sensors

Article
A “Skylight” Simulator for HWIL Simulation of
Hyperspectral Remote Sensing
Huijie Zhao, Bolun Cui, Guorui Jia * ID
, Xudong Li, Chao Zhang and Xinyang Zhang
School of Instrumentation Science & Opto-electronics Engineering, Key Laboratory of Precision
Opto-Mechatronics Technology, Beihang University, Ministry of Education, 37# Xueyuan Road, Haidian District,
Beijing 100191, China; hjzhao@buaa.edu.cn (H.Z.); boluncui@buaa.edu.cn (B.C.); xdli@buaa.edu.cn (X.L.);
xyliang@buaa.edu.cn (C.Z.); xy_zhang@buaa.edu.cn (X.Z.)
* Correspondence: jiaguorui@buaa.edu.cn; Tel.: +86-010-8231-5884

Received: 9 October 2017; Accepted: 2 December 2017; Published: 6 December 2017

Abstract: Even though digital simulation technology has been widely used in the last two decades,
hardware-in-the-loop (HWIL) simulation is still an indispensable method for spectral uncertainty
research of ground targets. However, previous facilities mainly focus on the simulation of
panchromatic imaging. Therefore, neither the spectral nor the spatial performance is enough for
hyperspectral simulation. To improve the accuracy of illumination simulation, a new dome-like
skylight simulator is designed and developed to fit the spatial distribution and spectral characteristics
of a real skylight for the wavelength from 350 nm to 2500 nm. The simulator’s performance was
tested using a spectroradiometer with different accessories. The spatial uniformity is greater than 0.91.
The spectral mismatch decreases to 1/243 of the spectral mismatch of the Imagery Simulation
Facility (ISF). The spatial distribution of radiance can be adjusted, and the accuracy of the adjustment
is greater than 0.895. The ability of the skylight simulator is also demonstrated by comparing
radiometric quantities measured in the skylight simulator with those in a real skylight in Beijing.

Keywords: HWIL; simulation; skylight; mineral mapping; sensor test; spectral measurement;
remote sensing experiment

1. Introduction
Simulation is an indispensable process in the calibration and validation of instruments or
analysis of algorithms. It is also an important way to analyze the characteristics of a target in
specific circumstances. Over the last two decades, digital simulation has become the major approach
to simulating hyperspectral imaging along the sun–target–observer image chain [1–4]. In the process
of digital simulation, solar irradiance, skylight, and reflected background radiance are considered to
illuminate the target. Then, the radiance to the sensor is calculated as a combination of directly reflected
radiance and upwelled radiance of atmosphere [5–7]. Lastly, the sensor characteristics are described
using common models. This is a good approach to simulating images in all kinds of radiation and
imaging geometric conditions. However, its performance is limited by the knowledge of characteristics
of the target, environment, and sensor.
Although the precision reflectance model of buildings and canopies [8,9] and the mixing model
of minerals (such as the Hapke model [10,11]) have been developed, and the precision model of a
spectrometer can also be specifically built [12], it is still important to measure reflected radiance in
a hardware-in-the-loop (HWIL) simulation when researching the influence of environment, such
as heavy metal toxicity in plants, nonlinear mixing of minerals, or weathered minerals [13–22].
The traditional approaches include conducting flight campaigns over experimental sites [13–15]
and experiments in labs [16–22]. The flight campaign approach has several shortcomings such as

Sensors 2017, 17, 2829; doi:10.3390/s17122829 www.mdpi.com/journal/sensors

Sensors 2017, 17, 2829 2 of 17

the great expense, time consumption, the complexity and difficulty of making simultaneous in-situ
measurements of atmospheric parameters and ground reflective characteristics. The result of using
a flight campaign is also limited by many environmental conditions such as solar zenith, weather,
and visibility.
Therefore, most experiments for researching or instrument testing are accomplished in laboratories
by measuring the spectral reflectance of the target. The common light source illuminating targets are
halogen lamps [16–19] or halogen lamps combined with a collimator or integrating sphere [20–22].
In this context, the indoor experiment faces some challenges. Firstly, the irradiance illuminated
on the target can hardly simulate the geometric characteristics of solar irradiance and skylights.
Secondly, the spectral characteristic of halogen lamps is quite different from that of solar irradiance or
skylights [23].
In order to simulate the remote multispectral imaging process indoors, an HWIL simulation
facility named Imagery Simulation Facility (ISF) was developed by ITEK Optical System (Lexington,
MA, USA) [24], which includes both solar and skylight simulators. However, the spectral characteristic
of the tungsten lamps selected for the skylight simulator was quite different from real skylights.
Therefore, they are respectively filtered with several different filters for the spectral range from 400 nm
to 2500 nm, or replaced by a different kind of lamp with accurate spectral characteristics from 400 nm
to 800 nm for color film imaging simulation [24]; this results in either of the following two problems.
First, the spectral isotropy over the hemisphere is decreased by different filters. Second, the facility
can only work in the visible region. Except for the above problems, the change in skylight irradiance
with different solar zeniths was not reported. Therefore, the ISF facility cannot be used to test a
hyperspectral system or algorithm, especially for studies on mineral identification, which are mostly
based on spectral features in the short-wave infrared (SWIR) region.
In this paper, a hemispherical lamp array with 80 identical lamps is designed as a skylight
simulator for the indoor simulation of hyperspectral remote sensing. A metal-halide lamp is chosen as
the light source in the simulator and designed to point to the center of the array. The spatial distribution
of lamps is modeled and tested considering the slightly different performances among individual
lamps. A series of experiments are conducted to test the performance of the simulator. As the result,
the spatial uniformity of irradiance is greater than 0.91. The spectral mismatch is about 1/243 that
of the ISF. The spatial distribution of radiance can be adjusted, and the accuracy of the adjustment is
greater than 0.895.
Section 2 analyzes the characteristics of the skylight and draws the requisite performance of
the skylight simulator. Section 3 indicates the development of the skylight simulator. In Section 4,
the accuracy of the simulator is tested. Conclusions are drawn in Section 5.

2. Theory
The spectral irradiance onto the surface, Eτλ , can be divided into three parts including direct solar
0 , skylight irradiance E , and background radiance L
irradiance Esλ dλ bλavg [25]:

0
Eτλ = Esλ cos σ0 τ1 (λ) + FEdλ + (1 − F ) Lbλavg π (1)

where F is the fraction of the hemispherical sky that could be seen from the target, σ0 is the zenith of
the target, and τ1 (λ) is the transmission from solar to the scattering volume. As shown in Figure 1 [25],
the downwelled irradiance (skylight) Edλ could be modeled as the integration of the directional
radiance from the scattering of solar light by a small unit volume over the hemisphere area above the
target [25]:
Z 0 τ ( λ ) τ (λ ) β ( λ, θ ) cos σdV
Esλ L1 L2 sca d
Edλ = (2)
r2
where σ is the angle between the normal to the target and the ray from the volume dV, r is the distance
from the target to the volume which will be integrated from the target to the top of atmosphere
Sensors 2017, 17, 2829 3 of 17
Sensors 2017, 17, 2829 3 of 17

atmosphere (TOA), and  L1 (  ) and  L 2 ( ) are the transmissions along the paths L1 and L2 .
(TOA), and τL1 (λ)Eand τL2 (λ) are the transmissions along the paths L1 and L2. This reveals that
that s is attenuated through the sun–volume path L1, scattered in a deflection of

This
0 isreveals
Esλ attenuated through the sun–volume path L1, scattered in a deflection of θd onto the target,
 onto the target, in the fraction of the angular scattering coefficient  sca (d ,  ) , and attenuated
ind the fraction of the angular scattering coefficient β sca (θd , λ), and attenuated again through the
volume–target
again through the L2.
pathvolume–target path L2.

Figure
Figure 1.
1. Contribution
Contribution to
to the
the downwelled
downwelled irradiance from aa unit
irradiance from unit volume.
volume.

Accordingly, the skylight generally keeps three characteristics. (a) The target is illuminated by a
Accordingly, the skylight generally keeps three characteristics. (a) The target is illuminated by
hemispherical skylight and the spatial radiation distribution is influenced by the solar direction. (b)
a hemispherical skylight and the spatial radiation distribution is influenced by the solar direction.
The total irradiance varies with the atmospheric status. (c) The total spectral irradiance arriving at
(b) The total irradiance varies with the atmospheric status. (c) The total spectral irradiance arriving at
the ground is spatially uniform.
the ground is spatially uniform.
In order to achieve a group of standard spectra of irradiance, MODTRAN 4.1 (Spectral Sciences
In order to achieve a group of standard spectra of irradiance, MODTRAN 4.1 (Spectral Sciences
Inc., United States Air Force, USA) was used to calculate the radiance from different directions and
Inc., United States Air Force, USA) was used to calculate the radiance from different directions and
irradiance onto the ground surface for different dates and times [26]. Ground locations were set at
irradiance onto the ground surface for different dates and times [26]. Ground locations were set at
Hami, Xinjiang, China, which is a typical zone with many porphyry copper–gold mineralization
Hami, Xinjiang, China, which is a typical zone with many porphyry copper–gold mineralization
subzones [27]. The parameters used in MODTRAN calculation are listed in Table 1.
subzones [27]. The parameters used in MODTRAN calculation are listed in Table 1.
Table 1. Parameters Used in MODTRAN Simulation.
Table 1. Parameters Used in MODTRAN Simulation.
Attribute Value
Attribute
Day of Year Value 15)
1–365 (interval
Day ofLocal
Year Time 10:00–14:00 (interval
1–365 (interval 15)1 h)
Local Time
Viewing Zenith Angle (°) 10:00–14:00
0–90 (interval 5) 1 h)
(interval
Viewing Zenith Angle (◦ ) 0–90 (interval 5)
Viewing Azimuth Angle (°) 0–360 (interval 5)
Viewing Azimuth Angle (◦ ) 0–360 (interval 5)
Location
Location 40°
40◦N,N,94°94◦EE
Spectral
Spectral RangeRange
(nm) (nm) 400–2500
400–2500
Spectral Response
Spectral Function
Response Function GaussianofofFWHM
Gaussian FWHM 10 10 nm
nm
Spectral Sampling
Spectral Interval
Sampling (nm) (nm)
Interval 1010
Model of atmosphere Sub-Arctic Summer Mid-Latitude Summer
Sub-Arctic Summer
Model of Aerosol
Model of atmosphere Rural
Visibility Mid-Latitude 40 km Summer
Model of Aerosol Rural
Visibility 40 km
As shown in Figure 2, the skylight irradiates the target from every direction and varies with
incident angle, holding the same spectral characteristic. The radiance comes to the maximum in the
As shown in Figure 2, the skylight irradiates the target from every direction and varies with
incident angle, holding the same spectral characteristic. The radiance comes to the maximum in the
Sensors 2017, 17, 2829 4 of 17
Sensors 2017, 17, 2829 4 of 17
solar direction (zenith 17°, azimuth 180°) and the minimum appears in the opposite direction of solar
incidence, as shown in Figures 2a,b. The spectral isotropy of radiance is tested by calculating the
solar direction
spectral (zenith
correlation 17◦ , azimuth
coefficients 180◦ ) and
between the minimum appears
azimuthal-average spectralinradiance
the opposite
withdirection
differentofzenith
solar
incidence, as shown in Figure 2a,b. The spectral isotropy of radiance is tested by calculating
angles and between zenithal-average spectral radiance with different azimuth angles, respectively, the spectral
correlation
as shown coefficients
in Tables 2between
and 3.azimuthal-average spectral radiance
The Pearson correlation withisdifferent
coefficient zenith
calculated angles and
according to
represent the radiances in band i with different
between zenithal-average X m spectral
i X n  i  with different azimuth angles, respectively,
radiance as shown
Equation (3), where and
in Tables 2 and 3. The Pearson correlation coefficient is calculated according to Equation (3), where
zenith/azimuth angles respectively. As the correlation in Tables 2 and 3 is greater than 0.980, the
Xm (i ) and Xn (i ) represent the radiances in band i with different zenith/azimuth angles respectively.
spectral characteristic of radiance can be summarized as stable with the change of zenith and azimuth.
As the correlation in Tables 2 and 3 is greater than 0.980, the spectral characteristic of radiance can be
The radiance at 16:00 shows the same characteristic in Figures 2c,d (the solar zenith is 55° and azimuth
summarized as stable with the change of zenith and azimuth. The radiance at 16:00 shows the same
is 180°). Caused by the change in solar direction, the◦ spatial distribution of the radiance changed. The
characteristic in Figure 2c,d (the solar zenith is 55 and azimuth is 180◦ ). Caused by the change in
correlation of the max radiance, min radiance, and middle one at 12:00 and 16:00 is also calculated,
solar direction, the spatial distribution of the radiance changed. The correlation of the max radiance,
as shown in Table 4. The spectral characteristic at different times can be defined as stable, because
min radiance, and middle one at 12:00 and 16:00 is also calculated, as shown in Table 4. The spectral
the correlation is greater than 0.957.
characteristic at different times can be defined as stable, because the correlation is greater than 0.957.

Corr=

E X i   E  X i   X i   E  X i 
E(( Xmm (i ) − E( Xmm (i )))( Xnn (i ) − E( Xn n (i )))) (3) (3)
Corr
DD(XXmm(ii))
 DD X(Xn ni(i))
p p

(a) (b)

(c) (d)

Figure
Figure 2.
2. The
The radiance
radiance of
of the
the skylight:
skylight: (a)
(a)The
Thespectral
spectral radiance
radiance from
from zenith 20°◦ and
zenith 20 and different
different azimuths
azimuths
at 12:00; (b)
at 12:00; The spectral
(b) The spectral radiance
radiance from
from azimuth 170°◦ and
azimuth 170 and different zeniths at
different zeniths 12:00; (c)
at 12:00; (c) The
The spectral
spectral
radiance
radiance from
from zenith 20◦° and
zenith 20 and different azimuths at
different azimuths at 16:00;
16:00; (d)
(d) The
The spectral
spectral radiance
radiance from
from azimuth 170°◦
azimuth 170
and different zeniths at 16:00.
and different zeniths at 16:00.

Table 2. The correlation between azimuthal-average spectral radiances and different zenith angles.

Zenith 0° 15° 30° 45° 60° 75° 90°

Sensors 2017, 17, 2829 5 of 17

Table 2. The correlation between azimuthal-average spectral radiances and different zenith angles.

Zenith 0◦ 15◦ 30◦ 45◦ 60◦ 75◦ 90◦

0◦ 1.000
15◦ 1.000 1.000
30◦ 0.998 0.999 1.000
45◦ 0.995 0.996 0.999 1.000
60◦ 0.992 0.994 0.997 1.000 1.000
75◦ 0.995 0.996 0.999 1.000 0.999 1.000
90◦ 0.993 0.993 0.990 0.985 0.980 0.985 1.000

Table 3. The correlation between zenithal-average spectral radiances and different azimuth angles.

Azimuth 0◦ 15◦ 30◦ 45◦ 60◦ 75◦ 90◦ 105◦ 120◦ 135◦ 150◦ 165◦
0◦ 1.000
15◦ 1.000 1.000
30◦ 1.000 1.000 1.000
45◦ 1.000 1.000 1.000 1.000
60◦ 1.000 1.000 1.000 1.000 1.000
75◦ 0.999 0.999 1.000 1.000 1.000 1.000
90◦ 0.999 0.999 0.999 0.999 1.000 1.000 1.000
105◦ 0.997 0.997 0.998 0.998 0.999 0.999 1.000 1.000
120◦ 0.996 0.996 0.996 0.997 0.998 0.998 0.999 1.000 1.000
135◦ 0.994 0.994 0.994 0.995 0.996 0.997 0.998 0.999 1.000 1.000
150◦ 0.992 0.992 0.993 0.993 0.995 0.996 0.997 0.999 0.999 1.000 1.000
165◦ 0.991 0.991 0.991 0.992 0.993 0.995 0.996 0.998 0.999 1.000 1.000 1.000

Table 4. The correlation of spectral radiance at 12:00 and 14:00.

12:00
Max Middle Min
Max 0.993 0.977 0.986
14:00 Middle 0.966 0.997 0.957
Min 0.993 0.972 0.991

The stability of the spectral characteristic ensures the probability of adjusting the irradiance of the
skylight simulator. It is obvious that the spatial uniformity of irradiance and spatial distribution of
radiance onto the ground are two very important characteristics of the skylight. When adjusting the
zenith distribution of the radiance, the difference in radiance between each target could be negligible.
However, when adjusting the azimuth distribution of radiance, lamps need to point to different
targets, respectively, to maintain the uniformity of irradiance, which will cause great differences in
radiance on targets. As a result, the simulation of azimuth distribution of radiance is almost impossible.
Therefore, the skylight simulator is supposed to be adjusted to simulate the zenith distribution of
radiance, and maintain the spatial uniformity of spectral irradiance at the same time. The irradiance
and the zenith distribution of the radiance would be similar to the MODTRAN-simulated result shown
in Figure 2.

3. Instrument Description

3.1. Instrument Design

3.1.1. Chosen Light Source

As mentioned above, adjusting the lamps with different filters will decrease the spectral isotropy
of radiance. In addition, it is obvious that combining different kinds of bulbs into one set of reflector
Sensors 2017, 17, 2829 6 of 17

and fore-optics to act as a lamp in the lamp array is nearly impossible, even when not considering
the increase
Sensors in power cost. Therefore, the best way to simulate the spectral characteristics6 of
2017, 17, 2829 of17a
skylight would be to find a light source whose spectral irradiance on the target is most similar to
characteristics
a skylight andofthen fourcalibrate
kinds ofit.lamp were to
In order compared,
choose the i.e.,best
LEDlight
lamp, tungsten
source for thehalogen
skylight lamp, xenon
simulator,
lamp, and metal-halide
the characteristics lamp.
of four kinds of lamp were compared, i.e., LED lamp, tungsten halogen lamp,
xenonThe spectral
lamp, irradiance of lamp.
and metal-halide these lamps, as shown in Figure 3a, was collected in a darkroom by
usingThean spectral
ASD FieldSpec
irradiance pro FR spectroradiometer
of these lamps, as shown in(Analytical Spectral
Figure 3a, was Devices,
collected Inc., Longmont,
in a darkroom by using
Colorado) with a full-sky irradiance Remote Cosine Receptor (RCR)
an ASD FieldSpec pro FR spectroradiometer (Analytical Spectral Devices, Inc., Longmont, Colorado) accessory. Their relative
with a full-sky
deviations fromirradiance
the skylight Remotespectrum band i (RCR)
CosineinReceptor accessory.with
are calculated TheirEquation
relative deviations
(4) and plottedfrom thein
skylight spectrum in band i
Figure 3b. The spectral correlation between lamps and skylight is calculated with Equation spectral
are calculated with Equation (4) and plotted in Figure 3b. The (3) and
correlation
listed in Table between
5. It islamps
obviousandthatskylight is calculated
the spectral irradiancewithofEquation
the LED(3) andonly
lamp listed in Table 5.
distributes in It
theis
obvious that the spectral irradiance of the LED lamp only distributes in the visible
visible region. The relative deviation of the tungsten halogen lamp, which is equipped in ISF, is much region. The relative
deviation
greater thanofthose
the tungsten halogen lamp,
of the metal-halide lamp which is equipped
and xenon lamp. A inxenon
ISF, is lamp
muchisgreater
a kind than
of high those of the
intensity
metal-halide
discharge (HID)lamp andwhich
lamp, xenon is lamp.
usedAasxenon lampprojector
a cinema is a kindlamp,
of high intensity
vehicle lamp, discharge
and solar (HID) lamp,
simulator
which is used as a cinema projector lamp, vehicle lamp, and solar simulator
for photovoltaic cell testing [28–32]. However, there is no compact xenon light source with a wide for photovoltaic cell
testingMost
beam. [28–32]. However,
of the thereavailable
xenon bulbs is no compact
now arexenon
barelight
bulbssource witha areflector
without wide beam. Most of the
or fore-optics. In
xenon bulbs available now are bare bulbs without a reflector or fore-optics.
order to improve the illuminating efficiency, the xenon bulbs must be combined with specific fore- In order to improve the
illuminating
optics efficiency,
and reflectors. the xenon
However, thebulbs
accuracymustofbe combined
assembly andwith specific fore-optics
the isotropy of the shapeand reflectors.
of free-form
However,
surface the accuracy
reflectors of assembly
and Fresnel lenses and the isotropy
are limited. of the the
Therefore, shape of free-form
compact surfacelamp
metal-halide reflectors and
is chosen
Fresnel lenses are limited. Therefore, the compact metal-halide lamp
to be the light source, so that we could select lamps performing uniformly enough from among is chosen to be the light source,
so that weproducts.
industrial could select lamps performing uniformly enough from among industrial products.

(a) (b)

Figure
Figure 3.
3. The
Theradiances
radiancesof
ofdifferent
different lamps
lamps compared
compared with
with skylight
skylight radiance.
radiance. (a)
(a) The
The spectra
spectra of
of lamps
lamps
and (b) the relative deviations of lamps.
and (b) the relative deviations of lamps.
By comparing
By comparing the
the luminous
luminousefficacy,
efficacy, mean
mean lumens,
lumens, and and energy
energy consumption
consumption of
of several
several types
types of
of
lamp produced by PHILIPS (Amsterdam, The Netherlands), OSRAM
lamp produced by PHILIPS (Amsterdam, The Netherlands), OSRAM (Munich, Germany), and GE (Munich, Germany), and GE
(East Cleveland,
(East Cleveland, OH,
OH, USA),
USA), the
the PHILIPS
PHILIPS CDM-RM
CDM-RM Mini Mini 20W/83020W/830GX10
GX10MR16
MR1640D
40D[33]
[33] was
was chosen
chosen
at last.
at last.
X m ( i ) − X0 ( i )
E (i ) = (4)
X m  i   XX00(i )i 
E i   (4)
X 0 i 
Table 5. The correlation between the spectral radiances of lamps and skylight.
Table 5. The correlation between the spectral radiances of lamps and skylight.
Metal-Halide Lamp Xenon Lamp Tungsten Halogen Lamp LED Lamp
Correlation Metal-Halide Lamp Xenon Lamp Tungsten Halogen Lamp LED Lamp
0.57 0.81 0.10 -
Correlation
Coefficient
0.57 0.81 0.10 -
Coefficient
Sensors 2017,
Sensors 17, 2829
2017, 17, 2829 77 of
of 17
17

3.1.2. Optical Model

3.1.2. Optical Model
The skylight simulator needs to simulate spatially uniform spectral irradiance, and can be
The skylight simulator needs to simulate spatially uniform spectral irradiance, and can be adjusted
adjusted to simulate different zenith distributions of radiance, as mentioned in Section 2. In order to
to simulate different zenith distributions of radiance, as mentioned in Section 2. In order to simulate
simulate the spatial uniformity of spectral irradiance, the lamps are designed to be distributed as a
the spatial uniformity of spectral irradiance, the lamps are designed to be distributed as a circle and
circle and orient to the center of the hemisphere. The skylight illuminates the target from different
orient to the center of the hemisphere. The skylight illuminates the target from different directions over
directions over the whole hemisphere, so the skylight simulator is designed as a hemispherical lamp
the whole hemisphere, so the skylight simulator is designed as a hemispherical lamp array. The lamps
array. The lamps at the polar region of the hemisphere are omitted to leave more space for the
at the polar region of the hemisphere are omitted to leave more space for the spectrometer when
spectrometer when performing the imaging simulation. In order to fit the irradiance on the
performing the imaging simulation. In order to fit the irradiance on the illumination region with the
illumination region with the result calculated by MODTRAN, a complicated model was developed
result calculated by MODTRAN, a complicated model was developed with LightTools (Synopsys,
with LightTools (Synopsys, Mountain View, CA, USA) [34], as shown in Figure 4. Table 6 lists parts
Mountain View, CA, USA) [34], as shown in Figure 4. Table 6 lists parts of the parameters of the model.
of the parameters of the model. The optical characteristics of lamps, including the size, the shape of
The optical characteristics of lamps, including the size, the shape of the beam, and the spectral radiant
the beam, and the spectral radiant exitance, were carefully tested with a Luminance Meter and a
exitance, were carefully tested with a Luminance Meter and a Spectrometer. The radius of the skylight
Spectrometer. The radius of the skylight simulator is designed to be less than 3.3 m, considering the
simulator is designed to be less than 3.3 m, considering the space of the lab.
space of the lab.

(a) (b)

Figure
Figure 4.
4. The
The model
model built
built by
by LightTools.
LightTools. (a)
(a) The
The spatial
spatial distribution
distribution of
of lamps
lamps in
in the
the hemispherical
hemispherical
lamp array, and (b) the normalized intensity distributed by the zenith of one lamp.
lamp array, and (b) the normalized intensity distributed by the zenith of one lamp.

Table 6. The parameters of the LightTools model.

Table 6. The parameters of the LightTools model.
Parameters Value
Parameters
Radius of the surface of lamps 25Value
mm
Radius of the surface
Beam angle of lamps 25 mm
30°
Beam angle 30◦
Intensity Normalized intensity distribution of each lamp
Intensity Normalized intensity distribution of each lamp
Exitance
Exitance Exitance distribution
Exitance distributionon
onthe
the surface oflamp
surface of lamp
Spectral
Spectralcharacteristics
characteristics Normalized spectral exitance
Normalized spectral exitance
Lumens
Lumens Lumensofofeach
Lumens each lamp
lamp
Zenith 30 ◦ /45◦ /60◦ /75◦ /90◦
Zenith 30°/45°/60°/75°/90°
Azimuth Every 45◦ /30◦ /30◦ /15◦ /15◦
Azimuth
Radius of lamp array
Every 45°/30°/30°/15°/15°
1650 mm
Radius of lamp array 1650 mm

As the
As thesimulator
simulatorisissupposed
supposedtotoadjust
adjust
thethe zenith
zenith distribution
distribution of radiance,
of radiance, the the distribution
distribution of
of the
the lamps, along with the zenith angle, needs to be carefully designed. As
lamps, along with the zenith angle, needs to be carefully designed. As the reflectance or the the reflectance or the
bidirectional reflectance
bidirectional reflectance distribution
distributionfunction
function(BRDF)
(BRDF)arearedescribed
describedwithwithirradiance—not
irradiance—not radiance—on
radiance—
the target [25], it is better to define the performance of the skylight simulator
on the target [25], it is better to define the performance of the skylight simulator with irradiance. with irradiance.
In
In addition, it is obvious that the irradiance on the ground is much more available
addition, it is obvious that the irradiance on the ground is much more available than radiance in than radiance in aa
specific zenith
specific zenith region.
region. The
The irradiance
irradiance contributed
contributed byby the
the skylight
skylight ininspecific
specific zenith
zenith regions
regions (every 15◦ )
(every 15°)
at different
at differenttimes
timesis is
thethe
integral of the
integral ofradiance calculated
the radiance using MODTRAN,
calculated using MODTRAN,as shownas in Table
shown7. in
Table 7.
Sensors 2017, 17, 2829 8 of 17

Table 7. Components of skylight irradiance from different zeniths on 22 June.

Zenith (◦ )
Irradiance (µW/cm2 ) Total Irradiance (µW/cm2 )
30 45 60 75 90
18.56 3905 3293 2419 1434 191 11,244
25.00 3778 3625 2722 1599 210 11,934
Solar zenith (◦ )
52.62 1733 2786 3463 1975 257 10,214
80.00 668 1055 1422 1487 222 4979

The irradiance on the center of the ground illuminated by the lamps on a specific zenith can be
described by Equation (5), where n is the number of lamps on a specific zenith, Ii is the intensity of
the lamp at solid angle Ωi , θi is the zenith angle, and A is the area on the ground. The irradiance
contributed by the skylight from the zenith region [θi − 0.5∆θ ∼ θi + 0.5∆θ ] is close to the irradiance
illuminated by the lamp array on zenith θi . The number of lamps n can be estimated with Equation (5),
and the data are listed in Table 7. R
n Ii dΩi
Ei = (5)
cos θi dA
At last, the number and distribution of lamps were designed as shown in Table 8 with three
typical modes. Mode 1 is used when the solar zenith is less than 40◦ , Mode 2 is used when the solar
zenith varies from 40◦ to 65◦ , and Mode 3 is designed to simulate skylight when the solar zenith is
greater than 65◦ . Due to the regular interval of lamps, the skylight simulator can be adjusted to fit
many more situations such as an atmosphere with a different extinction coefficient or atmosphere on
different dates.

Table 8. The design of the lamp array with different working modes.

Lamp Status (Irradiance (µW/cm2 ))

Layer Number of Zenith of Azimuth
Mode 1 1 Mode 2 2 Mode 3 3
No. Lamps Lamps (◦ ) Interval (◦ )
(11,047) (9762) (5031)
1 8 60 45 All on 1/2 on 1/4 on
2 12 45 30 All on 3/4 on 1/4 on
3 12 30 30 All on All on 2/3 on
4 24 15 15 3/4 on All on 2/3 on
5 24 0 15 All on All on All on
Relative deviation of irradiance 0.018 0.034 0.036
1 Mode 1: solar zenith 15◦ ~40◦ ; 2 Mode 2: solar zenith 40◦ ~65◦ ; 3 Mode 3: solar zenith 65◦ ~80◦ .

3.2. System Development

In order to prevent the interference of illumination, the lab was built as a darkroom. All of
the devices in the lab were painted matte black, which reduces the illumination reflected from the
environment as much as possible. The brace of the lamp array was manufactured with aluminum
profiles to give high rigidity. The curvature accuracy of the round components was carefully adjusted
to be greater than 0.999. Each lamp with specific electronic ballast was measured as mentioned above
and fixed according to the model designed by LightTools, as shown in Figure 5. All of the lamps were
divided into several groups with RS-485 bus controllers connected separately. Each lamp could be
controlled with an outside computer.
Sensors 2017, 17, 2829 9 of 17
Sensors 2017, 17, 2829 9 of 17

Figure
Figure 5.
5. The
The working
working status
status of
of the
the skylight
skylight simulator.
simulator.

4. Performance Validation
4. Performance Validation
In order to test the performance of the skylight simulator, a series of experiments were
In order to test the performance of the skylight simulator, a series of experiments were performed.
performed. The essential performance indicators, including the spatial uniformity, the spectral
The essential performance indicators, including the spatial uniformity, the spectral mismatch, and the
mismatch, and the accuracy of adjustment, were tested indoors. The experimental performance
accuracy of adjustment, were tested indoors. The experimental performance including the agreement of
including the agreement of irradiance, the spatial distribution of radiance, the spectral characteristic,
irradiance, the spatial distribution of radiance, the spectral characteristic, and the collected reflectance
and the collected reflectance were tested by comparing the radiometric quantities collected in the
were tested by comparing the radiometric quantities collected in the field with those collected in
field with those collected in the simulator.
the simulator.
4.1. Essential
4.1. Essential Performance
Performance
In order
In orderto
totest
testthe
thespatial
spatial uniformity
uniformity of the
of the skylight
skylight simulator
simulator in each
in each band,band, the illuminated
the illuminated plane
(50 cm × 50 cm at the center of simulator) was divided into a 10 × 10 grid. An ASDASD
plane (50 cm × 50 cm at the center of simulator) was divided into a 10 × 10 grid. An Fieldspec
Fieldspec Pro
Pro spectrometer equipped with an RCR accessory was used to collect the irradiance E
spectrometer equipped with an RCR accessory was used to collect the irradiance E(λ) on each node.   on each
node.
The The uniformity
spatial spatial uniformity
was thenwas then calculated
calculated with Equationwith(6) Equation
according (6)toaccording to thestandard
the IEC 60904-9 IEC 60904-9
[35]:
standard [35]:
Emax (λ) − E (λ)
U (λ) = 1 − E max     Emin min   
. (6)
U     1-Emax (λ) + Emin (λ) . (6)
E max     E min   
The results are shown in Figure 6a. The measured irradiance—and, hence, the uniformity—from
2200 The results
nm to 2500 are
nmshown in Figure
are invalid, 6a. The
because themeasured
transmittance irradiance—and, hence, the
of the RCR decreases uniformity—from
substantially in this
2200 nm to 2500 nm are invalid, because the transmittance of the RCR decreases substantially
spectral range, which is revealed by the signal-to-noise ratio (SNR) of the collected irradiance calculated in this
spectral range, which is revealed by the signal-to-noise ratio (SNR) of the collected irradiance
with Equation (7) and shown in Figure 6b. Generally, the uniformity is greater than 0.88 from 350 nm
calculated with Equation (7) and shown in Figure 6b. Generally, the uniformity is greater than 0.88
to 2200 nm. Moreover, the uniformity is greater than 0.91 in most bands. In a word, the performance
from 350 nm to 2200 nm. Moreover, the uniformity is greater than 0.91 in most bands. In a word, the
of the skylight simulator is good enough for simulating the spatial uniformity of spectral irradiance, as
performance of the skylight simulator is good enough for simulating the spatial uniformity of spectral
mentioned in Section 2.
irradiance, as mentioned in Section 2. ∑ E(λ)
n
n
u  E  ∑E(λ) !2
SNR(λ) = v (7)
u n
t ∑ E(λ)− n nn
SNR     n

 E  
2
 n
 (7)
 E    n 
n  n 
 
n
Sensors 2017, 17, 2829 10 of 17
Sensors 2017, 17, 2829 10 of 17
Sensors 2017, 17, 2829 10 of 17

(a) (b)
(a) (b)
Figure 6.
Figure 6. The
The spectral
spectral non-uniformity
non-uniformity of
of the
the skylight
skylight simulator:
simulator: (a)(a) The
The spectral
spectral non-uniformity
non-uniformity ofof the
the
Figure 6. The spectral
skylight simulator
simulator in non-uniformity
in 350–2500
350–2500 nm of the
nm spectral skylight
spectral region simulator:
region and
and (b)
(b) the (a) The spectral
the signal-to-noise non-uniformity
signal-to-noise ratio
ratio (SNR)
(SNR) of of the
of spectral
spectral
skylight
skylight simulator
irradiance inwith
collected 350–2500
an nmfield
ASD spectral region and (b) the signal-to-noise ratio (SNR) of spectral
spectrometer.
irradiance collected with an ASD field spectrometer.
irradiance collected with an ASD field spectrometer.
The spectral isotropy over the hemisphere of the skylight simulator was tested by collecting the
The spectral
The spectral isotropy
isotropy over
over the
the hemisphere of of the skylight simulator was was tested
tested by by collecting the the
radiance reflected by a target with hemisphere
specific zenith the skylight
angle facingsimulator
different directions. Thecollecting
zenith of the
radiance
radiance reflected
reflected by a target with specific zenith angle facing different directions. The zenith of the
target was 30°, andby
◦ , and
a target
the azimuth with specific
of it was setzenith anglefrom
to change facing0° different
to 180°
◦ to indirections.
◦ in intervals ofThe 5°. zenith of the
The radiance
◦ . The
target was
target was 30
30°, and the azimuth
the azimuth of of it
it was
was set
set to
to change
change from
from 00° to 180
180° in intervals
intervals of
of 55°. The radiance
radiance
is shown in Figure 7 (part of data), and the correlation between them was calculated to be greater
is shown
is shown in Figure
in Figure 7 (part of
of data), and the correlation between them was calculated to be
be greater
greater
than 0.9995. Because7 only
(partone data),
kind ofand lampthewas
correlation
used, thebetween
spectral them
mismatchwas calculated
of the skylight to simulator
than
than 0.9995.
0.9995. Because only one kind of lamp was used, the spectral mismatch of the skylight simulator
is similar to Because only one of
the performance kind of lamp
a single wasshown
lamp used, intheFigure
spectral mismatch
3, even if theof the skylight
spectral simulator
characteristic of
is similar
is similar to
to the
the performance
performance of of aa single
single lamp
lamp shown
shown in in Figure
Figure 3,3, even
even ifif the
the spectral
spectral characteristic
characteristic of of
each lamp is not exactly the same.
each lamp
each lamp is is not exactly
exactly the
notspectral the same.
same. performance indicates that the facility can be adjusted without
This high isotropy
This high
This high spectral
spectral isotropy
isotropy performance
performance indicates that
that the
the facility
facility can be
be adjusted without
decreasing spectral isotropy, which is defined asindicates
the first characteristic of acan
skylight adjusted
in Section without
2.
decreasing spectral
decreasing spectral isotropy,
isotropy, which
which is is defined
definedas asthe
thefirst
firstcharacteristic
characteristicof ofaaskylight
skylightin inSection
Section2.2.

Figure 7. The spectral radiance reflected by the standard target.

Figure
Figure 7.
7. The
The spectral
spectral radiance
radiance reflected
reflected by
by the
the standard target.
standard target.
According to Equation (5), the irradiance contributed by different zenith regions can be
According
calculated usingtoto
theEquation
irradiance (5), the irradiance
of lamps at different contributed
zeniths. by different
Therefore, zenith regions
the accuracy of be can the
adjusting be
According Equation (5), the irradiance contributed by different zenith regions can calculated
calculated
irradiance using the
contributionirradiance
to the of lamps at
simulated different
skylightzeniths. Therefore,
at three the accuracy
differentof times of
of theadjusting
the irradiancethe
day, i.e.,
using the irradiance of lamps at different zeniths. Therefore, the accuracy adjusting
irradiance
correspondingcontribution to the simulated skylight at three different times of the day, i.e.,
contribution to to
thethree different
simulated solar zenith
skylight at three angles, cantimes
different be measured with
of the day, a spectrometer,
i.e., correspondingastoshown three
corresponding
in Figuresolar to three
8. Comparing different solar
the irradiance zenith
in Tableangles,
8, the can be measured
relative deviation with a spectrometer,
of adjusting was as shown
calculatedthe to
different zenith angles, can be measured with a spectrometer, as shown in Figure 8. Comparing
in Figure 8. Comparing the irradiance in Table X 8, the relative deviation of adjusting X was calculated to
be less thanin0.105
irradiance Table with Equation
9, the relative(4), where Xof
deviation m adjusting
is the simulated irradiancetoand
was calculated 0 than
be less is the0.105
irradiance
with
be less than 0.105 with Equation (4), where m is the simulated irradiance and X 0 is the irradiance
calculated
Equation with
(4), where Xm is the simulated
MODTRAN. The spatial irradiance and Xaccuracy
distribution of radiance
0 is the irradiance was evaluated
calculated with the
with MODTRAN.
calculated
RMSRE (root with MODTRAN.
mean The spatial
square relative error) distribution
of irradianceaccuracy of radiance
in different zeniths. ThewasRMSRE
evaluated with the
is calculated
RMSRE (root mean square relative error) of irradiance in different zeniths. The RMSRE is calculated
Sensors 2017, 17, 2829 11 of 17
Sensors 2017, 17, 2829 11 of 17

The spatial distribution accuracy of radiance was evaluated with the RMSRE (root mean square relative
X
using iEquation
X
using of
error) Equation
irradiance differenti zeniths.
(8),inwhere is the simulated
The RMSRE irradiance in zenith
is calculated and i(8),iswhere
the irradiance
Xi is the
calculated with MODTRAN. The 0
results are about 0.189 in Mode 1, 0.259 in Mode
simulated irradiance in zenith i and Xi is the irradiance calculated with MODTRAN. The results 2, and 0.198are
in
Mode 0.189
about 3. Therefore,
in Mode 1, the0.259
irradiance
in Modeon 2, the
and ground
0.198 in and
Modethe spatial radiance
3. Therefore, can be on
the irradiance adjusted with
the ground
accuracy
and greaterradiance
the spatial than 0.895 canand
be 0.741,
adjustedrespectively. The first
with accuracy and than
greater second characteristics
0.895 and 0.741, described in
respectively.
Section
The first2and
can second
therefore be simulateddescribed
characteristics by the facility.
in Section 2 can therefore be simulated by the facility.
2
u nn  2
v
X  XX 
0

Xii −
u ∑  X0 i i 
u
t i=1   (8)
RMSRE = i 1

i
Xi  (8)
RMSRE  n
n

Figure 8. The
Figure 8. The spectral
spectral irradiance
irradiance contributed
contributed by
by different zenith regions.
different zenith regions.

Table 8. The irradiance of different modes.

Table 9. The irradiance of different modes.
Zenith (°)
Irradiance (μW/cm2) ◦ Total Irradiance (μW/cm2)
Irradiance 30
(µW/cm2 ) 45 Zenith
60 ( ) 75 90 Total Irradiance (µW/cm2 )
Mode 1 MODTRAN 3905 30 45
3293 60
2419 75
1434 90191 11,244
Mode
Mode11simulated
MODTRAN 29973905 3653 2465 1434
3293 2419 1727 191
140 10,982
11,244
ModeMode 1 simulated
2 MODTRAN 2997
1733 3653 2465
2786 3463 1727
1975 140
257 10,982
10,214
Mode 2 MODTRAN 1733 2786 3463 1975 257 10,214
Mode 2 simulated 1499 2740 2465 2302 140 9146
Mode 2 simulated 1499 2740 2465 2302 140 9146
Mode
Mode 3 MODTRAN
3 MODTRAN 668 668 1055 1422 1487
1055 1422 1487 222 222 4854
4854
Mode
Mode3 simulated
3 simulated 749
749 913 1642 1535
913 1642 1535 140 140 4979
4979

4.2. Experimental
4.2. Performance
Experimental Performance
In order
In order to
to validate
validate the
theability
abilityofofthe
theskylight
skylightsimulator,
simulator,anan experiment
experiment was
was performed
performed on on
24
August 2017 in Beijing, as shown in Figure 9. The skylight simulator was adjusted
24 August 2017 in Beijing, as shown in Figure 9. The skylight simulator was adjusted to simulate the to simulate the
irradiance when
irradiance when thethe solar
solar zenith
zenith angle
angle is
is 28.9
28.9°.
◦ . The
The total
total irradiances
irradiances collected both in
collected both in the
the field
field and
and in
in
the skylight simulator were compared to test the agreement at the irradiance level. The
the skylight simulator were compared to test the agreement at the irradiance level. The total radiance total radiance
reflected by
reflected by the
the target
target both
both inin the
the field
field and
and the
the simulator
simulator were
were also
also compared
compared using
using the
the metric
metric RMSRE
RMSRE
to test the agreement at the spectral characteristic level. The accuracy of the experiment
to test the agreement at the spectral characteristic level. The accuracy of the experiment was tested was tested
by
by comparing the reflectance of the target collected in the field and
comparing the reflectance of the target collected in the field and in the simulator.in the simulator.
 Sensors
Sensors 2017,2017, 17, 2829
17, 2829 12 of 17
12 of 17
Sensors 2017, 17, 2829 12 of 17

Figure 9. The flow chart of the experiment. RMSRE: root mean square relative error.
Figure 9. The flow chart of the experiment. RMSRE: root mean square relative error.
Figure 9. The flow chart of the experiment. RMSRE: root mean square relative error.
Beijing is dominated by a warm, temperate, continental monsoon climate [36]. The MODTRAN
Beijing
atmosphere
Beijing ismodel
dominated
is dominated byby
is defined aas
warm,
a warm, temperate,
mid-latitude
temperate, continental
summer monsoon
[26]. The
continental climate
cloud cover
monsoon [36].
is less
climate TheThe
than
[36]. MODTRAN
15%, with the
MODTRAN
atmosphere
solar zenith
atmosphere model
model is
at 28.9°. defined
A scale
is defined as mid-latitude
asmodel summer
of the Cuprite
mid-latitude summer [26].
mineral The cloud
area cloud
[26]. The cover
in China is less than
andisa less
cover 15%,
geographical with
than 15%,modelthe the
with
solar zenith at 28.9 ◦ . A scale model of the Cuprite mineral area in China and a geographical model
solar including
zenith at several
28.9°. Aslopes with different
scale model slope angles
of the Cuprite mineralarearea
placed on theand
in China ground with skylight
a geographical model
including several
illumination. slopes
The directwith different
solar slopeisangles
radiance are placed
blocked on theplastic
by a black groundfiber
with skylight
holder, asillumination.
shown in
including several slopes with different slope angles are placed on the ground with skylight
The direct
Figure 10.solar radiance is blocked by a black plastic fiber holder, as shown in Figure 10.
illumination. The direct solar radiance is blocked by a black plastic fiber holder, as shown in
Figure 10.

(a) (b)

Figure10.
Figure 10.Photographs
Photographsof ofthe
theexperiment:
experiment:(a)
(a)collecting
collectingthe
thereflected
reflectedradiance
radianceof
ofthe
thetarget
targetin
inthe
thefield
field
and (b) collecting the reflected
and (b) collecting the reflected radiance in the skylight simulator.
radiance in the skylight simulator.
(a) (b)

Figure The
10.irradiance
Photographs of the realexperiment:
of the skylight was(a)collected
collectingusing an ASD radiance
the reflected FieldspecofPro
thespectrometer with
target in the field
The irradiance of the real skylight was collected using an ASD Fieldspec Pro spectrometer with an
an RCR accessory as shown in Figure 11, with the total
and (b) collecting the reflected radiance in the skylight simulator. irradiance measured as 14,570 μW/cm 2. Since
RCR accessory as shown in Figure 11, with the total irradiance measured as 14,570 µW/cm2 . Since the
the radiance of the simulator at zenith◦ 0–30° is removed, the irradiance collected in the field was
radiance of the simulator at zenith 0–30 is removed, the irradiance collected in the field was scaled
scaled to 11,214 μW/cm 2 according to the irradiance simulated by MODTRAN. The relative deviation
to 11,214 µW/cm accordingskylight
The irradiance of
2 the real was collected
to the irradiance using
simulated byan ASD Fieldspec
MODTRAN. Pro spectrometer
The relative deviation ofwith
an RCRof the irradiance
accessory as from
shown theinskylight
Figure simulator
11, with compared
the total with the scaled
irradiance field-collected
measured as 14,570 irradiance
μW/cm 2. is
Since
the irradiance from the skylight simulator compared with the scaled field-collected irradiance is less
less than 0.021. The irradiance from the simulator in 350–600 nm is less than that collected in the field
the radiance
than 0.021.ofThetheirradiance
simulator at the
from zenith 0–30°inis350–600
simulator removed, nm isthe
lessirradiance collected
than that collected in in
thethe
fieldfield
as a was
as a result of the limitation of the light source, as shown in Figure 11.
scaled to 11,214
result μW/cm according
of the limitation 2 of the lightto the irradiance
source, as shown in simulated
Figure 11.by MODTRAN. The relative deviation
of the irradiance from the skylight simulator compared with the scaled field-collected irradiance is
less than 0.021. The irradiance from the simulator in 350–600 nm is less than that collected in the field
as a result of the limitation of the light source, as shown in Figure 11.
 Sensors 2017, 17, 2829 13 of 17
Sensors
Sensors 2017,
2017, 17,
17, 2829
2829 13
13 of
of 17
17

Figure11.
Figure11. The irradiance of
of field-collected and simulated skylight (four peaks in
in the irradiance of
of the
FigureThe irradiance
11. The irradiance field-collected
of field-collectedand
andsimulated skylight(four
simulated skylight (four peaks
peaks in thethe irradiance
irradiance of the the
simulator
simulator have
have
simulator been
been
have removed
removed
been removed as bad
asasbad bands).
badbands).
bands).

A
A scaled
scaled model
model
A scaled
with
with
model with
different
different targets
differenttargets
was
targets was
used in
used in
was used
the
inthe experiments.
theexperiments.
experiments. The
The
The
targets
targets
targets
on the
the model
on model
on the model
were made
werewere
made from
from
made several
several
from kinds
kinds
several kindsof rock-forming
ofof
rock-forming minerals
rock-formingminerals such
suchas
minerals such as feldspar,
asfeldspar,
feldspar, pyroxene,
pyroxene,
pyroxene, chlorite,
chlorite,
chlorite, epidote,
epidote,
epidote,
muscovite,
muscovite, olivine,
olivine,
muscovite, crystal,
crystal,
olivine, kaolinite,
kaolinite,calcite,
kaolinite,
crystal, calcite, serpentine,
calcite, serpentine,
serpentine, andand
andsoso on.
soon. The
on.The
The radiance
radiance
radiance reflected
reflected
reflected by
by the
by the the
targets
targets and
targets
andandthe reflectance
thethe of
reflectanceof
reflectance them
ofthem were
them were collected
werecollected
collected using
using a spectrometer
a spectrometer
using with
with an
a spectrometer an
FR 8an
with FR
DEG 8 DEG Field-of-
FRField-of-View
8 DEG Field-of-
View Lens
Lens Fore
Fore optic
optic [23][23]
as as shown
shown in in Figures
Figures 12 and 12 and
13, 13, respectively.
respectively.
View Lens Fore optic [23] as shown in Figures 12 and 13, respectively.

Figure
Figure12.
Figure 12.The
12. Theradiance
The radiance reflected by
reflectedby
radiance reflected targets.
bytargets.
targets.
 Sensors 2017, 17, 2829 14 of 17
Sensors 2017, 17, 2829 14 of 17

Sensors 2017, 17, 2829 14 of 17

Figure 13. The reflectance collected in the field and in the skylight simulator.
Figure 13.13.
Figure TheThe
reflectance
reflectancecollected
collected in
in the fieldand
the field andininthe
the skylight
skylight simulator.
simulator.
The radiance reflected by targets is related to the spatial distribution of the skylight radiance.
Therelative
The radiance
The radiancereflected
reflected
deviation byby
of the targets
targets
total isisrelated
radiance related to the spatial
to the
was calculated spatial distribution
withdistribution
Equation (4)oftoofbethe
the lessskylight
skylight
thanradiance.radiance.
0.184, as
The relative deviation of the total radiance
of the Xtotal was calculated with Equation (4) to be less than 0.184, as
0.184,
The
shown relative deviation
in Figure 14, where m is radiance
the radiance was collected
calculatedinwith Equation (4)
the simulator andto Xbe 0 less
is the than
radiance
as shown inthe
Figure 14, whereX m Xm is the radiance collected in the simulator and X0 Xis0 the radiance
shown in Figure
collected in 14, where
field. The RMSRE is the
of theradiance
spectral collected
radiance was in the simulator
calculated using andEquationis(8), thewhere
radiance
collected in the field. The RMSRE of the spectral
collected in the field. The RMSRE of the spectral radiance was radiance was calculated
calculated

using using Equation (8), Xwhere
Equation (8), where i
X
is itheis radiance in band
the radiance in band i collected
i collected in the simulator and Xi is
in the simulator
0
and
X
the radiance
i band i collected
is theinradiance in band in i
X i the field.radiance
collected The
in theRMSRE
field. varies i collected
from 0.613
variestofrom
0.744, as simulator
shown in as
Figure 14. Accounting
Xiin for the spectral i
is the in The
band RMSRE in 0.613
the to 0.744, shown
and Figure
is the 14. Accounting
radiance for
in band
mismatch
the spectral of radiance in the 460–510 nm460–510
and 600–660 nm600–660
regions,nm theregions,
RMSREthe is still high.is However,
collected in themismatch
field. TheofRMSRE
radiance in the
varies from 0.613nm and
to 0.744, as shown in Figure RMSRE still high. for
14. Accounting
the spectral mismatch can be calibrated with a spectral coefficient later for application
However, the spectral mismatch can be calibrated with a spectral coefficient later for application to to hyperspectral
the spectral mismatch of radiance in the 460–510 nm and 600–660 nm regions, the RMSRE is still high.
remote sensingremote
hyperspectral as the sensing
deviation as of
themost of the of
deviation bands
mostisoflow.
the The
bandsRMSRE
is low.ofThe radiance
RMSRE simulated
of radiance by
However,
halogen the spectral
tungsten mismatch
lamps can be
is also calculated calibrated with
to becalculated
170.7, which a spectral
is 170.7, coefficient
much higher compared later for application
with that of the to
simulated by halogen tungsten lamps is also to be which is much higher compared
hyperspectral
skylight remote
with thatsimulator.
sensing
The accuracy
of the skylight
as the
simulator.
deviation
of The
the radiance of most
accuracysimulation
of the bands
is improved
of the radiance simulation
is low. The
significantly
is improved
RMSRE
by choosing of radiance
significantlythe
simulated
lamps by
to behalogen
spectral tungsten
matched. lamps
by choosing the lamps to be spectral matched. is also calculated to be 170.7, which is much higher compared
with that of the skylight simulator. The accuracy of the radiance simulation is improved significantly
by choosing the lamps to be spectral matched.

Figure 14.
Figure 14. The comparison of
The comparison of radiance.
radiance.

As the information of the target is expressed as reflectance, the spectra of reflectance collected in
different situations are compared in Figure 13. The RMSRE and the correlation of reflectance are
Figure 14. The comparison of radiance.
shown in Figure 15. The RMSRE of reflectance is calculated with Equation (8), where X i is the
As the information of the target is expressed as reflectance, the spectra of reflectance collected in
different situations are compared in Figure 13. The RMSRE and the correlation of reflectance are
Sensors 2017, 17, 2829 15 of 17

SensorsAs
2017,
the17,information
2829 of the target is expressed as reflectance, the spectra of reflectance collected 15 of in
17
different situations are compared in Figure 13. The RMSRE and the correlation of reflectance are shown

reflectance in band i collected in the simulator

in Figure 15. The RMSRE of reflectance is calculated with X  Equation (8), where Xi is the reflectance in
and i is the reflectance in band i collected in
0
band i collected in the simulator and Xi is the reflectance in band i collected in the field, and is less
the field, and is less than 0.233. The correlation of reflectance is calculated with
than 0.233. The correlation of reflectance is calculated with Equation (3), where Xm (i ) is the reflectance
Equation
in band i collected
X m (i )
(3), wherein the simulator
is the reflectance
and Xn (i )in
is band
i collected
the reflectance in theisimulator
in band and
collected in
X n (i )
the field, is the
and is
greater than 0.966.
reflectance in band i collected in the field, and is greater than 0.966.

Figure
Figure 15.
15. The
The comparison
comparison of
of reflectance.
reflectance.

5. Conclusions
5. Conclusions
The skylight simulator is an important facility in the HWIL simulation of remote sensing,
The skylight simulator is an important facility in the HWIL simulation of remote sensing,
especially for hyperspectral remote sensing. The previous facilities, including the ISF developed by
especially for hyperspectral remote sensing. The previous facilities, including the ISF developed
ITEK Optical System, are not applicable in a hyperspectral remote sensing experiment. The two main
by ITEK Optical System, are not applicable in a hyperspectral remote sensing experiment. The two
reasons for this are the mismatch of spectral characteristics and lack of ability of adjustment for the
main reasons for this are the mismatch of spectral characteristics and lack of ability of adjustment
simulation of different scenes, because of the limitations of the light source and structure. A new
for the simulation of different scenes, because of the limitations of the light source and structure.
facility was designed and developed with a new wide-beam metal-halide lamp and hemisphere
A new facility was designed and developed with a new wide-beam metal-halide lamp and hemisphere
structure to spatially and spectrally simulate skylight illumination.
structure to spatially and spectrally simulate skylight illumination.
The performance of the proposed skylight simulator was tested using a spectrometer with
The performance of the proposed skylight simulator was tested using a spectrometer with
different accessories in different situations. The spatial uniformity of spectral irradiance is greater
different accessories in different situations. The spatial uniformity of spectral irradiance is greater
than 0.91 in almost all bands from 350–2200 nm. The spectral match of the simulator with the real
than 0.91 in almost all bands from 350–2200 nm. The spectral match of the simulator with the real
skylight improves by about 243 times in the visible and near infrared (VNIR) and SWIR regions
skylight improves by about 243 times in the visible and near infrared (VNIR) and SWIR regions
compared with a halogen-lamp-based system such as ISF. The spectral isotropy over the hemisphere
compared with a halogen-lamp-based system such as ISF. The spectral isotropy over the hemisphere is
is greater than 0.9995. The accuracy of adjusting the irradiance of the simulator is greater than 0.895
greater than 0.9995. The accuracy of adjusting the irradiance of the simulator is greater than 0.895 for
for three modes. The accuracy of irradiance distribution along different zeniths is greater than 0.741.
three modes. The accuracy of irradiance distribution along different zeniths is greater than 0.741.
To validate the spectral performance of the simulation, the spectra of irradiance, radiance
To validate the spectral performance of the simulation, the spectra of irradiance, radiance reflected
reflected by targets, and reflectance of targets were collected with the skylight simulator and
by targets, and reflectance of targets were collected with the skylight simulator and compared with
compared with those collected in the field. The relative deviation of irradiance is less than 0.021. The
those collected in the field. The relative deviation of irradiance is less than 0.021. The RMSRE of
RMSRE of spectral radiance is 0.6–0.7 with the relative deviation of radiance less than 0.184. The
spectral radiance is 0.6–0.7 with the relative deviation of radiance less than 0.184. The RMSRE of
RMSRE of spectral reflectance is less than 0.233, and the correlation is greater than 0.966. The spectral
spectral reflectance is less than 0.233, and the correlation is greater than 0.966. The spectral isotropy
isotropy over the hemisphere is greater than 0.997.
over the hemisphere is greater than 0.997.
All the above spectral accuracies in the 350–2500 nm region are about 243 times those of the latest
All the above spectral accuracies in the 350–2500 nm region are about 243 times those of the
facility. Simultaneously, the isotropy of the spectral characteristic is ensured, and the adjusting
latest facility. Simultaneously, the isotropy of the spectral characteristic is ensured, and the adjusting
accuracies of irradiance and radiance are greater than 0.895 and 0.741, respectively.
accuracies of irradiance and radiance are greater than 0.895 and 0.741, respectively.
Future work will be focused on developing new light sources working in 350–2500 nm to
improve the accuracy of the spectral simulation, and, in particular, to decrease the mismatch of the
spectral characteristic in the 350–600 nm region.
Sensors 2017, 17, 2829 16 of 17

Future work will be focused on developing new light sources working in 350–2500 nm to improve
the accuracy of the spectral simulation, and, in particular, to decrease the mismatch of the spectral
characteristic in the 350–600 nm region.

Acknowledgments: This work was supported by the China Geological Survey under Grants 12120115040701
and DD2016006816, the Chinese Major Project of High Resolution Earth Observation System under
Grants 50-Y20A38-0509-15/16, 04-Y20A35-9001-15/17 and 04-Y20A36-9001-15/17. The authors would like
to thank Wang Runsheng, Gan Fuping, Yan Bokun and Liang Shuneng in China Aero-Geophysical Survey and
Remote Sensing Center for Land and Resources for discussing the use of the skylight simulator.
Author Contributions: Z.H. and C.B. conceived and designed the facility; C.B. and J.G. performed the theoretical
analysis of the skylight; C.B. produced the facility; C.B. and J.G. analyzed the data; C.B. and J.G. wrote the paper.
Conflicts of Interest: The authors declare no conflict of interest. The founding sponsors had no role in the design
of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, and in the
decision to publish the results.

References
1. Kolb, K.E.; Choi, H.S.; Kaur, B.; Olson, J.T.; Hill, C.F.; Hutchinson, J.A. Digital imaging and remote sensing
image generator (DIRSIG) as applied to NVESD sensor performance modeling. Proc. SPIE 2016, 9820.
[CrossRef]
2. Steinberg, A.; Chabrillat, S.; Stevens, A.; Segl, K.; Foerster, S. Prediction of common surface soil properties
based on Vis-NIR airborne and simulated EnMAP imaging spectroscopy data: Prediction accuracy and
influence of spatial resolution. Remote Sens. 2016, 8, 613. [CrossRef]
3. Kim, S.; Kim, S. Target size prediction and verification by geometrical analysis and SE-WORKBENCH for
ground target detection. In Proceedings of the 2014 14th International Conference on Control, Automation
and Systems, Seoul, Korea, 22–25 October 2014.
4. Boukabara, S.; Moradi, I.; Atlas, R.; Casey, S.P.F.; Cucurull, L.; Hoffman, R.N.; Ide, K.; Kumar, V.K.; Li, R.;
Li, Z. Community global observing system simulation experiment (OSSE) package (CGOP): Description
and usage. J. Atmos. Ocean. Technol. 2016, 33, 1759–1777. [CrossRef]
5. Zhao, H.; Jiang, C.; Jia, G.; Tao, D. Simulation of Hyperspectral Radiance Images with Quantification of
Adjacency Effects over Rugged Scenes. Meas. Sci. Technol. 2013, 24, 5405. [CrossRef]
6. Jia, G.; Zhao, H.; Li, N. Simulation of hyperspectral scene with full adjacency effect. In Proceedings of the
International Geoscience and Remote Sensing Symposium (IGARSS) 2008, Boston, MA, USA, 7–11 July 2008;
Volume 3, pp. 724–727.
7. Karl, S.; Luis, G.; Christian, R.; Theres, K.; Sigrid, R.; Hermann, K.; Bernhard, S.; Valery, M.; Stefan, H.
EeteS-The EnMAP End-to-End Simulation Tool. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2012, 5,
522–530. [CrossRef]
8. Widlowski, J.; Mio, C.; Disney, M.; Adams, J.; Andredakis, I.; Atzberger, C.; Brennan, J.; Busetto, L.; Chelle, M.;
Ceccherini, G.; et al. The fourth phase of the radiative transfer model intercomparison (RAMI) exercise:
Actual canopy scenarios and conformity testing. Remote Sens. Environ. 2015, 169, 418–437. [CrossRef]
9. Kuester, T.; Spengler, D.; Barczi, J.; Segl, K.; Hostert, P.; Kaufmann, H. Simulation of Multitemporal and
Hyperspectral Vegetation Canopy Bidirectional Reflectance Using Detailed Virtual 3-D Canopy Models.
IEEE Trans. Geosci. Remote Sens. 2014, 52, 2096–2108. [CrossRef]
10. Ciarniello, M.; Capaccioni, F.; Filacchione, G.; Clark, R.N.; Cruikshank, D.P.; Cerroni, P.; Coradini, A.;
Brown, R.H.; Buratti, B.J.; Tosi, F.; et al. Hapke modeling of Rhea surface properties through Cassini-VIMS
spectra. Icarus 2011, 214, 541–555. [CrossRef]
11. Deb, S.; Sen, A.K. Laboratory light scattering from regolith surface and simulation of data by Hapke model.
Planet. Space Sci. 2016, 124, 36–47. [CrossRef]
12. Tao, D.; Jia, G.; Yuan, Y.; Zhao, H. A digital sensor simulator of the pushbroom Offner hyperspectral imaging
spectrometer. Sensors 2014, 14, 23822–23842. [CrossRef] [PubMed]
13. Cui, J.; Yan, B.; Wang, R.; Tian, F.; Zhao, Y.; Liu, D.; Yang, S.; Shen, W. Regional-scale mineral mapping
using ASTER VNIR/SWIR data and validation of reflectance and mineral map products using airborne
hyperspectral CASI/SASI data. Int. J. Appl. Earth Obs. Geoinf. 2014, 33, 127–141. [CrossRef]
Sensors 2017, 17, 2829 17 of 17

14. Altmann, Y.; Dobigeon, N.; Tourneret, J. Nonlinearity detection in hyperspectral images using a polynomial
post-nonlinear mixing model. IEEE Trans. Image Process. 2013, 22, 1267–1276. [CrossRef] [PubMed]
15. Schaepman, M.E.; Jehle, M.; Hueni, A.; D’Odorico, P.; Damm, A.; Weyermann, J.; Schneider, D.F.; Laurent, V.;
Popp, C.; Seidel, F.C. Advanced radiometry measurements and earth science applications with the airborne
prism experiment (APEX). Remote Sens. Environ. 2015, 158, 207–219. [CrossRef]
16. Robertson, K.M.; Milliken, R.E.; Li, S. Estimating mineral abundances of clay and gypsum mixtures using
radiative transfer models applied to visible-near infrared reflectance spectra. Icarus 2016, 277, 171–186.
[CrossRef]
17. Liu, D.; Li, L.; Sun, Y. An improved radiative transfer model for estimating mineral abundance of immature
and mature lunar soils. Icarus 2015, 253, 40–50. [CrossRef]
18. Taylor, L.A.; Pieters, C.; Patchen, A.; Taylor, D.S.; Morris, R.V.; Keller, L.P.; McKay, D.S. Mineralogical and
chemical characterization of lunar highland soils: Insights into the space weathering of soils on airless bodies.
J. Geophys. Res. 2010, 115, E02002. [CrossRef]
19. Zhang, X.; Lin, H.; Cen, Y.; Yang, H. A nonlinear spectral unmixing method for Abundance retrieval of
mineral mixtures. Proc. SPIE 2016, 9874. [CrossRef]
20. Knight, E.J.; Kvaran, G. Landsat-8 operational land imager design, characterization and performance.
Remote Sens. 2014, 6, 10286–10305. [CrossRef]
21. Lenhard, K.; Baumgartner, A.; Gege, P.; Nevas, S.; Nowy, S.; Sperling, A. Impact of improved calibration of a
NEO HySpex VNIR-1600 sensor on remote sensing of water depth. IEEE Trans. Geosci. Remote Sens. 2015, 53,
6085–6098. [CrossRef]
22. Kotov, I.V.; Haupt, J.; O’Connor, P.; Smith, T.; Takacs, P.; Neal, H.; Chiang, J. Characterization and acceptance
testing of fully depleted thick CCDs for the large synoptic survey telescope. Proc. SPIE 2016, 9915. [CrossRef]
23. Hatchell, D.C. FR Repeatability. In Analytical Spectral Device Technical Guide, 3rd ed.; Analytical Spectral
Devices, Inc.: Boulder, CO, USA, 1999.
24. Maver, L.A.; Scarff, L.A. Multispectral imagery simulation. Proc. SPIE 1993, 1904, 144–160. [CrossRef]
25. Schott, J.R. Remote Sensing: The Image Chain Approach, 2nd ed.; Oxford University Press: New York, NY, USA,
2007; pp. 119–121.
26. Berk, A.; Anderson, G.P.; Acharya, P.K.; Chetwynd, J.H.; Bernstein, L.S.; Shettle, E.P.; Matthew, M.W.;
Adler-Golden, S.M. MODTRAN4 User’s Manual; Air Force Research Laboratory, Space Vehicles Directorate,
Air Force Materiel Command: Hanscom AFB, MA, USA, 2000.
27. Du, W.; Yang, S. Mineral mapping using ETM+, ASTER and Hyperion imagery in Honghai area of Hami in
Xinjiang, China. Int. Conf. Remote Sens. Environ. Transp. Eng. 2013, 6, 961–964. [CrossRef]
28. Beck, B. Lasers light up the silver screen. IEEE Spectr. 2014, 51, 32–39. [CrossRef]
29. Dong, P.; Cheng, K.W.E.; Ho, S.L. Interpolation simulation model of steady-state and dynamic characteristics
for car automotive system. IET Power Electron. 2009, 2, 697–705. [CrossRef]
30. Khanh, T.Q. Lighting Quality for Automotive Lighting. Light Eng. 2014, 22, 59–63.
31. Xue, D.; Zhiwei, S.; Graham, J.N.; Peter, J.A.; Dahe, G. Time-resolved spectra of solar simulators employing
metal halide and xenon arc lamps. Sol. Energy 2015, 115, 613–620. [CrossRef]
32. Jawad, S.; Georgakis, G.; LaChance, R.; Ozalp, N. Description and characterization of an adjustable flux solar
simulator for solar thermal, thermochemical and photovoltaic applications. Sol. Energy 2014, 100, 179–194.
[CrossRef]
33. Koninklijke, P.N.V.; Royal Philips. MASTERColour CDM-Rm Mini 20W/830 GX10 MR16 40D.
Available online: www.philips.com/lighting (accessed on 15 December 2015).
34. Optical Research Associates. Modeling Sources in LightTools; Optical Research Associates: Pasadena, CA, USA, 2007.
35. IEC Center Office. International Standard Photovoltaic Devices—Part 9: Solar Simulator Performance Requirements;
IEC Central Office: Geneva, Switzerland, 2007.
36. Tian, Y.; Yin, K.; Lu, D.; Hua, L.; Zhao, Q.; Wen, M. Examining Land Use and Land Cover Spatiotemporal
Change and Driving Forces in Beijing from 1978 to 2010. Remote Sens. 2014, 6, 10593–10611. [CrossRef]

© 2017 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).