News & Updates
JCSDA News and Newsletters
Data Assimilation Changes in the January 2015 GFS Model Upgrade at NOAA
From the Winter 2015 issue of the JCSDA Quarterly
On January 14, 2015, an operational upgrade to the NOAA National Centers for Environmental Prediction (NCEP) Global Forecast and Analysis (GFS/GDAS) was implemented. The forecast model has been changed significantly, most notably by a change from Eulerian to Semi-Lagrangian dynamics and an increase in resolution from Eulerian T574 (~27 km) to Semi-Lagrangian T1534 (~13 km). The full list of model changes can be found at www.nws.noaa.gov/om/notification/tin14-46gfs.htm.
Before this upgrade, the analysis was performed on a linear grid that corresponds to the model truncation T574 (1152x576 grid boxes) while the 80-member ensemble that prescribes 75 percent of the solution was at T254 (512x256). For the new system, both the GSI analysis and the members of the ensemble are computed at T574, so now the resolution of the main background error term (and therefore the typical resolution of the increments) is consistent with the analysis. The increment is then transformed into wave space and added to the full-resolution background. This approach is more consistent with what is done at most other NWP centers, although the observation innovations are still calculated at the truncated resolution- which should be addressed in future upgrades.
The Ensemble Kalman Filter (EnKF) members themselves benefited from the addition of stochastic physics to help address system uncertainty in the forecast model. This supplements, but does not entirely remove, the need for additive inflation.
On the satellite radiance side, the first-guess departure (O-B) statistics for low-peaking microwave channels have benefited from significant improvements to the FASTEM microwave sea surface emissivity module in the CRTM. Figure 1 demonstrates how both the position of the peak of the histogram and the dependence on wind speed of the O-Bs for AMSU-A channel 2 are much improved on going from FASTEM-1 (the previous default version in the GSI) through FASTEM-4 to FASTEM-5.
Figure 1. The improvement in the innovation statistics for AMSU-A channel 2 on changing the CRTM microwave sea-surface emissivity algorithm from FASTEM-1, through FASTEM-4 to FASTEM-5.
Further improvements to the bias characteristics of this and other satellite measurements are coming from improvements to the radiance bias correction scheme (Zhu et al., 2014). This scheme removes the requirement for the previous two-step bias correction scheme (where the scan-dependent component is calculated outside of the variational framework) by including extra predictors to describe this through a fourthorder polynomial of scan angle. In addition, the pre-conditioning for the bias-correction coefficients is now prescribed based on the Hessian rather than through pre-specified parameters (which greatly improves convergence); new bias correction predictors have been introduced to handle large land-sea differences and for SSMIS; automatic initialization of new data is performed; and improved handling of data that go missing and then recover is introduced (see Figure 2).
Figure 2. An illustration of how a temporary loss of data results in an increase in the error variance used to calculate the bias correction coefficients once the data returnâ€”allowing the bias correction scheme to react to possible changes in instrument characteristics.
The above two changes plus a fix to a bug in the way AMSU-A radiances are used around the ice edge have resulted in a much improved analysis in the Southern Hemisphere with significant improvement in forecast skill.
This upgrade will also include the assimilation of the sounding channels on SSMIS which were previously not possible to assimilate because of large biases in the data, which were a function of the position of the satellite in its orbit but which also varied with season. This was addressed by adding two new SSMIS-specific bias correction predictors to the bias-correction scheme:node x cos(latitude) and sin(latitude). Here node is +1 for the ascending part of the orbit and -1 for the descending part. The effectiveness of the scheme is illustrated in Figure 3. The inclusion of SSMIS (initially examined as part of the gap mitigation strategy) results in a small but positive impact in the Southern Hemisphere.
Figure 3. The spatial coverage and histogram of innovation for F18 SSMIS channel 4 before (top) and after (bottom) bias correction illustrating the removal of the intra-orbit (descending and ascending) biases in these data.
Other upgrades to the system include replacing GOES 6-hourly winds with hourly winds plus quality control changes; improvements to the quality control of GPSRO observations in the lower atmosphere where the refractivity has high vertical gradients; turning on the assimilation of IASI on Metop-B; and adjustments to the ATMS observation errors.
Andrew Collard (IMSG @ NOAA/NCEP/EMC), Daryl Kleist (University of Maryland), John Derber and Russ Treadon (NOAA/NCEP/EMC), Lidia Cucurull (NOAA/ESRL), David Groff, Emily Liu, Xiujuan Su, Paul van Delst, and Yanqiu Zhu (IMSG @ NOAA/NCEP/EMC), Quanhua Liu (NOAA/NESDIS/STAR)
Zhu, Y., J. Derber, A. Collard, D. Dee, R. Treadon, G. Gayno, G. and J. A. Jung (2014.), Enhanced radiance bias correction in the National Centers for Environmental Prediction's Gridpoint Statistical Interpolation data assimilation system. Q.J.R. Meteorol. Soc., 140, 1479-1492.
Zhu, Y., J. Derber, A. Collard, D. Dee, R. Treadon, G. Gayno, J. Jung, D. Groff, Q. Liu, P. van Delst, E. Liu, D. Kleist, 2014. Variational bias correction in the NCEP's data assimilation system. The 19th International TOVS Study Conference (ITSC-19), Jeju Island, South Korea. ( http://cimss.ssec.wisc.edu/itwg/itsc/itsc19/program/papers/10_02_zhu.pdf)