|Title||Analysis and forecast impact of the main humidity observing systems|
|Publication Type||Technical memorandum|
|Secondary Title||ECMWF Technical Memoranda|
|Authors||Andersson, E, Hólm, EV, Bauer, P, Beljaars, A, Kelly, GA, McNally, A, Simmons, A, Thépaut, J-N, Tompkins, AM|
|Place Published||Shinfield Park, Reading|
The global analysis and forecast impact of observed humidity has been assessed by means of observing system experiments with the ECMWF 4D-Var data assimilation system. It is found that humidity data have a significant impact extending into the medium range (5-6 day forecasts), with a marked impact also on the wind and temperature fields. This contradicts some previous studies that have shown insignificant impact of humidity observations in general. The current, greater benefit of the humidity analysis may be due to improved model and data assimilation methods, and vastly increased availability of atmospheric moisture observations. The results show that each tested data type provides benefit to the analysis and forecast performance, which indicates that the humidity analysis is effective in extracting information from a wide variety of humidity observations. Data from the microwave sounding instruments (SSMI and AMSUB) dominate the humidity analysis over sea, whereas radiosondes, surface stations (SYNOP) and AMSUB dominate over land. The infrared sounders (GEOS, HIRS and AIRS) dominate in the upper troposphere, at 200-300 hPa. The lack of absolutely calibrated humidity data makes dealing with biases in observations and model one of the main issues for determining the global moisture distribution and a balanced hydrological cycle. In these experiments, SSMI adds water in the sub-tropical subsidence areas due to a bias with respect to the model. In several locations over land, radiosondes and SYNOP have opposite bias impacts in the boundary layer, resulting in local influence on precipitation when either data set is withheld. The SYNOP data are biased wet and the radiosondes are biased dry with respect to the model.