Global Forecast Dropout Prediction Tool (GFDPT) Project: Status and Preliminary Results

The enhancements in Global Forecast System (GFS) model resolution, parameterization schemes, and  Gridpoint Statistical Interpolation (GSI),  particularly the hybrid-ensemble variational data assimilation system, have reduced the occurrence and severity of GFS Northern Hemisphere (NH) and Southern Hemisphere (SH) forecast skill dropouts (percent 5-day Anomaly Correlations smaller than 0.7). This project provides evidence as to what causes GFS dropouts which continue to affect the GFS performance statistics compared to other international centers, eg., ECMWF, UKMO, CMC models and how they can be alleviated.

 Important goals of the GFDPT project are to develop a monitoring system to analyze differences between GFS and other models, identify quality control (QC) issues of conventional and satellite observations and make corrections in the subsequent forecast cycle.  Various components of the GFDPT system are prediction and detection of actionable volumes of conventional and satellite observations that cause the dropouts. The forecast-forecast correlations between ECMWF and GFS provide first a warning of GFS dropout potential, an extremes code sifts out and displays extreme volumetric integrals of squared differences of geophysical fields of GFS analysis differences compared with ECMWF and other background guess fields and finally a full diagnosis of actionable volumes are performed using the GSI diagnostics files and or an independent community observation assessment tool (COAT) consisting of passive microwave, broadband and hyper spectral IR radiance measurements and atmospheric motion vectors. The source region in an initial condition that caused a poor GFS forecast can be determined from forecast error information. Post-mortem studies are carried out for GFS dropout cases using ECMWF initial conditions for GFS forecast – “ECM” cycled runs to investigate whether dropouts originate due to model problems or assimilation issues. We show preliminary evidence that contaminated radiances can potentially cause poor skill GFS forecast when they are assimilated in a region that is sensitive to initial conditions in terms of the 5 day forecast error.  We illustrate examples from the GFPDT system applied to a few recent 2015 and 2016 GFS forecast dropouts focusing on satellite MW and IR radiance observations 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.