SolarEnergyVol. 28, No. 4, pp. 293--302,1982 003g-092X/$2/040293.-10503.

00/0
Printed in Great Bri~in, © 19~2PergamonPress Ltd.

ESTIMATION OF THE DIFFUSE RADIATION FRACTION

FOR HOURLY, DAILY AND MONTHLY-AVERAGE
GLOBAL RADIATION

D. G. ERAS,S. A. KLEINand J. A. DUFFLE

Solar Energy Laboratory,Universityof Wisconsin, Madison, WI 53706,U.S.A.

(Received 1 May 1981; accepted 18 August 1981)

Abstract--Hourlypyrheliometerand pyranometerdata from four U.S. locationsare used to establish a relationship

between the hourly diffuse fraction and the hourly clearness index kr. This relationship is compared to the
relationship established by Orgllland Hollands and to a set of data from Highett, Australia, and agreementis within
a few percent in both cases. The transient simulation program TRNSYS is used to calculate the annual
performance of solar energy systems using several correlations. For the systems investigated, the effect of
simulating the random distribution of the hourly diffuse fraction is negligible.A seasonallydependent daily diffuse
correlation is developed from the data, and this daily relationship is used to derive a correlation for the
monthly-averagediffuse fraction.

1. INTRODUCTION which includes the effect of multiple reflections between

Solar radiation incident on buildings or collection sur- the ground and sky.
faces must be known in order to perform thermal Within each group of existing correlations (i.e. the
analyses. In general, only measurements of the total hourly, daily, and monthly-average relationships) there is
horizontal (global) radiation are available. As most sur- considerable disagreement. The significance of this dis-
faces of interest are inclined, it is necessary to estimate agreement depends upon what the diffuse correlation is
the radiation on a tilted surface from measurements of used for. If the annual total radiation on a tilted surface
global radiation. Estimation procedures usually require is estimated using each of the existing relationships, the
the beam and diffuse components of global radiation. results will generally be within a few percent. If a
The beam and diffuse components of global radiation computer simulation is used to estimate the annual per-
can be estimated from empirical relationships. Existing formance of a system with concentrating collectors, the
relationships correlate the fraction of the global radiation results obtained using different correlations may vary by
which is beam or diffuse to an index of atmospheric more than 10 per cent. In addition, there are incon-
clarity. Correlations of this type have been developed for sistencies between daily and monthly correlations. These
use with hourly, daily, and monthly-average values of discrepancies may be the result of variations in in-
global radiation. strumentation and measurement techniques, different
The hourly correlations of Boes[1], Orgill and methods of correlating the data, locational dependence
Hollands [2], and Bruno [3] can be expressed as relation- of the data, or insufficient data.
ships of ISL the ratio of the hourly diffuse radiation to The objectives of this study are to develop, from a
the hourly global radiation, to kT, the ratio of the hourly new data base, relationships for estimating the diffuse
global radiation to the hourly extraterrestrial radiation, fraction of hourly, daily, and monthiy-average global
IIIo. Correlations of the hourly diffuse fraction, Ia]L to radiation, to determine the degree to which the relation-
kc, the ratio of the hourly global radiation to an estimate ships developed are dependent on season and location,
of hourly "clear sky" radiation, IIIc, have been and to compare these relationships to the existing rela-
developed by Bugler[4] and by Stauter and Klein[5], tionships.
each with different definitions of "clear sky" radiation.
The statistical algorithm developed by Randall and 2. THE DATA BASE
Whitson[6] cannot be expressed analytically; this al- The data used to develop the correlations presented
gorithm was used[7] to estimate beam radiation for the here are thought to be among the best data available at
SOLMET data base. this time. These data were recorded at the four U.S.
Relationships for estimating the beam and diffuse cities listed in Table 1. The data for Livermore were
components of daily global radiation have also been recorded by Sandia; those for Raleigh by the Environ-
developed by numerous authors. The correlations of Liu mental Protection Agency; and the data from Fort Hood
and Jordan[8], Choudhury[9], Stanhill[10], Tuller [ l l ], and Maynard were recorded by the Army Atmospheric
Ruth and Chant[12], and Collares-Pereira and Rabl[13] Sciences Laboratory. The Aerospace Corporation edited
all relate HSH, the daily diffuse fraction, to Kr, the ratio the raw data, reduced them to the International Pyr-
of daily global to daily extra-terrestrial radiation, H/Ho. heliometric Scale, and placed them on magnetic tape.
On a monthly-average basis, relationships between Information included for each hour are direct normal
HaJH and /(T have been developed by Liu and radiation, total radiation, mean solar altitude, declination,
Jordan[8], Page[14], Tuller [l l ], Collares-Pereira and date, and extraterrestrial radiation. The direct normal
Rabl [13], and Iqbal [ 15]. Hay [ 16] developed a correlation radiation was measured with a pyrheliometer and the

293
294 D. G. Exns et al.

Table 1. Cities and duration of records for aerospace data base

Fort Hood, Livermore, Raleigh, Maynard, Albuquerque,
Station Name TX CA NC MA NM

Station Number 03902 32899 32900 -0042 23050

Latitude (Deg. N~ 31.08 37.70 35.87 42.42 35.05

Longitude (Deg. N) 97.85 121.70 78.78 71.48 106.62

Altitude (ft) 1080 486 441 203 5314

Data Period

Begin
119174 118174 20/3/75 111175 1/1/81

End
(Day/l~/Yr) 3016178 30110175 114176 31112/76 31/12/84

total radiation was measured with a pyranometer. It kT of 0.025. The resulting correlation is shown with the
should be noted that while the use of a pyrheliometer average data in Fig. 1. It can be represented by:
eliminates the needs for shade ring corrections, there can
be problems with tracking systems and calibration. IJI = 1.0 - 0.09 kr for kr <-0.22
To test the applicability of the correlation at locations la]l = 0.9511 - 0.1604 kr + 4.388 k 2 - 16.638 k3r
other than the four from which it was derived, a set of + 12.336 k~-for 0.22 < kr <-0.80
data recorded in Highett, Victoria, Australia (latitude Ia/I = 0.165 for kr > 0.80. (1)
38'S) during the years 1966--69 was obtained[17]. These
data, measured using an unshaded pyranometer and a For values of kT greater than 0.8, eqn (1) was not fit to
pyranometer with a shade ring, are also thought to be of the data. Following the procedure of Orgill and
high quality. Hollands[2], a constant value of IJI was chosen for kT
In addition, hourly data for a month (Feb. 1980) in this range. Orgill and Hollands attribute the observed
recorded in Albany, NY were obtained from the increase in the diffuse fraction as kr increases from 0.8
Atmospheric Sciences Research Center of the State to beam radiation being reflected from clouds and
University of New York[18]. These data are numerical recorded as diffuse radiation during periods when the
integrations of minute values. Included in this data set sun is unobscured by the surrounding clouds. The data
are total insolation on a horizontal surface, direct normal in this region of kT represent only 0.2 per cent of the
insolation, diffuse insolation using a shadow band, points in the combined data set, and they are not under-
diffuse insolation using an occulting disk, and the number stood well enough to justify fitting a curve to them.
of minutes of beam radiation during each hour. The The mean bias error (d) and the standard deviation (~r)
presence of three independent measurements of the are used to indicate how closely the hourly correlation
diffuse radiation made it possible to monitor the internal agrees with the data. The mean bias error is the weighted
consistency of the diffuse data. difference between the diffuse fractions estimated from a
All of the data were checked for the following in- correlation and the measured diffuse fractions.
consistencies: zero global radiation after sunrise and
before sunset, beam radiation exceeding global, global N
radiation exceeding extraterrestrial, and no beam radia- ((I,d/) - (I,,..,//))1
tion when kr is large. Data exhibiting any of these d= N (2)
problems (less than 1 per cent of the total number of 5"./
hours) were deleted from the data set.
1.2 L, , . a ' ,' I ' , ' I ' , ' I ', ' i ' ,~e' , ' I ' , 'qrrTmT'~
All Locations
3. ESTIMATING T I ~ DIFFUSE FRACTION OF HOURLY GLOBAL
RADIATION

(a) Correlating the diffuse fraction with kr 65 Months Data

The diffuse fraction of the hourly total radiation is
strongly correlated with kr [8,2,3]. The parameter kr is
an indicator of the relative clearness of the atmosphere.
In general, when the atmosphere is clearer, a smaller
fraction of the radiation is scattered. A relationship was
developed between Idl and kr using the combined data
I
-0.[j
i,i,l,I,t,t,i,I,L,l,i,l,t,L,i,l~L,ILtJ
.I .2 .3 .4 -5 .6
"~N

.7
XXX

.8
×

.8
1
--

1 .0
of the four U.S. locations. The individual kr and Ia/l kr
values were weighted with the total radiation for the Fig. 1. Hourly correlation between I~IIand kr compared to average
hour, and average values calculated for each interval in hourly U.S. data.
 Estimation of the diffuse radiation fraction for global radiation 295
In eqn (2), Id.,, is the measured diffuse radiation, I is the 1'2-'''1'''1'~'1'~'1'~'1'. I'''1'''1'''1'

measured total radiation, and la.c is the diffuse radiation . Fort Hood, TX

estimated from the correlation. The standard deviation is :-. . . . . . ~ - -. Lot. • 31.08

an indication of how much the measured hourly diffuse

fractions vary from the correlation. The standard devia-
tion is defined by:
22 Mont "\~
~'U., Std. Oev.= 0 A 4 3 ~ ~ ~_

o"= ,/z (3) .0.0 I ' l ' i ' I ' ~ , I , I ,-L-~I , i , l , I , , I , I , I . I ,~

•1 .2 .3 .4 .5 .6 .? .8 .9 1.0
kT
In general, 95 per cent of the data lie within plus and
'7~T~ r~q-'rT'T~--~TTTTT' I i , I q~'T-r-r~FrlT'r~
minus 2~r of the correlation.
,,' ~ Moynord. MA
The mean bias error and standard deviation were 1,0
~ : , Lot.- 42.42
calculated for the data from each of the four U.S.
locations. Figure 2 shows the hourly correlation (solid .8

line), average values of the data for each interval in kr of .6

-\ \
0.0125 (x's), and plus and minus one standard deviation \a \
24 Months ".
of the hourly diffuse fractions from the correlation S'd' °t~'=°"~ '7 "~.~,
(dashed lines). The mean bias error exhibits only a slight -o.ooe -\ .~,~;~
.2
Iocational dependence, and except for Livermore, the
average data lie very close to the correlation for all ,t] ,, ~ ~ L ~ .~_~_1~ LJ_~ I=d~_L~J~L. 4-~J_A_;=&_~L
• [J .1 .:: .3 .4 ,S .6 .'? ,8 .9 .0
values of kr less than 0.8. The standard deviation is also kT
relatively independent of location. However, the lane
size of the standard deviation indicates that there may be I , 2 :q r~T'~'q~q- , [ , I ' [ ' "I ' I ' I ' I ~ ' q ' ~ 7 - r T ~ ' f ~
: ,~ R o l e gh N C
considerable error (roughly plus or minus two standard ! .0 ~ : ~ . 7 , ~ , Lot.=35.77
deviations) in estimating the diffuse fraction for any
particular hour.
A comparison between the Highett data and eqn (I) is
•6 \ k
shown in Fig. 3. For values of kr between 0.2 and 0.8
I0 Months ' "'
(which include 90 per cent of the data points), the _~., S,d O_ev..O,,2 "'.~i'
average data for kr intervals of 0.0125 (x's) are within 3
per cent of the correlation. The diffuse fraction of the
Highett data approaches 0.85 as kr approaches 0, which . . ~ . I , , , I , ~I._~.4_~L~_J_ l , I , L, l , I ,~L , I . J
does not appear reasonable. A similar trend in the data "0,0 .I ,2 ,3 ,4 ,S ,6 ,7 .8 ,9 ! -0
kT
from Raleigh, NC was traced to erroneous pyrheliometer
readings. The nature of the ttighett diffuse radiation
1.2 i.l.,,l,i,l.i,l.,.l, , i,,,i, i l,i,l, ,
measurements (shaded pyranometer) makes it impossible
I'L Livermore, CA
to determine whether erroneous data are present. If the l.O ~ ~ , LOt.- 37.68
diffuse fraction approached 1.0 as expected, the
agreement between the Highett data and this correlation
would be excellent.
Agreement between eqn (1) and the ttighett data sup-
II M o n t h s D a t a ~ .~ ~ h~---
ports the contention that the correlation is location- __'U.4 Sial. D e v . - O . 1 0 9 L,
~/,('Yx'~ \ , , •I t~"
independent. Further support can be drawn from a vo .=vv.v.3,,e
.3"eo % "~ ,..3(x)~
comparison between eqn (1) and the relationship ~ ,~ -I

developed by Orgill and Hollands. The two correlations, I ,I ,I , I *1 , J , I , l . l . I ,I,1.1, [,i , I,lll, l ,1,4
"0.0 .l .2 .3 .4 .5 .6 .7 .8 .9 1 .0
compared in Fig. 4, are within 4 per cent of each other kT
for all values of kr and within 2 per cent for values of kr
Fig. 2. Comparisons of the U.S. data with the modified kr
greater than 0.5. The nominal accuracy of the instru- correlations.
ments is 5 per cent; thus, the agreement between the data
recorded at Highett, Australia and Toronto, Canada and
eqn (I) is within the uncertainty of the measurements. winter and underpredict in the spring and summer. Due
The large standard deviation of the hourly data from to the large variation in the mean bias error for each of
eqn (1) is due, in part, to a seasonal variation in the the seasons, only the overprediction in fall and winter is
average diffuse fraction. The U.S. data were grouped statistically significant.
into four seasons, and the mean bias error and the
standard deviation were found for each of the four (b) Correlating the diffuse fraction with k~
locations; these results appear in Table 2. The correlation The large standard deviation of the hourly diffuse data
tends to overpredict the diffuse radiation in the fall and from the correlation between Ia]I and kr led to an
296 D.G. ERBSet al.
i.~,~.,., ,.,.,.,. .,.,.,.,.,.,.,.,.,.,-~
~ "~ " . ~ ~ ' ~ Highett, Victoria 4
i.0~. ~ -",, Lat=58.O0 S. l.O .... = ~ r e s e n t Correlation

-- - 6 ~ .6
\ -:- Sld. Dev. : 0.133 \ x~ ,
-~. ~:_- c~:o.o~ \ %', --'~.4
• Z~ ×\
•2 ......

• .0 .I .2 .3 .4 -5 ,6 .7 -8 .9 : .2 ,0.0 l l J ' l ' l ' J ' l ' q l l ' i ' l ' ' ' l ~ ' ' l ' l ' l ' ' l l ~ l A
kT •1 .2 -3 .4 k r .5 ,6 .7 .8 .9 1.0
Fig. 3. A comparisonof the hourly Highett data with the present
correlation between I,/I and kr. Fig.4. A comparisonof the present kr correlationand the Orgilland
Hollands correlation.

investigation of correlating IdI to ko the ratio of hourly cloud cover. The data were separated into three bins
global to hourly "clear sky" radiation, to determine if using the hourly per cent possible sunshine. The first bin,
including air mass, season, and receiver altitude as 0-20 per cent, corresponds to mostly cloudy conditions.
parameters could reduce this standard deviation. The The second, 21--80per cent, corresponds to partly cloudy
beam component of the "clear sky" radiation was skies, while the third, 81-100 per cent, is the clear sky
obtained from a model developed by Hottel[19], while region. A curve relating the diffuse fraction to kT was
the diffuse "clear sky" radiation was calculated using a then constructed for each per cent possible sunshine bin.
correlation of Liu and Jordan[8]. The three curves are shown in Fig. 5 along with eqn
The standard deviations of the data from the cor- (1). Despite the large uncertainty associated with the
relation were only slightly smaller when k~ was used in curves as a result of the small data base used, the
place of kr. There was no change in the mean bias errors dependence of IdI on per cent possible sunshine is very
on an annual basis, although the seasonal dependence of evident. At a value of kr of 0.5, the difference between
the diffuse fraction was reduced somewhat when the the average diffuse fractions of the first and third bins is
data were correlated with kc instead of kr. The use of kc 35 per cent. A correlation which includes the hourly per
as the independent variable did not reduce the un- cent possible sunshine as a parameter could significantly
certainty of the estimated hourly diffuse fraction reduce the standard deviation of the hourly diffuse frac-
sufficiently to warrant the extra calculations required, tions from the correlation.
when compared to the use of kr.
(d) The dependence o[ simulation results on hourly cor-
(c) The use o / p e r cent possible sunshine data in a relations
correlation The disagreements that exist among the correlations
The diffuse radiation fraction for an hour is strongly available for estimating the I d l lead to different esti-
dependent on the type and distribution of clouds in the mates of the radiation incident on an inclined surface. It
sky during the hour. Neither kr nor kc is a function of is of interest to know how large of an effect the choice of
the per cent possible sunshine. Intermediate values of kr diffuse fraction correlation will have on calculated solar
(or kc) can be the result of a thin, continuous cloud cover system performance. The correlations chosen for a
or a heavy, intermittent cloud cover. For a constant comparison of simulated system performance are: (1)
value of kr, thin continuous clouds will result in a higher The Liu and Jordan daily Kr correlation (often used on
diffuse fraction than will heavy intermittent clouds. an hourly basis); (2) eqn (1); (3) A statistical correlation
One month of data from Albany, NY were used to based on eqn (1); and (4) The Aerospace model of Ran-
develop a correlation which demonstrates the depen- dall and Whitson.
dence of the hourly diffuse fraction on the nature of the The statistical correlation adds a diffuse fraction devi-

Table 2. Seasonal bias errors and standard deviations

Season Winter Sprin 8 Summer Fall

Location

F o r t Hood, .092 -.035 -.040 .042

TX a .172 .140 .154 .140

Livermore, .o51 -.019 .048 .045

CA a .113 .116 .113 .105

Raleigh, .048 -.013 .000 .032

NC c .130 .102 .099 .130

Maynard, .058 -.031 -.027 .055

MA o .141 .131 .142 .142
 Estimation of the diffuse radiation fraction for global radiation 297

TRNSYS[21] was used to model the performance of

1.0 -- -~,-',~¢=-~-~L~ J , i Albony, NY
F:ebruary Ooto two different types of systems for a wide range of
system parameter values. The difference between the
O.e PercentPossible [ ~ ' ~ L l e annual solar fractions obtained using eqn (1) and those
Hourly ?un=shlne i I ~
obtained using the other three correlations were cal-
0.6 A =0-20% x\ T - ~ L/--Hourly Correlation
O=2t-80% \1 i ~ ( E q u o t i o n (2.1l)) culated from the simulation results and used as the
-~0 O "8' - '00"/. "1~/i ~,E L responses in a factorial design[22] for each system.
0.4
The solar fraction, 3;, is defined here as one minus the
fraction of the load met with auxiliary energy. A factorial
0,2 ....
design was used to determine the effect of changing
certain system parameter values on the differences be-
0.00~0 I ' 1 ' 1 'l'l'l'l'l'lll~l'l'l'l'l'lll'l'lJ
0.2 0.4 ~ 046 ON8 ~.0 tween the annual solar fractions. Several system
parameters were designated design variables, and a low
Fig. 5. The dependence of the hourly diffuse fraction on hourly
fraction of possible sunshine. and high level chosen for each. All possible combinations
of the design variables were simulated. The difference in
a response resulting from changing a design variable
ation to the diffuse fraction estimated from eqn (1). For from the low to the high level, referred to as a main
the first hour in each day, this deviation is random and effect, was found for each design variable and response.
normally distributed, with the variance a function of kT. The effects of varying more than one design variable at a
For all subsequent hours in the day, the diffuse fraction time, referred to as interactions, were also calculated, as
deviation is found using the AR(I) model[20]: were the average values of the responses for each of the
systems.
d~=p,dj_,+ay (4) The first series of simulations investigated five vari-
ables of a house heating system with a flat-plate collector.
where dj_,is the deviation for the preceding hour. The The five design variables, their low and high levels, and
residual aj is random and normally distributed with a the annual solar fractions are given in Table 3. The main
mean of 0 and variance (to. 2 The values of p, and o,o used effects (i.e. the changes in the differences between the
(0.604 and 0.135) are radiation weighted averages of the annual solar fractions resulting from changing the levels
values for the four U.S. locations. of the design variables) are given in Table 4; the inter-
SOLMET Typical Meteorological Year (TMY) data for action effects are not given, as all were less than 0.4 per
Seattle-Tacoma, Washington were used in the simula- cent solar fraction.
tions. Seattle is notable for low values of RT in the winter The results in Table 4 indicate that the only significant
and high values of RT in the summer. (Earlier simulations difference in annual solar fractions is between the Liu
using data from Madison, WI, Albuquerque, NM and and Jordan correlation and eqn (1). For the range of
Seattle, WA, indicated that the Seattle data produced the system parameters investigated, the effect of adding a
largest disagreement among correlations.) statistical variance to eqn (I) on the simulation results is

Table 3. Design variablesand annual solar fractions for flat plate system
Level of Design Variable Annual Solar Fraction

Collector Collector Tank Vo1~e Collector Gollector Llu and Equatlon Aerospace Statistical
Azlmuth Slope ~ r e a Quality ArR ~rdan [I] Eq. [I]

35 l/m! "Poor" 45 .~ .39 .37 .37 .37

35 1/m', "Good" 45 m2 .52 .48 .49 .49
35 l/re' "Good" 45 m2 .4S .41 .42 .43
35 1/m', "Poor" 45 m2 .28 .26 .26 .26
150 l/m', "Good" 45 ea2 .60 .55 .S6 .57
150 1/m', "Poor" 45 m2 .40 .38 .38 .38
150 l/re' "Poor" 45 m2 .36 .33 .33 .33
150 1/m', "Good" 45 m2 .4S .41 .41 .42
35 1/m' "Good" 90 m2 .72 .66 .67 .69
35 1/m', "Poor" 90 m2 .4g .45 .46 .47
35 1/m', "Poor" 90 m2 .44 .40 .41 .41
35 l/re' "Good" 90 m2 .57 .51 .52 .54
150 I/m' "Poor" 90 mE .S8 .54 .55 .55
150 l/re', "GOod" 90 m2 .69 .64 .66 .66
150 1/m', "Good" 90 m .70 .63 .65 .65
150 1/m' "Poor" go m2 .45 .41 .41 .42

Average for All Runs .Sl .46 .47 .48

Collector Type F' ~ a KL Ube (w/m2 - °C) Covers

"Poor" .75 .95 .gS .06 1.4 single

"Good" .g5 .lO .95 .012 .Z8 single

UAHousE - 333 w/m2 - °C UTANK - .28 w/m2 - °C MCOLLECTO

R ° - 60 kg/m 2 - hr
298 D. G. ERBS et al.

Table 4. Effect of design variables on differences in estimated solar fractions for fiat plate system
DifferenCe in Annual Solar Fraction

Main Effect of Design Variable

(Statistical
(Llu and Jordan Correlation (Aerospace
Design Variable - Equation [l]) -Equation [ l ] ) -Equation [ l ] )

Average .043 .014 .007

Azlmuth -. 005 -.003 -•004
Slope .006 -.001 • 000
Tank Volume Ratio .006 • 000 • O01
Collector Quality .017 .010 -•004
Collector Area .017 .008 • 006

negligible. The difference between the statistical Aero- 4• ESTIMATION OF THE DIFFUSE FRACTION OF DAILY GLOBAL
space model of Randall and Whitson and eqn (1) (both RADIATION
developed from the same data base) is also negligible. A correlation was developed between the daily diffuse
The second series of simulations investigated three fraction, ttd/H, and KT using the hourly data from the
variables in an industrial process heating system. The four U.S. locations. It was shown earlier that the cor-
collectors are two-axis tracking linear concentrators with relation between IJI and kr is seasonally dependent• To
a concentration ratio of 15. The load consists of generat- determine whether the correlation between HSH and g r
ing 100°C steam from 8:00am to 12:00pm, 5 days a was also dependent upon season, the daily data were
week. Table 5 lists the design variables, their low and grouped into seasonal bins. Following the procedure of
high levels (along with other system parameters), and the Collares-Periera and Rabl[13], 3 bins were chosen:
annual solar fractions. The main effects are presented in
Table 6. Once again, all interactions were less than 0.4 Winter ~, < 1•4208
per cent fraction by solar. Spring and Fall 1.4208 -< cos -< 1.7208
Based on the results in Table 6, the only significant Summer ~o, > 1.7208. (5)
difference in annual solar fractions is again between the
Liu and Jordan correlation and eqn (1). Although larger Separating the data in this manner resulted in stronger
than they were for the flat-plate system, the differences seasonal trends than were obtained when the data were
between the statistical version of eqn (1) and eqn (1) and separated into seasons by month.
between the Aerospace method and eqn (1) are still Only the correlation for winter was noticeably
negligible. The concentrating collectors only utilize beam different from the other two seasonal correlations• For
radiation, and as a result, they are much more sensitive values of Kr greater than 0.45, the winter data have a
to the split of global radiation into the beam and diffuse smaller average diffuse fraction than the data for the
components. These comparisons show that a con- remainder of the year• The air in winter is generally drier
sideration of the statistical variation of the diffuse frac- and less dusty than during other seasons, which tends to
tion, as in the Randall and Whitson algorithm, is un- lower the diffuse fraction during the winter for large
necessary for estimating the long-term performance of values of Kr. In addition, the lower solar altitude in the
solar energy systems. winter allows less of the scattered radiation to reach the

Table 5. Designvariables and annual solar fractions for concen~ating system

Level of Design Variable Annual Solar Fraction

Tank V o h m ~ Tank L o s s Collector Llu and Equation Stetlstlca;

Collector A r e a Coefficient Area Jordan [l] Aerospace [l]

50 I/m 2 0.70 w/m~-°C 40 m~ .38 •31 .29 •31

200 I/m 2 0.70 w/m2-°C 40 m2 .31 .24 .23 •24
50 I/m 2 0.14 ~/m~-% 40 m2 .44 .35 .33 .35
200 I/m 2 0.14 w/#-°C 40 m2 .43 .36 .35 •36
50 I/m 2 0.70 w/m2-°C 80 m2 .65 .56 .54 .58
200 I/m z 0.70 w/m2-°C 80 m2 .58 .49 .46 .49
50 I/m 2 0.14 w/m2-°C 80 m2 .72 .64 •62 .65
200 I/m 2 0.14 w/m2-°C 80 m2 .75 .65 .63 .66

Average f o r All Runs .53 .45 .43 .46

FD = .B5 ( ~ ) n = •75 Evacuated tubular receivers

C811ector area mass flow rate - SO kg/m~- hr *
Collector loss c o e f f i c i e n t - .13 w/m - C
Load - 16,670 W *per nomtnal aperature area
 Estimation of the diffuse radiation fraction for global radiation 299

Table 6. Effect of design variables on differences in estimated solar fractions for concentrating system
Difference in Annual Solar Fraction
Main E f f e c t o f Design Varlab]e
tStatlstlcal
(Liu and Jordan Correlation (Aerospace
Design Variable -Equation [l]) -Equation [ 1 ] ) -Equation [ 1 ] )
~verage • 083 • 008 -.016
Tank Volume Ratio • 006 -. 004 - . 004
Tank Loss Coefficient • 004 • 003 .004
3ollector Area .018 .012 -•003

ground, and if clouds are present, the fraction of the amount of data available for developing a daily cor-
hour during which there is beam radiation must be larger relation will be 30 times larger than that for the monthly-
due to the lower transmittance of the clouds and the average data. The daily correlation will be more
atmosphere• smoothly defined and there will be greater precision in
Each of the seasonal correlations was fit with an fitting the daily correlation when compared to the
equation. The equations obtained for the summer data monthly-average correlation developed from the
and the combined fall and spring data were virtually the data. In addition, the monthly-average correlation
same. The seasonal correlations are represented by the derived from the daily correlation should be
following equation: more accurate over the long-term than a correlation of
the monthly-average data because it reflects the long-
For o~s< 1.4208 term average radiation distribution. A final advantage is
HdlH = 1.0 - 0.2727 KT + 2.4495 K~-- 11.9514 K 3 that a monthly-average correlation derived using eqn (7)
covers the range of /~r from 0.3 to 0.7, even though
+ 9.3879 K~ for Kr < 0.715
measured values o f / ( r may not extend over this large a
HJ H = 0.143 [or K~ >-0.715. range.
For w, -> 1.4208 The relationship between ~Id]lTtand Hd/H suggested
Hd/H = 1.0+0.2832 Kr -2.5557 K2+ 0.8448 K]- by Liu and Jordan can be written as:

for Kr < 0.722

Hd/H = 0.175 for K~ - 0.722. (6) - ~ : NRT -~ Kr. (7)

Equation (6) is shown with the average seasonal bin

To use eqn (7), N values of HdIH and Kr must be
values in Fig. 6.
known for each value of/(St, where N is a large number,
sufficient to represent long-term average conditions. The
5. ESTIMATION OF THE DIFFUSE FRACTION OF MONTHLY- values of HJH are obtained from a daily correlation
AVERAGE RADIATION between HalH and Kr. All that remains is to determine
One method of developing a relationship between the distribution of Kr characteristic of the long-term
ftd[~I and/~T is to sum the daily values of Hal,H and Ho average for each value of/(T.
for all days in each month and correlate the resulting The definition of/~T is given by the equation:
values of/~u[/'t and/(T. However, it was shown by Liu
and Jordan[8] that a relationship between ffIu[~Iand/~T 1

can be obtained from the relationship between HdlHand Rr = KT df (8)

KT if the long-term average distribution of Kr is known.
There are several advantages to deriving the monthly- where jr is the fraction of the time Kr was less than a
average correlation from the daily correlation. The particular value. Liu and Jordan found that for widely
spread locations, cumulative distributions of Kr having
~ - All Four Cities -~ the same value of /(r were remarkably similar. They
1.0~1~I~¢~. /-~ Correlofion for used data from 27 locations to develop generalized cumu-
F '~.~ / ~'s`-q'42 t lative frequency distributions of Kr for the 5 values of
8F-
\7 r-c°,re'°''°°'°r 1 /~r between 0.3 and 0.7. Recently, Bendt et al.[23] used
b :"1~ 7 Cds>l'42 data from 90 locations, with 20 yr at each location, to
develop generalized KT distributions. Theilacker [2.4] has
also recently developed generalized Kr distributions
T'O'4 ~- +:Spring/Foll (I .42-~S-1.72) ~ x ~ .
using 23 yr records for several locations. The agreement
• 2~~ x = Winter (WsSl.42) x.x_ ~. . . . . with the distributions of Liu and Jordan is generally very
.OL t <1, I L L , ~ , l l ~ , 1, ~, I, x ~ ! _ ~ I , i , I , i , I , i ~ good.
•0 .l .2 .3 .4 .5 .6 .7 .8 .9 l.O
KT The generalized Kr distributions of Liu and Jordan
Fig. 6. Comparison of the seasomal U.S. data with the seasonal can be used to generate values of Kr for any value of
correlation between the daily diffuse fraction and KT. /(T. The procedure is to choose N values of f which are

SE VoL 28, No. 4--C

300 D. G. ERBSet aL

equally spaced between 0 and 1. The corresponding moisture and dust content of the air and in the dis-
values of Kr are then found from the appropriate /~'r tribution of cloud cover. The hourly data for the four
curve. This allows eqn (7) to be evaluated for values of U.S. locations were also used to develop a nonseasonal
/~r between 0.3 and 0.7. The seasonal daily diffuse daily diffuse correlation from which the following non-
correlation, eqn (6), was used along with eqn (7) and the seasonal monthly-average daily diffuse correlation was
Kr distributions of Liu and Jordan as curve fit by derived:
Cole[25] to derive a seasonal monthly-average daily
diffuse correlation. The following equation was fit to the For 0.3 -</~r -< 0.8
correlation: fta[tt = !.317 - 3.023/~r + 3.372/Cr - 1.769/C~. (10)

For co, -< 1.4208 and 0.3 -/~T -< 0.8 Equation (10) compares with a mean bias error of 0.002
#,d/Y = 1.391 - 3.560/~r + 4.189 g~-- 2.137/~]- to the monthly-average data for Highett and the four
For o~, > 1.4208 and 0.3 -</~r -< 0.8 U.S. locations.
The monthly-average correlations of Liu and Jordan,
HJH = 1.311 - 3.022/~r + 3.427/C]-- 1.821/~']-. (9)
Page, Collares-Periera and Rabl, Hay, and eqn (9) were
compared to the data from the 4 U.S. locations and from
The monthly-average data were grouped into 3 Highett. The relationship developed by Hay requires
seasonal bins according to the monthly-average value of
monthly-average values of ground reflectance; a value of
the sunset hour angle. The 3 bins are defined by eqn (5). 0.2 was used for all months and locations. The standard
Figure 7 is a comparison of the seasonal monthly- deviations and mean bias errors from each correlation
average correlation (eqn 9) with the combined United
were calculated for each location. The results are
States and Highett data. There are no distinguishable
seasonal trends in the Highett data, while the seasonal
variation in the diffuse fraction of the U.S. data is larger
than the seasonal variation of the correlation. The mon-
thly-average diffuse fraction is more strongly dependent ~. ;lI "-~28~-
~ - I' l ' l ' l ' l"' l ' l O
\1(///~/
' r 'Fl ' aES
Spring,
r'easlonIasI'elaWi
s'onlfano.f'lrt~eSummer
rCUrVecurve
i' - r,-' l ' for
r'''i''
upon season than the daily diffuse fraction. One explana-
tion for this, which is supported by Theilacker and by
Bendt et aL, is that the generalized Kr distributions have .6~- \"~ Aerospace and
more variation in Kr (They are "steeper") in winter and II L- "~~'~-~ Highett Data 1
less variation in Kr ("flatter") in the summer, which .4~zx-Sumrner(OJs~l.7~')"~'~ll3 Months Data ,1
would increase the dependence of the monthly-average lie • 21 +-Spring/Fall (l.42 ~wsc1.72)>t'~q-~'~.....
data from that of the daily data. ~- x=Winter(%-1.42) "~
For some locations, such as Highett, Australia, the OF, J ,],*, I,l,l, ~, I L I L J , I , I , I , I , I ,I,J,l,I
monthly-average diffuse fraction does not exhibit •0 .1 .2 .3 .4 ~.T.5 .6 .7 .8 .9 1.0
seasonal dependence. The proximity of Highett to the Fig. 7. Comparisons between the derived seasonally dependent
ocean may tend to damp seasonal variations in the correlation and the seasonal monthly-average data.

Table 7. Comparisonof existing monthly-averagecorrelations with present U.S. and Highett data
C-P
Location L+J Page + Rabl Hay (I) Eq. [8]
-0.041 0.007 0.017 -0.011 0.011
Fort Hood, TX
0.079 0.078 0.048 0.068 0.058
-0.018 -0.008 0.076 0.032 0.032
Livermore, CA
o 0.058 0.037 0.109 0.072 0.061
-0.070 -0.026 -0.014 -0.043 -0.026
Raleigh, NC
0.085 0.048 0,042 0.065 0.044
-0.069 -0.003 -0.000 -0.041 -O.Oll
~aynard, MA
o 0.086 0.058 0,033 0.068 0.046
Combined -0.051 -0.003 0.016 -O.01g 0.002
U.S.
o 0.080 0.063 0.058 0.068 0.053
Highett -0.049 0.002 0.024 -0.021 0.007
~ustralia c7 0.067 0.040 0.071 0.050 0.048
M1 Data -0.050 -0.001 0.019 -0.036 0.004
Combined
0.074 0.054 0.064 0.061 0.051
 Extimation of the diffuse radiation fraction for global radiation 301

presented in Table 7. The Liu and Jordan correlation and a absorptance

eqn (9) were derived from daily correlations, while the ~ emittance
other relationships were developed using monthly- p coefficientofautocorrelation
average data. cr standard deviation
~- transmittance
~, sunset hour angle
6. CONCLUSIONS
The correlation developed between the hourly diffuse Subscripts
b back
fraction and kr was found to be essentially the same as
c calculated
the relationship previously developed by Orgiil and e edge
Hollands [2], although different data were used in each m measured
case. Data recorded in Highett, Australia were also n normal
found to agree to within a few per cent with the hourly
relationship presented. While the uncertainty in the REFERENCES
estimated diffuse fraction for an hour is significant, the I. E. C. Boes, Estimating the direct component of solar radia-
correlation predicts the long-term average hourly diffuse tion. Sandia Report SAND75-0565, (1975).
fraction accurately. For the systems investigated, the 2. J. F. Orgill and K. G. T. Hollands, Correlation equation for
importance of simulating the distribution of hourly hourly diffuse radiation on a horizontal surface. Solar Energy
19, 357 (1977).
diffuse fractions about the long-term average in com-
3. R. Bruno, A correction procedure for separating direct and
puter simulations is minor. Only the long-term relation- diffuse insolation on a horizontal surface. Solar Energy 20,
ship between Id]I and kr, and not the random nature of 97 (1978).
Id/I, appears to be important. For simulations involving a 4. J. W. Bugler, The determination of hourly insolation on an
flat-plate collector system, the results obtained with inclined plane using a diffuse irradiance model based on
hourly measured global horizontal insolation. Solar Energy
different correlations were generally within 5 per cent of 19, 477 (1977).
each other, but for a concentrating collector system, the 5. J. A. Duffie and W. A. Beckman, Solar Engineering of
Liu and Jordan daily diffuse correlation resulted in Thermal Processes.Wiley New York (1980).
significantly higher estimates of system performance. 6. C. M. Randall and M. E. Whitson, Final report--hourly
insolation and meteorologicaldata bases including improved
The seasonal monthly-average daily diffuse correlation direct insolation estimates. Aerospace Report No. ATR-78
given by eqn (9), which was derived from a seasonal (7592)-1 (1977).
daily diffuse correlation, agrees closely with the monthly- 7. SOLMET, Volume 2--Final report Hourly solar radiation
average U.S. and Highett data. The monthly-average surface meterologicalobservations. TD--9724(1979).
correlations of Collares-Pereira and Rabl and of Page 8. B. Y. H. Liu and R. C. Jordan, The interrelationship and
characteristic distribution of direct, diffuse, and total solar
agree with the U.S. and Highett data nearly as well as the radiation. Solar Energy 4, 1 (1960).
monthly-average correlations presented. 9. N. K. D. Choudhury, Solar radiation at New Delhi. Solar
Energy 7, 44 (1963).
10. G. Stanhlll, Diffuse sky and cloud radiation in Israel. Solar
NOMENCLATURE Energy 19, 96 (1966).
a diffuse fraction residual 11. S. E. Tuller, The relationship between diffuse total, and extra
d bias error terrestrial solar radiation. Solar Energy IS, 259 (1976).
d mean bias error 12. D. W. Ruth and R. E. Chant, The relationship of diffuse
/ cumulative fraction of occurrence radiation to total radiation in Canada. Solar Energy 18, 153
3~ annual solar fraction (1976).
F' collectorefficiencyfactor 13. M. Collares-Pereira and A. Rabl, The average distribution of
F~ collectorheat removal factor solar radiation---correlations between diffuse and hemis-
H daily total radiation incident on a horizontal surface pherical and between daily and hourly insolation values.
Ha daily diffuse radiation incident on a horizontal surface Solar Energy 22, 155 (1979).
Ho daily extraterrestrial radiation incident on a horizontal sur- 14. J. K. Page, The estimation of monthly mean values of daily
face total short-wave radiation on vertical and inclined surfaces
monthly-average daily total radiation incident on a from sunshine records for latitudes 400N---40"S.Proc. UN
horizontal surface Conference on New Sources of Energy 4, 378 (1964).
Hd monthly-average daily diffuse radiation incident on a 15. M. Iqbal, A study of Canadian diffuse and total solar radiation
horizontal surface data--I. Monthly average daily horizontal radiation, Solar
Ho monthly-averagedaily extraterrestrial radiation incident on Energy 22, 81 (1979).
a horizontal surface 16. J. E. Hay, Calculation of monthly mean solar radiation for
I hourly total radiation incident on a horizontal surface horizontal and inclined surfaces. Solar Energy 23, 301 (1979).
/~ hourly "clear sky" total radiation incident on a horizontal 17. J. W. Bannister, Solar radiation records. Division of
surface Mechanical Engineering, Commonwealth Scientific and In-
ld hourly diffuse radiation incident on a horizontal surface dustrial Research Organization, Highett, Victoria, Australia
Io hourly extraterrestrial radiation incident on a horizontal (1966--69).
surface 18. Monthly solar climatological summary for the solar energy
K extinction coefficient meteorological research and training site--region 2. Atmos-
kc hourly clearness index (ratio of I to Ic) pheric Sciences Research Center, State University of New
kr hourly clearness index (ratio of I to I0) York at Albany, Albany, New York (1980).
Kr daily clearness index (ratio of H to Ho) 19. H. C. Hottel, A simple model for estimating the trans-
Kr monthly-averagedaily clearness index (ratio of/~ to Ho) mittance of direct solar radiation through clear atmosphere.
L length,thickness Solar Energy 18, 129 (1976).
N number of days 20. G. E. P. Box and G. M. Jenkins, Time Series Analysis
U loss coefficient Forecasting and Control. Holden-Day, San Francisco (1970).
302 D. G. ERBSet aL

21. S. A. Klein et al. 'TRNSYS--transient simulation program. 24. J. C. Theilacker, An investigation of the monthly-average
University of Wisconsin-Madison Engineering Experiment utilizability for flat-plate solar collectors. Masters Thesis in
Station Report 38-10, Version 10.1 (1979). Mechanical Engineering, University of Wisconsin-Madison
22. G. E. P. Box, W. G. Hunter and J. S. Hunter, Statistics for (19S0).
Experimenters Wiley New York (1978). 25. R. J. Cole, Long-term average performance predictions for
23. P. Bendt, A. Rabl and M. Collares-Pereira, the frequency CPC's Proc. Am. Section of I.S.E.S., Orlando, Florida
distribution of daily isolation values. Solar Energy Research (1977).
Institute, Submitted to Solar Energy (1980).