A new 3DVAR-based method is developed for retrieving three-dimensional moisture field from slant-path data from a hypothetical high-resolution ground-based GPS network. Compared to limited existing work using 3DVAR-based methods, our system considers flow-dependent background error covariances and models such covariances using a spatial filter. The method is tested using a set of OSSE (Observation Simulation Systems Experiment) experiments for which ground truth exists. The OSSE experiments also allow us to examine the effectiveness of high resolution networks not existing today.

The results show that our system is able to adequately retrieve the three-dimensional water vapor field even when the GPS slant-path water vapor observation is the only source of observations and when the number of observation data is less than ten percent of the number of grid points or retrieved values. Because the background error covariance is explicitly considered, we can know how and how much the observation information is spread in space. Furthermore, flow-dependent background error covariance improves the retrieval by taking advantage of known flow structures in the background and/or intermediate analysis. Statistic correlations between the analysis increment field and the truth-minus-background field are found to be more than 0.6. The use of flow-dependent background error covariances was particularly helpful to the analysis at the low-levels where the retrieval is most difficult due to lack of overlapping paths. The impact of additional data from direct surface observations is also examined.