Accurate estimation of mixing height is important, since it is an important parameter for lower atmospheric studies involving aerosol monitoring and pollutant dispersal models. Sodar happens to be one of the best instruments for monitoring the mixing height. But it suffers from the drawback of acoustic noise, which makes the measurement inaccurate. Conventional Kalman filter has been used to estimate atmospheric boundary layer by filtering the measurement noise involved in sodar data. But there are certain limitations of the accuracy available from conventional Kalman filter which may be overcome by proper adaptive design. The present work develops an adaptive scheme for estimation of mixing heights. It considers the selection of a proper meteorological system model from a bank of system models. Diurnal and seasonal changes in measurement noise statistics of the acoustic radar is taken into account by designing a fuzzy logic based adaptive scheme.