This section presents a study of the vertical profiles of horizontal mean winds calculated from data obtained with the University of Wisconsin Volume Imaging Lidar during the 1987 and 1989 FIFE programs. Wind profiles are determined from aerosol structure movements in volume scans of the convective boundary layer. An objective technique for identifying unreliable measurements and a method for estimating errors due to random noise in data are developed. The VIL wind profiles are then compared with radiosonde, aircraft-based, and ground-based measurements.
Mesoscale wind profiles are needed in a wide variety of meteorological applications. Wind profiles are traditionally measured using methods that provide wind speed and direction at certain points. However, point measurements in the convective boundary layer do not generally provide representative area-averaged wind estimates. Point measurements often include contributions from large eddies, roll vortices, and topographic influences, which causes them to differ from the area-averaged value. The averaging time often cannot be extended to dampen temporal fluctuations without losing some information about the changing value of the mean wind. Therefore, direct area-averaged measurements are required.
Several methods are used to remotely measure area-averaged winds of the atmosphere. They can be measured by determining cloud layer movements from satellite images . However, altitudes of cloud layers are difficult to obtain if concurrent ground based measurements are not available, and the technique works only at the altitude of the cloud layer. A Doppler radar measures mean winds in large volumes of the atmosphere when large scatterers (water droplets, insects, dust, etc.) are present . A Doppler lidar extends the principle of the Doppler radar using shorter wavelengths so smaller aerosols can be used to trace the wind . This enables Doppler lidar wind measurements in the CBL in a wide variety of atmospheric conditions. The technique can be used to reveal mean wind profiles using azimuthal scans with a constant elevation angle. The inaccuracy of the method is of the same order of magnitude as for the boundary layer turbulence. For example, Eberhard et al.  have reported about 0.7 ms RMS-errors in radial wind measurements for a single Doppler lidar profile.
The next generation of Doppler lidars  is expected to improve the accuracy to better than 0.1 ms.
Another lidar technique to measure winds is to trace inhomogeneities of the aerosol field from backscatter profiles. Eloranta et al.  and Sroga et al.  employed a cross correlation technique to determine the aerosol movements between subsequent lidar backscatter profiles measured with slightly different observation angles. Sasano et al.  adopted a correlation method used for interpretation of satellite images  and applied it to a scanning lidar data to estimate two-dimensional, area-averaged wind vectors. The three-dimensional aerosol mapping capability of the Volume Imaging Lidar allowed Schols and Eloranta  to determine vertical profiles of horizontal area-averaged winds. The wind profiling method was based on determining the maximum correlation between subsequent CAPPI scans, which are calculated from boundary layer volume scans of the VIL.
The wind profiling algorithm introduced by Schols and Eloranta  is refined in this study (see section 4.1). A method is developed to estimate the root-mean-square (RMS) errors when determining the wind speed and direction (see section 4.2.2). If turbulent mixing and high signal noise substantially change aerosol structures between CAPPI scans, the strength of the aerosol structure correlations may decrease to the level of random noise causing uncertain wind estimates with large errors. This study incorporates an objective method to determine the probability that the chosen correlation peak is caused by aerosol movements between two subsequent CAPPI scans instead of random signal fluctuations (see section 4.2.1). This analysis is employed in the accuracy estimations to reject results that may contain large errors. Wind estimates which have passed the previous analysis are then compared with measurements made with an aircraft-based instrument, radiosondes, and automatic weather stations (see section 4.3). Even if these traditional measurements had no noise, they may differ substantially from the area-averaged winds due to bias by large-scale eddies and roll vortices. Therefore, the error estimation method is tested by comparing the errors with the internal consistency of the VIL results. The FIFE program produced a large number of VIL wind profiles which are stored in the FIFE database . A summary of the VIL wind results is presented in section 4.3.4.