3-30€¦ · 2.3.1 introduction 49 2.3.2 discussion of results 50 references 54 3. 0 modeling 55...
TRANSCRIPT
2 r 3-30. 1'f , ,
R-758 NASA CR-K123MV
AEROSOL PHYSICAL PROPERTIES
t r
> r i f '
I "'• ,-, .'
FROM SATELLITE^ORIZQN INVERSION \^K* "• * * < t 1svt • ; ' , * i, i li',
'v- ''V:', .*/' ' 'V' ' f«,/ c i» * T V i *? i i .
'-, '• , , ;.'v** / .* ' ^ »• *
ufii^V'-';*, ' ik/"/.'1, '>*• ' ' '?' jjw£-£*L*v ' ' •* *' "*V$ 'J*1 * <>UJ
• * i?*t*$JF'f' /*" '* ' * " '<* ' ^J^i '-<5>"*w ?FIL t i S{ ,A' ?i ' * *
inff!/,*! , ;* ; • ',t% ^1' <* N•>" *" " ife^ *» * 5 *t» * * ,» R *** "* i, *I {* i V / { »
* /*« * * ' ' ' ! i * '/ f i '6/ ^ >t , »••' M. '1 * , , , !
t rt^^ » s^ « <&
4»% ! / , 4- ,, f't'a^'*'l* v',
M^S JS' » V yitf * l,
t % ' ^ "* T^* * f^| t ( * ' * w
I"1 # *i VBi!C3^ ' > *T* t f> J
V,, **^ T i, kj yff* * ft ^ » Vf , * J
^,v * » il! * ^ ^ *" !,. ^ t * * ? %*^ , * *rf j, *^^*f ^ ^ 1 * - ^ { ^ * *
^w^t /Vi.> |f% ^S|4 /. r ^ ^ \ * /•
>^ ^ e i^*^^^ K f * ""
, ' '^',y .'*'
i f f v'V , 4 ,
f •>
4 * /
,) M .' ^* s M ;>- l ' •
i. «. , /
,/; x\ '. . " *. «/? ' »< » %t ! ,
rt 1 *. , •
i , t
» ' V /f^^§\
v /*/ T^^^^a^^^\\U ^S i . Ifu* L) »
\ ~s V^4fr"^ ^»* ift/ /O/i , f y^ft a V"i'~Jrii"'IUlah /o/ t Jr ^ V^vs. rf ' feytl'y* ^ t s ** >^^* I__Z^^^^* / ^Vfop T^S^^^
^ * * % **
'' 11 '.\V^ . ' ''<
, . <!';-.X'wB'j.:'- . ""ISST. ;-'"j • , "
' s ^, * K t ' ^"/ r » * 1 « * ' « A «v ', *
. * ' *„»' ' « 1' » "-1
' e B Y ' 4 '' * *, < >^ » ' '»„ ', ^ -?
Carlton R.,Gray, Harv£y4L. Malchow, Dennis C Merntt, if* J * /; t
Robert; E:'Var, and ^Cynthia K! Whitney '" * *, ' 1 \
\ * '!/ ^ ,v> 1 ," *' , u , / ' i , . k
*i 'i^^ "V" «.' /r" s1 * » " '» - . •^s* " " * / / li ' ,4 '» H * V ' _ !„ t""*
v »' * '-1 •" ^ < *'. i ^ is x " >' ,/ « ( « ' > T » s r t • , , *^aa . *• A' ' % K2^ S^1 B' B f?*8*
( & ? • * a^Lf^^^^i KM B* ijH « * faC2_ I "•*^j& jr~i ^Sy us&5£ 3 i '••<, ^^3 swfe * * t^ ^ ^ f * ^ — i , t i l ^
% v *VJF r* ' I * H' ' * vs r *^"^ " r > , ' 1*,;
V;! , ':?;\ Vi ; ' . l .«> > \ \ > H '<> f V ' ,, > J -' 1 >
> < » ' ' * l , H, ' ' '-1" > ^^ ' f - f' . *V ' ' . . / ^\! v , ,,% v t \ ^ *
MASSACHUSETTS, INSTITUTE OFf TECHNOLOGY ^' < , , ' . ' , , ' * * » » »
DEPARTMENT (OF METEOROLOGY , ', •*' ^ 4 i t * * '• ' t >AND (,v *» .,; ; > ;.- .
CHARLES S. 'DRAPER ^LABORATORY, INC!/ * j
,'\ M *' ' ,
1 ^ ?c *^ ' v r ; ^
' » ' / V -FINAL REPORT *
% *
CONTRACT NAS, 1-H092- ,
^ • «j *
'
.
' t
)
^ > *V ' i
^PREPARED FOR
I f • »
NATIQNAL AERONAUTICS AND SPACE ADMINISTRATION
LANGlEY RESEARCH CENTER * ^ *
* Hampton, Virginia » 23365 ^ ,>i "" , > ~-v - A
"1 ! !
,' JULY 19731 ' > <
v\» 'r1 ^ I " f f '
* / *» / ^ n *
/ * *» ' ' ' f ' , , ' ", * * * « ' ' ' , - \ , ' , X
* *.',;v v : » -'
^
;*f
'
https://ntrs.nasa.gov/search.jsp?R=19730018603 2020-06-09T09:15:05+00:00Z
1. REPORT NO.
CR-112311
2. GOVERNMENT ACCESSION NO. 3. RECIPIENT'S CATALOGUE NO.
4. TITLE AND SUBTITLE
Aerosol Physical Properties from Satellite Horizon Inversion
5. REPORT DATE
JULY 1973
7. AUTHOR(S)
6. PERFORMING ORGANIZATIONCODE
DSR-73734
C.R. Gray, H.L. Malchow, D.C. Merritt,R.E. Var, and C.K. Whitney
8. PERFORMING ORGANIZATIONREPORT NO.
AER-24-F10. WORK UNIT NO.
9. PERFORMING ORGANIZATION NAME AND ADDRESS
Massachusetts Institute of Technology, Dept. of Meteorologyand the Charles S. Draper Laboratory, Inc.Cambridge, Massachusetts
11. CONTRACT OR GRANT NO.
NAS 1-11092
12. SPONSORING AGENCY NAME AND ADDRESS
National Aeronautics and Space AdministrationWashington, D. C. 20546
13. TYPE OF REPORT ANDPERIOD COVERED
Contractor Report
14. SPONSORING AGENCY CODE
15. SUPPLEMENTARY NOTES
16. ABSTRACT
An investigation is made into the feasibility of determining the physicalproperties of aerosols globally in the altitude region of 10 to 100 km froma satellite horizon scanning experiment. The investigation utilizes a horizoninversion technique previously developed and extended herein. Aerosolphysical properties such as number density, size distribution, and the realand imaginary components.of the index of refraction are demonstrated to beinvertible in the aerosol size ranges (0.01—0.1ju), (0.1—1.0/1), (1.0— 10/u).Extensions of previously developed radiative transfer models and recursiveinversion algorithms are displayed.
17. KEY WORDS [SUGGESTED BY AUTHOR(S)]
Aerosols, Ozone, Horizon Inversion,Remote Sensing, Scattering
18. DISTRIBUTION STATEMENT
Unclassified — unlimited
19. SECURITY CLASSIF.(OF THIS REPORT)
Unclassified
20. SECURITY CLASSIF.(OF THIS PAGE)
Unclassified
21. NO. OF PAGES
211
TABLE OF CONTENTS
Section
1.6 INTRODUCTION. . 1
REFERENCES 5
2.0 RADIANCE INVERSION AND SIMULATION 72.1 AEROSOLS - IMPLICATIONS OF INVERSION 7
2 .1 .1 Introduction 7
2 .1 .2 (0. 1 - 1. O j u ) Aerosol Inversion Simulations . . . . 8
2. 1.3 (0.01 - 0. I/Li ) Aerosol Inversion Simulations. ... 132 .1 .4 (1.0 - 10/j ) Aerosol Inversion Simulations 14
2.1.5' Modeling Errors 152.2 STELLAR OCCULTATION SIMULATION 44
2 . 2 . 1 Introduction 44
2 . 2 . 2 Sensitivity 44
2 . 2 . 3 Selected Simulation Results 452 . 2 . 4 Conclusion . 45
2. 3 HORIZON PROFILE PARAMETER SIMULATION 49
2 . 3 . 1 Introduction 49
2 . 3 . 2 Discussion of Results 50
REFERENCES 54
3. 0 MODELING 55
3.1 FILTER DEVELOPMENT 553.1.1 Introduction 55
3. 1.2 Comparison of Techniques 55
3.1.3 Filter Reviey;:. 573. 1.4 Sensitivity Review 59
3.1.5 Stellar Occultation 60
3.2 RADIATIVE TRANSFER MODELING 62
3.2.1 Introduction 62
3 . 2 . 2 The Computational Problem 62
3 . 2 . 3 Codes Applicable to Related Problems 63
3 . 2 . 4 The Radiative Transfer Codes 64
111
TABLE OF CONTENTS (Cont)
Section Page
3.3 AEROSOL MODELING . 71
' 3/3. 1 Introduction : / . • • • • • • - . ." ."". ."."". . ' ' . 71
3. 3. 2 Choice of Models . . . 71
3.3.3 Error Sources 74
3.3.4 Models of the Extinction, Scattering andAbsorption Cross Sections 83
REFERENCES 90
4.0 SUMMARY • • • 91
4. 1 CONCLUSIONS —AEROSOL PHYSICAL PROPERTIES . . 91
4. 2 RESEARCH EXTENSION . . . . . . . . . 92
REFERENCES 94
5.0 APPENDICES 95
5.1 THEORETICAL HORIZON PROFILES . 95
5.2 MODELED AEROSOL FUNCTIONS 183
5. 3 MULTIPLE SCATTERING CODE C.K.W. . . . . . . . . 191
5.3.1 Introduction 191
5.3 .2 Single Scattering . .. . . . . 192
5.3 .3 Forward Scattering on the Sun-Path. . 197
5.3 .4 Forward Scattering on the Detector Path . . . . . 201
5.3 .5 Full Multiple Scattering . . . 203
REFERENCES : . 211
IV
1.0 INTRODUCTION
This report represents an extension and refinement of programs arid,
techniques developed for determining the vertical distributions of various
atmospheric constituents (aerosols, ozone, and neutral atmospheric density
i.e. Rayleigh scattering air molecules) from radiance measurements of
the earth's horizon (Newell and Gray, 1972). Surveyed in the previously
reported . study were the techniques required for determining selected
constituent distributions and the observability conditions and instrument
characteristics necessary for the successful measurement inversion. This
report extends the radiance inversion technique to the determination of
aerospl physical characteristic distributions. In the previous study
stratospheric aerosols were found to exhibit gross extinction features so
that they could be observed by the proposed technique. Because of increasing
interest in these particles as products of various pollution sources, it was
decided to investigate further (Grayet al, 1972) the potential of the horizon
inversion technique for determining other physical characteristics of the
stratospheric and mesospheric aerosols besides total extinction. Thus an
analysis of the capability of the technique to deduce information about the
particulate size distribution and index of refraction was undertaken.
These investigations were centered around the refinement and extended
development of the techniques previously employed. Here the state vector
in the inversion computer code was extended to include the desired aerosol
characteristics. The radiative transfer simulation was refined to include
arbitrary (noncoplanar) azimuth angles and opaque (thick) clouds. These
refinements were tested for accuracy against other less efficient codes.
To supplement these activities an efficient empirical aerosol model was
developed from data generated by a more complex Mi'e code computation
of the aerosol optical properties. With the empirical aerosol model,
quantities such as the partial derivatives of the angular scattering function
with respect to the real part of the index of refraction can be easily computed.
These refined techniques and aerosol models were then used to simulate
an aerosol inversion procedure for obtaining the vertical distributions of
aerosols and their gross physical properties.
The measurement technique used here is to scan the earth's sunlit
horizon as shown in Figure 1.0-1 with a multiwavelength photometer and
to record the radiant intensities measured at a predetermined sequence of
angular positions. These measured intensities are then processed as shown
in the flow chart Figure 1.0-2. _
The estimation procedure begins with an input of the geometry which,
includes primarily the sun direction and the measurement platform altitude.
The geometric data along with an a priori estimate of the atmospheric
state (constituent densities and aerosol characteristics) is used to compute
a theoretical prediction of the intensity profile and its partial derivatives
with respect to state elements. This is done with a radiative transfer
simulation for the first selected wavelength and scan angle. The theoretical
partial derivatives, theoretical intensities, and the measured intensity are
then fed to the filter equations where an optimal estimate of the atmospheric
state is produced based upon the difference between measured and predicted
intensities, the state covariance, and the measurement noise. The optimal
state estimate is then fed back through the intensity calculations and a
new intensity is computed and compared to the next measurement. The
loop continues until all wavelength channels and scan angles have been
sampled. The result is an optimal estimate, in a minimum variance sense,
of the atmospheric state including aerosol number density, size distribution
parameters, and index of refraction. ,
Section 2 of the report discusses the aerosol inversion capability of
the scattered sunlight experiment, stellar occultation inversion, and radiance
effects from a horizon profile parameter study. The aerosol inversion
simulations show the relative invertability of the various aerosol
characteristics under a range of conditions including different altitudes
and assumed a priori uncertainties. The efficacies of several wavelength
bands are compared. Three aerosol size ranges are considered, (0.01-0.1//),
(0.1-1.0/v) and (1.0-10A/). The effects of changes in variables such as season,
latitude, satellite altitude, solar zenith and azimuth angles, and albedo on
horizon profiles are investigated.
2
Section 3-details'the filter and radiative transfer model development
and inversion procedure refinement work that was carried out to make
aerosol inversion possible. Included are discussions of the augmented state
filter, the refined radiative transfer simulation, and the empirical aerosol
models.
Section 4 provides a review of our conclusions and recommendations
on the invertibility of aerosols,'model development and satellite data
requirements.
Section 5 appends supplementary information.
Appendix 1. A parameter study of horizon profile intensity variability
as a function of season, latitude, albedo, sun azimuths, and zenith and
wavelength is illustrated.
Appendix 2. The regression formulation of the aerosol model used
for the size ranges (0.01-O.i/t/), (0.1-1.6/u) and (1.0-10.0//).
Appendix 3. The multiple scattering code C.K.W. is discussed.
Fig. 1.0-1 Limb Scan Geometry
InitializeComputation:
GeometryState EstimateCovariance
Use EstimatedState Elementsto ComputePredictedIntensities
Use EstimatedState Elementsto ComputePartialDerivativesof Intensities
MeasuredIntensity
UpdateStateEstimateandCovariance
FinalEstimate ofConstituentDistribution
Fig. 1.0-2 Atmospheric Data Inversion Procedure
REFERENCES
SECTION 1
1) Gray, C. R., H. L. Malchow, D. C. Merritt, and R. E. Var, "Aerosol
Monitoring by Satellite Horizon Scanning", J. Spacecr. Rockets, 10,
71-76, January, 1973.
2) Newell, R. E., and C. R. Gray, "Meteorological and Ecological
Monitoring of the Stratosphere and Mesosphere", NASA Contractor
Report, NASACR-2094, August, 1972.
2.0 . RADIANCE INVERSION AND SIMULATION
2.1 AEROSOLS - IMPLICATIONS OF INVERSION
2.1.1 Introduction
Using a Kalman-Bucy filter to invert horizon profile data produces
both an estimate of the atmospheric state and a covariance matrix describing
the accuracy of the estimates. By propagating the covariance matrix through
a proposed measurement schedule it is possible to predetermine which
parameters will or will not be affected by the measurements, and what
linear combinations of. parameters will be well known. A function called
sensitivity has been defined which is a measure of the amount of information
obtained about a particular state element in a measurement. The sensitivity
curves are computed very rapidly, but fail to predict quantitatively the
final variances and covariances of the parameters which are obtained by a
covariance propagation. They do, however, qualitatively predict which
parameters will have small variances at the end of a measurement schedule,
and so they are used extensively in the following discussion.
A sensitivity analysis was applied to the problem of determining which
aerosol parameters are observable from a horizon scan experiment. It
has been demonstrated (Newell and Gray, 1972) that aerosol extinction per
unit volume is readily recovered from limb scan data at all altitudes between-3
10 and 85 km with an idealized instrument having a noise value of 10
times the maximum horizon signal and between 10 and 100 km for a noise-4 ' .
value of 10 . The sensitivity study, therefore, is aimed at determining
the observability of four different parameters, namely aerosol number
density (/°), size distribution parameter (a), real index of refraction (n),
and imaginary (complex) index of refraction (n1).' The size distribution
parameter (a) is the radius exponent where the power law size distribution
is of the form n ( r ) = Ar . The results are broken into three groups
representing three assumed aerosol size ranges: 0.01-0.1/u, 0.1-1.0/y, and
1.0-10/c/. For these size ranges solar zenith angles were varied from +60
to -30 with resulting scattering angles varying from approximately 50
to 140 . Constituent results based on these variable conditions are developed
in Figures 2.1-1 to 2.1-12. Analysis shows how the invertability of the
parameters is affected by 1) the geometry 2) the instrument noise, and 3)
the statistics of the initial estimates of the aerosol parameters. This
analysis was conducted for spherical aerosols and nonspherical aerosol
shape factors which have a spherical equivalent.
2.1.2 (0.1-1.0/u) Aerosol Inversion Simulations
For this aerosol size range (0.1-1.OAT), .as is also the case for the
other size ranges, the aerosol number density is initially adjusted so that
extinction at 5500 A is equal to the extinction used as the standard state
(Newell and Gray, 1972) based on Elterman, 1966andl970, given the initial
estimate that a = 3, n = 1.5, and n1 = 0. Because of the proportionality
between extinction and cross section for a given number density there will
be an increased number density for the (0.01-0.1//) range over the (0.1-1.0/t/)
range and conversely a decrease for the (1.0-10//) range. For the range0 0
(O.I - I .OA-) the number density runs from approximately 10 /cm at ground
to 10"5 /cm3 at 100 km.
For all of the three aerosol size ranges the instrument white noise_ n o
was assumed to be 10 u watts /cm (RMS). It has been shown previously
(Newell and Gray, 1972) that varying the noise simply varies the maximum
altitude from which information can be retrieved. The maximum useful
altitude can be approximated by observing the intensity versus altitude
curves of Section 2.3 and noting at what altitude the intensity reaches the
noise level i.e. S/N = 1. At a few kilometers above this point the initial
estimates of constituents are as good as the filtered values and no significant
changes are affected in the.estimates.
Sensitivity curves were produced fora number of different scattering
angles and wavelengths with an instrument having a finite field of view.
Sensitivity is a function derived from the filter equations, which indicates
the relative information content from constituent to constituent and altitude
to altitude for every measurement condition i.e. (wavelength and tangent
height). Given a measurement, the information content for a particular
constituent and altitude (state element) is reflected in a decrease in the
variance associated with that element and it is natural to look at the
covariance update equation for the definition of sensitivity:
P(nH-l) = P ( m ) - K ( m + l ) B ( m + l ) P ( m ) (2.1-1)
the term of interest here is K(m+l)B(m+l)P(m), which is the decrease in
the covariance for a particular measurement condition. Normalizing by
P(m) yields the sensitivity matrix K(m+l)B(m+l). .Of particular interest
in the sensitivity matrix are the diagonal elements which indicate the relative
decrease in variance of the elements of the state vector. To isolate the
effects of the measurement vector on the sensitivity, the covariance is'
assumed to maintain its initial value. With this assumption the sensitivity
of the jth element of the state vector is
s . ( m ) =3h(x ,m)
dx.± ,
\2Pii + R
2.1-2
In choosing an instrument's wavelengths it is desirable to have as
many linearly independent measurements as possible. That is, each
measurement should be sensitive to a different combination of parameters.
The functions which relate extinction and phase function to aerosol
parameters change slowly with respect to wavelength. For this reason a
wide spread of wavelengths over the region 2500 to 7000 A is best suited
for the inversion (see Figure 2.1-1 or 2.1-4). Thus the wavelengths 3000,
4000, 5000, 6000, and 7000 A, which span the visible region with a minimum
of wavelength channels necessary for the inversion, have been arbitrarily
selected as working values. In this case the actual wavelengths chosen
are not as important as the separation between the wavelengths and in an
actual experiment more wavelengths would be run for greater accuracy. "
The total scattered sunlight radiance at the horizon is a composite
of the radiance contributed by aerosols and Rayleigh scatterers. For each
scattering constituent the radiance contribution is proportional to the product
of its scattering cross section and angular scattering function (phase
function). If the scattering cross .sections for both aerosol and Rayleigh
scatterers are equal then the relative contribution of each constituent to
the horizon radiance depends only upon the relative values of the phase
functions at a given scattering arigle. For most scattering angles away
from the forward region the aerosol phase function will be smaller than
the Rayleigh function thus reducing the relative contribution of aerosols to
the radiance, however this effect can be partly offset by an increase in the
relative scattering cross section of aerosols at longer wavelengths.
Both' the effects of changes in wavelength on cross section and changes
in scattering angle on the phase function can.be observed by examining the
sensitivity curves of Figures 2.1-2 and 2.1-4. In Figure 2.1-2 it can be
seen that the aerosol number density sensitivity "AEROSOL" increases at
longer wavelengths while the "NEUTRAL" molecule scattering decreases.
This is the cross-section effect. Figure 2.1-4 differs only by a change in
solar zenith from Figure 2.1-2. The scattering angle for Figure 2.1-4 is
less favorable (i.e. less forward, w l l O versus ~ 50 ) for aerosols.
Therefore, the aerosol sensitivity is lowered relative to Figure 2.1-2 while
the neutral molecule scattering sensitivity increases.
The most serious .problem encountered is in the uncertainty of the
initial statistics associated with the initial estimate. The updating of
estimates by the filter inversion routine depends in part on the RMS values
of the initial estimate. If a parameter is well known it will not receive
much of an update and its variance will decrease relatively slowly. This
is particularly critical in the case of the aerosol parameters where there
is a large sensitivity to aerosol extinction per unit volume, however, the
particular aerosol parameters which receive these updates are determined
partly by their initial variances.
10
Sensitivity computations in the (0.1-1.0/v) size range were made with
initial RMS error values of 600% for (P), 1 for (a), 0.05 for (n), and 0.005
for (n1) along with realistic estimates of neutral density and ozone (Malchow,
1971) RMS values as a function of altitude (Figure 2.1-3). Curves run
with these values showed a large sensitivity to aerosol number density
and only a small additional sensitivity to (n1) below 30 km. Other runs
with 100% RMS error on (P) (Figure 2.1-1) showed all parameters except
(n) to be readily observable at all altitudes. Thus varying the initial RMS
estimates varies the final RMS estimates and the sensitivity, and until
realistic data are collected about these parameters the covariance matrices
associated with the parameter estimates will be unreliable.
The sensitivity curves indicate a potential problem with regard to ob-
servability. Since the wavelength dependence of sensitivity is basically the
same for both number density(/°) "AEROSOL" and size distribution(o) "ALPHA"
there is the possibility that no two measurements are linearly independent in
these two parameters, therefore making it impossible to estimate either in
one . This similarity in wavelength dependence occurs because increasing
either (a) or (/>) has the effect of lowering the wavelength dependence of
the total received signal by increasing the effective aerosol to Rayleigh
scattering ratio. A complete propagation of the covariance matrix was
made to determine if this was the case. The results indeed showed a.
high correlation (0.6 to 0.8) between the two parameters but also showed a
significant factor of five decrease in variance for both parameters in a
coarse inversion (only three wavelengths and a four kilometer separation
of tangent heights). The final variances were VQ - ±0.2 and crp = ±20%
which indicates that the two quantities are separable and observable.
Covariance propagations were performed for each of the particle size
ranges. The results of these propagations show numerically how the initially
assumed variances of the aerosol parameters are reduced by the information
gained in a measurement sequence.
11
Initial values of the aerosol parameter RMS uncertainties were chosen
to reflect the range of values occurring in the literature (Newell and Gray,
197,2, and Malchow, 1971). Number density (.ft) was assumed to vary over
.one order of magnitude ±3cr, thus the ap used was 166%. This is consistent
with _the large observed, variations in the number density-of stratospheric .
aerosols related to volcanic activity. The size parameter (en) was assumed
to cover the entire observed range (Xa<6, thus aQ = 1. The real part of
the .index of refraction ranges from that of water (1.33) to fused silica
(1.65) thus yielding a ±3a range of 0.32 and a a of 0.05. Finally the complex
part of the index of refraction was assumed to be as large as 0.015 (+3cr)
which is felt to be conservative (Volz, 1973) with a a , value of 0.005.
Both the real and complex parts of the index of refraction are assumed to
be wavelength independent. All initial RMS values are assumed constant
with altitude. ,
Figure 2.1-13 shows graphically the reduction in RMS uncertainty
resulting from a covariance propagation for the (0.1-1.0/v) particle size range.
The solid line indicates the assumed initial value of a particular quantity,
and the error bars show the final RMS uncertainty. Numerical values
corresponding to the error bars are listed in Table 2.1-1. Considerable
reduction of the number density uncertainty occurs for all altitudes although
it is especially prominent at 10 and 20 km where the aerosol turbidity is
at a maximum. A similar pattern of error reduction is displayed for the
size parameter (a). Uncertainty in the complex part of the index of refraction
is reduced, however, the improvement is not as striking as in the case of
the other parameters. In general the covariance propagation for this size
range shows that substantial reductions in the initial uncertainties of aerosol
number density, imaginary index, and size parameters are effected by the
inversion. The uncertainty in the real part of the index of refraction is
however reduced only slightly. This result is consistent with the sensitivity
curves in Figures 2.1-1 and 2.1-2 which show little sensitivity to the real
index of refraction at any altitude.
12
2.1.3 (O .OI -O . IA / ) Aerosol Inversion Simulations
The aerosol number density was adjusted to preserve the standard
extinction versus altitude curve for 5500 A with a = 3, n = 1.5, and n1 = 0.7 3For this range the density varied from 1.4x10 /cm at the ground to 5.5x
- 2 310 /cm at 100 km. Since the phase function for this size range varies
less with angle than in the (0.1-1.0^) range only a few zenith angles were
run. The results of these runs show that the constituent sensitivities did
not significantly change with zenith angle (see Figures 2.1-6 and 2.1-7).
The most notable difference between this size range and the two others
is the lack of sensitivity to (n1). For these small particles only number
density and the size distribution parameter (a) are observable because the
partial derivatives of the scattering phase function with respect to (n) and
(n') for this size range, are near zero. A change from 300% to 100%
uncertainty in number density enhances the (a) sensitivity rather than
increasing (n1) sensitivity (see Figure 2.1-6 and 2.1-5 respectively).
Analysis of the two sensitivity curves for (/>) and (a) indicates a
potential problem in distinguishing between the two parameters. The fact
that the sensitivities for both parameters have the same wavelength
dependence implies that the measurement equations at different wavelengths
are not totally linearly independent (see Figure 2.1-8). However a coarse
covariance propagation was run (five wavelengths and four kilometer tangent
height separation) which indicates that the two are separable. The
correlations were indeed high (0.6 to 0.8) between the parameters but there
is a factor of two decrease in number density variance and a factor of ten
decrease in size distribution variance from the initial values.
The two parameters ( ft and a ) are highly observable from 10 to 100
km or until instrument noise becomes dominant. Typically an instrument_3 '
noise of 10 times the maximum horizon signal restricts the altitude-4sensitivity range to 85 km while a noise value of 10 extends the altitude
sensitivity to approximately 100 km. Thus the interference that comes
13
about from increased uncertainties in neutral density at higher altitudes
does not reduce the observability of the two aerosol parameters significantly.
Figure 2.1-14 illustrates the RMS error reduction for the particles
in this size range. The numerical values are listed" in Table 2.1-II. The
results are generally similar to those obtained for the (0.1-1.0/u) size range
except that there is a smaller reduction of the number density uncertainty.
Somewhat surprising, in view of the near zero sensitivity for (n1) shown in
Figure 2.1-6, is the substantial reduction of the uncertainty in (n1) after a
covariance propagation. This indicates that the factor causing the error
reductionisa buildup of correlations between (rt1) and the other parameters
as opposed to a partial derivative effect which dominates the sensitivity
results. When the sensitivity and covariance propagation results differ
significantly as in this case it is important to consider the differences
between the abbreviated covariance propagation used to produce these
working numbers and a real covariance propagation used in an actual
inversion. To insure the ultimate convergence of the state vector estimate
it is often required to reduce the covariance update by means of a numerical
multiplier. This results in a smaller reduction of the initial covariance
after a given number of reiterative updates than would' be indicated by the
unweighted propagation used here.
2.1.4 (1.0-10/t/) Aerosol Inversion Simulations
The large particles in this range are the furthest from Rayleigh in
their scattering properties, i.e., in their cross-section wavelength
dependence and in the shape of their phase function. Because of their largeo
cross section the number density is low, running from 23.4 /cm at the
ground to 7.7x10 /cm at 100 km.
The great difference between these particles and other atmospheric
constituents makes them the most readily inverted of all size ranges. It is
possible to invert all four aerosol parameters from a given horizon scan
when the phase function is not minimized by an unfavorable scattering angle
14
i.e. sun angle. Although the four sensitivities are reasonably close, results
from the previous size ranges indicate that all four physical characteristics
are separable. Variable scattering angles are readily obtainable for nearly
all conceivable orbits from equatorial to polar orbits with the exception of
the 600 and 1800 local hour angle near polar-orbits.
There is, however, a problem with this size range that does not appear
in the other ranges. That is, when the scattering angle approaches 110°,
the phase function is so small that the aerosol energy contribution and
respective sensitivities drop drastically and only number desity is highly
invertible from the data (Figure 2.1-12). The variations in sensitivity with
changing solar angle can be clearly seen in Figures 2.1-9 through 2.1-12.
Covariance propagation results for the large particles are shown in
Figure 2.1-15 and the corresponding numerical values are listed in Table
2.1-II. In the case of the large particles all the parameter uncertainties
are substantially reduced including the real part of the index of refraction
which was essentially unaffected for the other particle size ranges.
2.1.5 Modeling Errors
With the standard filter approach to the inversion problem it is
theoretically possible to drive the variances on all the parameters as close
to zero as desired by taking enough measurements. However, since there
are inherent errors in the radiative transfer modeling a near perfect variance
is unrealistic. The purpose of this section is to determine the effects of
aerosol modeling errors on the final accuracy obtainable in an inversion.
This is done by computing the vectors (k, Ax) and the scalar (AI) in the
equation Ax = k (AI) where Ax is the state error, (k) is the filter gain, and
AI is the intensity error.
The results are presented in sets of tables. The first table of each
set (Table 2.1-IV) illustrates the gain (k) which is used to compute errors
in the state given errors in intensity. This is done with a geometry having
15
a zenith angle = 30 and an azimuth angle = 0° for all size ranges, with
standard initial errors (see section 2.1.2), and using a 300% initial uncertainty
in aerosol number density. The additional tables of each set are the computed
state error Ax for errors in the albedo and aerosol models and for an
instrument Mas. ^The state errors are based upon the gain (k) of the first
table.
It is important to realize that the initial uncertainties are important
in the propagation of error. For example, a parameter which has both a
large partial derivative and a large initial uncertainty will receive a large
update and become more in error by inaccurate models than a parameter
with a small initial variance and partial derivative. The numbers given
represent the maximum error that will occur in an inversion using 3000,
4000, 5500, and 7000 A as wavelength channels. It should be noted that the
(k) values for 3000 A at 20 and 40 km, and for 4000 A at 20 km are set
equal to zero. This is because these wavelengths are saturated at the
designated altitudes and therefore yield no information about the densities.
Tables 2.1 -V, VI, IX, X, XV, and XVI list the errors in the state
vector elements produced by errors in the aerosol extinction and phase
function models. The model errors used here are representative of the
errors associated with the aerosol models presented in Section 3.3 of this
report. Tables2.1 -V, VI list the state errors for aerosols in the (0.01-0. IA/)
size range. The modeling errors for this size range are small («2% in
extinction and «5% in phase function) and consequently the effects on the
state are also small. The largest resulting error is due to the phase function
error, and is a 12.3% error in the particle number density at 60 km as
shown in the fourth column of Table 2^1 -VI.
The (O.l-l.O/^) particle size range has larger associated errors and
consequently larger effects on the state. Tables 2.1 -IX, X list the state
errors associated with this size range. As with the smaller size range,
the phase function error (10%) produces larger state errors than the
extinction error (5%). The largest error induced is in the aerosol number
16
density at 60 km, and is 26%. This error is still small compared to the
assumed initial number density uncertainty of 300%, and is less than the
60% uncertainty obtained after a coarse covariance propagation.
For the large particle range the state errors are presented in Tables
2.1 -XV, XVI. Again the number density estimates .are primarily affected,
and the errors are comparable to the final state estimates obtained via a
covariance propagation.
Table 2.1 -XII shows the effects of the assumption of the wrong size
range for aerosols, i.e., the table shows what errors result if the size
range is assumed to be (0.1-1.0/u) whereas the actual size range is (.01-0.1/u).
The resulting errors are: large as expected since the extinctions differ by
two orders of magnitude for. the two size ranges. Aerosol number density
is shown to be in error, by one order of magnitude or « 1000% at 40 km.
Other state elements are also strongly affected. The neutral density estimate
is in error by 52% at 80 km, and the ozone density by 242% at 60 km.
Errors in the other aerosol parameters are significant though not large
(these are given in absolute units). At 40 km, the error in (a) approaches
one which represents a significant alteration of the size distribution. The
results of this size range shift emphasize the importance of either knowing
rather accurately the particle size, limits, or expanding the state vector to
include these limits. .
Table 2.1 -XIII shows the effects of the introduction of a 10% instrument
bias error in each wavelength channel. The error induced in the aerosol
number density is moderately large because the signal contribution by the
aerosol to the total signal is small. Since the aerosol number density
uncertainty is large, the filter attempts to correct the signal error by
adjusting mainly the aerosol number density.
Finally, Tables 2.1 -VII, XI, and XVII show the effects on the state
of the introduction of anunestimated effective surface albedo deviation from
the expected value. The effect of an albedo uncertainty is similar to the
17
effect of an instrument bias, i.e., there is a broad-band shift in the measured
intensity away from, the expected value. Since aerosol density is assumed
to have a relatively large initial uncertainty, it is this quantity which is
mainly adjusted to account for the intensity shift. The amount of adjustment
done to the aerosol number density is inversely proportional to the
contribution that the aerosol makes to the total intensity. Error values in
Table 2.1 -VII, XI, and XVII are based upon a scattering angle that coincides
with the minimum in the angular scattering functions for the medium and
large-sized particles.
Thus these aerosols are making a minimal contribution to the total
intensity at the receiver, and therefore the errors related to an uncertain
albedo are maximized. Tables 2.1 -VII, XI, and XVII illustrate this effect
dramatically. Each aerosol size rangeis defined to have the same extinction.
As the particle size increases, the scattering phase function decreases (at
the chosen sun angle) and the error increases rapidly for the same extinction
in each size range. There are two implications of these results. One is
that the albedo should be added to the state vector and estimated to minimize
its uncertainty. The other implication is that the solar scattering angle
for primary radiation should be chosen to minimize the effect of albedo
uncertainties. At more desirable scattering angles the large particle errors
(Table 2.1 -XVII), are substantially reduced, and are comparable to the
small particle errors (Table 2.1 -VII). For example, the 1160% error in
number density for large particles at a scattering angle of 80 (Table 2.1
-XVII, 40 km) is reduced to 60% when the scattering angle is changed to
30°.
18
om
§
oro
Io
-H n-p w o- H S U
dPO
dprH
dPCO
dPVD
dPen
dpoCN
dpVOvoH
OOM
O
VDr~CN
O
•<*3*CN
O
oVD0
O
invoo*0
ro*rH• •O
0•H
oorooo -
0rHO
O
*O
0
ro*d*oo
*3*0•o
003*O• •O
in0
o
CNo0
o
rH00
O
H00
O
rHOO
0
CNOO•O
rooo•o
inoo•o
0) (0O AC -P10 e ^a) a
0)> a)n) N
6 w0 0)n > o
fo O01 -H<U £1 -P.pj 1 i t i
-P -H (0C IS ft
-Hn) C-P OM -H •(!) -P rHo mc CP oD nJ -P
HI
CN
0)
idEH
3-o
om
§
H oCM
(0•r\ H•P CO O•rl 2 >-)
H W
dPVD
dPmCM
dPmCM
dpr~CM
<*pom
VO
dpVDVDrH
Ht1
CM
O
in« «o••0
VDmO•o
••*o•
o
00ino•o
100
o
o•
•H
00"SfO
0
U5*e
O•o
in* *0•o
o*0•
o
in*3*0•o
r-oo
ino•o
CMO0
O
rHoO*
0
rHOO•O
rHO0•O
CMOO•O
CMO0
O
ino0. •o
Q) tOO J3C -P
tn •C 0)<u tr>
-HH(0
ou
§
<uoc
HHI
CM
0)rH
•3EH
0)
c•HrtS C-P O
-P(0
moSH
cuN
-HCO
(Ufo -H
CMCO rH
Q) .ti U-rl -P -cH
-P -H 4Jrl(0
oI
20
oin
O<n
H OCM
(0•H M
4J W O•H 2 r4
H W
dpco(N
dPV£>CM
dP•**"CM
dPOr-H
dPinCM
dPCMCM
dPVOVOrH
r-inVO•
0
unm0•
o
coino•
o
CMr-0•
o
minCM•o
inMCM•0
dPO•H
•CMO•
0
VOCM0•
O
inCM0*
o
rHrH0•
O
COHO•
0
To•
0
mo•0
H00•
O
rHOo•
O
rHoo•
O
HOO•0
H0O• • •O
CM0O•
O
in00•
o
0) (QU X!C -Pnj D^
Ou
o
a)oCID
CO
HHHI
rH
•
CM
a)
CO<u J3•H -P
4J -HC S
•H
O•rH-P
Q)0)> CtO (0S PH
0) 0)> N
U-H-P
3-o
21
H1!_|
,
C>4
W1—5COEH
WU
CrHH
WNHM
j
oW
^^-^,_).
oI
o
o• —
(ZO
.y
*—*
£5H<0
p£MEH
HCM
o
II
j-J )3g-HN
^0oCO
Q)(— 1
C«.
j^4J-HC0)N
W
^10)JJ(Ug(0
rfl
rH
oenPQ)
""C
o c
tu LU0 0O 0
* *C rj
£
0 O
UJ UJ0 0Q 0
ft ft
0 O
8
c o
UJ UJO O0 0
0 0
^^roeu
^"
-P•iHCDCQ)Q
M0)
gjj»2
r<
•
-P
r— |
O10
oCjj<J
o :3
LU LUo oO ~2• •
O .3
0)CoNo
0 0
J LUO 00 O• •0 O
rH(dJ_l4Jj3Q)S3
o o
LU m0 Oo o• •
O O
-^ O OOpt; o ov- o 0
co ^
9" . o o
o in
1 1UJ LUO Ln_r> r\j• •
1s- <tf°1
"1 '1
1 1LU LU>- ovt' —"•• »
-?• -0
—1 —1
1 1UJ LU
T C^<y co
^o co1 t
LTI sn
LU UJM \OCM M3• •
-H — •
0> ^3
UJ LUVD L-»rr> O• •
in o»1 1
"3" ^^4 -4
LU MJ-M fO"M C^• • .
roro
0 0O 0LPl OLTNt-
0 0rvl rj
O O 'J > r*?
I I 1LU UJ UJ U)O <y r inO (JO f<"> — <* • • •
C3 ; — 00 't1
O M -• -H
1 1 1Uj IJJ UJ LUo o ~? o0 vj CD 0• • » »
O sQ — i if
O O — 1 — >
UJ UJ 'Jl iUO fA — 4 ur\o LA en c«
o cr> CM ir\1 i 1
0 * -t LA
•M ...J '..'.I UJO LH 00 fO O 00 CM• • • •
O CM T — 1
O 00 ~« —i—4 —4
UJ lij 'i-J I>J
O CM CO -4O vp CM 00• • • •
O -4 l»"\ -4
1 1 1
Dm o o- 4 - 1 — 4
UJ Lu ULJ LU
O <T* ^o — i t CM• . • •
O oo — < ro
0 0 ^ 0O -5 'J Oo o ..r\ p .-:^ -T f~-
o o o o•4- •*• t vf
v^-
1. u
~4
o•
rO1
"3
6IT,•
—4
M
:; j. *.
"•\11
>J-
L.' 'CO\o•
(M
(M— 1
i.l-
CIJ4-•
—4
1
o—4
J-;
o-o•
o
,0oor>
o- <O
<r f-
1 1'.jj LU
<f 0SO J>• •
CM vO1
-> o
il LUO 0*-* .n• •
.-4 'M
CNJ \J
'. ! M :QD CMLTI >-
-4 rO1 1
f ~J"
'.. J :..J
t\J — *pv }-t •
—1 T
t~ — <•- (
:U UJ
CM J-r^ 4"• •
>*• -<1 1
O 0— «* "4
.Jj JJrn ^*ro ^\• •.f r-
0 0o oO -">J- J^
0 0so -•£>
CM
!•Jj— io•
,-«.
o
-0•«^
tr~.
.--)
i'.ii-O
—4 .
1
LA
-;jCM<Mt
— ( •
_j
-UO—4 '•
1T»
1
>•—4
^J
'.\J
oot
'
ooo"~ .o•o
sr >t M o
1 i iUJ -i-' UJ 1U(T* o ro Oy^ >. _ 4 fr\• • • •
CM CM n —11 1
—4 — 4 -J (N)
,.j 1J . J •;.;
^- AI r\l vO• C^ u"> ."M• • • • •:-vJ — 1 J" — 4
"O ?O - -J"
i : ; '.. .( j vjj tj.vt»- *• — ' ^<• !*• 1A 3
(^ .'M vfl — <1 1 1 1
t i- -f ,.n
,;j L.) JJ .jj•-H \Q c>sj o••T OO .f) !N|• • • •
'.M — * T -H
o in o> co:— 4
''jj , -J JJ JJ
03 ^ 0 0ca -3 — i ~-• • • •
-4 "» NJ !*.
1 1 1 1
• — 0 0 " --( ,-4 — I — 1
LU .jj JJ Ij
O >- O 03r\J r^- ^* -^t ' . • « •
rS LA 0s PJ
-> '.J '-> OO 0 O 00 O in 0m s- '-r\ r-
0 0 O O•a co a -~5
22
wU23wN
w,_q(-1
AE
RO
S(
-t
rH> .1 0
•H 1• rH
CN o
*W o1-3 N-'«
3 «EH O
fa
liX<
rt§PHM
WEHr<E-5W
ER
RO
R
&OHEHU"Z.MEHXW
OPCN
rtj
WEHH&
oo
II
£4->3
-HN<
•koOCO
II
a)Htr>C<j^4->•rl
C(1)tsl
T?orl• .M(U
0)4Jr-«P
rHO(0XIm
>H(U-P<L>errt'Url
idft
rH
O10orla)<
•C
r— f-- r— c-
I I I !LU UJ UJ (JJNO CO CN! COfO ~4 CO NO• • • »
— < ro <N ;M
C
in -"\ u-\ in
i i i iuJ iU UJ uj•f f U"\ ON
• ^* -—I >- • — 1• • • •
CM \r\ ir\ m
8
rn m n) <n
1 1 1 1UJ Ul UJ UJrj NO — i CMin ro CM <*•
• « • «•frt f— O3 f-
1 1 ! 1
"T7orlrl<U
4JC(UUrl0)
3
>4J-HtoCQ)arH<L>
rj
K
•
4JrH<
HOtoOrlQ)<
O O O O
LU UJ UJ UlNO — i (S| TO ^ <JN-'st-• « • •
CM •*• -4- \J-
<DCONO
(O CM v-i <M
1 1 1 1LU UJ LU UJfO rfi & Q3T CO 1— NO• • • •
<Sl -i Nf t\J
1 1 1 1
rHft
|4J3(Ua
-r rO !M <M
1 1 1 IUJ UJ UJ U!•o f— in \oO — 1 O ">• « • •
—4 ro >.r\ \J :
"g . o a o- oj«J CNJ -*• O -O
HC5
I
WNHW
K!owsw«:
3.H H> •1 O
rH 1• rH
CN O•
W 0I-H* ""CQ3 OJEH O
fa,
X<N_X
. «
D^wwEH<E-iW
^ O« 0
rt IIrt _.W fi
4->S 30 gH -rlEH NU <&Dfa O
oW "cnrf HHT|
ft a>rH
dp &»in C
<
< r*fi
ffi ^EH -:iH ?,
* £
17orlM(1)\w
a)4J3rHOtojQfd
S-i0)4J<L>
MraftrHOU]0M0)<
"c
r-
IUJvT«0>
sl-
fl
LTl
1UJ.
—*.* —
»<J>
8
rxj
1LU
O-r•H
1
^T7orlrlQ)
4->CQ)UrlQ)_a
>.4J-H10C0)Q
rlQ)
•i0s
•
4->H•<
HOCO0rl0)<:
O
LUOM•
<»
Q)CoNo
(O
1UJOvXi
•(/>1
Hf«MJJ3Q)&
^t1
LUNf
IM•
NT
l
•g oX j fM
NO 3 -0
1 1 1LU Ui a-CM NO —iC-J CJ CM
— 4 —4 _4
-d- ^ -J-
i i iLU LU LUm NT oen NJ- nr.• • •
~< — i — i
"\J '\l M
1 1 1UJ UJ UJNT r- r-«r- o oo• • •
— ( CM -H1 1 1
—t —4 —
UJ LU ULl\0 "<"> — <-- 4 (M. _|• • •
r-4 .-< r-4
(M O CNJ
1 1
UJ U4 UrM -< FMrj CM t-• • •
iO — i voI I 1
CM ~< -4
I I IUJ ;u LUin o NT-i in r-• • •
CM -H —i
0 O 0<;• -o -o
23
HH>
1iH
•
CN
WiJJm<!EH
W0JZ
2
ta«Hin
,J0eno%w*:^^p.
,-H•
O1HO•
O*w"
CV*CMoPM
^^
X<]•*— •"
tfo'§wMEHtfEHW
jqf?*!•
g
OPwfflJ13in
•
o
QPOJU
<cPS0faPH§m«woPpqoaa,-1
•
^^
ECEHH.s
o0
II
x;4-1se
•HN<
*.
OoCO
11Q)
1nn&ic<;A-P•Hc0
tSJ
^_»noMH0)
Q)4J2
r-JOco
XJrd
***"*''
H0)4->•Q)ardMrdcu
i-HowoMQ)
-<
„ — .VIO)-l^40)
-PcQ)UMtt)_a
>4->•HtoC(UQ
M(U
"g
2;^H
JJ
H•<
^g
c
C5
r-|OwoMQ)f3j
(UCONo
r-|rdM4J3Q)g;
'g'
^
O LA IA iT
I I I " , „UJ LL. .UJ U- W-<>»•««• o ^(7- CM in ir\ <U. . . . 4->
r\j —J -J — i OeCO
• n•4- fo en r°. fd
W Atl l l l U
U-' U.1 UJ LU 2 r-lin <M a; LT. <=S O.-< OJ U~i >t K W. . • . O
IA •-* — i ^J W ^N (L)HQ <
i W OOO- »~ * **J r
HH IIl l l l 0
UJ uj "- JJ MX;(\l CO — 4 O O 4J
>j I— rn j d J3. . . . H pq g
r>. r-i (\j (\i H rtj -Hl l l l H N
> — <I ;i— i rvj pg CM ^-) - ^~.
• <-H O ' f*>CN 1 0 g
Li.. UJ .0; uJ >H n Oif , m pp. c\j pj • \f" 0 m <^J (^ ° II. . i. . ffl — »—
xj **^ »^ *~^ rfj ' Q)EH « • H >.
: O O> 4Jfa C -H
fM -H — 1 CV rt! W
cl l l l ^ XI 0)
UJ UJ U-l OJ — ' 4J p-n o LO c 'H-• r- C « S C Vi
« • . « H <U <DLO vT or. • •«• <J N X)l l l l O g
^3; rt 2co — i —, ^ pq
Enl l l l t - 3
UJ UJ U_, LJJ (_|U1 xj- QJ r0, pq< J IA <) OO. . ....CNJ — 4 dl) W>
I
^<1
0 0 9 0rNi > -b co
•
4JH<C
—
C
C
s
tH0WoQ)<
0)C0N0
H(0M4J3Q)2
<!*'-N
"B
5
Q O n rn
1 1jj LJ aJ -iJ3 3 <f NJ3 3 -A -I. . . .
O 3 i<1 <M
.3 3 N ,-A .
1 1U U JJ -iJ
3 O (^ CTi- 5 3 ^ 0. . • •
O -3 -< O
3 O r-l -H
1 1LLJ .U Ui JJ
•3 O -f !*-3 3 A 0• . • •
O D -0 h-1 1
3 0 -i -«
LU .LI ai i/j3 ^ vo <MO ;3 O ~«
. ' . . .3 3 -T .-n
O 'O 3 CDp-4
lU JJ JJ IU3 a \i -t3 3 .3> -"... . .
• 3 3 - 1 T51 1
3 3 lA J-•"* •"*
.u ..u 'Jj JJ3 O vO CNI3 3 ^- O. : . . .
3 3 -» U-\
io o o o :0 0 0 O3 :O Tl O :
rr> •-!• -A r- ;
tO O O OrNi ;-N) N <M j
; 1
o*-^'-*^ i— * o o o
I I I 1JJ ^J JJ •'•_! u'J dl JJ UJ
3 r— .3 "r\j -j\ o .J- 3•3 J- NJ ! -O O J> *. . . . . . . .
.3 -- !—<• — I ;A — 4 *-< :M
O -l -I -H 3 3 3 3
I I Ii.U .J .U 'a.i i. ; .j.! .JJ JJ-3 to > > — < O3 m r-~3 co ' •— -* -H ^-i co r>. . . . « . . .
3 rO rO ~T -i 4- «A r-
.3 -H -H -< — < r\) \J :\|
JJ in ill ui .0,1 :u .J.I i. iO M l~- A 3 A 3 A,-> >- a^ T* A . o •». . . . . . . .
3 <M fM (^ >. M * OI I I | | l l
3 -H .— < -H 3 —< -^* — *
'.LI _U L.J '..:.) l.J ,.u U-.! .'Io - H r — m xo^ -^ooO 1— LA CO -4 3 rO «O. . . . . . . .
O H — < — • .-O — < — < — *
3 O\ M -I • ?n '0 ~" 3— 1 — 4 r-^ •-«.-<
^jj u JJ jj a.: .u uJ .,j3 O O M O — i O (VV•3 O •-< t . -O O CO <T. . . . . . . .
3 A CvJ -H -4 (^ Li ^*I I I l l l l
O-N- 0 O '.O: o ~ •"«•—4 — 4 -H — 4 — * .-H -^
JJ i.u JJ uJ aj .u JJ .uO » M •>- ' IA <3> t J-Oi-HC-J-* < M O L A - -... ' . * • . . .
O.ro "~* "^ — t ,-H rri ~<
oo 3 o a o o o•3 O O O , O O O OO 3 'A O O 3 JA ',3n: r LA r- ry .f 'A i-- .
1 ;
OJ3 O O O!O O O;
• -^xT <t <f sO:o -0 ^O;' i i 1
»— ( I—) —4 ^J
-U JJ UJ JJO .A r^- rn
i— 1 -A .A TN!. . . .
-" AJ rr» xf
— 4 — * r*4 ^|
!JJ LU OJ UJ
-4 ~O A r\lr\j ._< r— ro. . . .
r\| O O — I
"i ."O .-"i <r
'.U U_ iJ UJ•<1 3 \l -.»iA •*• 'XI -(. . . .
~i .i- r 1l l l l
3 3 -• -•
UJ LJJ UJ UJJ <f 0 •=}•
CD 00 ro f-. . * .
ff> oo — i —>
— 1 f^ CT> :-O^-4
;JJ 'U JJ J.Ieg sO <f OlA f>J rO st. . . .
c\j -J cO ^0l l l l
,O O <"- !»-~l — 4 —4 —4
i.Sj J ;.J UJCN| IA LA r-a> c\l IA evj....
— t ~» vt CVl
O-3 O 3o 3 o aO O LA O•r> -j- :JTI i»-
O 0 0 033, CO •» O
24
rl
Or)
V)(D
Q)•P3
rH0V>
X!W CCO -^
—<d MK QJ
-PW Q)»-•* r^
TA
BL
E
2.1
-IX
RO
R
(AX
) F
OR
(O
.l-ly
) A
ER
OSO
L S
I2
WIT
H
A
5%
EX
TIN
CT
ION
E
RR
OR
Zen
ith A
ng
le
= 3
0°,
Azi
mu
th
=
0°
ity
(p
erc
en
t err
or)
A
ero
sol
Para
n
K inw c
<Uw pEH< HEH OJCO Xj
$
»
C
17o•A .ft f> r> £]
1 l l I <°'JJ ,y -i .u Q,f\i co co ro rio -A o n T!. . . . »
r* fO fO *0 610ft
c
3
rHOu
8Q)
<
0)coNo
A * -^ ^ g 5
1 1 1 1 ^LLJ -u; uJ a ^ i]J> r- ^- o ^ ,3,-> _ fNJ -t "^
• • • • S e. . . N 6
7.J
DL
--
3 4
.56
c 0
-4.
/4L
- 3
ij7
.33
t-
2 :
9.9
Uh
C
-1.U
3E
- 2
1b
.7^
E-
1
1..0*b 1
-l.
uy
t-
2
1o
.io
u-
2
9.t
2c ti
-9
.bo
t-
3
1
TA
BL
E
2.1
-X
*RO
R
(AX
) F
OR
(O
.l-l
y )
AER
OSO
L S
I
WIT
H
A
10%
PHA
SE.
FUN
CT
ION
.E
RR
OR
Zen
ith
A
ng
le
= 3
0°,
Azi
mu
th
=
0°
sit
y
(perc
en
t err
or)
A
ero
sol
Para
:
1 : 1 1 1 5 cKM »-*rtlUJ
frl /""
rHOJrl
4J3Q)
-1- AJ -H -H p --
1 ^ ' ' EH ^LU '.jj uj iu tr xo :<J- O f\j W XJ(7i O n .-0 g. . . . P
a ^ .< ..^ >„ .„ (
•
4JrH<
_ j ig • o !o o o ! -P^ ^1 ist -O 00 i rj
-C
JT\ UTl •*• JT>
I I I !UJ UJ JJ .JJf*^ f\l f^ 3*o -n o ~c• • • •
f«^ n\ ^-J ;T\
C
8
rHO(00MQ)•<
Q)C0N0
•* sj- ^ -3-
1 1 1 1LU U.! ..U U io -4- -< o(\| <J> — 4 ^• • • • .
(M rg r-^ (\j
r\| IM r\| f\)
I I I !isj u aJ ;U0 ^ - ^ - 3>- LO f- -f. . • • •
<-> M rg ^jt i l l
-H ,_( -^ _,
.U Uj' .JJ 'J-JJ- O N IA00 L"> vO 1*1. • t •
-H (M N "M
M -1 O ~<
1 J 1.LJ iU JJ ;uO .*• 'J> ro\J O f~ r\l• ; • • •
(•T) ,^ | -^
1 1 1 1
flSl_lMl >
•t-*
0fl)w!S
CO rH -^ ^H
1 il 1 1lil LU U-i U.Irn in 'NJ o— 1 iO -3» f^-
• • •• • • ••N t\i r~ r*-
i
g , o a o o^ : f\J 4>» O ~O^5. r
— j I < '- \
inOMM(1)
fl)-P3
rHO
•& M
2 XIH ;; «0 •"• ^3 ^ u"=5 f£ ^a "• a)
o -^w Q cu« w i
TA
BL
E 2.1
- x
i
RR
OR
(A
X)
FO
R
(O.l
-ly
) A
ER
OSO
L
SI
) A
LBED
O
ER
RO
R
FOR
A
CLO
UD
0.
5 A
LB
Zen
ith
A
ng
le
=3
0°
, A
zim
uth
=
0
°
sit
y
(perc
en
t err
or)
A
ero
sol
Pars
W rH CQJW ~ Qr .
< £ ^EH EH Q)CO H- XI
& G3Z
, iri
^- rri rn f\
1 1 1 1LU UJ LL. LUC^ rr; N^- ff^O (NJ O •&. • . •
»*• (M fVJ — '
C
s
rHOwoQ)<c
(UCONo
m fr> rfi rn
l i l tLU UJ U- JJ— < CO n-. JDf 1 30 *•' C-• • . •
— ' ir\ LT\ m
• CM —I - — '
I I I !LLJ LU LL» LU
<x» <\j m oua — < >o a;• . . •
tf> ^- 01 <\J1 l | l
(M {\J (VI (\J
LU LU U, LL'vO ~i »o r-o ro »u o• . . .
>— i in «T fi
rH O O -«
LU LJ LU LL:• LTV CT vT ^>
CO 03 <M 00. . . .
_< i--l PTi _<t i l l
rHdM4J3Q)"Z.
fM O --• ^i
LU LU LU LL.CO f| Lfl CT.
fM <M rh rv.• • • •
-4 O (VI (\)i
• •£<•^s
e o o <i oAJ fvj >* j) eo
25
HHX1
•
TA
BLE
2
WO
1wNHCO •
l).
AE
RO
SO
L
j.rH
1H
*0
OCM
^cr?s
1H
WEH
W
QWtS3H
HEH
COUHEH
[AR
AC
TE
RIS
wU
' A
ER
OS
OL
' "^~
i — 1
O1
rH0
O
KEH
Co
II
4J
N
C *oCO
II
iHtnC.
%•HG<UN
*JHO^10)
CD4JPIoto
ja(0
<D-PQ)gf[J
>-l(u
AJJ
Ae
roso
l
x— st l
O
(1)
•sQJU}-!CD
3
•HtoC
Q
Q)
a
5
•••>
C
8
rH
Ae
roso
!O
zon
e
nj
(Ua
"H
• * ( i" ^^
i i i iUJ UJ UJ UJ>O ^ 0 ^^NJ O ' OO
..- -00- in — » CNJ-- -
rO. rg rr, .r, •
1 I I I
4- M >?> O, o -m * 33
• • » »r\J —4 fO. ^O
r— t — 1 — 1 — 1
I I 1 1^ -.LI L-J -U
• • t •
I I I 1
.•N ;.-n ?N -M
i J .U .U ,JJin 'cr* " 'n
<N i- CM X3
~~* O si >-^ .
1UJ UJ U LJ 'rO >t (MOr^ c\i \o
m c\i IM -4 'i l l l
rsj -4 -4 - •
1U) U, Ui LJ•ON o -4 r-4 :•J- -t 0 f\|
.rsj — 4 ro in •
o !o o o jOsl l-t '« <S .
i j
w2
w1 1
CO
t ^oCOo
M SiHH _.
. X 3.' rH
rH |
* i — 1CM .
OW ^
EH ptl
^^Xo
O
W
wEH
EHW
/•»!
SIT
BIA
S
imuth
=
0C
H N
r— «
g c *CO 0£3 coH n#>0 <UrH H
Cnrt^ £
K „
^ 5& "H
0)N
~
oHMCU
O-prj^*rHOW
JE
0)^j65M(0n.
Ae
roso
l ]
*
0
0)
C0)o0)
-p-HWC0)Q
M0)
|
-P
'i
C
8
l
Ae
roso
lO
zon
e
rH03
-P
CJ
"e
*f rO -*o
i i i iaJ Uj >jj UJ-o o -* in4- -t ro o• • • •
~~-i4~i-4~ — 4"". -4"
-^" C*) PO fft
l i t iuj LU UJ UJ^^ ^^ Ci ^Dvo NO m 10
* * * *^"
; - ,f
•'C^J r— - "H
1 1 1 1lij UJ tU UJrO CJN ro inin >A r<i oo
1 I I I
-4 (M ;NJ : r\l .
r\j rO O . *"CO in O CO
-a :o TO r\i
•NJ •- f^ O
1 IUJ i U :. JJ ! LJ
\0 \0 M • *•'•
\0 'O H CO1 1 ' .1
r, p -1 _ ',1 i J
(M (M CSJ rO '
Nj- p .-^ rsl ;
• . i
o b o o ;oj it so oo ;
26
.HX1
CM
W
«
EH
W
•3
c2wNHCO
^_q0
ow<
pi
o•Hi
r-H
.
B!O
2"
H
O
W
1-H1— I
o
JJ.i-HN^i
O
n
H
rH
£4
^ftf^
£
JJ
-ACQ)N
MQ)•PQ)etoMfdA, — iotooM0)<J
"VC
o o
UJ UJ0 0o o0 C3
c
O O
UJ LU0 0O 0• •
o o
Q
o o
UJ UJO r3O 0• *
0 0
^^rogU
,
>4J•Hto
<L>P
M(U
•ji3
^
oto0JJ.OJ
o o
:jj 'JJO 0o o
« *O -'->
(1)£ONO
3 O
- t-i -JJ:3 .3'.3 •-)
O :3
iH
U4J
Q)
3 0
JJ -UO :•>O O
• *c3-p>
0 0^^ O O°S. b o
f<"> <t
i•.
<c1 ° °X ^ ^— ' i
<M m
1 I-J UJ<M O>—4 vj- -
-t -M1 1
— 1 —4
1 1!i.J JJ
CO IA
• t(M -i
1 1
— < -H
1 1.11 UJ
D JO
• •
1
O :3
1 .1 UJ
? ^3• • -
>T <\(
,M 0—4 *-^
;J.J JJ
C5 AJ(M — *
-1 (M1 1
*- O»-4 -4
JJ .JJ> lA-» (A* •
-4 —4
-> "3O OA O :
:A r-,
..'|
0 O :f\J rsj i
O f\J -H tsj
I I I— tj UJ OJO rfS cr> O.3 IA A ?*-
O X5 — < CO1 1 1
O r-4 O O
1.ij i:v : ;J '., i
T> o -o obO ro -A n• • • •
O .-3 -T J-1 1 1
0.0 -• -H
ilj i L. : L— ^.iO J- sO LA•3 -i -t -:>
O rT ^i ,-<-i1 1
O r-J — 4 O
1 1
' U UJ LU UJ
"3 '3 tn -oO .0 CP 13
O r-4 'iJ —( 'i
O O m -g— 4 — t -4
JJ .U JJ ,U-3 '-JN >- .fTJ
o so - "^
"3 A IA -vl1 1 1
O '.O O >• ,~-t —4 -4
ijj UJ JJ UJ .3 c> o <»3^00• • • •
3 ^ (M \0
O o O o :O 0 0 O •O O LA Oro -. ijO ^**
,000 O !
. »$• •* ^ <h ;
i i
(NJ ~> O
1 1UJ ' U >aJIAIA :3-J- .-i n
fM CM — i1 1 1
•- 1 ."3 —i
1JI '../ IU
O rn csj"O r*- j-• • •
;M ;AJ -"01 1 1
O —4 rNJ
iJ U ' .Uoo J-
O --J* sO
•^ . — ( •— 4
1
^ (M -t
1 1 1UJ 1.; I'..'r- ro r-~O^ O ^0
CM rn rn
,"•> ;> 'NJ•-^ ~4
ID ilJ UJ13 ' \O
Sj f\J <NJ« • «
-< fM OO
1 '.I
o o o-4 .-4 —4
-JJ -U UAGO O•ri - -<• , • *
— ' tM IA
O O 0O 0 .3^3 O Arn 'f A
O 0 0.3 ;o c
i
o
LjJ
^Nt
— (1
—4
;U
3•
r-!
CJ
L.D—4O
•o1
—I
11 Jo~4
0^
H.—4
,uCDo>*
1
•X)-H
U; i-1
*
CO
O :
O :"~ •
o '••o :
—4
.1UJ-o•A
LT\1
O
._-jA
••t1:
-*
ilJNO
0 f-< -<
LU iJJ UJTM :M "• o-> a :-n• * *
(M -J N1 1 1
•-I M (A
L< j i I .i..'\0 — < (\irn *•— — j• • •
n 4~ — iI I i
— t co ^
c. i uj .jjo in !*-
(M IA O O
-o
^1
'—J
S
<T
-,—<
-JJ\Q
X3
1
X)~4
,jj
IAO
M
0
OOn
o^0
Cj rO -^1 1
•J, ^1 -J
1 1 1lU I.LJ ; _jO -< K~\O ."M M
rsi rn oo
r- ~3 o—4 —4
uJ -U LU;<1 t\J LAt O O
1 1 1
!O O .O. - 1 - 4 — 1
,u -J .u•* CM IA•d- ro "*-
* • •vi \i\ <n
o -3 oO 0 OO 'A O•r IA r-
O O 0CO 00 -X)i
27
noJHH0)
0)JJ0rHOW,Q
W (dU —
3 MPH Q)
JJW 0)N gH IdCO O rH
o idrJ. fto PH ilW O rHO « -C OPi Pi -P Mw w g o< E M
K. 2 -H (U> ^ o N <:« . H <
"c
rH
om \ \o <o >H
t i l l ( UUJ ;uj UJ LU :
rr\ ,.f\ -3 y> Q)3 _< _l ,M JJ. . . . r)
>• — i"Tx \n rH
' " ~ owJJ
c
ro -t -1- '-*• W (do —I 1 1 i 2
111 UJ Ol Ul < MSO VPi CD [\l PH (UO -1 <O \O -P. . . . W (U
•M CO (M — 1 N gH idW O M
8
PH 0 IS
« ^ .^ n g g „ ^1 1 1 1 CO PH rH
ui >jj uJ uj O W •£ O»- O sf u-\ PH -P Msoi^-io H a ' 3 °. . . . <q O 6 S-i
— 1 -< r-l — t H H -rl 0)
^- B < «
e
'vTo>r m m n LILI
1 1 1 1 (U•JJ UJ :JJ :!•
. l <T> r^ -a /nO O NO ,T> 4J. . . . 0
.H l/\ ,-f> CM .i ,i i i 'd
«< nn ja
e
m ro ,-<, ,-n w * -5
1. -1 1 1 Z ,L, "UJ LU IJJ OJ S1 ^ Ll-H t\l ,T| J- g <! g
r- f*. r- <t ,_ jj
• • • • w § o7" "^ T1 ^ N g g
1 1 ' 1 H m ^ IdCO !j ^o M
y ° rt
8
m ro rsi -M « ~ n .fti i i i O m1 1 1 1 C O . < 2 r H
N ^ " S S-° I S°---* H S g 1 8-^ g §.5 21 ?N ^— r )
*c
rO (Tl -J- st
I i i ii l 'LU LU U-1 LUo c; *j- ^>f^ CNJ Q-> f^.
* •rj ^H r^ mt il 1
C
r\; r\i (Vj r\i
I 1 I 1LU U. UJ U.1
•* co or. rr•J/ fNJ O -4J.....
r«-i ni rsj — iI i 1 i
a
<\i •— ' — i f— i
1 ) 1 1LU LL! 4- LUr-4 >JJ Q\ j
O» CU .£} Lf>
. . . .>T •— -H r-i1 1 1 1
" 3. EH : 1 3. 2 . : . ' . . 1 "3. ~ * 1rH 0 CJ - >-
- rH 2 O HCvl | M O O
rH EH ^ MW ^ .. &rq «— W II Q)CQ *> „,<2 Pi o <" -PH 0 CM H C
f e < | - 8^ <! M* K j- S<3 EH 5 ft-- H *J —
•^ "^
« § T9 N ^Pi N -rlPi OW C
0)W 0
< MEH 0)CO .Q
p
p
12
rHoinoMQ)<
•-1,-tOO r H o D " ^. rH fe O K
CN | 0 ouJ ;u viJ u ,_| W <*> •- MT, C\J .</• u-\ W LO „ }H-j- i-i M o hP ^ rtj • " Q>. . . . cQ W
(A i— ( 03 O rilDHCu lU jJ1 1 1 1 E-i O OP "i C
Pn 0 JT Q)m G o
(UGONO
- « - < - < M ^-v «i MX < (U
M i l l < -g &U iiJ iU :,J "-^ ffi •*] — '
in 03 03 O EH "d:J> -7>?-\J Is- PH H 5 >•. . . . 0 & S -P
O •*• Vf\ rxl CH N -H1 PH in
W CQ)
1-1idL|
JJ
3Q)2
•\J ' -« ~4 -4 W QEH
I I I ! f f , r l:.U ,11.1 UJ UJ EH QJfM —• — < m 03 oo o <M tsj: ^. . . . . r j
ir\ ;in >o in .21 : 1 1 1 ;
jj*rH<
^ . ig o o o o; jjX <M * <O 50' r-l— , i ; ft.
r-l
O(0oM<u<
- ^^- H o < 0 -• * rH O M
., , , , , : : <N 1 Pi 00 OUJ UJ JJ -U - rH O MN £ 5 ^ ! W h l l r H-H r~ o t3 j s . d)' ! * * , ' W « Q)r - x o v o L n . S c d H ^ j j
EH O PH D^ Ch PH C 0)
i M ef. r»
Q)C0N0
•H r> O •N ^ MX O XI 0)1 ' • < Q -P a
U UJ .U .LI ^_/ H -r| C;D >3- 1*1 00 ffl C-i^ooin p i J O ) ^* L! ! * 0 ^ N 4 J? rH r1 °? e — -H
1 I I 1 Pi — WW rH C
• m
rH
(drHJJ30)2
<M - :> o H - Q, , EHI I 5 f f i 'L l
UJ l.j in ui • S S mK.vo:r , rsj • • ^ S ntf. «. > . —. : UJ . ' H-(1O CSI —* ;> Jg p
« « • • ; 5.\0 )7» *NJ — < ; rg
r-|
010oM0)<!
CM r^ r\j r\i
UJ UJ LU j_.in * ir. r-u -H a) u\• • •:• .
cr» ^^ is_ ^u
(UcoNO
r-J -J ^ O
LU U-l UJ U_tf> (J> r-.ni
-H — 1 vC — '. . . .
—' Ln J- rv1 1 i 1
rHidLI4J30)z
— • — < ^j — i
1LU UJ LU UJ
ITi f>- rg ^*n o «<• h-. . ;• .
"3 J- IT J-
1 ' • 1I•'g o b o o ; _J^ .» ft -O -0 : ^
—• . i ; • rt;
"g" o o <P o^ CNJ ^ «O CD-^ • 1
—
p*—I(f)2UJ
9O 10O
ALTITUDE (KM)
Fig. 2.1-1 Scattered Sunlight Constituent Sensitivities
for Aerosol Size.Range (0,1-l.Oy..) , Zenith Angle = 60°,
Azimuth Angle = 0°, and la values (p=100%, a=l, n=0.05,
n'=0.005), Wavelength in Angstroms.
29-
0 30 40 50 60 7.0 80 90 10O
ALTITUDE (KM)
Fig. 2.1-2 Scattered Sunlight Constituent Sensitivities
for Aerosol Size Ranqe (0.1-1.On), Zenith Angle = 60°,
Azimuth Angle = 0°, and la values (p=300%, a=l, n=0.05,
n'=0.005), Wavelength in Angstroms.
30
>-H>
izLJto
0— H 1 >
10 20 30 50 60 70 80 9O 100
ALTITUDE (KM)
Fig. 2.1-3 Scattered Sunlight Constituent Sensitivities
for Aerosol Size Range (0.1-l.Oy), Zenith Angle = 60°,
Azimuth Angle = 0°, and 10 values (p=600%, a=l, n=.05,
n'=..005), Wavelength in Angstroms.
31
0)zuO)
•6000-I • h- H 1 (-
n1
n
.3.2
.1
O
.3
. 2| 4000.1.O.
7000 ,6000 X5000 4000
4000
.6000 AEROSOL
0 1O 20 30 50 60 7O 8O 90 10O
ALTITUDE (KM)
Fig. 2.1-4 Scattered Sunlight Constituent Sensitivities
for Aerosol Size Range (0.1-l.Oy), Zenith Anale = 0°,
Azimuth Angle = 0°, and lo values (p=300%, a=l, n=.05,
n'=.005), Wavelength in Angstroms.
32
COzLJCD
O0 10 50 60
ALTITUDE (KM)
'70 80 9O 100
Fig. 2.1-5 Scattered Sunlight Constituent Serisitivities ":
for Aerosol Size Range (0.01-0.ly), "Zenith Angle = 60 ,
Azimuth Angle =0°, and la values (p=100%, a=l, n=.05,
n'=,005), Wavelength in Angstroms;
33
i
COzLJ
,3.
2.
.1.
0
.3.
2.
. 1.
O
.3.
2
.uo
.3..
,2.
.1-
0.
.3..
,2..
. 1..
0
-t- »-
n1 •
n
7000>6000
1 1 1-
ALPHA
60007 / /
3000
OZONE
.4..
.3..
2..
-1-0._0
NEUTRAL
10 2O 30 40 50 60 ?O 8O 90 100
ALTITUDE (KM,)
Fig. 2.1-6 Scattered Sunlight Constituent Sensitivities
for Aerosol Size Ranae (0.01-0.ly), Zenith Angle = 60°,
Azimuth Angle = 0°, and la values (p=300%, a=l, n=.05,
n'=.005), Wavelength in Angstroms.
. 34
COzUJen
10 20 30 4O 50 6O 7O
ALTITUDE (KM)
80 . 9O 10O
Fig. 2.1-7 Scattered Sunlight Constituent Sensitivities
for Aerosol Size Range (0.01-O.ly)/ Zenith Angle = 0°,
Azimuth Angle = 0°, and la values (p=300%, a=l, n=.05,
n'=.005), Wavelength in Angstroms.
35
enZuto
0O 10 50 60 70 80 90 1OO
ALTITUDE (KM)
Fig. 2.1-8 Scattered Sunlight Constituent Sensitivities
for Aerosol Size Range (0.01-O.ly), Zenith Angle = 60°,
Azimuth Angle = 0°, and la values (p=300%, a=l, n=.05,
n'=.005), Wavelength in Angstroms.36
6O 7O 8O 90 10O
ALTITUDE (KM)
Fig. 2.1-9 Scattered Sunlight Constituent Sensitivities
for Aerosol Size Range (1.0-lOy), Zenith Angle = 60°,
Azimuth Angle = 0°, and la values (p=300%, <S=1, n=.05),
n'=.005), Wavelength in Angstroms.37
COZLJCO
00 1O 2O 3O 40 5O 60 70 80 90 10O
ALTITUDE (KM)
Fig. 2.1-10 Scattered Sunlight Constituent Sensitivities
for Aerosol Size Range (1.0-lOy), Zenith Angle = 90 ,
Azimuth Angle = 0°, and la values (p=300%, a=l, n=.05,
n'=.005)> Wavelength in Angstroms.38 .'• "
COzIdCO
--— ..... 5000-- 7000
.---^f^/-. ^-—60001O 20 30 40 5O 6O 70 8O 90 10O
ALTITUDE (KM)
Fig. 2.1-11 Scattered Sunlight Constituent Sensitivities
for Aerosol Size Range (1.0-lOy), Zenith Angle = 50°,
Azimuth Angle = 180°, and la values (p=300%, a=i,
n=.05, n'=.005), Wavelength in Angstroms.39
t:tozuen
AEROSOL
0 1O 30 40 5O 60 70
ALTITUDE (KM)
90 10O
Fig. 2.1-12 Scattered Sunlight Constituent Sensitivities
for Aerosol Size Range (1.0-lOy), Zenith Angle .= 0°,
, Azimuth Angle = 0°, and la values (p=300%, a=l, n=.05,
n'=.005), Wavelength in Angstroms.4 0 , . . . - . - • '
o
CO
UJQ
DiUJCO
40
- *
i-t-
-\—
T
60 80 0 20
ALTITUDE ( KM )
54
3
2
1
<cn:Q_
+1.6
1.5
1.4
•0.02
-0.01
040 60 80 100
Fig. 2.1-13 Covariance Propagation for Aerosol Size Range
(0.1-l.Oy), Zenith Angle = 60°, Azimuth Angle = 0°,
and la values (p=166%, a=l, n=.05,n'=.005), Wave-
lengths of 3000, 4000, 5000, 6000, and 7000 8,
and Tangent Heights Spaced at 4 km.
41
CO
O
on~2.LUO
CQ
•* +• +
20 40 60T t t T t
80 0 20 . 40ALTITUDE (KM) '
6
5
4
32
1
••1.6
-1.5
60 80 100
1.4•0.02
-0.01
0
Fig. 2.1-14 Covariance Propagation for Aerosol Size Range
(0,01-0.ly) Zenith Angle = 60°, Azimuth Angle = 0°,
and lo values (p=166%, a=l, n=.05, n'=.005), Wave-
lengths of 3000, 4000, 5000, 6000, and 7000 &,
and Tangent Heights Spaced at 4 km.
42
o
CO
UJo
O3-£EIT)'
1 1 1-
H 1-
0 20 40 60 80t t t T T
5
4
3
2 :
1
jl.6
•:.<:4
0.02
0.01
20 40 60 80 100
ALTITUDE (KM)
Fig. 2.1-15 Covariance Propagation for Aerosol Size Range
(1.0-lOy), Zenith Angle =' 60°, Azimuth Angle = 0°,
and la values (p=166%, a=l, n=.05, h'=.005), Wave-
lengths of 3000, 4000, 5000, 6000, and 7000 S,
and Tangent Heights Spaced at 4 km.
43
2.2 STELLAR OCCULTATION SIMULATION
2.2.1 Introduction
-4" One of the contract- subtasks was concerned with the application of
the Kalman-Bucy filter to the inversion of stellar occultation data. The
data, however, was unobtainable during the contract period. The inversionscheme was nevertheless developed and a sensitivity analysis was
performed. Several computer runs were made using simulated data which
illustrates the invertibility of high altitude aerosol layers with a star
occultation experiment. :
2.2.2 Sensitivity
The sensitivity function for stellar occultation is identical to that
used for scattered light. The partial derivatives are, however, applied to a
different radiative transfer model. Figure 2.2-1 shows the envelope of
peak sensitivities for a complete stellar occultation scan at several
wavelengths. It can be seen from this figure that both ozone and neutral
density will be invertible down to an altitude of 20 km. The curves indicate
ozone cannot be inverted below this level. This is because the intensity
attenuation of ozone sensitive wavelengths below 20 km is too great to permit
measurement of the signal with the assumed instrument signal to noise
ratio of 100:1.
Figure 2.2-2 shows the effect of a high altitude aerosol layer (50
km, one order of magnitude anomalous increase) on the stellar occultation
sensitivity curves. It can be seen that standard levels of aerosols do generate
a sufficient signal increase to be. detected, however, an anomalous increase
of this nature will be measured and inverted for the assumed 100:1 signal
to noise ratio. . . . . .
44
2.2.3 Selected Simulation Results
For the purpose of simulation, fictitious, but realistic, density
distributions other than standard were used to generate simulated data e.g.
anomalous layers were included based on measurements of Rossler, 1968
and Elliot, 1971. The data was then processed by the inversion routine.
In Figure 2.2-3 the crosses represent the estimate of the density with two
error bars given for each-point". The horizontal bar represents the un-
certainty in the altitude at which the average density occurs while the
vertical bar represents the density uncertainty. The density error
bars are derived from the filter while the altitude error bars are derived
a priori in simulations. The solid line represents the fictitious density or
"right" answer. If the densities fall within the RMS error bars 67% of the
time the simulation is considered a success.
Figure 2.2-3 is a simulation in which an order of magnitude anomalous
aerosol layer at 50 km was used to generate fictitious data. The inversion
results demonstrate the ability of the inversion technique to determine such
layers if they exist with a 100:1 signal to noise ratio and wavelengths of
3000, 4000, 5000, and 7000 A.
2.2.4 Conclusion
The Kalman-Bucy filter is easily adapted to the inversion of stellar
occultation data and produces excellent results. Although a simplified
radiative transfer model was used in these simulations there is no reason
why a more sophisticated model incorporating refraction and dispersion
effects could not be used. This would extend the range of invertibility well
below 30 km for both neutral density and aerosols in addition to providing
more accurate results above 30 km!.
45
V<IdQ-
enZLJQ
CDZLJCO
1 1 1 1 1 1
1O 2O 30 SO 60 70 8O 90 1OO
ALTITUDE (KM)
Fig 2.2-1 Stellar Occultation Constituent Peak SensitivityProfiles with 100:1 signal to noise and standarddensities.
46
LJQ_
tnzuo
enZuen
80 9O 1OO20 30 -tO SO 6O ?O
ALTITUDE (KM)
Fig. 2.2-2 Stellar Occultation Constituent Peak SensitivityProfiles with 100:1 signal to noise and an (orderof magnitude increase) anomalous aerosol layer at50 km.
47
CW>I/T> siosoyav
(WD 3NOZO
Is
V
LJQ
Ht—_I<
uj2 3_l<
C-H
Oin
c t-iO Q)•ri >i4-1 rtirt) H-PiH iH3 OO COO OO >-l
Q)t-l (0(0 '
iH tnrH 30) O-P HCO (8
e^3 O(U C
4->. (0
rH C
IW (U
o oC Q)O W
•VH 0)
H O4Q)> 0)
H 4J
roI
(N
tn•rl
(WO nO/N) A1ISN3Q48
2.3 HORIZON PROFILE PARAMETER SIMULATION
2.3.1 Introduction
In this section we discuss the results of a horizon profile parameter
study aimed at determining how horizon profiles are altered by changes in
the various measurement parameters. These include season, latitude,
satellite altitude, solar zenith and azimuth angles, and either the earth's
albedo or the altitude and albedo of a cloud layer. The newly developed
mulitple scattering code (C.K.W.) was used to produce 324 radiance profiles
representing various combinations of the measurement parameters. The
horizon profiles are plotted in multiples of from three to five profiles per
graph and are compiled in Appendix 5.1. Figure 2.3-1 is a sample profile
set showing zenith angle as a variable. The measurement parameter being
varied in each graph is identified at the top of the graph and values of a
given variable corresponding to a given profile are listed to the right of
each graph. The fixed parameters for each profile set are listed in the
legend. The right hand graph in each figure displays the fractional change
of each radiance profile relative to one member of the set.
Figure 2.3-2 shows the limb scan geometry, defining the tangent height
(H), the satellite altitude (S), and the line of. sight (L). The direction of
the sun's rays are specified in relation to a Cartesian coordinate frameA ' A '
attached to the satellite, with (Z) along the local vertical and with (Y) inA A
the plane defined by (Z) and (L). The solar zenith angle (9) is measuredA A A
relative to (Z) and the solar azimuth angle (0) is measured in the ( X , Y )A ' A A
plane relative to (Y). The angle which (L) makes with (Y), i.e., the scan
angle, is defined as (6).
All of the horizon profiles given in this report pertain to a satellite
altitude of 500 km. The scan angle (5) therefore varies from about 19.5
to 22 as the tangent height varies from 100 km to zero. The single scattering
angle is determined by the angles (0, $, and 6 ) through the expression
= (cos(f>sinOcos6 - cos0sin<5). Thus, for the coplanar cases (<£=0° and ^ = 180 )
0=90 +6-e and 0=90°+S+G, and for the ^=90° case cos^ is simply given by
(-cos0sin&).
49
2.3.2 Discussion of Results
For reasons of organizational convenience the 84 figures for this
parameter study are located in Section 5.1. In each case the atmosphere
was considered to have two-kilometer layers (DZ = 2) that extended to an
altitude of one hundred and twenty two kilometers (ATM = 122). The seasonal
and latitudinal distributions for the mean neutral and ozone densities used
in determining these profiles are shown in Figures 5.1-73 to 5.1-84. Figures
5.1-1 to 5.1-12 show the variations in horizon profiles resulting from
seasonal variations in the neutral and ozone densities for each of twelve
latitide-wavelength conditions (latitude = 0, 40, and 70°N and X= 3000,
4000, 5500 and 7000 Angstroms). At 0°N latitude the same neutral density
distribution was used for all seasons. The changes in the profiles with
season at 0 N latitude reflect, therefore, only the changes in the ozone
density distribution. This profile set represents the only case where the
differential horizon profiles can be interpreted simply in terms of the
changes'occurring in just one of the atmospheric constituents . These changes,
which are referenced to the winter profile, are greatest at 3000 A and
smallest at 4000 A in agreement with the known spectral characteristics
of the ozone absorption cross section (Wu, 1970). The remaining horizon
profiles in this parameter study are arbitrarily based upon the latitude
and seasonal choices of 40°N latitude, winter.
Figures 5.1-13 to 5.1-24 show the variations in horizon profiles with
wavelength for twelve cloud-zenith angle conditions (a zenith angle of 30
and 80° and a cloud at 4, 1, and 10 km with an albedo of 0.4 and 0.8). The
solar irradiances at 3000, 4000, 5500, and 7000 A are 5.14, 14.92, 17.25,2 :
and 13.25 u watts /cm -A respectively (Thekaekara and Drummond, 1970).
Figures 5.1-25 to 5.1-28 show the variations in horizon profiles with
wavelength for four zenith angles (0, 30, 50, and 80 degrees) and a ground
albedo of 0.3. These figures show that clouds are only observable at the
longer wavelengths like 5500 and 7000 A which penetrate deepest into the
atmosphere. In addition, these figures show that the enhancement of radiance
50
due to an increase in cloud albedo is strongly dependent upon the zenith
and azimuth angles, which determine the scattering angle (0) and hence
the fraction (sin0) of the solar irradiance that is actually incident upon the
cloud surface.
Figures 5.1-29 to 5.1-64 show the variations in horizon profiles with
zenith angle for 36 combinations of azimuth angle, wavelength, and ground
albedo (azimuth angle = 0, 90, and 180 degrees; X= 3000, 4000, 5500, and
7000 Angstroms; ground albedo = 0.2, 0.5 and 0.8). The differential horizon
profiles are all referenced to the zero zenith angle profile. These figures
show that the shape of a profile is not affected greatly by changes in the
albedo and in the solar azimuth and zenith angles. For the brightest profile
(80° zenith and 0° azimuth) there is little albedo effect because sin (0) is
relatively small and therefore most of the signal is produced by strong
forward scattering high in the atmosphere. Albedo effects are most
prominent for azimuth and zenith angle combinations that make the scattering
angle (0) fa i r 12.
Figures 5.1-65 to 5.1-72 show the variations in horizon profiles with
ground albedo for eight combinations of zenith angle and wavelength (zenith
angle = 30 and 80 degrees; X= 3000, 4000, 5500, and 7000 Angstroms).
The differential horizon profiles are referenced to the albedo = 0 profile.
These figures show that the enhancement of the line of sight radiance due
to the earth's albedo is strongly dependent upon wavelength and scattering
angle.
51
l-Ll l/>
-J UIj— UI
o oCT UQ- 9
z ui
s ^E <O -r
-J fi< N
U mcc <u SU. <u. >
o-to
..o
|o
O
..O
.oCO
..o
uQD
_l<
O O O O O O OO O O O O O O O. O Oi o ^ j - c o c \ l ' - | o < r » o o r v v O C n ^ - c o c \ j r H ^ - *
enQ)iH-Hmo
coN-H>-lOHi
Q)rH
30NVHD A1ISN31NI
0)W UJui u
I <£=J ou. uiO 9ccQ. U
z go <N
ulN
V
CM
O
cc<>
O
O
CMIO
roI
888 8S* O o" Ou» to O « .
CC
0.
2 t; to, ?
o o^1 •-!
30Nviava
iniO
o
o
..o
o
k•~ 12— T1*"
:i:i' 10• " too
CM !
te508 !
V
UJoD
-p•HCQ)
(D
•HS-l(0
CM
52
Fig. 2.3-2 Limb Scan Geometry
53
REFERENCES
SECTION 2
1) Elliott, D.D., "Effect of a High Altitude (50Km) Aerosol Layer on
Topside Ozone Sounding", Space Res., 11,857-861, 1971.
2) Elterman, L., "Vertical-Attenuation Model with Eight Surfaces
Meteorological Ranges 2 to 13 Kilometers", AFCRL Report 70-0200,
March, 1970.
3) Elterman, L., "An Atlas of Aerosol Attenuation and Extinction Profiles
for the Troposphere and Stratosphere", AFCRL Report 66-828,
December, 1966.
4) Malchow, H.L., "Standard Models of Atmospheric Constituents and
Radiative Phenomena for Inversion Simulation", MIT Aeronomy
Program Internal Report No. AER 7-1, January, 1971.
5) Newell, R.E., and C.R. Gray, "Meteorological and Ecological
Monitoring of the Stratosphere and Mesosphere", NASA Contractor
Report, NASA CR-2094, August, 1972.
6) Rossler, F., "The Aerosol Layer in the Stratosphere", Space Res.,
8, 633-636, 1968.
7) Thekaekara, M.P., and A.J. Drummond, "Standard Values for the Solar
Constant and its Spectral Components", Nature, Phy. Sci., 229, 6-9,
January 4, 1971.
8) Volz, F.E., "Infrared Optical Constants of Ammonium Sulfate, Sahara
Dust, Volcanic Pumice, and Flyash", Appl. Opt., 12, 564-598, March,
1973.
54
3.0 MODELING
3.1 FILTER DEVELOPMENT
3.1.1 Introduction
There are a number of approaches to the problem of inverting aerosol
parameters, neutral atmospheric density, and ozone density from horizon
profile data. Ideally the method chosen should provide a best estimate (in
some statistical sense) of the desired parameters incorporating not only
the data, but the instrument characteristics arid previous knowledge of the
parameters as well. What follows is a discussion of several techniques
which have been considered as well as the reasoning behind the choice of
the Kalman-Bucy filter.
3.1.2 Comparison of Techniques
The first and most obvious approach to the inversion problem is to
attempt an analytic solution of the equations of radiative transfer, that is
solving for density as a function of intensity. This technique, has been
used (Anderson, 1969) for atmospheric probing; however, even for simple
geometries with single scattering and only one unknown constituent several
simplifying assumptions must be made in order to reach an analytic solution.
Also, once the solution is reached it is difficult to perform a meaningful
error analysis on the final answer. Thus given the stated goals of the
method to be chosen and the complexity of the radiative transfer equations
for horizon scan geometry, it is clear that this approach to the inversion
problem is limited.
Since it is not practically feasible to solve the equations analytically
the next possibility to consider is an indirect solution to the equations using
partial derivatives and some iterative scheme to arrive at an approximate
solution to a set of simultaneous equations. This would probably be possible
and would give reasonable answers, however, there would be no statistical
55
error analysis. Also if there were more data points than needed to solve
the simultaneous equations only the first data taken would be used. Thus
this technique involves incomplete data utilization and does not provide
for the desired error analysis.
Next, a regression technique might be considered. A regression would
provide not only a best estimate in a least squares sense but error statistics
as well. However there are some problems with the application of a
regression and these are: 1) Previous knowledge of the parameters is not
incorporated in the estimate or the statistics, 2) Instrument noise is not
incorporated, and 3) The large amount of data must be processed
simultaneously rather than sequentially. However this would be a good
technique if there was not a better approach to the problem.
The use of Bayesian statistics makes possible the incorporation of
previous knowledge and instrument noise in the inversion scheme. It also
allows data to be processed one point at a time thus incorporating each
new datum into the estimate of the statistics of the estimate. There are a
number of Bayesian approaches to the problem such as maximum likelihood
estimation, Bayesian regression, weighted least squares, and the Kalman-
Bucy filter.
For the case in which the equations relating the measurement and
the state are linear, and noises are Gaussian, all these techniques are
equivalent. However in real physical situations these assumptions are often
invalid and the techniques differ. The Kalman-Bucy filter was chosen
because: 1) It has all the advantages of Bayesian statistics, 2) The filter
equations are simple and easy to understand and code, and 3) Because of
the simplicity of the equations the filter readily lends itself to iterative
techniques which ensure convergence in spite of the nonlinearity of the
equations.
56
3.1.3 Filter Review
The Kalman-Bucy filter is a recursive technique for estimating the
state (in this case aerosol parameters, neutral atmospheric density, and
ozone) of a system (the atmosphere) from measurements of that system
(horizon profile data). At each recursion the previous estimate of the state
is updated. The degree and distribution among parameters of the update
is determined by the measurement; the previous estimate of the state, the
instrument noise, and the statistics (covariance matrix) associated with the
estimate of the state. After the state is updated the covariance matrix is
also updated so that it reflects the current knowledge of the state.
The updating procedure is done in three steps. First is the calculation
of the filter gain which determines the degree and distribution of the update
among the elements of the state vector.
K(m+l) = P(m) BT(m+l) [B(m+l) P(m) BT(m+l) + R]"1 (3.1-1)
where
(m) is an index of measurement
(K) is the filter gain
(P) is the covariance matrix
(B) is the measurement vector whose elements are
• b. = 3h(x, m+l)/3x.
h(x, m+1.) is the predicted value of the intensity based, on the
state estimate and measurement parameters
and
(R) is the instrument noise.
57
The second step is the updating of the state estimate
x(m+l) = x(m) + K(m+l) [l (m+1) - h(x, m+1)] (3.1-2)
where (x) is the state estimate and (I) is the measurement.
Thirdly the covariance matrix is updated to incorporate the knowledge
gained of the state.
P(m+l) = P(m) - K(m+l) B(m+l) P(m) (3.1-3)
. As can be seen from the equations there are three steps which must
betaken before applying a filter to the problem of limb scan data inversion.
First a computer code giving a direct solution to the equations of radiative
transfer is needed for the calculation of h(x, m) and 3h(x, m)/9x (partials
are usually calculated by the approximation Ah(x, m)/Ax). Second the state
(x) must be defined, and third some initial estimate of the covariance matrix
P(0) must be made to start the recursive process.
Two computer codes are available for the first requirement and these
are discussed in detail in Section 3.2. For all of the inversion work to
date the hybrid single scattering code has been used since the multiple
scattering code had not been completed. However, the characteristic shape
of the hybrid profile agrees well with the multiple scattering code. Also,
the hybrid codes use less computer time and it is simpler to extract the
partial derivatives from the single scattering code. It is important to note
that one of the advantages of the filter inversion technique is the capability
of incorporating any solution into the equations of radiative transfer no
matter how complex. Thus in an actual experiment the more accurate
multiple scattering code would be used.
The definition of the state has been previously discussed (Newell and
Gray, 1972) for the purposes of inverting three constituent densities.
Basically that scheme had density points defined at arbitrary altitude
58
increments. The elements of the state were the logs of the densities (or
extinction in the case of aerosols) at each altitude. By using this definition
of the state at intervals of from 3-10 km along with a finer integration
step of 1 km in the radiative transfer codes it was found that the filter
would converge (Newell and Gray, 1972) for realistic densities that would
be encountered in the atmosphere.
3.1.4 Sensitivity Review
For this current work the state definition was expanded to include
the log of aerosol number density (P), size distribution parameter (a), and
both real and complex indices of refraction (n and n1) at each chosen altitude.
This expanded state was first analyzed through sensitivity studies and then
through covariance propagation to determine the feasibility of estimating
each parameter. The results of this study are discussed in Section 2.1. .,
The sensitivity vector is derived from the filter equations and indicates
the relative information derived about each constituent parameter at each
altitude for every measurement condition (i.e. wavelength and viewing angle).
Since after each measurement the covariance matrix records the information
gained from the measurement it is natural to look at the covariance update
equation for the definition of sensitivity. The term of interest here is
K(m+l) B(m+l) P(m), which gives the decrease in the covariance after the
(m+1) measurement. Normalizing by P(m) yields the sensitivity matrix
K(m+l) B(m+l) whose diagonal elements comprise the sensitivity vector.
For practical data presentation and speed in calculation, the sensitivity
vector can be defined by assuming that the covariance matrix maintains
its initial variances and has zero cross correlation. With these assumptions+Vi
the sensitivity of the j element of the state vector is
s. + R (3.1-4)
59
It is important at this time to realize both the usefulness and drawbacks
of sensitivity analysis; The main purpose of the analysis is to show just
which constituent parameters and altitude regions are important in a
measurement. Thus sensitivity analysis will reveal for example that at
4000 A and a tangent height of 20 km with a certain set of initial uncertainties
the greatest information will be derived about aerosol number density and
aerosol absorption in the region from 20 to 23 km. The analysis will also
reveal that at 3000 A and 20 km the only information derived is from ozone
between 40 and 50 km. (This is due to the strong ozone absorption which
prohibits this wavelength from penetrating below 30 km.)
The one disadvantage of the sensitivity analysis stems from the
assumption that each update has zero cross correlation. It often happens
that a set of measurements do not yield linearly independent sensitivity
vecfers. For example consider a simple experiment designed to determine
aerosol number density and neutral density at 50 km by two measurements
at 3800 and 4000 A. The sensitivity vectors would be high for both
constituents at both wavelengths seemingly indicating that both are readily
observable. However, the actual final covariance matrix of such an
experiment would show that little has been learned of either constituent
since there is no way to distinguish between constituents from the
measurements. Therefore, the covariance matrix would indicate that the
sum of both constituents is well known.
This is not a serious problem however since visual inspection of
sensitivity vectors is often sufficient to determine whether or not there
will be a problem with ambiguous measurements. If there seems to be
some doubt then a complete propagation of the covariance matrix can be
made to determine exactly what information is recoverable from the data.
3.1.5 Stellar Occultation
The stellar occultation inversion technique is essentially the same
as the inversion of scattered sunlight (Newell and Gray, 1972). There are,
however, some aspects of the stellar occultation radiative transfer model
60
that make improvements possible in the construct of the.-state elements.
These improvements yield finer data resolution and more accurate error
statistics. The next few paragraphs describe the essential differences
between the two inversions.
. The radiative transfer model used for the stellar occultation work
i s . . . • : • • ' . • • : • • ' • . • • • • • • . . • : • . '
I-= I0e"T (3.1-5)
where (I) is the measured intensity, (I ) is the star intensity, and (r) is
the optical depth along the line of sight. The model .used assumes a linear
line of sight and neglects dispersion and multiple scattering effects., These
assumptions will not cause errors in excess of 10% above 30 km .(Hayes
and Roble, 1968). This altitude is approximately the maximum depth to
which ozone information can be retrieved. , . , "
Since the Kalman-Bucy filter assumes linearity in the .measurement
equation it is desirable to obtain a radiative transfer model which is as'
nearly linear as possible. In the case of stellar occultation it is possible
to create an exact linear equation by measuring the log of the intensity
rather than the intensity thus giving the new measurement equation
ln(I) = ln(Io) - r ,. . .. (3.1-6)'
In the scattered light case discussed previously it is necessary, to
estimate densities at a resolution approximately two to three times as coarse
as the tangent height data. This was necessary along with a gain damping
term .to insure convergence of the filter. For the linearized stellar,
occultation case it is-possible to estimate density at each tangent height
and also to dispense with the gain damping factor. The final density estimates
are therefore average densities of a layer between two tangent heights.
61
3.2 RADIATIVE TRANSFER MODELING ;
3.2.1 Introduction
This section discusses the extension of computer algorithms for
simulating the horizon profile caused by radiative transfer within the earth's
atmosphere (previously reported in Newell and Gray, 1972). Here, we begin
by reviewing the computational problems to which these codes are addressed
and introduce briefly computer codes applicable to related problems. Their
applicability to present work objectives is discussed, with emphasis on
the unsolved problems that required the development of new codes. These
new codes, a simple, modified single scattering code and a more
sophisticated, full multiple scattering code, are compared in terms of
accuracy obtainable versus computational burden required. The various
applications to which they have been put are presented.
3.2.2 The Computational Problem
The problem of horizon profile simulation is a difficult one because
in the earth's atmosphere there are important polarization effects in the
presence of multiple scattering involving not only Rayleigh but also aerosol
scatterers, as well as absorption. Furthermore, there is the
nonhomogeneous structure of the atmosphere, and its slight curvature in
conformity with the earth. Many of these factors contribute substantial
difficulty. Polarization requires a matrix rather than a scalar treatment,
thus increasing the complexity of operations and the amount of storage
space required. Multiple scattering is a well-known primary source of
difficulty in all radiative transfer calculations. Rayleigh' scattering and
absorption are straightforward, but aerosol scattering is characterized by
an exceedingly complicated ill-behaved phase function, or angular pattern,
that is difficult to incorporate in a radiative transfer model. The
nonhomogeneous structure of the atmosphere imposes an altitude dependence
of the scattering into any given solid angle. Finally, the curvature of the
earth entails formidable geometry problems.
62
3.2.3 Codes Applicable to Related Problems
Basic computational techniques for radiative transfer modeling that
existed at the outset of our work are reviewed by Hunt (1971). We shall
refer to the two basic approaches he discusses as the matrix operator
method and the Monte Carlo method. The matrix operator method represents
the atmosphere (or any sub-layer of it) as a matrix operator that transforms
an input column vector of stream irradiances to an output column vector
of stream irradiances. This matrix formulation is not to be confused with
the matrix formulation required by the inclusion of polarization in the
problem; the matrix operator method has matrices even if polarization is
ignored. The Monte CarlOrmethod represents the atmosphere with a random
number generator that chooses scattering locations, angles etc. for individual
photons, very large numbers of which are followed to accumulate a horizon
profile.
The matrix operator method has been used mainly in homogeneous
plane parallel atmospheres, where application of the doubling technique
makes it impressively efficient. It has not, however, yet been developed
to handle inhomogeneous curved atmospheres. Thus, while it was available
in principle at the outset of our work, in practice, its use would have required
extensive development. By contrast, the Monte Carlo method has actually
been applied to essentially our problem in a program called FLASH written
by Collins and Wells, 1970. The FLASH program will indeed accommodate
every one of the sources of difficulty mentioned in the preceeding section.
Furthermore, the program existed at the outset of our activities. It would
have been an attractive solution to our problems except for one factor: it
requires a large amount of computing time (approximately 2000 seconds
of CDC 6600 cpu time per horizon scan to achieve an accuracy of
approximately four percent).
63
3.2.4 The Radiative Transfer Codes
Within the Draper Laboratory, there are now two radiative transfer
codes that simulate horizon profiles. One is a refinement and completion
of the code REV.HYBRID which is described briefly in the earlier report
by Newell and Gray, 1972 and in greater detail in an internal report by
Var, 1971. It is based essentially on single scattering plus an effective
planetary or cloud albedo as a secondary source. The second code (C.K.W.)
is a more recently developed full multiple scattering model with polarization.
It is based on mathematical techniques described in Newell and Gray, 1972
and in more detail in a recent journal article by Whitney, 1972. Some of
the computational procedures involved in implementing the model are
described in Appendix 5.3 of this document.
Considerable effort has been expended to provide a basis for judging
which of the two radiative transfer codes should be used for any particular
application. The remainder of this section reports on these activities.
There are three main areas of comparison: accuracy achievable, run time
requirements, and suitability for specific applications.
In attempting to establish the absolute accuracy of the C.K.W. code
we compared its results to FLASH results through the courtesy of the Air
Force Cambridge Research Laboratory. Differences between FLASH and
C.K.W. were found to be small and related primarily to fluctuations in the
Monte Carlo results or differences between the details of layer definitions
in the two codes.
The run times for both the C.K.W and the REV.HYBRID codes appear
to be significantly less than those required for Monte Carlo simulations,
in spite of the fact that specific comparisons are complicated by a number
of operational and statistical factors. Our comparisons have been based
rather arbitrarily on the times required to produce horizon profiles which
are in apparent agreement with each other. Typically, the C.K.W code
produces a horizon profile in approximately twenty seconds, and
64
REV.HYBRID in.approxirnately seven.seconds,:both on an IBM 360/75; while
the Monte Carlo, code of .FLASH uses approximately two thousand /seconds
of CDC 6600 time to produce comparable results; .The significaritcomparison
here, is between, the :C.K.W. code and the Monte Carlo code, since-only
they are really addressed to the full multiple scattering problem. A tentative
estimate of, the .time saving with C.K.W-..code is a factor of 'several hundred.
,-. ....:.. . In general, one would anticipate that multiple scattering would increase
.observed radiances'to .those that have-been scattered directly into the
.receiver field of:view. . The magnitude of that contribution as determined
by the C.K.W.; code ,is- illustrated in Figures .3:2-1 and 3.2-2. The
enhancement of the radiance occurs without substantially altering the profile
shape. Such an effect can be simulated also with the REV.HYBRID code
by introducing a fictitious planetary albedo. The ficticious albedo required
to match multiple scattering results depends upon sun angle and wavelength.
Figures 3.2-3 and 3.2-4 show the effects of an albedo range of zero to
unity upon REV.HYBRID horizon profile at 4000 and 5500 A. At 5500 A
the vertical optical depth of the atmosphere is smaller than at 4000 A,
thus more sunlight reaches the surface and is reflected back into the
atmosphere. The reflected radiation is scattered into the receiver field
of view adding substantially more radiation than when the albedo is zero
causing a significant variation in the enhancement for the two wavelengths.
There is no way that the enhancement required to make REV.HYBRID
conform to the C.K.W. code can be derived from basic principles. Thus
we feel it is wise to use the full multiple scattering code (C.K.W.) for
applications requiring absolute radiances, rather than simply horizon profile
shape. Where only shape is required, REV.HYBRID is sufficiently accurate.
We come now to the question of suitability for different applications.
There are two important applications to consider here: the generation of
horizon profiles for the inversion simulations used to produce the aerosol
invertibility results of Section 2.1, and the generation of .the 324 sample
horizon profiles in Appendix 5.1 illustrating the effects of varying wavelength,
65
latitude, season, cloud height, zenith and azimuth angles. For the inversion
simulations, the single scattering REV. HYBRID was judged suitable because
the simulation does not require absolute radiances but rather fractional
changes in radiances induced by small perturbations in each of the quantities
defining elements of the state vector. Furthermore, the calculation of these
differential changes would have required the calculation of many complete
horizon profiles, and hence a considerable amount of computer run time
with C.K.W. This is not the case with REV.HYBRID; differential changes
can be obtained without recalculating complete profiles. By contrast, for
the parametric horizon profile study, the computer code (C.K.W.) was used
because the effects of the parameters to be illustrated could not have been
accurately obtained from REV.HYBRID.
66
O)u>
l«- Ks gQ- <
z *'N u
2 o0 Z1 «
. .UJ Q-K PLJ _)li. i>u. £S
o oCM • o-
o°°CM
OvOCM
O^t-CM
O O O O O O O O O OCM O «o- «o •*• ,CM- O °° *o *•CM CM ^ ;rt - - •=<
30NVHO AiISN3iNl IVNOIlDVcdJ
OCM
O V
_J<
O. CM
I
CO
8eu . .coN•HH
S,
(U4J-P(0O
cxcooo
Q) O 'r-*t Otr> 1*1c•H IIW
0)T! rHC CH(0 C
Oj -P•H -H-P CrH (U
iCN,
•H
67
(A
OOoo
</>u
^- K0 yir tQ- <oZ «>
S yg l1 «>< <iaU O.(T PU =!
O OO 8
vO
OOvO
O1010
OOlo
O Olo O
OOn
OloCM
OOCM
O(f)
OO
.Orx ^
-* iu
_l<
30NVHO A1ISN31NI
CO
OS-l
cON
SCn -«c cw;
-rH
M o0) O4J in4J intf IIO <<
<u oH O
C(0
0)
0)
-p-HC<UIS)
CMI
cs•
n-
(cd31S V?ND /SUV/AD) IDNVIOVdo '
68
o0
01
1010
O10
10CO
Oro10CM
O<M
O
(x) 30NVHD A1ISN31NI
10I
V
U.QD
zv•+-S
U
§H
0)rH•H
o
coN
-HH
K '
QHtfCQJHW•
>
c-H
0):
IdO
Q)
OOo•
woT)0)
(04J0)C(8
ft
G M•H OCO M-l
roI
BONVIdVd
69
( > Jy i
N
"7u a §
rf§
(M
Oo Ooo
oCM
OO
Ooo
O•*-
OCM
30NVHO AiISN3iNI
oCMI
v
Ldo
to .<D CM
•HM-lOV)CM
oN
•H
OOmIT)
-pa
oK -acQ (0HPH Om>-i mw o•
> ww o« T3
0)
V
u
_J<
(0U
0)rH
w
ICM•
ro
&>•H
(0-P0)cmrHCM
70
3.3 AEROSOL MODELING
3.3.1 Introduction
.The application of recursive filter algorithms to limb profile inversion
of aerosols requires that at each recursion the aerosol optical properties
be recomputed given an updated physical characteristic estimate. Aerosol
optical and physical properties are related theoretically through the Mie
equations which express the angular scattering contributions and cross
section as a function of the physical parameters such as size and index of
refraction. At each recursion the filter requires recomputed values of
these quantities and also the partial derivatives of radiant intensity with
respect to those physical characteristics that are included in the state vector.
For example the partial derivative of intensity with respect to index of
refraction is of the general form 3I/3n = ( d l / d r ) (d r ldn) + (d l ldP (6>))
OP (0)/3n). Evaluation of drldn and 3P (0 ) /3n involves the use of detailed
aerosol models. Since computational times on the order of a minute are
involved in determining these quantities using the Mie series, the total
computer time required for inverting a single scan would be on the order
of hours per scan just for aerosol calculations. Such an amount of time
would be out of proportion to other calculatipnal requirements posed by
the filter algorithm, and inconsistent with the goal of near-real-time
inversion capability.
The implications of this calculational complexity when matched with
the desire for computational economy are that multicalculations with the
Mie series are to be avoided, and some alternate procedure must be provided.
3.3.2 Choice of Models
Several alternate procedures for the Mie computations are suggested.
For the simple one parameter size distribution with which we are concerned,
there are four parameters to consider within each size range. These are
wavelength (A), size distribution parameter (a) (used in the distribution
71
law n(r) = Ar ), and the two part index of refraction m = (n) - i (n1).
Thus each scattering phase function and associated cross section is a
function in four dimensions P(0; a,\, n, n1) and a(a, \, n, n1). The empirical
fit problem is thus to find a set of four dimensional functions to fit computed
data to within some accuracy criterion. In addition the (0) dependence of
(P) must be described for any chosen set of parameters.
One approach would be to establish a four dimensional array of
computed points and to interpolate between them. It was felt however that
the search times involved in this procedure would be longer than desirable,
and that the four dimensional interpolation procedure would be excessively
complex.
Another approach, and the one taken here, is to fit, in a least squares
sense, the computed data with continuous functions of the four variables.
This has been done by applying a multivariate regression analysis to the
.computed Mie data for the extinction, scattering, and absorption cross
sections (a,-,-., a and a), and to the computed Mie phase function data at
six selected angles, viz., 0°, 20°, 60°, 110°, 164°, 180°. The Mie data for
each function (P,a) was obtained for 81 points corresponding to all possible
combinations of the three values assigned to each parameter, viz., a = 2,
4, 7; X = 4000, 6000, 8000A; n = 1.3, 1.4, 1.5; and n1 = 0, 0.02, 0.04. The
continuous functions thus obtained for the phase functions and cross sections
are listed in Appendix 5.2.
There are several options for the choice of a "goodness of fit"
criterion. In general one can force fit with a least path moment weight
called the best Lp approximation (Kahng, 1972). When p = *) the result is
the Chebyshev "maximum residual" fit. When p = 2 we have the usual
least squares fit. For this problem some (p) in between would perhaps be
desirable; however since a multivariate computer code was readily available
only for p = 2, this fit criterion was used and the resulting functions are
least squares fits in four dimensions.
72
Phase Function Interpolation
To obtain P(0; a, A, n, n1,) for (0) values other than 0^= 0°, 20°, 60°,
110°, 164°, 180°, an interpolation formula is required. A least squares
polynomial fit could be made to the selected angle points, however,
comparable accuracy was found to be obtainable by fitting log-linear straight
lines between the function values P'j at the selected angles 0.. Thus between
the selected 0., values of'P(0) are given by
In P ( 6 . , , ) - ' In P ( 6 . )P ( 6 . ) e x p 1+1 x
(3.3-1)
Normalization
The interpolation formula has the advantage that it can be easily
normalized to insure proper conservation of scattering by shifting each
modeled (P) up or down by an amount determined by a closed form integral.
The constant of normalization is just
5 {exp(m.. 6 ...., ) Y •. -exp (m• .6 • ) . 3 • }c.= 1/2TT £ P ( e . ) e x p ( - m . e . ) ——i , ^^—i (3.3-2)
i=l 1 + m.
where '
. - Yi = misinei+1-cos0i+1. (3.3-3)
•Bi- = misin6i - C0s8i . (3.3-4)
and
ii^ = {In P(8 i+1) - 1" p(ei)
}/(9 i+1-e i) (3.3-5)
73
3.3.3 Error Sources
It is possible to have considerable error in the phase function models
for a particular set of parameters and still gain useful information from a
model. The important thing is that the model represent an average over
an ensemble of computed points that are representative of the phase function
over the range of the parameters. For example if a modeled P(0) has an
RMS deviation of 25% at some angle, but variations of a parameter (say
n') over its range cause P(#) to change by say 200% then reliable partial
derivatives and the concomitant inversion sensitivities can be obtained.
The error sources and related limitations are described in the following
subsections.
Several strong trends in the data that should outweigh the RMS
residuals are evident. These are noted below.
1.0-lOA/Size Range
1. There is a strong (n1) dependence at both 164° and 180° as well as
other angles in this region. The phase function is generally an order
of magnitude lower for n1 = 0.02 than for the n' = 0 case. The n1 =
0.04 case drops P(0) another 50%.
2. The P(0°) values are twice as high for the n1 = 0.02, 0.04 cases as
for the n1 = 0 case.
3. (n1) effects are also strong at 20° and 60° (200% to 500% variation),
as well as at other intermediate angles.
4. (X) effects are strong (300%) at 0° and 180°. As expected, the (X)
dependence changes from P(0°) <* 1/X to P(180°) or X.
5. (a) effects are as much as 400% at P(0°) and diminish with increasing
(0)/ Beyond 110° complicated forms result.
6. (n) effects are most noticeable and consistent at 164° (up to 300%
variation). As (n) increases, P(0) increases.
74
0.1 -1. Q/J Size Range
1. The(n') dependence at 164° and 180° is strong; (200% to 500%) unless
a = 7, elsewhere it is very weak (10%).
2. (n) effects are up to 200% at 164° and 180°, but diminish with increasing
' Q and X.
3. (X) dependence is strongly linked to (a) over all (0).
4. (n) effects at 164° are strong (200% to 300%) unless a= 7, X = 8000A.
5. (a) effects are strong at 0 (up to an order or magnitude), but correlated
with (X), P(0°) oc I / a -A . - .
0.01-0.1/v Size Range
1. (n1) dependence is negligibly small over the range of all the other
parameters.
2. (n) dependence is at most a 10% effect at the larger angles for a = 2.
3. (a) and (X) both strongly influence the phase function, but the effects
are on a smaller (200%) scale than for the larger size ranges.
Mie Code Errors
To normalize the phase function generated by the weighted size
distribution average, the Mie code performs a trapezoidal integration
between the selected defining points, and adjusts the amplitudes by the proper
constant to achieve conservative scattering. However, since much of the
accumulated value of the normalizing integral 2w / p(0) sin e do is obtained
at small angles, the adjustment of the entire functional amplitude is very
sensitive to the fineness of the angular defining intervals. This is
particularly important for the larger particles which have rapidly decreasing
phase function values near 0 = 0 . In the (1.0-10/u) size range, a decrease
in the small angle (0) step size from 10 to 0.2 results in a factor of
twenty adjustment of the phase function upward. The upwards adjustment
is caused by the excessive integral accumulation under the trapezoidal
function between 0° and the first defined point at 10° in the coarse sample.
75
Figure 3.3-1 shows an extreme case of this with angle step increments
of 0.2 and 10 . Both curves in Figure 3. 3-1 are normalized by integration to
unity. The dashed curve is the actual curve under which the trapezoidal
integration is performed rather than the computer-drawn straight line
connecting the 0 and 10 points. In general the error due to the overall
phase function adjustment for normalization is proportional to the difference
between the scattering cross section as predicted, by the evaluation of the
0 = 0° Mie series and the value obtained by integration of the phase function.
For the (1.0-10A/) size range the maximum value of this error is about
25% forthe Ae intervals that were used in the calculations. In the (O.l-l.O//)
range the error is about 5% or less, and forthe (0.01-0.1/u) range the error
is less than 0.3%.
Size Distribution Sampling Errors
The power law distribution n(r) = Ar is sampled at discrete (r)
values within the chosen size ranges. Each (r) choice corresponds to a
particular Mie scattering result for a particle of radius (r), and the resulting
phase functions are averaged by combining the individual (r) phase values
with the (-a) weight. Since this process has the effect of averaging out the
oscillations related totheBessel and Legendre series within the Mie series,
the results depend somewhat on the fineness of the Ar integration steps.
Table 3.3-1 shows an example for the size range (1.0-10A/) with complex
index of refraction. Differences between the phase function values for the
range of 10 to 200 radius sample points are generally of the order of a
few percent except for the 180° values which differ by 20%. In Table 3.3-II
similar values are shown for non-absorptive aerosols (n1 = 0). Thirty
percent differences in the phase function values occur here fora few angles,
and indicate that the number of radius steps required for a few percent
accuracy is nearer 100 than 10. However the required computer time is
directly proportional to the number of integration steps.
76
CD
CD
TABLE 3.3-1
PHASE FUNCTION VALUES
NOMINAL CASE, a=2, A=.4, 11=1.4, n' = .02, l-10y
PHASE ANGLE
O.TLU
s 10^ 50
g 100LU
£ 200u_CD
0
__P°372368368368
20°,0367,0373
,0371,0372
60°,00406
,00411,00418
,00418
110°,00120,00122,00119
,00120
164°,00126,00118,00124
,00123
180°,00122,00164
,00152,00154
TABLE 3.3-II
PHASE FUNCTION VALUES
I>ex:LU|
£ 105 50
£100i0/3 200
0°256249
253253
a=2,
20°,205,236
,203,204
A=.4, n=1.4, n
PHASE ANGLE
60°
,0151,0190
,0196,0196
'=0, l-10y
110°
,00264,00285
,00245
,00276
164°, 0114,0134
,0141,0138
1.80°,0578
,0676
,0728
,0731
77
The derived aerosol model is based upon a 10 step integration, thus
30% errors may be expected in the (1..0-10A/) size range from this error
source. Figure 3.3-2 shows a plot of the differences between the phase
function values resulting from 10 step and 200 step size distribution
sampling. The differences are expressed, as a percentage of the phase
function value for the 200 sample case. It shows that the effects of changing
the sample step size are essentially random with regard to angular position.
Errors Related to Piecewise Function Fitting
The errors just, discussed were related . to the computational
uncertainty of the phase function at a particular angle. The modeling
procedure we have chosen connects the points f(P(0.) , 6.) with exponential
curves of the form P(f>) = Aexp(B0); thus even if the parameter model P(0-;
n, n ' . o r . X ) were perfectly accurate there would be errors caused by the
fluctuations of P(0) between the select model points P(0.). This would be
true, whether the points were connected by exponential curves or by, e.g. a
least squares polynomial fit. This error source would decrease in magnitude
if the P(0.) sample points were.more closely spaced, however the six chosen
B- represent a compromise between accuracy and the need to limit model
complexity. Each new 9. requires a new Lp fit to find the modeled P(6.',
n, n1, a, X), however the data shows that the Q. choices should be optimized
for each size range separately. For example, Figure 3.3-3 shows the
log-linear fit to the (0. 01-0. In) case havingthe most curvature. The fit is in
error by a maximum of 14%. The RMS error for this case is 7%. For
some of the better fits in this size range the RMS error is about 4%. The
case illustrated in Figure 3.3-3 could be improved by the addition of points
defined around 30°, 90°, and 110°. However the choice of defining angles
for a given size range is not a simple matter to be solved by examination
of a single curve. The noticeable structure in the curves in the backscatter
region moves over a range of (0) when the various parameters are changed.
Figure 3.3-4 shows the six point fit to one of the worst-fit cases in
the (0.1-1.0/y) range. In this size range the points of fit correspond quite
78
closely with the major points where curvature changes, however it can be
seen from the figure that additional defining points would be helpful here
at 130° and 176°. The maximum error for the (0.1-1.0/y) case is 56%,
with an RMS error of 21.5%, l<r. '
In the (1.0-10/y) size range the standard points of definition do not fit
the Mie curves well, and three or four additional points should be used for
a better fit. A particularly bad-fitting case is shown in Figure 3.3-5.
The minimum for the phase function is at 130° and is missed by the
approximate fit point at 110° causing large errors at 130°. The 6-point
function also ignores the secondary peak at 150°, and fits badly at 10°.
Over the range of parameters considered, the position of the phase function
minimum for the (1.0-10A/) size range varies over the interval 110°-130°.
Thus, improved modeling of this size range would require inclusion of extra
defining points within this interval. The maximum error for this case is
630% at 130°. The RMS error is 400%.
The RMS errors listed in this section are based upon the sequence
of discrete angles used in the Mie computations. Since these angles are
defined with smaller increments at small and large angles, there is a built-in
weighting which may adjust the RMS error value somewhat either upward
or downward (depending on the particular fit) from the RMS error which
would result from a uniform disbursement of sample points.
In summary, the errors due to the piecewise log-linear fit at six
preselected angles are moderate but within useable limitsforthe (0.01-0.1/u)
and (0.1-1.0/u) size ranges. The errors are too large for the large particle
range (l.O-10/i/) which implies that an optimization procedure should be
constructed for choosing the angular points of definition.
Parametric Fluctuations
It is of interest to examine the effects of small changes in the defining
parameters (a ,A,n , n1) to see whether or not the phase function is undergoing
rapid fluctuations near the selected nominal parameter values. If this were
the case there would be no guarantee that the value of P(9) at the nominal
79
parameter value was actually representative of P(0) in the region around
the selected value.
This problem was investigated by means of a set of Mie calculations
with small perturbations from the standard points. Deviations from the
case a = 2, X = 4000A, n = 1.4, n1 = 0.02 were considered by running cases
with a= 2.1, X.= 4100A, n = 1.41 and n' = 0.021. The (1.0-10/u) size range
was considered because it exhibits the most erratic behavior and should
represent a worst case. The results are tabulated in Table 3.3-III.
TABLE 3.3-III
LINEARITY PARAMETER 5 FOR A l-10y CASE
(a) (n) (n') (X)
5=2.3
9.0
0.36
0.91
1.2
0.39
2.4
1.6
5.0
0.18
1.6
0.63
0.12
0.2
2.7
0.27
0.12
0.027
4.2
7.7
18.0
1.0
2.1
6.7
0°
20°
60°
110°
164°
180°
AP Ax- The number in the table is J- = =- -^— where AP and Ax aretherangeso tr o x
of the phase function and the parameter expressed as a percent of thenominal
value. The quantities 6P and 6x are the perturbed phase function and
parameter values also expressed as a percent or fraction of the nominal
values. If all the functional behavior were linear these values would be
unity. Typical changes in the phase function indicate the £= 4 corresponds
80
to quadratic dependence. The results show that at X = 4000A one has a
rapidly varying phase function at 60°, however the excursion from the value
at 4000A to the value at 4100A is a variation of 21% which is comparable
to the scale of other errors for this size range. Otherwise the fluctuations
are relatively calm. The small values for (n1) are related to the fact that
P changes rapidly in the forward and backward directions for very small
deviations of (n1) from zero. Thus the range of P is large between values
corresponding to n1 •= 0 and n1 = 0.02, hence the small values of (£) for
(n1). Since the modeling functions used were primarily quadratic, one expects
to find relatively smooth behavior of the phase functions with respect to
parametric variations over the parameter range.
Summary of Phase Function Modeling Errors
The models listed in Appendix 5.2 exhibit a complex error structure
in general. However one can estimate the maximum RMS errors by assuming
that the various sources of error are independent and thus that the errors
can be added. The error sources are, in approximately decreasing order
of importance, regression residuals, piecewise fitting errors, . size
distribution sampling errors, Mie Code normalizing errors, and higher
order parametric fluxuations.
The regression residuals are listed in Table 3.3-IV with both maximum
and RMS values given.
At the defined angles the total error consists of the regression error,
the size distribution sampling error, and the normalizing error. If we
take the typical size distribution sampling errors of Table 3.3-1 as la values,
and the normalizing errors discussed above and combine them by means
of a root mean sum of squares, the errors in Table 3.3-V are obtained at
the defined phase function angles.
81
TABLE 3.3.-IV
RMS AND MAXIMUM RESIDUALS FOR PHASE FUNCTION PARAMETER MODEL
Phase 0.01-O.ly 0.1-i.Oy 1.0-10y
Angle a(%) Max.(%) a(%) Max.(%) o(%) Max.(%)
0° 0.3 0.9 38 3 10
20° 0.2 0.5 1.8 5 11 45
60° 0.1 o.3 4 9.5 8 23
110° 0.4 2.7 4.5 17 18.9 97
164° 0.25 0.6 6 15 31 120
180° 0.2 0.5 10 25 32 127
TABLE 3.3.-V
RMS ERRORS AT DEFINED ANGLES
INCLUDING NORMALIZATION AND SAMPLING INTERVAL ERRORS
Phase 0.01-O.ly o.l—l.O/i 1.0-10 yAngle
0°
20°
60°
110°
164°
180°
2%
2%
1%
3%
4%
4%
6%
5%
9%
15%
9%
16%
25%
27%
35%
32%
43%
46%
82
In .between these "defined" values the piecewise fitting errors must
be added in proportion to the distance from the defined angle. These errors
have been discussed in one of the previous paragraphs, and since no attempt
has been made to assess them, statistically they will not be combined with
the above table to produce a grand model error table. For the (0.01-0.1/u)
size range the "in-between" errors are of the order of 10% at maximum.
For the (0.1-l.Oju) range the maximum fitting error is roughly 50%, and
for the (1.0- 10A/) range it is approximately 100%. A graphical display of
the error sources is shown in Figure 3.3-6.
3.3.4 Models of the Extinction, Scattering and Absorption Cross Sections
The cross sections were modeled by applying the regression
procedure described above for the phase functions. Since the cross sections
for absorption and scattering sum to the extinction cross section, only two
of the three cross sections need be modeled in each case. Thus the two
with the smallest variances were chosen from each set of three models
and used to produce the third by algebraic combination. The errors for
these quantities are listed in Table 3. 3- VI.
TABLE 3. 3 -VI
CROSS-SECTION ERRORS
0.01-O.ly 0.1-T.Oyi 1.0-lOy
cr(%) Max(%) a(%) Max(%) a(%) Max(%)Error Error Error
a (absorption) 3 1 0 2 5 2 9cl
a (scattering) 0 . 3 1 - -s
a (extinction) - - 10 34 18 52e
The standard errors for the unlisted cross sections can be determined
if the correlations are known, however, since these errors were not used
explicitly in the inversion procedure the correlations were not computed.
83
H 1
A = 1.1110 +00•a = -2.0X = .100 Mn = 1.400 n1 = -0.02EXT = 6.6575 -O7
Fig. 3.3-1 Dependence of Normalization Adjustment
Upon A6 Sample Size.84
w.H(0
Q)-PC
-H
QJ
g(0
OJN
•HU)
O41
OJ
Q)
(NI
ro
Cn•H
Q)tnC(0
OJN
-HCO
C -HO O
-H W-P O
-H (0
-P '-tn P.
•H O
1
-PC0)U
tna<u-PU5
ooCM
O 'OH CLt (0Q)
c•HW
°°zdaNV°ld
102t
A. = i.mo -02o = -2.0X = .400n = 1.500 n1
EXT = 1.033* -11
PIECEWISE MODEL
1 II 1 • I I I 1 1 1 h 1 1 1 1
30 60 90 120THETA
150 180
Fig. 3.3-3 Worst Case Fitting Errors,
0.01-0.ly Range
86
10*;
zg
uzZ)L_
LJin
10-2
H 1 1 1 1 (- H 1 1 1 1 1 1
A = 1.1110 -Ola = -2.0X = . «K)0-jjn = 1.400 n1 = 0EXT = 8.2688 -O9
1Q"34 1 H 1 1 1 1 1 1 1 1-
O 30 6O 90THETA
H 1 1 f-12O 150 18O
Fig. 3.3-4 Worst Case Fitting Errors,
0.1-l.Oy Size Range87
1 H—1 1 1——t
A = 1.1110 +OOct = -2.0X = .400 »n = 1.400 n' = 0DOT = 6.5609 -07
10°
OI— •
I-oz
uin<Q.
10"
10"
t PIECEWISE MODEL
In.
!,,+ I1
H H 1 1- t I 1 h-
30 60 90 120THETA
150 ISO
Fig. 3.3-5 Worst Case Fitting Errors,
1.0-10y Size Range88
Ul0)o
OwS-lo
0)
-pu3
M-l
Q)W
eu
iCO
ro
-H
REFERENCES
SECTION 3
1) .Anderson, G.P., "The Vertical Distribution of Ozone between 35 and
55 Km as Determined from Satellite Ultraviolet Measurements",
Thesis for Master of Science, Department of Astro Geophysics,
University of Colorado, 1969.
2) Collins, D., and M. Wells, "Flash, A Monte Carlo Procedure for Use
in Calculating Light Scattering in a Spherical Shell Atmosphere",
AFCRL Report No. 70-0206, 1970. ;
3) Hayes, P.B., and R.D. Roble, "Stellar Spectra and Atmospheric
Composition", J. Atm. Sci., 25, 1141-1153, November, 1968.
4) Hunt, G.E., "A Review of Computational Techniques for Analyzing
the Transfer of Radiation Through a Model Cloudy Atmosphere", J.
Quant. Spectrosc. Radiat. Transfer, 11, 655-690, 1971.
5) Kahng, S.W., "Best Lp Approximation", Mathematics of Computation,
26, 505-508, 1972.
6) Newell, R.E., and C.R. Gray, "Meteorological and Ecological
Monitoring of the Stratosphere and Mesosphere", NASA Contractor
Report, NASA-CR-2094, August, 1972.
7) Var, R.E., "A Hybrid Algorithm for Computing Scattered Sunlight
Horizon Profiles", MIT Aeronomy Program Internal Report No. AER
7-3, June, 1971.
8) Whitney, C.K., "implications of a Quadratic Stream Definition in
Radiative Transfer Theory", J. Atm. Sci., 29, 1520-1530, November,
1972.
90
4.0 SUMMARY
4.1 CONCLUSIONS — AEROSOL PHYSICAL PROPERTIES
The vertical distribution of aerosols from approximately 10 km
upwards to 100 km can be determined globally by a satellite horizon inversion
technique. Additionally, information such as the number density, the size
distribution, and the complex parts of the index of refraction can be obtained.
Three particle size ranges (0.01-0.1/v, 0.1-1.0/w, 1.0-10/c/) were
considered in the simulations, and the results show that the quality of the
physical characteristic information is size range dependent, i.e., that the
relative observability of the physical characteristics changes from one size
range to the next. For the smallest size range (0.01-0.1/u) the number
density (f t) and the size distribution parameter (Q-) are observable, but the
indices of refraction are not. However this result rests partly upon the
assumption that the initial uncertainties of the four modeled physical
parameters are as follows: a = ±300%, o• • = ±1, a , . = ±0.005,
CT = ±0.05. Reductions in a and cr have the effect of increasing the
observability of (n1). A simple covariance propagation case in this size
range, using four kilometer horizon sample intervals and five wavelength
channels, reduces the initial variances as follows: a (±300%)->0p = ±150%,
= ±0.1. cr - + c r , a , -»• a , .•' n no' n1 n'o
For the medium sized particles (0.1-l.OA/), using the same initial
covariances, the simulations show that the complex part of the index of
refraction becomes generally observable. The peak sensitivity of (n1) jumps
from a value of «0.01 to«0.4 which is in the highly observable range. A
simple covariance propagationcase, using four kilometer horizon sample
intervals and three wavelengths showed initial covariance reductions as
follows: or (±300%)-*o>>= ±60%, crao(±l)-«ra = ±0.2,
an,o(±0.005)-*an, = ±0.001, anQ(±0-05^an = ±0.05.
91
All four physical parameters are found to be observable for the large
particle (1.0-10yu) size range. However the observability of the index of
refraction is strongly dependent upon the scattering angle, i.e. the sun
azimuth and zenith angles. For a scattering angle in the forward region
(«50 ), (n) is found to be unobservable. In the side-scattering region of
the phase function(n') is unobservable and the observabilities of (n) and (a)
are considerably reduced. However in the backscattering directions all
four aerosol parameters are observable.
An estimate of the state errors resulting from errors in the models
was made based upon the filter gain and computed horizon intensity
deviations. The calculations show that the aerosol modeling errors induce
physical parameter estimation errors that are generally smaller than the
final variances resulting from a covariance matrix propagation. However
the error analysis also shows that a deviation of the albedo of 0.5 from
the assumed value can cause large uncertainties in the state vector elements.
Thus the albedo should be included as an element of the state vector.
4.2 RESEARCH EXTENSION
The results obtained in study show how well one can determine the
parameters of a specific aerosol model. For example the inchoate model
developed for this research uses a single parameter, power law size
distribution. If the real aerosol size distribution is of some other functional
form, then the results of applying the chosen model in an inversion would
be to obtain a best power law fit to the real function. If the best power
law fit to the real function can be closely related to the real distribution,
then the results are useful for aerosol research. To this end, then, it is
important to attempt to generate the most realistic aerosol models possible.
A first step in this direction is to insure that the ranges of the chosen
variables cover the real physical range. This has been done in the model
used here in every respect except the aerosol size range. Hence one of
the next steps in a continuation of aerosol model development would be to
reconstruct the model with variable particle radius limits. This would
92
increase the size of the state vector by two elements, however, the present
model could be used to optimize the selection of wavelength channels. By
fixing wavelength, the dimension of the model could be reduced by one
element for a net increase of one dimension in the model.
To insure that fitting errors are minimized in extended aerosol models,
a procedure should be constructed which determines a minimum set of
phase function defining angles consistent with accuracy goals. The models
should also make maximum use of the analytical relationships derived from
ray optics (Van de Hulst, 1957) and information theory (Whitney, 1972).
The design of the filter algorithm should be further extended to include
the additional states resulting from improvements in the aerosol model
and from the inclusion of polarization measurements. The effects of model
type error sources such as approximation errors in the radiative transfer
simulation, fitting errors in the aerosol models, and relative fluctuations
in the solar power output should be explicitly included in the filter
formulation. As the aerosol models become more complex, additional
sources of information will be required besides the measurements at
different wavelengths. Two possible sources are measurements at different
solar angles and measurements of polarization. The present radiative
transfer simulation for a curved atmosphere includes the necessary
polarization calculations, however, these simulations would have to be
checked against other results to insure their accuracy. Accurate inclusion
of cloud scattering effects at low sun angles (<10°) would require
modifications of the existing code as well as simulations of the transfer
problem within a cloud.
93
REFERENCES
SECTION 4
1) Van de Hulst, H. C., Light Scattering by Small Particles, John Wiley
& Sons, Inc;, New York, 1957.
2) Whitney, C. K., "Stream Transformation and Phase-Function Integrals
in Radiative Transfer Theory", MIT Aeronomy Program Internal
Report No. AER 22-3, January, 1972.
94
5.0 APPENDICES
5.1 THEORETICAL HORIZON PROFILES
This appendix contains the theoretical multiple scattering horizon
profiles discussed in Section 2.3. The profiles illustrating the effects of
changes in various measurement parameters are arranged as follows.
Figure Numbers Variable Measurement Parameter(s)
5.1-1 to 5.1-12 Season; for 12 combinations of latitude
and wavelength.
5.1-13 to 5.1-28 Wavelength; for 16 combinations of cloud
cover, ground cover and zenith angle.
5.1-29 to 5.1-64 Zenith angle; for 36 combinations of azimuth
angle, wavelength, arid ground albedo.
5.1-65 to 5.1-72 Ground albedo; for 8 combinations of
zenith angle and wavelength.2The units of solar irradiance (SUN PWR) are in /uwatts/cm A.
Figures 5.1-73 to 5.1-84 show the vertical distributions of neutral
and ozone densities for 12 combinations of season and latitude. These
figures illustrate the mean and one sigma standard deviation values about
the mean.
Neutral density data was obtained from Valley, 1965 and from the
data compiled by Groves, 1970. The designations winter, spring, summer,
and fall refer to the months January, April, July, and October. The ozone
density data was generated from the modified Fermi distribution function
developed by Wu, 1970 and the peculiar appearance of the ozone density
distribution for the minus one sigma curve is due to the fact that the one
sigma values are equal to or greater than 100 per cent of the mean value
below approximately 10 km and again above approximately 50 km. Aerosol
extinction data was compiled from a variety of sources by Malchow, 1971
and the composite model illustrated in Figure 5.1-85 was constructed.
95
Latitudinal variations in the aerosol extinction data were accounted for on
the basis of data gathered by Salah, 1971 as illustrated in Figure 5.1-86.
Due to a lack of sufficient seasonal distribution, this aerosol extinction
data was used to represent the four seasons.
96
(x) 30NVH3 A1ISN31NI
V
uQ
UlQD
oII
exooon
inQ)
^H-HM-lOS-lPi
CON
-H
cotn(0Q)en
0)iHXJ(0
•HS-l(8
IiH
•
ID
V..WO0 2
97
05uilafy . Zt 8*80; wt! wQC•- Jo 5
I1Vo
30NVHD A1ISN31NI
,si§
O
O
CM
O o»•
0 2
•riO
LJC5DH
LJOZ)
03ooo
WQ)•H-H
O
04
CoN
-rH
cow(0a)wCD
to•H
rr>I
99
£V
uQD
(x) 30NVH3 A1ISN31NI
ooII
03ooo
to0)H•HmO
U)U
tra.
NU3 UJ
Vo
| o. .
si:"O - S j "
.O
:ITW
CM
C
»-t
) 0
P (-" CO Z Z S
0 V <JI « T vo
O O O O O O .
• lo
o
uQ
oN
•H
COCO(00)
CO
0)
(0•H
100
30NVHO A1ISN31NI
idQD
zv«*-*
u
Oo
ooo
CO
(U•H•Hmo
coN•HMOfficoCQ(00)
(U•HXI(0
•H
mI
V..NDa <•
101
~ ft 33 y S
Z ' i f c
oCO
OvD
OCM
Oo
o03
O OCM
OCM
IO•*-I
.OCO
.OfN.
o
o
O•*-
.OCM
UJQ
oo
30NVHO A1ISN31NI
ooo
tnQ)
M-lOM
uQ
coN
•rH
co(AfC0)en0)
•HM(0
IiH
•in
(cd31S V..WDo <J
102
LJQ
oo
ooinin
caQ)
(x) 30NVHO A1ISN31NI
O1O
co
tou
Coa:CL
W
8
a:og
8
inCM
cea.
Ooo
Orv.
OvO
O
Oro
uQ
coCOm0)
Q)
(0-HH(0
r-i
m
z
_| 01 < I
oCM
O
CM
ooo
CMIo
V - N D0 2
103
ocu
u
oCEa.
o <§*0 UJ1 5i
UCU
oCM
u.H 1-
o«-lI
in Oeg
in<\j
iO<*>I
(x) 30NVHO A1ISN31NI
loTrx
Din
O•*-
Oc\J
inCOi
UlO
£V*k^
LJ
Oo
0<
oominII
CQ0)rH•HIWOMO4
coN
Couim0)
0)rH
•3-•HM(t)
00I
in
104
o o o o o o o o ov I 3 l O ' 4 - < " > C \ l ' ^ O c r > C O
o. o o o o(X J> in *• «">
O O O O O OCM "" «j< CM ro
O
I
Ev
LJQ
(x) 30NVHO A1ISN31NI
Oor~II•
EH
ooinm
I Ir<
in
ot-i
U)uiZocc
O °>N .,C J
3 $
vo
Sii
CLO
o'Syg-TgII
^u3*8t .LJ1-1-0)< < -I_!«> <
Ct
0.
158g:ao 'i "°0 Z„ - < •
, - "CK
|ig<
g t » -N U " M< N
U) «) J Q
• o»-l
o
.ooo
io•so
OTIO
On
.Q*C\J
kL
V
U
ON
-Hi-lOffi
cow(0Q)
(8•H
CTlI
in
oO
CM
O
inio>i rv
O
105
(x) 33NVHO A1ISN31NI
LJQ
Q
Kh-
ooII
03ooo
ID0)
oM
CON
-H
COwrt)(U
(D
(0-HM
Or-tI
iH•
in
30NVIOVH
106
(*) 33NVHO A11SN31NI.;
V
LJQD
UJQ
^-
oo
cx:ooo
CQ0)rH-Hmo
GON
-H
oCOtOQ)W
XI-(0
•HM(8
IrH
•m
107
UQ
0<
ooor-
tn0)r-i-H
(x) 33NVHO A1ISN31NI
t . 8 a » JoH H DO 2 Z S 1^
_1 01 < U> <
o5 0
.r-4
1
0
CM1O
CO1
o*-i0
w>iO
vOiO
' ' (v'"i0
uQ
oN
•H
coCO(0<uw-0)
-H
10
CM
I
in
108
uQ
en0)iH•HM-lo
CON
0)XI
(*) 30NVH3 A1ISN31NIoEH
v
uQD
_J<
C OQ) iHiH UQ)
Hi O12 o
ro(U II
X! -P(0 -H
•H CM Q)rd N
rorHI
•
in
tn-Ht,
3DNViaVH
109
oCOto
O O O OO O O O
o o o o o oOvon
O O O*- CM oo n oo
o o o o o o o o o o o o o oo o o o o o o o o o o o o oCM
C\J C\J C\J CM C\Jo °°rt I *-*
CO 10 CM
(*) 30NVHO A1ISN31NI
LdQ
HP
<
CO Q)0) Ar-l H-H <IWO •>H g
C
•8(0•H
MoS3 O
C0)
U
O
CM
O
O
O
roI
LJQ
(0 O& o
n0) II
XI• (0
(0>
I•H•
in
Cn-H
•HcQ)
VZND0 «
110
3ONVHO A1ISN31NI
v
UJQ
_l<
uo
_l<
(UrH-HM-l
S
coN•H
0-1CQ)M0)
IIo
T3Q)
CXO
U
(0 O£; o
ro0) II
i-1 (UfO N
•" *•—*"•
inrHI
^H•
IT)
(«3iS 30NVIQVH
111
S gd O
2 %Q- <j
O IM OCE ZO VJ
-I 3< >
i u
Id mo: <u su. <u. >
ooooto
OOO
OOOo
ooolo
OOOOCO
OOOloc\j
(*) 30NVH3
ooooA1ISN31NI
ooo10
OOoo
oooto
o•*•
o
ooo10
V
UQD
p_J
UQ
<IJ
00• •
II
0)XIr-l
O "
G r-ON -P
•H (0MO CUK o
EHx;-P T(tr> 3C O
0}U
to o3= onQ) IIH £,0 -P(0-r)In(0
I
in•
•H
•HC0)
112
30NVHO A1ISN31NI
V
LJQ
_J•<
IIO•OQ)
tn Xi0) H
r-H <-HM-l •»O E
C HON -P
•H rdMO &ffi 0
ttX4J T3tn 3C OQ)
0)U
rti O
0)n
ZV• X
.Ld
§
_J<
XI -P.rd -H
-M C^ (1)(0 to
ItH
•' •
in
v. wo /siivnrf)o <•
113
(x) 30NVHO A1ISN31NI
Zv
uQD
v
UlQ
00
Q)to X!0) iH
rH <•H
C i-HON 4J
O O.K O
^X!-P T3
H O<U> *(fl OIS o
co0) IIrH £
(0
00
in
Cn-H
C(1)
114
(x) 39NVH3 A1ISN31NI
UoDI-
IIO•c
W Q)0) XI
M-lO
CO
O &<ffi O
EH
C0)
rH0)
O
HV• ^
uDt-
(d oS o
CO0) II
($ "H
S-l Q)
> ~
(Ti
IT)
fa
(H31S
115
N
30NVHO A1ISN31NI
o
UloID
zv
idQZ)
00• •
IIO•d
in a)0) ,QrH rH
o1 iM
c •*ON 4J
•H (0
C OQ) iHH UQ)>a o
oCOII£-P-HC(1)
0)
(TJ•H
OCM
CM
O
O
O
c\jiO
116
O O
8 8«- COco n
O OO OO OOco
OOO
oo \oCM CM
o o o oo o o oo o o o•<»- CM O °°CM CM (M •"•
0 0 0O O Oo o ovO CM
oooo
o oo oo oOO
o oo oo o*- CM
o oooCM
30NVH3 A1ISN31NI
V
LJo
_J<
uQD
IIo•o
M 0)
-H
oeucoN
O Oiffi O
c o<D iH•H U
n) o& o
00Q) II•H J3XI -Pm -H
•H Gii <"<0 N
IrH
•IT)
117
(x) 30NVH3 A1ISN31NI
E-v
UJo
00•
II•8
CO 0)(D ,q
•H4-1On
CON
•HMo
0)
e
(0
oHU
(0 O
(0-P-H
Ev
uQ
_J<
CM (1)(0 IS3
<NCN
I
CJ1•H
118
30NVH3 A1ISN31IMI .• .
V
LJ
O'O0)
UQDI-
oM
C •-!oN -P
•H ITJMo aK O
C O<U rH•H U0)
(0 O ^IS O
COa) IIH .£Q -P(0 -H
-H CM (1)
fOOJI
in
119
o o o oo o o oo o o oO °o \& *t-,*- CO CO CO
O O _ .o o o o o o o o o o o o o o o oO O Q O O O Q O O O O . O O O O Q
<O vD ^- CM O °°CM Ovl CM CM CM
CM
(«) 30NVHO A1ISN31NICM
OOCMI
\L •
uQ
0)
•H
O
04
c
•H
o
C(U
00
.§0).a
OEH
U
ZV
uQ
(d O£ O
COd) IIrH XA -P(fl -H
•H CM (U
IrH
in
lONVIQVd
120
(x) 30NVHO A1ISN31NI
10IO
sg:s*R .£ I- r- 00 2 Z« < < _J D 5 M3 _1 o-> < ';> 3* Q
inn i i i—t-
te
uo
UQDHP_J
W0)rH•H4-1O
oN
T3(1)
-Pid
C 3Q) OH iH0) U
(0 •>» O
O(U UrH x:
XI +J(0 ft
•H CtH 0)(0 N> ^
in(NI
03I
(XI
121
(x) 30NVH3
uQD
LUQ
H
O•ora a)
0) X)i-l iH-H rtjmo *fi IC oON -P
•H (0MO Q<K O
G O<D t-l
•-H U0)
(0 OIS o
0) II
nJ-H
vorg
I
C0)
122
UlO
in0)
rH•HM-lO
t-H
C OON -P
•H nsnO ftffi O
EH
COi-lQ)
uoD
_J<
o
(0
r~-
irH
•
in
•Hfa
oin
II••£!-P•HC<u
CO
wo 33NVK1Va
123
(x) 30NVHO A1ISN31NI
.0.
V
LJQDHH_J
zv
uoD
IIO-o
CO 0)Q) XI
rH rH•H <•WO -
C ooN JJ
•H (0MO ftffi O
-P 73Cr> 3C OQ) rHrH U0)> *a o!S o
0)00
,XI -P(0 -H
•H C>H (U(C isi
00(NI
IT)
tr>-rH
124
C/5
LJ a-I U— UJ^ aO Ocr uiQ- 9
cc
2U ma: <Ld CCu. <L. >
OO
O00
OIV
O•3-
oCM
o—to
o
OCO
OIS.
.0'n
o
LdaDI—
ca(U
V-l <NC4
IIc oO 13N a)
•H jai-l iHo <K
DI OC o
O'r-4
I
30NVH3 .A1ISN31NI
.V
Uo
I-_J<
c0) OIS! O
IIQ) A
X!(0•H
(0>
E•HN
tr>•H
125
o(T-
Ooo
ovD
Oin
On
Ooj
o
30NVHO A1ISN31NI
o
0
01
oCNJ
1o
1o
1o
1o
1o
1o
ooIc
v
LJo
Ev
uiHiDHP_J
10<U
o ~J-i in
cO
IIoT!
N 0)
o <;
Ui oC orti o
nX II-P ^•HCQ) ON O
II0) A.H -P
m•HS-l
or-HI
H•
in
Cn•H
•HN
126
U 05
o oct uia. 9z u
§8ct <
yUl mcr <ui 5u. <u >3
oo
ocr>
Ooo
oCO
OCM
30NVHO A1ISN31NI
JO
18
Jo
JO
joiI
ilo
o•tI
uo
UQ
_J
0)
to<u,-H-HM-JO
C! OON
-HHO
tJI OG O
<C on
.C II-P ^<•HC -0) 0NJ O
II<U fl
ns•H
ni
in
•HEn
N
(H3IS V.. WO0 6
127
inUi G5_i u.— • ufe oco e>cc uQ- 9
-au itu. <u. >3
in Ooo
inin
in(M
O(M
in
(x) 30NVHO A11SN31NI
o
UoD
V
UQDHH-
U]
t-l-H
M (N
IIC OO T3N Q)
•H XI
O i^K
Q) CXCrH0^ OC ortl O
x: u4J ^<•HG ^(U ON O
II0) XIrH -Pxi 2<^J 6-H -HM N(0 <C> ^
CMCO
I
in
Cn•H
128
C 8fegCC UJQ- 9
z 4
u. >3
oCT--
OCO
O(X
O O10
O O(M
O
30NVHO A11SN31NI
SSoS
rIs
teiJorisJo
_V
LdQD
_J
UJQD
_J
030)H•HIHo
cO,N
in
tr> oC ortj O
.C II-P <<•r\
0) O *N O
II
•PQ)HX!m
•H
<c
rom
I
tn
129
S oto <
z
Ono
inu a-J UJP ujtjr CtO oC£ UJ0. 9
o: <
~- UJ
u aitr <Ld Eu. <u. >
ID invo
in OCO
inCM
O(M
o
V
LdQDHH_J
to0)
M oo
O T)N 0)
o
Cr> oC O< o
c\j
O
(x) 30NVH3 A1ISN31NI
S S S 8 o
!<» i i i—i inn i i i—i-£r
/ / '
cc.UJ
z5
.... iiiii i i i i inn I i I 1 in ii i i l
t—
_)
1-
«
K
U)
II
§IIICD
y?
N
Ul
*" " 1» <M :CD,, |
52
O
oc\jiO
COi 10I
£v
-p <•Hfi0) O
(D
(d•H
n
g-HN
(«31S
130
inU 55
ui_Jr uJi: CCO oo: uiQ- 9
LC <
LJ 2u. <LL. >5
o. oC1
o(M
OO
ocr>
OCO
Orx
O0-)
o
EV
LJQDh-
tnQ)iH•HIHo
CN
c oO -GN Q)
•H X2M r-H
O <ffi(U
•HCnC(<
oo
m
(x) 30NVH3 A1ISN31NI
S O O _ o<j> rS O n
V
UoD
_l<
OoIIx:-PI•HN<
IT)roI
O O
131
(x) BONVhD "A1ISN31NI
£V
uODI-
V
LJQ
COQ)
IIc oo -oN 01
•H £1M .HO <
0) CX•Hty> oC oi< in
inA II-P r<•Hc(U O
Csl OII
0) 43r-i 4J
(0•H •H
N(d <;> ~
roI
in•
Cn
132
UJ_) us.—• uj^ Ko ocr. ua. 9
. 2
CC <
3! -S< N
i y• LJ coo: <u. 5lu. <u. >
oID
On CM
OCNJ
? o--—to
10Too
O
.oro
OCM
.O
30NVHO A1ISN31NI
ot-HI
" K ' ffl - E " t°5 < < .j D < N ;'"'3 _l -,T < '/> J o !
i—i i i i—-lo
V
QJQ
HI—_J<
uoD
_J<
in<u•H•HIWO
cON
co
IIoT!0)X!r-1
Q) CXiHDI OG O<; in
LT>
£1 II-p ^-HcQ) OC<3 O
XI(C
-H
nI
IT)
e•HN
133
8 2-i 1 1 1 < 1——i 1
coU 55_J LJ
O o<r uiQ- 9
Z W
§^Ct <
CE <u 2L. <U. >o
30NVHO A1ISN31NI
.o
.O
jo
i°T^
0CM
0O
0OO
0
>
(oCM
)
oCMCM
OOCM
0CO
d d o" D I3~ C\J
i
00
ooo
0vD
0•J-
0 0CM
'
OCM
o
rV• ^
u
_i<
£v
uQD
to0)
iH-Hmo
cON
oas
CM
IIo
,Q
exooo
C<
.C-pCQ) OIS!
II<U
to>
COnI
Cn•H
-P3e
-HN<
V-.NDo «•
BONVIQVd
134
inLJ 55_l u— uU. CCO oCE UQ- 97- u,
cr <
U mcc <U 5u. <u >3
. o*• CMCM CM
OOCM
OCO
OvD
O OCM
OO
Ooo
O O•f
"6"CM
(x) 30NVHO A1ISN31NI
oTo
OCMI
£v
uoD
_J
UlQ)rH•Hmo
cON
•H
A-P
•H
mIIoti0)
D- CW
QOOr>-
o o _ ovj> n O c»Q) ON O
.V
uoD
tu
ft•H
CO>
rn
in
s:-P
3g
-HN
135
Oto'
Id 5>
C ^O O
Q- 9z wS sK <
U 05cc; <
<>
"P oco
OCM
OO
OCO
o10
O .0 OCM
o-to
. o
.0og
.0
zV
UQ to
Q)rH-HMHO
co
C OO *&
a) crtrHtjl O
30NVH3 A1ISN31NI
zV
UoD
_J
-P •<-HcOJ ON O
0) .£rH -P
03•HS-l
o•I
•HtSl
136
Zy
uo
a)
M-l x-O CM
Oc -ao <uN X!
0)rH Ot7> OC O
(x) 39NVHD A1ISN31NI
uoD-
_J<
•H -c o0) ON (Tl
II0) £rH -P
-H -HH N
> —
V7ND /Silver!)o <•
137
r.to(Ur-T•HII I
O
cON
•H
0)
as -CM0}
•H otn oo
(*) 30NVHO A1ISN31NI
.V
UQD
X! •<4J•H -C O0) Ots: en
II0) JCi -Pxi g
(o 6•H -HW N(d <C
<N
IrH
•
in
O
138
uOD
o
M
-HH | ^^'O ooi-i
CM II
C •§O (UN X!
-H rHn <;oK
c«Q)rH Ot7> OC O< rn
(x) 30NVHO A1ISN31NI
o i?82§
Ev
uoD
c oQ) ON CT,
II0) ,CrH 4J
to eM Nfd <C^^ ^-^*
ro^<
I
tr>-rH
V0
NO • 30Nviava
139
iny gfe g^ s
cr
U mCC <
U.u.3
10i o*-<
1'.
T*dc\j
inCMI
d01lof>I
O
O
..O
te
V
UQD.h-
W<UH•Hm x-.O CN
qoN•H
cwOo
30'NVHO A1ISN3.LNI
^ o o o oo «<-> «^ *
UJQ
-p•H -G OQ) ON CTi
IIQ) ,£H -M
(0•H
IH
•
IT)
•
tr>
-HN
140
U)Ld ui_J U."- UJ
n "•ft uj0. 9
DC <
-! U)< N
u Scc <u 5L_ <Lu >Q
O O»-l
1OC\J
1ioC\J
1OCO
1
o12
O
.O
.o
O^>1
V
u
30NVHO A1ISN31NI
050
i-H•Hi| j ^^O ininp; II
OC 'OO (UN XI
•H rH
M <:
crtQ)H Otn oG o
II,c ^-p•HC O0) O
UJQ
in«*I
•Ht,
•HN
V^30Nv:ava
•' 141'
U 55_i uP u§'§
CL 9
(£.<
U m(T <ui 5u. <U. >
O«-HI
oCMI '
CMI
OCOI
O
I
(x) 30NVH3. A1ISN31NI
Iotoo
Y:V
UQDh-
UJoZ)
w(UrH•HII I
O ooH
PM II
CON•H
a)
C<!
0)
OOo
-P-rt »C OQ) O
tS3 CTlII
Q) X
rcJ £H•H -HM N(C ff,> —
IiH•
in
V-.ND0 2.
142
1
LiJ Oi_J u
O 0
Q. 9
2T U
E <
< N
z u
tf ®U £u. <L- >3
J
1 \Lo O' ^J O ^> O ^o O ^o O ^o
I ' I I 1 I 1 I
OoV- «
0en
Ooo
O(X
O
0
O
0
o<M
O
0
V
uo a)
•H•HinOn
CM
ON
•H
0)•HtjiC
CX
OOmm
(x) 30NVHD AIISN31NI
uiQ
_l<
•H •>G O<U ON
(U
(0•H
II-p
-HN
tn•H
(d31S V.,N30 '
143
-r - i . < - f — ..
COU 05_J LJ
O oK U)0. 9
Z kJ
N <*£ <
1 t
< N
H jj
U CDcr <u 5!u. <u. >3
O 'o O
1 1 1
._ .
uo Oi
1 1 1 — » ' 1 1 1—
'
ol-H
0
o00
OIX
0•X)
Oen
O•*-
1
rfe• j
ui o wi O in O co*7* C\J C\J CO fO ^- <J-1 i i i i T i
(x) 30NVHO A1ISN31NI
Ev
uD
uQDHH_J
tnQ)iH•HH |
O
oC T5O 0)N ^3
•H rHM <:oK
crt0)rH Otn oC m< in
IIA ><4J-H -C O0 ON CTl
(!)-P
ig•9(0-H -HS-l N(0 <
oo
IH
•
IT)
tr>•H
144
PR
OF
ILE
S
(DE
GR
EE
S)
2 ^
|x
1 Z-J LJ< Ni —\L, LJ
U mCC <U (tli. <U_ >3
• " >— — . ..,.,, • ' • • ' .t • • . . •• | i ^
£ijgin
O•t-
O
oCV
or-l
r~\
>o O 10 O in O ^o Ci lri <5^" T T 1 ^ i ' * ? ' i T
(x) 30NVH3 A1ISN31NI
V
uQD
.V
UQD
_J
(1)rH•HUH ^~.O CO
fi-oN•H
ow(D
II
Sa)jarH<*
crtOOIDin
•H -c oQ) ON CTi
II0) .C
(0 g-H -HM N(0 <
CTl
33NVIC1VH
145
h *O o
Q- 9
o: <
?ii Z
U mo: <u 2
io O'«-•i
O'ooi
C\JO'ni
o
ZV
uQ
(00).-(•HII I
On
ON
-HMOK
(1)
-§0)
OH
OO
30NVHO A1ISN31NI :
uQ
4J•H ••c o0) ON CTi
II(U ,£rH 4J
n N(0 <C^ ^"^'
oinI
tr<
146
"J U)-J UJ
O oct uQ- 9
u £u <u. >3
O<Mi
O'<*>l
O't-i
O00
TIV
|0TvD
PTio
i:V
Q
I- 0)
O Lf>M& II
C -§
8 'J8•H i-H
Q)crtOoo
(x) 30NVHO A1ISN31NI
oS2S
uQD
•HC OQ)tsi
XI(0
•H
(0>
inI
Cn•ri
IIx:-p
-rlN
147
8 o Si 2 2j | i 1 ' -» L. - i l - - 1 ' j
(/>Ul oj_J Ur~ ujO 0Ct UJQ- 9
z ^
5 <
J 1
isu £u. <u. >o
1 _ 1
••.
OO1-i
ocr-
O00
Orx
0
oCni
0'to
\ 1°\ TCM1 1, , , , \
0t-t
<-)
i i v . . i i
(x) 30NVHO A1ISN31NI
8 0 o oO (-> <J> to
Ey
Q
y
LJS
_j
to0)H•Hti i ^_^
O oo
o tuN XI
•H i-lH <O
.03 «•cw;Q)H Ot? OC O< r-
II45 •-<4J•H vC O(U O
tSl <T>II<D .e
CM.inI
•HEM
g-HN
(y31S-V 7N3o <• . BONVIQVH
148
U u>_l u— ub- itO OCK UJQ. 9
!-; Z
is
z ri' Lu CD
(X <
U. <U. >Q
10
O10
Oro
10CO
Ofio
te
|o•C\J
o*-H
I
LJQ
I-
W0)H•H x-xM-l CMO •H II& O
•OG Q)O J3N rH
<u orH Otr> oC n
30NVHO A1ISN31NI
uQ
•H OC O<U 00
0)
•H14
nir>
I
-P3g
-HN
149
• — 1
CO
UJ 55_J ur ui-u. Q;o o(£. UQ. 9-7- U -
£ <
j|
1 — —
\P u \5 a! \cr < \u a \u. < \L- >Q
to Ot- 1-
1 -f —1 1 — t • r — - ,-, •. .-.. 1
I 1 1 1 ._ j_ j ^
I -
1
V
1
i
u
.0cr>
OCO
O(X
O
0
o
o
oC\J
o
, f~\' o O l o O L o O l ^ ' O ^ Ooo ro c\J CM • " ' • - ' ' I •-"
V
UJQ
CO
CON'
o•o(U
(x) 30NVH3 A1ISN31NI
Offi 0<
(U Oi-l OCn oC n
UQD
_J
-P -•H OC O0) oo
<D ArH 4JA 3(d £
-H -HM N
g 5
ini
IT)
-HPn
(H31S
150
oWl
tou </>H afe gCC U)Q. 9
TT UJo r»N 2Ct <
< N
U m
,.u. <LL >
in«-
o•*-
O
UQ
CO0)
14-1 00oM
coN
•HM
s<D
€
n (NJ (\1
OOO
(x) 30NVHO A1ISN31NI
(y31S /SllV/Art) 30NVIQVa
151
i:V.
uQD
J<
-p •»•H Oa o0) 00
a,rH -P
•9 §m g•H -HV-l Nn) <
ininI
in
•en
(f)y ss *O OK UQ. 9
Z U
§§o: <
< N
iaUJ mo: <u 5L_ <u. >
co (v vo in CN! -» <MI
COI
Y.V
QZ)
_J
Q)H
<4HO
CON
IIOTlQ)
O "SB CM;Q) O
Oo
(x) 3ONVHO A1ISN31NI
V
UQ
I-
-P•H OC o0) 00
inI
Cn
-P3e.
-H
152
u
CK U0. 9Z UJ0 dNao
z yU mcc <U 5u. <L. >
CM ^q o <r> oo rv *jj ^- co
V
uoD n
a)
M-I inOs-i II04 O
cON
-H
a)
O -ffi crt0) OH O&1 ^3
c •*
U) 30NVHO A1ISN31NI
ZV
uQ
Hh-_J
-P•H OG O(U oo
a> IIx:-P
-HN
in
CM
O
mus V P W O /siivnri)O 2
153
uQD
•H
en
| t ^_i
M-l 00OH II
CU OTf
C 0)O -QN H
•H SC
K 0<
<U OrH Otr> o
(x) 3ONVHO A1ISN31NI
c\jo
O
O
CMIO
?"s^h- H
UJ j
l_ _ > UJ2 1- 1- (D
5 _) U) <
mil i 1 I i
3 «• .. •
n "• cym it
z rin _i a
inIO
CM
zv—/
uD
-P
•rH OC O0 CON H
0)
(0
J-l(0>
00in
I
in
.-PP
(W31S V-.N30 «
154
u
Q. 9z uj
a
U 2u. <u. >
OCO
toU ®Q H
t -B *?i- H II*
C 0)O *QN r-l
•H <MO *as o=c0)
2 ?
ooininII
30NVHO A1ISN31NI c o0) 00
808
UJ
S
0)•HAa
ifl
mi
J:
•HN
30NVIOVW
155
CM
O
35Ul
Ct UJa. 9z. u
CNJ exl
(x) 30NVHO A1ISN31NI
808
. 5510 UlUJ Ul
l Q™c 80 9Q. Uj
o <N
§1
*J
-* yv 3o I
o*-
II
O
O
<\JlO
COI
O•to
0.
e,-PU CMQ .. c\|
CO ti II
^f! -rOH
y t . 8 <o uz»- H m z JIS < < _) D < N
O
.Ooo
O
:O
nVfc^
u
UQ
03(U
MH inO •M II
Ck O
(!)CON
1(U OrH OIji inC in< II£ ^-P•H OG o0) ooN rH
II
rH -PrQ »3
ITS g•r) -HM -N(U <!
o
IrH, •
m
oin
O O
156
U ui
r 8fegcc u0. 9
tt <§5
<
t-
zuN
Z 4'UJ 5)fC <u 2u. <L_ >
T\l oI I
vOi
COI
O
O
o*-tI
LJQ
wQ)iH
.mO
oo
IIOxi(!)C
ON
•H «CS-lO
EG (X
Q)
C<
OOin
(x) 30NVHO A1ISN31NI
O
v.
uaD
-p-H OC O0) COSI rH
Q)
to•H
voI
IT)
e•rHN
BONVIQVd
157
UJQ
.D.raa)rH•H -^U-J CM
04 O
T)CON
(U
ex0)
(x) 30NVH3 A1ISN31NI
oO .
&> oC r^< II
en uU u=J ou. uO 9Q. U
O <N
II-L UN
3 u. _j
o
cco
(T
Q.
O 'UJa
is
" <M
O-to
lotoo
fe
is
;S j?:". lor !<*>
r-1
2 2
1
oo
(«31S
«-H
1
-0-«-4
0 2
OJ
O
/S11V/
io.—I
^rt) 33f^
io•—i
iviavy
iniO
. . U
1
O•H
uQ
b_J
-P --H OC OQ) 00
Q) x;iH -PXI 3fl g:•H -HH N
(Nvo
IT)
158
01U 55_l u)C £O otr u
uo
0
U) m<£ 5U a
V °o' IN! ^ ^ ^ co1 exl
35NVH3 A1ISN31NI
u
U. UJO 9o:Q. Ul
z ^o <
UJ
CD<Z<>
§O -eg
t .8I- r- CD< < -I
CM
O
O
O
o
OCO
.oCM
c\jiO
•*-i• o
Ev
u
I—•
V
UQ
_J
<D
CON•H
inIIO
13(U
Ow CM;0 o
rH Otji OC r-<, II
•P•H oc oQ) 00
•Q)
(0•HMHJ
COvoI
in
.-P3g-HN
159
o—(-
in
-O O(X UlO. 9
oN
•5o
< N
U) CDa: <ui 5u. <u. >Q
CM
o-to
..o<T-
looo
.Orx
&;iOiorte|0Tro
UoDt
COQ)
bs-i
c Q)O X!N H
•H rijMOK CXC
Q) OrH OCn oC P-
(x) 30NVHO A1ISN31NI
SoS 0tmt » t i——JQ
ZV
uQ
A-P•n oC o0) ooN iH
•H -Pja 3m g•H -rlM N(0 <
vcI
160
(%) 3ONVHD A1ISN31NI
LJQ
H
V
uQ
M0)
O OM oPi ^5
CON
•H
O
II
"-8Q) Q).Q rHH tr>
Q)
XI
•H
(0
IT)
IH
•
in
-rHfa
4J•HC(U
V,N3 /SliV*nr()o <•
161
in ;:UJ §
C uO Q
T t
§ 5 ;N Q.
0 *~X R§— 1 3 CO
s °
i i oU £ o- u. < z
• u. > <i".. Q
. 1 /*-' oJ. ^ °°
a1
J
/(
o •«O -; ; t \J
«^> - CJ O °°
u
5
J
/
'
•J3
• 1
< l - C v J O C O t - O ^ ~ ^ J O
.
.
•
-
c\l1
OO
o
o00
O
O
f>lo
o
.O
3ONVH3 A1ISN31NI
ZV
uQD
v
uo
_l<
U]0)
•H"4-1
O OOO
O -<N
•H •«n oO Offi n
O II•dQ) Q)
(1)
^ ^!XI -P(0 -H-H CH 0)
VOVDI
Cn•Ht,
'•' (162
o<MCvl
U 2
U- uO q£ t
Q O3 Qa u-i a>U _j
u 5
ooC\J
Is
1°T
O
CJ
uQ
_J<
w<U
•H
o oo
inC IIo ^N
-H -. n oO oK ro
(x) 30NVHO A1ISN31NI
V..WD
o o«-l *-*
30NVIGVd
Ey
uoD
0) 0)XI rH
0)^H £XI -Pm -H•H GM 0)(0 N
Cr«•H
163
o o o o o o o o O OOO vC
ot-
oCM
o oC\JI
UoDI-
CO
r-l ^-•H CX4-1O
coN
•H
OOOr-II
(x) 30NVHD A1ISN31NI
uQD
_l<
O
0)
H
0)
(!).Htn
4J-HC(1)
COV£>I
(cd31S
164
(x) 30NVHD A1ISN31NI
inu
iZo<rQ.
•z.oNEoI_J<p
LJCCUU.u.Q
£
uQ
Z3t—
•J
8)
Q 82 oO UJo ffl
<U) "
1CO ^~•< pCC 0< z
.
t— 1
o0'
.000
OTIN.
OVo
I
T. n
• . kIoCO
1°tei
1 Lo «-|V*'
oo
(Aa)
M-lO
CXooOnIIC
ON
•HH OO Offi oo
O II"3(1) 0)
Q)•HXI(T3
-HM
uQ
•Hfa
C«:
-HC0)
(y3is V W Do
165
f-ro
(ft Pyio "cr ^
- °- P -z 50N ft
O *~I Q 3
^ Q ,1 O UJ
< d 3 < 'P < :
Z jli) m KK S °
t.
_.5
c\i o' °°.CO CO C\l
^ . -
' . . ' . /^> *•' c\l o1 «! ^c\I c\I c\i csi "^ *^
'
^ • o j o 0 0 * ^ 5 ^ J ~ c \ J O^» ^ ^
CM
1
OO
ocr>
000
Orx
O*&
Oin
O•f
OCO
O
O1-4
30NVH3 A1ISN31NI
V
UQ
•I-
v
uQ
(0(UH•H
O OOo
C IIo ^<N
•HM OO offi 00
II
Q)
OT3
<UXI•H
"Q)
XI -P(0 -H
-H CM 0)
Or-I
in
Cn•H
166
o5 o
12Lo
loTto
oc\j
o
Zv
uoD
CO
•n cXM-IO on opu in
inC IIO -<N•rl -M OO Offi 00
(x) 30NVHO A1ISN31NI0)
V
uQZ)
<1J"HtT>
-P-H
IT)>
I,H
«
in
167
CvjPO
(/>
yU. U)
§§Q. t
2 <ON Q.
o OE>3d3
Ucru 2<5
30NVHO A1ISN31NI
o
o
.o
.0
UQD
_J<
T.V
UQD
U]0)
14-1o(-1
exooor-
O <<N
•H vn oO oK oo
OrC
Q)J3rH<0)
II
d)rHtnG
XI -P«3 -H
•H Ci-l 0)
168
WQ
oo
i-iQ)-PC
•HIS
CO0)
(WO nO/N) A1ISN3Q 2
zV
uo
-P W-H <UW 3C rH0) (CQ >
Q) DC rHON , +1
•C CC (01C
CrH • (C0) 0)
ror-I
Mo
(WO m/N) 3NOZO
169
CNJ
C
OCMO o o o
*—I *—t r-i
(wo m/N) A1ISN3Q
ZV
u
WQ
Oo
•H
U)Q)rH-HIHo
uoD
-HinC0)Q
U)
n)>
0) OC rHON +1,o•a cC (8
CrH (0flj 0)W K
0) O
IiH•
in
tn-HPL|
(NO m/N) 3NOZO
170
I 1 Mil I t I I I H-H-t;! I I I . • Q
<NJ
O
oCMo
cr>.-4O
COv-l
O
.0
i f' i i II M ' I—I 1—-—IH I I I I—I 1 •—II I) I t-» I—-t KD '•IXr-4
O
<J3»-»
O
inT-^
O
•*-«-<o
uo
_J<
(wo no/N)
wQ
H
oo
0)
to0)rH•HIHo
(WO 3NOZO
:• -171
EV
uOD
4J W-H 0)W 0C rH(U (0Q ><U DG ioN +1o
10
(0
CB J -
C(0(U
Q) O2 M-l
V.
uQ
_l<
QD
(WO n O / N ) A1ISN3Q
oo
(0En
-HmO
^J [fl•H 0)W 3C M0) (0Q ><U DC rHON +1O
T3T) CC (0(0
CH (0nJ d)
0) OS >w
vor-~
I
(W3 HD/N) 3NOZO
172
zV
LJoD
-J<
(wo no/N) AHSN3Q
wQDEHH
aoo
0)-pc
M0)iH•H<4-lOM04
L.V
uOD
_J<
4-> W•H 0)CD 3C rH(1) (0,
QJ DC rHON +1O
TiC<0
<U2
r-r-
I
in
C(0
<Ug
inO
173
I t -+" -*4H H-t »*>•• MM M 4- 4 f MH-t * 4—4—4- HI M I I t—1 HH I 4-f- It MM t-4 I
CUD 00/N) A1ISN3Q 1Vain3N
o o o o—1 —• r-l r-1
(wo na/N) 3NOZO
CO
o(X
o
UlQ
wQDEHH
oo
-HMa
to(D
iH-H4-1Oi-lcu
-p-HWc<i)Q
<u t>G iHON +1
•a cC (0R)
cH (8(0 (UM g
Q) O
oo
I
W<u
174
o o,-* T-H
(wo no/N)
wQ
EH
$
oo
S-l0)
01<urH-HIHO
l I > l IMII ) t I—I lllli III I KD
•H 0)tn 3C rHo) <oQ >a) e>C rHON +|O
Ti13 CC (0m c
rH <0(0 Q)
-P
0 O
tn•HCn
O O O^H •-! -1
(WO nO/N) 3NOZO,
175
(ND nO/N) A1ISN3Q IVdlfGN
paQDE-iH
12Oo
(0
CO<u
scu
ZV
LJQD
_J<
-P CO•H <Uw 3C i-l0) (0Q ><]) OC HON +1O
TS
(0
®13
o00I
Cn)
Il-lO
(NO nO/N) 3NOZO
176
(wo no/N)
zV
LJQD
_J<
wQ
H
2;oo
i-i(U-pc
-H
to(U.
On
-p•HtoC
0)CONo
C(0
(0
4J30)
ooI
r-l
•
in
•Hfe
inQ)
D
(00)e
O O O O«—* r-t w-t «H
(WO 00/N) 3NOZO
177
QDH-
WQ
HEH
2Oor-
C-HMD4
(0<u
cH•HIWO
(WO 00/N) AlISN3a.~IVyj.n3N
COC0)Q
0)CON
rt
-P
0)
ooI
COQ)
>
t>
§
a)6
(WD H3/N) 3NOZO
178
1-H-I-+—t- Illl I I I I 1 HIM I I I 1 -H++1-H— I 1
WQDE-i
oo
(WO 00/N) A1ISN3Qoi-lfX;
-P CO•H 0)10 3C rHQ) rOQ >
Q) D
ON +|O
0)2
ooI
C(0
(WO 00/N) 3NOZO
179
O O O O*-4 T-l <-4 -1
ra/N) AHSN3Q ~ivyin3N
HQ
HEH
3
OO
(0
en0)rH•HM-lO
04
V
uo
.p in•H 0)w 3C H<U <flQ >
Q) OG HON +1O •o•d GG (0
CrH (8id a)
a)3
ooi
in
Mo
(WD 3NOZO
180
10
10 ~::
10 -*::
Z 10 •*=:
Aerosol Extinction @X = 5500 A
~2LgHo-2LH-X
10 "•::
10 -"::
_JOCDOcrLJ
10 -"::
10 -7.
10 -6.
0-5 km Elterman (1970)»
5-35 km Elterman (1966) '
35-80 km Exponential (0 .57 x Rayleigh @5500 A)
10 20 30 60 70 80
ALTITUDE (KM)
Figure 5.1-85 Aerosol Extinction Profile181
0.1 2.50 10 20 30 40 50 60 70 80 90
v LATITUDE l°l •
Figure 5.1-86 Stratospheric Aerosol Latitude Factors
182
5.2 MODELED AEROSOL FUNCTIONS
This appendix gives the analytic expressions obtained (by a
multivariate regression procedure discussed in Section 3.3 for scattering
phase-functions P(0.; a , X , n, n1) at 6L = 0°, 20°, 60°, .110°, 164° and.180°,
the total cross-section v,(a,\, n, n1) and either the absorption cross-section
v ( c t , X, n, n1) or the scattering cross-section <r (a, X, n, n'). Thesea sexpressions are grouped according to the particle size ranges (Q.Ql-O. l / j ,
O . I - I . O A / , and 1.0-10/v).
In the 0. 01 to 0. 1/y range we have:
P(0°) = .exp(-2.937E-2 + 5.234E-2 aX - 4 . 7 4 7 A - 5.812E-2 aA2
- 3.388E-2 a2 + 3.259 X2 + 6.963E-2 a2A + 1.195E-1 n2
- 3.961E-2 a 2 A 2 - 3.744E-1 nA)
n
P ( 2 0 ° ) = exp(-2.396E-l + 4 .447E-2 aA - 4 . 6 8 7 , A - 4.803E-2 aA
- 3.229E-2 a2 + 3 .069 A2 + 6.728E-2 a 2 A . + 1.295E-1 n2
- 3.887E-2 a 2 A 2 - 1.451E-1 n 2 A )
n
P ( 6 0 ° ) = exp(-1.776 + 6 .964E-2 aA - 1.906 A - 9.225E-2 aA
-1.402E-2 a2 + 1.388 A2 + 2.051E-2 a2A + 9.914E-2 n2
- 2.343E-1 nA - 4.050E-3 n a 2 A 2 ) .
0 ' OP(110°)=exp(-4.482 + 4 . 9 2 8 A + 3.946E-2 a - 3 .747 A - 1.013E-1
a2A - 2.372E-1 n2 + 6.485E-2 a 2 A 2 + 2.946E-1 nA .
+ 1.534 E-l n2A + 1.100E-2 n 2 A 2 a)
183
P(164°)=exp(-6.612 - 1.685 aX + 1.227E+1 X + 1.166 aX2
- 8.577 X2 - 3.798E-1 n2 - 7.140E-3 a2X2 + 4.126E-1 n2X
+ 5.990E-1 a + 5.850E-3 na2)
P(180°)=exp(-6.686 -1.711 aX + 1.249E+1 X + 1.179 aX2
'-'8.693 X2 - 3.715E-1 n2 - 6.800E-3 a2X2 + 4.027E-1 n2X
+ 6.117E-1 a + 5.800E-3 na2) ,
a = exp(1.51lE+l - 1.779E-1 aX +1.189E-2 naX2 +3.959 X2s
- 1.604 a - 9.541 n"1 - 1.043E+1 X + 2.689E-2 a2)
a = .007259 + 77.895 n1{(2.58(a/2)2'35 - .79a) (X/.4)-1* 3} ~1
{1 + .0208 X(a-2) (7-a){| (n-1.234) |/.066}
- {.1667(a-2) (a-4) (n/1.3)5'75(.06-n1)}{1+7.5(X-.4) (X-.6)}}
In the 0.1 to ly range we have:
P(0°) = exp(6.415E-l + 2.302 naX + 1.992E-1 nX2a +5.082 nX2
+ 1.841E-1 a2X2 + 9.519E-1 na - 1.404E-1 na2 + 9.720 nn1
- 1.533 nn'aX - 1.560E+1 nX + 1.679E-1 (na)2
+ 2.865E-1 nn'(aX)2 - 2.336 aX2 + 3.003 n2 - 1.569 n2a
+ 9.547. X - 2.570E-1 na2X + 3.095E-1 n2aX - 6.137 nn'X2
- 9.415E-1 nn1)
P(20°)= exp(-3.668 - 9.038E-3 (na)2 + 2.348E-1 naX - 3.470E-3
(na)2 + 2.986 n"1 - 8.749 a2X + 4.534E-2 (naX)2 - 4.293 X2
+ 1.473E-1 na - 3.111E-1 (nX)2a + 7.388 nX2 + 1.166E+1
Xn'n2 - 2.672E+1 n'X - 1.666E-1 n'(Xna)2 - 2.416 (nX)2
+ 2.292E-1 n'(Xa)2 + 8.322E-2 a + 2.412E+1 X2n' + 1.235E-1
n'n2a - 9.250(Xn)2n')
184
P(60°)= exp(-5.344 - 1.100E-2 n2a + 3.714 n3X + 1.168E-1 n
- 2.729E-2 (naX)2 - 4.697 n2X + 2.213E-1 (n)n a + 2.068E+1
n'X/a - 1.715 Xn + 2.363 naX - 1.429 n2aX - 4.949E-1 nn
+ 3.470 n2/{(l+5n')a> - 2.016E+1 n2a~3 -I- 5.328E-1 (n1)'5
4- 9.547 n2a~2 - 3.757E-1 n2(l+5n')~2 - 2.719 n1 + 1.058E-1
exp(-10 n'a) - 4.793E-3 (na)2X - 2.276E-1 n2 + 2.996E+1
(n'X/a)2 + 4.266E-2 n2(l+5n')~1)
P(110°)=exp(4.751 - 4.099 n2aX - 8.511 n + 3.866 aX + 9.015E-2
(naX)2 - 3.342E+1 n1 + 8.520 Xn'a - 3.278E-12 na12/X
+ 6.585E-1 a2X + 2.164 n2a - 5.439 aX2 - 3.228E+1 nX2
- 9.783E-2 nn'(aX)2 - 7.001E-1 (aX)2 + 4.486E+1 nX
+ 3.729E+1 X2 - 1.196E-1 na2 - 5.399E+1 X + 1.259E+1 nn1
- 3.436 nn'aX2 - 5.098 n'(na)2 + 1.143E-I-2 X(n')2 + 6.578
naX2 - 2.303 na - 2.545E+1 a(n'X)2)
P(164°)=exp(2.472E+l + 7.151E-1 (a-2) (a-4) (X-.4) (X/.6)"la°8
+ 2 .546 (n-1.35) |a-7| (n/1.4)"3 '88 - 4.058E+1 nn1 + 7.673E-1
n n ' ( a X ) 2 - 1.056 a2 - 5.101E+1 X2 - 1.843 a2X -f 7.514E-1
nn 'a 2 X - 1.471 a + 1.769E+1 aX - 7.267E-1 (1+50 n ') /ct
+ 2.921 na2 + 5.524E-1 X ( n a ) 2 + 7.434 ( n a X ) 2 - 1.854
n ( a X ) 2 - 2 .226 X/U+50 n1) + 4.555E+2 ,n (n 1 ) 2 + 3.278E+1
nX2 - 7 .245 aXn2 - 1.336E+1 na - 1.815E+1 n + 9.171 n2a
- 1.379 (na) 2 + 1.352 ( a X ) 2 + 2.562E-1 a ( n X ) 2 )
2P(180°)=exp(1.340 + 5.193E-2 ln{ (a-2/X) (2/n ) (1+50 n')
- 2.180E+1 n - 1.070 n n ' ( a X ) 2 - 1.092E-1 (na) 2 - 1.379E-1
na2 + 3.134E-1 Xa2 + 3.732 nn 'aX 2 + 5.469E-1 (naX) 2
185
- 3.927E+1 nn1 + 8.804 n'aX - 8.464 n2aX+ 1.187E+1 X2
- 3.627E+1 nX2 + 3.261 aX - 1.255 (aX)2 + 1.495E-1
n'(na)2 + 1.427 n' + 3.153 naX2 + 3.236E-1 (a-2/X)2
+ 4.157E+1 n(n')2 + 3.377 aX2 + 8.490 nX + 3.580 an2
- 5.849E+1 X)
4 ' 4 .a,,, = exp{2.3{X A.a3 - exp ( B .aD) Hl+. 036a (n-1.4) I'1}. j=0 D j=0 D
.where AQ = -8.896 E-l A, = 1.288
A2 = -5.316 E-l A3 = 6.727 E-2
A = -2.855 E-3
BQ = 1.948 Bj^ = 1.159 E-l
B2 = -2.551 E-2 B3 = 2.491 E-3
B = -8.889 E-5
a = -1.518E-3 + FX {1 +12.5U-.6) (X-.8) (.215 + DX + .5(n-1.3)3.
CX + (n-1.3)(ri-1.4) BX>. + 25(X-.4) (X-.8)(.25-.03(7-a)
- .7(n-1.3)) T- 12.5(X-.4) (X-.6) (.47 - .032(7-a) - .7(n-l.
EX - .2(n-1.3)(a-4)(a-2)/15)}(1.61(a/2)2'32 - .ei)'1
where FX = 1.372 (n'/.02)AX
AX = .585 + 2.5U-.4) (.138 - .0163(a-2)) + .0644(a-2)
CX = |(a-2)|a~a/8
BX = a2 CX/8
DX, = | (n-1.3) | (n/1.4)~10*21
. EX = I (a-7) la/12
186
In the 1 to 10y range we have:
P(0°) = exp(8.457 - 1.357 a + 6.576 X + 9.534E+1 n1 - 6.383 a2
- 2.374E+4 (n1)2 - 8.454 X2 - 1.596E+1 Xn'a2 + 2.076E+3
X(n'a)2 - 7.498E-1 aX2 - 1.161E+1 (aX)2 4- 9.258 na2
+ 8.853 X(na)2 - 3.303 (na)2 - 3.573E+1 nn1 + 1.211E+2
Xn'a + 1.375 Xn'n2 - 6.465E+1 Xa(n')2 - 3.425E+3 Xn(n')2
- 9.796E+1 (Xan1)2 - 3.656E+1 an1 + 1.262 n2 + 5.706E+1
na(Xn')2 + 4.348 n'a2 + 5.356E+2 a(Xn')2 + 2.365E+3
n(Xn')2 + 1.036E+3 X(nan')2 - 7.508 n'(Xna)2 - 6.004
(naX)2 + 1.221 aX + 1.677E+1 n(aX)2 - 2.909E+3 nX(an')2
- 1.155E+4 (nn1)2 + 3.359E+4 n(n')2 - 1.199E+2 an'X2
+ 9.740 nX2 + 1.736E-3 nn'a2 - 1.195E+1 nX - 2.378E+1
nXa2 + 1.717E+1 Xa2 + 2.148 nn'(aX)2)
P(20°)= exp(-6.551 - 5.221E+1 nn1 - 6.469E+2 naX(n')2 + 6.866E+1
n(aXn')2 + 3.327E+2 nn'(aX)2 - 8.600E+4 (nn1)2
+ 3.156E+3 aXn1 - 1.193E+2 n'(anX)2 - 2.377E+2 n'(aX)2
+ 7.901/n + 2.525 n2 - 5.764E+1 n1 - 1.585E+3 X2
+ 5.231E+2 Xn2 + 9.972E-2 (na)2 - 1.822E+1 Xa2
+ 2.823E+1 (aX)2 - 8.998 X(na)2 + 2.520E+1 nXa2
+ 2.304E+3 nX2 - 8.260E+2 (na)2 - 1.941E-2 na2 + 2.430E+5
n(n')2 - 1.689E+5 (n1)2 - 3.983E+1 n'X2 - 1.402E+1 nan1
+ 9.910 a(nX)2 + 1.379E+1 (naX)2 - 4.360 naX+ l.OOOE+3 X
- 3.913E+1 n(aX)2 - 1.560 na + 1.587E+3 Xan'n2 - 4.434E+3
nn'aX - 1.463E+3 nX - 1.903E+1 naX2 + 1.422E+1 aX + 6.747
an'n )
187
P(60°)= exp(-2.065E+2 - 1.942E+3 n' + 1.348E-1 aXn2 + 3.908
(aX)2 + 5.889E+2 an'X2 - 2.647E+1 (n1)2 + 3.968E+2 a(Xn')2
+ 1.789E+2 n"1 + 3.761E+1 n2 + 3.453 aXn1 + 2.123E+2 nX
- 3.399E+2 na(n')2 - 2.754E+1 an2 - 1.225E-2 a2 - 8.192E+1
Xn2 - 1.354E+2 X + 5.389 Xa2 - 4.490E+2 na(Xn')2
- 1.054E+3 n'n2 + 7.534E+1 na + 1.800 (naX)2 + 1.274E+3
(nn1)2 + 3.513E+2 an'(nX)2 - 2.453 n'(Xna)2 + 3.865E+1
n(aXn')2 + 1.520 an'n2 - 4.816 aX2 - 5.124E+1 a - 5.320
n(aX)2 + 2.786 E+3 nn1 - 9.080E+2 nn'aX2 - 5.976E+2
X(nn')2 + 1.106E+1 (arm1)2 - 7.938 nXa2 + 2.916 X (na) 2
+ 3.193 naX2)
P(110°)=exp(2.558E+2 + 5.608E+1 a(nX)2 + 1.412E+4 h1 - 2.236E+4 nn'
+ 8.719E+3 n'n2 + 9.753E+3 (n1)2 - 2.447E+2/n - 4.403E+1 n2
- 1.587E+2 naX2 + 2.201 a(nn')2 - 1. 782E+1, aX - 1. 075 an2
- 1.409E+4 n'X2 + 1.440E+4 X(nn')2 + 1.113E+2 aX2
- 4.937E+3 Xn'n2 - 1.838E+1 an'n2 - 2.539E+2 aXn' + 7.063
nn'X + 1.474E+4 nn'X2 - 3.422E+3 n1 (nX) 2 + 1.419 (ann1)2
-5.161 Xn' a2 - 5.923 n' (na)2 - 4.134E-3 (na)2 - 2.600E+3
an'X2 - 5.367E-1 X(ann')2 - 3.557E+3 (nn1)2 - 2.589E+4
nX(n')2 - 7.125 aXn2 + 2.336E+1 naX + 6.357E-3 a + 1.431
na + 5.565E+3 (Xn')2 - 1.618E+3 an'.(nX)2 +.4.130E+3 ann'X2
.+ 1.549E+2 aXn'n2 + 1.079 nn'a2)
P(164°)=exp(2.689E+2 + 1.618E+4 n' + 8.978E-2 a(nX)2 -2.602E+2/n
- 2.900E+4 (n1)2 -- 3.299E+3 an'X2 - 3 . 436E+4 ah (n1 X) 2
+ 6.153E+3 n'n2 - 2.036E+4 nn' - 4.332 n'(aX)2 + 1.493 aX
+ 1.669E+4 (nn')2 + 2.443E-2 na2 + 2.274E+3 nn'aX2,
188
+ 4.619E+1 cm'(n-A)2 - 5.877E+1 na - 1.269 E-l (aX)2
..- - 1.429 Xn2 - 9.151E+3 aX(nn')2 '+ 7.659E+1 X(ari')2
- 1.297E+3 aXn'n2 - 8 . 899E+4; ,n.(Xn') 2 - 5.749E+3 Xn1
- 4.313E+1 n2 - 8.996E+3 a(nn')2 + 2.629E4-3 aXn1
+ 1.798E+4 a(n')2 - 1.885E+3 an1; + 1.339E-f 3 annf
- 1.059E+5 aX(h')'2 + 5.003E+4 a(Xn',)2 + 8.643E+4 aX(n')2
+ 4.100E+1 a + 1.062E+1 an2 + 4.159E+3 Xnn' + 1.225E+5
(Xn')2)
P(l'8d°)=exp(1.759E+2 .+ 1.334E+2 n'.+ 3.142E+4 aX(nn')2 - 2.056E+3
X(an')2 -. 3.018E-fl n2 - 1..074E+3 Xn'a2 + 4.978E+3 (n1)2
+ 5.162E+2 aXn'n2 - 2.187E+1 (aX)2 - 1.068E+1 (naX)2
,. +:6.399E+2 rin'Xa2 + 2.924E+5 naX(n')2 + 4.239E+1;Xn1(na)2
+ 1.371E+2 aX' -n'6.876 aXn2 - 9.354E+4 a(n.')2 + 5.932E+2
• X(nn')2 - 6.690E+2 nn1 (aX)2 - l'.508E-f3 Xn1 — l,663E+2/n
- 1.945E+2 naX - 6.090 n'an2 + ;1.004E+3'n1(aX)2 - 9.508E+3
a(nn')2 + 8.024E+4 an(n')2 + 1.169E+3 n'X2 + 1.629E+5
na(Xn')2 - 2.267E+5 a(Xn')2 + 3.438E+5 aX(n')2 + 3.062E+1
n(aX)2 - 3.701E+2 an'(nX)2 + 1.939E+2 n'a2 - 1.748E+2 nn'a
- 8.555E+1 n'(na)2 + 1.618E+3 nX(an')2)
OT = exp(-1.498E+2 - 1.020E+1 a + 3.730 a2 + 8.622E+2 nX
- 2.967E+2 n'(aX)2 + 1.548 (na)2 - 5.009 na2 - 2.402E+2
n'(naX)2 - 4.761E+2 n'Xa2 + 1.484E+2 (nX)2 + 4.393E+1 n'a2
+ 3.614E+2 X2 - 6.377E+2 X + 2.016E+1 n2 - 4.702E+2 nX2
- 2.852 Xn2 + 5.636E+2 nn1(aX) 2 + 1.086E+1 n'(na)2
- 3.906E+1 nn'a2 + 1.545E+2/n - 1.476E+2 Xn'(na)2
1&9
+ 5.258E+.2 nn'Xa2 + 8.306 na + 3.476E+3 an1 (nX)2
- 5.662E+2 aXn'n2 - 1.181 aX2 + 3.284 a(nX)2 •+. 1.239E+1 aX
- 1.423E+1 naX + 5.827E+3 an'X2 - 9.137E+3 nn'aX2
+ 9.489E+2 nn'aX - 8.068E+1 n'a + 3.763E+1 n'n2
- 1.981E+3 ,Xn' +. 2.406E+3 nn'X - 1.159E-1 X'(na)2
- 6.879E-1 (aX)2 + 8.424E-1 nXa2 - 7.663E+2 Xn'n2
+.. 8.076E+1 (Xn1)2)
a = 1.195E-3 - 2.507E-4 a + 1.087E+3 n' + 3.008E+3 a(n')2a+ 2.789E-5 a2 - 2.547E+2 (an1)2 + 1.191E+1 n'a2 - 1.091E+4
(n1)2 - 1.940E+2 an' - 6.414E-4 Xn2 + 4.157E+3 X(n')2
- 3.911E+2 Xn1 - 6.612E^-2 nn' + 1..380 nan1 - 4.455E+1
a(Xn')2 + 2.947E-2 n' (aX)2 - 2.262 n' (na)2 - 2.715E+2
h'(nX)2 +. 2.815 an'X2 -i- 1.853E+2 n'n2 + 2.201E+1 X(an')2
+ 2.609E+2 nn'X + 5.240 nn'a2 - 3.396E+2 n'X2 - 2.056E+2
aX(n')2; - 5,251 nX(an')2 + 1.3T7E+3 h(n')2 r- 1.977E+3
nX (n-1) .
190
5.3 MULTIPLE SCATTERING CODE C.K.W.
5.3.1 Introduction
Because it has not been given more than a cursory description in
print, and because its speed makes the model concept potentially valuable
to the scientific community as a whole, and because it is the model to be
adapted for all our future work we present in this and the following
subsections a brief description of the multiple scattering code C.K.W. in
it's present state of development.
For the program C.K.W. the atmosphere is modeled as a set of
concentric spherical layers, each characterized by extinction coefficients
for Rayleigh scatterers, aerosols, and ozone, and by an index of refraction,
all 'assumed constant within a given layer. The acceptability of assuming
constancy depends on the thickness of layers. Typically, we use 50 to 60
layers of 2 km thickness, and the calculated horizon profile is insensitive
to further refinement.
The calculations naturally fall into a hierarchy of increasing difficulty.
The simplest problem is single scattering, which is handled in great detail,
with the full spherical geometry, exact integrations including refraction,
detailed phase function data with polarization, and a Feugelson law for
albedo ( Kondratyev , 1969). The next case includes some nearly forward
scattering on the path from the sun to the site of one major scatter into
the detector line of sight. This is done with almost as much refinement as
single scattering. The next, more difficult case includes some nearly forward
scattering on the path to the detector. For this case, an approximate
estimation of coupling between scan lines is introduced. The final case is
full multiple scattering, which requires some additional .approximations
for the geometry. The following sections detail the methods and
approximations used to treat each of the successively more difficult cases.
191
5.3.2 Single Scattering
Basic Geometry
The modeling of -the-horizon profile requires that we first of all
establish a coordinate frame in which all the vectors that occur in the
calculation can be expressed unambiguously. For this purpose, we affix a
coordinate frame to the detector. The (z) axis is along the local vertical
at the detector, the (x) axis is perpendicular to the plane of scan, and the
(y) axis lies in the scan plane, pointing roughly in the direction from which
the detector seeks light (Figure 5.3-1). This choice of coordinate frame
is arbitrary, but convenient on several counts. First, this coordinate frame
is independent of the detector altitude. This is useful because one of the
applications for horizon profile modeling is in spacecraft navigation, .for
which the altitude may be unknown. Second, the zenith and azimuth of the
sun can be measured directly from the craft, so the vector representing
the input sunlight can be defined directly for this frame. Third, in the
special but common case that the sun lies in the scan plane, the (x) axis is
perpendicular to the scattering plane, so there is a coincidence in regard
to the naming of polarization states: both the usual geometric conventionand the convention used in scattering theory agree that polarization along
(x) corresponds to the same direction in Poincare space.
A second geometric requirement to be met is the specification of
the earth and its atmosphere. Thus we introduce the height of the detector,
(HDET), and the earth radius (RE), shown in Figure 5.3-1. Also we fix
the radii to the bottoms of each of the layers, RT :i_j
RL = RE + L KM (5.3-1)
where (KM) is the single-layer thickness and (L) is an integer member of
set (NL) defining the number of layers. Next we precalculate geometric
quantities that will be used over and over in the horizon profile calculation.
192
These are the scan path increments through the layers, DXT for L"= 0(1 )NL.J _ j - ' . , ' . ' '
For the scan path tangent to the earth, the path increment through layer
(L) is
-C' • • ' - " • - . • . DXL £ ^2 RE KM (/L+T - ,/L) (5.3-2)
Almost these same path increments occur on all the other scan paths too,
and so the numbers are saved for use on all the scan paths.
The final item of geometric information to be handled is the location
of any cloud. We denote the altitude of a cloud by the variable SOLIDTOP
in Figure 5.3-1. If the cloud happens to fall between layers, then it occludes
aportionof some layer. This portion is called SOLIDTIP. In the subsequent
calculations, any path increment through the partially occluded layer should
be adjusted from the appropriate DX^ above. The adjustment factor is
SQ = /KM - /SOLIDTIP (5-3.3)/KM
Average Attenuation
The attenuation within a path increment (DD) is one of the factors
that determine how much radiation is scattered to the detector from (DD).
• At any point in (DD), the radiation arriving from the sun is damped by
passage through a total optical depth denoted by (TTS). Similarly, the
radiation scattering to the detector is further attenuated by passage through
' 'a total optical depth denoted by (TTD). Thus at a point, the attenuation is
-TTS - TTDe
193
A full contribution to the output from the whole increment (DD) is of the
form
J a-TTS - T T D - E a . p . W dxDD i
where a. and p.(0) are the scattering coefficient an'd phase function for the11i constituent/and (dx) is. a differential increment of length. Assuming
constancy of all extinction coefficients in the layer (or at least their local
ratios), we split the contribution into two factors:
'*-' /.-DD DD
TTS - TTD ,dx
and
DD
The first factor is the average attenuation, and its evaluation is the subject
of this subsection. .
To perform the integration in the average attenuation we introduce a
change of variables: .
dx = J d(TTS + TTD) (5.3-4)
Here (J) is the Jacobian of the transformation,
J = l/(d(TTS + TTD)/dx) (5.3-5)
194
We assume the derivative in (J) can be approximated by
A (TTS + TTD)/DD
Thus we reexpress the average attenuation by
f -TTS - TTDJ eA(TTS + TTD) DD6 ^" d-<TTS .*. TTD)
This can be evaluated immediately as
: •:.-.- .... DAMP: = -Ae~TTS TTD/A..(TT_S + TTD), , , (5.3-6)
The above trivial result.is modified ;slightly by the inclusion of the
phenomenon of refraction. The basic principle involved is that light travels
not exactly in a straight line, but rather in a slightly curved line. At a
point on the line, we let (n) denote the refractive index, (r) the distance to
(the earth center, and sine the angle of incidence on a surface parallel to
.the earth. The curved line is described by - . , . - . _ ;
(n)(r) sine = constant (5.3-7)
where the constant is the nominal tangent height RR of the scan line.
Referring to the greatly exaggerated Figure 5.3-2, we see that where the
unrefracted path would be intersecting the bottom of layer (L) at an angle
whose sine is Ru/R T , the refracted path is passing through a somewhat• - • • i-P- ' . V * ' ' • ' • , • : • • .
lower point and at a lesser angle. Let (y) be the nominal scan line, and
(z) be the perpendicular displacement of the true path from the unrefracted
oheV Then the true angle is
f l . . /M • '/dz-.-^V , . " • : : " - - . , . - . . . -9 ^ arcsmr—1- arcsmf^ — J ( 5 3_8 }• • • - . ' , ' • ' • - • • \ L I / \ T J / - • • • . • - . '
195
We also have
/RH \arcsinfjp RiNDEXLJ (5.3-9)
\ L /e
v
where RINDEXT is the refractive index in layer (L). The derivative dz/dy\-ican be isolated and simplified by appropriate series expansions. It is used
in the code to develop the true path as a function of the nominal path, and
so to provide the correct boundaries, etc., at which to evaluate (TTS) and
(TTD).
Scattering Factor
Besides the average attenuation discussed in the preceding subsection,
the contribution to the output from a path increment (DD) depends on the
previously introduced scattering factor of the form
£ a± PI(T|») DD
Since the atmosphere being modeled contains both Rayleigh and aerosol
scatterers, there are two a's and p's that have to be evaluated. In both
cases, the (CT) is the local extinction coefficient, denoted as (RDTAU) for
Rayleigh scatterers, and (ADTAU) for aerosol scatterers. Also in both
cases, the polarized output (if any exists) can be expected to be linearly
polarized in the plane perpendicular to the scattering plane, so that in both
cases two polarization parameters of the phase function constitute sufficient
input data. Finally, for both cases it is essential to specify the scattering
angle (0). Actually, this is done by specifying the cosine of that angle
(COPSI). We simply form the dot product of a vector representing the
input sunlight and a vector representing the light traveling to the detector.
In the case of Rayleigh scattering, the phase function can be calculated
analytically within the program. Ideally, the two polarization parameterssy .
are 6/1677 and 6 COPSI /16j7 . Sometimes it is appropriate to modify this
196
slightly. But for aerosol scattering, the situation is much more complicated.
At present, we read in the phase function for selected values of (COPSI),
usually 46 in number, and interpolate on the basis of (COPSI).
Albedo
Scattering from the atmosphere is only one of two possible mechanisms
whereby light reaches the detector. There is also scattering from the
surface of the earth, or some cloud layer above it. The total fraction of
the incident light scattered from such a surf ace is its albedo. The simplest,
.and most commonly used angular. dependence for surface scattering is the
well-known Lambert law:
• - - ' . yradiance out = (irradiance in) ALBEDO -2. (5.3-10)
where (A/ ) is the cosine of the incidence angle. It is known that natural
surfaces usually are not Lambertian. Nevertheless, better models are
only occasionally available. Our model has incorporated a Feugelson law
for single scattering from clouds, but otherwise uses Lambert's Law.
The Feugelson law for the approximate albedo (A ) of a cloud of opticalC*
thickness (r) is given by
2y + 1
A = 1 -
(5.3-11)uo/3 + c 2 ( 3 v - D/8 {1 - 2u (ci/3 - 1)}
where c^ = 2.48 and Cg = 3.70. : .
5.3.3 Forward Scattering on the Sun Path
After single scattering, the next less simple mechanism whereby light
reaches the detector includes one or more nearly forward scatters on the
path from the sun to the primary scatter. This subsection details the way
197
in which this contribution is handled by adapting the methods already
described in the preceding sections. That is, we discuss modifications to
the geometry, the average attenuation, the scattering fraction, and the albedo
calculation.
Geometry
The geometry used to calculate the contribution to the output that
includes some forward scattering on the sun path differs from the single
scattering geometry only in that what is assumed to propagate along the
sun path is a stream rather than a plane wave. The explicit definition foro
the stream (Whitney, 1972) is radiance integrated against a cos weighting
function. There occurs in the calculation a factor (FPRL), which is part
of the stream definition, and factors (NORM) and (ESTPRL) that relate to
the inversion of streams back to radiances at the output. This inversion
is discussed in detail in a subsequent subsection on full multiple scattering.
Attenuation
The notion of a stream rather than a plane wave propagating along
the sun path modifies the average attenuation in two ways. First, since
forward scatters do not remove anything from the stream, forward scatters
have to be excluded from the optical depths along the sun path. This is
done by modifying the optical depths by a factor of (1-PHS), where (PHS)
is a combination of streamlike phase function integrals that represents
forward scattering from a stream into itself. The construction of (PHS)
is detailed in the above mentioned paper. The modified optical depth from
the sun is (MSTTS). With that replacement, the analog of (DAMP), the
average attenuation, is
MSDAMP = -Ae~MSTTS ~ ^^/ L (MSTTS + TTD') (5.3-12)
198
The second modification required is that we exclude from (MSDAMP)
the case in which absolutely no scatters occur on the sun path, since we
have already counted single scattering. This done by decreasing (MSDAMP)
by that part of the average that represents just single scattering. Thus
we put
MSDAMP = MSDAMP - DAMP
Scattering
The scattering factor is also affected by the inclusion of forward
scattering along the sun path. Where before it was appropriate to have a
plane-wave transformation, or phase function p(0), it is now appropriate
to have a stream transformation, or (PHS), evaluated for the nominal
scattering angle (0). For both Rayleigh and aerosol scatterers, these PHS's
are entered as data, precalculated at the angles that occur in the dodecahedral
arrangement of streams introduced in (Whitney, 1972) namely 0°, 180°,
63 and 117°. An interpolation routine then obtains the (PHS) for the desired
angle (</>).
Albedo
The final modifications required to accommodate a stream on the
sun path concern the albedo calculations. For a Lambertian surface, a
discretized version of Lambert's law is required. The continuous law
involves the factor | cos(0)| In-, where
TT = / | COS (9) |dfihemisphere
199
That is, the law is formulated to say that uniform radiance input provides
uniform radiance output, just scaled down by the albedo. The discrete law
must be formulated to say the same thing, but with the integral replaced
by a sum over the dodecahedron angles, and the |cos(e) l replaced by
averages of |cos(0)| with quadratic weighting functions. The normalization
for all the averages is
(p-k)
hemisphere= 2TT/3 (5.3-13)
The actual averages are as follows: for 0 ,
< | c o s ( 9 )•j I /\. /N.
= ^ I JU-k)
hemisphere
dfi = 3/4 (5.3-14)
ofor 63 ,
<|cos(8)|> = -^ I (z-k)
hemisphere
hemisphere
hemisphere
= 9/20
A A A A O4z-k(x-k)
(5.3-15)
200
Thus the sum analogous to the integral is
5 terms =
3/4 + 9/20 = 3 (5.3-16)
Thus for the discrete case, the (77) in Lambert's law is replaced by a 3.
5.3.4 Forward Scattering on the Detector Path . •
We suppose now that the photons sustain one or more nearly forward
scatters on their path from the primary scatter to the detector. This case
is slightly more complicated than that .discussed in the preceding section,
and so entails a few more modifications, in regard to geometry, average
attenuation, and scattering factor.
Geometry
The major geometric consideration with forward scattering on the
detector path is the possibility of transfer of radiation from one line of
sight to another at a slightly different angle. To handle this, we formulate
for each line of sight an approximation for the probability that forward
scattering ultimately contributes to that particular line of sight. This
probability varies from zero for the uppermost line of sight to unity for a
scan tangent to the earth. In between, the probability is approximated as
follows: the difficult feature requiring approximation is that where forward
scatters are -concerned we must think not of contributions from a path
increment, but rather from whole three dimensional regions. That is, if
subsequent forward scatters are allowed, contributions to a given scan line
come not from the line, but from a region surrounding the line. We formulate
the above mentioned probability function as a line integral of the form
PMm = / e~ T ' dT1 (5.3-17)
201
where the path over which the integral is done traverses, roughly, the optical
region that can, with no mechanism other than subsequent forward scattering,
contribute to the m scan line. The path proceeds from the detector,
through atmospheric layers down to and including the mth one, and then
out again. The AT' to be associated with any particular layer has to be
approximated in a rather gross way, because with the curvature of the
earth, the different propagation .directions in a stream can traverse
drastically different optical depths. We begin by formulating Ar1 in such
a way that if the whole atmosphere were condensed in one uniform layer
(whose geometric thickness would therefore be the atmospheric scale
height), that layer would have the.optical depth tangent through itself:
AT' = EXTINCTION /2 RE SH (5.3-18)
Thus for a layer of thickness KM, we put
AT' = EXTINCTION /2 RE KM /KM/SH . (5.3-19)
This formulation guarantees that for very small (KM), (r1) converges to a
well-defined function of the tangent height associated" with (PM) and is
insensitive to the number and geometric thickness of layers that happen to
be used in the atmospheric model. Finally, we must consider the effect of
the scattering phase function. Clearly in the case of extreme forward
scattering, the contributing region that Ar1 represents should shrink to
zero; that is, forward scattering becomes ineffective for transfer from
one line of sight to another. We represent this volume effect by modifying
Ar1 to Ar1 (1-PHS)3. ,
Attenuation .
The attenuation for contributions to the output that will undergo one
or more forward scatters on the detector path is modified in several ways.
First, the attenuation on the detector path is modified by a factor (1-PHS)
202
to omit forward scattering just as the attenuation along the sun path was.
The modified optical depth on the detector path is (MSTTD), so that at
each point the attenuation is
-MSTTS - MSTTDe
Second, as we have already counted single scattering and forward scattering
on the sun path, we have to subtract the portion of the function representing~MSTTD~TTDthose events: e . Third, and novel to the case where there
are to be subsequent forward scatters, there is no one unique and readily
identified path over which this function should be evaluated and averaged.
Recall that instead of contributions from a path increment, we are considering
contributions from a spatial region. Instead of attempting to average the
attenuation over a region, we simply leave it as a point function, evaluated
at representative points (e.g. the tangent points of scan lines directed to
the bottom of each atmospheric layer).
Scattering
The absence of a particular path to integrate over is again evident
in the scattering factor, for no particular path length (DD) appears. Instead,
(DD) is replaced by a standard length for subsequent forward scattering,
which is just the /2 RE KM /KM/SH previously introduced in connection
with the PM's.
5.3.5 Full Multiple Scattering
The final and most complicated mechanism whereby light reaches
the horizon profile detector involves two or more scatters that cannot be
classified as nearly forward. This section discusses the application of a
method that handles these events, and also incidentally provides some
additional information concerning the other mechanisms already discussed.
The method is basically that introduced in Whitney, 1972. In that paper,
the integro-differential equation of radiative transfer is discretized as a
set of twelve coupled differential equations for streams. For symmetry
the streams are arranged pointing to the faces of a regular dodecahedron.
203
Geometry
The geometry of the horizon scan situation is modified slightly to
achieve a speedy application of the method without significant loss of
accuracy. Specifically, thicker layers are used (usually 20 km), and they
are imagined to be flat until the final step, where curvature is introduced
as a correction. Because the layers are flat, the introduction of a cloud
is simplified; the analog of
SQ = /KM - VSOLIDTIP (5.3-20)
/KM
becomes simply KM - SOLIDTIP
. . KM
Also because the layers are flat, the original continuous equation of radiative
transfer can be expressed with inverse cosines of incidence angles. In
the discrete case, these are all replaced by inverses of cosines averaged
with quadratic weighting functions; that is, inverses of the values 3/4 and
9/20 previously derived in connection with the discrete analog of Lambert's
Law.
Attenuation
The attenuation of contributions to the output is calculated in much
the same way as it was in the preceding section. All optical depths are
scaled by a factor of the form (1-PHS), in order to eliminate forward
scattering from the calculation. Then the integrals are done exactly. The
general form of the integral required is
204
where r, , r*, TO • • • are the optical depths at which the first, second, third
. . . scatters occur, and CQ, c, , c2 ... are the cosines for the stream
traveling from the sun, from the first scatter, from the second scatter, . . .
. . . The various T'S are constrained so that each of the exponentials is
less than unity. The constraints are handled by a tree structure, or set of
nested do-loops, that governs the calculations. The program first chooses
the layer for the first interaction (Ll), then the inclination for the scattered
stream (PI), then its direction (Ql), and then the layer for the second scatter
(L2), and so on. If the first inclination (PI) is upward (downward) the
second layer (L2) is required not to be below (above) the first layer (Ll).
In the code, the multiple attenuation integral is not done all at once,
but in pieces; since the TV integral is done in a loop which is within the
loop that does the r_ integral, and which is in turn inside the loop that£
does the T integral. As an example, the T-, integral is just
(I/Co - 1/Cl) -
Because it has been isolated from the functions of ?•„, it is no longer true
that the exponential is necessarily less than unity; it may as well be greater.
However, when all the required integrals are multiplied together, the result
is less than unity, as is physically required.
A small novelty is required to handle the case where two or more
scatters occur in sequence within the same layer. The problem for that
layer is then a micro version of the problem for the atmosphere as a whole.
We could handle it the same as the full atmosphere, breaking the layer
into smaller sub layers. But such a procedure would be time consuming,
so we attempt to get an approximate answer without resorting to it. For
the case where all the scatters from(i) through (i+n) occur in one layer, we
approximate the required integral with the product of the' r., T. + , , . . . T.+n
integrals, but with an important modification. Consider the case where
205
all the inclinations are the same. The problem then reduces to a familiar
one in Poisson statistics, and the required integral is known to be simply
the above product, but divided by (n.1). Thus we induce that generally a
factor of (1/n!) should be incorporated.
Scattering
We come now to the actual scattering operations for full multiple
scattering. These are constructed by a procedure whose goal is to minimize
the storage space and computational burden required. The main source of
difficulty to be overcome is the inclusion of polarization, requiring four
Stokes parameters s~, s., s2, s3 to characterize a stream and, in general,
sixteen parameters to characterize a stream transformation. The input
stream transformations are provided for streams in the Z-X plane separated
by the dodecahedral scattering angles 0°, 63°, 117°, and 180°. For these
cases, symmetry reduces the sixteen parameters to far fewer, so that storage
space is minimized. In particular, for 63° and 117°, there are only eight
parameters: one 2x2 matrix operating on ( gO j and another operating
(s )» and for 0 and 180 there are only three parameters, only two
on
which are independent. For any particular scatter that occurs in the code,
the stream transformation is synthesized by combining the minimal stored
information with appropriate rotations of the plane-polarization parameters
(s / • That is, first a 2x2 rotation is applied so that the polarization
state in the nominal scattering plane acquires the name 1. Then the
appropriate stream transformation is applied. Then another 2x2 rotation
is applied to restore the polarization state in the nominal scattering plane
to its final correct name.
Program Control
In the case of multiple scattering it is not enough simply to attenuate
and scatter. The program must be prevented from expending excessive
effort on negligible contributions to the output, and must be prevented from
reconsidering the contributions it has more accurately calculated in the
single scattering part of the code. These goals are accomplished by several
tests in the code.
206
Of primary importance is the so-called twig pruning, or (ENDTEST)
for multiple scattering. This test compares the current contribution to
the output with a numerical standard. The standard is based on the output
that was obtained from single scattering, on the total number of terms of
the particular scattering order that are left to consider, and on an input
parameter expressing the programmer's judgement of how accurate he
wants to be. If the test is failed, it is judged unprofitable to further pursue
that particular branch of the multiple scattering tree. Were it not for this
test, the program would run for outlandishly long times without appreciably
changing the output.
Inversion of Streams
The final step in creating the horizon profile is to take the streams
SA provided by the multiple scattering code, and invert them to estimates
of physical radiance, S(k). This problem is the inverse of that involved inA
forming the stream S A from the radiance S(k). Since forming SA is analogousP A P A
to performing an integral transform on S(k), it follows that estimating S(k)
is analogous to inverting the transform. Because we do the original
transform only for power n = 2, and not arbitrary (n), it is possible to
invert only approximately. This section discusses what is thought to be
an optimum procedure for approximating S(k).
A
We begin by considering a simple case where S(k) is assumed to beA
equal to S(-k), and to have the form
6 terms~ SP
(k) = 2, (5.3-21)
with this form, it can readily be verified that
5 terms
S(e> '• - - (5.3-22,
207
Referring to Whitney, 1972, we see that the inverse of this relationship
would be
5 terms
:. £ V (5.3-23)
As in Whitney, 1972, we have, to separate forward-traveling from, . • • : • ) - • : . ' • ; . 7 " , . .
backward traveling radiation by extending these simple six term
relationships to twelve terms. For the numbers used, the analogous
inverse is
. 1+ 8"
5 terms 5 terms3 V^ 1 NT20" X S+6' + 40" Z S-6'
(5.3-24)
This is the basic relationship currently used to invert the array of output
streams in the code,
208
SCAN LINES
Fig. 5.3-1 Horizon Scan Geometry
209
NOMINAL. UNREFRACTEDSCAN PATH
REFRACTED SCAN PATH
Fig. 5.3-2 Effect of Refraction
210
REFERENCES
SECTION 5
1) Groves, G.V., "Seasonal and Latitudinal Models of Atmospheric
Temperature, Pressure, and Density", Air Force Surveys in
Geophysics, No. 218, May, 1970.
2) Kondratyev. K.Ya.. "Radiation in the Atmosphere", 386-399, Academic
Press, New York and London, 1969.
3) Malchow, H.L., "Standard Models of Atmospheric Constituents-and
Radiative Phenomena for Inversion Simulation", MIT Aeronomy
Program Internal Report No. AER 7-1, January, 1971.
4) Salah, J.E., "On the Nature and Distribution of Stratospheric
Aerosols", Private Communications, February, 1971.
5) Valley, S. L., Handbook of Geophysics and Space Environment,
McGraw-Hill, New York, 1965.
6) Whitney, C.K., "implications of a Quadratic Stream Definition in
Radiative Transfer Theory", J. Atm. Sci., 29, 1520-1530, 1972.
7) Wu, M.F., "Ozone Distribution and Variability", MIT Aeronomy
Program Internal Report No. AER 1-2, August, 1970.
21-1