Abstract
The feasibility of running a limited-area numerical weather prediction model with initial conditions from a network of profilers is investigated. The observing system, a combination of microwave radiometer and UHF/VHF Doppler radar, provides low-resolution vertical temperature profiles and highly accurate wind measurements. The initial fields are created by a 12-hour model run. The high-resolution wind field is represented by the model output, simulating a network with profilers located at every grid point of the model domain. The model-created temperature field has to be smoothed vertically and horizontally to match the accuracy of the radiometer observations. A variational analysis technique is applied to infer the missing frontal-scale thermal structure in the three-dimensional temperature field from the more accurate wind observations. A double twin experiment is carried out, using an isentropic primitive equation model. First, the model is initialized with the smooth temperature field and the model-created wind field. Second, the model is initialized with the variationally analyzed mass field and the associated wind field. The two forecasts are compared to a control forecast.Four different synoptic cases are chosen to investigate the impact of smooth and upgraded mass fields on the numerical weather forecast: a cold air outbreak in January 1985, two springtime cyclogenesis events which took place in March 1983 and in April 1982, and one severe weather event which occurred in June 1983.