next up gif contents
Next: Noise in the Up: Error Analysis Previous: Error Analysis
Return to the Publications. Return to the Index.

Random Cross Correlation Function Regions

 

A problem of the wind profiling method is the possibility that the selected maximum correlation region results from random noise correlations instead of aerosol structure correlations. Normally, a cross correlation function between two CAPPI scans is dominated by a single peak. Sometimes, however, random noise fluctuations and a weak aerosol structure coherence between subsequent CAPPI scans lead to random correlation regions, which are stronger than the aerosol correlation peak. Then the wind estimate may be calculated from the strongest random correlation region, which causes a random result. This prompts a need to identify the spurious correlation peaks.

An earlier study[44] indicated the intensity of wind estimate fluctuations from the hourly mean values correlated with the CCF maximum strength. In this study, an objective method is developed to distinguish between reliable and unreliable wind estimates by indicating the probability that the chosen maximum correlation region represents the mean aerosol movements. Probability that the strongest region of the CCF represents the mean wind is given by

 

where is the mass of the global maximum correlation region and is the mass of the correlation region i. The masses are calculated by summing the intensities around the peak with amplitudes higher than of the global maximum correlation amplitude.

If there is only one correlation region, then is 1. If there are numerous similar correlation regions, then tends to zero.

Figure 29 shows scatter plots for vs. wind speed and direction for quarter-hourly averaged results on July 27, 1989. Most of the results have a and are within the natural wind deviation of that day; the results with are more widely spread. In general, a provides good results, while a < 0.5 is poor.

  
Figure 29: vs. speed (left) and direction (right) for quarter-hourly averaged wind estimates on July 27, 1989, from 8:00 to 15:00 CDT. All measurements with are within natural deviation of the wind; the most widely fluctuating wind results have .

Figure 30 shows the fraction of reliable VIL wind estimates as a function of averaging time interval and . Averaging the cross correlation functions effectively increases the fraction of reliable results. When compared with the individual measurements without time-averaging, even a fifteen-minute averaging more than doubles the ratio of the good measurements to all measurements. About 76% of the hourly averaged wind estimates in the boundary layer are calculated from a unique correlation maximum region. Note that most of the measurements with between 0.9 and 1.0 are consistent.

  
Figure 30: Fraction of reliable results as a function of averaging time interval and from July 26 to August 11, 1989. Open circles show fraction of results with and altitudes ranging from 50 m to 2000 m; black rectangles show measurements in the convective boundary layer with ; open diamonds show measurements in the convective boundary layer with . Averaging the CCF increases the fraction of reliable results by reducing random correlations.



next up gif contents
Next: Noise in the Up: Error Analysis Previous: Error Analysis
Return to the Publications. Return to the Index.



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