Campbell Scientific PWS100 Present Weather Sensor - Manual (Page 92)

Section 8. Functional Description
8.6 Algorithm Description
8.6.1 Detecting and Classifying Precipitation
The PWS100 has a structured detection volume consisting of four sheets of
light each 0.4 mm in depth with 0.4 mm spacing. The area of detection is
approximately 40 cm
2 as defined by the overlap of the two detectors each of
which is 20 ° off of the light sheet propagation axis.
Detection of precipitation is carried out in real-time. Time and frequency
domain analysis of the signals from the two sensor heads is carried out to
ensure all particles are detected. Time domain analysis allows quick stripping
of larger particles while the frequency domain analysis picks out smaller
particles using a slightly more computationally intensive routine.
Using a series of thresholds and checks precipitation signals from particles as
small as 0.1 mm diameter passing through the area of detection are identified
while any sources of noise or particles not passing through the defined area are
discounted. A routine then strips the signal from the incoming data for each
real particle passing through the defined area in order to minimize the data that
is to be further processed.
When a particle passes through the four light sheets it provides a signal
characteristic of the particle type and provides information on the velocity of
the particle, by means of auto correlation, and size, by means of cross
correlation of the two channels of data. Particles for which cross correlation
could be poor (e.g., snow) may have their size measured by the analysis of
particle transit time or signal amplitude (in the same way that present weather
sensors without structured detection volumes measure particle size). Due to the
Gaussian nature of the light sheet intensity there is more error involved in these
transit time or signal amplitude measurements than those using the cross
correlation which is independent of the signal intensity.
The structure of the particle is to some degree given by the analysis of the
signal peak to signal pedestal ratio. The signal pedestal is found through
stripping higher frequencies from the particle signal (i.e., those produced by the
four light sheets) and is higher for particles with higher crystallinity and
therefore more scattering sites. Particles of water have almost no pedestal
compared to the peak signal value whereas snow can have large pedestals
compared to the peak signal value.
Analysis of separate snow and rain/drizzle events has shown that there is very
little overlap in the signal peak to pedestal peak ratio for the solid and liquid
particles, though drizzle values extend further towards the snow values and
snow grains extend further towards the liquid precipitation values, while the
larger rain and snowflakes remain almost entirely separate. This is shown
graphically in Figure 8-6 which represents two separate events one comprising
snowflakes and snow grains (particle count on the left vertical axis) and one
representing a mixed drizzle and rain event (particle count on the right vertical
axis). This particular differentiator, only obtainable because the PWS100 has a
structured detection volume, is extremely useful in the fuzzy logic processing
as described below and further in Section 8.6.4.1.
8-6