Testing Stability of PDFs in Time/Space Determining Relative Errors; Error Characterization

Physical Validation Detecting, Quantifying and Reporting Errors in Satellite Algorithms Understanding the Processes Responsible for the Estimate Uncertainties

Globalization Extrapolating Uncertainties in Satellite Estimates to Non-GV Regions Characterized by the Same Rain Types

NASA/TRMM Products for Improved NMQ QPE Products

TRMM GV Data Flow
TRMM GVS Radar Data
Kwajalein, Melbourne, Houston, Darwin

1B-51 Radar QC 1C-51 2A-54 2A-55
QC can be an iterative process

1B-51: 1C-51: 2A-52: 2A-53: 2A-54: 2A-55: 2A-56: 3A-53: 3A-54: 3A-55:

Raw radar reflectivity QC radar reflectivity Rain existence Instantaneous rain rate Stratiform/convective rain type Three-dimensional reflectivity Rain gauge data 5-day rainfall accumulation Monthly rainfall accumulation Monthly 3-D reflectivity


**All products sent to TRMM Science Data & Information System (TSDIS), then to the Goddard Distributed Active Archive Center (GDAAC)**

Automated, Gauge-adjusted Z-R table creation to use as input for standardized rainfall products




Input: 1C-51, 2A-54, & gauge-adjusted WPMM ZR table Output: Instantaneous rain rate (mm/hr) Data truncated at 150 km Horizontal resolution: 2 x 2 km2

Input: 2A-53 Output: Monthly rainfall accumulation Data truncated at 150 km Horizontal resolution: 2 x 2 km2

NASA/TRMM Products for Improved NMQ QPE Products
TRMM PR : 2A-25 (3-D Reflectivity, Near Surface R)

QuickTime™ and a Animation decompressor are needed to see this picture.

NASA/TRMM Products for Improved NMQ QPE Products
TRMM GV: 2A-53, 3A-54 TRMM PR : 2A-25
Additional information regarding TRMM GV climatological product generation, development, and rainfall statistics:
Official TRMM products can be ordered from: