A New Inter-Comparison of Three Global Monthly SSM/I Precipitation Datasets Matt Sapiano, Phil Arkin and Tom Smith Earth Systems Science Interdisciplinary.

A New Inter-Comparison of Three Global Monthly SSM/I

Precipitation Datasets

Matt Sapiano, Phil Arkin and Tom SmithEarth Systems Science Interdisciplinary Center,

University of Maryland

2

Motivation• Currently working on a new reanalysis of

precipitation – Aim to use Optimal Interpolation to combine data sources

• Special Sensor Microwave/Imager (SSM/I)– One definite constituent of the reanalysis– Longest MW precipitation dataset (starts 1987)

• Several algorithms exist for estimation of precipitation– Goddard Profiling algorithm– NOAA/NESDIS algorithm (Ferraro)– Remote Sensing Systems algorithm (Wentz)

• Last comparison of these data was several years ago– So: compare them to inform precipitation analysis

→ Monthly averages, 2.5º resolution

3

Some SSM/I facts…• Defense Meteorological Satellite Program

Special Sensor Microwave/Imager– 7 channels: 19.35 (H+V), 21.235 (V), 37.0 (H+V), 85.5

(H+V) • Data from 1987 - present

F08 Jul 1987 – Dec 1991 F10 Dec 1990 – Nov 1997 F11 Dec 1991 – May 2000 F13 May 1995 – present F14 May 1997 – present F15 Dec 1999 – Aug 2006F16 Oct 2003 – presentNote: Not all channels were available during the record; notably, the 85GHz channel onboard the F08 satellite was unavailable from June 1990.

4

NOAA/NESDIS (Ferraro)• Scattering technique over land

– Grody Scattering Index (SI) from 19, 22 & 85 GHz channels

– Precip occurrence determined by SI>10– Screening for snow and ice– Precip empirically estimated from SI

• Scattering and emission over ocean– Precip occurrence from SI or emission (Q)– Precip empirically estimated from SI or Q

• Used 37GHz channel when 85GHz unavailable in 1990-91

• No overlapping periods for satellites that have similar local equator crossing times

5

RSS (Wentz)• Physically based retrieval of rain, wind, water

vapor– Estimate transmittance of liquid water from brightness

temperature, apply beam filling correction and derive atmospheric attenuation

– Mie scattering theory used to estimate columnar rain rate

– Columnar rain rate converted to surface rain rate using assumed column height from SST

• New version of algorithm released September 2006 (Version 06)– Improved beam filling– Improved relationship between column height and SST

6

GPROF SSM/I Version 6• Goddard Profiling algorithm

– Inversion scheme to retrieve vertical structure

• Instantaneous rainfall rates calculated from weighted average of existing hydrometeor profiles created using numerical cloud model– Goddard Cumulus Ensemble Model

• Land: Scattering technique• Ocean: Emission technique• Most recent version (V7) not applied to full

SSM/I dataset, so V6 is used here– Don’t be confused by naming conventions!!!

7

GPROF V6 Sea Ice Issue• Problem of sea ice contamination in GPROF SSM/I

Version 6– First NH (20-60º) EOF shows unphysical anomalies– Clearly an artifact (larger over Sea of Okhotsk)

• Correction applied here to remove anomalously large values– Gridpoint mean plus five times the zonal mean standard deviation

Pre

cipi

tati

on, m

m d

ay-1

8

Between satellite comparisons• Same local crossing times • RSS (Wentz) has more consistently higher correlations

and lower bias

GPROF V6 SSM/I F11 – F13

RSS V06 (Wentz) F14 – F15

GPROF r(F11,F13) after spatial 1-2-1 smoothing

→ Small spatial errors cause noisy correlation field

mm day-1

9

Some SSM/I facts…• Defense Meteorological Satellite Program

Special Sensor Microwave/Imager– 7 channels: 19.35 (H+V), 21.235 (V), 37.0 (H+V), 85.5

(H+V) • Data from 1987 - present

F08 Jul 1987 – Dec 1991 F10 Dec 1990 – Nov 1997 F11 Dec 1991 – May 2000 F13 May 1995 – present F14 May 1997 – present F15 Dec 1999 – Aug 2006F16 Oct 2003 – presentNote: Not all channels were available during the record; notably, the 85GHz channel onboard the F08 satellite was unavailable from June 1990.

10

Different time measurement - correlations• Correlations from different overpass times for

overlapping periods• Differences reflect diurnal cycle

F13

vs

F1

4F

10 v

s F

11

NOAA/NESDIS GPROF V6 SSM/I RSS V06 (Wentz)

11

Different time measurement - biasF

13 v

s F

14

F10

vs

F1

1

NOAA/NESDIS GPROF V6 SSM/I RSS V06 (Wentz)

• Bias from different overpass times• Wentz has good agreement between satellites• Different biases over land and ocean

– High tropical land diurnal variability is of consistent sign– Problem with biases at high latitudes in GPROF due to sea ice

12

Algorithm comparison - ocean• Zonal mean precipitation from all three algos

– Multiple lines represent the different satellites – diurnal cycle is evident• Good agreement between Ferraro and Wentz• Annual cycle dominates extra-tropics

Ocean only GPROF SSM/I Wentz Ferraro

20ºN – 60ºN

20ºS – 20ºN

60ºS – 20ºS

13

Wentz comparison• Wentz algorithm is quite different• Good advertisement for the benefits of re-processing

Ocean only GPROF SSM/I Wentz Ferraro

Wentz V05

Wentz V06

14

Algorithm comparison - Land• Only NOA/NESDIS and GPROF V6 as RSS is ocean only• Good agreement in annual cycle at higher latitudes, but magnitudes

disagree – GPROF V6 gives higher winter precipitation– Is this a problem with snow contamination?

Land only GPROF SSM/I Ferraro

20ºN – 60ºN

20ºS – 20ºN

60ºS – 20ºS

15

Gauge validation• Correlation with Chen et al. (2002) [GHCN+CAMS] and GPCC gauge

analyses (monitoring product)• NOAA/NESDIS data better correlated with gauges at higher latitudes

– Lack of profiles at high latitudes for GPROF V6?– Snow contamination problem again?

NOAA/NESDIS GPROF V6 SSM/I

Che

n et

al.

GP

CC

16

TAO buoy validation• Correlations with TAO/TRITON

buoy rain gauge data– Data from ATLAS 2 self

siphoning gauges– Data has been quality

controlled and an empirical wind correction was applied

• All three algorithms have high correlations with oceanic precipitation

• RSS (Wentz) V06 data has the highest correlations (not statistically significant though!)

NOAA/NESDIS

GPROF V6

RSS V06

60.0r

65.0r

62.0r

GPROF SSM/I Wentz Ferraro

17

Conclusions and Further Work• SSM/I data continues to increase in value as a climate data

record• RSS V6 algorithm performs well over oceans

– RSS also most homogeneous over the changing satellite record– RSS V06 bias appears to be superior to V05 bias

• Over land, NOAA/NESDIS appears to have better properties than GPROF SSM/I V6 at higher latitudes– GPROF SSM/I V6 is more homogeneous over the tropics– Lower correlations at mid/high latitudes is a problem

• Results from GPROF V6 SSM/I not applicable to most recent TMI product– Need for reprocessing of SSM/I using most recent GPROF algorithm

[This would make a nice recommendation for this workshop!]• Single satellite available before 1992

– Is data homogeneous? Effect of 85GHz failure?