Impacts of GPS-derived Water Vapor and Radial Wind Measured by Doppler Radar on Numerical Prediction of Precipitation

Abstract

This study conducted data assimilation experiments using the operational mesoscale four-dimensional variational data assimilation (4D-Var DA) system of the Japan Meteorological Agency (JMA). Experiments investigated the impacts of GPS-derived water vapor and Doppler radar-derived radial wind (RW) on precipitation prediction for a heavy rain event on 21 July 1999. RW data were obtained from Doppler radars at Narita and Haneda airports. GPS data were obtained from the GPS Earth Observation Network (GEONET) of the Geographical Survey Institute (GSI). Both precipitable water vapor (PWV) and slant water vapor (SWV), which is the amount of water vapor integrated along the slant path between GPS receivers and GPS satellites, were derived from the GPS data. Because SWV contains three-dimensional water vapor distribution information (Seko et al. 2000), we anticipated that assimilating SWV data would more accurately reproduce the moist air inflow at lower layers. Comparisons between observed and model-predicted precipitation regions helped define the impacts of assimilating RW, PWV, and SWV into the model. If the assimilated data included only conventional meteorological data, the model yielded small precipitation regions over a mountainous area far from Tokyo. If the assimilated data included both GPS-derived water vapor data and conventional data, lowlevel inflow in the model was more humid and precipitation occurred along the low-level convergence zone. Because the predicted position of the convergence zone differed from observations, however, the position of the precipitation region was not reproduced correctly. When RW and conventional data were assimilated into the model, low-level northerly flow was reproduced in the northwest of Tokyo. This northerly flow intensified the low-level convergence where precipitation was observed, and the position of the forecasted precipitation was more similar to that of observations. In this model run, however, lowlevel inflow from the south was less humid than observed, and precipitation onset was delayed by 1 hour. If GPS-derived water vapor data, RW data, and conventional data were all simultaneously assimilated, the precipitation position was modeled correctly, and precipitation onset occurred as observed. Comparisons between the vertical cross sections of analyzed water vapor fields and first-guess water vapor fields helped measure the impact of data assimilation on the modeled water vapor distribution. When PWV, RW, and the conventional data were assimilated, water vapor on the windward side of the low-level inflow decreased. In contrast, water vapor in the low-level inflow did not decrease when SWV data were used, instead of PWV data. For this rainfall event, the assimilation of RW and GPS-derived water vapor data improved the precipitation prediction. Assimilation of SWV data improved the representation of the vertical water vapor distribution.