1. An Inter-comparison of 5 HRPPs with 3-Hourly Gauge Estimates
Matt Sapiano
Cooperative Institute of Climate Studies (CICS), University of Maryland

2. Aim PEHRPP Suite 2: High time resolution comparisons
Inter-compare high-resolution precipitation
products with high time resolution gauge data
Aiming to keep it very simple – summary statistics commonly used and understood by the wider community
Also interested in performance of model forecast data…

3. HRPP Data used Validate at 0.25˚, 3-hourly Product Provider Data
Method
TRMM Multi-satellite precipitation analysis (TMPA, a.k.a. 3B42 or 3B42RT for Real Time)
GSFC (G. Huffman)
Geo-IR, microwave from SSM/I, TRMM, AMSU, AMSR
Merged microwave and microwave-calibrated infrared (IR)
CPC Morphing Technique (CMORPH)
NOAA CPC
(J. Janowiak,
B. Joyce)
Passive microwave (PMW) rain rates advected and evolved according to IR imagery
Hydro-Estimator (HE)
NOAA NESDIS ORA (B. Kuligowski)
Geo-IR, NWP
Tb in geostationary-IR, modulated by cloud evolution, stability, total precipitable water, etc.
NRL blended algorithm
(NRL-Blended)
NRL (J. Turk)
Histogram-matching calibration of geo-IR to merged microwave
Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks (PERSIANN)
UC Irvine (K.-L. Hsu)
Geo-IR, TRMM microwave
Adaptive neural network calibration of geo-IR to TRMM TMI
Validate at 0.25˚, 3-hourly

4. Validation data Long record, sub-daily gauges required:
ARM Southern Great Plains (SGP - Oklahoma, Kansas)
TAO/TRITON Buoy array (Tropical Pacific)
2 undercatch corrections applied based on wind and threshold
Evaluate between Dec 2002 and March 2006
Separate 6-month seasons; require >1 year of data
Use weighted average of nearest 4 grid-points from HRPP
Match gauge data to HRPPs (offset of TMPA)
Exclude buoy stations with prob of precip <0.1
Also include Stage IV radar and NARR in analysis
Use these as benchmarks for skill

5. 3-hourly Correlations SGP ONDJFM AMJJAS TAO Buoys TMPA CMOR HE NRL PER
NARR
ST-IV
TMPA
CMOR HE
NRL
PER
NARR
ST-IV
ONDJFM
AMJJAS
TAO Buoys
TMPA
CMOR
NRL
PER
TMPA
CMOR
NRL
PER

6. 3-hourly Correlations by month
Plots show the mean 3-hourly correlation over all gauges in both areas as a time series
Inputs are similar – do HRPPs have same skill?
NARR and Stage IV perform best
CMORPH and other “MW” datasets perform well
PERSIANN has different skill – relies heavily on IR
Suggests that IR contains information that MW misses (duh!) – further combination might be useful…
SGP
TAO Buoys

7. Bias SGP ONDJFM AMJJAS TAO Buoys TMPA CMOR HE NRL PER NARR ST-IV TMPA

8. Model Forecast data: NOAA GFS
GFS 12 and 15 hour forecast precipitation used as a test of performance of model data
Models have well-known problems with diurnal cycle of precip: sub-daily precip a challenge
Treat GFS as a HRPP and include CMORPH and Stage IV for comparison
At 3-hourly resolution, GFS is inferior to CMORPH
GFS performs much better at daily resolution
Generally poor in cold season
ONDJFM
AMJJAS
CMOR
GFS
ST-IV
CMOR
GFS
ST-IV
CMOR GF
ST-IV
CMOR
GFS
ST-IV
3-Hourly
Daily
3-Hourly
Daily

9. Summary
HRPPs all do well; CMORPH out-performs others over these particular validation areas
TMPA, PERSIANN and Hydro-Estimator are close behind
Bias correction of TMPA is a huge advantage over land, but biases appear comparable to other data over ocean
HRPPS have different strengths based on use of IR, MW or other data
Further combination of techniques might yield skill increases
Model data might be superior at daily scales
BUT: inferior at sub-daily scales – users should apply caution
These results might not be typical of all areas, but do give interesting insight into sub-daily behavior

10. Relevance to PEHRPP objectives
Workshop Objective 1 /Precip Product issues 1: Common strengths and weaknesses
Results show different inputs yield slightly different types of skill (combination good)
Precip Product issues 2+3: use of NWP
Example of GFS shows skill at daily, but difficulty with sub-daily (need for further evaluation of different datasets)
Regional Validation 1: how to validate techniques at sub-daily scales
Biggest issue is data length and quality (methods from daily are suitable)

11. Diurnal Cycle HRPPs mostly capture the shape of the diurnal cycle
All have max slightly early; PERSIANN especially - shouldn’t this be the other way around?
Gauge corrected TMPA has advantage over land and correctly captures amplitude
Same is not true over ocean where it has similar pitfalls to CMORPH
GFS not as bad as might be expected