A set of observation system experiments (OSEs) are conducted over three seasons using the hourly-updated Rapid Refresh (RAP) numerical weather prediction model to demonstrate the importance of various observation types for 3-hr to 12-hr RAP forecasts. These experiments selectively remove observations from assimilation to quantify the increase in forecast error across a variety of metrics. Removal of aircraft observations yield the largest increase in forecast error for wind, relative humidity, and temperature forecasts averaged over the 1000-100 hPa layer, but many other observation types also have significant positive impacts, when assimilated, including satellite, surface, GPS-precipitable water, radar, and rawinsonde observations. Seasonal differences in the error characteristics exist, with impacts on relative humidity forecasts strongest during the spring which is characterized by strong forcing and sharp moisture gradients. Convective processes limit mesoscale observation impacts in the summertime. Retention of assimilated radar data will also be highlighted where impacts tend to persist for relatively short time periods of less than six hours.
Finally, we will demonstrate the impact of modifications to assimilation methodologies of surface observations including cloud information from ceilometers. In total, these results provide insight into the relative importance of different components of the observing system as we move towards a global rapid refresh capability and provide additional motivation for mesoscale ensemble data assimilation.