Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

570

J. Opt. Soc. Am. A/Vol. 4, No. 3/March 1987

Kenneth Sassen

Polarization and Brewster angle properties of light pillars


Kenneth Sassen Department of Meteorology, University of Utah, Salt Lake City, Utah 84112
Received May 6, 1986; accepted September 19, 1986

Numerical simulation of the nocturnal light pillar, an atmospheric optical phenomenon inadvertently caused by humans, reveals that the pillars are virtually completely polarized at the Brewster angle for ice as a result of the
geometry of rays reflected off near-horizontally aligned ice crystals from a nearby light source.

plate crystals that the first-order internal reflection contributes importantly to the display and that the depth of the crystal-containing layer and the effects of atmospheric attenuation serve to limit the height above the horizon to which the pillars are visible. The model findings have been verified with experiments involving the generation of artificial pillars from linearly polarized light sources. Both observations and model predictions support the view that the plate ice crystals causing the display have tilt angles that are distributed normally from the horizontal
plane.

It is also shown for

INTRODUCTION The evolution in scientific thought leading to the discovery of the polarized nature of light grew from early accounts of the unusual property of Iceland spar, a birefringent crystal
of calcite, to cause the phenomenon of double images.'

whose attainments

are of a higher order, that al-

Both Huygens and Newton proved by experiment that the cause of this double refraction was intrinsic to some property of light and not the result of modifications induced by the
crystal itself. Newton, in Query 25 of his book Opticks,

most all the phenomena of double refraction and polarisation, intricate and capricious as they appear to be, have been brought under the dominion of general laws, and can be calculated with as much accuracy as that with which the astronomer can compute the motions and positions of the heavenly
bodies. The angle of polarization, or the Brewster angle, was shown

concludes "the unusual Refraction is therefore performed by an original property of the Rays. And it remains to be enquired, whether the Rays have not more original Properties than are yet discovered." 2 Newton then goes on in Query 29 to liken those "Sides of the Rays" causing the unusual refraction to the poles of a magnet. Although Newton was unable to explain this property in terms of corpuscular theory, just as Huygens had earlier been constrained by his adherence to the longitudinal wave theory, this analogy to the poles of a magnet seems to have been the source of the term polarized light adapted by later researchers such as Malus and Fresnel, who contributed significantly to the understanding of polarization. The preoccupation with the phenomenon of double refraction, which proved so profitable to the founding of modern theories, led to the development, in the field of optics, of
various forms of polarizing prisms by Rochon, Wollaston,
3 and Nicol. But it was the work of Sir David Brewster that provided an explanation of the relationship between polarization and reflection at a dielectric interface. His confi-

to follow the simple law: "The index of refraction is the


tangent of the angle of polarisation." 3

In this paper the polarization properties of light pillars, a nocturnal atmospheric optical phenomenon produced inadvertently by humans, are examined with the aid of experiments and a numerical model based essentially on the theory of Fresnel and Brewster, with the model inputs constrained by observations. It is demonstrated that the conditions under which these often startling displays are generated provide a unique opportunity to observe the Brewster angle in a natural setting. LIGHT PILLAR FORMATION Nocturnal light pillars are a simple reflection phenomenon related to sun pillars in the atmosphere and the glitter paths. often observed on a gently rippling water surface. The two kinds of pillar, however, have quite different structures as a result of the geometries of rays reflected from light sources positioned essentially at infinity (i.e., the Sun or the Moon)
or locally (i.e., nearby lights).4 Moreover, since light pillars are normally seen during very light snowfall, when the observer is located within a cloud of ice crystals, it is a relatively

dence in these discoveries is apparent in the following quotation, reproduced from Ref. 1: One of the principal objects of the present series of

papers is to correct these absurd misapprehensions;


and we have no doubt that we shall be able to

render the subject intelligible to such of our readers as have but a very slender portion either of physical or mathematical knowledge; and to convince those
0740-3232/87/030570-11$02.00

simple matter to study the nature of the particles causing the display. These properties make the man-made light pillar unique among atmospheric optical displays. A climatological study of light pillar displays has shown that they occur almost exclusively at temperatures well be 1987 Optical Society of America

Kenneth Sassen low freezing when the atmosphere just above the surface

Vol. 4, No. 3/March 1987/J. Opt. Soc. Am. A

571

5 With contains diffuse concentrations of planar ice crystals. their large basal reflecting faces, simple-plate and branchedplate crystals constitute the best scattering media for these displays. As with all optical displays in the atmosphere, however,the structure of light pillars can be understood only with a knowledge of both the shapes and orientations of the particles contributing to the display. It has long been recognized that sun pillars must be associated with ice crystals that fall with their reflecting faces oriented very nearly parallel to the ground, but it was not until computer simulations were performed that the exact effects of crystal orientations could be appreciated.6 It was established, given that. the crystals must wobble about the perfect horizontal alignment, that the pillars became more conspicuous in their height above the horizon and amount of spreading as the crystal tilt angle increased. It was a microphysical and photographic study of light pillars themselves which revealed important information on the nature and cause of these ice crystal fluttering motions.7 It is well known in the field of fluid dynamics that the most stable, or "preferred," orientation of a regularly shaped falling object is that in which the particle aligns itself with its major axis perpendicular to the direction of motion. In other words, the object falls vertically at a constant velocity

LIGHT PILLAR MODEL


A variety of types of light sources can generate pillars; how-

ever, to describe the most general case, a light source that isotropically radiates unpolarized light of intensity Io is used
in our model. (This condition is approximated by light

pillars generated from exposed incandescent light bulbs, as is often the case.) The unpolarized light incident upon an
ice crystal face will then be reflected in accordance with the

Fresnel reflection formulas, producing a degree of partial


polarization that depends only on the angle of incidence Oi, or, as shown schematically in Fig. 1, on the observer's elevation angle 0 for a given crystal tilt angle A from the horizontal

plane. To determine the components of reflected light that

v, which is minimized by presenting the maximum geometric

cross section to the flow. This stable fall attitude is maintained over a range of particle diameters d in a fluid of

kinematic viscosity v. To characterize the falling behavior of particles of a specified shape, it has been found convenient to use the dimensionless Reynolds number, defined as Re = vd/v. Since the Re regime corresponding to stable fall patterns for particles with the arbitrary shapes of ice crystals can be found only through experimental means, the falling behaviors of "model" ice crystals in more convenient fluid media, such as water, have been studied in the laboratory for application to the atmosphere. By using a thin disk as the model of a plate crystal, for example, it has been shown that the kinds of crystals that generate light or sun pillars should exhibit a stable horizontal orientation in a quiescent fluid through the range of 1 < Re < 100.8 Studies of the planar ice crystals collected at the ground during light pillar displays have confirmed that this approximate range of Re for stable fall is valid in the atmosphere.
7

r /"'I//
S

/ / // //

Fig. 1. The scattering geometry of a ray from a local isotropic to an observer. source reflected off an ice crystal with tilt angle 4s

/
50
I

0
1.0

10 20

30

40

0.9

0.8 0.7 0.6

However, microdensitometer analyses of pillar photographs


have revealed additional information on the fall attitudes of

ice crystals under atmospheric conditions that cannot be reproduced in the laboratory tank. It was found that plate crystals with Re - 10 display the most stable fall attitudes and that horizontal brightness variations across photographed pillars would indicate that the ice crystal tilt angles are distributed normally from the horizontal plane. It was concluded that these departures from the laboratory model experiments could be attributed to the effects of atmospheric turbulence, which acts to perturb the crystals slightly from their stable fall attitudes. In terms of planar ice crystal diameters, the most stable particles have dimensions of
-0.5 to 1.0 mm. Crystals <0.1 mm in diameter are too

0 0.5 Id 0.4 IL
Id

0.3 0.2
0.1
0 10 203I

10 20 30 40 50 60 70 8090

influenced by Brownian and turbulent motions to maintain their preferred orientations, while crystals larger than a few millimeters develop spiraling or pitching motions as a result of the unstable flow induced in the wake of the particle.

Fig. 2. The change of parallel and orthogonal reflectivitieswith the


incident and transmitted ray angles for an ice surface in air (n = 1.30). The Brewster angle occurs at Oi= 52.4.

572

J. Opt. Soc. Am. A/Vol. 4, No. 3/March 1987

Kenneth Sassen from a single crystal can be expressed in terms of the obser-

Table 1. Forms of the Relationships for Reflectivities and Effective Crystal Reflecting Areas for the Surface (n = 0) and First-, Second-, and nth-Order Internal Reflections
n 0 1 2 Reflectivity R (1 -R) 2 R (1 -R) 3 R 3 (1 - R)n+'R 2 n-I Effective Area AcosOi A cos Oi-2r tanOt A cos Oi-6T tan Ot A cos Oi- (4n - 2)r tan Ot

vation angle from the equation


-

IOA 2 4- r

2(O)

R(O)1,,IP(X),

(2)

0.08 0.07
(I)

where r is the range from the light source to the crystal, Ae is the effective ice crystal reflecting area defined below, and P(X) is the probability that an ice crystal will have the proper tilt angle to contribute to the display. The first term in Eq. (2) describes the amount of energy intercepted by the crystal, and since the incident rays are essentially parallel to one another, it can be assumed that all the reflected energy is collected by the detector (e.g., a camera lens). The dependence of the orthogonal and parallel reflectivities on the
angle of incidence is shown in Fig. 2. The angle of polarization occurs where Oi+ Ot= r/2, or, from Brewster's law, when tan Oi= n.
(3)

z
0 0
> W

0.06 0.05 / 0.04

H
iLu -J IL

| l

An ice surface with n = 1.30 displays a Brewster angle of 52.430.

/ // /

/^\1 l / l /

0.03
-J Lu

0.02
0.01

R 1V 1
// //
A/

1
l l
|

Ray-tracing simulations have been performed to assess the impact of internal reflections on the effective crystal reflecting area Ae. The plate crystals were treated as disks with a cross-sectional area of A and a thickness-to-diameter ratio of r. The effective area of the bottom crystal face producing the surface reflection is simply A cos 0i. In Table 1 are given the effective areas of the top crystal face that contribute to pillar scattering from the first-, second-, and nth-order internal reflections, neglecting the negligible
amounts of energy passing through the sides of the disk.

/L~g X 0
10 20 30 40 50 60 70
ANGLE OF IN CIDENCE,81

8 80

|
90

The total reflecting area is the sum of the various effective


areas, but the intensity of each order of internal reflection is

Fig. 3. Relative intensities reflected iti the twopolarization planes for a thin disk with thickness-to-diamel ter ratio r = 0.1 as a function of 6,. Dashed lines represent the intensdtiesfromthe surface reflection alone, while the solid lines include i ;hecontributions frominternal reflections.

a function of the sequence of reflectivities involved in the internal ray paths, which are also compiled in Table 1. In Fig. 3 are shownthe variations of the relative reflected intensities with the angle of incidence for a thin disk (r
=

are polarized orthogonal and parall.elto the scattering plane (i.e., the plane containing the incii lent and reflected rays), 9 the reflectivities R 1 and R1I are cal culated from R
1
=

dashed and solid lines represent the contributions from the surface reflection alone and the total of the surface and firstand second-order reflections, respectively. Although it is apparent from Table 1 that n ' 2 internal reflections must contribute negligibly, the differences between the solid and dashed curves in Fig.,3 demonstrate that the single internal reflection cannot be neglected in calculating the intensities
light pillars. The actual thickness of a crystal is of no significance for a nonabsorbing substance such as ice in the

0.1); the

od )of 2(0i sin 2 sin (OR + o*)


2 o

2(0i tan tan2 (0i 4

visible spectra, but the thickness-to-diameter ratio does affect the angular distribution of internally reflected light, as
. 00 , k---/~~\.L

examined below. Perfect Horizontal Alignment

where the angle of the transmitted ray Otfrom the normal


(see Fig. 1) is found from Snell's law by using the refractive

index of ice in air (assumed to be 1.30 for the calculations). If the crystal tilt angles are small, and we know this to be the case from polarization lidar observations of cirrus clouds'0 and from light pillar studies,7 then the reflected intensities polarized orthogonally and parallel to the scattering plane are, essentially, horizontally and vertically polarized, respectively.

The intensities in the two polarization planes reflected

Ice crystals maintaining near-perfect horizontal alignments in the atmosphere generate subsuns from parallel-rayed sunlight. In contrast, because of the finite distance of the light source, nocturnal light pillars produced under these conditions will be extremely thin and can extend to great heights. If the crystal tilt angle 4 = 0, it can be seen from Fig. 1 that only crystals positioned exactly at the midpoint between the source and observer, and in the vertical azimuth plane, can reflect light to the observer. Assuming that the

Kenneth Sassen
100 100

Vol. 4, No. 3/March 1987/J. Opt. Soc. Am. A

573

>(n

10-,
C')
z

10o-

z z
t -J
g-

0-

o-

-j
>

a: 10-2

90

10-30 0

10

20

30

40

50

60

70

80

90

OBSERVATION ANGLE, 6 OBSERVATION ANGLE, 8 (b) (a) Fig. 4. The total (IT), orthogonal (II), and parallel (III) polarized relative intensity components versus observation angle for light pillars generated from crystals with perfect horizontal alignment. The two crystal models in (a), where r = 0.1, and (b), where r = 0.4, represent thin plate and thick plate ice crystals, respectively. Solid curves give the intensities from surface and internal reflections, while the dashed IL and Ij curves consider only the surface reflection.

crystal sizes and concentrations are uniform with height and that the concentrations are low enough to prevent significant attenuation and multiple-scattering activity, Eq. (1) can be applied directly to describe the variations of pillar intensity with observation angle. For this relatively simple
case, P(X) is a constant and r = s/2 sin (7r/2 - 0), where s is

the separation distance between the source and the observer.


Thus it is only necessary to include the inverse range-

squared dependency in the relative reflected intensity curves of Fig. 3 to arrive at the variations in pillar intensity
with height. These results are given in Figs. 4(a) and 4(b).

that the intensity values are identical in this region. The peak intensity occurs at 90, and since the rate at which the total intensity decreases with increasing 0 is reduced by the internally reflected contributions, these components are important for increasing the height to which the pillars can be observed. On the other hand, the apparent height of a light pillar can be limited by the depth Zmax of the atmospheric layer containing the oriented crystals. The results given in Fig. 4 were obtained by assuming that the crystals extend to
a great height, or that
Zmax >>

s. An abrupt cutoff in pillar

The variations with observation angle of the total (unpolarized), orthogonal, and parallel polarized intensity components are shown in Fig. 4 after normalization of the curves to

height will occur at the observation angle 0 = tan-'(2 zmax/s) if the relative distance to the source is not small in comparison with the depth of the layer containing crystals.
The presence of the Brewster angle is quite obvious in Fig.

unit intensity at the IT value maximum. As in Fig. 3, the dashed curves represent the intensities produced solely by
the surface reflection, while the solid curves include the

4. Observations of a thin light pillar with the aid of a rotating linear polarizer, therefore, should reveal the structure to be completely polarized at an observation angle of
37.60.

contributions of the first- and second-order internal reflections. The two thickness-to-diameter ratios represent the ice crystal models of a thin plate (r = 0.1) and a thick plate (r = 0.4) crystal. Both crystal types have been associated with
light pillars, 7 so this range of r should be representative of

Normally Distributed Alignment

It has been demonstrated through computer simulations of sun pillars6 that the subsun gradually elongates and broadens in azimuth angle as the maximum ice crystal tilt angle is increased from zero. This also has the effect of allowing the sun pillar to be visible at increasing solar elevation angles.

the crystal types involved in pillar generation. In general, the contributions from internal reflections are significant
except at low observation angles, with the effects being more pronounced for the thin plate model. At 0 < 140 no internal

reflections can contribute, even for the thin disk model, so

Although the subsun and its light pillar counterpart have little in common, simple geometric considerations dictate

574

J. Opt. Soc. Am. A/Vol. 4, No. 3/March 1987

Kenneth Sassen ZMAX


sI s sin(20

L2ip)
i

sin(7r-20
+ 'MAX

2q/

(6)

provided that only real reflections are involved (i.e., that 0


2i lies within the range of 0 to 7r/2). For 0 < 2t, 10= 0, and values of 1 = lo are not a simple function of 0 or independent of v.. In view of the importance of the function 1 = lo in
1-1

determining the light pillar structure, Fig. 6 presents calculated values for lt'ma = 0.3 and 3.00 with q5 = 0 and zma/s =
10. The curve for
Climax =

0.30 is shown multiplied by a factor

of 10 to illustrate that the form of the function is independent of a, except at low and high observation angles as the maximum tilt angle becomes significant. It is clear that, in contrast to the case for perfect horizontal alignment, this scaling factor will contribute significantly to the illumination of light pillars at both low and high observing angles. The polarization properties at 0 = 0 of light pillars produced by crystals distributed normally from the horizontal
plane with a = 0.1 and 1.0 (dashed IT curve only) are present-

ed in Fig. 7. For these computations and those to follow,the surface and first-order internal reflections of a crystal with r = 0.1 are modeled. In this section, the Zmax/s ratio is set to 10in order to examine the high elevation features produced,
for example, by a relatively close light source. In general,

the intensities of light pillars in the vertical azimuth plane are relatively invariant with the standard deviations of the
Fig. 5.

tals constrained to orientations between :limax The function I-l represents the path length over which crystals can reflect energy into angle 0, and Zmax is the top of the crystal-containing layer.

The light pillar scattering geometry for fluttering ice crys-

l0o

that light pillars will also broaden into azimuthal scattering planes 0 equal to the maximum permissible tilt angle )'rax. The scattering geometry of crystals constrained to orientations between
clmax

is shown schematically in Fig. 5, where

the path length containing crystals capable of reflecting light into angle 0 is designated as I - lo. By a process
analogous to that used for Eq. (2), we may now derive the integrated intensity along 0 and in the 0 plane from 1(0, o)iiI =

R(0) 1 ,1 IP(X)dl.

(4)

As discussed earlier, and as is indicated in Fig. 5, the maximum height of the layer containing the uniformly mixed
crystals can limit the length I as a function of 0.

10

The probability function P(X) describing the distribution of crystal tilt angles has been chosen to simulate light pillar observations indicating a Gaussian distribution of crystal orientations around the horizontal plane.7 This function is
defined as

P(X) = +/exp -(\2 +2

(5)

where a. is the standard deviation of a normal curve with a

mean of zero. For the purpose of our calculations it is adequate to specify 4l'max= 3o-. A minimum P(X) that would permit a reflection is then calculated for each simulation,
using f = 3o-and 0 = 0. The path length I - 10can be considered as a scaling factor

l0-l L
0o

30

60

90

OBSERVATION ANGLE,
Fig. 6.

representing the number of crystals, for a given particle concentration, that contribute to the display. A value of 1 can be derived trigonometrically from the expression

significant impact on the structure of light pillars from fluttering


crystals, shown for two Gaussian crystal tilt angle distributions and by a factor of 10 for this comparison.

Variations in the function I - lo (see Fig. 5), which has a

the vertical (4 = 0) azimuth plane. The solid curve is multiplied

Kenneth Sassen

Vol. 4, No. 3/March 1987/J. Opt. Soc. Am. A

575

must assume a larger range of tilt angles. For a <50.1, the intensities are essentially constant for any value of a.,and the
10-

solid curves of Fig. 7 are produced. As the maximum tilt angles increase, however, the intensity curves display a slight peak centered around 0 = 2 4'max as a result of the

special scattering geometry (Fig. 6). At the same time, the minimum intensity of IIl in the Brewster angle region increases slightly as crystals with increasing tilt angles contribute to the scattering at the Brewster angle. Nonetheless, the pillars remain strongly polarized in the Brewster
10-2
H

angle region.

z
Lu

To illustrate the azimuth angle dependency of the intensity components, the predicted structure of a light pillar with
a. = 3 and zmax/s = 10 is given in Fig. 8. dence for
4

This example

represents perhaps an extreme case, since there is little evi'max

exceeding 90 from earlier sun or light pillar

Ld a:

0-

10

20

30

40

50

60

70

80

90

Fig. 7.

OBSERVATION ANGLE, e Relative intensity components in the vertical azimuth plane

studies. The relative intensities are contoured in intervals of the logarithm of intensity, as given in the figure, in order to compress the large dynamic range of the values. The most notable features of this simulated light pillar are the relatively strong decreases in total and orthogonal intensities with increasing 0 and the indications of the very low parallel-polarized intensities in the Brewster angle region. The tendency for the IT values to increase at large 0 is due both to the behavior of IIIabove the Brewster angle and to ratio the scattering geometry permitted by the large Zmax/s used in this simulation. Examined below are the effects produced by varying the depth of the crystal layer as well as by the inclusion of atmospheric attenuation effects.
Observational Variables

for pillars generated from thin plates with normally distributed tilt
= 0.1 and a = 1.0 (dashed angles displaying standard deviations of a. is the ratio of the depth of the crystalIT curve only). zmax/s containing layer to the source distance.

normal distribution of tilt angles. This results from the fact that, although the scaling factor 1- lo increases linearly with
a., values of P(X) decrease correspondingly as the crystals

An examination of photographs of light pillar displays (see Ref. 5) reveals that the pillar images often do not extend to very great heights above the horizon. In the pillar simulations presented so far, it has been assumed that the depth of the crystal-containing layer is an order of magnitude greater than the distance to the light source, and the findings sug-

AZIMUTH ANGLE, q
-10 -5
II,,

5
I,,,

10
1-

-10
L

10

-10

-5

10

9Or
80F
LU

IL.,,,.

I,,.

>-~~~~~~3.5

-0 I2-.5

7
-

~~~~~~-4.0
3.5

-275
-2.5 -3 0

-J

70k
60F1

z z
LU

40
C/) 30 --

40 -~ ~ ~ ~ ~ ~ ~ ~ ~~~~~~~~~~~4
-2.5 -2.5 40

-4 0

2.25 co 2 0 20

- 2.25

ol

~~~~-175

-2.0
75 -1.5 -1.25
=

II
-/7 ,

~~~~~~~~~~~~~~~~~~ 0 -275
217

- 2 75

1 0
0

~~~~~~-1.5
IT
-1.25

~~~~~~~~~~~~~~~~-202

-.

= 3.0 and Zmax/s Fig. 8. A two-dimensional representation of the structure of a light pillar with a.

10. The unpolarized (IT) and horizontally


Brewster angle effects are

(Ii) and vertically (III) polarized components are contoured in intervals of the logarithm of relative intensity. apparent in the III display.

576

J. Opt. Soc. Am. A/Vol. 4, No. 3/March 1987

Kenneth Sassen

r =3.0 0 = 00
10-,

I-_

visibility,5 such that the anisotropic scattering properties of oriented plate crystals may produce k values of the order of 0.1 km-' or less. Hence, under the best observing conditions, optical extinction would not appear to have a significant impact on the brightness or structure of light pillars. (Of course, if the distance to the light source is great, even such low extinction rates could produce a considerable optical thickness.) On the other hand, pillars are also noted during snowfall that are composed of a mixture of oriented crystals and spatial or aggregated ice particles, producing much greater values of k. Under such conditions only relatively close light sources will produce bright pillars, while the upper pillar segment will also fade rapidly in brightness. EXPERIMENTAL COMPARISONS

ILU

10-2 ZMAX/S = /.0

Total (Unpolarized) Intensity

Er

A basic assumption involved in our light pillar simulations is that the fluttering motions of ice crystals from the horizontal plane can be modeled through the Gaussian probability
function of Eq. (5). On the left-hand side of Fig. 11 are

shown the findings reproduced from Ref. 7, on which this assumption is based. At the center of the figure is a photograph of a light pillar that was selected for the microdensitometer analysis shown at left. The instrument was scanned at high resolution across the pillar image at .1.30 0 10

20

30

40

50

60

70
l0-'

OBSERVATION ANGLE, 6
Fig. 9. Alterations in the total intensity curves of pillars induced by changing the relative distance to the light source in the vertical
azimuth plane. The large crystal tilt angles (a = 3.0) produce a

gradual intensity decline across the top of the crystal layer.

gest that it should be possible to observe, under these conditions, faint light pillars extending to large observation angles. However, given in Fig. 9 are the 4 = 00 total intensity curves (a = 3.0) for zmax/s ratios that are more realistic when the light source is remote from the observer. It is clear
U)

that as, for example, the observer moves away from the light source, the apparent heights of the pillars will diminish. Moreover, unlike the case of the extremely thin pillar produced by perfectly aligned crystals, large crystal tilt angles will cause the upper pillar terminus to fade gradually. Photographs of light pillars show this to be the case. Finally, in addition to the effect of the zmax/s ratio on the pillar structure, it is also obvious that the pillar brightness
will be decreased by optical attenuation along each reflected

Li

10-2

I-

-J
LU

ray path through the atmosphere. This process is easily simulated by including the factor exp -(r + 1)kin the integral of Eq. (4), where k is the atmospheric extinction coefficient. After specifying that s = 1.0 km, the IT curves of Fig. 10 illustrate the effects of increasing k from 0 to 1.0 km-'. Since the ray paths at higher observation angles involve

10-3

greater source-to-observer distances, the upper portions of pillars suffer the most from attenuation. The extinction coefficients of 0.1 to 1.0 km-' correspond to atmospheric conditions ranging from, say, a hazy sky to light snowfall, respectively. It is apparent that vivid light pillar displays are often accompanied by excellent horizontal

I0

20

30

40

50

60

70

OBSERVATION ANGLE, 8 Fig. 10. Alterations in the total intensity curves at 0 = 00 of pillars produced by changing the atmospheric extinction coefficient k.
The light source was assumed to be 1.0 km from the observer.

Increasing k or the distance to the source preferentially reduces the intensity of the upper pillar.

Kenneth Sassen 15

Vol. 4, No. 3/March 1987/J. Opt. Soc. Am. A

577

-'15

Li
CD

10

I0

z
0
IX
C)

CD

0
Fig. 11. Comparison of a microdensitometer

-o
j__
analysis (left) of the light pillar photograph at center with a model simulation (right) based on the

observations (see the text). Arrows indicate the observation angles of each total intensity profile across the pillar. The tilt angles of the crystals causing the display are demonstrated to be normally distributed from the horizontal plane.

observation angle increments (note the arrows at the approximate 6of each scan), producing intensity profiles that closely resemble Gaussian curves. This light pillar appears in the photograph to extend 13 above the horizon and was
generated by a brilliant light source estimated to be 5 km

the intensity peak at 6 = 24maxdecays characteristically (see only for 0 > 120 Fig. 10). With the assumed Zmax/S = 0.115,

are the intensity profiles significantly reduced by limiting the crystal layer depth, and the agreement in this region is
again favorable. Since this pillar was confined to rather low observing angles, the effect of atmospheric attenuation is

distant from the observation point. Under these conditions,


a shallow crystal-containing layer -0.6 km in depth would

fairly uniform at all 6 angles through this range.


Polarization Properties

have been responsible for producing the light pillar, according to our current analysis. Finally, measurements of the spreading angle of the lower pillar image yielded a half-

width divergence angle, indicative of the maximum crystal


tilt angle, of 3.2.

The model simulation of this light pillar is presented for


comparison on the right-hand side of Fig. 11. Shown are the

To verify the model's predictions of the Brewster angle in light pillars, a pair of linearly polarized light sources was constructed and deployed in a remote rural area where light 3
pollution was minimal." Each source consisted of a 10 -W

total intensity profiles at 20 observation angle increments across the simulated pillar, using s = 5.0 km, Zmax = 0.575 km, k = 0.1 km-', and a. = 1.0. (The extinction coefficient
was selected on the basis of the excellent horizontal visibility

incandescent light bulb encased in a sheet metal box that lacked portions of the top and front side such that light could be radiated outward through a somewhat greater than 0-90 range of elevation angles. Laminated plastic polarizing filters were then curved over the missing corner of each

that prevailed during this display.) Both the relative width (between the {30 azimuth angles) and the height of the
intensity profiles were normalized by maximizing the fit

between the 6= 20 profile and the corresponding microdensitometer line scans. The results of this qualitative comparison between simulation and measurement are quite favorable, particularly in the range of 0 from about 6 to 100, where

housing and mounted in place, one vertically and one horizontally polarized. Although these were clearly not isotropic sources, light was emitted into the proper volume of space to generate a complete light pillar if viewed from the front of the apparatus. Time-exposure photographs of light pillars produced by
the linearly polarized light sources are shown in Figs. 12-14.

578

J. Opt. Soc. Am. A/Vol. 4, No. 3/March 1987

Kenneth Sassen

090~ _

MMIN&L

fIZMMWMW

Fig. 12. Three photographs of light pillars produced from horizontally (left of each pair) and vertically polarized light sources. Photographs were taken, from right to left, through vertically and horizontally polarized filters and with the filter removed from the camera, demonstrating that no depolarization of the incident light occurs during pillar scattering.

The observations were collected at a temperature of -12 0 C


during a very light snowfall that consisted of a mixture of

simple plate crystals with 0.6-mm modal diameters and, in

generally smaller amounts, 2 .5-mm-modal-diameter


Reynolds numbers of -10 and 75, respectively. The struc-

branching plate and dendritic crystals. The modal dimensions of the crystals in these two categories correspond to ture of a light pillar produced under these conditions should therefore reflect the contributions from crystals with both
very small and relatively large maximum tilt angles. A care-

tions, which preserve the incident polarization state, have been included in our model, and it is also apparent that the difference in the intensities of the polarized pillars points to the presence of the Brewster angle, as predicted. The light pillar photographs shown in Fig. 13 span an observation angle range of nearly 90. They were obtained 830 field of view. The photograph on the right was taken with the camera tilted upward -45 and at roughly twice the exposure time, and it has been situated in the figure so that the observation angles correspond to those of the lowerphotograph. The horizontally (left of each pair) and vertically polarized intensities seem to compare quite well with the
model predictions of In and III shown in Fig. 7. Although at a distance of 20 m, using a wide-angle camera lens with an

ful examination of the pillar photographs taken at close range reveals that the light pillars appear to be composed of a thin and intense pillar surrounded by a cone of single reflections attributable to the larger dendritic ice crystals.
The three photographs of Fig. 12 were taken from a distance of 40 m, using a normal lens, and extend about 400 in

the reemergence of the vertically polarized pillar above the 14, obtained by pointing the camera vertically from a position between the two light sources, more clearly shows the polarized pillars converting at the zenith. This final experiment corresponds to light pillar model predictions in the limit of Zmax/S - -D. CONCLUSIONS The various factors controlling the structure and polarization properties of the nocturnal light pillar have been examBrewster angle is not easily visible in these photographs, Fig.

observation angle above the ground. The left- and righthand sources in each photograph are horizontally and vertically polarized, respectively. From right to left, the photographs were taken with a polarizing filter aligned on the camera for vertical polarization, with a filter aligned for horizontal polarization, and with the filter removed. Despite a small amount of leakage of unpolarized light from the sources, these photographs demonstrate that there is no depolarization of the incident linearly polarized light produced during light pillar scattering. Only simple reflec-

Kenneth Sassen ined with the aid of model simulations. Although only relative reflected intensities have been calculated by normalizing certain variables, use of the model permits quantitative

Vol. 4, No. 3/March 1987/J. Opt. Soc. Am. A

579

predictions. In the model it is assumed that disklike ice crystals are well mixed throughout the depth of the crystalcontaining layer, that the crystal tilt angles are distributed normally from the horizontal plane, and that there is a relatively narrow size distribution of crystals with average area A. However, any atmospheric conditions inferred from ob-

servations can be incorporated into the model, including the


situation shown in Figs. 12-14, in which the size distribution of the crystals was bimodal in character. Analogous to the case for solar reflection phenomena, the differences in scattering geometries corresponding to perfectly aligned and fluttering crystal plates are responsible

for producing quite different displays. The analog to the subsun is an extremely thin and (potentially) high light pillar, but any degree of crystal fluttering produces significant alterations in the reflected intensity distribution with observation angle and also causes scattering into nonvertical

Fig. 14. Photograph taken looking vertically from between the two

polarized light sources, showing the pillars converging at the zenith after passing through the Brewster angle region (upper part of pillar at right).

azimuth planes. In view of the strictness of the requirement for perfect horizontal alignment, it can be questioned whether true subsuns or their light pillar counterparts actually
occur in nature, however.

It has been found that, in addition to the reflection off the bottom face of the crystal, the first-order internal reflection utilizing a portion of the top crystal face must be considered in pillar formation. At near-normal incidence the surface and internal reflections are nearly of the same intensity, and, depending on the thickness-to-diameter ratio of the crystal, the internally reflected contributions can be significant even to low pillar-observing angles. Although model predictions of the total, unpolarized intensity indicate that light pillars
should be observable to very high elevation angles, we have

shown that the height above the horizon to which pillars are visible is limited by the depth of the crystal-containing layer in relation to the distance to the light source. In addition, atmospheric attenuation can preferentially diminish the intensity of the upper portions of light pillars. The most interesting feature of the simulated light pillars is that their polarization properties demonstrate the presence of the Brewster angle for ice reflections in the atmoFig. 13. Montage of artificially polarized light pillar photographs

spanning an observation angle range of nearly 900. The horizontally (left of each pair) and vertically polarized pillars can be compared with the model predictions of In and III, respectively, in Fig. 7, confirming the presence of the Brewster angle. Note the star

sphere. The phenomenon of complete polarization at 0 = 37.60 is seen in its purest form in the pillars produced from near-perfectly aligned crystals, but the pillars retain a very
high degree of polarization in the Brewster angle region even

for crystal populations containing reasonable, maximum tilt


angles. Hence all light pillars visible at 0 ' 400 are predict-

streaks visible at the top.

580

J. Opt. Soc. Am. A/Vol. 4, No. 3/March 1987

Kenneth Sassen
2. I. Newton, Opticks (Dover, New York, 1952, unabridged repro-

ed to display the Brewster angle effect when viewed through

a linear polarizer, at least in the usual case, when plate crystals are causing the display.'2 The extent to which these numerical findings have been verified through experimental observations appears to be appreciable. The Gaussian distribution of crystal tilt angles from the horizontal plane, previously inferred from the analysis of light pillar photographs by using a microdensitometer, has now been confirmed with model simulations. The occurrence of the Brewster angle in the vertically polarized component of an artificially generated light pillar has also been illustrated photographically. It is unlikely that any other atmospheric optical phenomenon will prove to provide
us with such a vivid demonstration of the law formulated by

duction).

3. The paper by D. Brewster, "On the laws which regulate the polarisation of light by reflexion from transparent bodies" first appeared in 1815 in Phil. Trans. 105, 125-130 and 158-159, but the discovery of Brewster's law dates from 1812 [see G. P.

Sir David Brewster in the year 1812. ACKNOWLEDGMENTS This report is based on an invited paper presented at the
Optical Society of America Topical Meeting on Meteorological Optics held in Honolulu, Hawaii, in 1985. The current

bridge, 1985)]. 4. R. Greenler, Rainbows, Halos, and Glories (Cambridge U. Press, Cambridge, 1980). 5. K. Sassen, "Light pillar climatology," Weatherwise 33, 259-262 (1980). 6. R. G. Greenler, M. Drinkwine, A. J. Mallmann, and G. Blumenthal, "The origin of Sun pillars," Am. Sci. 60, 292-302 (1972). 7. K. Sassen, "Remote sensing of planar ice crystal fall attitudes," J. Meteorol. Soc. Jpn. 58, 422-429 (1980). 8. W. W. Willmarth, N. E. Hawk, and R. L. Harvey, "Steady and

K6nnen, Polarized Light in Nature (Cambridge U. Press, Cam-

unsteady motions and wakes of freely falling disks," Phys. Fluids 7, 197-208 (1964).

York, 1975). 10. C. M. R. Platt, "Lidar backscattering from horizontal ice crystal plates," J. Appl. Meteorol. 17, 482-488 (1978).

9. M. Born and E. Wolf, Principles of Optics (Pergamon,New

research has been supported by National Science Foundation grant ATM82-10709. The author would like to thank G. Vali and other personnel at the University of Wyoming for their assistance in initiating these studies and S. A. Bennett for manuscript preparation. REFERENCES AND NOTES
1. For an anthology of early papers on experiments with light, see
Polarized Light, W. Swindell, ed. (Dowden, Hutchison Ross, Strondsberg, Pa., 1975). and

11.The production of these "artificial" light pillars poses an interesting question. Although the light pillar is alwaysproduced by humans, the radiation is provided inadvertently and so can be considered to be a natural part of the nocturnal landscape.
Thus I regard the previous experiment

remote sensing but the study of the polarized pillars as active remote sensing.
12. As discussed in Refs. 5 and 6, light or sun pillars may occasionally be generated by long, columnar ice crystals falling with their major axes aligned horizontally. In such cases, the simple reflection geometry for wobbling plates is no longer valid, and

(Fig. 11) as passive

complex internal reflections may contribute importantly to the display, causing the light pillar intensity and polarization properties to differ.

You might also like