Chapter 3

METHODOLOGY

3.1 Cirrus Cloud Data Base using Ground-based Lidar

3.1.1 Background and Objectives

Long-term, highly stable, well-calibrated, precise and accurate data are pre-
requisite for climate studies. Thus, the raw (primary) data must undergo through
stringent quality checks before deriving parameters. As mentioned in Chapter 2, the
Rayleigh-Mie lidar i s operated to obtain backscattered signals from troposphere to
mesosphere through two different receivers, one for temperature measurement
(Rayleigh receiver) and the other to measure aerosol and cloud properties (Mie
receiver) since 1998. Over a period of 16 years (1998-2013) 93,690 individual
backscattered photon profiles are recorded. One needs an automated algorithm to
classify such a large number of profiles into clear, cloudy and noisy profiles. When
the profiles are classified using an automated algorithm, it makes it possible to
reproduce the results by others - a foremost requirement for any scientific study. The
laser beam attenuation caused due to the occasional presence of low or middle level
clouds prevents the laser beam to reach cirrus altitudes. Thus, the profiles affected
with such clouds have to be removed from the data set.
In the following sections of this chapter, the methods used to create a 16 year
(1998-2013) cirrus cloud data base using Mie lidar system are described. Overall the
method involved pre-processing of data (Section 3.1.2), cirrus cloud detection
(Section 3.2), and the retrieval of optical properties of cirrus clouds (Section 3.3). The
CALIOP Level 2 algorithms for the cloud detection and the retrieval of extinction
coefficient from CALlOP backscatter profiles are also discussed in sections 3.4 and
3.5, respectively. The definitions of the TTL characteristics (like cold-point
tropopause, lapse-rate tropopause, convective-outflow level and ITL thickness) and
methods used for their estimation from radiosonde temperature profiles are presented
in Section 3.6.

3.1.2 Detection of faulty and noisy profiles from the Mie-Lidar data set
It i s very essential to filter out faulty, poor-quality profiles from a multi-year
data set. Individual profiles obtained by lidar system are 4 minute averaged raw
backscattered photon counts. These profiles are labelled as level 1 profile d&. The
Chapter 3 Methodologl

criterion for the elimination of these noisy profiles is based on a preset threshold value
of 500 photon counts at 5 km altitude bin. All the lidar profiles having photon counts
less than 500 at 5 km altitude bin are discarded. Figure 3.1 illustrates an example for
the elimination of noisy profiles from the lidar observations acquired on February 18,
2007 during 0058 IST to 06:08 IST.

Fiure 3.1 A @pica1example showing the elimination of noisyprofilesfi.om the lidarprofiles

acquired on Feb 18, 2007from 00:58 IST to 06.48 IST. Profiles represented by red colour
are the eliminated ones while those represented by purple colour are the selected profiles.

Each dot represents photon counts at a given altitude bin. Noisy profiles are
represented by red colour while the profiles with good signal to noise ratio are
represented by purple colour. The threshold value of photon counts was arrived at
using the frequency distribution of photon counts at 5 km altitude bin. It should be
notes that the threshold value is system specific and it will be different for different
system. The database of raw photo count profiles obtained after removing noisy
profiles is named as level 2 profile data. The level 2 data set is then used to identify
cloudy profiles and detect their base and top altitudes using WCT method as
described in the next section.
Chapter 3

3.2 Cirrus Cloud detection

3.2.1 Cloud Detection Methods
Cloud detection or identification of cloud boundaries (cloud-base and top
altitudes) is essential for cloud classification into low, middle or high level clouds and
waterlice-clouds. The height (or temperature) of a cloud layer controls the radiative
balance of the Earth-Atmosphere system in a fundamental way [Lynch, 20021 and
hence cloud-boundaries are important variables for climate studies. The presence of
cloud in a lidar profile is manifested as significant enhancement in the backscattered
signal above the background scattering level of the ambient atmosphere. An
enhancement in the background scattering level can also arise from the co-existence
of aerosols, dust layers, smoke plumes or haze layer [Pal et a/., 19951. The above-
mentioned ambient conditions are very common in the lower troposphere. The
detection of a cloud layer in such ambient conditions is tedious and requires
additional available information (like relative humidity) for unambiguous detection
and classification. However, cirrus clouds which generally occur in the free
troposphere are easier to detect except under conditions perturbed by a volcanic
eruption. There are numerous methods available in the literature, which were initially
designed for atmospheric boundary layer determination but later found their
usefulness in cloud detection as well. Some of these methods are: gradient method
[F/amant el al., 1997; Menut et al.,19991 which uses first or second order derivative
of the lidar signal with respect to height, variance method [Piironen and Eloranfa,
1995; Menu1 et al., 19991 uses variance of the signal; scattering ratio method involves
a priori calculation of scattering ratio using lidar inversion algorithm (described in
subsequent section). These methods either use raw signal or range-corrected signal
(described in subsequent section) to calculate quantity of interest (derivative or
variance or scattering ratio). In all these different methods, a region of cloud layer is
said to be detected when the quantity of interest exceeds a pre-defined threshold for
sufficient number of consecutive range bins. Since, a significant level of noise is
superimposed on the lidar signal; the derivative method is not suitable for cloud
detection [Young, 1995; Kovalev and Eichinger. 20041. Scattering ratio method is also
not suitable as it requires the knowledge of cloud boundaries in advance; in addition
to it the scattering ratio calculation involves a lot of assumptions and computational
efforts. Apart from these methods, the detection of cirrus cloud layers is also done by
curve fitting methods. For instance, Sassen and Cho [I9921 used a method to d
41 hi
AM la
Chapter 3 Methodology

the layers of thin or sub-visible cirrus cloud by fitting lidar data with the molecular
backscatter profile calculated from temperature and pressure data obtained
independently using radiosonde observations. A cloud layer was identified whenever
the backscattering signal exceeded the molecular backscattering level. Young I19951
used an algorithm for the detection of optically thin clouds which determines the
cloud base and top altitudes by extrapoletirig and fitting the reference cloud-fm
signal (signal below and above cloud Layer) to the measured signal. A method by
Clothiuta et al., [I9981 identifies a cloud by comparing a constntcttd clear-sky lidar
profile with cloudy profile and uses a pn-deftned threshold estimated from the least
square e m of signals acquired on many clear-sky days. The curve-fitting methods
work well for single-layered optically thin clouds but gives problem for multi-layered
cirrus clouds. Thii k becaw intha case of multi-layered clouds the cloud-he region
(number of altitude bins) between the bdjaccnt layers is not sufficient for curve fitting.
In the pmsent study, a aew advanced matkmatical tacrhniuc, named Wavelet
Covarianoe i'Pm&m m*) method hadl bean wed. 'Ibis madrod dobs aot mquin,
any other infamution bafomhmd, wcmp thC M ~ ~ & M I,WCT
, mddsOlrJ ir h p l 8
and very sensitive to the d W m of opCically thin d,mkl.llryersddW#8 h
noisy signal. Recently, this method has bem used by r > I ~eii d, [U)133 M
classifying various types of cirrus clouds observed using tidar data over W.3hc
present study utilizes this method for the detection of tmpical cirrus clouds the details
of which are described below.

3.2.2 Wavelet Covariance Transform method

The base and top altitudes of cirrus cloud layers are determined using Wavelet
Covariance Transform (WCT) method [Brooks, 2003; Haij et al. 20071,Originally,
this method was developed to identify boundary layer top, and later its application has
been extended to detect optically thin clouds [Baars et al. 2008 and Dionisi et al.
20131. The WCT of lidar backscatter raw profile P (2) is defined as
-
1 &' z-b
a JP(~I~(-)&,
P, bj = -
a
Zh

where, z is altitude, za and 4 are lower and upper limits of the profile, h(?) is a

step function known as Haarjknction defined as,

b is called translalion and a is called dila~ionof Haar function. The translation h
locates the centre of the step while the dilation a gives the width of the step. An
appropriate selection of a is essential for the correct cloud detection. AAer going
through several profiles and experimenting with various values. a = 3.0 was chosen.
Figure 3.2 shows a typical example for the application of this method in the detection

2008-11-20 00:22 IST 2008-11-20 00:22 IST

loa 10' ion 10' -0.2 -0.1 0.0 0.1 0.2 0.3
Photon Counts WCT

Figure 3.2 Illustration of the application of WC'T method on u lidur profile observed on 20
Novemhrr ZOO8 ut OO:Z2 IST. (a) Raw profile with multi-luyercd clouds. (b) WCl'prnJilr
corresponding to the raw prnjle. Single dotted doshed mugentu linm represent the top
altitude while dushed blue lines represent the huse u1titudv.v nfc1od.s.

of multi-layered cloud layers. The base of a cloud layer corresponds to the altitude
where Wp(a,b) becomes negative showing a local minimum with its value less than a
pre-defined threshold. While the top of the cloud layer corresponds to a positive value
of W p (a,b) greater than the threshold value exhibiting a local maximum. An
appropriate value of threshold for cloud detection was selected after inspecting its
performance on many cloudy profiles. A judicious selection of threshold is essential
Chapter 3 Methodology

otherwise a large value o f threshold may leave many profiles o f thin c i m s clouds
undetected whereas a small value o f threshold may lead to false detection o f noisy
profiles as cloudy. We used a threshold value that varies linearly as a function o f
altitude. This approach takes benefit o f low noise in near range and avoids false
detection at far end due to high noise. Further to avoid false detection, if raw photon
counts at cloud layer are not greater than mean background plus three times the
standard deviation then those profiles are excluded. Cloud base and top heights of five
different layers can be obtained very accurately using this method. The lowest
geometrical thickness of cirrus cloud layer that NARL lidar can detect is 600 m
example for which is shown in Figure 3.3. Even a slight variation in the photon counts

2008-09-1323:16 IST
- - - .
,

Figure 3.3 An example ofgromefricullv and op/icully /hin cloudohserved on 13 September

2008 a/ 23:16 IST. (a) and (b)are same as Figure 3.2 (a) and (hl hut,fi)rl3Sep/emher 2008.

is reflected in the WCT profile as demonstrated in Figure 3.3 (e.g. at 16.5 krn),
thereby making WOT method highly sensitive to thin cloud like sub-visible cirrus.
The application o f WCT algorithm on level 2 lidar profile data for each year yielded
cloud base and top altitudes which are saved as level Icloud layer data file. Due to
the large atmospheric variability, different cloudy situations like multi-layered clouds,
thin clouds above a convective anvil etc., are acquired in the lidar profiles, llnder
these different cloudy scenes, the automated cloud detection algorithm which uses
same criterion for cloud detection may fail sometimes leading to false cloud detection.
Chapter 3 Methaiologv

Therefore, cloud base and top altitudes determined in level 1 cloud layer data files
were manually rechecked to ensure that the method has worked consistently over the
complete data-set. Wrongly detected cloud base and top altitudes are corrected
manually and then saved to get level 2 cloud layer datafile.

Figure 3.4 Steps involved in the generation ~fcloud layer data set.

Algorithmic determination of cloud ensures that our findings are reproducible,

whereas manual inspection ensures that we made judicious choice of threshold values.
Figure 3.4 illustrates steps involved in the process of creation of cloud layer data set
base starting from the elimination of noisy and faulty lidar profiles to the generation
of level 2 cloud layer data tile. The level-2 cloud layer data set contains cloud base
height, top height, date, and time information.
Chapter 3 Me!hodologv

3.2.3 Criterion used for discriminating cirrus clouds from other clouds

Cirrus clouds usually occur at temperatures below -20 'C [Liou, 19861 and at
altitudes above 8 km [Lynch et a/., 20021. To distinguish cirrus cloud layers from
those of low and middle level clouds, a temperature threshold value of -20 'C has
been chosen. Only those cloud layers with a base temperature below this threshold
value are considered as cirrus cloud layers. The temperature of -20 'C corresponds to
an altitude of about 8.4 km as obtained from the long-term (2006-2013) temperature
measurements using radiosonde over Gadanki (shown in Figure 3.10). The base
temperature of cirrus cloud layers (observed using Mie-Lidar) during the period 1999-
201 3 are obtained from NCEP FNL temperature profile data over Gadanki.

3.3 Retrieval of Extinction coefficient from Lidar data: Kaestner's inversion

algori thrn

In the previous chapter, single-scattering elastic lidar equation (Fquation. 2.1) was
discussed which has two unknown parameters known as backscattering coefficient (m
and extinction coefficient (a). Both of these coefficients are pre-requisite for deriving
optical properties of aerosol and cirrus cloud layers like transmission, depolarization
ratio, backscattering ratio, and most-importantly optical thickness. Optical thickness
( I , , ~ of
~ ~a) cloud layer is a measure of the amount of light extinguished due to
scattering and absorption of light by the cloud particles and is mathematically
expressed as:

'cloud = [a(r)dz
Here, a(z) is the extinction coefficient as a function of altitude, zb and z, are
base and top altitudes of a cloud layer. Optical thickness gives information about
density of clouds. Higher the optical thickness, denser the cloud is. To estimate the
optical thickness of cirrus cloud layers it is essential to retrieve extinction coefficient
profile from the lidar backscattered photon counts. A plethora of methods are
available in the literature for the retrieval of extinction coefficient. However, the most
popular ones are described by Fernald et al. [1972], Klett [1981], Fernald [1984],
Klett [1985], and Kaestner [1986]. The method proposed by Kaestner (19861
explicitly derives the extinction coeficient from the lidar equation unlike the methods
Chapter 3 Methodology

by Fernald and Klea which derive backscattering coefficient and hence Kaestner's
method has been chosen in this study for the retrieval of extinction coefficient.
Replacing the range (r) by altitude (z'), considering the full overlap of the laser and
telescope FoVs (i.e. O(z Y = I ) and after rearranging the Lidar equation (Equation 2.1)
can be rewritten as

This form of lidar equation is known as noise and range-corrected lidar equation.
CT
After substituting z' (P(z7 - P,,,,, ]= P1(zl), C. Pn,A.- = C ' and taking natural
2
logarithms on both sides, the above equation takes the form of

Differentiating equation (3.5) with respect to z yields

U(Z)and P(z) are two unknowns which have to be derived from single lidar equation.
This problem is simplified by assuming a linear relation between a(z) and ,8(z) given
as

where, S(z) is known as lidar ratio which is expressed in the units of sr. Lidar ratio
depends on the size and shape distribution of particles. Both, a(z) and P(z) can be
expressed as a summation of contributions from molecules and particulate. So,

where, subscript " M denotes molecular contribution while the subscript "P" is for the
contribution from particulates. SM (= 8 d 3 ) and Spare the lidar ratios for molecules
47
Chapter 3 Methodology

and particulates, respectively. By multiplying equation (3.6) by SPJ?(') from both

sides and replacing a(z) and P(z) using relations (3.8) and (3.9), an inhomogeneous
da,.
Riccati diferentiai equation in - can be obtained which has the form:
dz

where,

d 1 dS,,
MI) = -[1n(z'.p)]+2.~~.(.\, +s,)+--.-,
dz S,, dz

'The stable solution of equation (3.10) for the particulate extinction coefficient (a,) is
given as

Z(z)
(z) = -Sf, (z).& (z) + -
a,.
N(z)

where,
I:
Z(z) = Sl,.(z).z2.P(z).exp 2.J[sI,(zt)- SM (3.14)

z,, is the reference altitude which is taken as 25 km in this study. Retrieval begins
from the reference altitude to successive altitude bins. Pw (z) is the molecular
backscattering coefficient which is calculated from the number density of air
molecules @), by using the equation (3.16):

p, o - " 550
(z) = 5 . 4 5 ~ 1 (~) .p,,(z)

where, h is the lidar wavelength which is 532 nm. pw (2) is expressed in the units of
km-'si'. Number density of air molecules @d is calculated by using ideal gas law
Chapter 3 Methodology

and pressure-temperature @-7') profiles. pa,, in units of number per cubic centimetres
is expressed as

where, NA = 6.023 X 10" molecules per mol, Avogadro 'snumber,

Rd = 287.0 J K-I kg',universal gas constant for dry air and

Md = 0.02896 kg mol-I, mass of dry air.

Input parameters like p-T profiles, lidar ratio and multiple-scattering factor are
the pre-requisites for the estimation of extinction coefficient profiles using NARL
lidar data. The selection of these parameters plays an important role in the reliable
estimation of extinction coefficient. The selection procedure of these parameters is
described in the subsequent sub-sections.

3.3.1 Selection ofp-T profiles

The best choice for the p-T profiles is the p-T profiles obtained from the
radiosonde observation over the lidar site. However, radiosonde observations are
expensive and are not carried out frequently throughout the lidar operation period
except during some field campaigns. In addition to this, many-times radiosonde
observations are limited to the altitudes below the chosen reference altitude (2,) due
to the early burst of the balloon. Due to the above-mentioned limitations and
unavailability of the daily radiosonde observations over Gadanki during the period
from 1998-2005, 6 hourly NCEP-FNL p-T profiles (available since 1999) are used for
the calculation of molecular backscattering coefficient (PM).Figure 3.5 shows an
comparison of NCEP-FNL and radiosonde p-T profile for a typical day at Gadanki. In
general, a good agreement is found between the two for the altitude ranges that matter
for cirrus clouds. For the year 1998 when no NCEP-FNL data were available,
monthly mean temperature profiles are used for the estimation of the molecular
backscattering coefficient.
 -100-80-80-40-20 0 20 40
Temperature (C)
Figure 3.5 Comparison o f NCEP-FNL (1 ' x 1 " grid, shown with red dashed curve)
temperature proJiles with a radiosonde temperature profile (shown with dashed red curve)
over Gadanki on OR August 2008 at 12 UTC (1 7:30 IST).

3.3.2 Selection of Lidar ratio for cirrus clouds

The lidar ratio greatly depends on the size and shape distributions of ice-
crystals present in the cirrus clouds. Due to the large variability in the ice-crystal size
and shape distributions the value of lidar ratio varies from 10 sr to 60 sr [Chen et al.,
20021. Since, one of the objectives of this study is to compare the extinction
coefficient obtained from Mie lidar with that of CALIOP, a similar range-independent
lidar ratio of 25 sr has been chosen following CALIOP's extinction retrieval
algorithm. A lidar ratio of 40 sr is selected for the aerosols in the cloud free regions
below and above the cirrus cloud layers. Lidar ratio inside the cirrus cloud layer is
assumed to be constant (25 sr).
Chapter 3

3.3.3 Selection of multiple-scattering correction factor (q)

The retrieval algorithm discussed in the previous section accounts only for
single-scattered photons. Multiple-scattering effects must be accounted when back-
scattering signal from clouds especially thick cirrus clouds are considered. The effect
of multiple-scattering which is a function of laser power, cloud range (or height),
receiver

'74 Level 2 Profile data

Noise and Range

Correction

Number density
( calculation from I

Calculation of Rayleigh
backscattering
9 = 0.75, SM= 8x13 sr,
z,= 25 krn
Calculation of extinction
1 coefficient using 1
Kaestner's method
u
Calculation of optical
I thickness of cirrus cloud (
I lavers I

Figure 3.6 Steps involved in the retrieval of exfinction coefficient und the calculation of
optical thickness.

FoV, size and shapes of ice-crystals [Eloranta, 19981 cannot be neglected in the
measurement of cirrus cloud properties using a lidar with a receiver FoV of 1 mrad.
Several studies [Chen el al., 2002; Chepfer eet al., 1999; Hogan, 2006; Sassen and
Cho, 1992; Sassen and Cornstock, 20011 have suggested different values of multiple-
scattering correction factor ('1) ranging from 0.1 to 0.9 based on different crystal
Chapter 3 Methodology

habits and optical properties of cirrus clouds. In this study, the effect of multiple-
scattering is taken care of by assuming q = 0.75 following Sassen and Cho 119921 and
Sassen and Cornstock [2001]. Sassen and Cho [I9921 used telescope with FoV of 3
mrad which is comparable to our telescope. Sassen and Cornsrock [2001] used three
different values of q depending on cloud type which are 0.6 to 0.7 for thick clouds,
0.8 for thin clouds and 0.9 for sub-visible cirrus cloud. A single intermediate value, 11
= 0.75 for all the cloud types has been selected in this study. An effective lidar ratio
obtained from the product, q.S (0.75 x 25 sr = 18.75 sr) has been used in the retrieval
of extinction coefficient for cirrus cloud layers.
The entire process involved in retrieval of extinction coefficient using
Kaestner's method and the calculation of optical thickness from lidar profiles (level 2
profile data) is summarized in Figure 3.6.

3.4 CALIOP Level 2 Algorithms

In the present study, 5 km CALIOP level 2 cloud layer and cloud profile data
products have been used as mentioned earlier in Chapter 2 (Section 2.2.3). The
nomenclature Level 2 for CALIOP data products refers to derived geophysical
parameters using Level 1 data. Whereas the. nomenclature Level 1 refers to three-
dimensional geo-location and calibration [Winker et al., 20091 of Level 0 data. Geo-
location is the process of finding geodetic location of CALIOP footprint by using the
CALIPSO spacecraft altitude and ephemeris data. The accuracies of geo-location in
horizontal and vertical directions are about 50 m and 10 m, respectively. The
calibration of CALIOP 532 nm parallel channel night-time profiles involves
normalization of the measured signal to the predicted molecular signal in the region
between 30 and 34.2 km [Powell el al., 20091. The predicted molecular signal profile
is obtained from molecular number density by using gridded Global Modelling and
Assimilation Ofice (GMAO) meteorological profiles of pressure, temperature and
ozone along the CALIOP track. In order to avoid the effects of solar background
noise, the calibration coefficients during daytime are obtained by interpolating the
adjacent night-lime calibrations. The 532 nm perpendicular channel profiles are
calibrated relative to the 532 nm parallel profiles with the help of a depolarizer as
shown in Figure 2.8 (Chapter 2). The Level 1 data processing generates calibrated
profiles of 532 nm perpendicular attenuated backscattering coefficients and total
attenuated backscattering coefficients at 532 and 1064 nm [Winker et al., 20091. The
52
Chapter 3 Methodology

definition of these and other derived parameters (scattering ratio, depolarization ratio
and colour ratio are given in Appendix I. The CALIOP Level 2 data processing is an
interactive multi-stage process which involves following three primary modules:
I . Identification of aerosol and cloud layers in the 532 nm attenuated
backscatter profiles by a set of algorithms called Selective Iterated
Boundary Location (SIBYL),

2. Classification of aerosol and cloud layers into their different types by

Scene ClassrficutionAlgorithms (SCA),

3. Retrieval of particle backscatter and extinction coefficient profiles by

the Hybrid Extinction Retrieval Algorithm (HERA).

An overview of these algorithms is given in Winker el al. [2009]. Each of these

modules is described briefly in the subsequent section.

3.4.1 Selective Iterated Boundary Location (SIBYL)algorithm:

CALlOP fundamentally measures the range-resolved profiles of backscatter
intensity at a native horizontal resolution of 333 m in which the signals from a variety
of geophysical entities like clear air, aerosol layers, cloud layers and the earth's
surface are embedded [Powell el al., 20091. These geophysical entities are CALlOP
targets and are called "features". A feature is defined as any extended, vertically
contiguous region of enhanced backscatter signal which rises significantly above the
expected clear air value, i.e. signal from a purely molecular atmosphere [Vaughan et
al., 2009; Young and Vaughan, 20091. Figure 3.7 (a) illustrates an example of these
features in a CALlOP 5 km horizontal average profile of 532 nm attenuated
backscatter coefficients observed on 13 August 2006 at 17:14:18 UTC over the Yellow
Sea (36.IsN, 126.6 'E). Along its orbit track, CALlOP provides succession of such
profiles constituting a two-dimensional atmospheric section as a function of height
called "CALIPSO curtain". An example of CALIPSO curtain is shown in Figure 3.7
(b). A section of CALlPSO curtain consisting of 16 consecutive protiles (each of 5
km horizontal resolution) of attenuated backscattering coefficient extending 80 km
along CALlOP ground track containing a number of features between the surface and
an altitude of 30 km is called a "CALIPSO scene". For the distinct identification of
features in a CALIPSO scene characterized by backscattering intensities of different
Figure 3.7 (a) An example of CALIOP 5km horizontal average profile of 532 nm attenuated
backscatter coefficient measured over Yellow Sea (36.1 W, 126.6 XE) on 13 Augur1 2006 a1
17:14:18 UTC. (b) CALIOP curtain as observed during the same overpass (Adapted from
Figure 1 of Vaughan el al. [2009U

1-r1-J Apply Attenuation

Figure 3.8 Flow diagram of SIBYL algorithm ,for multi-resolution layer detection. The
variable K = I , 2, 3 correspond^ to horizontal resolution of 5, 20 and 80 km respectively
(Adapted.fiom Figure 5 of Vaughan et a/. [2009U.

orders of magnitude, a Selective Iterated Boundary Location (SIBYL) algorithm was

developed, details of which are presented in Vaughan el nl., [2009]. SIBYL algorithm
mainly consists of two modules: an averaging engine and a profile scanning engine as
shown in Figure 3.8. The purpose of the averaging engine is to increase the signal to
Chapter 3 Methodology

noise ratio by averaging the profiles at different successive horizontal resolutions (5,
20 and 80 km) in order to achieve distinct identification of features in the CALIOP
scenes. The amount of averaging depends on the scattering strength of the feature.
Stronger features like boundary layer clouds are detected in single laser shot (333 m)
while weaker features such as sub-visible cirrus clouds require significant averaging
at coarser horizontal resolutions for their detection. The profile scanning engine
locates the boundaries of the features present in the averaged Level 1 532 nm total
attenuated backscatter profiles of a given CALIPSO scene by using an adaptive
threshold technique. This technique is adaptive because the selection of threshold
depends on the characteristics of the signal and noise associated with it. The threshold
is computed for each profile and its magnitude (which is above the molecular signal)
varies with the altitude. The selection of threshold is optimized in a judicious manner
which maximizes the possibility of feature detection while simultaneously minimizing
the false detections due to noise [Vaughanel al., 20091. The thresholding algorithm is
applied to the Level 1 attenuated scattering ratio profile. SIBYL first detects the top
layer followed by underlying layers in a 5 km horizontally averaged profile. Once the
features are detected at this stage, they are further detected at finer horizontal
resolutions of 1 and 113 km (not shown in the flow diagram). This is followed by the
removal of backscatter data between the detected features from the profile and then
the data below the detected features are corrected for the attenuation caused due to the
top layers. To further detect the fainter features in the same scene, corrected profiles
are averaged to a new coarser horizontal resolution (20 km) and an updated threshold
based on this averaged profile is used for layer detection. This process is iterated till
the profile scanning is performed at 80 km. The maximum number of layers that
SIBYI, can detect in a single 5km resolution profile is ten. Once the features are
found out, the mean values of the attenuated backscatter (at 532 and 1064 nm),
attenuated colour ratio, and volume depolarization ratio for each feature are computed
and are stored along with physical properties (base, top and mid-altitude), latitude,
longitude, time etc., in the CALIOP Level 2 layer products. Based on these physical
and optical properties, the classification of each feature into cloud-aerosol, aerosol
types and ice-water phase is performed by Scene Classification Algorithm as
described in the next sub-section.
Chapter 3

3.4.2 Scene Classification Algorithms (SCA)

The classification of the features (present in a CALIPSO scene) detected by

SIBYL algorithm into cloud and aerosol layers is based on the differences in the
physical (layer height) and optical properties (layer mean attenuated backscatter and
layer mean attenuated total colour ratio) of aerosols and clouds. This is done by a set
of three algorithms which are collectively known as Scene ClassificalionAlgorithms.
A Cloud-Aerosol Discrimination (CAD) algorithm identifies and discriminates the
aerosols and clouds in a profile data by using a confidence function defined in terms
of the multi-dimensional Probability Distribution Functions (PDFs) of the layer-
averaged properties [Liu el al., 20091. CAD algorithm assigns a signed floating point
number called CAD score to each of the detected features. The value of CAD score
lies between -100 and +loo. A feature with positive CAD score is classified as a
cloud layer while a feature with negative CAD score is an aerosol layer. The
magnitude of CAD score quantifies the confidence of cloud-aerosol discrimination.
The present study considers only those cloud layers for which CAD score lies
between 80 and 100. If the feature is identified as an aerosol layer, SCA further
determines the type of aerosol layer (dust, smoke, clean continental, polluted
continental, clean marine and polluted dust) by using the AERONET (AErosol
Robotic NETwork) sun-photometer based aerosol models [Omar el al., 20051, layer-
averaged depolarization ratio and 532 nm attenuated backscatter obtained from
CALIOP [Omar er al., 20091. Depolarization ratio is used to identify dust layers.
Based on the aerosol type, a specific lidar ratio is assigned to it for the retrieval of
aerosol extinction coefficient by HERA. However, if the feature is identified as a
cloud layer, the SCA further determines the phase (water or ice) of the cloud based on
the magnitude of volume depolarization ratio (61,) and layer-integrated backscatter
signal. The volume depolarization ratio typically greater than 20% are associated with
the scattering from cirrus clouds [Sassen and Benson, 2001). Stronger backscatter
signal and smaller depolarization value exhibited by water clouds are basis for their
discrimination from ice-clouds. The depolarization ratio is key parameter used in the
discrimination of ice-water phase as well as identification of dust aerosol type. The
value of depolarization ratio is sometimes affected due to the multiple scattering from
the dense water clouds, noisy signal from weaker layers and horizontally ariented
planar ice-crystals which pose difficulty in the unambiguous discrimination of ice-
Chapter 3 Methodology

water phase. In such case, a joint probability function o f depolarization and cloud-top
temperature is used [Hu el al., 20091. To circumvent the effects of the horizontally
oriented planar ice-crystals exhibiting low depolarization values due to the specular
reflection o f near-nadir pointing lidar pulses, the CALIOP nadir angle was increased
from 0.3" to 3" in November 2007.

3.4.3 Hybrid Extinction Retrieval Algorithm (HERA)

HERA is an iterative algorithm which retrieves the profiles of particle
backscattering and extinction coefficients by utilizing the results from preceding steps
involving SIBYL and SCA [Winker el al., 2009; Young and Vaughan, 20001. The
interaction o f HERA with other Level 2 algorithms and generation o f level 2

Figure 3.9 Flow char! showing the interaction ofHERA with SIBYL and SCA in the retrieval
ofextinction coefficient profiles (Source: Young and Vaughan et al., [20091).

aerosol/cloud layer products are shown in Figure 3.9. The basis of this algorithm i s to
solve the two components (molecules and panicles) lidar equation for the unknown
particulate extinction and backscatter coefficient either by using a priory lidar ratio
(unconstrained retrievals) or using layer two-way transmittance as a boundary
condition (constrained retrievals). SCA identifies the layers for which two-way
transmittance at 532 nm can be reliably estimated from the clear-air regionjust above
Chapter 3 Melhodology

and below the layers. By using the method suggested by Young [1995], SIBYL
estimates an initial lidar ratio at 532 nm from the layer transmittance apan from
providing layer boundaries. In case of the features for which the transmittance
measurements are impossible, SCA assigns initial lidar ratios at 532 nm and 1064 nm
for those features. This initial lidar ratio is used for the retrieval of extinction
coefficient from the backscattering coeficient using equation 3.7. In CALIOP version
2 data processing algorithms, the lidar ratio used for water clouds is 18 sr while that
for ice-clouds is 25 sr. The value of 25 sr is the mean of lidar ratio for ice-clouds
obtained from the ground-based lidars [Sassen and Cornstock, 20011. The effects of
multiple scattering due to cirrus cloud layers have been corrected by a using a
constant multiple scattering correction factor of 0.6 while the multiple scattering
effects due to aerosols have been neglected. The background noise for the CALIOP is
estimated from the measurement in the altitude range from 112 to 97 km, a region
which has negligibly small molecular contribution to the total backscatter. The
profiles of molecular number density and ozone absorption coefficient are estimated
from the pressure, temperature and ozone profiles provided by GMAO along the
CALIOP track. HERA consists of two modules one the profile solver and the second
the extinction averaging engine. After the removal of background noise and
instrument offset from the lidar signal, the profile solver calculates the two-way
particulate transmittance using two methods: (a) Constrained retrievals and (b)
Unconstrained retrievals. In constrained retrievals, the solution of extinction
coefficients are constrained by the two-way transmittance supplied by SIBYL by
iteratively adjusting the particulate lidar ratio until the difference between retrieved
particulate two-way transmittance and a supplied constraint is less than a predefined
tolerance parameter. However, the unconstrained retrievals involve the solution of
particulate backscattering coefficient by forward integration method. The retrieval is
performed first for the topmost layer and then proceeds further to successive range
bins. The signals from the lower features are corrected for the attenuation caused by
the higher layers. Incorrect estimates of lidar ratio, multiple scattering function or
attenuation correction due to overlying layers may lead to divergent solution of
backscattering coefficient. Since the uncertainties in the lidar ratio is largest and the
retrieval is usually most sensitive to adjustments in lidar ratio, the algorithm
iteratively adjusts the value of lidar ratio relative to the initial lidar ratio provided by
SCA until a stable and convergent solution is obtained. More details of HERA and
58
Chapter 3 Methodology

illustration for its application on CALIPSO scenes can be found in Young and
Vaughan [2009]. The details on the uncertainty and errors involved in the extinction
coefficient retrieval are provided in Young el al. 120131.

3.5 Determination of Tropical Tropopause Layer (TTL) Characteristics

The I T L characteristics over Gadanki are obtained using temperature profiles
from routine GPS radiosonde launches. A typical temperature profile obtained over
Gadanki on 8 October 2009 is shown in Figure 3.1 0 (a) along with a lidar profile in

10' 10' 10' 10' 10" -0.4 -0.2 0.0 0.2 0.4 -100 .80 -80 4 0 -20 0
Raw photon counb, WCT 1emperalure (C)

Figure 3.10 (a) Typical example qfraw lidar signal in the presence ( f a cirrus cloud layer
observed on 8 Ocmber 2009 between 19:22 IST and 19:26 IST. Temperature profile observed
at Gadanki on the same day ohiained.from GPS radiosonde launched at 17:30 IST is shown
by red line with its axis on the top. (IPZ,LRZ and C(IZ are respectively shown by pink solid
line. green triple dotted dashed line and brown dushed line. (b) WCT corresponding to the
raw profile along with the detected buse (blue da~hedline) and top alfi~udes(magenta single
dotted dashed line) ofihe cirrus cloud layer. (c) The clima~ologicalmean temperatureprqfile
(shown hy red line) over G'adanki ohiainedfrom radiosonde observations during 20062013
with corresponding CPZ, LRZ and COZ shown by d~fferenthorizontal lines.

presence o f cirrus cloud layer obtained on the same day during 19:22 to 19:26 IST.
The altitudes o f cold point tropopause (CPZ) and maximum convective outflow
(COZ) are identified as altitudes o f the coldest point below 20 km and minimum
potential temperature gradient, respectively. In Fig. 3.10 (a), CPZ and COZ are
located at 17.3 km and 1 1.1 km, respectively, and shown with pink and dashed brown
lines. Lapse rate tropopause altitude (LRZ) is also calculated which according to
Chapter 3 Methodology

World Meteorological Organization [WMO,19571 is defined as the level at which

lapse rate becomes -2 Kfkm, provided the average lapse rate from this level to next
two kilometres i s less than 2 K h . The LRZ for the example in Figure 3.10 (a)
coincides with the cold-point tropopause altitude (17.3 km). TTL thickness for each
temperature profile is also estimated by taking difference of CPZ and COZ and is 6.2
km in this example. The climatological (2006-2013) values of CPZ, LRZ, and COZ
over Gadanki are 17.1, 16.7 and 12.7 km, respectively (Figure. 3.9 (c)). The seasonal
variation of these TTL parameters along with the characteristics o f cirrus clouds
obtained from Mie Lidar data are discussed in Chapter 4.

3.6 Summary
This chapter describes the methods used in creating a data base for cirrus
cloud properties using ground-based lidar profiles needed for studying climatology o f
tropical cirrus clouds over Cadanki. The algorithms used for the quality checks, cloud
detection, retrieval of extinction coefficient using lidar profiles are described in detail
with illustrations. Unlike previous studies which used conventional methods for cloud
detection, this study takes advantage of a new algorithm for automated detection of
cloud layers using for the first time an advanced mathematical tool called Wavelet
Covariance Transform (WC'I') method for studying climatology of tropical cirrus
clouds. This method provides accurate base and top altitudes for five different cloud
layers. The algorithms used for the identification of cloud layers and the retrieval of
extinction coefficient for the CALlOP Level 2 cloud layer and cloud profile data sets
have also been described. The definitions and computation of TTL characteristics
have also been discussed. The data analysis methods described in this chapter are the
basis for the results obtained in the subsequent chapters (Chapter 4 to 7) of this thesis.