1. How the Products Relate
TRMM and GPM Products
George J. Huffman
NASA/GSFC
Introduction
How the Products Relate
Getting the Data
Schedule
A Little More IMERG
Final Comments

2. 1. INTRODUCTION
TRMM started with a goal of 5°x5° monthly data for the tropics
• the mission expanded to a much more applications-oriented focus
• user-oriented services developed
GPM built on TRMM’s success
• better sensors
• constellation as an integral part of the project
• improved algorithms
But, GPM is not just TRMM-plus
• zero-base review
• next-generation algorithms
• requires significant time to tune up the system

3. 1. INTRODUCTION – Data Sources
A diverse, changing, uncoordinated set of precip-relevant input data
• all have passive microwave (PMW) sensors
• TRMM and GPM have radar
• constellation was experimental in TRMM; only used for multi- satellite
• constellation a formal part of GPM
• we also use the global constellation of geo-IR
Precessing satellites gradually change their observation time
• depends on orbit parameters
• TRMM – 46 days
• GPM – 83 days
• Megha-Tropiques – 51 days

4. 1. INTRODUCTION – Data Coverage The current PMW coverage is “dense”
• about 90% of the day it meets the ≤ 3 hr criterion, needed to
• capture the diurnal cycle
• cover the decorrelation time of precipitation systems
• TMPA data interval of 3 hr gives ~1 sample in 3 hr
• a finer time interval is needed for smoother sampling
• for a ½ hr accumulation there is lots of empty space
• the state-of-the-art solution is “morphing”
• CMORPH, GSMaP, IMERG (uses CMORPH)
[PMW] HQprecipSource [identifier] Z 3 April 2014
GMI
TMI
AMSR2
MHS
SSMIS

5. 2. PRODUCTS – Data Levels
Level 0
• original instrument data (i.e., counts)
Level 1
• 1A: time-reference and ancillary information computed and appended, but not applied, to Level 0 data
• 1B: radiometrically corrected and geo-located Level 1A data that are processed to sensor units
• 1C: brightness temperature (Tc) products intercalibrated to the Level 1B GPM (GMI)
Level 2
• derived geophysical parameters at the Level 1 resolution and locations
Level 3
• parameters spatially and/or temporally resampled from Level 1 or 2 data

6. 2. PRODUCTS – TRMM and GPM Data Sets
Note – see packet for more details on GPM data sets
TRMM
GPM
VIRS [ 1B01 ] –
TMI [ 1B11 ]
GMI, partner [ 1B.GPM.GMI, 1C.<sat>.<sensor>.XCAL,
1C-R.<sat>.<sensor>.XCAL ]
PR [ 1B21 ]
DPR [ GPMCOR_KAR, GPMCOR_KUR ]
TMI [ 2A12 ]
GMI, partner [ 2A.<sat>.<sensor>.GPROF,
2A-CLIM. <sat>.<sensor>.GPROF ]
PR [ 2A21, 2A23, 2A25 ]
DPR [ 2A.GPM.<DPR,Ka,Ku> ]
TMI & PR [ 2B31 ]
GMI & DPR [ 2B.GPM.DPRGMI.CORRA ]
TMI [ 3A11, 3A12 ]
GMI [ 3A-<DAY,MO>.<sat>.<sensor>.grid,
3A-CLIM-<DAY,MO>.<sat>.<sensor>.grid ]
PR [ 3A25, 3A26 ]
DPR [ 3A-DAY-<ASC,DES>.GPM.DPR,
3A-<DAY,MO>.GPM.DPR ]
TMI & PR [ 3B31 ]
GMI & DPR [ 3B-<DAY,MO>.GPM.DPRGMI.CORRA ]
Latent heat [ 3G25, 3H25,
3G31, 3H31 ]
Latent heat [ pending ]
Multi-sat. [ 3B42, 3B43 ]
Multi-sat. [ 3B-HHR<none,-E,-L>.MS.MRG.3IMERG,
3B-MO.MS.MRG.3IMERG ], [ GSMAP at JAXA ]

7. Data access pages (should) give all GSFC instances of data sets
3. GETTING THE DATA
Data access pages (should) give all GSFC instances of data sets
• start with “Get Data” button on
Example access page (IMERG)
• General notice at top
• algorithm description
• links to documentation
• summaries and DownLoad links for various formats/runs
• scroll down for other Level 3 data
• click other Level tabs for 1 and 2
Feasible latency for NRT a subject of study 7

8. automatic registration
It’s all free!
3. GETTING THE DATA – A Few More Details
Systems that provide GPM data
• ftp://jsimpson.pps.eosdis.nasa.gov – NRT
• ftp://arthurhou.pps.eosdis.nasa.gov – production
• – production (order)
• – production (order)
• – production (interactive)
Latency
• Near-Real Time is within hours
• production is (at least) a few days
• IMERG has successive Runs
• Early – 6 hr (goal is 4) … forward morphing
• Late – 16 hr (goal is 12) … forward and backward morphing
• Final – 3 months … forward/backward morphing, better input, precip gauges
More about IMERG
• “Meet the Developer” over lunchtime
• “IMERG: Background and Early Results” poster
• ARSET training tomorrow, and webinars posted on the web •
system requires free,
automatic registration 8

9. 3. GETTING THE DATA – Shapefile Averaging
Giovanni 4 now provides averages over non-rectangular areas
• areas are represented in “shapefiles”
• initial selection includes watersheds and political entities
Example for Maryland, 1-10 April 2014, from IMERG Final Run 9

10. The TRMM satellite is done • fuel was depleted in July 2014
4. SCHEDULE – TRMM
The TRMM satellite is done
• fuel was depleted in July 2014
• PR products stopped 8 October 2014
• TMI was shut down 8 April 2015
• re-entry is “soon”
• TMPA-RT uses climatological calibration, so continues to run “as is”
• production TMPA partly depends on PR for calibration
• production switches to climatological calibration with October 2014
• performance is being scrutinized
• gauge calibration over land continues to yield consistent results
• climatological calibration over ocean causes a discontinuity
• loss of legacy sounder estimates or computer hardware could raise issues for continuing TMPA 10

11. GPM Day-1 products were brought up during 2014
4. SCHEDULE – GPM
GPM Day-1 products were brought up during 2014
• launch on 27 February 2014
• early release of Level 1 GMI in June 2014 to support tropical cyclone forecasts
• Level 2 data sets released in September 2014
• IMERG
• Final Run released in December 2014 (back to 12 March 2014)
• Late Run starts 7 March 2015
• Early Run starts 1 April 2015
First reprocessing
• Cross-Calibration Team has finalized radiometric calibration for PMW
• Version 2 Level 1 and 2 data sets to be released for GPM era in September 2015
• Version 2 IMERG to be released for GPM era in early 2016
Second reprocessing
• Cross-Calibration Team to finalize radiometric calibration for TMI in Spring 2016
• Version 3 Level 1 and 2 data sets to be released for TRMM/GPM era in Fall 2016
• Version 3 IMERG to be released for TRMM/GPM era in Winter 2017
• TMPA and TMPA shut down in late Spring 2017 11

12. “Day 1 IMERG Final Run Release Notes”
5. IMERG – Release Notes
“Day 1 IMERG Final Run Release Notes”
• an introduction and first cut at comparisons
• a living document
• IMERG_FinalRun_Day1_release_notes.pdf
• this and all documents hot-linked on the IMERG data access page, accessible on the Level 3 tab on
• specific features, problems, behaviors
• effects due to IMERG’s structure
• cautions due to the input data
• biggest overall issue is that none of the inputs or IMERG are fully GPM-based 12

13. 5. IMERG – Final Run vs. 3B43 for June 2014
(Mostly the) same input satellites, different algorithms, different calibrator
• 2BCMB vs. 2B31
Similar features, but not identical
• features (SPCZ)
• bias (ITCZ)
IMERG Final (mm/d) June 2014
ITCZ
SPCZ
TMPA 3B43 (mm/d) June 2014
Dashed lines on IMERG show the 50°N,S limits of TMPA 13

14. 20 surface gauges in a 6x5 km region near Wallops Island, Virginia
5. IMERG – Daily Comparison to Pocamoke Fine-Scale Grid, April-August 2014
20 surface gauges in a 6x5 km region near Wallops Island, Virginia
Excellent correlation for event occurrence (warm season)
Both over- and under-estimates for largest events
Days after 1 April 2014
[Courtesy J. Tan (UMBC; WFF)] 14

15. The transition from TRMM to GPM datasets is well underway
6. FINAL COMMENTS
GPM algorithms and processing differ significantly from TRMM counterparts
Guidance on access to TRMM and GPM data and documentation has been upgraded and consolidated in one set of pages
The transition from TRMM to GPM datasets is well underway
• Fully GPM-based retrievals will be available for Levels 1 and 2 in September 2015
• GPM-era IMERG should be reprocessed in early 2016
• TRMM/GPM-era products will be processed about a year later
• TMPA and TMPA-RT to be run until Spring 2017
The “Get Data” emblem is on a separate key-stroke. 15

16. Reserve Slides 16

17. 2. IMERG DESIGN – Processing
IMERG is a unified U.S. algorithm that takes advantage of
• Kalman Filter CMORPH (lagrangian time interpolation) – NOAA
• PERSIANN with Cloud Classification System (IR) – U.C. Irvine
• TMPA (inter-satellite calibration, gauge combination) – NASA
• all three have received PMM support
• PPS (input data assembly, processing environment) – NASA
The Japanese counterpart is GSMaP
Institutions are shown for
module origins, but
• package will be an integrated system
• goal is single code system appropriate for near-real and post-real time
• “the devil is in the details”
UC Irvine
GSFC
CPC
Compute even-odd IR files
(at CPC) 3
IR Image segmentation
feature extraction
patch classification
precip estimation 9
Receive/store
even-odd IR
files 1
Compute IR
displacement vectors 4 11
Recalibrate
precip rate
Build IR-PMW precip calibration 10
Import PMW data;
grid; calibrate; combine 2
Forward/backward
propagation 5
Build Kalman
filter weights 7
Apply Kalman
filter 8
Import mon. gauge;
mon. sat.-gauge
combo.;
rescale short-interval
datasets to monthly
Apply climo. cal. RT
Post-RT 13 12
prototype6

18. 5. VALIDATION – 3-Hourly, 0.25° IMERG, 3B42, MRMS for 15 June 2014
IMERG better than 3B42 for precip occurrence
IMERG performs modestly better for precip volume
Note: Original footprint GPROF retrievals below 0.1 mm/hr are thresholded to zero
• how this affects IMERG depends on the resolution of the input sensor and subsequent averaging (here 0.25°)
[Courtesy J. Wang
(SSAI; NASA/GSFC 612)] 18

19. 5. VALIDATION – Snow in IMERG, NWS WSR88D, 12 March 2014
IMERG converted to dBZ, WSR88D in dBZ; both original resolution
Hang-back line in radar missing in IMERG
2-5” of snow with near-blizzard conditions at Cleveland, Ohio around 1900 UTC
IMERG (dBZ)
1700 UTC
1800 UTC
1900 UTC
WSR88D (dBZ)
1700 UTC
1800 UTC
1900 UTC
[Courtesy B. Rose (Weather Channel)] 19

20. • width shows relative contribution for each difference bin
5. IMERG – Half-Hourly Statistics for Individual Input Sources and Pocamoke Fine-Scale Grid, April-August 2014
“Violin diagram” for individual sources of the half-hourly IMERG estimates
• width shows relative contribution for each difference bin
Pocamoke has 20 gauges
in one IMERG grid box
GMI is best; AMSR
and SSMIS less so
The outliers for no-PMW
(interpolated) are partly
driven by the large
number of cases
No-PMW (interpolated)
data are competitive
with the skill for most
of the sensors
No PMW
TMI
AMSR
SSMIS
MHS
GMI
Number of cases
816 23 27
105
138 16
There’s a lot going on here, so you have to be succinct to avoid getting bogged down.
Not sure why AMSR is all positive, but the number o cases is small.
“No PMW” is mostly morphed interpolations, but perhaps also includes some IR. Jackson didn’t analyze that.
[Courtesy J. Tan (ORAU; WFF)] 20