Low-altitude nearly horizontal radio propagation is strongly affected by ducts associated with the capping inversion of the atmospheric boundary layer especially over the oceans. The key variables that influence the ducting effects are the height, depth and intensity of the inversion. These refractivity variables are not primitive and cannot be well assimilated by the conventional 3DVAR due to the following undesirable features: (i) As a linear estimator, the conventional 3DVAR cannot correct phase errors in the vertical profiles of refractivity. Instead, it smears the vertical structure around an inversion. (ii) The conventional 3DVAR may produce spurious double inversion layers when the observed inversion height changes significantly between observations (sounding stations) or the background inversion height changes significantly away from an observation. (iii) Analysis is not performed directly for refractivity which is a nonlinear function of the primitive variables (pressure, temperature and humidity). To solve these problems, modifications are made to the conventional 3DVAR and to the conventional innovation vector analysis for estimating observation and forecast error covariances. The modified method is applied to the observational data collected from the Variability of Coastal Atmospheric Refractivity field experiment and background data produced by the Navy's Coupled Ocean/Atmospheric Mesoscale Prediction System. The results will be presented at the conference.