Sie sind auf Seite 1von 17

Solar Energy 149 (2017) 60–76

Contents lists available at ScienceDirect

Solar Energy
journal homepage: www.elsevier.com/locate/solener

Developing a spectroradiometer data uncertainty methodology


Josh Peterson a,⇑, Frank Vignola b, Aron Habte c, Manajit Sengupta c
a
Solar Radiation Monitoring Laboratory, Department of Physics, 1274 University of Oregon, Eugene, OR 97403-1274, United States
b
University of Oregon, United States
c
National Renewable Energy Laboratory, United States

a r t i c l e i n f o a b s t r a c t

Article history: The proper calibration and measurement uncertainty of spectral data obtained from spectroradiometers
Received 18 January 2017 is essential in accurately quantifying the output of photovoltaic (PV) devices. PV cells and modules are
Received in revised form 22 March 2017 initially characterized using solar simulators but field performance is evaluated using natural sunlight.
Accepted 26 March 2017
Spectroradiometers are used to measure the spectrum of both these light sources in an effort to under-
Available online 11 April 2017
stand the spectral dependence of various PV output capabilities. These chains of characterization and
measurement are traceable to National Metrology Institutes such as National Institute of Standards
Keywords:
and Technology, and therefore there is a need for a comprehensive uncertainty methodology to deter-
Spectroradiometer
Uncertainty
mine the accuracy of spectroradiometer data. In this paper, the uncertainties associated with the respon-
Cosine response sivity of a spectroradiometer are examined using the Guide to the Expression of Uncertainty in
LI-COR 1800 Measurement (GUM) protocols. This is first done for a generic spectroradiometer, and then, to illustrate
the methodology, the calibration of a LI-COR 1800 spectroradiometer is performed. The reader should
be aware that the implementation of this methodology will be specific to the spectroradiometer being
analyzed and the experimental setup that is used. Depending of the characteristics of the spectrora-
diometer being evaluated additional sources of uncertainty may need to be included, but the general
GUM methodology is the same. Several sources of uncertainty are associated with the spectroradiometer
responsivity. Major sources of uncertainty associated with the LI-COR spectroradiometer are noise in the
signal at wavelengths less than 400 nm. At wavelengths more than 400 nm, the responsivity can vary
drastically, and it is dependent on the wavelength of light, the temperature dependence, the angle of inci-
dence, and the azimuthal orientation of the sensor to the light source. The expanded uncertainties in the
responsivity of the LI-COR spectroradiometer in the wavelength range of 400–1050 nm can range from 4%
to 14% at the 95% confidence level.
Ó 2017 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://
creativecommons.org/licenses/by/4.0/).

1. Introduction methodology (BIPM et al., 1995) is increasingly used for evalua-


tions of the uncertainty in solar irradiance measurements and
The energy produced by a solar energy system is directly related sensors. It also provides a consistent terminology for discussions
to the incident irradiance. Any uncertainty in the irradiance trans- of uncertainties as well as a standardized method for uncertainty
lates to uncertainty in the predicted system performance. There- analysis.
fore, uncertainties in the irradiance directly impact the system The performance of photovoltaic (PV) modules depends on both
design, performance estimates, and the financial risk associated the total incident solar radiation and the spectral distribution of
with system production. With larger solar electric systems being the incident radiation. Although current performance models use
deployed and costs becoming more competitive, considerable air mass and other atmospheric parameters as a proxy for changes
effort is going into fully characterizing the solar resource and eval- in the spectral distributions, the need for more precise models has
uating the accuracy and uncertainties in the solar measurements. resulted in the need to incorporate spectral data into PV system
The Guide to the Expression of Uncertainty in Measurements (GUM) performance analysis (Marion, 2012; Mavromatakis and Vignola,
2016). As with other solar irradiance measurements, knowledge
of the uncertainties of the spectral irradiance measurements is
⇑ Corresponding author. important.
E-mail address: jpeters4@uoregon.edu (J. Peterson).

http://dx.doi.org/10.1016/j.solener.2017.03.075
0038-092X/Ó 2017 The Authors. Published by Elsevier Ltd.
This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
J. Peterson et al. / Solar Energy 149 (2017) 60–76 61

Calibrations of spectral radiometers are extremely accurate an angle of incidence of 4°, the cosine error was 0.2%. Spectrora-
under laboratory conditions, where the light source is well cali- diometers used in making global horizontal irradiance measure-
brated, a constant temperature is maintained, and the incident ments observe light from a significantly larger field of view and
irradiance is perpendicular to the sensor. When the instruments require an uncertainty calculation which does likewise. In addition
are used in the field to measure the spectral distribution of global not every instrument employed in the field is equipped with a dif-
horizontal irradiance (GHI) or global tilted irradiance, the condi- fuser dome.
tions can be quite different from those in the laboratory under These examples demonstrate the need for a consistent method-
which spectroradiometers are normally calibrated. These different ology to accurately and consistently measure the uncertainty of
conditions lead to an uncertainty calculation that requires the spectroradiometers. Applying the GUM method would accomplish
inclusion of other important information. this. In many of these previous studies a modified version of the
The motivation of this work is to outline a methodology to GUM method is used, the methodology is obfuscated by details
determine the uncertainty of a spectroradiometer that is consistent pertaining to the experiment. In addition, most of the previous
with the GUM model. This methodology is applicable to all spec- studies have not included information pertaining to the cosine
troradiometers and is not specific to a particular instrument. More- response of the instrument and none have performed cosine
over the methodology aims at characterizing the uncertainties of a response measurements at large angles of incidence.
spectroradiometer in the laboratory under conditions that resem- The paper is separated into four sections and three related
ble those in the field. Specifically, tests on the deviation from the appendices. First an outline of the GUM method is given for a gen-
true cosine response and the temperature sensitivity of the instru- eric spectroradiometer. Then as an example, the GUM method is
ment are added to the uncertainties typically considered when cal- applied to the calibration of a LI-COR 1800 spectroradiometer. A
ibrating a spectroradiometer. discussion of the results and conclusions then follows. There are
To illustrate how this study improves upon previous body of three appendices which give supplemental details on the measure-
knowledge, a brief review of some of the previous work is in order. ment techniques used on the LI-COR 1800.
Several groups have performed intercomparison experiments This paper is intended to provide guidance on how to preform
using a detector based calibration procedure to compare spectrora- uncertainty estimation using the GUM method. The paper uses
diometers. An intercomparison study by Habte et al. (2014) found two spectroradiometers and associated source of uncertainties as
that over the wavelength range 400–1100 nm the differences an example. The sources of uncertainties described here are not
between ten different spectroradiometers were less than 10%. A complete. It is up to the user to modify the methodology described
similar study performed by Martinez-Lozano et al. (2003) on the here to their specific application, based on measurement equation
visible portion of the spectrum found a 5% variation between and known sources of uncertainties.
instruments. An intercomparison is an excellent way to compare
instruments however there are some difficulties associated with
this method. 2. Outline of the GUM method

d It is logistically challenging to gather and operate numerous The responsivity of a spectroradiometer is measured over a
spectroradiometers simultaneously at one location. range of angles of incidence with respect to the normal. The rela-
d The angle of incidence between the instruments and the light tive responsivity at each angle of incidence is determined with
source cannot be controlled and is not well defined, as the dif- respect to the responsivity of the instrument at an angle of 45°.
fuse light is from all portions of the sky. The expanded uncertainty of the instrument’s responsivity was
d The reasons for the variations between different instruments calculated over a range of angles from 30° to 60°. The uncertainty
are either unknown or difficult to determine from the data. calculation was performed using the GUM model described in
d In a particular instrument, the sources of the uncertainties are Habte et al. (2013), Reda et al. (2008), and Wilcox et al. (2002).
not subdivided into itemized components making further inves- The GUM process can be summarized in seven main steps.
tigation and reduction of the uncertainty difficult.
d A source of error that is common among all the instruments I. Determine the measurement equation for the quantity
involved in the intercomparison will not be observed and will under investigation. The measurement equation should
be neglected. This will result in an overly optimistic estimate explicitly list all variables used in making the measurement.
of the uncertainty. II. Identify the sources of uncertainty associated with each
variable in the measurement equation. Some variables in
Another common method to determine the uncertainty of an the measurement equation may have multiple sources of
instrument is to do a source based calibration. Yoon and Gibson uncertainty.
(2011) did an extensive study on the uncertainty of the calibration III. Compute the sensitivity coefficient (Cj) for each variable in
process of the NIST lamps used in spectroradiometer calibrations. the measurement equation. The sensitivity coefficients are
In their study they determined the expanded uncertainty was less computed by taking the partial derivative of the measure-
than 1.2% for wavelengths between 350 and 1100 nm. However, ment equation with respect to each variable in the measure-
this uncertainty estimate does not include the Type A uncertainty ment equation. The sensitivity coefficients describe the
of the spectroradiometer and does not include any information degree to which the measurement result will change if the
about the cosine response of the instrument as the measurements input quantity changes.
are taken at an angle of incidence of zero degrees (Habte et al., IV. For each variable in the measurement equation, calculate the
2013; ASTM Standard G-138-12). Both of these sources of uncer- standard uncertainty (lower case u) using either statistical
tainty significantly increase the uncertainty of instrument under methods (Type A uncertainty) or nonstatistical methods by
realistic conditions. Other groups have performed the source based estimating the uncertainty using available knowledge (Type
uncertainty measurements of spectroradiometers equipped with B uncertainty). The computation of the Type A and Type B
diffuser domes. These studies include information about the cosine uncertainties are described below. A variable may have sev-
response of the instrument however they typically only go up to eral sources of uncertainty, including mixtures of both Type
angles of incidence of 10°. Eppeldauer et al. (2009) found that at A and Type B uncertainties, that factor into the overall
62 J. Peterson et al. / Solar Energy 149 (2017) 60–76

uncertainty of that variable. For such situations, the stan- For a spectroradiometer, the measurement equation needed to
dard uncertainty is computed using the root-sum-squared compute the irradiance is given by Eq. (3). This equation computes
method of the various sources. the spectral irradiance from the measured millivolt signal detected
V. The responsivity of the spectroradiometer may vary as the and the spectral responsivity of the device.
angle of incidence changes. As such, this variation in respon-
VðkÞ
sivity should be measured and included in the overall uncer- IrrðkÞ ¼ ð3Þ
RðkÞ
tainty of the instrument.
VI. Calculate the combined standard uncertainty for the quan- In Eq. (3), Irr(k) is the spectral irradiance of the light source
tity under investigation. This is done using the root-sum- (W/(m2 nm), V(k) is the electrical signal output by the spectrora-
squared method for all standard uncertainties determined diometer usually measured in mV, and R(k) is the responsivity
in steps 4 multiplied by the corresponding sensitivity coeffi- factor (mV/(W/m2 nm)). Each variable in Eq. (3) depends on the
cients found in step 3. The combined standard uncertainty wavelength, specified by subscript (k). In the forthcoming discus-
should also include the variation in responsivity determined sion, the wavelength specifier (k) is omitted from the expressions
in step 5, also added in quadrature. to make the equations easier to read. Readers should note that each
VII. Calculate the expanded uncertainty by multiplying the com- variable is wavelength dependent.
bined standard uncertainty by a coverage factor (k) deter- By rearranging Eq. (3) to solve for the responsivity in terms of
mined for the desired level of confidence (usually 95%) the voltage and the irradiance, the responsivity of the instrument
with a value that is typically derived from a student’s ‘‘t” can be determined.
table. The responsivity of the device is recorded at an angle
V
of incidence of 45°. The expanded uncertainty of the instru- R¼ ð4Þ
ment is reported over a range of angles from 30° to 60°. Irr
Through the use of Eq. (4), the responsivity of the instrument
The standard uncertainties computed in step 4 are either one of can be experimentally determined by measuring a light source of
two types, Type A or Type B uncertainties. known irradiance and recording the voltage. The light source could
Type A: Uncertainties pertain to statistical variations in the data be either the sun as with an intercomparison of instruments or a
set due to random effects. These are the natural variations within a reference lamp. If the light source is the sun, a second spectrora-
data set that occur with repeated measurements. The uncertainty diometer is required to act as a reference instrument. If the light
is the standard deviation of the data set and is computed using source is a reference lamp, the irradiance of the lamp must be
Eq. (1). known. The experimental setup used in this study will measure
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi the irradiance of a reference lamp.
Pn  2 For a source based calibration using a lamp, the spectrora-
i¼1 ðX i  XÞ
uA ¼ ð1Þ diometer being calibrated is typically located a known distance
n1
from the light source, and the sensor is placed such that the plane
where uA is the Type A standard deviation, n is the number of data of the sensor is perpendicular to the incident light. At this location,
 is the
points in the sample, X i are the individual data points, and X the irradiance of the lamp is specified and follows the inverse
mean of the sample. square law (Yoon et al., 2012). Under these conditions the corre-
Type B: Uncertainties that pertain to estimates using available sponding uncertainty can be determined. Depending on the cali-
information. These are the uncertainties in the device itself that bration setup and the service provider, the uncertainty of the
are not directly measured using statistical means and are obtained irradiance, may or may not include fluctuations associated with
from various sources including: manufacturer specifications, cali- variations in the distance between the instrument and the lamp.
bration results, and experimental or professional judgment. These For spectroradiometers calibrated at NIST and NREL among others,
uncertainties get divided by a factor that depends on the type of the uncertainty of the lamp irradiance includes any variations in
distribution associated with the measurement. For a rectangular the irradiance due to variations in distance (Yoon and Gibson,
distribution, such as is associated with the precision of the data 2011; Habte et al., 2013). If the service provider does not include
logger, the standard uncertainty is computed by dividing the the variation in distance in their uncertainty estimate, this should
uncertainty by a square root of three. For a normal distribution, be included additionally. The uncertainty of the lamp is discussed
the standard uncertainty is computed by dividing the uncertainty in greater detail in the upcoming sections.
by the coverage factor (k), typically k = 2 for a 95% confidence In this standard calibration practice, the angular uncertainty
interval. Other distributions exist and are discussed in the depth has minimal effect because the sensitivity to a small misalignment
in the GUM model. is near zero. However, when the spectroradiometer is used in the
field, the incident angle is rarely zero, and the light intensity is
uncertainty estimate ideally proportional to the cosine of the incident angle. This is
uB rectangular ¼ pffiffiffi ð2AÞ
3 called a Lambertian response. The deviation from the Lambertian
response is often called the cosine response or cosine error of
uncertainty estimate the instrument and will be referred to as such in this document.
uB normal ¼ ð2BÞ
k To determine the cosine response of the instrument, the angle
between the instrument and the light source must be varied dur-
where uB is the Type B standard uncertainty and k is the coverage ing the calibration procedure. As the angle increases, the amount
factor. of light incident on the detector decreases, and it is determined
according to Eq. (5):
2.1. Determining the measurement equation
Irr ¼ I CosðhÞ ð5Þ
Using the steps outlined above, the basic formula and proce- where Irr is the irradiance of light reaching the detector at a partic-
dures are established for a generic spectroradiometer; however ular angle and location, I is the irradiance of light incident on a per-
the GUM methodology can be applied to other similar devices. pendicular surface to the light source at the desired location, and h
J. Peterson et al. / Solar Energy 149 (2017) 60–76 63

is the angle of incidence between the plane of the detector and the 2.3. Determining the sensitivity coefficients
incident light. Under these laboratory calibration conditions the
light is collimated and only from one direction. In Eq. (5), the units Using Eq. (6) as the measurement equation, the sensitivity fac-
of both the left and right sides are W/(m2 nm). tors are calculated by taking the partial derivative of the measure-
The experimental setup may require that the spectroradiometer ment equation with respect to each variable.
be moved back from the light source to be able to rotate the device
without hitting the light. In doing this, the irradiance should be dR 1 R
CV ¼ ¼ ¼ ð7AÞ
measured at the location where the experiment will take place. d V I CosðhÞ V
The irradiance of the lamp at the reference location is measured
with a second, separate calibrated spectroradiometer. dR V R
CI ¼ ¼ ¼ ð7BÞ
Combining Eqs. (4) and (5) results in the measurement equation d I I2 CosðhÞ I
used in the GUM model:
d R V SinðhÞ
V Ch ¼ ¼ ¼ R TanðhÞ ð7CÞ
R¼ ð6Þ dh I CosðhÞ2
I CosðhÞ
The sensitivity coefficients in Eq. (7) are all linearly dependent
This is the responsivity at a given incident angle and distance. on the responsivity (R).

2.4. Determining the standard uncertainty


2.2. Sources of uncertainties

The next step in the GUM procedure (step 4) is to determine the


The sources of uncertainty are specific to the instrument in
standard uncertainty of each variable in the measurement equa-
question and the experimental conditions used. The uncertainties
tion. The following notation is applicable for a generic GHI spectro-
associated with the measurement equation include any uncer-
radiometer. The numerical values of the standard uncertainties
tainty involved in the voltage, irradiance, or angle of incidence.
will depend on the specifics of the measurements and the instru-
Each of these variables may depend on one or more parameters.
ment being used. The values for a LI-COR 1800 spectroradiometer
For example the uncertainty in voltage may be dependent on the
are listed in Table 2 and used in Section 3.3 when evaluating the
noise associated with the data logger and an uncertainty associated
uncertainties of the LI-COR spectroradiometer.uv: The standard
with the temperature response of the instrument. In addition, if the
uncertainty of the voltage coming from the spectroradiometer
responsivity varies as a function of the angle of incidence, then that
when measuring the lamp is denoted as uV. This is the net variation
to must be included as a source of uncertainty as described in step
in the voltage at each wavelength that is recorded by the data log-
5 of the GUM model. Table 1 lists a general overview of the uncer-
ger after all signal processing has occurred within the device. The
tainties that would likely be present in most calibrations. Table 2 at
standard uncertainty of the voltage may be composed of both Type
the conclusion of Section 3.2 gives a detailed list of the specific
A and Type B uncertainties. Also, each type of uncertainty may
uncertainties associated with the calibration of the LI-COR 1800
have several components. For example, the Type B uncertainty of
spectroradiometer.
the voltage depends on the uncertainty due to variations in the
The uncertainty parameters listed in Table 1 correspond to
temperature of the sensor and the uncertainties associated with
uncertainty values of an instrument for a single calibration. If mul-
noise in the measurement electronics. Combining the Type A and
tiple calibrations are taken over an extended period of time, the
Type B uncertainties using the root-sum-squared method results
instrument stability can be monitored and the change in respon-
in the following standard uncertainties in the voltage:
sivity over time can be studied. With information on the change
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
X X ffi
in responsivity over time an additional uncertainty related to the
uV ¼ ðu 2
VA Þ i
þ ðu 2
VB Þ j
length of time from the previous calibration and the amount of i j
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
exposure should be included (BIPM et al., 1995; Wilcox et al.,
¼ uVA1 þ uVA2 þ    þ u2VB1 þ u2VB2 þ   
2 2
ð8Þ
2002). The uncertainty model developed in this paper considers
the uncertainty in the responsivity due to a single calibration and where both the Type A and Type B uncertainties are summed over
does not include any uncertainty that may be introduced due to the various uncertainties associated with the voltage measurement.
instrument degradation. If a long period of time has passed since In Eq. (8), this is denoted with the subscripts ‘‘i” and ‘‘j”. The Type B
the instrument has been calibrated, the degradation of the instru- standard uncertainties in Eq. (8) have all been divided by the appro-
ment should be considered. The degradation rate can be obtained priate divisor according to Eq. (2).
from the manufacturer specifications or from the measured uI: The standard uncertainty of the irradiance in the lamp is
responsivity history of the instrument. denoted as uI. This is the uncertainty in the irradiance of the lamp
when the spectroradiometer is perpendicular to the light source at
the reference distance from the light source. The spectral distribu-
Table 1
tion and magnitude of the light is obtained by a calibration that
List of expected sources of uncertainty of spectroradiometer measurement. should be traceable to a National Institute of Standards and Tech-
nology (NIST) reference. Sources of uncertainty to consider in the
Uncertainty Brief description Associated
label with
lamp irradiance are the accuracy of the calibration of the lamp
and uncertainties associated with the placement of the spectrora-
uV Uncertainties in the voltage measurement of Voltage
diometer at the calibration distance. These various standard uncer-
the spectroradiometer
uI Uncertainties in the irradiance of the lamp Irradiance tainties associated with the irradiance are also combined using the
used to calibrate the spectroradiometer root-sum-squared method.
uh Uncertainties in the angle between the Angle (h) uh: The standard uncertainty of the angle (h) when measuring
spectroradiometer and the lamp
the lamp is denoted as uh. The uncertainty in the angle is deter-
uangle Variations in the responsivity due to changes Responsivity
in the angle of incoming light
mined by how accurately the angle of incidence can be measured
during the calibration process. This is not an uncertainty of how
64 J. Peterson et al. / Solar Energy 149 (2017) 60–76

the responsivity varies with incident angle but merely how accu- individual trials for each azimuthal orientation at each angle of inci-
rately the incident angle can be measured. dence according to Eq. (6). All azimuthal angles measurements are
The three terms uV, uI, and uh are the standard uncertainties included in the R30–60 data set. The minimum and maximum
associated with their corresponding measurements. Each term responsivity in the range between 30° and 60° is determined and
may be composed of several subparts. The details of each uncer- subtracted from the average at 45°. In this way, the largest varia-
tainty are specific to the instrument in question and the technique tions one would expect in the responsivity are obtained. Assuming
used by the calibration service provider. These terms are a measure that the uncertainties are rectangular, the deviations are divided by
of how accurately the instrument can be measured for a particular a square root of three. Two uncertainties are associated with the
angle of incidence and azimuthal angle. angle: one corresponds to a maximum responsivity, and the other
corresponds to a minimum responsivity. These two uncertainties
do not necessarily have the same value.
2.5. Standard uncertainty in the responsivity due to variations in angle
A reference responsivity at an incident angle of 45° was chosen
as it is typically used for the calibration for a pyranometer. A differ-
Due to limitations in instrument design, optical systems such as
ent reference angle could be chosen to better represent the ‘‘typi-
spectroradiometers and pyranometers are prone to Lambertian
cal” incident angle on the pyranometer, but this typical angle will
cosine response errors (Myers, 1997). This deviation often varies
vary depending on location and time of year
with the angle of incidence and can also depend on the azimuthal
angle of the incident radiation (Cannon, 1986).
Under global horizontal irradiance field measurements, light is 2.6. Determining the combined standard uncertainty
incident on the spectroradiometer from all angles, both as direct
and diffuse light. The angle of incidence of direct light is well The combined standard uncertainties of the responsivity are
defined; however the angle of incidence of diffuse light is from computed using Eq. (10). The notation uangle is used to denote
all parts of the sky and is not well defined. Furthermore, without one of the angle uncertainties, either the plus or minus one (Eqs.
the use of two spectroradiometers (one global and one diffuse), (9A) and (9B)).
or other such means of separating the global irradiance into its qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
direct and diffuse components, the ratio of diffuse to direct light uR ¼ ðCV uV Þ2 þ ðCI uI Þ2 þ ðCh uh Þ2 þ ðuangle Þ2 ð10Þ
is not easily determined under all sky conditions. An integrating
sphere may reduce the uncertainty due to the cosine response of Substituting the sensitivity coefficients from Eq. (7) into Eq. (10)
the instrument; however integrating spheres have their own set results in:
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
of uncertainties ranging from spectral response to cosine response  2  2
(Eppeldauer et al., 2009; Hanssen, 2001). The discussion of inte- R R
uR ¼ uV þ uI þ ðR TanðhÞuh Þ2 þ ðuangle Þ2 ð11Þ
grating spheres and their uncertainty is outside the scope of this V I
study. The general form of the measurement equation (Eq. (6)) will
Pulling the responsivity, R, outside the square root and moving
be that of a spectroradiometer not using an integrating sphere.
it to the other side gives a ratio of the uncertainty in the responsiv-
The variations in the responsivity that occur due to deviations
ity relative to the responsivity. The negative signs inside the
in the cosine response of the instrument are often times neglected
squared terms have been ignored because these terms are squared.
in fieldwork, in that only a single responsivity is used for each rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
wavelength to describe all angles of incidence. Often times correc- u 2 u 2 u 2
V I angle
tions are not applied that would take into account the angle of
uR ¼R þ þ ðTanðhÞuh Þ2 þ ð12AÞ
V I R
incoming light. Therefore, if the angle of incidence during the mea-
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
surement deviates from the calibration angle of incidence, then an uR u 2 u 2 u 2
V I angle
additional uncertainty due to these deviations should be included ¼ þ þ ðTanðhÞuh Þ2 þ ð12BÞ
R V I R
in the overall uncertainty in the responsivity. These deviations
should be over a range of angles of incidence and azimuthal angles.
2.7. Determining the expanded uncertainty
If the responsivity is determined at each angle of incidence and
azimuthal angle, then the uncertainty in the responsivity would
The expanded uncertainty of the responsivity (U95R) is calcu-
simply be the combination of uncertainties listed above (uv, uI,
lated by multiplying the combined standard uncertainty by a cov-
uh), and there would not be an angular uncertainty. If, on the other
erage factor (k):
hand, the responsivity is given only at a one particular angle of
incidence and azimuthal orientation (as is typically done in field- U95R ¼ k uR ð13AÞ
work), then the uncertainties listed above do not contain all the
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
necessary information, and the uncertainty in the responsivity u 2 u 2 u 2
V I angle
due to angular variations must be added. U95R ¼kR þ þ ðTanðhÞuh Þ2 þ ð13BÞ
V I R
uangle: The standard uncertainty in the responsivity due to vari-
ations in the incident and azimuthal orientation is denoted as uangle where k = 1.96 (assuming infinite degrees of freedom) for a 95%
and is calculated using Eqs. (9A) and (9B): confidence level.
Moving the responsivity, R, to the left side results in a ratio of
MaxðR30—60 Þ  R45 the expanded uncertainty of the responsivity relative to the value
uangleþ ¼ pffiffiffi ð9AÞ
3 of the responsivity:
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
U95R u 2 u 2 u 2
V I angle
R45  MinðR30—60 Þ ¼k þ þ ðTanðhÞuh Þ2 þ ð14Þ
uangle ¼ pffiffiffi ð9BÞ R V I R
3
The expanded percent uncertainty of the responsivity (%U95R)
R45 is the average responsivity over all azimuthal rotations when is computed by multiplying Eq. (14) by 100:
the angle of incidence is 45°. R30–60 is a set of responsivity values for
U95R
each angle of incidence in the range between 30° and 60°. The val- %U95R ¼  100 ð15AÞ
ues of R30–60 are calculated by taking the average responsivity of the R
J. Peterson et al. / Solar Energy 149 (2017) 60–76 65

rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u 2 u 2 u 2 standard lamp with intensities known at specific wavelengths.
V I angle
%U95R ¼ 100k þ þ ðTanðhÞuh Þ2 þ Also, as with pyranometers, the spectroradiometer does not have
V I R
ð15BÞ a perfect Lambertian response; therefore, the changes in the
responsivity need to be measured at various angles of incoming
The expanded percent uncertainty of the responsivity depends light as well. To measure the responsivity of the device and the
on four terms, each of which is unitless. Each term in Eq. (15) is Lambertian response of the instrument, the device was placed in
for a given wavelength of light and a given intensity of light source. front of a standard lamp, as shown in Fig. 1.
Also, the uncertainty terms (uI, uV, uh, uangle) are all standard uncer- An alternative to the source based calibration setup shown in
tainty values. Fig. 1 would be a detector based calibration which has two spectro-
The responsivity term, R, in the denominator of the fourth term radiometers simultaneously observing a single light source (Habte
is the responsivity of the instrument at an angle of incidence of 45°. et al., 2014; Martinez-Lozano et al., 2003), typically natural light.
The responsivity is wavelength dependent and may also have some While a detector based calibration could reduce certain errors it
variations associated with the azimuthal angle. In Eq. (15B), the would also create other challenges such as the cosine and temper-
responsivity in the denominator is computed by taking the average ature response of the reference instrument must be well under-
responsivity over all azimuthal directions. stood, not being able to easily obtain all angles of incidence and
The first three terms in Eq. (15) are related to how well the light azimuthal angles in order to map the space, and for a global irradi-
source can be measured. The fourth term is related to how varia- ance measurement the light source will not have a well-defined
tions in the angle of incidence and azimuthal angle affect the angle of incidence. While there are solutions to these problems
responsivity. As mentioned earlier, uangle± consists of two different for the purposes of this study a source based calibration setup
values that were derived from Eqs. (9A) and (9B). In the computa- was chosen in order to have control over the experimental
u
tion of the term, angle
R
, both the top and bottom terms are inversely parameters.
related to the irradiance according to Eq. (6). In this way, the irra- From the experimental setup shown in Fig. 1, the responsivity
diance of the lamp will cancel. The term is a ratio of the voltages of the device at the various wavelengths and angles of incidence
and angles. Any stable light source can be used to measure the was computed using Equation (6).
cosine response of the instrument. Moreover the light source only The reference lamp used to calibrate the spectroradiometer was
needs to be stable to provide a reliable cosine response. a LI-COR 1800-02 Optical Radiation Calibrator (LI-COR Inc, 1990).
Under typical indoor calibration conditions, the spectrora- This lamp uses a power-regulated heated quartz tungsten halogen
diometer is calibrated perpendicular to the light source, and filament. The light source is totally enclosed by a light shroud with
h = 0. When this is the case, the uncertainty associated with the a small aperture to allow light to pass through to the detector. The
angular measurements becomes negligible because Tan(0) = 0. In actual filament is located 20.3 cm behind the aperture of the
addition, the cosine response of the instrument is ignored under shroud. The lamp used in this study was calibrated from 335 to
these experimental conditions, so uangle = 0. The first two terms 1060 nm through the use of an EKO MS-700 spectroradiometer.
are independent of the angle, h, and remain unchanged with vari- The EKO spectroradiometer used to calibrate the lamp has its
ations in the angle. As the angle increases, the angular terms own uncertainty which is incorporated into the uncertainty of
become increasingly important. the lamp. The calibration procedure and uncertainty measurement
of the lamp is discussed in in Section 3.2 and in Appendix B.
The front surface of the spectroradiometer detector was placed
3. Example of GUM method application on a LI-COR 1800
14.5 ± 0.2 cm away from the end of the calibrator aperture of the
spectroradiometer
reference lamp. This was done to allow the spectroradiometer to
rotate throughout the range of incident and azimuthal angles as
The GUM uncertainty methodology discussed in Section 2 gives
necessary. Even with the extra distance, some angles under certain
an overview of the general methodology for a GHI spectroradiome-
orientations were unattainable because the end of the spectrome-
ter. The calibration and uncertainty analysis of a LI-COR 1800 spec-
ter hit the lamp enclosure at large angles. In moving the spectrora-
troradiometer (LI-COR Inc., 1989) is next examined to give a
diometer back, the irradiance is decreased. The 14.5 cm distance
practical demonstration of how the GUM methodology is used.
was selected to give the most angle of incidence measurements
The LI-COR 1800 spectroradiometer has had widespread use in a
while still maintaining the largest irradiance. The spectroradiome-
number of research areas and has been considered a standard in
ter detector was positioned in the direct line of sight of the light
spectral irradiance measurements. This model of spectrometer is
source through the aperture, and the distance between the detec-
no longer manufactured due to changing availability of electronic
tor and the aperture was kept constant with a fixed-length remov-
components over time. A basic description of the LI-COR 1800
able spacer bar.
spectroradiometer is given in Appendix A. The unit used in this
study was equipped with a Teflon diffuser dome and has been
modified with a thermoelectric temperature control unit mounted
to the photodiode.
In Section 3.1 a description of the measurement technique is
given along with a description of how the responsivity varies with
the angle of incidence (GUM model step 5). In Section 3.2 a detailed
description of the uncertainties of the spectroradiometer is given
(GUM model steps 2, 3, and 4). In Section 3.3 the expanded uncer-
tainty is computed (GUM model steps 6 and 7).

3.1. Measuring the responsivity at various angles of incidence

The LI-COR 1800 uses a silicon detector to measure the spectral


irradiance. The silicon detector’s responsivity changes with wave- Fig. 1. Schematic of the experimental setup for determining the responsivity of the
length, which requires the instrument to be calibrated against a spectroradiometer.
66 J. Peterson et al. / Solar Energy 149 (2017) 60–76

To determine the cosine response of the spectroradiometer, the were explicitly included in the calculation, as an additional term
spectroradiometer was rotated in front of the light source. The in the uVV term, this uncertainty would be double counted.
angle of incidence, h, is the angle between the normal to the plane Using the voltage values given in Fig. 3, the responsivity was
of the detector and the angle of the incoming light. The angle of computed using Eq. (6). The responsivity of the instrument at a
incidence was varied in both the positive and negative directions variety of incident angles is shown in Fig. 4.
from h = 75° to +75°. The uncertainty in the angle of incidence The curves in both Figs. 3 and 4 have the same general shape,
measurement is ±0.3°. Under this experimental configuration, the with large jumps in the data set. The drastic changes in voltage
cosine response of the instrument could be determined over a wide are due to the preliminary filter wheel changing filters inside the
range of incident angles. spectroradiometer. The different filters have different transparency
In addition, the spectroradiometer was rotated so that the azi- characteristics, so whenever the filter changes, there can be a sharp
muthal angle of the incoming light was varied. A schematic of change in the output voltage. It follows that because the responsiv-
how the azimuthal angle was varied is shown in Fig. 2. In this fig- ity is proportional to the voltage, if the voltage changes drastically,
ure, two orientations of the device are shown; either Side A or Side then the responsivity will change as well. The change in filters often
B of the device is resting on the tabletop. In each of these two ori- causes the disjointed data points in the spectral responsivity file.
entations, the angle of incidence was varied from 75° to +75°. In At wavelengths less than 400 nm, the responsivity has signifi-
this way, the light was incident on the detector from all four azi- cant amounts of noise in the voltage signal, resulting in significant
muthal directions. The four azimuthal directions are labeled (0, noise in the responsivity. The sensitivity to noise at short wave-
90, 180, 270). An example light source is shown for Orientation lengths is visible in the left panel of Fig. 4 and is documented in
1, wherein side 270 is closer to the light source. For the other other studies (Riordan et al., 1989). At wavelengths greater than
angles of incidence, the light source is moved relative to the 400 nm, the responsivity does not vary much for different incident
spectroradiometer. angles, except when the incident angle increases to 75° or more.
With the spectroradiometer in the experimental setup In the upcoming discussion on the uncertainty associated with
described above, the voltage output by the spectroradiometer the responsivity of the spectroradiometer, the uncertainty in the
was measured as a function of wavelength for a variety of inci- responsivity due to variations in the incident and azimuthal orien-
dence and azimuthal angles. The average voltage of the four azi- tation (uangle) will be used (step 5 of the GUM model). Depending
muthal directions was computed for every angle of incidence and on the design and the materials used, an integrating sphere may
is shown in Fig. 3. In Fig. 3, the voltage curves for different angles reduce the size of the deviation from true cosine response of the
of incidence have the same general shape because the filter wheel instrument.
of the spectroradiometer always occurs at the same wavelengths. Fig. 4 corresponds to the average responsivity, with the results
To better show the small voltages present in the range from 300 from all azimuthal angles averaged. To understand the effects of
to 400 nm, a separate scale is used in the panel on the left. angle, the individual trials must be studied individually. Differ-
In Fig. 3, representative error bars of one standard deviation are ences that exist in the responsivity between these individual trails
shown at selected wavelengths. The size of the error bar was com- can offer some perspective on the effects of the uncertainty associ-
puted by taking the standard deviation of all voltage measure- ated with the angle.
ments at a particular angle of incidence, including various Using the voltage values corresponding to each angle of inci-
azimuthal angle orientations. These variations in voltage con- dence and azimuthal orientation, the responsivity was computed.
tribute the overall uncertainty of the responsivity and are a combi- Fig. 5 shows the responsivity for each of these variations at four
nation of noise in the device, variations in the lamp brightness, discrete wavelengths. The average responsivity at an angle of inci-
variations in the placement of the spectroradiometer with respect dence of 45° is also shown for perspective. The different azimuthal
to the lamp, and variations associated with different azimuthal orientations are listed as separate data sets. A legend for the figure
angles. The different sources of uncertainty cannot be easily sepa- is given in the 1000-nm plot. Note that the responsivity at 400 nm
rated into isolated components. It will be assumed that the terms is approximately an order of magnitude smaller than the respon-
uV u
V
and angle
R
incorporate these uncertainties. If this uncertainty sivity at higher wavelengths.

Fig. 2. Schematic of the experimental setup used to measure the azimuthal dependence of the spectroradiometer.
J. Peterson et al. / Solar Energy 149 (2017) 60–76 67

Fig. 3. Voltage output by the spectroradiometer compared to the angle of incidence. The voltage values shown here have been averaged throughout all azimuthal angle
measurements. Representative error bars of one standard deviation are shown. The error bars for the 0 and 15 degree angle of incidence are too small to be visible in the plot.

Fig. 4. Responsivity compared to wavelength for various angles of incidence. The responsivity is nearly constant as the angle is varied. Only when the angle of incidence is
increased to 75° is there a noticeable change in the responsivity compared to the other angles. A detailed view of the wavelengths from 300 to 400 nm is shown in the left
panel.

The error bars shown in Fig. 5 are the expanded uncertainty of from h = 30° to 60° considering all azimuthal orientations. This is
the responsivity excluding the angular terms (uangle±). The deriva- done for every wavelength of light. An example will help illustrate
tion of this uncertainty will be presented in Section 3.3. Significant the point.
differences exist in the responsivity due to the angle of incidence At 600 nm, the maximum responsivity is for an angle of inci-
and the azimuthal variations. These differences are larger than dence of h = 60° when the instrument was in the azimuthal orien-
the uncertainties in the data due to non-angle of incidence effects. tation of 90°. Under these conditions, the responsivity of the
Fig. 5 offers insight into how the responsivity changes when the mV
instrument is 32,500 W=m 2 nm.

angle of incoming light is varied. When the angle of incidence is At 600 nm, the minimum responsivity is for an angle of inci-
smaller than 45°, it is generally observed that the responsivity is dence of h = 45° when the instrument was in the azimuthal orien-
greater than the average responsivity at 45°. When the angle of tation of 270°. Under these conditions the responsivity of the
incidence is 45°, the responsivity values are centered around the mV
instrument is 28,200 W=m 2 nm.
average responsivity, as would be expected. As the angle of inci-
The maximum and minimum responsivity values are subtracted
dence is increased to 75°, it is generally observed that the respon-
from the average responsivity at 45° (Eqs. (9A) and (9B)) to deter-
sivity decreases sharply, with a few exceptions. It has been
mine the standard uncertainty associated with the angular varia-
suggested that the cause of this drastic change is likely the shade
tions of the incoming light, uangle ±. This is a measure of how
ring surrounding the diffuser dome (Cannon, 1986).
much the responsivity varies over the angle of incidence range
Observing the variations that exist in the different azimuthal
from h = 30° to 60°, and it is similar to the methodology used for
angles, one gains perspective on how the responsivity can vary
the pyranometer calibration made using the Broadband Outdoor
depending on the orientation of the light. The solar zenith angle
Radiometer Calibration methods prior to the year 2015 (Wilcox
and solar azimuthal angle vary during the day and during the year.
et al., 2002).
Because the responsivity of the spectroradiometer varies with inci-
In Eqs. (12)–(15), when the uncertainty in the responsivity is
dent angle and orientation, it is important to characterize these
computed, the standard uncertainty, uangle±, is divided by the
variations and understand how the angles affect the accuracy of
responsivity at 45°. The standard uncertainty divided by the
the measurements.
average responsivity offers a convenient way to compare the size
In Eqs. (9A) and (9B), the uncertainty term, uangle±, is a measure
of the uncertainty to the responsivity. The ratio of the uncertainty
of how much the responsivity changes over the angle of incident
relative to the responsivity is shown in Fig. 6, with the maximum
range from h = 30° to 60°. This was done by taking the maximum
and minimum uncertainties shown separately.
(and minimum) responsivity over the angle of incidence range
68 J. Peterson et al. / Solar Energy 149 (2017) 60–76

Fig. 5. Responsivity compared to angles of incidence. The different azimuthal orientations are shown explicitly. The legend in the 1000-nm plot shows the different azimuthal
orientations.

The ratio in the uncertainty is greatest at wavelengths less than standard uncertainty ‘‘u” variables listed in Section 2. A discussion
400 nm, shown explicitly in the panel on the left. In this wave- of each uncertainty and how it was calculated is given below.
length range, the magnitude of the responsivity is relatively small Table 2, at the conclusion of this section, lists the uncertainty val-
while the change in responsivity is relatively large, resulting in a ues for each variable as well as the notation used to describe each
large ratio. At wavelengths greater than 400 nm, the ratio of the term. The uncertainties are separated into four terms following the
uncertainty drops to less than 10% for all wavelengths. In Sec- notation of Eq. (15).
tion 3.3, it will be shown that the variations in responsivity due The first term in Eq. (15) is uVV . The uncertainties associated with
to variations in the angle are the largest source of uncertainty. the voltage measurement of the LICOR spectroradiometer are the
electronic noise in the signal (uv-Noise), the accuracy of the data log-
3.2. Uncertainties associated with the calibration technique of the LI- ger (uv-datalogger), and the effects of temperature of the spectrora-
COR 1800 spectroradiometer diometer (uv-Temp).
uV-noise: The voltage signal recorded by the data logger will have
With an understanding of the device and the measurement some natural variations. The uncertainty due to these variations is
technique used to calibrate the instrument, the possible sources referred to as ‘‘noise” and is a Type A uncertainty.
of uncertainty associated with the responsivity of the LI-COR The noise in the signal can be determined by placing the device
1800 will now be discussed. These uncertainties are the various in a totally dark environment and making a measurement. Ideally,
J. Peterson et al. / Solar Energy 149 (2017) 60–76 69

Fig. 6. The standard uncertainty associated with angular variations of the incoming light. The standard uncertainty, uangle±, has been divided by the average responsivity at
45° for comparison purposes. The maximum and minimum have been computed separately.

if there was no noise, the voltage would be zero. The dark signal enough accuracy such that the inaccuracy of the voltmeter does
was measured, and the voltage of the device was recorded. Fig. 7 not contribute to the overall uncertainty of the responsivity. Only
shows the results of the scan. The dark voltage can be positive or the round-off errors of the voltmeter are assumed to contribute.
negative. There is not a significant change in the noise as a function uV-Temp: The temperature of the spectroradiometer sensor
of wavelength. affects the voltage output by the sensor. The LI-COR 1800 spectro-
Using the data presented in Fig. 7, the standard uncertainty of radiometer used in this study has been modified by adding a tem-
the data is computed using Eq. (1). perature control unit to the instrument. The temperature control
unit maintains the temperature of the sensor to within ±0.1 °C.
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Pn  2 Under both calibration conditions and during field measurements,
ðV i  VÞ
uVnoise ¼ i¼1
¼ 0:980 mV ð16Þ the temperature of the sensor is maintained at 18 °C.
n1 To measure the effect of temperature on the instrument output,
These results are consistent with the LI-COR user manual (LI- the temperature of the sensor was varied while the device was
COR Inc., 1989, pp. 7-3, 8-1), which states that the noise should observing a LI-COR 1800-02 calibration lamp. The output voltage
be less than 1 mV. Given that the voltage readings at wavelengths from the device was recorded as a function of temperature. This
greater than 400 nm are much greater than the noise (see Fig. 3), enabled an estimate of the uncertainty of the voltage output as a
the uncertainty due to noise is insignificant. Only at wavelengths function of temperature. The uncertainty associated with the tem-
less than 400 nm, when the voltage becomes smaller, does the perature is a Type B normal distribution. See Appendix C for a
noise become a significant source of uncertainty. detailed outline of this procedure.
uV-Datalogger: The voltage readings produced by the LI-COR 1800 The temperature of the spectroradiometer is assumed to be in
are measured in 1-mV increments. The voltage signal is measured thermal equilibrium with the surrounding environment. For the
by the device and rounded to the nearest millivolt when the signal device used in this paper, which was equipped with a thermal con-
is made digital. The voltage measurements could be off by as much trol unit, the photodiode was maintained at a constant tempera-
as ½ mV due to round-off errors of the data logger. This is a rect- ture. However for the device in question, the rest of the
angular Type B distribution. The standard uncertainty due to instrument was not thermally controlled. The optical components
round-off errors in the data logger is given by Eq. (17): not thermally controlled will experience a larger range of tempera-
tures. For the purposes of this study, the additional uncertainty due
1=2 to fluctuations in temperature of the rest of the device is neglected.
uVDatalogger ¼ pffiffiffi ¼ 0:2887 mV ð17Þ Further studies are necessary to determine the uncertainty associ-
3
ated with a larger range of temperature on the rest of the optical
The manufacture does not give any information on the accuracy components. The methodology outlined at the end of Appendix C
of the voltmeter used as part of the spectroradiometer. It is assumed would take into consideration theses larger temperature fluctua-
that the voltmeter is able to read the millivolt signals to a great tions. This would need to be done in an environmental chamber
such that all optical components are in thermal equilibrium.
The ratio of the standard uncertainty in the voltage relative to
u 
the voltage VTemp
V
is shown in Fig. 8. The ratio of the uncertainty
is independent of the incident angle and the distance to the light
source because both the voltage and uncertainty in voltage change
at the same rate. As the angle of incidence increases, the voltage
decreases, and the uncertainty in the voltage decreases by the
same amount, resulting in the ratio remaining unchanged when
the angle of incidence is varied. Therefore, the ratio shown in
Fig. 8 applies to all orientations of the lamp and the sensor.
The uncertainty due to temperature uncertainties becomes
prevalent at wavelengths greater than 900 nm. At wavelengths
between 400 and 900 nm, the uncertainty due to temperature vari-
ations is negligible. At wavelengths less than 400 nm the noise in
the responsivity signal obscures any effects due to temperature.
Fig. 7. The dark signal voltage compared to wavelength. The detector was placed in If the spectroradiometer did not have a temperature control
an entirely dark environment and voltage was recorded. unit, the instrument would experience a much larger fluctuation
70 J. Peterson et al. / Solar Energy 149 (2017) 60–76

uVTemp 
Fig. 8. The ratio of the standard uncertainty in the voltage due to temperature variations compared to the voltage V
for a temperature change of 0.1 °C.

with temperature and the uncertainties in voltage would be signif- uI-EKO Calibration: The EKO spectroradiometer had recently been
icantly larger. For such an instrument, the uncertainties due to calibrated at NREL against a NIST-certified lamp (Yoon and
temperature variations become the predominate uncertainty in Gibson, 2011). The standard calibration procedure of NREL reports
the voltage for wavelengths greater than 900 nm. A brief descrip- the expanded uncertainty of the instrument which has been con-
tion of an instrument undergoing larger temperature variations is verted to a standard uncertainty. The ratio of the uncertainty in
given at the conclusion of Appendix C. the irradiance due to inaccuracies in the EKO calibration is shown
Combining the three kinds of standard uncertainties that exist in Fig. 10. The uncertainty value reported in Fig. 10 is the standard
for the voltage by using the root-sum-squared method results in uncertainty of the calibration process.
the standard uncertainty in the voltage: uI-EKO Noise: To determine the noise in the lamp signal, the EKO
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi spectroradiometer was placed at 34.8 cm and allowed to record 12
uV ¼ ðuVNoise Þ2 þ ðuVDatalogger Þ2 þ ðuVTemp Þ2 ð18Þ values of the lamp during a 12-min time period. The standard devi-
ation of the irradiance of these measurements was recorded. Any
Dividing both sides of Eq. (18) by the voltage results in the ratio of noise variations that are inherent in the EKO spectroradiometer
the standard uncertainty in voltage relative to the voltage: and also in the LI-COR lamp will be measured. The two effects can-
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi not be distinguished in this method. This is a Type A uncertainty;
uV u 2 u 2 u 2
VNoise VDatalogger VTemp
¼ þ þ ð19Þ thus, according to Equation (1), the standard uncertainty is the
V V V V standard deviation of the signal. The uncertainty due to the noise
This ratio will be used to compute the uncertainty in the in the signal is shown in Fig. 10. The noise is insignificant at all
responsivity using Eq. (15). wavelengths greater than 400 nm. At wavelengths less than
Fig. 9 shows the ratio of the standard uncertainty in the voltage 400 nm, the noise is the same order of magnitude as that of the
of the LI-COR spectroradiometer due to noise, round-off errors in irradiance, and the uncertainty becomes large.
the data logger, and temperature uncertainties. The uncertainty uI-distance: Both the EKO spectroradiometer (used to calibrate
due to the noise in the signal is the significant source of uncer- the lamp) and the LI-COR spectroradiometer (the instrument in
tainty. The uncertainty due to temperature variations is an order question) were positioned at the same location from the lamp. In
of magnitude smaller than the noise. the experiments conducted for these tests, the distance from the
Next, we turn our attention to the uncertainty in the irradiance sensor to the lamp filament had an uncertainty of ±0.2 cm. This
of the LI-COR lamp calibration system. The lamp was calibrated 2 mm was a judgement call made by the experimenters based on
using an EKO MS-700 spectroradiometer. A detailed description uncertainties on the size of the filament, challenges in placing
of the lamp calibration process is given in Appendix B. and measuring the spectroradiometer to the desired location, and
The uncertainty in the lamp is determined by three factors: uncertainties in the curved diffuser dome of the LICOR spectrora-
the uncertainty of the calibration of the EKO spectroradiometer diometer. The irradiance was measured at distances of 34.6 and
(uI-EKO Calibration), the noise in the EKO spectroradiometer signal 35.0 cm (2 mm on either side of 34.8). The irradiance at these
(uI-EKO Noise), and the uncertainty in the distance (uI-distance). two locations was determined and the amount of change in the

Fig. 9. The ratio of the standard uncertainty in the voltage due to variations in noise, round-off errors in the data logger, and temperature. The data shown here was for an
angle of incidence of 45°. The magnitude of the voltage varies with incident angle for a given irradiance intensity.
J. Peterson et al. / Solar Energy 149 (2017) 60–76 71

Fig. 10. Ratio of the standard uncertainty in the LI-COR lamp relative to the irradiance of the lamp. The various sources of uncertainty are plotted for comparison purposes.

irradiance was determined for a 2-mm spread in the distance. The It should be mentioned that the EKO spectroradiometer also has
standard uncertainty in the irradiance to variations in the distance a temperature response: However since the instrument was at
was assumed to be a normal distribution of Type B uncertainty. The room temperature when it was calibrated at NREL and during
variations in the irradiance were divided by two to give the stan- the lamp calibration in this study the temperature response
dard uncertainty according to Equation (2B). The ratio of the stan- of the instrument is included in the calibration uncertainty
dard uncertainty due to distance is shown in Fig. 10. uI-EKO Calibration. Had the two temperatures been outside the operat-
The three components of the uncertainty in the irradiance are ing temperature of the instrument 25 ± 5 °C, the uncertainty due to
combined using the root-sum-squared method. The ratio of the total temperature response would need to be included as well.
uncertainty in the irradiance of the lamp relative to the irradiance of uh: The uncertainty in the measurement of the angle of inci-
the lamp is shown in Fig. 10. The uncertainty is less than 1.1% for all dence, h. Note that uh is the uncertainty associated with how accu-
wavelengths greater than 400 nm. The major contributors to the rately the angle of incidence can be measured, and it is not the
uncertainty in the irradiance are due to variations in the distance same as the variation in the responsivity due to changes in the
of the sensor and the uncertainty in the calibration of the EKO. angle. The digital angle finder used to measure the angle of inci-
dence had an uncertainty of ±0.3° (.005236 rad), which is assumed
to be the uncertainty in the angle between the spectroradiometer
and the light from the calibrator. Assuming that the distribution
of uncertainties is a normal distribution, the 0.3° uncertainty is
divided by a factor of two to give the standard uncertainty associ-
ated with the angle of incidence according to Eq. (2B):
0:005236 radians
uh ¼ ¼ 0:002618 ð20Þ
2
Multiplying the standard uncertainty by the tangent of theta
according to Eq. (15) results in the overall uncertainty due to the
angle of incidence. The uncertainty associated with the angle of
incidence is shown in Fig. 11. Under the experimental conditions
that were used, the uncertainty can become as great as 1% at 75°.
uangle: The uncertainty associated with the change in responsiv-
ity when the angle of incidence is changed was discussed in Sec-
tion 3.1. Fig. 6 shows the ratio of the uncertainty relative to the
responsivity.
Fig. 11. The overall uncertainty associated with the angle of incidence. At large A brief synopsis of the various uncertainties previously dis-
incident angles, the uncertainty increases significantly.
cussed is given in Table 2. This table is intended to clarify any con-

Table 2
List of uncertainty parameters. See Figs. 6, 9, 10, and 11 for the relative size of each uncertainty.

Uncertainty Brief description Type of Number of data points Type of distribution


label uncertainty (Type A Only) (Type B Only)
uV-noise Noise in the LI-COR spectroradiometer voltage Type A 800
uV-Data Logger Round-off error in the LI-COR spectroradiometer Type B Rectangular
uV-Temp Variations in the voltage of the spectroradiometer due to fluctuations in temperature
Type B Normal
uv-Total Total uncertainty in voltage
uI-EKO Calibration Uncertainty of the EKO calibration process Type B Standard
uncertainty
uI-EKO Noise Noise in the lamp and EKO spectroradiometer (inseparable) Type A 12
uI-EKO Distance Variations in the irradiance due to uncertainties in the distance Type B Normal
uI-Total Total uncertainty in irradiance
uh Uncertainty in angle measurement Type B Normal
uh Tan(h) Total uncertainty in responsivity due to inaccurate angle measurement
uangle Uncertainty in responsivity due to changes in the angle of incoming light Type B Normal
72 J. Peterson et al. / Solar Energy 149 (2017) 60–76

fusion on the notation. The magnitudes of the uncertainty values in the ratio of the standard uncertainty in the responsivity relative
for each variable can vary drastically depending on several factors. to the responsivity. In Fig. 12, the ratio of the uncertainty in the
These factors include the wavelength of light, the angle of inci- responsivity relative to the value of the responsivity is shown
dence, and the azimuthal angle. See Figs. 6, 9, 10, and 11 for guides along with the four components of Equation (12B). The uncertainty
to the relative size of each uncertainty. shown in Fig. 12 is the standard uncertainty of the responsivity;
The accuracy of the spectroradiometer data depends not only the 95% uncertainty will be shown later.
on the specifications of the instrument but also on the calibration Fig. 12 shows two different plots. The upper plot, titled uanlge(+),
procedure, measurement set up, and meteorological conditions. shows the uncertainty that exists due to variations in the respon-
Therefore, statements about the overall measurement uncertainty sivity that are greater than the responsivity at 45°. These were
can be made only on an individual basis, taking all relevant factors derived from Equation (9A). The lower plot, titled uanlge(), shows
into account. Other potential sources of error that are not quanti- the uncertainty that exists due to variations in the responsivity
fied or included in this study include: wavelength offset, wave- that are less than the responsivity at 45°. These were derived from
length bandwidth, nonlinearity factors, how the electrical signal Eq. (9B).
is handled by the electronics of the device, the impact of air tem- In both plots, it is clear that the uncertainties, uangle±, are signif-
perature, relative humidity and other atmospheric parameters, icantly larger than the rest of the uncertainties combined. This is
temporal changes in the responsivity, and the effects of a non- true for nearly all wavelengths. The uncertainties vary depending
collimated light source. The reader should note that these sources on the wavelength of light being considered.
of uncertainty may need to be included depending on the experi- With an understanding of the various components associated
mental setup. with the uncertainty of the responsivity, the expanded percent
uncertainty was computed according to Eq. (15). The expanded
3.3. Calculation of the expanded uncertainty uncertainty was computed for both the uangle+ and uangle- terms.
The expanded percent uncertainty in the responsivity for these dif-
Combining the uncertainties of each type (voltage, irradiance, ferent computations is shown in Fig. 13.
measurement of angle, and variation of responsivity with angle) For comparison purposes, the expanded uncertainty was also
using the root-sum-squared method according to Eq. (12B) results computed with the uangle± terms set to zero. By not including

Fig. 12. Ratio of the standard uncertainty in the responsivity compared to the responsivity. The various components of the uncertainty in the responsivity are separated to
show their relative magnitudes. The uncertainty associated with the angle (uanlge±) has been separated into two separate plots.

Fig. 13. Expanded percent uncertainty in the responsivity. The uncertainty associated with the angle (uanlge±) has two different values corresponding to the two curves, uR95%
(+) and uR95%(). For comparison purposes, the expanded percent uncertainty is shown if the angle terms are ignored.
J. Peterson et al. / Solar Energy 149 (2017) 60–76 73

The uncertainties in the LI-COR spectroradiometer are similar to


uncertainties encountered in most spectroradiometers that mea-
sure the GHI spectrum. The leading sources of uncertainty in the
LI-COR spectroradiometer have been identified. First, there are
large uncertainties in wavelengths less than 400 nm that can be
attributed to the poor response of the spectroradiometer detector
to ultraviolet light. The second is the variation in the responsivity
when the incident angle of light is varied. Variations exist in both
the angle of incidence and the azimuthal angles. These deviations
from true cosine response can be associated with the optics of
the instrument. The uncertainty associated with the lamp used to
calibrate the device is the third leading source of uncertainty; how-
ever, this is significantly smaller than the other two previously
mentioned errors. The uncertainty due to temperature variations
Fig. 14. The responsivity curve compared to wavelength. The error bars shown on
in the sensor at near-infrared wavelengths (greater than 900 nm)
the curve correspond to the expanded uncertainty uangle± shown in Fig. 13.
has been significantly reduced by using a water-cooled plate with
a thermoelectric controller that is placed behind the detector. This
uangle±, it is clear how influential the angular uncertainty is in the
maintains the detector at a near constant temperature.
overall uncertainty.
A methodology to calibrate GHI spectroradiometers and to
The uangle± is a measure of how much the responsivity changes
quantify the uncertainty of these devices has been demonstrated.
from the average responsivity at an angle of incidence of 45°. The
The future studies listed below would improve upon this technique
amount the responsivity varies was measured over a range of
to reduce several of the leading sources of uncertainties for the
angles of incidence from 30° to 60°. The responsivities can change
spectroradiometer.
significantly at angles greater than 60°, whereas there is little
The diffuser dome of the spectroradiometer used in this study
change in the range between 0° and 45°.
deviates from the ideal cosine response curve in a predictable man-
At wavelengths less than 400 nm, the expanded percent uncer-
ner, and these deviations are the largest source of uncertainty for
tainty, including the deviation from true cosine response, has an
the spectroradiometer. Integrating spheres may reduce the size
expanded uncertainty between 20% and 200%. At wavelengths
of the uncertainty in the cosine response of the instrument. How-
greater than 400 nm, there is a 5% and 20% expanded uncertainty.
ever all instruments, even those using an integrating sphere, will
If the responsivity was determined and applied for each angle of
exhibit a cosine response uncertainty. It should be possible to
incidence, the uncertainty would be significantly reduced at all
determine a cosine response adjustment algorithm that would
wavelengths. This is shown in Fig. 13 as the data set labeled ‘‘No
apply a responsivity function that is dependent on the angle of
uangle uncertainty”.
incidence of light. The cosine response adjustment would work
The uncertainty estimates presented in Fig. 13 are consistent
well to correct the spectral direct irradiance contribution, which
with the uncertainty values from other studies (Habte et al.,
has a well-defined angle of incidence relative to the sensor. The dif-
2014, 2013; LI-COR Inc., 1989; Myers, 1989). These studies include
fuse spectral irradiance comes from all portions of the sky and is
both detector based calibrations and source based calibration tech-
crudely the weighted average angle of 45°. A challenge with this
niques. The LI-COR user manual states that the cosine response of
method is how to separate a singular global horizontal measure-
the instrument is ±8% (standard uncertainty). The user manual
ment into its two constituents, direct horizontal and diffuse hori-
does not give any information about the wavelength dependence
zontal. Under clear sky conditions this should be possible. Under
of the cosine response.
partly-cloudy conditions this becomes more challenging. One pos-
As is often the case when making field measurements, a single
sible solution is to use two spectroradiometers, one direct normal
responsivity is used regardless of the angle of incoming light. As is
and one diffuse, to measure the two components.
standard practice for pyranometers, the responsivity of the LI-COR
The uncertainties in wavelengths less than 400 nm would be
1800 was measured at 45°. This was done using the data set shown
significantly reduced, if the light source used had a brighter ultra-
in Fig. 4. The expanded uncertainty is applied to the responsivity
violet spectrum or the diffuser dome was replaced with something
and is presented in Fig. 14 as the error bars.
that had a better ultraviolet transmission. It may be necessary to
perform a calibration using two different light sources with two
4. Results and conclusions different spectrums. One of these light sources should be a strong
emitter of ultraviolet light in the wavelength range from 300 to
Accurate spectral irradiance measurements are important as 400 nm. With this light source, the uncertainties associated with
one tries to better understand and compare the performance of dif- the poor responsivity to ultraviolet should be reduced, which in
ferent PV technologies; therefore, quantifying uncertainties of turn would reduce the uncertainty in the irradiance in the wave-
spectral irradiance measurements using a consistent methodology lengths as well. The responsivity of the instrument as a function
is crucial. The GUM methodology has been used for estimating of wavelength would be the combination of the responsivity pro-
uncertainties in GHI spectral irradiance measurements, and as a duced by the two lamps.
specific example the uncertainties of a modified LI-COR 1800 spec- One of the largest sources of uncertainties associated with the
troradiometer have been calculated. The GUM method provides irradiance of the lamp was the location of the sensor relative to
uncertainty analysis that ensures the estimated uncertainty for the lamp. This uncertainty was present in both the calibration of
data collected by the spectroradiometer can be compared based the lamp and also with the calibration of the LI-COR spectrora-
on standardized methods of equations which supersedes the his- diometer. Care was taken to try to minimize this uncertainty, but
torically commonly used method to estimate uncertainty using due to experimental limitations, the uncertainties associated with
concept of precision and bias. Further, the method assists to iden- distance cannot be further minimized. The uncertainties associated
tity the different sources of uncertainties such as cosine and tem- with the distance are on the order of the curvature of the LI-COR
perature response errors independently which in turn will assist in spectroradiometer diffuser dome. Without a more accurate way
understanding and reducing the leading sources of uncertainty. to measure the location of the sensor, information about how the
74 J. Peterson et al. / Solar Energy 149 (2017) 60–76

azimuthal angle of the diffuser dome affects the responsivity of the A schematic of the LI-COR 1800 spectroradiometer is shown in
instrument cannot be determined. The uncertainty associated with Fig. A1.
distance would be reduced if the distance between the lamp and Incident light first passes through a cosine receptor diffuser
the spectroradiometer was increased. However, if this was done dome made of Teflon. The cosine receptor used in this study was
a brighter light source would also have to be used. not an integrating sphere. The light is redirected using a mirror
When a spectroradiometer is used in the field to make global set at 45° to the top plane of the instrument. The spectroradiome-
horizontal spectral irradiance measurements, there may be some ter then filters the incoming light using a preliminary color filter
additional uncertainties that need to be included in the final irradi- wheel. This filter blocks unwanted colors from entering the detec-
ance uncertainty calculation. In addition to the uncertainties in the tor. The filter wheel cycles through seven colors during the course
responsivity discussed here, there may be uncertainties due to of a scan. The wavelengths of light passed by each filter are listed in
degradation in the instrument, soiling of the detector surface and Table A1.
cleaning, and physical uncertainties such as leveling and shading Next, the light is separated into a spectrum of colors using a
to consider. To compute the uncertainty in the irradiance these diffraction grating. During the course of a run, the diffraction grat-
additional uncertainties should be included and the GUM model ing is rotated so that the various colors are incident on the detec-
should be applied to the measurement equation which computes tor. Before the light reaches the detector, it must first pass through
the irradiance. This is a process very similar to the one performed an exit slit, ¼-mm wide, which produces a narrow beam of light for
here, simply with a different measurement equation. the detector. The detector is a silicon photodiode, manufactured by
As better estimates of the performance of photovoltaic modules EG&G Optoelectronics Canada (model HUV-2000B).
become more important, it will be necessary to understand the As the diffraction grating moves, it sweeps different wave-
spectral characteristics of the incident solar irradiance. Measuring lengths across the slit. The slit width limits the wavelengths that
the spectral irradiance in the plane of array of a PV system has pass through and directly determines the spectral bandwidth that
large uncertainties because of the deviation from true cosine reaches the detector. The measurements described in this paper
response as well as the many other uncertainties associated with had a slit width of ¼ mm. This corresponds to a full width at half
spectral measurements. Models of the spectral distribution of inci- maximum bandwidth of 4 nm. Thus, when the grating is set at
dent solar radiation are only as good as the data that is used to val- 500 nm, the detector senses wavelengths from 498 to 502 nm.
idate the models. Therefore it is very important to understand the Therefore, although the solar irradiance data recorded by the spec-
uncertainties associated with these of spectral measurements. troradiometer is in 1-nm increments, these measurements include
a Gaussian distribution 4-nm wide. More information is available
in the LI-COR user manual (LI-COR Inc., 1989).
Acknowledgments Each optical component of the spectroradiometer will have an
effect on the amount of light reaching the photodiode. These com-
This work was supported by the U.S. Department of Energy
under Contract No. DE-AC36-08GO28308 with the National
Renewable Energy Laboratory. The authors also thank Bonneville Table A1
Power Administration, Energy Trust of Oregon, Emerald People’s Wavelengths of light passed through the various filters on the LI-COR 1800
Utility District, Portland General Electric, and Oregon BEST for their spectroradiometer.
support of the University of Oregon’s SRML activities. Start wavelength (nm) End wavelength (nm)
Filter 1 300 348
Filter 2 349 418
Appendix A. Basic overview of the LI-COR 1800 Filter 3 419 558
spectroradiometer Filter 4 559 678
Filter 5 679 775
It is useful to have a basic understanding of how the LI-COR Filter 6 776 938
Filter 7 939 1100
1800 operates to characterize the calibration and the uncertainty.

Fig. A1. Schematic of the LI-COR 1800 spectroradiometer (LI-COR Inc., 1989).
J. Peterson et al. / Solar Energy 149 (2017) 60–76 75

ponents include the diffuser dome, mirror, filter wheel, diffraction guishable from a linear fit. The line drawn in Fig. B1 is the best
grating, entrance and exit slits. The efficiency of the photodiode at fit through the data.
converting light to an electrical signal will determine the voltage The average value of the data at 34.8 cm is shown along with
that the photodiode produces. The optical properties of each com- the error bars that exist. The horizontal error represents the 2-
ponent will likely be dependent on the wavelength of light and mm uncertainty in the location of the LI-COR spectroradiometer.
may have a temperature response as well. Other models of spectro- The vertical error bars include uncertainties associated with the
radiometers or those operating with an integrating sphere would calibration of the EKO spectroradiometer, the noise in the EKO
have their own set of optical properties but the same principles measurements, and the variation that exists in the irradiance
can be applied. throughout the 2-mm change in distance. The details of these
The responsivity of the device is the ratio of how much voltage uncertainties are beyond the scope of this paper. The vertical error
was produced by the diode at the end of the process to how much bars encompass the variations in the data throughout the 2-mm
light was incident on the diffuser dome at the start according to Eq. spread in distance, as would be expected. The variation in the irra-
(6). Changing the properties of a single optical component, for diance as a function of distance is in good agreement with the
example the reflectance of the mirror, will change the light making expected r12 distance dependence of light. The uncertainty values
it to the final sensor, and in turn how much voltage the diode pro- in the irradiance of the lamp are the vertical error bars shown in
duces. This will have the effect of changing the responsivity of the Fig. B1. This procedure was done for all wavelengths.
instrument and is the underlying principle behind calibrations.
The set of optical components of a device are all internal to the
Appendix C. Effect of temperature on spectroradiometer
device and light will interact with the entire set of components
measurements
under both calibration conditions and during fieldwork. Since the
device is always used as a complete unit, details about the surface
The University of Oregon Solar Resource Monitoring Laboratory
aluminized mirror are not included.
added a temperature-control mechanism to the LI-COR 1800 spec-
troradiometer to minimize the effect of temperature variation on
Appendix B. Calibration procedure of the reference lamp the spectral measurements. This control also enables a detailed
analysis of the effect of temperature on the performance of the
To calibrate the LI-COR 1800 spectroradiometer, first the irradi- spectroradiometer. The spectroradiometer was mounted to a LI-
ance of the reference lamp had to be well understood. The refer- COR 1800-02 calibrated lamp. The temperature of the spectrora-
ence lamp used to calibrate the spectroradiometer was a LI-COR diometer was varied using the same temperature control mecha-
1800-02 Optical Radiation Calibrator. The lamp was calibrated nism that is used during normal operation. The temperature
using an EKO MS-700 spectroradiometer, which had been recently control unit can be varied to attain a wide range of temperatures.
calibrated at NREL with traceability to the NIST standards. In this The temperature of the spectroradiometer was varied from 10° to
way, the calibration of the LI-COR 1800 could be traced to a NIST 26 °C. The normal operating temperature of the instrument is
source. However, with each of these transitions comes an increase 18 °C when making measurements in the field. The temperature
in the uncertainty in the measurement. is controlled to accuracies of ±0.1 °C. The irradiance of the lamp
The EKO MS-700 spectroradiometer measured the reference was assumed to be constant for the tests.
lamp from 336 to 1060 nm. The lamp was measured such that At each temperature setting, the voltage output by the spectro-
the angle of incidence was h = 0°. radiometer was recorded. The ratio in voltages between the volt-
When the LI-COR spectroradiometer was calibrated, it was posi- age output by the device at T = 16 °C and T = 18 °C is shown in
tioned 34.8 cm from the filament of the lamp. The uncertainty in Fig. C1. The ratios in the voltages of T = 20 °C and T = 18 °C are also
this measurement was ±0.2 cm. Because the location of the LI- shown. The other temperature readings are not shown in this plot
COR sensor was not known to accuracies greater than 2 mm, the because these three temperature readings were the closest to the
changes in the irradiance of the lamp must be known throughout normal operating temperature of 18 °C when making the field
this range. To determine the irradiance of the lamp compared to measurements.
the distance, the EKO spectroradiometer was placed at varying dis- At wavelengths less than 400 nm, the uncertainty in the calibra-
tances in front of the lamp, and several measurements were made tion system produces large uncertainties that are not related to the
at each distance. temperature of the device but instead are related to the low ultra-
Fig. B1 shows the irradiance of the lamp at 600 nm throughout a violet intensity of the light source. Between 400 and 900 nm, the
small range of distances centered on 34.8 cm. Similar plots can be temperature of the device does not significantly change the voltage
made for other wavelengths. Over the small range of distances
shown in Fig. B1, the r12 dependence of the irradiance is indistin-

0.07
0.069 EKO measurment
Irradiance (W/m2/nm)

0.068 Average Value


0.067
0.066
0.065
0.064
0.063
0.062
34.0 34.2 34.4 34.6 34.8 35.0 35.2 35.4 35.6 35.8 36.0
Distance (cm)

Fig. B1. Irradiance of the lamp compared to distance from the filament. The
irradiance was measured at all wavelengths in the range of 336–1060 nm. The Fig. C1. The ratios of voltages at a variety of temperatures. The voltage at 16 °C (and
irradiance value shown here is for 600 nm. 20 °C) is compared to the voltage at a temperature of 18 °C.
76 J. Peterson et al. / Solar Energy 149 (2017) 60–76

output by the sensor, and there is very little difference between the late the uncertainty in the voltage associated with the larger tem-
readings made at T = 16 and 20 °C. At wavelengths greater than perature variations.
900 nm, a 2 °C change in temperature can change the responsivity Using the same methodology as above, the temperature of the
by up to 2% or 3%. This is fairly typical of the behavior of photodi- LI-COR 1800 was varied from 10 to 26 °C and the voltage was com-
odes that are used in the measurements (Riordan et al., 1989), and pared over this 16° temperature range using a modified version of
it demonstrates the importance of temperature control or adjust- Eq. (C1). Over this temperature range, the voltage varies in a nearly
ment for spectral measurements in the range from 900 to 1100 nm. linear manner with temperature at every wavelength.
To compute the uncertainty in the voltage due to a 0.1 °C tem- For outdoor field work, the instrument can experience ±25 °C
perature change, first the absolute value of the difference in volt- fluctuations in temperature over the course of the year. The larger
age between T = 16 and T = 20 °C was computed. This difference variation in temperature will produce standard uncertainties in
in voltage was caused by a four-degree variation in temperature. voltage of greater than 10% at wavelengths greater than 1050. This
To compute the variation in the voltage due to a 0.1 °C change in will be a leading source of uncertainty at wavelengths greater than
temperature, this difference must be multiplied by a scale factor 1000 nm. Wavelengths less than 900 nm are not significantly
as shown in Eq. (C1). It is assumed that within this temperature affected by the variation in temperature. These errors could be
range, the voltage changes linearly with temperature: reduced by measuring the temperature of the instrument and
applying a responsivity function that is temperature dependent.
0:1
uTemp ¼ ABSðVT16  VT20 Þ  ðC1Þ If a spectroradiometer does not have a temperature control unit,
4 one of the challenges would be how to maintain and accurately
The variations in temperature are assumed to be a Type B nor- measure the temperature of the device. Measurements would need
mal distribution. To compute the standard uncertainty in the volt- to be done in an environmental chamber that has a variable tem-
age caused by variations in the temperature, the uncertainty is perature control unit.
divided by a factor of two, as shown in Eq. (C2). This is the standard
uncertainty reported in Section 3.2: References
uTemp
uVtemp ¼ ðC2Þ ASTM Standard G-138-12. Standard Test Method for Calibration of a
2 Spectroradiometer Using a Standard Source of Irradiance. ASTM International,
West Conshohocken, PA <www.astm.org>.
The standard percent uncertainty in the voltage is computed by BIPM, IEC, IFCC, ISO, IUPAC, IUPAP, OIML, 1995. Evaluation of Measurement Data -
dividing by the voltage at 18 °C, Eq. (C3): Guide to the Expression of Uncertainty in Measurement. ISO TAG 4, Geneva
<http://www.bipm.org/utils/common/documents/jcgm/JCGM_100_2008_E.
uVtemp pdf>.
uTemp% ¼  100 ðC3Þ Cannon, T.W., 1986. Spectral solar irradiance instrumentation and measurement
VT18 techniques. Solar Cells 18, 233–241.
Eppeldauer, G.P. et al., 2009. Radiometer standard for absolute responsivity
The standard percent uncertainty in the voltage due to a 0.1 °C calibrations from 950nm to 1650nm with 0.05% (k = 2) uncertainty.
uncertainty in the temperature is shown in Fig. C2. The uncertainty Metrologia 46 (2009), S139–S145.
results shown in Fig. C2 apply only to temperatures in the imme- Habte, A., Andreas, A., Reda, I., Campanelli, M., Stoffel, T., 2013. Uncertainty Analysis
of Spectral Irradiance Reference Standards Used for NREL Calibrations. NREL/TP-
diate vicinity of 18 °C. If the device were to be used at a different
5500-58617 (adapted with permission).
operating temperature, the uncertainty at that new temperature Habte, A. et al., 2014. Spectroradiometer Intercomparison and impact on
would need to be recomputed. Only at wavelengths greater than characterizing photovoltaic device performance. In: Conference Paper SOLAR
2014.
900 do the uncertainties in the voltage increase by an appreciable
Hanssen, L., 2001. Integrating-sphere system and method for absolute
amount. These results are in agreement with the results reported measurement of transmittance, reflectance, and absorbance of specular
in the LI-COR 1800 user manual (LI-COR Inc., 1989, pp. 2–4). samples. Appl. Opt. 40 (19), 3196–3204. 1 July 2001.
For spectroradiometers used in outdoor fieldwork that do not LI-COR Inc, 1990. LI-1800-02 Optical Radiation Calibrator Instruction Manual.
Lincoln, Nebraska <https://www.licor.com/documents/
have a temperature control unit there will be significantly larger zx7hyrmk072qqiye1pvl>.
changes in temperature than the 0.1 °C used in this study. This will LI-COR Inc., 1989. LI-1800 Portable Spectroradiometer Instruction Manual. LICOR
cause significantly larger uncertainties in the measured irradiance. Inc., Lincoln, Nebraska.
Marion, B., 2012. Influence of atmospheric variations on photovoltaic performance
For such instruments, a similar methodology can be used to calcu- and modeling their effects for days with clear skies. In: 38th PVSC.
Martinez-Lozano, J.A. et al., 2003. Intercomparison of spectroradiometers for global
and direct solar irradiance in the visible range. Am. Meteorol. Soc. 20, 997–
1100.
Mavromatakis, F., Vignola, F., 2016. Spectral performance of PV modules of different
technologies. In: 11th International Conference on Deregulated Electricity
Market Issues in South Eastern Europe.
Myers, D.R., 1989. Estimates of uncertainty for measured spectra in the SERI
spectral solar radiation data base. Sol. Energy 43 (6), 347–353.
Myers, D., 1997. Radiometric Instrumentation and Measurements Guide for
Photovoltaic Performance Testing. National Renewable Energy Laboratory,
Golden, Colorado. NREL/TP-560-21774 April 1997 <http://www.nrel.gov/docs/
legosti/old/21774.pdf>.
Reda, I., Myers, D., Stoffel, T., 2008. Uncertainty estimate for the outdoor calibration
of solar pyranometers: a metrologist perspective measure. NCSLI J. Measur. Sci.
3 (4), 60 (NREL Report No. JA-581-41370).
Riordan, C. et al., 1989. Spectral solar radiation data base at SERI. Sol. Energy 42 (1),
67–79.
Wilcox, S., Andreas, A., Reda, I., Myers, D., 2002. Improved methods for broadband
outdoor radiometer calibration (BORCAL). In: Proceedings of the ARM Science
Team Meeting, St. Petersburg, Florida, April 2002.
Yoon, H.W., Gibson, C.E., 2011. Spectral Irradiance Calibrations NIST Special
Publication 250-89. National Institute of Standards and Technology,
Gaithersburg, Maryland, pp. 18–22 <http://www.nist.gov/calibrations/upload/
sp250-89.pdf>.
Yoon, H.W. et al., 2012. The distance dependences and spatial uniformities of
Fig. C2. Uncertainty in the voltage due to a 0.1 °C temperature variation in the spectral irradiance standard lamps. In: Earth Observing Systems XVII
vicinity of 18 °C. (Conference 8510) SPIE Vol. 8510, 85100D.

Das könnte Ihnen auch gefallen