1. Forecast Sensitivity - Observation Impact (FSOI)
Inter-comparison Experiment
Tom Auligné, Joint Center for Satellite Data Assimilation (JCSDA)
Ron Gelaro, NASA, Global Modeling and Assimilation Office (GMAO)
Rahul Mahajan, David Groff, NOAA, National Weather Service (NWS)
Rolf Langland, Naval Research Laboratory (NRL)
Jianjun Liu, NOAA’s Satellite and Information Service ( NESDIS)
James Cotton, Larry Morgan, UK Met Office
Yoichiro Ota, Japan Meterological Agency (JMA)

2. Forecast Sensitivity – Observation Impact (FSOI)
Fcst Error
observations assimilated
Langland and Baker (2004)
Time
-6 h
0 h
+24 h
Build the illustration in a few steps, click by click....
Adjoint-derived (single outer-loop) observation impact
Ensemble-derived observation impact

3. Motivation
Several NWP centers are computing FSOI routinely to monitor/understand/tune their DA system.
Are relative impact of various observation types comparable?
Can we learn from similarities/differences to improve NWP systems?
Multiple NWP Centers expressed interest in this study NRL, GMAO, EMC, Met Office, ECMWF, Meteo-France, KMA, JMA, Env. Canada

4. Motivation (cont.)
Multi-staged and multi-faceted set of experiments. Participants free to contribute commensurate with interests and capability.
Initially a narrowly defined FSOI experiment—much in the mold of the previous THORPEX inter-comparison (Gelaro et al. 2010)
Original aspects
Broader participation of NWP centers (Adjoint & Ensemble-based)
Latest global NWP systems
Operational (not baseline) observing network
Approach
centrally collect FSOI output without aggregation of the data
Convert into common data structure
Flexibility to stratify information.

5. Experimental Design
Time period: 3-month DJF (planned JJA 2014) 00UTC & 06UTC cycles
Verification:  24h forecast against self analysis
Metric: global total dry energy (surface-100hPa)
Adjoint: dry plus moist physics, as available
Ensemble: flow-following localization
Results shown here are VERY preliminary (only global summary plots of impact at 00UTC will be shown)

6. Participating NWP Centers
NRL
GMAO
Met Office
JMA Adjoint
JMA Ensemble
EMC
Analysis System
3DVar In Observation Space
Hybrid 3DVar
4DVar
LETKF re-centered via 4DVar
EnKF re-centered via 4DEnVar
FSOI Technique
Adjoint
Ensemble
Experiment Resolution
T119L60
Model: 25km
DA: 50km
Ens: 100km
Model N320 (40km)
Adjoint N216 (60km)
Model TL959L100
Adjoint TL319L100
Ensemble (x50) TL319L100
Ensemble (x80) T254
Specific Considerations
QC = channel selection + dynamical observation error
~30% cycles discarded due to spurious impacts
Additional thinning of observations except for aircraft data

7. COMMON

8. Observation Impact at 00UTC: Observation Count
GMAO
NRL
Met Office
106
106`
105
JMA Adjoint
JMA Ensemble
EMC

9. Observation Impact at 00UTC: Observation Count

10. Observation Impact at 00UTC: Total Impact
0.35
0.3x2
0.3
0.06
3.0
GMAO
NRL
Met Office
0.3
0.06
JMA Adjoint
JMA Ensemble
EMC

11. Fractional Impact at 00UTC: Satellite Radiances

12. Fractional Impact at 00UTC: Other Observations

13. Observation Impact at 00UTC: Fraction of Beneficial Observations
Impact < 0  Beneficial
Impact > 0  Detrimental
ε=10-10
Impact < -ε  Beneficial
Impact > ε  Detrimental
-ε < Impact < ε  Neutral

14. Observation Impact at 00UTC: Fraction of Neutral Observations

15. Total Impact: METOP-B AMSUA Ch. 6
Ensemble
Adjoint
GSI
JMA

16. Adjoint

17. Conclusion and Perspectives
Confirmation of results from Gelaro et al. (2010), noticeable additional impact from hyperspectral IR sounders
Discrepancies between adjoint- and ensemble-based FSOI values. Further investigation needed.
Many more composite plots can provide detailed analysis
Mean/variance maps (across centers)
Time/spatial correlations
Ancillary interests not covered so far
investigation of different types of norms and verifications methods,
more participating NWP centers: global & regional NWP systems.
Possibly set up Near-Real-Time multi-agency monitoring capability

18. Questions?