An Analysis of VAD Radar Wind Impact on the NCEP GFS Slow Speed Bias

The NCEP Global Forecast System (GFS) model background is shown to be too slow compared to both radiosonde and MDCRS aircraft winds in the CONUS area in January 2012, especially around 300 to 200 hPa where the bias is roughly 1 m/s. This model slow speed bias is also shown to become worse as the observed winds get stronger. Diagnostics from the European Centre for Medium-Range Weather Forecasts (ECMWF) shows that their background speed is much better. The Velocity Azimuth Doppler (VAD) radar winds used at NCEP indicate slow speed biases relative to the background and cause an average slowing of the analysis wind speed that also gets stronger at higher wind speeds. This paper addresses how much the VAD wind data are causing the slow speed bias of the background as well as different procedures to reduce the slow speed bias while attempting to improve the model forecast skill.

Since data assimilation and forecast experiments use considerable computer resources, statistics and speed scatter diagrams are used to suggest where and when the VAD winds have excessive slow bias.

A number of sensitivity experiments are performed using different processing or data quality control (QC) on the VAD winds in the NCEP Global Data Assimilation System (GDAS). These include using no VAD winds, a reject-list for select sites with significant slow speed biases, speed dependent QC, where observations with slower speeds relative to the background are subject to stricter QC, and use of raw level II radar winds from the site to be quality controlled before producing VAD profiles. After studying the speed biases and general fit of the background to observations, tests will be performed with the Global Forecast System (GFS) to measure impact on forecast skill.