Doppler wind lidars are capable of measuring multi-scale turbulent flows in the atmospheric boundary layer. With proper scanning geometries and wind retrieval algorithms, lidar measurements provide robust estimates of time-evolving threedimensional mean flow and turbulence fields. This dissertation focuses on uncertainty quantification and reduction of (i) mean wind velocity, (ii) turbulence statistics and (iii) wind turbine wake characteristics derived from lidar measurements. To reduce the uncertainty in mean wind velocity measurements from lidar arc scans, a model is developed and verified for uncertainty prediction. Both observations and model predictions show that the uncertainty is proportional to turbulence intensity and is reduced by aligning arc scans with wind directions and using large arc spans with large azimuth intervals. The limitations of lidar turbulence measurements are evaluated through a field experiment and statistical simulations, both of which show that the volumetric averaging of lidar measurements inflates the autocorrelation of lidar radial velocity and consequently causes large errors in the estimated radial velocity variance. It is further shown that, given atmospheric flows are commonly stationary for no more than one hour, the systematic error is negligible but the random error has about 8% uncertainty, imposing a limit on the accuracy of current lidar turbulence measurements. Uncertainties in lidar wake characterization are evaluated through a field experiment. A wake detection algorithm is developed to retrieve wake characteristics from small sector scans. Because the wake moves as the wind turbine yaws, lidar measurement locations relative to the wake change with wind direction. This change introduces large uncertainties in the retrieved wake characteristics, especially when vertical wind shear is large or wind directions are nearly orthogonal to the laser beams. By focusing on uncertainty reduction in three aspects of lidar applications of wind measurements, this dissertation demonstrates that with careful scan design and data processing and by accounting for site conditions, it is possible to model, quantify and minimize both the systematic and random errors in lidar-derived wind velocities.