Evaluation of National Centers for Environmental Prediction Global Forecast System at the Atmospheric Radiation Measurement Program Southern Great Plains Site

Since 2001 output from the National Centers for Environmental Prediction global weather forecast system (GFS) has been routinely processed to produce single column profiles at locations corresponding to the Atmospheric Radiation Measurement (ARM) Program sites. In the present study, GFS forecast was examined and compared with ARM observations at the Southern Great Plain Central Facility for the years from 2001 through 2004. The comparison was made for the first 48 hours of forecast and primarily for surface energy fluxes and clouds. For most of the variables examined, the forecast compared well with the observations. The performance of the model has been improving along with model upgrades. For certain variables, persistent biases exist. For instance, the model overestimated latent heat flux by 80 W/m over the 1-3 PM time period of the day and, coincidently, overestimated surface downward solar flux by 43 W/m, and underestimated sensible heat flux by 48 W/m. As a result, the surface net heat budget in the forecast was balanced due to cancellation of errors. For cloud, the model was able to capture its observed temporal evolution and vertical distribution during major synoptic events. However, on average, the model largely underestimated cloud fraction in the lower and middle troposphere, and slightly overestimated in the upper troposphere. The forecast missed daytime non-precipitating low clouds seen in the observations. Diurnal cycle of clouds in the middle to lower troposphere was much weaker in the forecast than in the observation. The bias in cloud diurnal cycle was found to be responsible for a phase shift in the diurnal cycle of the forecast cloudy-sky surface downward longwave flux (SDLW). Further assessment was made for time before and after an upgrade of the GFS longwave radiative transfer module. After the upgrade, a systematic low bias of cloudy-sky SDLW found in the earlier forecast was eliminated; however, clear-sky SDLW at night became too large. The latter was probably caused by a nighttime warm bias in the lower and middle atmosphere. This investigation demonstrated the unique value of ARM observations in diagnosing and understanding forecast model errors, and for improving the forecast model physics.