Solar Energy Vol. 72, No. 5, pp.

385–395, 2002
 2002 Published by Elsevier Science Ltd
Pergamon PII: S0038 – 092X( 02 )00019 – 1 Printed in Great Britain
0038-092X / 02 / $ - see front matter
www.elsevier.com / locate / solener

DYNAMIC ANNUAL DAYLIGHT SIMULATIONS BASED ON ONE-HOUR

AND ONE-MINUTE MEANS OF IRRADIANCE DATA
OLIVER WALKENHORST † , *, JOACHIM LUTHER*, CHRISTOPH REINHART** and
,
JENS TIMMER*** ****
*Fraunhofer Institute for Solar Energy Systems ISE, Heidenhofstraße 2, 79110 Freiburg, Germany
**Institute for Research in Construction, National Research Council, 1200 Montreal Road, Ottawa,
ON K1A 0R6, Canada
***Faculty of Physics, University of Freiburg, Hermann-Herder-Straße 3, 79104 Freiburg, Germany
****Center for Data Analysis and Modelling, University of Freiburg, Eckertstraße 1, 79104 Freiburg,
Germany

Received 11 July 2001; revised version accepted 15 February 2002

Communicated by JEAN-LOUIS SCARTEZZINI

Abstract—This study investigates the influence of the short-term dynamics of daylight on simulation-based
predictions of the annual daylight availability in a building. To this end annual indoor illuminance simulations
are carried out for a two-person-office using the RADIANCE-based dynamic daylight simulation method
DAYSIM. As of yet, all available daylight simulation methods are typically based on 1-h means of irradiance
data and thus tend to neglect the short-term dynamics of daylight. In the first part of this study the dependence
of the annual daylight availability on the underlying simulation time step interval is quantified. Assuming two
different automated daylight-dependent artificial lighting strategies, the predicted annual artificial lighting
demand is systematically underestimated by up to 27% on the simulations based on 1-h means instead of
1-min means of measured beam and diffuse irradiances. The general validity of these results is ensured by
employing irradiance data from five stations world-wide. As measured 1-min means of irradiance data are
generally not available for practical applications, the stochastic Skartveit–Olseth model, which generates 1-min
means of irradiance data from hourly means, is adapted for daylight simulation purposes in the second part of
this study. The utilization of modeled 1-min means of irradiance data reduces the above described systematic
simulation errors to below 8% for both automated lighting strategies and all five stations. Accordingly, the
modified version of the Skartveit–Olseth model is able to enhance the quality of dynamic daylight simulations
— without any additional planning effort for the lighting designer.  2002 Published by Elsevier Science
Ltd.

1. INTRODUCTION To compare different daylighting concepts dur-

ing the planning phase of a building one has to
Within the present discussion dealing with new
rely on simulation methods that allow one to
ideas on the construction of buildings the concept
predict the annual daylight availability in the
of a so-called lean building plays an increasing
interior of a building. To this end the
role (Voss et al., 2000). Lean buildings harmonize
RADIANCE-based simulation environment DAY-
with their given climatic boundary conditions and
SIM which represents a reliable and easy-to-use
exploit naturally available energy sinks and
tool to perform dynamic indoor illuminance simu-
sources in order to provide increased thermal and
lations has recently been developed and validated
visual comfort for their inhabitants while reducing
(Reinhart and Herkel, 2000; Reinhart and Wal-
the energy demand.
kenhorst, 2001). Similar to the ESP-r /
One aspect of a lean building is the conscious
RADIANCE link (Janak, 1997) and the ‘Dynamic
use of daylight to light the interior of buildings.
Lighting System’ (Cropper et al., 1997; Mardal-
The benefits of a carefully planned daylighting
jevic, 2000), DAYSIM is based on a daylight
concept encompass an enhanced visual comfort
coefficient approach (Tregenza and Waters, 1983)
for the inhabitants providing them with glare-free
and the Perez sky model (Perez et al., 1990,
natural daylight and visual contact to the outside
1993). Daylight coefficient methods are able to
as well as a reduced electric energy demand for
simulate the time development of indoor illumi-
artificial lighting.
nances for arbitrary time step intervals based on
time series of beam and diffuse irradiances. As
†
Author to whom correspondence should be addressed; e-mail: widely available test reference years usually pro-
walkenho@web.de vide hourly means, most dynamic daylight simu-

385
386 O. Walkenhorst et al.

lation methods are presently based on this time the annual artificial lighting demand for two
step, i.e. the short-term dynamics of daylight automated daylight-dependent lighting control
introduced by clouds intermittently hiding the sun strategies. Measured 1-min irradiance data sets
(see Fig. 1) or a temporarily varying cloud cover are chosen as a lower threshold with respect to
thickness is discarded. As daylight cannot be time resolution based on the assumption that
stored, the usage of 1-h irradiance data in dy- they contain all information about the short-
namic daylight simulations may thus lead to term dynamics of daylight which is relevant in
considerable errors in the prediction of the annual the context of daylight simulations and as
daylight availability. measured irradiance data with a higher time
Janak suggested to use a stochastic model resolution are scarcely available.
introduced by Skartveit and Olseth (Skartveit and 2. Reduction of the simulation errors with respect
Olseth, 1992) to model the intra-hour dynamics of to time resolution: Measured 1-min irradiance
the beam radiation and thereby reduce these errors data are generally not available for practical
(Janak, 1997). He implemented the Skartveit– applications. To reduce the simulation errors,
Olseth model into ESP-r without further validat- 1-min irradiance data are modeled from hourly
ing its applicability for dynamic daylight simula- means using a modified version of the Skar-
tions. This paper extends Janaks original work by tveit–Olseth model (Skartveit and Olseth,
both quantifying the simulation errors introduced 1992). This modified version is developed for
by using 1-h instead of 1-min time steps for use in dynamic daylight simulations. To quan-
annual daylight simulations (1) and by presenting tify the error in the prediction of the annual
and validating a practical method to reduce these daylight availability which remains if one uses
simulation errors based on the Skartveit–Olseth modeled 1-min irradiance data, dynamic annu-
approach (2): al indoor illuminance simulations are carried
1. Quantification of simulation errors with respect out with measured 1 -min means and modeled
to time resolution: To quantify the errors in the 1 -min means of irradiance data as input. As in
prediction of the annual daylight availability (1) the two resulting annual indoor illuminance
which one incurs by neglecting the short-term data sets are then compared with respect to the
dynamics of the daylight, dynamic annual predicted annual artificial lighting demand to
indoor illuminance simulations are carried out ensure the capability of the modified Skar-
for a test office with two different types of tveit–Olseth model to reduce errors in dynamic
underlying irradiance data sets: measured 1 -h daylight simulations. The organization of this
means and measured 1 -min means. The re- paper is as follows: Section 2 describes the
sulting two different types of annual indoor methodologies to investigate problems (1) and
illuminance data sets are then used to predict (2), simulation results are presented in Section

Fig. 1. Measured 1-h (solid line) and measured 1-min (dashed) means of global irradiance on May 17th 1998 in Freiburg,
Germany.
 Dynamic annual daylight simulations based on 1-h and 1-min means of irradiance data 387

3 and finally Section 4 contains concluding

remarks and an outlook on remaining research
tasks.

2. METHODOLOGY

2.1. Dynamic daylight simulations with DAYSIM

In this study all dynamic simulations of indoor
illuminances due to daylight are performed using
the RADIANCE-based simulation environment
DAYSIM. DAYSIM merges the backward
raytracer RADIANCE (Ward and Shakespeare,
1998) with a daylight coefficient approach and
permits reliable and fast dynamic indoor illumi-
nance simulations.
DAYSIM requires two essential informations Fig. 2. Sketch of the test office including the position of the
for an annual indoor illuminance simulation: three simulation points.
(a) Description of the geometry and the materi-
als of the building: This information is provided
in the form of regular RADIANCE input files.
(b) Description of the light sources for each 2.2. Irradiance data sets
time step: Ambient daylight is physically de- To examine the two problems described in the
scribed by the sky luminance distribution. As Introduction, annual indoor illuminance simula-
measurements of this two-dimensional function tions are carried out for three different global and
are usually not available for practical applications, beam irradiance data sets:
DAYSIM makes use of the Perez sky luminous • measured 1-min means (1)
efficacy (Perez et al., 1990) and sky luminance • measured 1-h means (2)
distribution model (Perez et al., 1993) to model • modeled 1-min means (3)
the sky luminance distribution using global and Fig. 3 provides a schematic survey of the
beam irradiance — which are widely available — daylight simulations and the underlying irradiance
as input values. Therefore DAYSIM needs to be data sets employed in this study.
fed with an annual data set of global and beam
irradiances. 2.2.1. Measured 1 -min means. Measured 1-
Given these two input files, DAYSIM allows min means of global and beam irradiances are the
one to simulate an annual indoor illuminance data basic irradiance data set from which the other two
set for any specified point and orientation inside sets, measured 1-h means and modeled 1-min
of a given building. More details regarding the means, are derived.
underlying simulation algorithm of DAYSIM are To support the general validity of the simula-
provided in Reinhart and Herkel (2000) and tion results, eight annual data sets of measured
Reinhart (2001). 1-min global and beam irradiance from five
In the present study indoor illuminances are different weather stations world-wide are used.
simulated for three simulation points inside and The five stations are listed in Table 1.
outside of a southward oriented 18 m 2 test office All five stations are affiliated to the Internation-
which is sketched in Fig. 2. Two simulation al Daylight Measurement Programme (IDMP)†
points inside the office are situated at work plane launched by the CIE in 1991.
height on the middle axis of the office 2 and 4 m
from the facade, both pointing upwards. They 2.2.2. Measured 1 -h means. Each of the eight
represent the work places of the two-person-of- annual data sets of measured 1-min global and
fice. The third simulation point is situated on the beam irradiance described in the previous subsec-
facade of the test office and points southward. All
simulations are based on this test office. Only the
underlying annual global and beam irradiance
†
data sets are varied. http: / / idmp.entpe.fr
388 O. Walkenhorst et al.

Fig. 3. Schematic survey of the simulation pathways in the present study. The numbers pertain to the enumeration in the text.

tion yields — by averaging — an annual data set (AR(1) process, see, e.g. Brockwell and Davis,
of measured 1-h global and beam irradiance. 1991).
6. Back-transformation to absolute quantities
2.2.3. Modeled 1 -min means. In this study a using the McMaster clear sky model which
modified version of the stochastic Skartveit– results in 1-min global irradiances gi 5 k i ? gc,i
Olseth model is used to generate 1-min global and and beam irradiances gb,i 5 k b,i ? gbc,i .
beam irradiance data from measured hourly 7. Output: 60 1-min means of global irradiance gi
means. The basic structure of the original Skar- and beam irradiance gb,i (i 5 1, . . . ,60).
tveit–Olseth is sketched in the following. The The output 1-min global and beam irradiances
model consists of seven steps which are carried for each hour are then concatenated hour by hour.
out for each hour of the year with non-vanishing This way one can obtain annual data sets of
ambient irradiances: modeled 1-min global and beam irradiance from
1. Input: 1-h means of global irradiance K # 0.3 data sets of hourly means. The reader should bear
and beam irradiance 0.3 , K # 0.6. in mind that the Skartveit–Olseth model is sto-
2. Transformation to relative quantities using the chastic, i.e. different initializations of the em-
McMaster clear sky model (Davies and ployed pseudo-random number generator result in
McKay, 1982) which results in 1-h means of different realizations of annual 1-min irradiance
the so-called global clearness index K 5 G /Gc data sets.
and beam clearness index Kb 5 Gb /Gbc . While preserving the basic structure of the
3. Realization of the intra-hour standard deviation original Skartveit–Olseth model, the following
sk 5 (o i6051 (ki 2 K)2 )1 / 2 and sk b 5 (o i6051 five modifications have been carried out to render
(k b,i 2 Kb )2 )1 / 2 of the 60 1-min global and the model more suitable for dynamic indoor
beam clearness indices k i and k b,i using a illuminance simulations:
parametrized probability distribution. 1. Synchronization of global and beam ir-
4. Realization of 60 1-min global and beam radiances: As mentioned above DAYSIM em-
clearness indices k i and k b,i using parametrized ploys the Perez sky luminance distribution
probability distributions. model which requires consistent global and
5. Temporal rearrangement of the 60 1-min glob- beam irradiance as input, i.e. global and beam
al and beam clearness indices using realiza- irradiances have to be mutually matched at
tions of a first order autoregressive process each time step. In the original Skartveit–Olseth

Table 1. Latitudes, longitudes and station heights of the five weather stations (a.s.l.5above sea level)
Station Latitude [8] Longitude [8] Height [m a.s.l.]
Albany / USA 42.70 N 73.85 W 79
Bratislava / Slovakia 48.17 N 17.08 E 195
Freiburg / Germany 47.98 N 7.83 E 279
Geneve / Switzerland 46.33 N 6.02 E 425
Tsukuba / Japan 36.15 N 140.05 E 43
 Dynamic annual daylight simulations based on 1-h and 1-min means of irradiance data 389

model the output 1-min global and beam model scarcely influences the suitability of the
irradiances are put in order according to two original probability distributions.
independent AR(1) processes and are thus not 3. New parametrizations to improve the modeling
synchronized. Therefore it may happen that for of the minima of the 1-min global indices: A
a single time step the beam irradiance exceeds thorough analysis based on the eight annual
the global irradiance which leads to a negative data sets of measured and modeled 1-min
diffuse irradiance. To obtain consistent 1-min global irradiances indicated that the original
global and beam irradiances, the Reindl model Skartveit–Olseth model generates too many
(Reindl et al., 1990) which estimates the very small 1-min global indices, especially
diffuse fraction of a given global irradiance is outliers below k i 5 0.2. Dynamic indoor il-
used instead in this study. The Reindl model is luminance simulations aiming at the prediction
used in its simplest version which merely of the annual daylight availability are very
requires global irradiance and solar elevation sensitive to this type of shortcoming as too
as input parameters. In the modified Skartveit– many extreme global irradiance minima result
Olseth model steps (2)–(6) are solely per- in an underestimation of the annual daylight
formed for the global irradiance and between availability and consequently in an overestima-
step (6) and step (7) the Reindl model is tion of the annual artificial lighting demand.
utilized to determine the beam irradiance at Therefore, a new parametrization of the three
each time step. relations s k* (K, s3 ), k min (K, sk ) and k max (K, sk )
2. Usage of the ESRA clear sky model: To (Eqs. (6b), (9b) and (9c) in Skartveit and
reduce the number of required site-specific Olseth, 1992) which pertain to step (3) and
atmospheric input parameters for the clear sky step (4) of the original model) was necessary
model — which are not available for arbitrary as these three quantities determine the extrema
sites — the ESRA clear sky model (Rigollier of the 1-min global clearness indices of each
et al., 2000) is employed in step (2) and step hour, using the average root squared deviation
(6) to replace the McMaster clear sky model. of the 3-h global clearness indices of the
The ESRA model is a broadband model which preceeding, the actual and the following hour,
predicts global and beam irradiance for a given s3 . For the new parametrization the following
site and time under clear sky conditions. It approach has been chosen for the three rela-
requires the Linke turbidity factor, which tions:
summarizes the overall turbidity of the atmos-
phere, as the only atmospheric input parameter.
s *k (new) (K, s3 ) 5 bs ? s *k (K, s3 ) (1)
In addition, the monthly mean Linke turbidity
k (new) 1.4
factors for a given site can be estimated from min 5 (K 2 0.03) ? exp(211 ? bmin ? s k )

the input hourly means of beam irradiance † . 2 0.09 (2)

Thus the input of site-specific atmospheric
k (new) 1.3
parameters becomes optional which makes the max 5 (K 2 1.5)? exp(29 ? bmax ? s k ) 1 1.5
modified Skartveit–Olseth model easier to use
(3)
for practical applications. An investigation
based on the eight annual data sets of 1-min This approach originates from a careful com-
global irradiance showed that ESRA model parison of the eight annual data sets of measured
and McMaster model yield very similar global and modeled (using the original Skartveit–Olseth
clearness indices. As, due to modification (1), model) 1-min global irradiances. The three pa-
the modified Skartveit–Olseth model is solely rameters bs , bmin and bmax have been determined
based on global clearness indices, the replace- on the basis of the eight annual data sets by
ment of the McMaster model with the ESRA minimizing a suitable objective function. For the
objective function the sum of the squared relative
†
To estimate the monthly mean Linke turbidity factors we deviations between the smallest measured and the
have used an empirically based algorithm which calculates smallest modeled 1-min global clearness index of
the Linke turbidity factor for the 4 h of each day with the each hour plus the same sum of the deviations
highest solar elevation by inverting the ESRA formula for between the biggest 1-min global clearness in-
the clear sky beam irradiance. From about the 120
dices of each hour was used. To allow for a more
calculated hourly Linke turbidity factors for each month
the smallest three are then selected and the mean of these flexible adjustment of the three parameters the
serves as an approximation for the monthly mean Linke minimization has been carried out separately for
turbidity factor for this month. four different clearness index classes to which
390 O. Walkenhorst et al.

every hour has previously been assigned to ac- be run at several coarser time steps ranging
cording to its hourly mean of the global clearness between 2 and 30 min by averaging the originally
index. Table 2 contains the results for the three generated 1-min data but has not explicitly been
parameters for all four clearness index classes. validated with measured data in this range.
4. Minimization of artificial discontinuities be-
tween subsequent hours: In the original Skar- 2.3. Comparison of the resulting simulated
tveit–Olseth model the realizations of the illuminance data sets
AR(1) process for the temporal rearrangement The three different types of annual irradiance
of the 1-min global clearness indices (step (5)) data sets described in Section 2.2 result in three
are completely independent for subsequent different simulated annual illuminance data sets
hours. Therefore, artificial discontinuities usu- for the three simulation points inside and outside
ally arise at the transition between subsequent of the test office.
hours. To minimize these artificial discon- These three annual illuminance data sets are of
tinuities, special realizations of the AR(1) different practical relevance:
process are selected. The AR(1) realization 1. The simulated illuminances based on measured
selected for each hour depends on the last 1-min irradiance data are of limited relevance
1-min global clearness index of the preceeding for practical applications due to the lack of
hour: one chooses the AR(1) realization in available measured 1-min irradiance data, but
which that 1-min global clearness index of the they serve as a reference case in the following
current hour with the smallest difference to the comparison. An accompanying study showed
last 1-min global clearness index of the pre- that they are in good agreement with measured
ceeding hour is arranged at the first position of illuminances in the test office.
the current hour. 2. The simulated illuminances based on measured
5. Input option for a horizon: The modified 1-h irradiance data represent the time resolu-
Skartveit–Olseth model allows to input the tion of conventional dynamic indoor illumi-
horizon of the simulation site which may nance simulations using test reference years.
consist of mountains or surrounding buildings. 3. The simulated illuminances based on modeled
If this information is available the horizon can 1-min irradiance data using the modified Skar-
be entered via 36 horizon heights which repre- tveit–Olseth model constitute a new possibility
sent azimuth segments of 108 width. Consider- to cope with the short-term dynamics of day-
ing the horizon can reduce artifacts in the light. They are derived from data set (2).
hours when the sun is partly below the horizon The criterion which is chosen to compare these
by taking into account the absence of beam three different illuminance data sets is the annual
irradiance during time steps when the sun is electric energy demand for artificial lighting using
hidden behind the horizon. On the one hand it the following two different automated lighting
becomes possible to calculate hourly global control strategies:
clearness indices more accurately in step (2) (a) The closed loop strategy operates with two
and on the other hand the output 1-min ir- ceiling-mounted lamps with integrated illumi-
radiances in step (7) become more precise. nance sensors facing downward which are ideally
To run the modified Skartveit–Olseth model calibrated and measure the simulated illuminance
requires no additional information besides the on the two work plane positions at 2 and 4 m
geographic coordinates of the simulation site. The distance from the facade. As soon as one simu-
generation of an annual 1-min irradiance data set lated work plane illuminance falls below a given
from hourly means requires about 90 s on a threshold illuminance the dimmed lamp lighting
Pentium Pro 200 MHz Linux workstation. this work plane instantaneously provides the
The modified Skartveit–Olseth model can also lacking illuminance to maintain the threshold
illuminance.
(b) The facade sensor strategy operates with
two ceiling-mounted lamps and one facade-
Table 2. Results for the three new fitting parameters for each
of the four clearness index classes (K is the hourly mean mounted illuminance sensor facing southward
global clearness index) which measures the simulated southward outdoor
K # 0.3 0.3 , K # 0.6 0.6 , K # 0.9 K . 0.9 illuminance. To any given indoor threshold il-
bs 1.3 1.3 1.0 0.7 luminance corresponds a much larger outdoor
bmin 20.03 0.20 0.47 2.00 threshold illuminance which is different for the
bmax 1.66 4.26 0.89 0.38 two work plane positions. As soon as the simu-
 Dynamic annual daylight simulations based on 1-h and 1-min means of irradiance data 391

Fig. 4. Predicted specific annual electric energy demand for artificial lighting assuming the facade sensor and closed loop
strategy on the basis of measured 1-min irradiance data vs. the indoor threshold illuminance for Freiburg (1998).

lated southward outdoor illuminance falls below Finally, the attendance time of the work places is
the outdoor threshold illuminance of either of the assumed to be weekdays between 8 a.m and 6
two lamps, the corresponding lighting system is p.m. Fig. 4 illustrates the order of magnitude of
fully switched on. Moreover, the lamp remains the specific annual electric energy demand for the
activated for at least 15 min after being activated. two strategies vs. the underlying threshold illumi-
This inertia of the system is introduced to supress nance.
frequent and irritating switchings of the lamps in Some remarks concerning the criterions will
the presence of temporarily varying clouds. conclude this section. The two automated lighting
Both control strategies do not take into account control strategies are chosen because they cover a
any glare protection systems or shading devices. wide range of possible automated daylight-depen-
More precisely, as criterion the specific † annual dent artificial lighting strategies: on one side
electric energy demand for artificial lighting of closed loop as a dimmed and instantaneously
the whole test office is calculated for both strate- reacting strategy and on the other side facade
gies which is measured in kW h per net office area sensor as an undimmed and inertial strategy. The
and year. The mean for the whole test office is authors do not imply that the investigated systems
obtained by averaging the specific energy demand represent the most common automated lighting
of the two separately treated simulation points systems which are in use nowadays.
with their respective lamps. To enhance the In this study only automated strategies as
general validity of the investigations, the annual opposed to manual control strategies are consid-
electric energy demands are computed for a wide ered because validated models for manual control
range of given indoor threshold illuminances It strategies are currently not available. The de-
within 200 and 2000 Lux (e.g. in Germany for a velopment of such models will be part of the
small office a minimum of 500 Lux is prescribed newly proposed IEA task 31, which will focuss on
by code (DIN, 1990)). Depending on the indoor the influence of the occupants on the daylight
threshold illuminance It , the specific delivery rate availability in a building. The authors assume that
of the lamps Ps is set to 1-h illuminance data will presumably be insuffi-
cient to model manual control strategies so that
Ps (It ) 5 2.5 W/ m 2 1 0.025(W/ m 2 ) / Lux ? It (4) short-term illuminance data — whose generation
is enabled by means of this study — will be
following the reference of the Swiss Society of required by these models.
Engineers and Architects dealing with energy in It is worthwhile to mention that the annual
building construction (SIA, 1995). The first term electric energy demand for artificial lighting is
of Eq. (4) pertains to an electronic ballast. only one possible measure for the annual daylight
availability. But every other possible measure will
supposedly depend in a similar way on the time
†
In this study ‘specific’ means ‘based on the footprint’. resolution of the illuminance data. Their sensitivi-
392 O. Walkenhorst et al.

Fig. 5. Relative deviation according to Eq. (5) for the closed loop lighting control strategy vs. the indoor illuminance threshold
for 1 year of data from each of the stations Freiburg (1998), Bratislava (1998), Geneve (1999), Albany (1996) and Tsukuba
(1994).

ty to time resolution may be smaller or even 1 -min irradiance data are compared to each other
larger than the sensitivity of the artificial lighting with respect to predicted specific annual electric
demand presented in the following section. energy demand for artificial lighting for the above
described office scenarios.
Fig. 5 presents results for the closed loop
3. RESULTS
strategy for the five investigated sites and varying
This section presents the results of the com- indoor illuminance thresholds. The figure shows
parison of the three different simulated illumi- the specific annual electric energy demand for
nance data sets. In Section 3.1 simulation errors artificial lighting based on measured 1-h ir-
of the annual artificial lighting demand due to radiance data, W1- h,measured , to measured 1-min
reduced time resolutions are quantified. Section irradiance data, W1- min,measured . To facilitate the
3.2 shows how far these prediction errors can be comparison between different stations, relative
reduced using the modified Skartveit–Olseth deviations
model. Drel 5(W1-h,measured 2W1-min,measured ) /W1- min,measured
3.1. Quantification of simulation errors with (5)
respect to time resolution are plotted.
In the following simulated annual illuminance Fig. 6 displays the same results for the facade
data sets based on measured 1 -h and measured sensor strategy. The two figures reveal that for

Fig. 6. Relative deviation according to Eq. (5) for the facade sensor lighting control strategy vs. the indoor illuminance threshold
for 1 year of data from each of the stations Freiburg (1998), Bratislava (1998), Geneve (1999), Albany (1996) and Tsukuba
(1994).
 Dynamic annual daylight simulations based on 1-h and 1-min means of irradiance data 393

both lighting control strategies and all investi- scatter around rather low values (,500 Lux) in
gated sites the annual artificial lighting demand is winter and around rather high values (2000 Lux)
systematically underestimated using 1-h instead of in summer.
1-min irradiance data. This systematic underesti-
mation lies for all stations and for all investigated 3.2. Reduction of simulation errors with respect
threshold illuminances within the range of 6–18% to time resolution
for the closed loop strategy and within the range In this section simulation results based on
of 15–27% for the facade sensor strategy. modeled 1 -min and measured 1 -min irradiance
This effect is primarily caused by the fact that data are compared.
during most annual working hours the hourly Figs. 7 and 8 correspond to Figs. 5 and 6 from
mean illuminance exceeds the considered thres- the preceeding section. They reveal that for both
hold illuminances. For these hours the simulation lighting control strategies and all investigated
based on 1-h means predicts sufficient daylight sites the time-resolution-related simulation errors
during the whole hour whereas 1-min illumi- are significantly reduced using the modified Skar-
nances may occasionally fall below the threshold tveit–Olseth model. The remaining errors amount
illuminance thereby activating the artificial light- to less than 2% for the closed loop strategy and to
ing system. During these hours one can only less than 8% for the facade sensor strategy for all
‘forfeit daylight’ if one simulates with a time considered sites and all investigated illuminance
resolution of 1 min compared to hourly simula- thresholds.
tions. This forfeiture has a stronger influence on The remaining overestimation for the facade
the facade sensor than on the closed loop strategy sensor strategy originates from the tendency of
as for the former one single 1-min illuminance the modified Skartveit–Olseth model to still gen-
that lies below the threshold illuminance results in erate too many very small 1-min irradiances.
15 min of activated artificial lighting due to the Without the improved parametrization explained
built in inertia of the control system. under 2.3.3 the overestimation for the facade
We also analyzed the seasonal variability of the sensor strategy would lie above 20%.
relative deviation for different indoor illuminance However, the remaining errors are small if one
thresholds. We found that for low indoor illumi- keeps in mind that the annual outdoor solar
nance thresholds (,500 Lux) the largest part of irradiance supply at one fixed site varies by
the overall annual relative deviation stems from around 610% (Reise, 2001) and thus leads to
the winter months. For increasing indoor illumi- natural variations of the annual daylight availabil-
nance thresholds the winter share of the overall ity for different years.
annual relative error decreases until the summer For the potential user of the model it is
share becomes the dominant part of the overall important to mention that different realizations of
annual relative error for high indoor illuminance the modeled annual 1-min irradiance data, i.e.
thresholds (2000 Lux). This is due to the fact that different runs of the stochastic Skartveit–Olseth
winter hours are on average darker than summer model, only have a minor impact on the simula-
hours, i.e. intra-hour indoor illuminances tend to tion outcome. The relative standard deviation of

Fig. 7. Relative deviation between modeled and measured 1-min irradiance data in analogy to Eq. (5) for the closed loop lighting
control strategy vs. the indoor illuminance threshold for 1 year of data from each of the stations Freiburg (1998), Bratislava
(1998), Geneve (1999), Albany (1996) and Tsukuba (1994).
394 O. Walkenhorst et al.

Fig. 8. Relative deviation between modeled and measured 1-min irradiance data in analogy to Eq. (5) for the facade sensor
lighting control system vs. the indoor illuminance threshold for 1 year of data from each of the stations Freiburg (1998),
Bratislava (1998), Geneve (1999), Albany (1996) and Tsukuba (1994).

the specific annual electric energy demand for brighter offices more hourly indoor illumi-
artificial lighting resulting from ten different nances lie above the threshold illuminance.
realizations never surmounts 0.7% for all stations, • The investigated indoor illuminance thresholds
investigated threshold illuminances and lighting comprise the wide range between 200 and
strategies. This implies that for practical purposes 2000 Lux which covers the complete range
one single realization of the model should usually which is thought to be relevant for daylighting.
yield a sufficient simulation accuracy. The modified Skartveit–Olseth model is easy to
use as it requires only marginal user input and
thus allows a more accurate prediction of day-
4. CONCLUSION
light-relevant planning quantities without any
This study shows that the neglection of the additional working effort for the lighting designer
short-term dynamics of natural daylight can intro- compared to conventional dynamic daylight simu-
duce substantial errors in the simulation of the lations based on hourly data. The modified Skar-
specific annual electric energy demand for auto- tveit–Olseth model as well as the DAYSIM
mated control strategies of artificial lighting sys- simulation environment can be downloaded from
tems. These systematic errors can be significantly www.nrc.ca / irc / ie / light / daysim.html.
reduced if one simulates indoor illuminances Finally we give an outlook on further possible
based on modeled 1-min irradiance data using the applications of modeled 1-min irradiance data:
modified Skartveit–Olseth model. 1. This study focusses on the differences between
The authors’ confidence in the general ap- 1-h and 1-min irradiance data concentrating on
plicability of the modified Skartveit–Olseth model the distribution of intra-hour minima. The
for dynamic daylight simulations is based on the distribution of intra-hour irradiance maxima
following three arguments: presumably also depends on the time resolu-
• The employed data stem from five stations tion of the data. For this reason modeled 1-min
worldwide which are situated in diverse cli- irradiance data could be used for enhanced
mates within densely populated regions. For all predictions of the frequency of glare effects
these sites a comparable quality of simulation which might in turn lead to an enhanced
results has been achieved. performance assessment of shading devices.
• Further simulations have confirmed that this 2. As mentioned in Section 2.3 the modeling of
high quality could be maintained for other manual control strategies for both artificial
office geometries with smaller facade aper- lighting systems and shading devices necessi-
tures, room depth up to 10 m and varying tates short-term irradiance data.
facade orientations. There is a tendency to- 3. More generally, short-term irradiance data
wards increased simulation errors with increas- might be of use in modeling any kind of
ing brightness of the office, i.e. the brighter the non-linear solar-driven system, especially if it
office the bigger the relative discrepancy be- exhibits a threshold behaviour as, e.g. solar
tween simulations based on 1-h and 1-min thermal devices or photovoltaic pumping sys-
irradiance data. This is due to the fact that in tems.
 Dynamic annual daylight simulations based on 1-h and 1-min means of irradiance data 395

NOMENCLATURE Kittler, Pierre Ineichen, Tsunebumi Mikuni and Richard Perez

for sharing their irradiance data with us.
Finally the authors are indepted to their colleagues at the
gi 1-min global irradiance pertaining to the ith Solar Building Design Group at the Fraunhofer Institute for
minute [W/ m 2 ] Solar Energy Systems for valuable comments and fruitful
gb,i 1-min beam irradiance pertaining to the ith discussions.
minute [W/ m 2 ]
gbc,i 1-min beam irradiance under clear sky pertain-
ing to the ith minute [W/ m 2 ]
gc,i 1-min global irradiance under clear sky per- REFERENCES
taining to the ith minute [W/ m 2 ]
G 1-h global irradiance [W/ m 2 ] Brockwell P. J. and Davis R. A. (1991). Time Series: Theory
and Methods, Springer, New York.
Gb 1-h beam irradiance [W/ m 2 ]
Cropper P., Lomas K. J., Lyons A. and Mardaljevic, J. (1997)
Gbc 1-h beam irradiance under clear sky [W/ m 2 ]
A Dynamic Lighting System: Background and Prototype.
Gc 1-h global irradiance under clear sky [W/ m 2 ] LuxEuropa 97 Proceedings, Amsterdam, pp. 480–492.
It indoor threshold illuminance [Lux] Davies J. A. and McKay D. C. (1982) Estimating solar
ki 1-min global clearness index pertaining to the irradiance and components. Solar Energy 29(1), 55–64.
ith minute [–] ¨
DIN (1990) DIN 5035, Teil 2: Beleuchtung mit kunstlichem
k b,i 1-min beam clearness index pertaining to the Licht, Richtwerte fur¨ Arbeitsstatten
¨ ¨
in Innenraumen und im
ith minute [–] Freien.
k max upper threshold for 1-min global clearness Janak M. (1997). Coupling building energy and lighting
indices during hour [–] simulation. Fifth International IBPSA Conference, Prague,
8–10 September 1997, Vol. II, pp. 313–319.
k (new)
max upper threshold for 1-min global clearness
Mardaljevic J. (2000) Simulation of annual daylighting pro-
indices during hour (according to new
files for internal illuminances. Light. Res. Technol. 32(3),
parametrization) [–] 111–118.
k min lower threshold for 1-min global clearness Perez R., Ineichen P., Seals R., Michalsky J. and Stewart R.
indices during hour [–] (1990) Modeling daylight availability and irradiance com-
k (new)
min lower threshold for 1-min global clearness ponents from direct and global irradiance. Solar Energy
indices during hour (according to new 44(5), 271–289.
parametrization) [2] Perez R., Seals R. and Michalsky J. (1993) All-weather model
K 1-h global clearness index [–] for sky luminance distribution — preliminary configuration
Kb 1-h beam clearness index [–] and validation. Solar Energy 50, 235–245.
Reindl D. T., Beckman W. A. and Duffie J. A. (1990) Diffuse
Ps specific delivery rate of the lamps [W/ m 2 ]
fraction correlations. Solar Energy 45(1), 1–7.
W1- h,measured specific annual electric energy demand for
Reinhart C. F. and Herkel S. (2000) The simulation of annual
artificial lighting based on measured 1-h ir- daylight illuminance distributions — a state-of-the-art com-
radiance data [kW h / m 2 a] parison of six RADIANCE-based methods. Energy Build.
W1- min,measured specific annual electric energy demand for 32, 167–187.
artificial lighting based on measured 1-min Reinhart C. F. and Walkenhorst O. (2001) Validation of
irradiance data [kW h / m 2 a] dynamic RADIANCE-based daylight simulations for a test
bs fitting parameter [–] office with external blinds. Energy Build. 33, 683–697.
bmax fitting parameter [–] Reinhart C. F. (2001) Daylight Availability and Manual
bmin fitting parameter [–] Lighting Control in Office Buildings — Simulation Studies
and Analysis of Measurements. Ph.D. Thesis, Technical
Drel relative deviation with respect to the predicted
University of Karlsruhe, Germany, ISBN 3-8167-6056-2.
specific annual electric energy demand for Reise C. (2001) Private communication.
artificial lighting [–] Rigollier C., Bauer O. and Wald L. (2000) On the clear sky
s3 average root squared deviation between the model of the ESRA — European Solar Radiation Atlas —
hourly global clearness indices of three sub- with respect to the heliosat method. Solar Energy 68,
sequent hours [–] 33–45.
sk standard deviation of 1-min global clearness SIA (1995) Empfehlung SIA 380 / 4: Elektrische Energie im
indices during 1 h [–] Hochbau. Schweizerischer Ingenieur-und Architekten-Ver-
s k* expectation value of the standard deviation of ein.
1-min global clearness indices during 1 h [–] Skartveit A. and Olseth J. A. (1992) The probability density
and autocorrelation of short-term global and beam ir-
s k* (new) expectation value of the standard deviation of
radiance. Solar Energy 49(6), 477–487.
1-min global clearness indices during 1 h Tregenza P. R. and Waters I. M. (1983) Daylight coefficients.
(according to new parametrization) [–] Light. Res. Technol. 15(2), 65–71.
sk b standard deviation of 1-min beam clearness ¨
Voss K., Reinhart C. F., Lohnert G., Wagner A. (2000)
indices during 1 h [–] Towards lean buildings — examples and experience from a
German demonstration program for energy efficiency and
solar energy use in commercial buildings. Conference
Acknowledgements—We would like to thank Arvid Skartveit Proceedings of the EUROSUN, Kopenhagen, 19–22 June
and Jan Asle Olseth for making the source code of their 2000, CD-ROM.
original model available and to Guy Newsham for providing Ward G. and Shakespeare R. (1998). Rendering with
valuable comments on a draft version of this paper. Many RADIANCE. The Art and Science of Lighting Visualization,
thanks to Christian Reise, Stanislav Darula and Richard Morgan Kaufmann Publishers.