Publicidad

The identification of second trip echoes within the COPE dataset has shown that they have distinct signatures when observed using normalised coherent power, especially when averaged over a moving window, and median azimuthal phase shift difference. These sig- natures have been used to remove second trip echoes during processing, removing the effects previously seen in Figure 4.3 and Figure 4.5, using a logical decision process. Po- tential options for the logical decision process were based on those used previously in radar filtering and included threshold filtering, a decision tree and fuzzy logic (Chan- drasekar et al., 2013). As the filter only has to identify and remove one signal, based on at most two radar moments a simple threshold approach was taken, akin to a very basic decision tree. The distribution of both mean SQI and azimuthal phase shift change were re-examined in a two dimensional distribution space (Figure 5.10) to determine the relationship between echo type, mean SQI and phase difference. Two possible filtering approaches were developed based on these distributions. Option 1 is a simple threshold filter, based only on the mean SQI in the 25 gate moving window, with any observations with a mean SQI below 0.3 being removed as a second trip echo. Option 2 applies a mean SQI threshold which varies as a function of the median azimuthal phase difference, as shown in equation 5.4, whereδΦ is the median azimuthal phase difference and SQIthr is the variable threshold.

SQIthr(δΦ)                0.2 + 0.02δΦ, if 0₆δΦ₆10 0.4, ifδΦ>10 0.3, whereδΦ is missing (5.4)

Figure 5.10: Two moment heat maps, indicating the number of observations identified that fall within the given hexagonal bins. Identified rainfall echoes are indicated with the purple colour scaling, while second trip echoes are shown with the red colour scale. A hexagonal bin is only coloured if it contains at least 20 observations, which is less than 0.01% of the total observations in each category. The observations are from the same

events shown in Figure 5.1 for precipitation and Figure 5.7 for second trip echoes.

To assess the merits of each filtering approach the 111 volume scans taken on the 17 Aug 2013 were processed using both filters. The 17 August was chosen as it has a significant number of second trip echoes, located across a range of azimuths at varying intensities. Figure 5.11 shows the impact of applying the filter to the number of echoes observed as a percentage of the total number of scans at 0.5◦elevation. The raw data shows the alternating pattern also seen in Figure 4.5, where the second trip echoes are visible in every other radial due to the staggered PRF. Application of the simple threshold filter removes this affect, suggesting it is successfully removing second trip echoes and the variable filter also successfully removes the affect. The difference between the two

Chapter 5. Quality control 71

filtering options is that the variable filter is more aggressive, removing more echoes as second trip signals in comparison to the simpler option, removing 1-2% more echoes on average. Analysis of the data using animated comparisons suggests this aggressive filtering is robust and indicated much better performance of the variable filter compared to the simple filter in turbulent conditions.

Figure 5.11: Azimuthal variation of mean echo occurrence within a radial on the 17

August 2013. Solid black line shows the echo occurrence for the raw reflectivity data, the blue line shows the impact of using a single value threshold filter of 0.3 and the green

line indicates the echo occurrence for the variable filter shown in equation 5.4.

Figure 5.12 shows an example of the two filtering options, compared to the raw data, from a single elevation scan (2013-08-17 09:48 UTC, 0.5◦elevation). In this example both filters are successfully removing the second trip echoes occurring to the north east of the radar, with the variable filter being more aggressive in its filtering of the interlaced first and second trip echoes at about 30 km range. The variable filter also removes some of the ground clutter signals resulting from Dartmoor due east of the radar, which is a primary reason for the differences in echo occurrence percentages between the two filters seen between 75◦and 120◦in Figure 5.11. The main difference between the two filtering options occurs to the north west of the radar, at around 40 km range, where the simple filter is removing a large region of reflectivity which shows no visible indication it is a second trip echo (using reflectivity and the other available radar moments). The variable filter does not remove this region as it has a lowδΦ, which results in a lower SQI_thrbeing applied in this region. Further investigation reveals this area to have unusually low SQI values for

first trip echoes, which are coincident with a zone of convergence indicated in the Doppler wind field and in the Doppler spectral width. These observations are consistent with the presence of a warm front indicated on UKMO analysis charts which passes through this area at this time. This signature and the associated removal of first trip echoes by the simple filter can be seen in the preceding and following volume scans. Dixon and Hubbert (2012) indicate this potential issue with SQI in turbulent areas, only suggesting that elevated spectral width may help in moderating the threshold used, although this is also susceptible to increase in areas of noise and second trip echoes, indicating the two moments are not mutually exclusive. It is noticeable that the introduction of δΦ to the filter (Fig. 5.12, panel C) greatly reduces this erroneous filtering, and strongly supports the use of the two moment variable filter for the removal of second trip echoes.

Figure 5.12: Error in simple second trip filter due to turbulent mixing as a result of frontal convergence. Data shown is the 0.5◦elevation angle reflectivity data from 2013- 08-17 09:48 UTC in raw form (A), after the application of the simple filter (B) and after application of the variable filter (C). Range rings shown at 25 km and 50 km from the

Chapter 5. Quality control 73