Validation of precipitable water from ECMWF model analyses with GPS and radiosonde data during the MAP SOP

Abstract

Precipitable water vapour contents (PWCs) from European Centre for Medium-Range Weather Forecasts (ECMWF) analyses have been compared with observations from 21 ground-based Global Positioning System receiving stations (GPS) and 14 radiosonde stations (RS), covering central Europe, for the period of the Mesoscale Alpine Programme experiment special observing period (MAP SOP). Two model analyses are considered: one using only conventional data, serving as a control assimilation experiment, and one including additionally most of the non-operational MAP data. Overall, a dry bias of about −1 kg m−2 (−5.5% of total PWC), with a standard deviation of ∼2.6 kg m−2 (13% of total PWC), is diagnosed in both model analyses with respect to GPS. The bias at individual sites is quite variable: from −4 to ∼0 kg m−2. The largest differences are observed at stations located in mountainous areas and/or near the sea, which reveal differences in representativeness. Differences between the two model analyses, and between these analyses and GPS, are investigated in terms of usage and quality of RS data. Biases in RS data are found from comparisons with both model and GPS PWCs. They are confirmed from analysis feedback statistics available at ECMWF. An overall dry bias in RS PWC of 4.5% is found, compared to GPS. The detection of RS biases from comparisons both with the model and GPS indicates that data screening during assimilation was generally effective. However, some RS bias went into the model analyses. Inspection of the time evolution of PWC from the model analyses and GPS occasionally showed differences of up to 5–10 kg m−2. These were associated with severe weather events, with variations in the amount of RS data being assimilated, and with time lags in the PWCs from the two model analyses. Such large differences contribute strongly to the overall observed standard deviations. Good confidence in GPS PWC estimates is gained through this work, even during periods of heavy rain. These results support the future assimilation of GPS data, both for operational weather prediction and for mesoscale simulation studies.