| [Français] |
| Home | Participants | Model table |
Index: Introduction |Brightness temperature results |Jacobian and transmittance results | HIRS results | AMSU results
1.1 INTRODUCTION
An intercomparison of radiative transfer models (RTM) with specific
application to NOAA-12 channel 12 (6.7 micron) was recently carried out
under GVaP (GEWEX Global Water Vapor Project, Soden et al, submitted to
BAMS, 1999). At the ITSC- 10 1999 meeting in Boulder, the working group
on radiative transfer proposed to extend this intercomparison to several
HIRS and AMSU channels, with a special focus on the Jacobian intercomparison.
This activity is formally organized here. The Jacobian of a RTM expresses
the derivative of brightness temperature (Tb) with respect to model state
variables influencing the RTE, such as temperature, T, specific humidity,
Q, surface skin temperature, Ts, and pressure, Ps. The Jacobian is a fundamental
quantity required for the direct assimilation of radiances. Physically,
the Jacobian is similar to the so-called weighting function which is the
derivative of total transmittance with pressure. However it is more specific
because it expresses separately the sensitivity to T, Q, Ps or Ts. In numerical
weather prediction (NWP), Jacobians are computed by a routine referred
to as the Jacobian or gradient routine. Analysis increments in NWP will
differ if the RTM and Jacobians differ. The GVaP experience mentioned above
has already indicated significant differences in Jacobians of several RTMs.
This intercomparison will explore this issue in detail along with a standard
forward Tb validation. This invitation is addressed to anyone who wishes
to test a particular RTM. Many, in fact most RTMs do not have an accompanying
Jacobian routine. However Jacobians can easily be evaluated by brute force
from imposed perturbations to input profiles of the RTM.
1.2 CHANNEL SELECTION
A subset of 7 HIRS (infrared) channels from NOAA-14 are to be processed: channels 2, 5, 9, 10, 11, 12, 15. In addition, results are requested for AMSU (microwave) channels 6, 10, 14 and 18 from NOAA-15. These channels were chosen as most representative to study the impact of the quality of RTMs on temperature and humidity retrievals and to evaluate the influence of ozone. Technical details concerning the channels are provided in the Appendix. Those participants who cannot process all 11 channels (most models will not apply both to infrared and microwave channels) are invited to submit partial results, even for a single channel.
1.3 PROFILE SELECTION
A group of 42 profiles (txt,
gz) representative of most
meteorological situations (including
extremes in temperature, integrated water vapor and ozone) are to be processed
for the forward Tb (and transmittance) validation. Jacobians for only 5
profiles are requested due to the cost of its calculation from line-by-line
models. Thus in total, 11 channels times 42 profiles are requested.
1.4 REQUESTED QUANTITIES
For each vertical level, the following vector quantities are requested: 1) the top-of-the atmosphere (TOA) total transmittance (from level to space) 2) the TOA specific humidity transmittance (lines and continuum together) 3) the TOA ozone transmittance 4) the temperature Jacobian 5) the specific humidity Jacobian 6) the ozone Jacobian The following non-vector quantities are also requested: 7 ) the TOA brightness temperature 8) the Ts Jacobian 9) the Ps Jacobian The vertical profiles are defined on the 43 pressure levels of the RTTOV-5 RTM. The output format is provided in Appendix. Ozone transmittances and Jacobians are only requested for HIRS 2, 5 and 9. Specific humidity inferences are requested for all HIRS channels and for AMSU-18. Jacobians are only requested for five profiles (6,18, 19,30, 31). Where not calculated, Jacobians and transmittances should be set to 999.
Important note: If your RTM cannot operate directly on the chosen 43 levels, for instance if it is designed to work on fixed levels which differ from the chosen 43, then provide the same results as asked here but on your set of vertical levels. Do not interpolate to the 43 levels. The organizers will do it for you and use the same procedure for everyone to interpolate Jacobians to the 43-level grid.
1.5 JACOBIAN DEFINITION
The temperature Jacobian has units of Tb/T, hence K/K. The H2O and O3 Jacobians, formally in units of Tb/Q, hence K/(kg/kg) are to be multiplied by -10% Q(H2O) and -10% Q(O3). Hence the temperature Jacobian refers to the Tb change due to a 1 K local increase in T and the water vapor and ozone Jacobians refer to the Tb change due to a 10% decrease in concentration. Participants are free to do their vertical integrals as they chose. In principle, for a level I , the Jacobian should be representative of the layer between (P(I-1)-P(I))/2 and (P(I)-P(I+1))/2. Jacobians at the bottom/top levels should be representative of the lower/upper half of the lowest/highest layer. The Ts Jacobian is in K/K and the Ps Jacobian in K/mb.
1.6 COMPUTING A JACOBIAN BY BRUTE FORCE
For those who do not have a Jacobian routine, here is the proposed method to get Jacobians by brute force.
1.6.1 Temperature Jacobian
For a given level with temperature T compute Tb(T+0.5) corresponding to a local increase of 0.5 K on T. Similarly compute Tb(T-0.5) corresponding to a local decrease of 0.5 K. The temperature Jacobian for that level will be: Tb(T+0.5) - Tb(T-0.5). The Ts Jacobian is defined similarly. Your code should be precise to at least 7 significant digits.
1.6.2 H2O and O3 Jacobians
For a given level with mixing ratio Q, compute Tb(Q+5%Q) and Tb(Q-5%Q) corresponding to a local increase/decrease of 5% in Q(H2O) or Q(O3) in kg/kg. The Jacobian will be: -(Tb(Q+5%Q) - Tb(Q-5%Q)) or Tb(Q-5%Q) - Tb(Q+5%Q). Thus for N levels, the computation of each Jacobian requires 2N realizations of the RTM. Participants should all use this method, not the less precise one consisting in defining the T Jacobian as Tb(+1K) - Tb(unperturbed) which requires N realizations.
1.6.3 Ps Jacobian
Similarly, the Ps Jacobian is defined by Tb(Ps + 0.5 mb) - Tb(Ps - 0.5 mb)
1.7 LBL REFERENCE AS "TRUTH"
Line-by-line models (LBL) will serve as a reference for "truth". At least one LBL will provide results, the FLBL model. Participation from other LBL schemes is most welcome.
1.8 DEADLINE
Results are expected to be received via email. New deadline is end of January 2000.
1.9 PUBLICATION
A publication on this intercomparison is likely to be submitted in early 2000. The goal of this publication will be to evaluate the level of differences between the various models and to explain these in a sensible way. There is no doubt that the information provided on the total, H2O and O3 TOA transmittances will help in the interpretation of the results. In the end, weaknesses and strengths of fast RTMs used in radiance assimilation in NWP will be better understood.
APPENDIX: 1 ,
2 ,
3 ,
4 ,
5 ,
6 .
For each variable a GIF image (coarser resolution) are done. The PostScript (PS) file contains graphics from the table and more.
| Value of STD(K) for BT | Evaluation |
| < 0.1 | Excellent. |
| 0.1-0.2 | Very good. |
| 0.2-0.3 | Good. |
| 0.3-0.5 | Weak. |
| >0.5 | Poor. |
A mesure of goodness of fit, M, is defined as follows:
M=100.0*( SUM(Xi-Xref)2 / SUM(Xref)2)1/2
where Xi and Xref are model and reference Variables, and the sum is
over the 43 vertical levels. M here is evaluated for single profiles. This
measure can be interpreted as an overall percentage of error. The measure
has the same meaning for all channels (as opposed to straigt standard deviation
and bias measures). Values of M can be associated with the following evaluation
table:
| Value of M for Jacobian | Evaluation |
| 0-5 | Excellent fit, typical among LBL models. |
| 5-10 | Very good fit, achievable by fast models in some, but not all channels. |
| 10-20 | Fair fit, likely acceptable for NWP applications, some concerns above 18. |
| 20-30 | Weak fit, improvements strongly suggested for use in NWP. |
| >30 | Bad fit, non acceptable. |
| Value of M for Transmittance | Evaluation |
| < 0.5 | Excellent fit. |
| < 1.0 | Very good fit. |
| 1-2 | Good fit. |
| 2-4 | Weak fit. |
| >4 | Bad fit. |
Note: Values of M have little meaning if the minimum TOA transmittance is very close to unity or if the maximum Jacobian value is less than 0.005K.
For each variable a GIF images (caorser resolution) are done. For J's and transmittance, the PostScript (PS) file contains graphics from the table. For transmittance, the optical depth is calculated. A scaterred plot is produced to compare with each LBL model.
N.B.
The GIF files (gif1=log scale, gif2=linear scale) associate with the table have a coarser resolution. To see details download the PS (with scatter plot) or PS_linear (without scatter plot) file and use a viewing program (UNIX: ghostview). You can select or print the graphics you want.
The following files PS.Z or PS_lin.Z are UNIX compressed file containing the PostScript. To uncompress use the UNIX command uncompress. (ex: uncompress PS.Z)
The PostScript (PS) files have a lot of graphics.
| Brightness temperature | AMSU06 gif | AMSU10 gif | AMSU14 gif | AMSU18 gif | PS.Z |
| Temperature jacobian | AMSU06 gif1 gif2 | AMSU10 gif1 gif2 | AMSU14 gif1 gif2 | AMSU18 gif1 gif2 | PS.Z PS_lin.Z |
| H2O jacobian | AMSU06 gif1 gif2 | AMSU10 gif1 gif2 | AMSU14 gif1 gif2 | AMSU18 gif1 gif2 | PS.Z PS_lin.Z |
| Total transmittance | AMSU06 gif1 gif2 | AMSU10 gif1 gif2 | AMSU14 gif1 gif2 | AMSU18 gif1 gif2 | PS.Z PS_lin.Z |
| H2O transmittance | AMSU06 gif1 gif2 | AMSU10 gif1 gif2 | AMSU14 gif1 gif2 | AMSU18 gif1 gif2 | PS.Z PS_lin.Z |
The PostScript (PS) files have a lot of graphics.
AMSU_LBL (AER_LBL, MSC_MWLBL, ATM, CIMMS_MWLBL) results in PS format. ( PS.Z , PS_lin.Z )
| Commentaires/Comments | Remerciements/Acknowledgements |
Last Modification/Dernière modification: 28-Aout/August-2000. .
