
The following overview of version3 MST radar signal processing is
included in the global attribute 'comment' in netCDF files. The
sections are arranged into three groups:
Group 1: appears in spectral, radial and Cartesian data files
 Introduction
 Sampling Procedure
 From Samples to Spectra
Group 2: appears in radial and Cartesian data files
 Spectral Processing
 Time Continuity Testing
Group 3: appears in Cartesian data files
 Deriving Cartesian Wind Components
 The Accuracy
of Horizontal Wind Components
 Vertical Beam Data Products
 Beam Broadening Corrected
Spectral Width
 Aspect Sensitivity
Introduction
The Natural Environment Research Council (NERC) MesosphereStratosphere
Troposphere (MST) Radar at Aberystwyth (UK) is a 46.5 MHz
windprofiling instrument. It transmits short pulses of radio waves
which are scattered back to it from atmospheric targets. The distance
of a target from the instrument is determined by the time delay
between the transmission and reception of a pulse. The main targets
are metrescale refractive index irregularities, which are referred to
as clearair targets (the term does not necessarily imply clearsky
conditions as the radar is able to see through clouds). Hydrometeors,
aircraft, and groundbased objects can also give rise to radar
returns. The receiver signal is occasionally contaminated by
interference. The refractive index irregularities are caused by
variations in atmospheric humidity (within the lowest 10 km of the
atmosphere), in atmospheric density (within the lowest few 10s of km)
and in free electron density (above 50 km). The radar return signal
power is typically proportional to the square of the mean vertical
gradient of the (potential) refractive index and inversely
proportional to the square of the range of the clearair targets from
the radar. The refractive index irregularities are assumed to be
advected with the wind. Consequently the radial velocity, i.e. the
component of the wind vector along the radar beam pointing direction,
can be inferred from the Doppler shift of the radar return
signal. Atmospheric turbulence gives rise to variability in the radial
velocity when observed over a time scale of a few tens of
seconds. However, as a result of the radar's finite beam width, the
observed spread tends to be dominated by a beambroadening component,
which is proportional to the horizontal wind speed. Consequently it
becomes increasingly difficult to infer turbulence intensities as the
wind speed increases.
Sampling Procedure
The radar receiver signal is sampled at 1.0 us intervals following
the transmission of a pulse. This corresponds to sampling at 150.0 m
intervals in range from the radar. It is necessary to sample both the
inphase (I) and quadrature (Q) components of the receiver signal
(i.e. complex values) in order to allow both the magnitude and the
sign of the Doppler shift to be inferred. In the mmode samples are
recorded between ranges of approximately 60 and 90 km. In the stmode
samples are recorded between ranges of approximately 2 and 20 km. It
is possible to operate the radar in mstmode so that both altitude
ranges are sampled simultaneously. The range resolution of the radar
returns is determined by the length of the transmitter pulse (not by
the sampling interval), to which the receiver bandwidth must be
matched. The range resolution can be increased by using complementary
coding. This requires the phase of sublengths of the transmitter
pulse to be offset by either 0 or 180 degrees according to a set
coding pattern. The range resolution is then determined by the
sublength of the transmitter pulse (to which the receiver bandwidth
is matched).
From Samples to Spectra
No attempt is made to derive radar return signal parameters until
samples have been acquired for a large, predetermined number of
pulses  typically covering a few tens of seconds. The term dwell is
used to refer to this collection interval or to range profiles of any
of the data products associated with it. A dwell initially consists of
a complex time series, for each range gate, of I and Q samples which
are separated in time by the inter pulse period (of the order of a
millisecond). The nature of the samples changes only slowly from pulse
to pulse and so coherent integration is applied  for each range gate,
groups of consecutive samples are summed together. The number of
complex samples in the resulting time series is thus reduced, and the
time interval between them is increased, by a factor equal to the
number of coherent integrations (typically of the order of a few
hundred). The time interval between the new samples determines the
Nyquist velocity, the maximum radial velocity that can be
unambiguously determined. Decoding must be applied to the coherently
integrated samples if a complementary transmitter code has been
used. For each range gate, a complex Doppler frequency spectrum is
derived by applying a weighting window to the complex time series data
followed by a discrete Fourier transform. This spectrum is multiplied
by its complex conjugate to give a power spectrum. Doppler frequencies
are converted into Doppler velocities by multiplying by half the radar
wavelength. The sign must be changed so that movement away from the
radar (which gives rise to a negative frequency shift) is represented
by a positive velocity. If desired, consecutivelyobserved Doppler
velocity power spectra may be incoherently integrated by adding them
together. This increases the detectability of signals. In general a
Doppler velocity power spectrum contains one or more signal components
superimposed on a background of nominally white noise. The noise power
is dominated by broadband lowerVHF cosmic radiation, which undergoes
a diurnal variation by a factor of approximately 2.
Spectral Processing  click here for more details
For each Doppler velocity power spectrum, the noise power spectral
density (PSD) is determined by the method of Hildebrand and Sekhon
(1974). The noise power is equal to the noise PSD summed across the
width of the spectrum. The velocity bin limits of the strongest signal
component are determined by first locating the peak value of the of
the runningmeansmoothed PSD. The smoothed PSD is then followed to
either side until one of the following conditions is encountered: the
smoothed PSD has dropped below the noise PSD, the smoothed PSD has
dropped below a set fraction of the peak value, or a local minimum is
encountered (and the smoothed PSD is below a set fraction of the peak
value). The final criterion is particularly important under stratiform
precipitation conditions in order to separate partiallyoverlapping
clearair and hydrometeor signal components. The zeroth (m0), first
(m1) and second (m2) order moments are calculated within these limits
in order to derive the signal power (m0), the radial velocity (m1/m0)
and the spectral width (sqrt[(m2/m0)  (m1/m0)**2]). For stmode
observations it is desirable to identify more than one signal
component per spectrum (typically two). A radial continuity algorithm
is then used to identify the primary signal component for each range
gate, i.e. that which leads to the most likely overall clearair radar
return profile. A second attempt may be made to identify the primary
signal components if the first profile is deemed to be contaminated by
interference. A final attempt is made to improve the selection of
primary signal components for the lowest range gates in case of
contamination by hydrometeor returns. Subsequently attention is
confined to the primary signal components. Nevertheless, the radar
return parameters for the nonprimary signal components are saved in
the radial data files as they may contain scientifically useful
information, e.g. concerning precipitation.
Time Continuity Testing
For windprofiling purposes, MST radar observations are cycled
through a sequence of dwells with different beam pointing
directions. The radial velocity observed by a vertical beam (i.e. a
dwell with a beam pointing zenith angle of zero) is assumed to
represent the vertical component of the wind velocity. The radial
velocity observed by an offvertical beam (i.e. a dwell with a small,
nonzero beam pointing zenith angle) is assumed to represent the
vector addition of the vertical component of the wind velocity
multiplied by the cosine of the zenith angle, and the component of the
horizontal wind velocity along the the radar beam pointing azimuth
multiplied by the sine of the zenith angle. Consequently a component
of the horizontal wind can be derived for each vertical/offvertical
beam pair. When more than one vertical beam observation is made per
cycle, that closest in time to the offvertical beam observation is
used for deriving the horizontal wind components. When combining
vertical and offvertical beam radial velocity components, vertical
beam data are taken from those range gates which are most closely
matched in altitude to the offvertical beam range gates. A time
continuity algorithm is applied to the horizontal wind components for
the offvertical beams, and to the vertical wind components for the
vertical beams, as a further test for reliability. Time continuity is
first established unidirectionally, i.e. by comparing the
observations to be flagged only with those made at earlier times. This
allows windprofile data to be made available with only a short time
delay. However, the process is repeated as soon as there are
sufficient subsequent observations to allow bidirectional flagging to
be applied. The overall reliability of signal components requires them
to have passed both radial continuity (when the tests have been made)
and time continuity tests.
Deriving Cartesian Wind Components
In order to derive Cartesian components of the wind vector,
observations must be made in the vertical direction and at an
offvertical angle in two orthogonal azimuths. A typical cycle of
observation includes many more beam pointing directions. The
availability of additional information allows improved wind component
estimates to be made. Attention is confined to the horizontal wind
components (derived from vertical/offvertical beam pairs, as
described above) arising from offvertical beam observations made with
a single, predetermined zenith angle (indicated by the global
attribute cart_horiz_wind_zen_angle_deg;
this is typically 6.0 degrees). If observations are made more than
once per cycle with the same offvertical beam, only those from the
first instance are considered. If observations have been made with the
complementary offvertical beam (i.e. that with the same zenith angle
but with an azimuth angle which differs by 180 degrees), the
horizontal wind components are averaged (if they are both flagged as
being reliable). The difference between them gives a measure of
reliability of the averaged value. It gives an indication of the
degree to which a fundamental assumption of the windprofiling
technique is valid: that the wind field is stationary over the time
taken to complete a full cycle of observation (typically a few
minutes) and over the horizontal distance separating the radar
observation volumes for the different beam pointing directions (of the
order of 2 km at 10 km altitude). If the difference exceeds a
predefined value (given by the global attribute cart_max_compl_beam_horiz_vel_diff_mps), the
averaged horizontal wind component is flagged as being unreliable. The
horizontal wind complementary beam variability values stored in the
Cartesian files represent the root of the sum of the squares of the
values for the primary and secondary (i.e. orthogonal) radar
azimuths. The primary radar azimuth is given by the global attribute
cart_horiz_wind_primary_azi_angle_deg. Owing
to the aspect sensitivity of radar returns (i.e. to the fact that the
radar return signal power often decreases with increasing zenith
angle) and to the finite radar beam width, the effective pointing
angle for an offvertical beam is typically slightly closer to the
vertical than the nominal zenith angle. This leads to slight
underestimates of the magnitudes of the horizontal wind
components. The amount can quantified, and the horizontal wind
components are compensated, through consideration of the theta_s
parameter, which is derived from the ratio of the radar return signal
powers at two nonzero zenith angles (Hooper and Thomas, 1995). If
observations are made more than once per cycle at these zenith angles
(given by the global attributes cart_theta_s_low_zen_angle_deg and cart_theta_s_high_zen_angle_deg) the radar
signal powers are first averaged. Finally the horizontal wind
components are rotated from their native azimuths (i.e. from cart_horiz_wind_primary_azi_angle_deg and an
orthogonal direction) to give northward and eastward components. A
single flag indicates the reliability of both horizontal wind
components.
The Accuracy of Horizontal Wind Components
The characteristics of thirtyminuteaverage windprofile data
derived from this (version 3) processing scheme have been evaluated by
the Met Office. The random errors, evaluated over 7 days, are
typically in the range 1.0  3.0 m/s for altitudes between 2 and 15 km
and in the range 3.0  4.0 m/s between 15 and 20 km. The number of
reported winds begins to decrease with increasing altitude in the
upper altitude range. Winds are reported at least 80% of the time up
to 18 km but no more than 30% of the time at 20 km. Comparisons have
also been made against the winds from the Met Office's numerical
weather prediction model. The magnitudes of the component biases are
less than 0.5 m/s at all altitudes. The root mean square component
differences are in the range 2.0  3.0 m/s. The directional bias is
approximately 1.0 degree and the magnitude of the speed bias is less
than 1.0 m/s. These values are comparable to those for radiosonde
data.
Vertical Beam Data Products
The vertical wind component (for which the standard name is upward_air_velocity) is given by the radial
velocity of the vertical beam observation. The accuracy of these
estimates is difficult to estimate owing to the lack of comparable
measurements. The value of 0.2 m/s represents the typical magnitude of
fluctuations (about zero) under quiet conditions. Mountain wave
conditions give rise to peak values of the order of 1.0 m/s and
convective conditions to peak values of the order of 10.0 m/s. When
more than one vertical beam observation is made per cycle, only the
first instance is considered. The vertical beam radar return signal
power and spectral width also relate to this instance. A single flag
indicates the reliability of all three vertical beam
products. Vertical beam data products are taken from those range gates
which are most closely matched in altitude to the range gates for the
the offvertical angle (cart_horiz_wind_zen_angle_deg) used for the
horizontal wind estimation. This is true of all data products stored
in the Cartesian files. Changes in the vertical beam radar return
signal power from the uppertroposphere/lowerstratosphere region
tends to be closely related to those in the vertical temperature
gradient. This allows both the altitude and the sharpness of the
tropopause to be determined by the method of Hooper and Arvelius
(2000).
Beam Broadening Corrected Spectral Width
The spectral width of vertical beam radar returns tends to be
dominated by the beambroadening component, which is equal to the
product of the horizontal wind speed and of the radar beam twoway
halfpower halfwidth. The beam broadening corrected spectral width
values represent the the root of the difference between the squares of
the observed widths and of the beambroadening components. It has its
own reliability flag. The accuracy of 0.1 m/s is estimated from the
typical spread of those beambroadening component values which exceed
the observed values. No beam broadening corrected values can be
calculated under such conditions.
Aspect Sensitivity
In addition to quantifying the aspect sensitivity of radar returns
through the theta_s parameter, a simpler measure is given by the ratio
of the signal power observed by the vertical beam to that observed at
an offvertical angle (given by the global attribute cart_asp_sens_high_zen_angle_deg, which is
typically 6.0 degrees). Signal powers are first averaged when there is
more than one observation per cycle for each of the required zenith
angles. Small values of aspect sensitivity (< 5 dB) imply
backscatter from quasiisotropic refractive index irregularities,
which suggests that the atmosphere is well mixed. Large values (>
10 dB) imply that the coherence length of the refractive index
irregularities is far greater in the horizontal than in the
vertical. Such structures are consistent with the atmosphere being not
well mixed. The aspect sensitivity tends to be low in the troposphere
and high in the stratosphere although this is not always the case. The
aspect sensitivity has its own reliability flag.
Internal Links:
 Return to the top of the page
 Gaining access to the data
 File naming convention
 Data archiving convention
 Data locations
 The differences between
signal processing versions
 The contents of other data
files
 An overview of v3
signal processing
 A general description of netCDF file
structure.
External Links:
 The Unidata
netCDF website
