Josa 54 001031

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

JOURNAL OF THE OPTICAL SOCIETY OF AMERICA VOLUME 54, INUMBEHR8 AITGIJST 196i4

Spectral Distribution of Typical Daylight as a Function of Correlated


Color Temperature
DEANE B. JUDD
National Bureau of Standards,Washington,D. C. 20234
AND

DAVIDL. MACADAM
Research Laboratories, Eastman Kodak Company, Rochester, New York 14650
AND

GtTNTERWYSZECKI
National Research Council, Ottawa 2, Ontario, Canada
WITHTHE COLLABORATION
OF
H. W. BUDDE,*H. R. CONDITt S. T. HENDERSON,$AND J. L. SIMONDSt
(Received 29 April 1964

Spectral distributions of 622 samples of daylight (skylight, and sunlight plus skylight) have been subjected
to characteristic vector analysis, as composite data and in three subgroups (99 distributions measured by
Budde; 249, by Condit; and 274, by Henderson and Hodgkiss). The chromaticity coordinates (x,y) com-
puted from these distributions have been compared with direct visual determinations of chromaticity coordi-
nates of daylight by Nayatani and Wyszecki, and by Chamberlin, Lawrence, and Belbin. It was found that
the chromaticities indicated by the spectral distributions and by direct visual colorimetry cluster about
the curve: y = 2.870x-3.000x2 -0.275. This curve of typical daylight chromaticities falls slightly on the green
side of the Planckian locus. From the mean and the first two characteristic vectors of the composite data,
spectral distribution curves have been reconstituted by choice of scalar multiples of the vectors such that the
chromaticity points fall on the curve of typical daylight chromaticities at places corresponding to correlated
color temperatures of 48000, 55000, 65000, 75000, and 10 0000 K. The representative character of these re-
constituted spectral-distribution curves has been established by comparison with the measured curves from
each subgroup yielding the closest approximation to the same chromaticities. The agreement so found sug-
gests that this family of curves is more representative of the various phases of daylight between correlated
color temperatures 48000 and 10 0000 K than any previously derived distributions.

INTRODUCTION increasingly used in the reduction of spectrophoto-


EVERsince light sources B and C were recom- metric data to the calorimetric variables (Y,x,y) of the
mended in 1931 by the International Commission 1931 CIE standard-observer system.'
on Illumination at its meeting in Cambridge, England, The development of the Davis-Gibson liquid filters,
they have served as standard sources for the colorimetry and their use to realize standard sources for colorimetry,
of materials.' These sources are combinations of a gas- was an important step toward an internationally recog-
filled lamp at a color temperature of 28540 K with nized general method of colorimetry, but standard sources
Davis-Gibson2 liquid filters intended to approximate sorealized have limitations that prevent them from lead-
within the visible spectrum the spectral distributions ing to a completely general method. For example, the
of the complete radiator (blackbody) at temperatures ultraviolet content' is too low to permit these sources
of 4800° and 6500'K, respectively. Source B (4800'K) to be an adequate representation of sunlight or day-
has served in the colorimetric laboratory as a repre- light in the colorimetry of fluorescent materials.
sentative of noon sunlight, and source C (6500'K) has Furthermore, the standard sources cannot yield the
served as a representative of average daylight. This high luminous flux required for the most precise in-
laboratory use of these standard sources has steadily spection of industrial products because the liquids in
diminished in recent years because of the increasing the filters would become heated above the boiling
dependence placed on indirect colorimetry by means point. Because of these limitations, because of minor
of the spectrophotometer; but the spectral energy inconveniences arising from the use of liquid filters
distributions of sources B and C have been widely and (difficulties of preparation according to the careful
prescription given, need to replace the filters every
* National Research Council, Ottawa 2, Ontario, Canada. few months because of drift in their spectral character,
t Eastman Kodak Company, Rochester, New York 14650. temperature coefficient of spectral transmittance), and
1 Thorn Electrical Industries Limited, Great Cambridge Road,
Enfield, Middlesex, England. because improved glass filters could be made available,4
I Proceedings of the International Commission on Illumination,
8th Session, Cambridge, 1931 (Cambridge University Press,
Cambridge, England, 1932) p. 19. I R. Davis, K. S. Gibson, and G. W. Haupt, J. Res. Natl. Bur.
2 R. Davis and K. S. Gibson, "Filters for the Reproduction of Std. 50, 31 (1953); J. Opt. Soc. Am. 43, 172 (1953).
Sunlight and Daylight and the Determination of Color Tempera- 4 G. N. Rautian, N. V. Lobanova, and M. A. Znamenskaya,
ture," Natl. Bur. Std. (U. S.), Misc. Pub. 114 (1931). Z. Tech. Fisiki 26, 193 (1956). D. B. Judd, Farbe 10, 31 (1961).
1031
1032 1032'JUDD, MAdACADAMl, ANTD \NYSZECI( Vl
Vol. 54
54

a proposal was made by the Russian delegate to the A meeting of U. S. Technical Committee on Colorim-
CIE Committee on Colorimetry in 1957 to define etry was convened in Rochester in October 1962 to
standard sources solely as blackbody radiators and consider the results of these questionnaires and to
to substitute glass filters for liquid filters in their explore what next steps ought to be taken in the
5
realization. development of standard sources for colorimetry. All
As a part of the studies of this Russian proposal, members of the committee were present (I. A. Balinkin,
undertaken at the request of the CIE Committee on University of Cincinnati; D. B. Judd, National Bureau
Colorimetry, a questionnaire devised by one of us of Standards; D. L. MacAdam, Eastman Kodak
(GW) was circulated by the Canadian Committee on Company; D. Nickerson, U. S. Department of Agri-
Colorimetry, and a similar questionnaire devised by culture; and W. T. Wintringham, Bell Telephone Lab-
D. Nickerson was circulated by the U. S. Technical oratories). Present also as guests at one time or another
Committee on Colorimetry. were W. E. K. Middleton and G. Wyszecki, National
The replies to the Canadian questionnaire have been Research Council of Canada; Norman Macbeth,
6
summarized as follows: Macbeth Corporation; and H. R. Condit, 0. E.
Miller, and J. L. Simonds, Eastman Kodak Company.
"Question 4.-Do you suggest that the present CIE
Macbeth criticized existing standards of daylight, in-
standard sources should be changed? If so, which
sources do you recommend? Please give references and cluding Abbot-Gibson daylight7 and Middleton's com-
puted distributions8 for the overcast sky with an urban
reasons for your choice."
terrain suggested by the committee in 1955 as a target
The replies received can be separated into three 9
standard, as well as CIE sources B and C, as being
groups, each of which is about equal in size:
too pink to be acceptable representations of typical
" (i) No comments; leaves it up to the CIE to daylight. These recommended spectral distributions
investigate the problem and come forward with correspond to chromaticities on the pink side of the
recommendations." Planckian locus on the chromaticity diagram. He called
" (ii) No change recommended; emphasizes the attention to the fact that over a period of several
possible confusion a change may bring about and the decades his company has sold thousands of artificial
considerable effect it may have on existing tabulations daylighting units not only with a color temperature
and instrumentation." range, but also with a pink-green range. Many units
" (iii) New sources or changes are desirable; it would yielding colors on the pink side of the Planckian locus,
be desirable to have one or more standard daylight and even those falling very slightly on the green side,
sources which are of more practical use than the have been returned by customers with the complaint
present Source C and thus encourage the use of the that they were too pink to give color renditions char-
standard source in inspection work. A new standard acterizing natural daylight."0
source should provide a high level of illuminance over Examination of a chromaticity diagram supplied by
a large area of the visual field. The ultraviolet content Nickerson comparing the chromaticities corresponding
should be comparable with that of natural daylight to the spectral distributions of sunlight and daylight
both in amount and spectral distribution." by Abbot"' as summarized by Moon,'2 by Taylor and
Kerr,'3 and by Henderson and Hodgkiss supplied in
Nickerson6 summarized the replies to the question- advance of publication'4 was found to support Mac-
naire as follows: "No change should be made in CIE beth's indirect finding that typical daylight is more
standard light sources until a 300-800 nm definition greenish than existing standards of daylight.
for a target standard for daylight can be agreed upon, MacAdam reported that extensive spectral distribu-
and a sufficiently close match to this can be obtained tion data for various phases of daylight had been ob-
in an actual lamp, or in a lamp-filter combination, tained in Rochester in recent years independently in
that can be agreed upon for use as a standard source
for calorimetric and visual use. Further, that any
7 K. S. Gibson, J. Opt. Soc. Am. 30, 88 (1940).
series adopted for light-source standards should include 8 W. E. K. Middleton, J. Opt. Soc. Am. 44, 793 (1954).
one that in color temperature is at least as high as 9Proceedings of the International Commission onl Illumination,
7400'K. This is for use as a daylight substitute in 13th Session, Zfirich, 1955 (Bureau Central de la Commission
Internationale de l'Eclairage, 57, rue Cuvier, Paris 5c, France,
colorimetry and for visual use, to be adopted either 1955), Vol. I, p. 1.3.1 U-9.
in replacement for, or in addition to, the present 10Norman Macbeth and WV.B. Reese, "Some practical notes
source C." on standard illumination practices for color matching in the
U. S. A.-past, present, and future," 7th International Confer-
ence on Color; Florence, Prato, and Padua, Italy; 2-7 May 1963.
'Proceedings of the International Commnission on Illon1ination, 11C. G. Abbot, F. E. Fowle, and L. B. Aldrich, "The distribu-
14th Session, Brussels, 1959 (Bureau Central de la Commission tion of energy in the spectra of the sun and stars," Smithsonian
Internationale de l'Eclairagc, 57, rue Cuvier, Paris, 5', Frame, Miscellaneous Collections 74, No. 7, Publ. No. 2714 (1923).
1960), Vol. A, p. 93. 2 P. Moon, J. Franklin Inst. 230, 583 (1940).
a Proceedings of the International Commnissionon Illnmination, 3
A. H. Taylor and G. P. Kerr, J. Opt. Soc. Am. 31, 3 (1941).
15th Session, V'ienna, 1963 (Bureau Central de la Commission 4 S. T. Henderson and D. Hodgkiss, Brit. J. Appl. Phys. 14,
Internationale de 1'Eclairage) (to be published). 125 (1963).
August1964 SPECTRAL DISTRIBUTION OF TYPICAL DAYLIGHT 1033

two projects of the Eastman Kodak Company, one For n (622, 249, 274, or 99) sample sets of data, the
(Condit)"5 by spectroradiometer, the other (Miller) by response vectors can be arrayed to form a data matrix
interference filters, and that some of the data gathered of n rows and r columns.
by Condit had been subjected to characteristic vector It is possible to find a set of characteristic vectors,
analysis by Simonds."6 Wyszecki said that a project of which, when added in the proper amounts to the mean
the National Research Council was about to get under response vector, will adequately approximate any of the
way to measure the spectral distribution of north sky- original family of n response vectors. Details of the
light in Ottawa simultaneously with visual determina- computational procedure are given by Simonds.6 The
tions of chromaticity as a check. In view of the fact procedure involves the computation of the variance-
that data recently taken, or about to be taken, were covariance matrix S from the original n-by-r data
far more numerous, and sampled much more completely matrix of response vectors. Characteristic vectors of
the range of spectral distributions of daylight than this matrix are obtained corresponding to the latent
previously published data, it was decided that the roots of the determinantal equation
technical colorimetry committees of Canada and the
United States would jointly undertake to evaluate from
IS-LIJ =0, (1)
these recent data (Ottawa, Rochester, and Enfield, where I is an r-by-r unit matrix and L is a diagonal
England), a series of related spectral distributions of matrix (r-by-r) of the latent roots. The characteristic
typical daylight extending over a considerable range vectors, like the response vectors themselves, are sets
of correlated color temperatures. It is the purpose of of r numbers. Mathematically stated, the sample re-
this paper to describe the method of evaluation, to sponses at each value of wavelength are given by
present the series of spectral distributions, and to Eq. (2):
present evidence supporting the view that each in-
dividual spectral distribution of the series is typical E,= E1+MV ,l+M2V2,1
of daylight of a particular correlated color temperature. * -MIV',,
+M3V3,l+.
E2 = E2+M1 V1,2 +M2 V2,2
COLLABORATION OBTAINED
Condit, Henderson, and Budde supplied spectral Er=Er+MlVr+M
+M3 V3,2+*- M Vp,,2,
2 V2,
+M3 V 3,r+ **Mp p,r.
J p<r. (2)

distribution data relative to 560 nm for various samples


of daylight in the form of relative spectral irradiance The M's are the amounts of the characteristic vectors
for each 10-nm interval between 330 and 700 nm in which must be added to the mean response vector in
advance of publication. Condit supplied distributions order to produce the sample response vector. The
for 249 samples of daylight; Henderson, 274; and characteristic vectors Vp are uniquely determined for
Budde, 99. a given family of response curves. The same character-
Simonds supervised derivation of the first four istic vectors apply to all response vectors belonging
characteristic vectors not only for the composite data to the original family from which the vectors are
(622 distributions), but also for the three subgroups derived. Only the values of Ml, M2, - * , Mp vary
.

(Condit, Henderson, Budde) over the same spectral from one response curve to another. The M's, there-
range, and for the Henderson data over the range fore, are a complete specification of the response vector
330 to 780 nm. The program and computer time for to which they apply. Together with the uniquely de-
computing the chromaticity coordinates for each input termined characteristic vectors and the mean response
distribution, for evaluation of the characteristic vec- vector, the M's are sufficient information with which
tors, for obtaining reconstituted distributions from the to reconstruct the entire response vector from which
means and the first two characteristic vectors, and for they were derived.
checking these against the corresponding individual The number p of characteristic vectors required to
input distributions were supplied by the Eastman explain all the differences among a family of curves,
Kodak Company. each represented by r responses, will be equal to, or less
than r. For a perfect fit to all the response vectors, r
CHARACTERISTIC VECTOR METHOD characteristic vectors may be required. The power of
A mathematical statement of the method, adapted the tool, however, comes from the fact that a large
from Simonds,"6 might be given as follows: percentage of the variability among the family of
Response data Ex (spectral irradiance) are available homologous response vectors may be explained by
for r(38) levels of the variable X (wavelength). For using only a few characteristic vectors.
each experimental condition, then, the r values of Ex The p vectors, when graphically presented, are basis
constitute a one-row r-column vector of response data. curves or vectors. The following statements are given
"$H. R. Condit and F. Grum, J. Opt. Soc. Am. 53, 1340 (1963); without proof:
54, 937 (1964).
16 R. H. Morris and J. H. Morrissey, J. Opt. Soc. Am. 44, 530
(1) The p characteristic vectors are orthogonal;
(1954); J. L. Simonds, ibid. 53, 968 (1963). that is, they represent independent types of response
1034 13JUI)D,MACADAM, AND WXYSZECKI Sol. 54r

variability. Mathematically stated, ium oxide screen.'4 Narrow equal slits gave a spectral
r
bandwidth of 1.5 nm. "With the instrument's ac-
E VaiVbi=O, awAb. ceptance angle of about 60, a comparatively small area
was viewed and consequently the signal was sensitive
to changes in this small region of the sky. . . . With
(2) The M's, called scalar multiples, can be de- readings taken every 100 A . . . a run could be com-
termined for each sample response vector as a simple pleted in about 10 min. . . . Attempts to compensate
linear combination of the response data at the r values for the variation of intensity with time by using a
of wavelength. The weighting coefficients are uniquely second monochromator set at a fixed wavelength
determined for a particular set of vectors. throughout the scan gave some improvement but this
(3) The derivation of the vectors ensures that the could not be carried out regularly. It was decided that
first vector accounts for the largest amount of the a program involving many measurements over an ex-
total response variability; the second vector accounts tended period would compensate for the random varia-
for the second largest amount of variability; and so tions during any one run. With this direct-view ar-
forth. rangement it was desirable to make the measurements
EXPERIMENTAL DATA at times of apparently steady sky conditions. On a
good day the variation over a period of 10 min would
The 249 spectral distributions of Rochester daylight be within 3% of the mean value at any given wave-
supplied by Condit were obtained by comparing the length, but often it was worse than this, up to 10%
light reflected from a magnesium oxide test plate variation being common. . . . The majority of the
illuminated by incandescent lamp light of known color observations were made on the laboratory roof and
temperature with that reflected from a barium sulfate the white plate was used to give an average distribu-
test plate tilted 150 off the vertical and illuminated
tion for most of the hemisphere. The measurement
either by light from the sky and light from the sun
procedure was as before, with some reduction in varia-
at various altitudes in the plane perpendicular to the bility during a run on account of the integrating effect
test plate, or simply by light from the sky with the
of the white plate."
sun being back of the plate or obscured by clouds. The 99 spectral distributions of Ottawa daylight
The instrument used was the Beckman DK spectro-
were obtained by Budde by measuring the light
photometer whose slitwidths are automatically adjusted received at wavelength X either from the total sky, or
to maintain the output of the photodetector constant. from the north sky, relative to that at wavelength
The spectral bands transmitted varied from about 1 560 nm. This measurement relative to the irradiance
to 3 nm at half-height. The scanning time was about
at 560 nm eliminated fluctuations of the total irradiance
1 min. The measured spectral distribution curves ex-
during scanning of the spectrum. The apparatus con-
tended from 330 to 700 nm, and were read at somewhat
sisted basically of an integrating sphere, a Hilger
irregular intervals dictated by an attempt to obtain
quartz prism double monochromator (D191) and a
values of relative spectral irradiance at the middle of
1P28 RCA photomultiplier tube. The spectrum was
each major absorption or transmission band detected.
scanned with a bandwidth ranging from 1 nm in the
The input data were derived at intervals of 10 nm by
ultraviolet to approximately 7 nm in the red. The
interpolation of these data. Small corrections were
spectral range covered was from 300 to 720 nm. Simul-
applied for the difference in spectral reflectance be-
taneous measurements of the same daylight by means
tween magnesium oxide and barium sulfate. The meas-
of a Donaldson six-primary colorimeter 17 were made
urements refer to sky conditions not yielding any
by two observers as a closing check on the accuracy
visual evidence of industrial contamination. of the measurements of spectral distribution. The
The 274 spectral distributions of daylight in Enfield, agreement in chromaticity coordinates, calculated from
England, supplied by Henderson, were obtained partly the visual and the spectroradiometric measurements,
by light reflected from a horizontal diffusing white was better than 0.006 in x and y. The data supplied
plate of known spectral reflectance relative to mag-
for analysis were those derived from averaging the
nesium oxide, consisting of mat-finish Vitrolite receiv- original distribution curves over bands of 10 nm width.
ing light from nearly the whole hemisphere including The composite input data consisted of 622 spectral
direct sunlight if it was not obscured by clouds. The distributions Ex (Condit 249, Henderson 274, Budde
remainder of the observations refer to the north sky 99) weighted in inverse proportion to the residual
at 450 with an acceptance angle of about 60. The variances V(Ex /FE6 0o) when the data from each source
Hilger and Watts D290 grating monochromator was were separately analyzed and reconstituted with four
used, and the scan was divided between two photocells
vectors. The variance for each set of data was computed
(Mazda 27M3 from 300 to 580 nm, RCA 1P22 from
in the usual way as the sum of the squares of the
540 to 780 nm). The resulting spectroradiometer was
differences between each input data and the corre-
calibrated by light from an incandescent lamp at a
color temperature of 28540 K reflected from a magnes- 17R. Donaldson, Proc. Phys. Soc. (London) 59, 554 (1947).
August1964 SPECTRAL DISTRIBUTION OF TYPICAL DAYLIGHT 1035

TABLE I. Mean and first four characteristic vectors for


composite data on spectral irradiance of daylight.

Wavelength -o V
1 X
(nm) Mean V1 V2 V3 V4 0'

330 553 420 85 91 12 200- )


340 573 406 78 78 21
350 618 416 67 66 19
360 615 380 53 52 19
370 688 424 61 43 11 00
c-\
0

380 634 385 30 4 -41


390 658 350 12 - 2 -40
400 948 434 -11 -22 -32
410 1048 463 -15 -35 -28
420 1059 439 - 7 -50 -28 o -200 -
430 968 371 -12 -51 -11
440 1139 367 -26 -50 2
450 1256 359 -29 -60 15 300 400 500 600 700
460 1255 326 -28 -63 18 Wavelength,nm
470 1213 279 -26 -65 21 FIG. 1. Vectors V1 (open circles) and V2 (solid circles) derived
480 1213 243 -26 -59 20
from the composite data (622 spectral distributions).
490 1135 201 -18 -58 20
500 1131 162 -15 -47 23
510 1108 132 -13 -45 21 It will also be noted that the values of the first
520 1065 86 -12 -33 22
characteristic vector are greater than zero for wave-
530 1088 61 -10 -20 17
540 1053 42 - 5 -15 13 lengths less than 560 nm, and less than zero for wave-
550 1044 19 - 3 -9 8 lengths greater than 560 nm. The most common varia-
560 1000 0 0 0 0
570 960 -16 2 11 -7 tion from one measured curve of daylight spectral
580 951 -35 5 -10 -5 distribution to another is thus such as to correspond
590 891 -35 21 -3 -7 chiefly to a change in correlated color temperature of
600 905 -58 32 -5 -7
610 903 -72 41 -10 -7 the daylight. This yellow-to-blue variation corresponds
620 884 -86 47 -20 -9 to presence or absence of clouds in the sky, and to the
630 840 -95 51 -22 -8 inclusion or exclusion of direct sunlight.
640 851 -109 67 -36 -5
650 819 -107 73 -48 -3 Note also that the second vector has low values from
660 826 -120 86 -55 -2 380 to 560 nm and higher values in the longwave (red)
670 849 -140 98 -61 2
680 813 -136 102 -65 2 part of the visible spectrum. This second most common
690 719 -120 83 -57 3 variation thus corresponds to a pink-green variation
700 743 -133 96 -64 5 such as might be caused by the presence of little or
much water in the form of vapor and haze.
Figure 1 is a plot of the first and second vectors V
sponding value reconstituted from the mean and the and V derived from the composite data. By adding1
first four characteristic vectors by Eq. (2) divided by various2 scalar multiples M and M of these vectors to
1 2
the number of input data. On this basis, each of the the mean curve, it is evident that variations either in
monitored distributions supplied by Budde received the yellow-blue sense or the pink-green sense may be
full weight, the unmonitored distributions supplied by introduced into the reconstituted curves. Thus, by
Condit and Henderson receiving, as expected, some- adjustment of the scalar multiples M and M , re-
1 2
what lower weights. constituted daylight distribution curves would be
generated to correspond to any chromaticity within
MEANS AND CHARACTERISTIC VECTORS
the daylight range; in particular, values of these scalar
Table I gives, for the composite data, the means and multiples exist which make the chromaticity point cor-
the first four characteristic vectors. The analogous responding to the reconstituted curve fall at any
means and vectors derived from the three subsets of desired point on a locus of typical daylight chromatici-
data (Condit, Henderson, Budde) for the same spectral ties. Any two linear combinations of characteristic
range are available from the authors on request as are vectors would similarly yield an unique distribution
also means and characteristic vectors derived from the curve for each chromaticity point. It was found, how-
Henderson data for the range 330 to 780 nm. ever, that reconstituted curves satisfactorily reproduc-
It will be noted from Table I that, at the normalizing ing the measured curves were generated by using only
wavelength 560 nm, the mean of the spectral distribu- the first two vectors. The remainder of this paper
tions is set at 1000, and the values of the characteristic describes the derivation of such a locus on the chroma-
vectors are all identically zero there. Reconstitutions ticity diagram, the derivation of the corresponding
of spectral distributions of daylight obtained by Eq. 2 scalar multiples for correlated color temperatures 48000,
will thus also have values of 1000 at the normalizing 55000, 65000, 75000, and 10 0000 K, and demonstra-
wavelength 560 nm. tions of the degree to which curves reconstituted in
i(036 JUDIl), MAcAIDA M, ANDIVYSZECKId V 54
Vol.

and as solid circles the chromaticity points computed


from the 99 spectral distributions supplied by Budde.
In general, the plotted points derived from the spectral
distributions agree well with the two lines representing
the visual observations, and they cluster closely about
the line defined by: y=2.870x -3 .00x 2 -0.275, and
shown as squares connected by a solid line on Fig. 2.
The Nayatani-Wyszecki line diverges from the Cham-
berlin-Lawrence-Belbin line for correlated color tem-
peratures greater than 10 0000 K. Since there are too
few chromaticity points, computed from measured
spectral distributions to decide between these two lines
above 40 0000 K, the adopted locus has been drawn
about half-way between the two lines. This adopted
locus below 40 000 0 K is drawn closer to the Nayatani-
NVyszecki locus because the computed chromaticity
points indicate that course. The adopted locus follows
X the Planckian locus (shown as open circles connected
FiG. 2. Chromaticities of daylight compared to the locus of by a solid line on Fig. 2) fairly closely, but stays
chromaticities implied by the Planck radiation law (open circles slightly on the green side. Figure 2 also shows lines of
connected by solid lines). The temperatures of the complete
radiator corresponding to the points at the centers of these open correlated color temperature computed by Kelly20 from
circles are indicated in degrees Kelvin. The straight lines inter- the 1960 CIE-UCS diagram for correlated color tem-
secting this locus at 48000, 55000, 65000, 75000, and 10 0000 K
correspond to lines of constant correlated color temperature peratures of 48000, 5500°, 65000, 75000, and 10 0000 K.
computed by Kelly from the 1960 CIE-UCS diagram. The center Table II gives the chromaticity coordinates at which
of the chromaticity range found for daylight by Nayatani and these lines of correlated color temperature intersect the
Wyszecki is shown by solid circles connected with a solid line;
that found by Chamberlin, Lawrence, and Belbin, by a dotted locus adopted as typical of daylight chromaticities
line. Chromaticity points computed from the measured spectral defined by y =2.870x -3.000x 2 -0.275.
distribution curves are indicated by open circles (Condit), by
crosses (Henderson and Hodgkiss), and by solid circles (Budde).
The locus of chromaticities taken in the present paper to be RELATION BETWEEN SCALAR MULTIPLES AND
tvpical of daylight conforms to the relation: y=2.870x-3.000x 2 CHROMATICITY COORDINATES
-0.275, and is shown by squares connected by a dotted line.
Any spectral distribution curve Ex reconstituted from
accord with these scalar multiples may be considered a mean and two characteristic vectors derived from a
reliable representatives of typical daylight spectral family of measured spectral distributions is computed
distributions. by Eq. (2) as:
E>,=E>+M 1 V 1 ,X+M2 V 2 ,?, (2a)
CHROMATICITIES OF VARIOUS
PHASES OF DAYLIGHT and the tristimulus value X of the irradiance specified
by this distribution may be written:
To derive a curve of typical daylight chromaticities,
use has been made of recent visual measurements by X=E ExtxAX= E(Ex+MlVl,x+M 2 V 2 ,x)zXAX,
Chamberlin, Lawrence, and Belbin15 on the north sky where the weighting function x, is one of the color
in southern England, and by Nayatani and Wyszecki'9 matching functions of the 1931 CIE standard observer
on a near-north sky at Ottawa, Canada, and of the for colorimetry.' From the fact that the scalar multiples
chromaticities computed from each of the 622 spectral M1 and M 2 are constants independent of wavelength
distributions subjected to characteristic vector analysis. for any single reconstituted curve, the expression for
Figure 2 is a central portion of (x,y)-chromaticity
diagram of the 1931 CIE coordinate system. It shows TABLE II. Chromaticity coordinates (xy) of typical daylight
as solid circles connected for various correlated color temperatures.
by a solid line the locus
derived by Nayatani and Wyszecki, as a dotted line
that derived by Chamberlin, Lawrence, and Belbin, as Correlated color
temperature Chromaticity coordinates
open circles the chromaticity points computed from (0 K) x Y
the 249 spectral distributions obtained by Condit, as 4800 0.3518 0.3634
vertical crosses the chromaticity points computed from 5500 0.3324 0.3475
the 274 spectral distributions supplied by Henderson, 6500 0.3127 0.3291
7500 0.2991 0.3150
10 000 0.2787 0.2919
18 G. J. Chamberlin, A. Lawrence, and A. A. Belbin, Light and
Lighting 56, 73 (1963).
1Y Y. Nayatani and G. Wyszecki, J. Opt. Soc. Am. 53, 626
20
(1963). K. L. Kelly, J. Opt. Soc. Am. 53, 999 (1963).
August1964 SPECTRAL DISTRIBUTION OF TYPICAL DAYLIGHT 1037

TABLE III. Scalar multiples of the first two characteristic calculated and found to be as follows: Xo=102434,
vectors of the composite data required to reconstitute spectral Y 0 = 106769, Z 0 = 123630; X 1 = 1866, Y1 = 1914, Zi
distribution curves of typical daylight of five correlated color
temperatures. = 34810; X 2 = 2133, Y 2 = 762, Z 2:= -2355. Substitution
of these tristimulus values into Eqs. 4 and 5 yielded
Correlated color Scalar multiples of Second both the explicit direct connection between the scalar
temperature first characteristic characteristic multiples for the composite data and the chromaticity
(0 K) vector vector
coordinates of the reconstituted spectral distributions
4800 -1.140 0.677 of irradiance, and the reverse connection as follows:
5500 -0.784 -0.195
6500 -0.293 -0.689
7500 0.145 -0.752 0.30776+0.00561M 1 +0.00641M2
10 000 1.005 -0.378 x=
1.00000+0. I1594Mi+0.00162M2
(4a)
0.32079+0.00575M
1 +0.00229M 2
X may be written: y
1.00000+0.1 1594M1+ 0.00162M2
X= XO+MlXl+M2 X 2 , (3)

value of the mean dis-


- 1.3515- 1.7703x+5.9114y
where X 0 is the X-tristimulus
tribution, and X1 and X 2 are the X-tristimulus values 0.0241+0.2562x- 0. 73 4 1y
of the two characteristic vectors; for example, Xi (5a)
= 7 Vlx.xvZAX.By analogous argument for Y and Z 0.0300- 31.4424x+30.0717y
we may write: 0.0241+0.2562x-0.7341y
Y= YO+MlYl+M2 Y2, (3)
Z= ZO+MlZl+M2 Z 2. SPECTRAL DISTRIBUTIONS OF DAYLIGHT FOR
VARIOUS CORRELATED COLOR
Note that values of Xo, Yo, Zo are nowhere less than TEMPERATURES
zero, but the tristimulus values of the characteristic To derive a series of related spectral distributions of
vectors may be less than zero because the vectors typical daylight over a considerable range of correlated
themselves (see Table I and Fig. 1) are adjusted to color temperatures from the composite data, values of
zero at 560 nm with both positive and negative values scalar multiples M1 and M2 were computed by in-
for other parts of the spectrum. serting into Eq. (5a) the values of chromaticity co-
The connection between scalar multiples M1 and M2 ordinates x and y given in Table II for typical daylight
and chromaticity coordinates x, y follows directly from chromaticities of correlated color temperatures 48000,
Eq. 3 and from definitions: x=X (X+Y+Z), y
55000, 6500°, 75000, and 10 000)K. Table III shows
= Yj (X+ Y+Z). A convenient form of the connec-
tion, for X+ Y+Z abbreviated as S, is: TABLE IV. Extension of the mean and the first two character-
istic vectors of the composite data given in Table I to the spectral
X 0 /S 0 +MlXl 1/S 0+M2 X2/So ranges 300 to 330 nm and 700 to 830 nm from Moon's compilation
of the spectral absorptance of the earth's atmosphere due to
1+M 1S 1/S 0+M2 S 2 /So ozone and water vapor.
(4)
1 Y
Y 0 /S 0 +MA 1yS 0 +M 21y 2 /So Wavelength
(nm) Mean VI V2
I 1+M 1 S 1 /So+M 2 S 2 /So
300 0.4 0.2 0.0
310 60 45 20
To find the scalar multiples M1 and M2 required to 320 296 224 40
yield a reconstituted spectral distribution of irradiance
having any arbitrarily chosen values x and y of chroma- 330 553 420 85
(See Table I)
ticity coordinates, it is necessary to solve for M1 and 700 743 -133 96
M 2 from Eq. 4. A convenient form of solution is:
710 764 -129 85
720 633 -106 70
XoY 2 -X 2Yo+ (YOS2 - Y 2So)x+ (X2 S0 -XoS 2)y 730 717 -116 76
740 770 -122 80
X2 Y1 -X 1 2 (Y 2 8 1- YlS 2 )x+ (X 1 S 2 -X 2Sl)y 750 652 -102 67
760 477 -78 52
(5) 770 686 -112 74
780 650 -104 68
X1 Yo-XoY 1 + (Y1 SO- YoSi)x+ (XOS1-XiSo)y 790 660 -106 70
M2= 800 610 -97 64
X 2 Y 1 -X 1 Y 2+ (Y2 S 1- YlS 2 )x+ (XlS 2 -X2SS)y 810 533 -83 55
820 589 -93 61
The tristimulus values of the mean of the composite 830 619 -98 65
data and of their first two composite vectors have been
1038 1JUDD, M1 cADAMI, AND \WYSZECKI V
Vol. 54

TABLEV. Relative spectral irradiance of typical daylight istic vectors of the composite data given in Table I in
reconstituted from mean and characteristic vectors of the com-
posite data (Tables I and IV) by the scalar multiples of Table III. accord with Eq. (2a). To extend these spectral dis-
tribution curves to cover the wavelength ranges 300 to
Wavelength Correlated color temperature ( 0 K) 330, and 700 to 830 nm, resort was had to Moon's"2
(nm) 4800 5500 6500 7500 10 000 compilation of data on the spectral absorptance of the
300 0.2 0.2 0.3 0.4 0.6 earth's atmosphere due to ozone and water vapor.
310 23 21 33 52 97 Extensions of the mean and first two characteristic
320 68 112 202 298 506
330 132 207 371 550 943
vectors of the composite data in this way are shown
340 163 240 400 573 952 in Table IV. Table V shows the reconstituted spectral
distributions of typical daylight over the extended
350 190 279 450 627 1011
360 218 307 467 630 977 range 300 to 830 nm, and Fig. 3 shows a plot of these
370 246 344 522 703 1091 distributions.
380 215 326 500 668 1010 The spectral distributions of typical daylight were
390 267 382 547 700 1006
also evaluated in an analogous way from the means and
400 446 610 828 1019 1388 first two characteristic vectors derived from the four
410 516 686 916 1119 1515
420 554 716 935 1128 1503
subsets of data (Condit 249, Henderson 274, from 330
430 537 679 868 1033 1346 to 700 nm; Henderson 274, from 330 to 780 nm; and
440 704 856 1049 1211 1518 Budde 99) as a check on the distributions derived
450 827 981 1171 1330 1628
from the composite data comprising all 622 of the
460 864 1004 1178 1323 1594 measured distributions. The agreement within the
470 878 999 1149 1272 1503 spectral range 400 to 700 nm was found to be very
480 916 1026 1159 1269 1469
490 894 980 1088 1177 1344 satisfactory, but the amount of the ultraviolet com-
ponent (330 to 390 nm) indicated by the four subsets
500 936 1007 1094 1165 1300
510 949 1008 1078 1137 1246
of data showed significantly different dependences on
520 959 1000 1049 1086 1156 correlated color temperature. One might be tempted to
530 1011 1042 1077 1105 1153 ascribe these indicated different dependences of ultra-
540 1002 1021 1044 1063 1097 violet content on correlated color temperature to real
550 1020 1030 1040 1049 1064 differences between the atmospheric conditions at
560 1000 1000 1000 1000 1000 Rochester, Ottawa, and Enfield, were it not for the
570 979 973 964 956 943
580 995 977 957 942 914 fact that the two Henderson subsets of data (330 to
590 945 914 886 870 848 700 nm and 330 to 780 nm, identical from 330 to 700 nm)
600 993 944 900
also showed this different dependence. It is our view
873 835
610 1012 951 896 862 816 that ultraviolet content is poorly correlated with cor-
620 1014 942 876 836 780 related color temperature, and that the amount of
630 983 904 833 787 726
640 1020 923 837 785 716
ultraviolet indicated by the composite data should be
taken as typical.
650 990 889 800 748 683 In the opinion of the authors, the spectral distribu-
660 1021 903 802 745 673
670 1075 940 822 755 671 tions of irradiance produced by daylight at the earth's
680 1037 900 783 717 638 surface, shown in Fig. 3, are the most typical that can
690 912 797 697 640 567
700 960 829 716 652 573
710 969 849 743 681 602
720 801 702 616 565 500
730 901 793 699 643 572
740 963 850 751 692 617
750 814 719 636 587 524
760 601 528 464 427 379
770 864 759 668 614 545
780 815 718 634 584 520 C,
790 828 729 643 592 527
0)
800 764 674 594 548 488
810 665 587 519 480 429
820 736 650 574 530 472
830 775 683 603 556 496
, . .
. .
.

I0a 400 500 600 700 800


the values of the scalar multiples so obtained. Recon- Wavelength,nm
stituted spectral distributions were then obtained for FIG,.3. Spectral distribution of typical daylight for correlated
color temperatures: 48000, 55000, 65000, 75000, and 10 000°K
the wavelength range 330 to 700 nm by applying these reconstituted from the mean and the first two characteristic
scalar multiples to the mean and first two character- vectors of the composite data (622 measured distributions).
August1964 SPECTRAL DISTRIBUTION OF TYPICAL DAYLIGHT 1039

TABLEVI. Chromaticity coordinates (xy) corresponding to the


measured spectral distributions shown in Figs. 4-6.

Correlated color
temperature Chromaticity coordinates
(0 K) Spectral corresponding to the
(and Fig. number) distribution distribution
o I l 1V-~0,.X01 -0.8001

x y
5500 From Table V 0.3324 0.3475
(Fig. 4) Condit 51 0.3318 0.3465
Henderson 107 0.3324 0.3473
Budde 84 0.3256 0.3334 C U20 D
6500 From Table V 0.3127 0.3291
(Fig. 5) Condit 120 0.3124 0.3285
Henderson 67 0.3116 0.3280
Budde 12 0.3134 0.3295
7500 From Table V 0.2991 0.3150
400 500 I.000 170 00 0C00 600 700
(Fig. 6) Condit 241 0.2982 0.3157 Wavelength,
om
Henderson 247 0.2990 0.3146
Budde 44 0.2968 0.3100 FIG. 4. Comparison of measured spectral distributions of day-
light, whose chromaticities are near0 to that of typical daylight of
correlated color temperature 5500 K, with distributions recon-
stituted from the mean and the first two characteristic vectors
be derived from the experimental data gathered to derived from the composite data (622 measured distributions).
date. Although the initial experimental data employed The measured distribution chosen from each subset of data is
that whose chromaticity point is nearest on the 1960 CIE-UCS
in this analysis refer to different spectral bandwidths diagram to that for typical daylight of correlated color tempera-
of 10 nm and less, the spectral distributions proposed ture 5500'K; see Table VI. Quadrants A, B, and C compare
here should be taken as the average of the true values measured distributions (A-Condit 51; B-Henderson 107; C-
Budde 84) with the reconstitution giving the best least-squares
of spectral irradiance over wavelength intervals of 10 fit. The values of the scalar multiples (ll and M 2) so found are
nm throughout the spectrum relative to that for the indicated together with the variance of the measured distribution
interval 555 to 565 nm with its central wavelength at from the reconstitution. Quadrant D compares all three of these
measured distributions with the reconstituted distribution shown
560 nm. They are recommended as guides in the de- in Fig. 3 for 55000 K.
velopment of sources of artificial daylight that might
be proposed as standard sources for colorimetry; in corresponding to the nine measured curves selected.
particular, the spectral distributions found for corre- Figures 4-6 compare the measured relative spectral
lated color temperatures 55000, 65000, and 7500'K irradiances with the corresponding curves reconstituted
are proposed for the role of target curves to CIE from the mean and first two characteristic vectors by
Committee E-1.3.1 which has already indicated' its applying the scalar multiples. These scalar multiples
intention to develop standard sources of these correlated give the best least-squares fit and are shown on the
color temperatures. Should phases of typical daylight plots. Figures 4-6 also compare (lower right quadrant)
of other correlated color temperatures between 40000 these measured distributions of spectral irradiance with
and 40 000 0 K be desired, derivation of other curves of the reconstituted curves of Fig. 3 and Table V for
this family can be accomplished by means of Eq. 5a by correlated color temperatures 55000, 65000, and 75000K.
substitution of other chromaticity coordinates satisfy- The agreement is seen to be satisfactory. There is no
ing the relation: y=2.870x-3.OOx 2-0.275. Reliable indication that spectral distributions reconstituted from
curves of relative spectral irradiance of the pinker or the mean and the first two characteristic vectors of the
greener phases of daylight may also be derived by
means of Eq. 5a provided that, for 0.25<x<0.38, the
value of the y-coordinate departs from the typical
value by not more than 0.008. This range in y-coor-
dinates includes about 90% of the 622 computed
chromaticities.
COMPARISONS OF A SAMPLING OF THE MEASURED
SPECTRAL DISTRIBUTIONS WITH RECONSTITU-
TIONS FROM THE MEAN AND FIRST TWO
CHARACTERISTIC VECTORS OF THE
COMPOSITE DATA
Measured curves were selected from each subset of
data by taking those whose corresponding chromaticity
points on the 1960 CIE-UCS diagram are closest to
the chromaticity points given in Table II for correlated Wavelength,nm

color temperature 55000, 65000, and 7500'K. Table VI FIG. 5. Same as Fig. 4 except that the correlated color temperature
compares these chromaticity coordinates with those is 6500'K. A-Condit 120; B-Henderson 67; C-Budde 12.
1040 1JUDD, MACADAM, AND WYSZECEId V
Vol. 54

274 Henderson curves, 113; so of the 622 curves


comprising the composite data, 424 agree with the
reconstituted curves more closely than the agreement
R shown on Fig. 6 for Condit curve 241. The finding that
the Henderson measured curves agree less well with
.9 their reconstitutions than do the curves measured by
I
Budde and Condit is to be expected from the longer
time taken by Henderson to scan the spectrum; but
this finding should not be construed to mean that
averages of many Henderson curves corresponding to
nearly the same chromaticity are in error. The assump-
tion by Henderson and Hodgkiss'4 "that a programme
involving many measurements over an extended period
Wovelengfh, nm would compensate for the random variations during
FIG. 6. Same as Fig. 4 except that the correlated color temperature
any one run" seems to have been amply justified by
is 7500'K. A-Condit 241; B-Henderson 247; C-Budde 44. the good agreement already noted between curves re-
constituted from the Henderson subsets of data and
the Condit and Budde subsets.
composite data fail importantly to agree ivith measured
spectral distributions corresponding to t'he same chro- COMPARISON WITH SPECTRAL DISTRIBUTION
maticity. The variance V(Ex/E56o) for each curve is CURVE DERIVED FOR THE BRITISH
shown on the plot. The average variancce for all nine STANDARDS INSTITUTION FOR
curves is only 0.0011 which corresponds to a standard COLOR MATCHING AND
APPRAISAL
deviation of about 3 to 4 on the scale of 100 at the
normalizing wavelength 560 mu. These curves show A Draft British Standard Specification circulated
somewhat better than average agreement .The average 13 June 1963 for consideration as a revision of British
variance of the 99 Budde curves is 0.001 8; for the 249 Standard 950:1941 for artificial daylight for color
Condit curves, 0.0020; and for the 2'74 Henderson matching and color appraisal includes a definition of
curves, 0.0053. Perhaps a more striking indication of the "spectral power distribution of artificial daylight."
the degree to which the individual me;asured curves This distribution was derived by taking an average of
agree with the curves reconstituted from the mean and a group of spectral distributions measured by Hender-
4
the first two characteristic vectors is to count the son and Hodgkiss.1 These measured spectral distribu-
number of curves having variances no greater than tions all had chromaticities in the neighborhood of CIE
that (V=0.0028) for Condit curve 241 ';hown on Fig. source C but on the green side of the Planckian locus.
6. Of the 99 Budde curves, 89 have varian Lcesno greater The average curve corresponds to the chromaticity
than 0.0028; of the 249 Condit curves, 2.U2;and of the coordinates: x=0.314, y=0.329. Note that these chro-
maticity coordinates agree well with those derived by
us (x=0.313, y=329; see Table VI) to characterize
I I typical daylight of correlated color temperature 6500'K.
1.50
Figure 7 compares our recommendation (solid curve)
a for this correlated color temperature with the spectral
'a distribution (dotted curve) derived for the British
1.00
-

A Standards Institution and with the spectral distribu-


-a I tion (shown by open circles) derived from the mean
Co
, .50
and first two characteristic vectors of the Henderson
2~
330-780 nm data. The agreement shown is very en-
<,,.5
.- couraging. It demonstrates that, for the spectral range
330 to 700 nm, average daylight agrees well with
) I eI IA 80 British daylight of the same correlated color tempera-
300 400 500 600 700
- .

Wavelength,nm ture. The discrepancies between 700 and 780 nm may


1IG. 7. Comparison of three spectral dlistribut
be at least partially ascribed to the smaller bandwidth
represent typical davlight at a correlated color temperatureof (1.5 compared to 10 nr) used by Henderson and
6500'K. 'the distribution reconstituted from th(e mean and first Hodgkiss, and to the presence of more water vapor in
twvo cbiaracteristic vectors of the composite dat
solid line; this distribution is also shown in Fig.
a is dishtrabuathe Enfield atmosphere than in Rochester and Ottawa
tion reconstituted from the Henderson data for* 330 to 780 nm atmospheres. The good agreement affords a basis for
is shown by open circles, and the distribution derived for the the hope that a single set of related spectral distribu-
British Standards Institution by taking an avers
of measured spectral distributions from the sair e dfataiselshowin tions, such as those presented in this paper, may be
by a dotted line. found suitable for international use.

You might also like