A Constrained Ratio Aerosol Modelfit CRAM Approach

1
A Constrained Ratio Aerosol Model-fit (CRAM)
Approach for Improved Aerosol Retrievals from
Dual-Wavelength Observations
John Reagan, Xiaozhen Wang, Christopher
McPherson, and Kurtis Thome University of
Arizona, Department of Electrical and Computer
Engineering, and the College of Optical Sciences,
Tucson, AZ 85721
The Reality With GLAS and CALIPSO now in
orbit, global measurements of aerosol by
satellite lidar are now a reality with ever
growing amounts of data to draw on. The Problem
It is well known that aerosol backscatter and
extinction profiles cannot be unambiguously
retrieved from lidar observations without an
assumption linking aerosol extinction and
backscatter e.g., specifying the aerosol
extinction-to-backscatter ratio, or lidar ratio,
Sa Starting Point Basic algorithms initially
in use for processing these satellite lidar
observations rely on location/situation based Sa
look-up tables Improved Approach Constrained
Ratio Aerosol Model-fit (CRAM), approach which
relies on the information content of backscatter
and extinction spectral ratios at two lidar
wavelengths.
2
Overview

• Motivation To improve the accuracy and
  certainty of spaceborne lidar retrievals of
  aerosol backscatter and extinction, thereby
  providing the means for obtaining more accurate
  global aerosol characterizations to assist
  climate modeling assessments of the radiative
  impact/forcing of aerosols
• Outline
• Lidar Relations and Retrieval Approaches
• AERONET Based Aerosol Modeling Model Properties
• CRAM Approach
• 4. Examples of GLAS and CALIPSO CRAM Assisted
  Retrievals
• 5. Conclusions

3
Lidar Relations and Retrieval Approaches

• The normalized (range-squared and pulse energy
  normalized) attenuated backscatter lidar signal,
  X(r), versus range r depends directly upon the
  atmospheric backscatter, ?(r), and extinction,
  ?(r), coefficients

Aerosol backscatter/extinction retrievals
typically employ one of three constraints Auxil
iary-Transmittance boundary value transmittance
from auxiliary measurements. Direct-Transmittanc
e boundary value transmittance from lidar
signal decrease through an isolated
layer. Modeled Lidar Ratio aerosol
extinction-to-backscatter ratio, Sa , assumed
known.
4
Normalized Lidar Equation
GLAS
The range and energy normalized lidar
signal, X(r), at range r is given by
Modeled Sa (two-scatterer solution)
and??a(r) obtained by multiplying ?a(r) with
assumed Sa (Sa?a/?a) .

• APPROACH AERONET (success, contribution)
• create a bounded set of Sa values, representative
  of specific aerosol types, from precise analysis
  of AERONET data,
• examine associated parameters to help bound
  lidar retrieval

5
Comparison of Effects of C, Sa, Ta2 Uncertainties
on Aerosol Extinction Retrievals
6
AERONET Based Aerosol Model Determinations

• AERONET is a globally distributed network of
  sun/sky radiometers, in operation over a decade,
  for improving knowledge of global aerosol
  properties. AERONET observations allow retrieval
  of aerosol size/refractive index information
  (e.g., Dubovik et al., 2000, JGR, 105 (D8),
  9791-9806)
• Cattrall et al. (2005, JGR, 110, D10511, 13 pp.)
  have analyzed AERONE data from numerous sites to
  determine optical parameters (e.g., Sa) for the
  relatively few aerosol model/types that
  predominantly characterize aerosols observed
  around the world
• Biomass Burning
• SE Asia
• Urban/Industrial
• Oceanic
• Dust (spheres)
• Dust (spheroids) by T-matrix modeling

7
The ModelsAERONET Based Modeling Results

• SUMMARY OF LIDAR PARAMETERS RETRIEVED FROM
• SELECTED AERONET SITES (Cattrall et al., 2005)

SD Standard Deviation of Gaussian fit,
typically within 15 of Sa mean.
8
Modeling Results
9
Spectral Ratio Windows for AERONET Based Models
10
What is CRAM?

• CRAM A Constrained Ratio Aerosol
• Model-fit retrieval approach.
• CRAM is an approach for improving dual-wavelength
  lidar aerosol retrievals. It is not an
  inversion, but is a way of maximizing the aerosol
  information that can be extracted from
  dual-wavelength lidar data via modeling
  constraints. CRAM works on the most basic
  aerosol information available in the lidar
  signal, namely, the aerosol backscatter and
  extinction coefficients and spectral ratios
  thereof.

11
CRAM Assisted Lidar Retrieval Approach

• Lidar signals, X(r), are used in lidar
  retrieval relations to retrieve ßa(r) and
  sa(r) at 532 and 1064 nm for each model set of
  assumed Sa values (i.e., for Sa, mean and Sa,
  mean SD for given model).

• Resulting ratios of and
  from retrievals are compared to
• expected ratios for assumed aerosol model type
  to verify if retrievals are in agreement/consisten
  t with model assumption (i.e., retrieved spectral
  ratios, if correct, should fall within model
  spectral ratio windows due to model spread in
  Sa).

• A performance function, Q, can be used to
  quantitatively assess agreement in a
  leastsquares sense between the spectral ßa(r)
  and sa(r) ratios for a given model and
  thecorresponding ratios, obtained from the X(r)
  signals for different assumed model Sa values.

• Model assumption yielding minimum Q taken as best
  solution. But ratios that clearly fall outside
  model windows are obviously not acceptable fits,
  without need of Q assessment.

12
Performance Function, Q

• A performance function, Q, was formulated to
  assess the agreement in a least squares sense
  between the ratios of and
  for a given aerosol model and the corresponding
  ratios computed from the simulated X(r) signals,
  at 532 and 1063 nm, for different assumed Sa
  values

W? , W? and Ws are weighting constants,
generally set to unity. Rs term not used (i.e.,
Ws set to zero) unless some auxiliary estimate of
is available (e.g., from independent
determination of Angstrom exponent, ?, which can
provide model fit to ).

• Model assumption yielding minimum Q taken as
  best estimated solution.

13
(No Transcript)
14
Example Simulation Model Fit Results(for
elevated layer model shown earlier)
15
Example Simulation Model Fit Results
16
GLAS Color Images
GLAS image of assumed elevated dust layer off
West African coast.
17
GLAS Spectral Ratio Results for Dust Layer(Oct.
03, 2003)
x From Self-Transmittance Sa Determinations
18
GLAS Spectral Ratio Results for Dust Layer(Oct.
03, 2003)
The two red lines are the window-limits for
modeled aerosol extinction ratio for smoke case
while the two blue lines are for dust case.
19
GLAS Color Images
GLAS image of assumed elevated smoke layer along
Southeast African coast.
20
GLAS Spectral Ratio Results for Smoke Layer
x From Self-Transmittance Sa Determinations
21
GLAS Color Images
GLAS image of assumed Urban/Industrial mixed
boundary layer aerosol over India
22
GLAS Spectral Ratio Results for Assumed
Dust,Smoke and Urban/Industrial Layer
23
GLAS Spectral Ratio Results for Assumed
Dust,Smoke and Urban/Industrial Layer
24
Modeled and Retrieved (Direct-T Approach)
Spectral Backscatter Ratio for Assumed Dust, and
Smoke Layer
Average and SD for Direct-T Sa Sa,dust ? 45 ?
6, Sa,smoke ? 59 ? 9
25
Layer Optical Depths (532nm) at Sample Positions
along GLAS Tracks
26
CALIPSO Data and CRAM Assisted Retrievals
6 September, 2006 CALIPSO overpass off West
African coast (flightpath segment of interest
highlighted in red)
27
CALIPSO Data and CRAM Assisted Retrievals
28
CALIPSO Data and CRAM Assisted Retrievals
Example of a CRAM assisted aerosol retrieval
using an assumed dust Sa. The mode of the
spatial distribution is in this case very close
to 0.9, confirming the validity of the assumed
model.
29
CALIPSO Data and CRAM Assisted Retrievals
A second retrieval assuming an Sa corresponding
to smoke. In this example, the mode of the
spatial distribution is shifted slightly higher,
but is still well away from a value of 1.8
predicted by the smoke model, suggesting a poor
fit of the data to the model.
30
HSRL/CALIPSO Coordination
31
HSRL/CALIPSO Coordination
CALIOP 532nm attenuated backscatter (top), HSRL
532nm attenuated backscatter (center), and HSRL
532nm measured Sa (bottom). Spatial/temporal
coincidence point shown in red.
32
HSRL/CALIPSO Coordination
Histogram illustration of HSRL measured 532nm Sa
variability within aerosol layer extending from
37.15 to 38.15 N Latitude and from 0.9 to 1.95 km
in altitude. The mean of 72.599 and standard
deviation of 5.3731 demonstrate the applicability
of CRAM in this case, as well as the probable
validity of the Urban/Industrial model.
33
HSRL/CALIPSO Extinction Retrievals 532 nm
Extinction retrievals, 20km horizontal spatial
averaging, 20km sample separation.
34
Mixture Modeling Results
Spectral extinction ratios (?a,532/?a,1064) retrie
ved from mixtures of dust and urban/industrial
pollution.
Modeled and retrieved lidar ratios at 532 nm,
(Sa,532) using linear mixtures of dust
and urban/industrial pollution aerosol models.
35
Extensions/Enhancements to CRAM
Additional constraints may be added to the basic
CRAM approach (i.e., using more than just
backscatter and extinction spectral ratios) by
using auxiliary inputs of various types. Some of
these include

• Angstrom exponent estimates from MODIS or AERONET
  observations (less restrictive than requiring
  absolute optical depth value)
• Transmittance/optical depth estimate of one
  wavelength, as from direct transmittance solution
  to an elevated layer, from MODIS/AERONET or from
  lidar reflectance (e.g., from water at higher
  wind speeds)
• Differential transmittance for two wavelengths,
  T?1/T?2, as can be estimated from lidar spectral
  surface reflection ratio or from MODIS (e.g.,
  from water and perhaps certain land types)
• Sa at one wavelength from auxiliary HSRL
  observations

36
Conclusions

• Employing CRAM on dual-wavelength spaceborne
  lidar data in conjunction with AERONET based,
  improved aerosol models/parameterizations enables
  1) obtaining more accurate/bounded profile
  retrievals of aerosol backscatter and extinction
  and layer optical depths and 2)
  confirming/discriminating assumed aerosol types.
• CRAM successfully employed on GLAS data to
  confirm/discriminate assumed aerosol dust, smoke
  and urban/industrial layers. Aerosol model
  parameters independently verified by
  direct-transmittance lidar retrievals for the
  elevated dust and smoke layers. Similar
  successful results are being obtained from
  CALIPSO observations, but results can sometimes
  be misleading without proper interpretation/checks
  (e.g., layers too optically thick or merged
  layers of distinctly different aerosol types).
• HSRL data providing validation of Sa
  modeling/statistics as well as examples of
  inhomogeneity effects that limit applicability of
  CRAM.
• Extensions and enhancements to CRAM incorporating
  additional constraints enabled by combining lidar
  and passive satellite observations offer the
  promise for further error reductions in
  space-based aerosol retrievals.