Variational Assimilation of Precipitable Water Using a Nonhydrostatic Mesoscale Adjoint Model. Part I: Moisture Retrieval and Sensitivity Experiments

Abstract

Recently it has been proposed that the phase delay associated with the radio signals propagating from GPS satellites to a ground-based GPS receiving station can be used to infer the vertically integrated water vapor (precipitable water?PW) with a high degree of accuracy. Since a ground-based GPS receiving station is relatively inexpensive, a specially designed, dense GPS network can provide PW measurements with unprecedented coverage. Such a data set can potentially have a significant impact on operational numerical weather prediction. In this paper, a series of numerical experiments were conducted using a variational (4DVAR) data assimilation system based on The Pennsylvania State University ?National Center for Atmospheric Research mesoscale model MM5 and its adjoint. The special soundings collected in SESAME (Severe Environmental Storms and Mesoscale Experiment) 1979 wore used in two sets of experiments. In the first set, a 1-h assimilation window and an analysis of the observed PW data were used. All data were assumed to be available at the end of the assimilation window. The assimilation of PW data was found to effectively recover the vertical structure of water vapor and improve the quality of moisture analysis. The use of surface humidity data in addition to PW analysis resulted in further improvement in the quality of the retrieved moisture fields, particularly in the lower troposphere. The assimilation of PW and surface humidity data reduced the rms errors in the initial moisture analysis by as much as 40%. Such improvement cannot be achieved by assimilation of wind and temperature data, because they do not carry sufficient information on the moisture field. The authors also found that the assimilation of PW and surface humidity data can lead to significant improvement in short-range precipitation forecasts when used along with the wind and temperature data. The use of PW and surface humidity data in 4DVAR increased the threat score from 0.01 to 0.48 for 3-h forecasts and from 0.43 to 0.65 for 6-h forecasts. SESAME 1979 is a case with intensive convective activity, and the forecast is strongly affected by moist diabatic processes. The intent here is to test the impact of including adjoints of moist physics (adjoints of the Kuo cumulus convective scheme and the grid-resolvable precipitation) in the 4DVAR system to PW assimilation results during the initial stage of the storm case. Thus, the assimilation window is extended from 1 to 3 h, and it is assumed that PW data were available at an interval of 3 h in the second set of experiments. The PW data assimilated are generated by the model simulation. It was found that the inclusion of moist physics in the 4DVAR system reduced the systematic biases of the model, allowed a better fit between the model and observed data, and resulted in an improved ?optimal? initial condition and, consequently, a better short-range prediction. The threat score was increased from 0.30 to 0.50 in the 6-h forecasts following the assimilation cycle. These results suggest that the effects of physical parameterization should be included in a 4DVAR data assimilation system, especially for a situation with significant precipitation over a relatively long assimilation window (greater than 3 h). The sensitivity of the 4DVAR results to the initial guess field was also tested. The results of 4DVAR were found to be relatively insensitive to the quality of the initial condition (the guess field). Even with a very poor initial moisture field, 4DVAR was able to produce a high quality moisture analysis after the PW data were assimilated, although the number of iterations required had to be increased from 30 to 50.