i 5.11 validation of the gmao osse prototype runhua yang 1,2 and ronald errico 1,3 1 global modeling...

1
i 5.11 Validation of the GMAO OSSE Prototype Runhua Yang 1,2 and Ronald Errico 1,3 1 Global Modeling and Assimilation office, GSFC, NASA 2 Science Systems and Applications Inc. 3 Goddard Earth Sciences and Technology Center, UMBC Characteristics of the prototype OSSE 1.Simulating cloud effects on IR radiances-use a cloud probability as a function of NR grid-mean cloud fractions. The procedure is tunable to account for possible deficiencies in NR’s clouds and some implicit representativeness errors due to different models, resolution, etc. 2. Explicit random errors are drawn from a normal distribution (0,0.65R), where R is the sum of the instrument plus representativeness errors found in the GSI observation error tables. 3. GMAO DAS system includes a GSI adjoint to efficiently estimate the reductions in 1-day forecast error due to individual sets of observations. This technique is applied for OSSE validation. 4.Experiments: A reference run and an OSSE run for Jan. 2006, spin-up starts from Dec. 16, 2005 with the resolution of 2 0 in lat, 2.5 0 in lon, L72. Observing systems: conventional data (Approx: 1.4MB/day) + radiance data from HIRS2/3, AMSUA/B, AIRS, and MSU (Approx. 3MB/day) Validation It is imperative that a baseline OSSE be well validated. The OSSE should faithfully reproduce many metrics used to assess observations and data assimilation systems when similar existing observing systems are considered. The metrics include an estimation of observation impact, statistics of analysis increments and characteristics of innovation (observation minus background). . This table shows: 1. for surface pressure, wind, and wind speed, OSSE’s values are close to that of the Reference run; 2. for temperature, OSSE produces much smaller values than in the Reference; 3. for radiance, OSSE produces much larger values than in the Reference. The following two sets figures show the comparison of simulated radiance versus the real observation for a few selected observing systems. HIRS3_N17 observation location and histogram of OMF (accepted after GSI Quality Control) for the +/- 3hrs of 00z Jan. 30 2006. Right column: CTL. Left column: OSSE Introduction The Global Modeling and Assimilation Office at NASA has been participating in the international Joint Observing System Simulation Experiment organized by ECMWF and NCEP. Our contribution to this effort includes the development of the method and software for simulating observing systems; conducting OSSEs; and the development of metrics/technique for the validation of the OSSEs. Design of an OSSE Capability at the GMAO: Goals: 1.Be able to estimate the effect of proposed instruments on analysis and forecast skill by “flying” them in a simulated environment. 2. Be able to evaluate present and proposed data assimilation techniques in a simulation where “truth” is known perfectly. Requirements: 1. A self-consistent and realistic simulation of nature. One such data set has been provided to the community by ECMWF through NCEP (called Nature Run, NR). 2. Simulation of all presently-utilized observations, derived from the “nature run” and having instrument error characteristic plus representativeness error characteristic of real observations. 3. A validated baseline assimilation with the simulated data that produces various relevant statistics similar to corresponding ones in a real DAS. Summary 1.In general, the agreement between OSSE and real results is remarkable, especially since the specification of 2005 specifies only the data coverage and sea surface temperature but not the variants of synoptic situations present. 2. There are clear differences in both RMS and the means of analysis increments between the OSSE and the reference, particularly over the storm track region. Need further study. On-going work 1. Simulation of more realistic representativeness and instrument error characteristics. 2. Understand/study the satellite bias correction. 3. Use different RTMs, say RTTOVS and CRTM, for simulation and assimilation to study source of representativeness error. i i i g y e ~ 3 Where is innovation, and is an estimate of the impact of a single observation computed using adjoint model of GMAO forecast model and GSI. i y i g ~ Reference OSSE Surface pressure 0.320 /48416 0.279 /48305 Temperature 2.448 /46282 1.383 /45776 Wind 1.109 / 307505 0.933 /300412 Moisture 1.330 /6010 1.178 /5892 Wind speed 1.183 /16594 1.373 /16594 Radiance 0.258 / 919018 0.341 /820302 HIRS3-N17 Ch7 (peak of weighting function: ~700mb) This figure shows the mean change in E-norm of the 24- hour forecast error due to assimilating the indicated observation types at 0 UTC for OSSE (top) and real assimilation, or CTL (bottom) for the period of January 2006. 2.Characteristics of analysis system Table: January mean of Jo/n and observation count (in blue) 3. Root Mean Square and mean of analysis increment HIRS3-N17 Ch4 (peak of weighting function: ~200mb) AMSUA_N16 channel 2 observation location and histogram of OMF (accepted after GSI Quality Control) for the +/- 3hrs of 00z Jan. 30 2006. Top panels: CTL. Bottom panels: OSSE-where observation over land areas are not included. We have now addressed the problem of big radiance bias in OSSE: two different versions of CRTM are used, one for simulating observation data, and the other in the assimilation system. The latter uses a pre-released version of CRTM that perhaps has deficiency in surface emissivity model. Since we do not use satellite bias correction in the current OSSE, the bias is not eliminated during the assimilation process. Ignore colors

Upload: andra-hoover

Post on 19-Jan-2016

216 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: I 5.11 Validation of the GMAO OSSE Prototype Runhua Yang 1,2 and Ronald Errico 1,3 1 Global Modeling and Assimilation office, GSFC, NASA 2 Science Systems

i

5.11 Validation of the GMAO OSSE Prototype

Runhua Yang1,2 and Ronald Errico1,3

1Global Modeling and Assimilation office, GSFC, NASA2Science Systems and Applications Inc.

3Goddard Earth Sciences and Technology Center, UMBC

Characteristics of the prototype OSSE

1.Simulating cloud effects on IR radiances-use a cloud probability as a function of NR grid-mean cloud fractions. The procedure is tunable to account for possible deficiencies in NR’s clouds and some implicit representativeness errors due to different models, resolution, etc.

2. Explicit random errors are drawn from a normal distribution (0,0.65R), where R is the sum of the instrument plus representativeness errors found in the GSI observation error tables.

3. GMAO DAS system includes a GSI adjoint to efficiently estimate the reductions in 1-day forecast error due to individual sets of observations. This technique is applied for OSSE validation.

4. Experiments: A reference run and an OSSE run for Jan. 2006, spin-up starts from Dec. 16, 2005 with the resolution of 20 in lat, 2.50 in lon, L72.

Observing systems: conventional data (Approx: 1.4MB/day) + radiance data from HIRS2/3, AMSUA/B, AIRS, and MSU (Approx. 3MB/day)

Validation It is imperative that a baseline OSSE be well validated. The

OSSE should faithfully reproduce many metrics used to assess observations and data assimilation systems when similar existing observing systems are considered. The metrics include an estimation of observation impact, statistics of analysis increments and characteristics of innovation (observation minus background).

1. Observation impact –OSSE versus Control Estimation of observation impact is approximately computed

as (Ref. Langland and Baker 2004; Errico 2007 Tellus; Gelaro Zhu and Errico 2007 Meteorol Z.):

.

This table shows: 1. for surface pressure, wind, and wind speed, OSSE’s values are close to that of the Reference run; 2. for temperature, OSSE produces much smaller values than in the Reference; 3. for radiance, OSSE produces much larger values than in the Reference. The following two sets figures show the comparison of simulated radiance versus the real observation for a few selected observing systems.

HIRS3_N17 observation location and histogram of OMF (accepted after GSI Quality Control) for the +/- 3hrs of 00z Jan. 30 2006. Right column: CTL. Left column: OSSE

IntroductionThe Global Modeling and Assimilation Office at NASA has

been participating in the international Joint Observing System Simulation Experiment organized by ECMWF and NCEP. Our contribution to this effort includes the development of the method and software for simulating observing systems; conducting OSSEs; and the development of metrics/technique for the validation of the OSSEs.

Design of an OSSE Capability at the GMAO:Goals:1. Be able to estimate the effect of proposed instruments on

analysis and forecast skill by “flying” them in a simulated environment.

2. Be able to evaluate present and proposed data assimilation techniques in a simulation where “truth” is known perfectly.

Requirements: 1. A self-consistent and realistic simulation of nature. One such

data set has been provided to the community by ECMWF through NCEP (called Nature Run, NR).

2. Simulation of all presently-utilized observations, derived from the “nature run” and having instrument error characteristic plus representativeness error characteristic of real observations.

3. A validated baseline assimilation with the simulated data that produces various relevant statistics similar to corresponding ones in a real DAS.

Summary1.In general, the agreement between OSSE and real

results is remarkable, especially since the specification of 2005 specifies only the data coverage and sea surface temperature but not the variants of synoptic situations present.

2. There are clear differences in both RMS and the means of analysis increments between the OSSE and the reference, particularly over the storm track region. Need further study.

On-going work1. Simulation of more realistic representativeness and

instrument error characteristics. 2. Understand/study the satellite bias correction.3. Use different RTMs, say RTTOVS and CRTM, for

simulation and assimilation to study source of representativeness error.

4. Expanding the system to include more obs. types.

Acknowledgment: Meta Sienkiewicz, Emily Liu, Ricardo Todling, Ronald

Gelaro, Joanna Joiner, Tong Zhu, and Michele Rienecker

i ii gye ~3

Where is innovation, and is an estimate of the impact of a single observation computed using adjoint model of GMAO forecast model and GSI.

iy ig~

Reference OSSE

Surface pressure 0.320 /48416 0.279 /48305

Temperature 2.448 /46282 1.383 /45776

Wind 1.109 / 307505 0.933 /300412

Moisture 1.330 /6010 1.178 /5892

Wind speed 1.183 /16594 1.373 /16594

Radiance 0.258 / 919018 0.341 /820302

HIRS3-N17 Ch7 (peak of weighting function: ~700mb)

This figure shows the mean change in E-norm of the 24-hour forecast error due to assimilating the indicated observation types at 0 UTC for OSSE (top) and real assimilation, or CTL (bottom) for the period of January 2006.

2. Characteristics of analysis system

Table: January mean of Jo/n and observation count (in blue)

3. Root Mean Square and mean of analysis increment

HIRS3-N17 Ch4 (peak of weighting function: ~200mb)

AMSUA_N16 channel 2 observation location and histogram of OMF (accepted after GSI Quality Control) for the +/- 3hrs of 00z Jan. 30 2006. Top panels: CTL. Bottom panels: OSSE-where observation over land areas are not included.

We have now addressed the problem of big radiance bias in OSSE: two different versions of CRTM are used, one for simulating observation data, and the other in the assimilation system. The latter uses a pre-released version of CRTM that perhaps has deficiency in surface emissivity model. Since we do not use satellite bias correction in the current OSSE, the bias is not eliminated during the assimilation process.

Ignore colors