A four-dimensional variational data assimilation (4DVAR) algorithm, implementing the adjoint of a large eddy simulation (LES) has been developed and tested. The 4DVAR algorithm assimilates Doppler lidar radial velocity data from repeated volume scans and simultaneous temperature profiles from radiosondes in order to retrieve three-dimensional, time varying wind and temperature fields. The method works by finding the optimal initialization of the LES in which the Doppler lidar and radiosonde measurements function as weak constraints on the prognostic variables. The forward model of the 4DVAR algorithm uses standard flux-gradient relationships to relate SGS fluxes to resolved scale variables. The eddy diffusion and diffusivity coefficients are functions of the sub-grid scale (SGS) turbulent kinetic energy (TKE). As a simplifying assumption the SGS TKE is a prescribed function of height. Within the lowest 60 m AGL the SGS TKE profile is estimated from tower measurements. The 4DVAR algorithm is applied using data collected under convective conditions during the CASES-99 field experiment in south-central Kansas. This study presents results from a series of numerical experiments investigating the sensitivity of the retrieved wind and temperature fields to changes in the prescribed SGS TKE profile.