next up gif contents
Next: Summary of Wind Up: Results and Discussion Previous: Comparisons with Flight-Path
Return to the Publications. Return to the Index.

Accuracy Analysis

Turbulence causes large variations in point and line-averaged wind measurements. These variations often dominate the difference between traditional measurements and the VIL observations. Therefore, the point and line-averaged measurements are not referenced when analyzing the accuracy of the wind profiling method. Calculated error estimates of the VIL wind profiles are compared with the local consistency of the measurements. The local consistency is calculated for each wind measurement by subtracting the average of its vertically adjacent wind measurements. If the vertical wind shear is linear, the variations between adjacent vertical values presents the measurement inaccuracy.

The Cartesian mapping procedure introduced in Section 4.1.2 may cause profile smoothing by mixing the aerosols from different layers. Therefore, the Cartesian mapping routine is slightly modified for the consistency analysis: each backscatter signal element is averaged into the closest grid point instead of dividing it into the eight closest grid points. Also, the filling of the empty pixels is performed only in xy-plane. This eliminates the vertical mixing of the CAPPI scans due to the mapping algorithm. Data from July 26, 1989, is selected (see Figure 35) for accuracy calculations, since the wind shear is more linear on that day than on any other day in the FIFE program. The same case was also analyzed with the original cartesian mapper. The maximum difference between the wind results calculated from differently mapped CAPPI scans is about 0.1 ms in speed and 0.5 in direction. The RMS errors were smaller for the original Cartesian mapper, indicating the noise in the CCF planes was also smaller because of more efficient averaging.

  
Figure 35: Hourly averaged wind profiles on July 26, 1989, between 11:00 and 16:00 CDT. Only results with = 1 are plotted. The mixed layer mean depth rises rapidly to 1000 m before 11:00 CDT when the clouds start forming. The base of convective clouds varies between 1000 m and 1800 m during the measurement period. The wind speed shear is linear except in the first 200-300 meters and at the top of the CBL. Error bars show estimated root-mean-square errors.

Figure 36 compares the local deviations and calculated RMS error estimates for hourly averaged wind profiles on July 26, 1989. The local deviations are calculated from the averages of vertically adjacent results. The calculated error estimates represent RMS errors in wind determination from noise in the CCF

after the wind estimates with are rejected. The calculated RMS error estimates are consistent with the internal consistency of the results.

In the convective boundary layer (below 1400 m), the local deviations are less than 0.1 ms in speed and 1 in direction. Part of the local variation is probably due to real wind fluctuations. Thus, it may provide a conservative estimate of the accurasy. The calculated profiles show minimum RMS errors at about 1 km: the contrast between scattering from aerosol structures and mixed air is very good just below the CBL top, since the clear air penetrating down to the CBL is not yet mixed with the boundary layer aerosols. Closer to the ground, more intense mixing smears the aerosol structures, decreasing the signal-to-noise ratio of the CCF and flattening the CCF peak. Above the convective boundary layer, clouds block the signal decreasing the signal-to-noise ratio.

  
Figure 36: Local deviations of hourly averaged wind speed and direction (open circles) and calculated RMS error estimates (black circles) on July 26, 1989, between 11:00 and 16:00 CDT. The error bars indicate the minimum and maximum error of hourly averaged wind estimates during the 5-hour observation session. The calculated error estimates are consistent with the local deviations.

Figure 37 shows calculated RMS error estimates of wind measurements as functions of the averaged time and normalized altitude from July 26 to July 31, 1989. Error estimates are first linearly interpolated into altitudes normalized by the convective boundary layer mean depths and then averaged over time. In the CBL, the error estimates are almost constant with altitude; above the CBL, they grow almost exponentially. In the convective boundary layer, the RMS errors are on the average less than 0.2 ms in speed and 3.0 in direction. In the middle of the convective boundary layer, where aerosol structures have good contrast against the background, the RMS errors are 0.03--0.05 ms in speed and 0.3--0.8 in direction. Above the CBL, the RMS errors vary between 0.1--3.0 ms in speed and 1.0--30 in direction, since the correlations get poorer due to aerosol structures diminishing with altitude and clouds blocking the signal. The RMS errors decrease as the averaging time interval increases, since the time averaging of the cross correlation functions reduces random correlations.

  
Figure 37: Root-mean-square error estimate profiles of reliable results as functions of averaging time interval and normalized altitude from July 26 to July 31, 1989. The altitude is normalized with the convective boundary layer mean depth. Note the logarithmic horizontal axes. The profiles show RMS errors of wind measurements with = 1 averaged over a 5-hour period. Errors of 3-minute (open circles) and hourly averaged (black circles) profiles are shown. The RMS errors decrease as the averaging time interval increases. The minimum errors are obtained in the middle of the convective boundary layer. Below it, turbulent mixing smears the aerosol structures and decreases the signal-to-noise ratio of the CCF. Above it, the clouds block part of the data and decrease the signal to noise ratio.



next up gif contents
Next: Summary of Wind Up: Results and Discussion Previous: Comparisons with Flight-Path
Return to the Publications. Return to the Index.



Antti Piironen
Tue Mar 26 20:53:05 CST 1996