TJ5.3 GFS Model Dropouts and Improving Satellite Observation Quality Control

Tuesday, 24 January 2017: 9:00 AM
607 (Washington State Convention Center )
Jordan C. Alpert, NOAA/NWS/NCEP, College Park, MD; and V. K. Kumar, A. Eichmann, and S. A. Boukabara

Poor forecasts or skill score “Dropouts” occur in the National Weather Service (NWS) global forecast system (GFS) when other national center forecasts, for example, the European Centre for Medium-range Weather Forecasts (ECMWF), often do not exhibit a similar loss in skill. NCEP’s current operational GFS model is a spectral T1534 (13 km) and includes a Grided Satistical Interpolation (GSI) analysis for initializing the deterministic and other operational global and regional models. Recent upgrades have reduced NH and SH dropouts so they occur with less frequency and severity. The occurance of GFS Northern Hemisphere (NH) and Southern Hemisphere (SH) forecast skill, (percent 5-day Anomaly Correlations smaller than 0.7), dropouts shows a steady reduction from ~10-20 % per year over the NH and ~ 30-40 % per year over the SH during 1996-2001 to about 2 – 5 % dropouts per year since 2007*.

In the past, an attempt to quantify the skill differences when there are poor GFS forecasts compared to other national centers, eg., ECMWF, was studied with the focus on quality control (QC) of conventional observations. Global and regional area(s) were defined at initial condition (IC) time that had an impact on improving GFS 5-day forecasts when ECMWF analysis information was substituted. QC issues for conventional observations were investigated as the cause of these so called “dropouts” in forecast skill but no direct cause could be found for the low skill event cases, however, there was some evidence that non-conventional observations could influence the 5-day forecast outcome in a few cases.

Coordination efforts between JCSDA and NCEP/EMC Model Evaluation Group (MEG) are underway to evaluate the daily performance of NCEP’s operational T1534 Global Forecast System (GFS) forecast and analysis, identify model biases and conduct post-mortem studies of high-impact poor forecast events. The Global Forecast Dropout Prediction Tool project (GFDPT) with a goal to detect, analyze and improve QC by developing a monitoring system to analyze differences between the NCEP and ECMWF global models operationally and determine if the “dropouts” originate from QC problems in the assimilation especially the assimilation of radiances. Using dropout cases to magnify differences in the forecasts between the two centers, for a particular satellite platform, one can compare the co-located satellite radiance instrument observations to that derived using the community radiative transfer model (CRTM) from the GFS production analysis at the same time. That is, a satellite radiance comparison of the analysis’s derived radiance observations compared to the observed radiance, in a perfect world, would be the same and differences represent retrieval and assimilation errors.

We will provide examples of the prediction and diagnosis components of the GFPDT applied to a few recent 2015 and 2016 GFS forecast dropouts. We will highlight radiance observation and analysis differences (O-A) in regions where there are large analysis differences between GFS and ECMWF that cause poor 5-day forecasts in the GFS. The method used to find such regions and their influence on 5-day forecasts will be discussed as well as radiance observation QC issues that could cause the differences between ECMWF and GFS assimilation and dropouts.


1) NOAA/NESDIS/STAR 2) NOAA/NCEP/EMC 3) Joint Center for Satellite Data Assimilation (JCSDA)

4) Riverside Technology, Inc

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner