A simple and robust method for calibrating ceilometers has been
tested in an operational environment, demonstrating that the calibrations are
stable to better than ±5 % over a period of a year. The method
relies on using the integrated backscatter (B) from liquid clouds that
totally extinguish the ceilometer signal; B is inversely proportional to the
lidar ratio (S) of the backscatter to the extinction for cloud droplets. The
calibration technique involves scaling the observed backscatter so that B
matches the predicted value for S of 18.8±0.8sr for cloud droplets,
at ceilometer wavelengths. For accurate calibration, care must be taken to
only use profiles where the range correction is implemented and to exclude
any profiles having targets with different values of S, such as drizzle
drops and aerosol particles, profiles that do not totally extinguish the
ceilometer signal, profiles with low cloud bases that saturate the receiver,
and any profiles for which the window transmission or the lidar pulse energy
falls below 90 %. A range-dependent multiple-scattering correction that
depends on the ceilometer optics should also be applied to the profile. For
ceilometers operating at around 910 nm wavelength, a simple correction for
water vapour attenuation is applied to the signal using the vapour profiles
from a forecast analysis. For a generic ceilometer in the UK the 90 d
running mean of the calibration coefficient over a period of 20 months is
constant to within 3 % with no detectable annual cycle, thus confirming
the validity of the humidity and multiple-scattering correction. For
Gibraltar, where cloud cover is less prevalent than in the UK, the
90 d
running mean calibration coefficient was constant to within 4 %. The more
sensitive ceilometer model operating at 1064 nm is unaffected by water
vapour attenuation but is more prone to saturation in liquid clouds; such
profiles can be recognised and rejected and, despite the more restricted
sample of cloud profiles, a robust calibration is readily achieved. In the
UK, the running mean 90 d calibration coefficients varied by about 4 %
over a period of 1 year. The consistency of profiles observed by nine
pairs of co-located ceilometers in the UK Met Office network operating at
around 910 and 1064 nm provided independent validation of the calibration
technique. In all cases, if quantitative and reliable backscatter
observations are to be obtained it is essential to keep the window clean.
This may be a challenge in dusty locations. EUMETNET is currently networking
700 European ceilometers so they can provide ceilometer profiles in near
real time to European weather forecast centres and has adopted the cloud
calibration technique described in this paper for ceilometers with a
wavelength of around 910 nm.
Introduction
Ceilometers are simple, relatively inexpensive vertically pointing lidars
that typically operate at wavelengths of 905–910 or 1064 nm. They can be
left unattended for long periods and, as the name suggests, have mainly been
used for detecting cloud base height at airports where they are valuable for
air safety issues. Recent studies have shown that, in addition to detecting
the large backscattered return signal from the cloud base, they can also provide
vertical profiles of backscatter from both clouds and aerosols every 5–30 s
with a range resolution as low as 5 m. Ceilometer profiles have been
used in many research contexts; some examples are the Cloudnet scheme for
validation of the representation of clouds in operational numerical weather
prediction (NWP) forecast models (Illingworth et al., 2015), for aerosol
profiling (Markowicz et al., 2008; Madonna et al., 2015), fog
observations (Dupont et al., 2012) and the retrieval of mixing
height levels (Münkel et al., 2007).
Operational weather forecasting models such as those operated by the ECMWF
and MeteoFrance now represent both clouds and aerosols by prognostic
variables. Remote-sensing observations are needed to show that these models
are providing unbiased estimates of aerosol and cloud properties and
ultimately for data assimilation into such models to improve forecasts of
hazardous weather such as pollution episodes and severe convective storms
producing flash floods. The European Ground-Based Observations of Essential
Variables for Climate and Operational Meteorology (EG-CLIMET), which was a
recent Cooperation in Science and Technology (COST) action financed by the
European Union, noted that there are hundreds of ceilometers deployed over
Europe which are currently underexploited. EG-CLIMET recommended that the
ceilometers be networked to provide users easy access to calibrated
backscatter data (Illingworth et al., 2015). At the time
of writing, profiles from 200 ceilometers from 17 countries are being
distributed in near real time by the E-Profile programme of European
Meteorological Services Network (EUMETNET) with the number expected to
rise to about 700. The data formats, calibration techniques and retrieval
algorithms are being developed by COST action 1303: Towards operational
ground based profiling with ceilometers, Doppler lidars and microwave
radiometers for improving weather forecasts (http://www.toprof.imaa.cnr.it, last access: 15 October 2018).
If ceilometer data are to be used in an operational context, and potentially
for data assimilation, accurate calibration is essential when verifying
model performance by forward modelling the attenuated backscatter; for
example, Illingworth et al. (2019) show that a calibration accuracy of
10 % is needed when deriving O-B statistics obtained by comparing the
observed ceilometer backscatter (O) from Saharan dust with the forward
modelled backscatter (B) from the ECMWF CAMS model. The World
Meteorological Organisation requirements (OSCAR, 2018) suggest the goal for
ice water content (IWC) observations is to have an accuracy of 10 % and
for aerosol optical extinction to have an absolute accuracy of 0.01 km-1,
but no fractional accuracy is quoted. In most models the ice
particle density is assumed to be inversely proportional to particles size
(e.g. Brown and Francis, 1995), so IWC is proportional to extinction and for
a given lidar ratio and small amounts of attenuation, the requirement is for
a ceilometer calibration to be accurate to 10 %.
The use of theoretical calibrations for lidars and radars based on an
accurate budget of the losses and gains in the transmission and reception
optics and in the electronics together with atmospheric attenuation can
cause large errors (Protat et al., 2011). Accordingly, it is preferable to
find some natural target that has a known backscatter value. There are two
such candidates for ceilometers: firstly, the backscatter from the molecules
in the atmosphere and, secondly, the integrated backscatter profile from
water clouds that totally extinguish the lidar beam. In this paper, we will
focus on the second method. This method, using the attenuated backscatter
signal from liquid water clouds, relies on the fact that the backscatter to
extinction ratio (S) is a known value of 18.8 sr for wavelengths of
relevance to ceilometers (O'Connor et al., 2004). The advantage of this
method is that the backscatter values from liquid water clouds are very high
(typically peaking at 0.3 km-1sr-1) so the signal-to-noise ratio
of water cloud returns is very large. By contrast, the molecular signal
close to the ground is over one hundred times lower than the cloud returns
and of the order 10-3km-1sr-1, falling off exponentially
with height. For an accurate estimate of the molecular return it is
necessary to average the ceilometer profiles over several hours on selected
cloudless nights when there is negligible backscatter from thin cirrus
clouds or aerosols (e.g. Tsaknakis et al., 2011;
Wiegner et al., 2014).
In this paper, we present a development of the calibration technique using
liquid clouds that can be implemented operationally and which avoids the
aforementioned potential problems. We report on the values of the
calibration for the Met Office network of ceilometers and show the
calibration stability in time. In Sect. 2, we review the specifications
and performance of the two ceilometer models in widespread use in Europe.
The calibration algorithm is described in Sect. 3 and the instrument model-dependent corrections and calibration results are addressed in Sects. 4
and 5. Finally, in Sect. 6, we report on collocated ceilometer comparisons
and statistics of the stability and accuracy of the calibration.
InstrumentationThe Met Office ceilometer network
Figure 1 shows the locations of the 40 ceilometers in the UK that are
presently reporting the full vertical profiles of the attenuated atmospheric
backscatter and are referred to in this paper as the “Met Office ceilometer
network.” The purple crosses show the locations of the 29 Vaisala CL31
ceilometers and the red circles show the 11 Jenoptik CHM15k Nimbus
ceilometers that have been used to test the ceilometer calibration
technique. Nine sites have collocated Vaisala and Jenoptik ceilometers.
Other Met Office ceilometers, many of which are the Vaisala CT25K model,
report only cloud base height and are not discussed here, although the
calibration technique can be applied to both the CT25K and the newer CL51
models. Note that Jenoptik no longer produces ceilometers; the manufacturing
of them has been taken over by Lufft. From here on, we refer to these
ceilometers as Lufft ceilometers, including those manufactured before
production passed from Jenoptik to Lufft.
Location of Met Office ceilometers which record the full profile of
attenuated backscatter. The Vaisala CL31s are indicated by a purple cross and the Lufft CHM15k by red dots.
Vaisala CL31 ceilometers
The key technical properties of the ceilometers used by the Met Office are
summarised in Table 1. In brief, the Vaisala CL31 ceilometers use an InGaAs
diode laser which emits pulses with an energy of 1.2 µJ at a pulse
repetition frequency (prf) of 10 kHz with a central wavelength of 910±10nm, though the typical spectral width is more often 4 nm (Kotthaus
et al., 2016; Markowicz et al., 2008). At these wavelengths, attenuation by
water vapour is significant, a fact overlooked by O'Connor et al. (2004).
Ceilometers generally operate at low power, so, to increase the
signal-to-noise ratio, they tend to have higher pulse repetition rates
compared to high-power lidars; the returns from distant signals are
generally very low so “second trip” echoes are not usually a problem. The
CL31s have a single-lens design, with the centre of the lens collimating the
laser beam and the outer part of the lens used for focussing the
backscattered light onto the receiver, which uses an avalanche photodiode
(APD) detector to process the signal (Münkel et al., 2007). Complete
overlap of the transmitted beam at the receiver sample is achieved at a
height of approximately 70 m (Martuccci et al., 2010) and the maximum range
is 7.7 km.
Summary of some technical characteristics and parameters of the
Vaisala CL31 and Lufft CHM15k, as operated in the Met Office network.
a The
Met Office CL31 ceilometers have a vertical resolution of 20 m, apart from
at Exeter CL31, where the vertical resolution is 10 m.
There are currently several different versions of the firmware in use by the
Met Office ceilometer network. The various versions process the signal in
different ways, applying “cosmetic” shifts to the data to avoid unphysical
negative backscatter values. The original users for ceilometer data were
aviation forecasters and air traffic control and these cosmetic shifts were
applied so that it was easier for non-experts to interpret the displays.
Full details of the shifts and methods for correcting can be found in
Kotthaus et al. (2016). These effects should certainly be corrected for in
the study of smaller particles such as aerosols and ash; however, for the
stronger signal from cloud particles the effect of these shifts on the
calibration method shown here is negligible.
Lufft CHM15k Nimbus ceilometers
The Lufft ceilometers use a Nd:YAG laser and operate at a slightly longer
wavelength of 1064 nm where the attenuation by water vapour is negligible.
The APD detector employs a photon-counting method. Due to the biaxial design
of the Lufft ceilometers, full overlap is not reached until 1 km rather than
70 m for the CL31. The pulse repetition frequency is in the range
5–7 kHz
and the pulse energy is 8 µJ, which is six times higher than the
Vaisala CL31 ceilometers. This higher pulse energy, combined with the
different overlap configuration, results in a much higher sensitivity of the
CHM15k ceilometer, for detection of elevated aerosols such as volcanic ash
plumes.
The calibration algorithm
Autocalibration of ceilometers using liquid water cloud was proposed by
O'Connor et al. (2004) as a simple method that requires no
additional instruments to compute a calibration coefficient. The technique
relies on the use of the lidar ratio (ratio of extinction to backscatter,
denoted S), which is a constant for the droplets in liquid water cloud.
Several studies have derived S from Mie theory: Pinnick et
al. (1983) found that, for a wavelength of 1064 nm, S=18.2sr;
Wu et al. (2011) calculated an S of 18.5±0.47sr for
a wavelength of 1064 nm. O'Connor et al. (2004) calculated
an S of 18.8±0.8sr for a wavelength of 905 nm and showed that this
was essentially constant for the observed cloud droplet size distribution
for a mean droplet size ranging from 10 to 50 µm, but S values were
lower for drizzle having larger droplets. Since S is very similar at 905
and 1064 nm, we follow O'Connor et al. (2004) and use S=18.8sr for both
wavelengths.
The method compares this theoretical S to a calculated apparent S. When
the ceilometer signal is completely extinguished by the cloud, the total
path integrated attenuated backscatter B is equal to the reciprocal of
twice the lidar ratio:
B=∫0∞βobserveddz=∫βTrue(z)exp-2τzdz1=1ηS∫exp-2τdτ=12ηS,
where B is the total integrated attenuated backscatter, τ is the
optical thickness, S is the theoretical lidar ratio, and η is a
multiple-scattering correction which is dependent on laser wavelength, beam
divergence, telescope field of view and altitude (z). The multiple-scattering corrections are height dependent and calculated for each gate
using the fast method and code described by Hogan (2006; code available to
download at http://www.met.reading.ac.uk/clouds/multiscatter/, last access: 15 October 2018). η is
usually between 0.7 and 0.85 for wavelengths between 905 and 1064 nm in liquid
water clouds. The calibration technique involves multiplying the observed
backscatter signal βobserved by a calibration coefficient, C,
until Bη=0.0266m-1, which is the value for water drops when S=18.8sr. Note that C is a scaling factor and is the reciprocal of the widely
used calibration constant, CL, which is often used for photon-counting
receivers and is the factor by which the count should be divided to obtain
a calibrated value (e.g. Wiegner et al., 2014).
The calibration technique will fail if there are targets contributing to B
that have an S that is not equal to 18.8 sr. At ceilometer wavelengths,
aerosols generally have S values above those for cloud droplets; marine
aerosols have an S close to 20 sr, but most aerosols have values that are
much higher and in the range 40 to 100 sr for dust, smoke and ash (e.g. Omar
et al., 2009). If aerosols with S higher than 18.8 sr are included in
profiles leading to total attenuation of the signal, then the value of B
will be less than for cloud alone, and the apparent value of the calibration
coefficient, C, will be too high. Conversely, drizzle has S values below
those for cloud droplets, so if drizzle is included in the profile, the
value of B will be higher than for cloud alone, and the value of C would be
too low. The magnitude of the error due to aerosol depends on its optical
depth beneath the cloud layer; therefore, we can circumvent this uncertainty
by not selecting profiles which have large backscatter from aerosol. The
inclusion of profiles that do not totally extinguish the ceilometer
return will also lead to values of C that are too high, as will occasions
when the window transmission is reduced or the pulse energy falls.
Profiles of attenuated backscatter through stratocumulus cloud. Panel (a) shows an example of a suitable profile for calibration. The
integral of the profile (grey shaded area) is equal to 12ηS
and, when calibrated, should give an S of 18.8±0.8sr. Panel (b) shows
an example of a profile unsuitable for calibration due to the high
levels of aerosol in the first 200 m, indicated by the grey shading up to
200 m, and due to the drizzle below the stratocumulus cloud, indicated by
the slight increase in attenuated backscatter underneath the peak.
Figure 2a shows an example of an uncalibrated attenuating backscatter
profile typical of those from stratocumulus clouds that is ideal for use in
the liquid cloud calibration algorithm. Cloud is observed as the sharp peak
in attenuated backscatter just above the cloud base, rising to a maximum value of
0.28 km-1sr-1 within a few range gates and clearly dominating
the observed ceilometer return. The shaded area indicates the area of
integration used in computing the total attenuated backscatter of the
profile. The profile in Fig. 2b is for a stratocumulus cloud that completely
attenuates the ceilometer return. However, in this case, it is unsuitable
for calibration because there is a significant return from aerosol in the
lowest 200 m of the profile and a more gradual increase in attenuated
backscatter below the peak at the cloud base, indicating the presence of drizzle
below the cloud.
(a) Uncalibrated attenuated backscatter vertical profiles (colours
shown on a log scale) for 25 October 2014 from a CL31 ceilometer at Middle
Wallop airfield (51.1489∘ N, 1.5700∘ W) and (b) the apparent lidar ratio for
the same day. In (b), the grey line shows the apparent S for profiles
that pass the step 1 filtering of the calibration algorithm and the black
line shows the profiles that pass the step 2 filtering and are used to
calculate the calibration coefficient.
A new algorithm has been designed to automatically sift through all profiles
of attenuated backscatter, selecting only those suitable for the cloud
calibration according to a strict set of criteria. The method is fairly
simple, ensuring that it can be applied operationally with minimal impact on
processing time. No absolute values of βatt are required by the
algorithm to evaluate the criteria below, so the instrument can be
completely uncalibrated, or the calibration currently applied can have a
large error. The algorithm only requires a minimum of 10 suitable profiles
in a day for a calibration coefficient to be calculated. This means the
calibration algorithm is suitable for ceilometers at sites where liquid
water cloud can be sparse and infrequent. There are two main sets of
criteria that must be met by the profile of attenuated backscatter for it to
be used to calculate a calibration coefficient:
Unsuitable individual profiles.
Aerosol filter. In any single profile, if the aerosol under the cloud
contributes more than 5 % to the total integrated backscatter (as shown in
Fig. 2b), then this profile is removed from the calibration. The
transmission through the aerosol below the cloud attenuates the ceilometer
beam and this attenuation increases with greater concentrations of aerosol.
If the aerosol has a lidar ratio value twice the value assumed for cloud
droplets, then this filter should limit the calibration error to a maximum
of 5 %.
Peak sharpness filter. The peak backscatter magnitude must be a factor of 20
greater than the value 300 m above and below that peak. A liquid water cloud
suitable for calibration must fully attenuate the ceilometer beam;
therefore, the backscatter values should decrease rapidly in the gates
immediately above the peak value. Additionally, drizzle or rain below the
cloud may give a large backscatter signal and, like the aerosol, will
distort the apparent lidar ratio. Hogan et al. (2003)
report that individual liquid-water layers do not tend to occupy more than
300 m of the ceilometer profile due to their strong attenuation. Our own
observations of the data lead to the same conclusion. This filter should
therefore remove profiles that do not fully attenuate the beam and those
that contain drizzle or rain.
Window transmission and pulse energy check. A check is made on the recorded
instrument transmission (given as a percentage of how much of the instrument
window is clear) and on the reported pulse energy (given as a percentage of
a nominal amount). Both of these conditions can affect the true value of
attenuated backscatter. For considering instrument and calibration
stability, periods affected by reduced window transmission and/or reduced
pulse energy are filtered out at a threshold of 90 %. For quantitative
calibrations and observations of backscatter it is essential that the window
be kept clean. It may be possible to correct the observed backscatter for
low pulse energy but seems most unlikely that corrections can be made for
the low window transmission because any dust or dirt covering on the window
is probably not homogenous. It may be difficult to keep the window clean in
locations where dust is common.
Consistency of neighbouring profiles.
Lidar ratio stability. This filter traps errors due to patchy cloud cover
or drizzle that may not have been identified by the first filter by checking
that the apparent lidar ratio is the same as its nearest neighbours. The
recommendation is to compare to three profiles on either side; however, if the
ceilometer is at a site where liquid water cloud is infrequent, this could
be reduced to one or two profiles either side, with consequent degradation of
the accuracy of the calibration coefficient. There must be at least 10
acceptable profiles for a calibration coefficient to be recorded for that
day.
The operation of these filtering procedures in removing unsuitable profiles
is illustrated in Fig. 3a, where a stratocumulus cloud layer located at
about 1 km for the whole day is ideal for calibration. The liquid cloud
backscatter signal, which has values greater than 10-0.5km-1sr-1
(in the red region of the colour scale), appears as a thin
layer above which there is only noise. Within the noise, the diurnal cycle
of the skylight is visible. The noise in the data is visible as speckling
and is of the magnitude of less than 10-3km-1sr-1. There
are also limited periods of broken, patchy cloud, which are identifiable by
breaks in the layer of high backscatter, and limited periods of drizzle,
which are identifiable by the fall streaks (cyan colours of the order
10-2.5km-1sr-1) below the cloud.
Figure 3b illustrates the two main filtering steps of the new calibration
algorithm. The thick, light grey line shows the apparent S values for each
individual profile that is acceptable and has removed those profiles between
3.30 and 4.30 h where there is drizzle and aerosol below the cloud
base, but this still leaves some large apparent S values from 12.00 to 15.00 h
that are due to broken cloud that does not totally extinguish the
ceilometer return. The second filter checks for consistency between
neighbouring profiles and successfully identifies and removes these spurious
profiles where there is broken cloud.
The remaining profiles (i.e. those in black in Fig. 3b) give the values of
apparent S which would be used to calculate C using Eq. (1). It is evident
that these values remain very constant over the course of the day, implying
that the calibration of the instrument is very stable on this time scale.
This is important, since our method can only be applied during cloudy
conditions, which may be separated by intervals of several days. The
stability of C implies we can interpolate between calibration events.
Estimated transmission loss due to the atmospheric water vapour
content. The blue crosses are the values calculated by Markowicz et al. (2008) for a ceilometer with a wavelength of 910 nm.
Comparison of water vapour transmission correction methods using
ECMWF water vapour density profiles for 25 October 2014 at Middle Wallop,
England. In blue, the transmissivity is calculated using WAPL (Wiegner and
Gasteiger, 2015) and in red the transmissivity has been calculated using the
empirical function shown in Fig. 4. The black lines show the instrument-reported cloud base height at that time.
Calibration of 910 nm ceilometersWater vapour attenuation
To complete the calibration of the Vaisala CL31 ceilometers (and others of
similar wavelength – e.g. Vaisala CL51, CT25k, CT75k, Campbell Scientific
CS135), the effect of atmospheric water vapour below the cloud on the laser
signal must be considered. This is because the wavelength of these
ceilometers (910 nm) is in a weak water vapour absorption band. Note
that, because the Lufft CHM15k ceilometers operate at 1064 nm, where there is a
water vapour absorption window, those ceilometers do not require a
correction; however, the Lufft CHM8k operates at 905 nm and so would require
a water vapour absorption correction.
A recent paper by Wiegner and Gasteiger (2015) describes a
method of correcting for water vapour attenuation for ceilometers at
wavelengths around 910 nm by performing detailed line-by-line radiation
transfer calculations and investigating the impact of the instrument
emission spectrum (e.g. Vaisala states that for a CL31 the wavelength is 910±10nm at 25 ∘C and with a drift of 0.3 nmK-1). As
the housing of the ceilometer lasers and detectors are
temperature-controlled environments, the effect of laser wavelength drift
due to temperature can be considered insignificant. However, even if the
potential for drift is ignored, Wiegner and Gasteiger's method still
requires the use of a radiative transfer model or access to their WAPL
database of absorption coefficients. Because the liquid cloud calibration
method presented in this paper is intended for operational, real-time use, a
simple, robust and computationally cheap method was required.
A simplified technique for correcting for the two-way water vapour
attenuation has therefore been devised based on Markowicz
et al. (2008), who show that the normalised spectrum of laser emission is
wide enough to smooth out the individual water absorption lines so that, for
a water vapour path of 2 cm, a typical summer value in the UK, the change in
water vapour transmission varies from about 0.77 to 0.75 (about 3 %) as
the peak laser emissivity increases from 900 to 916 nm. The typical water
vapour path in winter is 1 cm leading to a transmission of about 0.85, so if
no water vapour correction was made, one would expect an apparent annual
cycle of the calibration coefficient of about 12 %. The water vapour could
be estimated using a microwave radiometer. Alternatively, it can be obtained
from a numerical weather prediction (NWP) model. In this paper we take the
latter approach. Cossu et al. (2015) have compared NWP output with the water
vapour path derived from microwave radiometers and find that the mean bias
of the NWP water vapour path is only 0.7 mm.
A simple monotonic function has been fitted to data from
Markowicz et al. (2008) in order to parameterise the two-way attenuation by water vapour as a function of integrated water vapour
(IWV) up to 2 cm at wavelengths of ∼910nm depicted in Fig. 4:
Twv=1-0.17IWV(z)0.52,
where Twv is the two-way transmission as a percentage of the
transmission without water vapour attenuation and IWV(z) is the atmospheric
water vapour content from the surface to height z in gcm-2. The
attenuated backscatter is then corrected using the following:
B=∫βatt×Cwvdr,
where Cwv=1Twv.
The transmission calculation for each range gate requires the water vapour
content obtained by integrating the water vapour density from the ground to
each specific range gate, resulting in a transmission profile. For the
automatic operational calibration of the Met Office ceilometers, water
vapour density would be calculated from the Met Office UKV model, a
convection-permitting variable-resolution regional NWP model run
operationally over the UK (Tang et al., 2013), using pressure, temperature
and specific humidity. A comparison of the detailed line-by-line Wiegner and
Gasteiger method with the simpler approach using the water vapour density
profiles obtained from the ECMWF operational forecast model provided by
Maxime Hervo (MeteoSwiss, personal communication, 2016) is shown in Fig. 5.
The WAPL method is depicted in blue and the new, simple method is in red.
The transmissivity profiles differ by a maximum of 2 % for a total
transmissivity of 0.85.
Region of integration
For the Vaisala ceilometers in the Met Office network, a cosmetic feature in
the firmware suppresses the range correction to the received power for
heights above 2.4 km, except when there are clouds present. This is done to
avoid the background noise signal leading to apparent clouds at high
altitudes that might confuse the non-expert, so for the calibration
procedure, profiles above 2.4 km are not suitable as the return signal may
not have been range corrected. In addition,
Kotthaus et al. (2016) found that the attenuated
backscatter in the lowest 200 m may be subject to artefacts so, in this
calibration study of the Met Office's Vaisala ceilometers, the cloud returns
above 2.4 km and below 200 m are not used.
2-D histogram of integrated attenuated backscatter with range, with
height-dependent multiple-scattering correction applied. Darker colours
(towards red) indicate a higher density of profiles. The values shown along
the right-hand side give the mean±SD of the
integrated attenuated backscatter (units sr-1) at 100 m intervals.
Figure 6 shows a histogram of the integrated attenuated backscatter, B, from
liquid cloud as a function of the height of the maximum attenuated
backscatter (used as an indicator of the height of the cloud), for profiles
from an uncalibrated Met Office Vaisala CL31 situated at Middle Wallop
(51.15∘ N, 1.57∘ W). Multiple scattering and water
vapour attenuation below the cloud have been accounted for. Over 100 000 profiles
were used from the period September 2014 to December 2015. The
numbers superimposed on the right side of the plot show the mean and
standard deviation (SD) at 100 m intervals of the range. For 16 of the 21 heights
shown, the mean value is 0.021 sr-1. The other five gates differ by a
maximum of only 0.002 sr-1. This provides confirmation, both of the
validity of the range-dependent multiple-scattering correction and the
assumption of constant S for different water clouds.
Calibration coefficient (C) for Middle Wallop CL31 from September 2014 to April 2016.
Each black cross represents a single day, calculated from
profiles deemed suitable by the calibration algorithm. Panel (a) shows the
mode of C for each individual day, (b) shows the mean of C for each day, with
the standard deviation shaded in grey and (c) shows a 90 d running mean for
the 20-month period. The average of the daily modes is 1.38±0.14,
the average of the daily means is 1.41±0.13 and the average of the
90 d running means is 1.40±0.021. The water vapour absorption
correction has been applied, without which there would be an annual cycle in
the calibration of about 12 %.
Below 500 m there is, however, a slight change. The distribution of the
integrated attenuated backscatter is still concentrated in a similar region
to other heights, but it also has a slight tail to the left. For profiles
in this tail region below 500 m, the attenuated backscatter is smaller,
which will result in a larger apparent lidar ratio. The mean value of the
integrated attenuated backscatter at heights below 500 m decreases by
9.5 %, with the standard deviation increasing by 17 %. We suspect that
this is a result of the instrument detector saturating or maybe range-dependent multiple scattering not being calculated correctly at close ranges
because of imperfect telescope alignment. When the cloud is very low, the
cloud signal may be so strong at its peak that the true magnitude of the
backscattered signal is not fully detected and, therefore, the integrated
attenuated backscatter appears smaller when compared to other heights. It is
also possible that this may, occasionally, be due to microphysical processes
within the cloud. Nicholls (1984) showed that there is a reduction in
droplet number concentration below 450–500 m (Powlowska et al., 2000). This
may, in some cases, be significant enough to affect the backscatter at this
height. Therefore, we also reject profiles where the cloud is between 200 and 500 m.
Calibration results for Middle Wallop
Figure 7 shows a time series of the calibration results for the CL31 at
Middle Wallop in southern England over a period of 20 months. The top panel
shows the mode of the calibration coefficient, C, for each day with
sufficient (minimum 10) attenuated backscatter profiles deemed suitable by
the calibration algorithm. For example, the black cross on 25 October 2014
is the mode of the calibration coefficients calculated from the filtered S
values (per profile) shown (in black) in Fig. 3b.
The results are for almost 2 years of data and establish that the
calibration remains stable over time. The number of profiles used for the
calculation of the daily value is different depending on the occurrence of
cloud on each day. As the calibration algorithm requires only a minimum of
10 profiles to be included in the daily value of the mode, even a short
period of cloud will be included for calibration purposes. This ensures that
the technique can be applied to ceilometers in locations with climates that
have relatively little cloud occurrence. For this site, the algorithm found
a minimum of 8 d every month with profiles suitable for calibration, with
slightly more suitable days during autumn and winter. The water vapour
correction profiles are calculated from the Met Office UKV model at the grid
point over the Chilbolton Observatory, which is approximately 15 km from
Middle Wallop. The variables needed to calculate the transmission profiles
were available every hour and have been interpolated to the observational
time.
The middle panel of Fig. 7 shows the daily mean and standard deviation for
the same station and data. For the 20-month (574 d of data available)
period, a calibration was possible on 320 d, or 56 % of the days, and
the average number of profiles per day was 292. There were just 7 d out
of the 320 when the calibration coefficient, C, was approaching 2.0 rather
than the median value of 1.4, so a 90 d running mean was calculated and is
displayed in the lower panel of Fig. 7. This running mean had a value of C
of 1.40 with a standard deviation of 0.021; this is less than 2 % of the
mean. The 2 % of outliers all have high lidar ratios: they are probably
from occasional profiles that do not completely attenuate the lidar signal.
As both the mean height of the cloud base (and therefore the amount of
multiple scattering) and the water vapour attenuation have a pronounced
annual cycle, this low value of standard deviation is evidence of the
appropriateness of the algorithms that correct for these two effects.
Accordingly, it is recommended that for automatic, operational use for a
ceilometer, without window transmission or pulse energy issues, a 3-month
running average of the calibration coefficient be used.
Calibration results for the Met Office network
The calibration of all the Vaisala CL31 ceilometers in the Met Office
network has been collated and is summarised in Fig. 8, where box and whisker
plots are shown of the calibration coefficient for each of the CL31s,
calculated from data for the period January–March 2015. All instruments
have a calibration coefficient larger than 1.0, with the majority of the
instruments having a coefficient of around 1.5. The range of coefficients
for each station is small, with 50 % of the data (contained within the
box) being within 10 % of the mean value. The anomalously high-calibration
coefficients for Benson and Exeter are probably due to some unknown
instrument malfunction as the window transmission and transmit power are
recorded as normal. The large value of the calibration coefficient is
correcting for this effect but also flags that there is a malfunction in
the instrument. The colour code in Fig. 8 indicates the different firmware
versions installed on the instruments within the Met Office ceilometer
network. Stations using the 202 firmware, which are shaded pink (for
example, Aberporth, Coningsby, Middle Wallop), tend to have an even smaller
range of C values, with 50 % of the data being within 8 % or less of the
mean. The network includes stations from Lerwick (60.16∘ N,
1.15∘ W) down to Gibraltar (36.14∘ N, 5.35∘ W),
demonstrating that the calibration method has been successfully applied
to a range of different climates, from the North Sea down to the
Mediterranean Sea and from both coast and inland sites.
Calibration coefficient for each of the CL31 ceilometers in the UK
Met Office network: 3 months of data (January–March 2015) have been used for
each instrument. The number of suitable calibration profiles will be
dependent on occurrences of cloud and, therefore, will vary for each
instrument. The box outline represents 50 % of the calibration profiles
and the whiskers extend to include 95 % of the profiles (outliers have
been excluded from plot). The horizontal red line in the box shows the
median calibration coefficient and the smaller, filled box shows the mean.
The box plot is shaded by firmware version as given by the ceilometer files
on 1 January 2015: pink for version 202 and blue for versions 170 and 172.
The water vapour correction of the data has been applied for the
calibrations depicted in Fig. 8, as described in Sect. 4.2. Ideally, the
water vapour profiles for each specific site should be used to calculate the
transmission correction. Due to data availability, only the model data for
Chilbolton were available at this time. As the calibration specifically
requires a cloud base below 2.4 km and the air is generally well-mixed below
the cloud base, the water vapour path mixing is generally fairly constant and
depends on the temperature and height of the cloud base. Therefore, it is
assumed that the season is more important than the location and so the same
water vapour profiles are used for all the ceilometers. In future, for
operational implementation, the site-specific vapour profile would be used.
Calibration coefficient (C) for Gibraltar CL31 from
January to December 2015. Each black cross represents a single day, calculated
from profiles deemed suitable by the calibration algorithm. Panel (a) shows
the mode of C for each individual day, (b) shows the mean of C for each day,
with the standard deviation shaded in grey and (c) shows a 90 d running
mean for the 12-month period. The average of the daily modes is 1.48±0.21, the average of the daily means is 1.51±0.19 and the average of
the 90 d running means is 1.50±0.053.
Figure 9 shows the calibration of the Gibraltar CL31 ceilometer in more
detail and has the same format as Fig. 7, for 12 months at Gibraltar rather
than the 20 months at Middle Wallop. As the UKV does not cover Gibraltar,
the water vapour correction was calculated using data from the Met Office
Global Unified Model. Due to the climate, the number of occasions when there
are suitable clouds for calibration is reduced at the Gibraltar site. In 1
year there were 51 d of suitable clouds, with each day having on average
128 profiles. However, from mid-May to mid-September there were only 2 days
on which calibration was possible and in December there were none. While this
is in part due to a lower amount of stratocumulus compared to the UK, it was
also caused by the window transmission. The Gibraltar ceilometer requires
regular cleaning as the dust tends to build up on the window, reducing the
transmission. Therefore, several days on which the window transmission dropped
below 90 % have been filtered out by the algorithm. The four crosses in Fig. 9a and b
which show a calibration coefficient closer to 2.0 correspond to
days on which the profiles only just pass the 90 % window transmission
check. Nevertheless, Fig. 9c confirms that the 90 d running mean
calibration coefficient over the 12-month period was 1.5 with a standard
deviation of 0.05, or about 3 %, and, as with the data in Fig. 7, there is
no sign of an annual cycle in the calibration coefficient.
Calibration of 1064 nm ceilometers
We now address the issue of cloud calibration for the Lufft CHM15k
ceilometers, which operate at a wavelength of 1064 nm. It should be noted
that many high-power lidars have a channel at 1064 nm and can also be
calibrated with the liquid cloud method. However, as they do not have the
same firmware and hardware issues as ceilometers, high-power lidars are not
directly discussed here.
Saturation issue
Before the cloud calibration can be applied to the Lufft ceilometers, the
issue of saturation must be addressed. Due to the greater pulse energy
(compared to Vaisala ceilometers) and the receiver type (photon counting),
the Lufft ceilometers are much more prone to saturation (Whiteman, 2003).
When saturation occurs, the backscatter reported for this profile is false
– it is too low. Hence, these profiles that saturate need to be avoided.
The exact magnitude of power at which the Lufft power saturates is unknown.
However, it is possible to detect the majority of saturated profiles,
because the saturation of the receiver usually causes the output to
overshoot to an unphysical negative value just above the cloud echo
(Holger Wille, Lufft, personal communication, 2017).
Two profiles of attenuated backscatter that detect liquid cloud
from the Lufft CHM15k ceilometer at Aberporth on 20 March 2015 (b shows same plot on different scale). The profile in blue has a
negative overshoot above the cloud, whereas the red profile does not.
The first panel of Fig. 10 demonstrates the impact of saturation and the
subsequent negative overshoot: the blue profile, from the lower cloud base
where saturation has occurred, has a smaller magnitude than the red profile
of the higher cloud that has not saturated. If a saturated profile were to
be used for calibration, then the total attenuated backscatter recorded by
the ceilometer would appear lower than an unsaturated profile and would,
therefore, systematically skew the calibration coefficient to be larger than
it should be.
Because the profiles that saturate have this apparent layer of negative
attenuated backscatter, this can be used to check if these profiles should be used in the calibration
algorithm. There is a correlation between the
negative backscatter and the magnitude of saturation: the larger the
negative backscatter value, the greater the magnitude of saturation, but
this relationship is not linear and so the saturation cannot be easily
corrected (Holger Wille, Lufft, personal communication, 2017). Hence, in what
follows, we simply filter out such profiles completely. To ensure it is the
negative attenuated backscatter of a saturated profile that is detected and
not just the random noise in the profile above the cloud (which appears as
small positive and negative values varying randomly from gate to gate), any
profiles which have a layer of negative backscatter greater than 100 m are
removed from the calibration. To increase confidence that only unsaturated
profiles are used, a cloud height threshold was also imposed, as
demonstrated in Fig. 11; this shows a histogram of the uncalibrated
integrated attenuated backscatter for profiles in liquid water cloud at
Aberporth. The multiple-scattering correction has been applied, but profiles
where the instrument saturates have not been filtered out. Therefore, one
can see clearly the impact of saturation.
As for Fig. 6 but illustrating the saturation of the CHM15k
ceilometer for most clouds with bases below 2 km. 2-D histogram of the value
of the integrated attenuated backscatter in profiles used for calibration
with range. A height-dependent multiple-scattering correction has been
applied. Darker colours (towards red) indicate a higher density of profiles.
The values shown along the right-hand side give the mean±SD
of the integrated attenuated backscatter (units sr-1) at
100 m
intervals.
As shown for the Vaisala ceilometer calibration (Fig. 4), the integrated
attenuated backscatter should be a constant, independent of the height. This
is not the case for the Lufft ceilometer. This is because when the
instrument saturates, the received power becomes limited to some (unknown)
maximum power. Backscatter is proportional to received
power × range2,
which means the integrated backscatter measured will appear to
be a function of range. This is in contrast to the Vaisala ceilometer in
Fig. 6, where saturation is not occurring. Therefore, Fig. 11 shows that
saturation is occurring below a height of 2.2 km because of the systematic
change in backscatter with range. Above 2.2 km, the integrated backscatter
does not change systematically with height, showing that these higher-level
clouds are not saturating the ceilometer receiver – since they are further
away, the received power is weaker and below the level at which saturation
occurs. The exact height at which the integrated attenuated backscatter becomes
constant will be instrument-specific as it will be dependent on instrument
power and on the individual receiver. However, with this simple test, the
height threshold required can be easily found, thus allowing for the
saturated profiles to be removed and calibration to be correctly calculated.
Calibration coefficients for the Lufft CHM15k nimbus ceilometer
at Aberporth (52.06∘ N, 4.33∘ W). Each black cross
represents a single day, calculated from profiles deemed suitable by the
calibration algorithm. Panel (a) shows the mode of C for each individual day,
(b) shows the mean of C for each day, with the standard deviation shaded in
grey and (c) shows a 90 d running mean for the 12-month period. The average
of the daily modes is 0.46±0.05, the average of the daily means is
0.48±0.05 and the average of the 90 d running means is 0.48±0.02.
Calibrated results of the Lufft CHM15k ceilometers
The calibration algorithm can now be applied to the Lufft ceilometers in a
way similar to the calibration of the Vaisala ceilometers. A couple of
changes are included. The Lufft ceilometers have a range correction applied
to the full attenuated backscatter profile; therefore they are not
restricted by a change in processing at 2.4 km. The upper range limit of
integration to compute B is increased to 4 km, which incorporates the vast
majority of liquid clouds in the UK. Additionally, the higher cloud range
means the ceilometer beam must travel through a larger portion of the
atmosphere, so the ratio filter (criterion 1a) is increased from 5 % to
10 %. Note that this may lead to a slightly larger uncertainty in C, but
we choose to make this compromise in order to obtain a reasonable number of
calibration estimates per month. The lower height limit is also changed, so
that clouds below 1 km are not used. This is to avoid using profiles in the
region where an overlap correction is applied as there is a potential
temperature dependency in the overlap function that has not been accounted
for (Hervo et al., 2016). The lower height limit is often higher than 1 km,
however, due to the instrument-specific region of saturation.
At the 1064 nm wavelength there is no absorption by the water vapour
molecules, so no water vapour correction is required. Figure 12 shows an
example of the cloud calibration applied to a Lufft CHM15k ceilometer
situated at Aberporth, west Wales, for the 12 months of 2015. Because of
the requirement to remove the low-level clouds that resulted in saturation,
calibration was only possible on 70 d or about 20 % of the days. Each
day had an average of 58 profiles; nevertheless, the 90 d running mean
calibration coefficient over the year was 0.48 with a standard deviation of
0.02 % or 4 %, with no sign of any annual cycle. This standard deviation of
4 % over the year is slightly higher than for the Vaisala ceilometers,
probably because of the relaxation of the threshold required for aerosol to be
considered negligible, but is well within the specified requirement of
10 %.
Calibration coefficient for each of the CHM15k Nimbus ceilometers
in the UK Met Office network: 3 months of data (January–March 2016) have
been used for each instrument. The number of suitable calibration profiles
will be dependent on occurrences of cloud and therefore will vary for each
instrument. The box outline represents 50 % of the calibration profiles
and the whiskers extend to include 95 % of the profiles (outliers have
been excluded from plot). The horizontal red line in the box shows the
median calibration coefficient and the filled box shows the mean.
The calibration has been applied to the rest of the Lufft ceilometers in the
Met Office ceilometer network, as shown in Fig. 13. Most of the sites have a
relative calibration of less than 1.0; however, Coningsby has a particularly
large calibration coefficient. This highlights the importance and need for a
calibration of each instrument. For each site, the relative standard
deviation is small.
Collocated comparisons
The majority (9 out of 11) of the Lufft ceilometers are collocated with a
Vaisala ceilometer, allowing comparisons between the two types. Figure 14
compares the observations of attenuated backscatter from the two ceilometers
at Aberporth, which have both been calibrated using the cloud method. To
make a fair comparison between the two instruments, it is necessary to
choose the meteorological situation carefully. Aerosols are problematic,
because the ceilometers operate at different power, different detector
sensitivities and different wavelengths, and the backscatter from aerosols
is wavelength-dependent in a way that we do not know a priori. We could
analyse profiles in liquid clouds; however, we have already used these for
calibration (so it would not be a truly independent test). In addition, the
backscatter profile in liquid clouds contains very large gradients which
make any comparison extremely sensitive to small offsets in range and/or
differences in range-gating between the two instruments. Rain profiles could
potentially be used for the comparison: however, rain which reaches the
ground may wet the telescope optics and affect the data.
Quicklooks for the observed attenuated backscatter between 1 and
2 km are shown for the (a) Vaisala ceilometer and the (b) Lufft ceilometer.
Panel (c) shows 5 min averaged attenuated backscatter comparison for the
Lufft and Vaisala ceilometers situated at Aberporth for 22 April 2016 between
00:00 and 15:00 GMT. The colour scale indicates the number of data points. Vaisala data
have been corrected for water vapour attenuation and difference in wavelength
and have been interpolated to match the resolution of the Lufft ceilometer.
The black line shows the linear fit of the data and the dashed red line is
the 1:1 line.
Better targets for such comparisons are drizzle drops and ice particles.
Drizzle drops are large compared to the wavelength of the lidar and hence
the scattering is almost wavelength-independent for 1064 and 910 nm lidars
(since we are close to the geometric optics regime), as shown by Westbrook
et al. (2010a). At the same time, the extinction of the lidar beam is much
more gradual than in liquid cloud, providing smoothly varying backscatter
profiles, which can be interpolated onto a common grid with little error. If
we use an ice case, we would need to account for the influence of specular
reflections from oriented ice crystals (e.g. Westbrook et al., 2010b).
Therefore, in this example, a drizzle scene has been chosen. The code of
Hogan (2006) confirms that for drizzle, multiple scattering is negligible.
To establish quantitatively whether the backscatter for drizzle drops at
1064 and 910 nm are actually equal, Mie calculations were performed, assuming
Gamma drop size distributions (Westbrook et al., 2010a). The results show
that the backscatter at 1064 nm is very similar to that at 910 nm but
systematically smaller. The differences are very modest: between 5 % and 8 %
for median drop diameter in the range 0.1–0.6 mm, with most of the calculated
values in this range close to 7 %. Meanwhile the extinction is essentially
identical for both wavelengths. Thus, if the calibration has been successful
it would be expected that the backscatter profiles in drizzle would match
very closely. However, the Lufft is systematically 7 % smaller than the
Vaisala if no adjustment to account for the different wavelengths is made.
Therefore, for this comparison, the Vaisala attenuated backscatter data have
been reduced by 7 %.
It is also necessary to consider the various technical issues already
discussed earlier in the article when selecting profiles, in particular the
need for the drizzle to be high enough to be in the fully overlapped region
for the Lufft instrument, and below the 2.4 km height, above which the Vaisala
ceilometer data range correction is variable. Therefore, the data used
cover the period 00:00 to 15:00 GMT on 22 April 2016, during which
time there is drizzling cloud. The data are 10 min averages of attenuated
backscatter between 1.0 and 2.4 km and the Vaisala data have been regridded
from 20 m resolution to 15 m resolution using linear interpolation. The
quicklooks of the attenuated backscatter for the Vaisala and Lufft
ceilometers are shown in Fig. 14a and b.
Figure 14c shows a joint histogram of the attenuated backscatter measured by
the two instruments. If the calibration of both instruments has been
successful, we expect the data to lie around the 1:1 line (dashed red line),
and indeed our data do lie close to this line. Performing a linear
regression of the backscatter values from the two instruments, we find an
intercept very close to zero, a slope of 0.91 and a high correlation
coefficient of 0.95. This indicates that our calibration process has been
successful and that the combined errors in the calibrations from both
instruments together are less than 10 %. The spread of the individual data
points is rather larger than 10 % and can be accounted for by the
different resolutions and interpolation errors. This comparison of the two
different types of ceilometers confirms the reliability of this calibration
method – the two independently calibrated ceilometers, each with their own
challenges (e.g. water vapour, saturation), are consistent with each other.
This result is important for an operational network such as the Met Office
ceilometer network because it helps maintain a reliable, comparable stream
of calibrated data, with water vapour and saturation successfully accounted
for, from each instrument at each site.
Conclusion
In this paper, we have presented a robust algorithm to calibrate ceilometers
based on the cloud calibration technique that relies on the fact that the
lidar ratio of liquid water clouds is a known constant. This new method can
be run operationally, removing unsuitable profiles where the cloud does not
fully attenuate the ceilometer beam or where there is significant
backscatter from aerosols. By excluding profiles when the low cloud leads to
instrument saturation (particularly in the Lufft instruments) or when the
window transmission is low, and by accounting for the attenuation of the
ceilometer beam by water vapour (in the Vaisala instruments), we show that
ceilometers from different manufacturers can be successfully calibrated
using this method. Instrument malfunction can be identified by sudden
changes in calibration. If either the window transmission or the pulse
energy falls below 90 % of normal values, then clearly the instrument
sensitivity and the calibration will be different. When the window
transmission falls below 90 % the backscatter becomes noisy, probably due
to the inhomogeneous nature of the layer of dust or dirt on the window.
Consequently it is essential that the window is kept clean if reliable data
are to be obtained. If the pulse energy falls below 90 %, then it
should be possible to correct the backscatter signal, but the precise
accuracy of this technique remains to be determined.
It has been demonstrated that the running 90 d mean calibration
coefficient for each instrument over a year is constant to better than 5 %
with no detectable annual cycle. At the time of writing, profiles from 200
ceilometers from 17 countries are being distributed in near real time by the
E-Profile programme of European Meteorological Services Network (EUMETNET)
with the number expected to rise to about 700. E-Profile has decided to
calibrate the Vaisala ceilometers using the cloud calibration technique
described in this paper.
Data availability
The Met Office data are available via BADC CEDA at
http://data.ceda.ac.uk/badc/ukmo-nwp/data/ukv-grib (last access: 20 November 2018) and
http://www.ceda.ac.uk/blog/new-dataset-met-office-lidarnet-ceilometers (last access: 20 November 2018).
Author contributions
EH wrote the paper. All authors contributed to the scientific ideas in the
paper EH and CCP wrote the code to perform the calibration. CCP and SB
provided EH with access to observational data. All authors discussed the
results and edited the manuscript.
Competing interests
The authors declare that they have no conflict of
interest.
Acknowledgements
The
authors would like to acknowledge the contribution of the COST Action ES1303
(TOPROF) and in particular Ewan O'Connor and Maxime Hervo for many helpful
discussions. We would also like to thank Mariana Adam, Joelle Buxmann and
Jacqueline Sugier (Met Office) for their help and support during this PhD
project.
Financial support
This research has been supported by the NERC Industrial CASE (grant
no. F4027100).
Review statement
This paper was edited by Andrew Sayer and reviewed by two anonymous referees.
References
Brown, P. R. and Francis, P. N.: Improved measurements of the ice water
content in cirrus using a total-water probe, J. Atmos. Ocean. Tech., 12, 410–414,
1995.
Cossu, F., Hocke, K., Martynov, A., Martius, O., and Mätzler, C.:
Atmospheric water parameters measured by a ground-based microwave radiometer
and compared with the WRF model, Atmos. Sci. Lett., 16, 465–472, 2015.
Dupont, J. C., Haeffelin, M., Protat, A., Bouniol, D., Boyouk, N., and
Morille, Y.: Stratus–Fog Formation and Dissipation: A 6-Day Case Study,
Bound.-Lay. Meteorol., 143, 207–225, 2012.Hervo, M., Poltera, Y., and Haefele, A.: An empirical method to correct for
temperature-dependent variations in the overlap function of CHM15k
ceilometers, Atmos. Meas. Tech., 9, 2947–2959, 10.5194/amt-9-2947-2016, 2016.
Hogan, R. J.: Fast, approximate calculation of multiply scattered lidar
returns, Appl. Optics, 45, 5984–5992, 2006.
Hogan, R. J., Illingworth, A. J., O'Connor, E. J., and Baptista, J. P. V.
P.: Characteristics of mixed-phase clouds. II: A climatology from
ground-based lidar, Q. J. Roy. Meteor. Soc., 129, 2117–2134, 2003.
Illingworth, A. J., Cimini, D., Gaffard, C., Haeffelin, M., Lehmann, V.,
Löhnert, U., O'Connor E. J., and Ruffieux, D.: Exploiting Existing
Ground-Based Remote Sensing Networks to Improve High-Resolution Weather
Forecasts, B. Am. Meteorol. Soc., 96, 2107–2125, 2015.Illingworth, A. J., Cimini, D., Haefele, A., Haeffelin, M., Hervo, M., Kotthaus, S., Löhnert, U., Martinet, P., Mattis, I., O’Connor, E. J., and Potthast, R.: How can Existing Ground-Based Profiling Instruments Improve European Weather Forecasts?, B. Am. Meteorol. Soc., 100, 605–619, 10.1175/BAMS-D-17-0231.1, 2019Kotthaus, S., O'Connor, E., Münkel, C., Charlton-Perez, C., Haeffelin, M., Gabey, A. M.,
and Grimmond, C. S. B.: Recommendations for processing atmospheric attenuated
backscatter profiles from Vaisala CL31 ceilometers, Atmos. Meas. Tech., 9, 3769–3791, 10.5194/amt-9-3769-2016, 2016.Madonna, F., Amato, F., Vande Hey, J., and Pappalardo, G.: Ceilometer aerosol profiling
versus Raman lidar in the frame of the INTERACT campaign of
ACTRIS, Atmos. Meas. Tech., 8, 2207–2223, 10.5194/amt-8-2207-2015, 2015.
Markowicz, K. M., Flatau, P. J., Kardas, A. E., Remiszewska, J.,
Stelmaszczyk, K., and Woeste, L.: Ceilometer Retrieval of the Boundary Layer
Vertical Aerosol Extinction Structure, J. Atmos. Ocean. Tech., 25, 928–944, 2008.Martucci, G., Milroy, C., and O'Dowd, C. D.: Detection of Cloud-Base Height Using Jenoptik CHM15K and Vaisala CL31 Ceilometers, J. Atmos. Ocean. Tech., 27, 305–318, 10.1175/2009JTECHA1326.1, 2010.Münkel, C., Eresmaa, N., Räsänen, J., and Karppinen, A.:
Retrieval of mixing height and dust concentration with lidar ceilometer,
Bound.-Lay. Meteorol., 124, 117, 10.1007/s10546-006-9103-3, 2007.
Nicholls, S.: The dynamics of stratocumulus: Aircraft observations and
comparisons with a mixed layer model, Q. J. Roy. Meteor. Soc., 110, 783–820, 1984.
O'Connor, E. J., Illingworth, A. J., and Hogan, R. J.: A Technique for
Autocalibration of Cloud Lidar, J. Atmos. Ocean. Tech., 21, 777–786, 2004.Omar, A. H., Winker, D. M., Vaughan, M. A., Hu, Y., Trepte, C. R., Ferrare,
R. A., Lee, K., Hostetler, C. A., Kittaka, C., Rogers, R. R., Kuehn, R. E.,
and Liu, Z.: The CALIPSO Automated Aerosol Classification and Lidar Ratio
Selection Algorithm, J. Atmos. Ocean. Tech., 26, 1994–2014,
10.1175/2009JTECHA1231.1, 2009.OSCAR: Observing Systems Capability Analysis and Review Tool,
available at: https://www.wmo-sat.info/oscar, last access: 27 June 2018.
Pawlowska, H., Brenguier, J. L., and Burnet, F.: Microphysical properties of
stratocumulus clouds, Atmos. Res., 55, 15–33, 2000.
Pinnick, R. G., Jennings, S. J., Chýlek, P., Ham, C., and Grandy, W. T.:
Backscatter and extinction in water clouds, J. Geophys. Res.-Oceans, 88, 6787–6796,
1983.Protat, A., Bouniol, D., O'Connor, E. J., Klein Baltink, H., Verlinde, J.,
and Widener, K.: CloudSat as a Global Radar Calibrator, J. Atmos. Ocean. Tech., 28, 445–452,
10.1175/2010JTECHA1443.1, 2011.Tang, Y., Lean, H. W., and Bornemann, J.: The benefits of the Met Office
variable resolution NWP model for forecasting convection, Met. Apps., 20,
417–426,
10.1002/met.1300, 2013.Tsaknakis, G., Papayannis, A., Kokkalis, P., Amiridis, V., Kambezidis, H. D.,
Mamouri, R. E., Georgoussis, G., and Avdikos, G.: Inter-comparison of lidar and
ceilometer retrievals for aerosol and Planetary Boundary Layer profiling
over Athens, Greece, Atmos. Meas. Tech., 4, 1261–1273, 10.5194/amt-4-1261-2011, 2011.Westbrook, C. D., Hogan, R. J., O'Connor, E. J., and Illingworth, A. J.: Estimating drizzle
drop size and precipitation rate using two-colour lidar
measurements, Atmos. Meas. Tech., 3, 671–681, 10.5194/amt-3-671-2010, 2010a.Westbrook, C. D., Illingworth, A. J., O'Connor, E. J., and Hogan, R. J.:
Doppler lidar measurements of oriented planar ice crystals falling from
supercooled and glaciated layer clouds, Q. J. Roy. Meteor. Soc., 136, 260–276, 10.1002/qj.528,
2010b.
Whiteman, D. N.: Examination of the traditional Raman lidar technique. I.
Evaluating the temperature-dependent lidar equations, Appl. Optics, 42, 2571–2592,
2003.Wiegner, M., Madonna, F., Binietoglou, I., Forkel, R., Gasteiger, J., Geiß, A.,
Pappalardo, G., Schäfer, K., and Thomas, W.: What is the benefit of ceilometers for
aerosol remote sensing? An answer from EARLINET, Atmos. Meas. Tech., 7, 1979–1997, 10.5194/amt-7-1979-2014, 2014.Wiegner, M. and Gasteiger, J.: Correction of water vapor absorption for aerosol
remote sensing with ceilometers, Atmos. Meas. Tech., 8, 3971–3984, 10.5194/amt-8-3971-2015, 2015.
Wu, Y., Gan, C. M., Cordero, L., Gross, B., Moshary, F., and Ahmed, S.:
Calibration of the 1064 nm lidar channel using water phase and cirrus clouds,
Appl. Optics, 50, 3987–3999, 2011.