Plain Language Summary
The Sun’s irradiance spectrum is used in many applications, such as
constraining the solar forcing in climate models and converting measured
satellite radiance to reflectance. A growing body of literature has
provided evidence that the currently available solar reference spectra
differ by more than their reported uncertainties. Such differences lead
to biased results when different reference spectra are adopted in the
aforementioned applications. This motivates our work to provide a new
high-resolution solar reference spectrum at higher accuracy than any
previously reported. Our ability to produce such a dataset is due to the
state-of-the-art measurements of the Sun’s irradiance spectrum made
since March 2018 by the next-generation Spectral Irradiance Monitor
(SIM) instrument on the Total and Spectral Solar Irradiance Sensor-1
(TSIS-1) satellite mission and the Compact SIM (CSIM) technology
demonstration mission. The TSIS-1 SIM and CSIM have order-of-magnitude
reduction in uncertainty relative to predecessor instruments primarily
because of a first-of-its-kind spectral radiometric calibration facility
capable of characterizing the instruments to higher fidelity. We develop
this new, high-resolution, solar irradiance reference spectrum by
adjusting high spectral resolution solar line data to the irradiance
scale of the more accurate, but lower spectral resolution, TSIS-1 SIM
and CSIM observations.
1 Introduction
Reference solar irradiance spectra have broad utility in atmospheric
science and climate applications. For example, the solar spectral
irradiance (SSI) is used to convert measured satellite radiance to
reflectance (e.g., Wielicki et al., 2013) and as the upper boundary
condition in radiative transfer models used, for example, in remote
sensing algorithms and renewable energy research (e.g., Berk et al.,
2014; Apell & McNeill, 2019). Some instruments use solar absorption
lines for wavelength calibration (e.g., Kang et al., 2020). Some
instruments also use the Sun for radiometric stability monitoring, which
requires a baseline solar spectrum to quantify instrumental changes
against (e.g., Pan & Flynn, 2015). Instruments that monitor radiometric
calibration stability relative to the moon (e.g., Werdell et al., 2019)
indirectly rely on a solar reference spectrum to convert lunar radiance
to reflectance using, for example, the RObotic Lunar Observatory (ROLO)
model (Kieffer & Stone, 2005). Solar reference spectra also constrain
solar irradiance variability models (e.g., Coddington et al. 2016),
which climate models use to specify solar forcing of climate change
(e.g., Kunze et al., 2020).
Various solar reference spectra exist for these applications. Some are
from direct solar irradiance observations from one or more satellite
instruments. These have relatively high reported accuracy and relatively
low (0.1 nm or poorer) spectral resolution compared to ground-based
observations and are typically specific to certain solar activity levels
(e.g., Thuillier et al., 2004). Other solar reference spectra are
constructed by normalizing high spectral resolution solar lines to a
higher accuracy, lower resolution, spectrum (e.g., Dobber et al., 2008).
Still others are created by concatenating independent datasets from
different spectral regions (e.g., Gueymard, 2003). Disagreements have
been identified between the available solar reference spectra and
independent measurements that exceed quoted accuracies particularly at
near-infrared wavelengths where 8% differences have been reported
(e.g., Elsey et al., 2017).
Since March 2018, NASA’s Total and Spectral Solar Irradiance Sensor-1
(TSIS-1) Spectral Irradiance Monitor (SIM) hosted on the International
Space Station (ISS) has observed SSI with lower radiometric uncertainty
(< 0.3%) over the majority of the spectrum than that attained
by previous instruments (Richard et al., 2020). Since 2019, independent
SSI observations have also been made by the CubeSat Compact SIM (CSIM)
instrument (Richard et al., 2019; Tomlin et al., 2020). CSIM
observations span 200 nm to 2800 nm, thereby extending further into the
infrared than the TSIS-1 SIM that spans 200 nm to 2400 nm. A mutual
validation of the TSIS-1 SIM and CSIM irradiance scales was demonstrated
by < 1% disagreement in concurrent observations (Stephens et
al., 2020). In this work, we produce a new reference spectrum, theTSIS-1 Hybrid Solar Reference Spectrum (TSIS-1 HSRS ), by
adjusting high spectral resolution solar line data to the SI-traceable
irradiance scale of the TSIS-1 SIM and CSIM instruments. Such an
approach is necessary because the technology does not exist to measure
the Sun’s spectrum over a broad spectral range from a single instrument
with both high accuracy and high (0.01 nm or finer) spectral resolution.
Figure 1 shows spectral differences of three solar irradiance reference
spectra to TSIS-1 SIM of order 10% in portions of the spectrum and that
cannot be explained by differences in solar activity. While ultraviolet
solar cycle variability reaches 10% at 200 nm, it drops to 5% by 210
nm and reduces even further to ~1% by 300 nm with the
exception of the Mg II line near 280 nm. Visible and near-infrared solar
cycle variability is on the order of 0.1% or less (Ermolli et al.,
2013). The ATLAS-3 spectrum (Thuillier et al., 2004), perhaps the most
widely-used solar reference in Earth science applications, is a
composite of observations from November, 1994 by five different
instruments including the SOLar SPECtrometer (SOLSPEC). Additionally,
high resolution modeled solar absorption features from Kurucz (1995)
were inserted into the lower resolution observations from the visible
through the near-infrared. Reported ATLAS-3 uncertainties are 2-3%.
Another solar reference spectrum is the Laboratory for Atmospheric and
Space Physics (LASP) Whole Heliospheric Interval (WHI) (Woods et al.,
2009). The LASP WHI is a composite of observations from April, 2008 with
the majority of the spectrum measured by the SOLSTICE and SIM
instruments on the SORCE satellite. Observations from SORCE SIM, the
predecessor to the TSIS-1 SIM, were adjusted by up to +8% for
wavelengths above 1350 nm to agree with the ATLAS-3 spectrum in a
recalibration that has been discussed with reference to a systematic
bias (Harder et al., 2010). Therefore, the LASP WHI and ATLAS-3
reference spectra are not independent above 1350 nm. Reported LASP WHI
uncertainties are 1-3% for wavelengths above 300 nm. The SOLAR-ISS
version 2 reference spectrum (Meftah et al., 2020) is from a newer
version of the SOLSPEC instrument (Thuillier et al., 2009) integrated on
the ISS from 2008 to 2017. The SOLAR-ISS reference irradiance baseline
spectrum is from April, 2008 for wavelengths spanning 165-656 nm and an
average over a six year period at wavelengths above 656 nm. Revised
engineering corrections, improved calibrations, and advanced thermal and
degradation corrections are reported as the reason for the changes in
baseline between the earlier ATLAS-3 composite spectrum and the newer
SOLAR-ISS spectrum (Bolsée et al., 2017). However, particularly in the
near-infrared, a thorough understanding of the offset remains under
study (Thuillier et al., 2013). Similar to ATLAS-3, higher spectral
resolution lines have been incorporated into SOLAR-ISS. The mean
reported SOLAR-ISS uncertainty from 165 to 3000 nm is 1.26%, with
uncertainties as low as 0.4-0.6% between 800 to 1700 nm and reaching,
or exceeding, 2% below 400 nm and above 2200 nm. Hilbig et al. (2018)
further summarize these and other solar reference spectra.
The methodology to develop the HSRS is described in Section 2 and the
datasets are described in Section 3. In Section 4, we present results of
our uncertainty assessment and comparison to independent datasets.
Concluding statements follow in Section 5.
2 Methodology
We develop the HSRS using a modified version of the spectral ratio
method. In this method, a wavelength-dependent scaling factor adjusts
high spectral resolution datasets (\(\mathbf{\beta}\)) to match a lower
resolution but higher accuracy spectrum (\(\mathbf{\alpha}\)). The
scaling factor, Q , is the ratio of the \(\mathbf{\alpha}\) and\(\mathbf{\beta}\) datasets after first convolving both to the same
spectral resolution and interpolating to a common sampling grid. The\(\mathbf{\alpha}\) and \(\mathbf{\beta}\) datasets are described in
Section 3.
Typically, Q is derived after a single-step convolution (Eq. 1)
of the \(\mathbf{\beta}\) dataset with the instrument line shape of the\(\mathbf{\alpha}\) dataset (\(\mathbf{\text{ILS}}_{\mathbf{\alpha}}\))
that degrades the resolution of the \(\mathbf{\beta}\) dataset
(\(\mathbf{\beta}^{\mathbf{*}}\)) to match that of the\(\mathbf{\alpha}\) dataset (e.g., Kang et al., 2017; Dobber et al.,
2008). Instead, we derive Q from a two-step convolution: the
first step is as described by Eq. 1 and the second step degrades both\(\mathbf{\beta}^{\mathbf{*}}\) and \(\mathbf{\alpha}\) datasets to a
common spectral resolution (denoted \(\mathbf{\beta}^{\mathbf{**}}\) and\(\mathbf{\alpha}^{\mathbf{**}}\), respectively) that is coarser than
that of the original \(\mathbf{\alpha}\) dataset. We accomplish this
with a Gaussian filter (\(\mathbb{N}\)) of specified standard deviation
(\(\mathbf{\sigma}\)) (Eq. 2). The two-step convolution reduces the
impacts of any uncertainty in \(\mathbf{\text{ILS}}_{\mathbf{\alpha}}\)on Q (Eq. 3), where the subscript \(\mathbf{\ddagger}\) denotes
an interpolation of the \(\mathbf{\beta}^{\mathbf{**}}\) dataset to the\(\mathbf{\alpha}^{\mathbf{**}}\) sampling grid. Finally, the adjusted\(\mathbf{\beta}\) dataset (denoted by \(\mathbf{\Upsilon}\)) is
computed from the product of the native \(\mathbf{\beta}\) dataset andQ , where the subscript † denotes an interpolation of Q to
the native \(\mathbf{\beta}\) sampling grid (Eq. 4).