The operational observation of the atmospheric water vapor field for the initialization of numerical weather prediction models is usually restricted to the standard radiosonde network. This includes typically 2 to 4 ascents per day in a network, where the stations are about 50 km apart from each other. Furthermore, the horizontal drift is usually not considered during the assimilation adding to the errors introduced bye the coarse spatial and temporal resolution. These systematic errors in the initial field are, apart from deficiencies in the physical parameterizations used, the main reason for forecast errors in general, and errors in quantitative precipitation forecasts (QPF) in particular.
State of the art lidar systems provide high-resolution and high accuracy observations of key atmospheric variables like temperature, water vapour content, and wind, especially in the important pre-convective environment. The observed quantities are very close to what is used by models, so that only little expense is necessary to process the raw data for the use in the model system. However, such temporally high-resolution observations pose certain demands to the used assimilation system. Recent results in nowcasting and short-range weather forecasting demonstrated that variational assimilation techniques in combination with high-resolution modeling is essential for optimising quantitative precipitation forecast (QPF) on the mesoscale.
Besides our lidar assimilation activities we also work on the assimilation of GPS slant-path measurements. Although the resolution is not as high as that of lidar systems, the spatial coverage of the GPS observations is much larger and they are not affected by clouds. Thus it is very interesting to compare the impact of assimilating the data of these different instruments.