Calibrated spectral radiance measurements from multispectral and imaging spectrometer instruments are a baseline requirement for producing geophysical data products that can be used to study Earth's land, ice, water, and atmospheric environments (Green, 1998; Green et al., 2006; King et al., 1996; Schaepman-Strub et al., 2006; Thome, 2001; Vane et al., 1993). Optical instrument calibration is based on a traceable radiance standard determined by the National Institute of Standards and Technology (NIST) in the United States, for example, where radiance measurements are collected from a stable illumination source in a controlled laboratory environment (Chrien et al., 1990; Schaepman and Dangel, 2000; Strobl et al., 1997; Tansock et al., 2015; Parr and Datla, 2001). Using this stable NIST-traceable source, periodic assessments of an optical instrument's response are made to monitor its long-term repeatability, mechanical functionality, and responsivity to variable light intensities. While radiometric calibration is fundamental to spectral instrument data acquisition, this is especially critical for missions bound for deployments in polar regions because the range of measured snow, ice, and liquid water surfaces spans the entire solar spectrum dynamic range. For airborne missions, precise and accurate pre-flight, in-flight, and post-flight calibration procedures are therefore of paramount importance to achieve targeted instrument stability and measurement requirements. Commitment to characterize instrumentation, instrument foreoptics, and supporting aircraft hardware during pre- and post-airborne mission timelines helps to produce remote sensing measurements in which uncertainty has been quantified and fully calibrated data products are available to support algorithm development and remote sensing science applications.

In this paper, we describe NIST-traceable laboratory, in-flight, and field radiometric calibration procedures necessary to obtain science-quality measurements from a visible-to-shortwave infrared (VSWIR) non-imaging airborne spectrometer. We used the MODerate resolution atmospheric TRANsmission (MODTRAN^®) code version 6.0 (Berk et al., 2005) to assess the measurement performance of the airborne nadir-viewing spectrometer over bright Greenland ice and dark bare rock/soil targets during a 29 July 2015 in-flight radiometric calibration experiment. Prior to this Greenland campaign, MODTRAN's capability, usefulness, and performance with regard to Arctic and Greenland ice sheet airborne remote sensing science remained largely unrealized. Thus, application of MODTRAN radiative transfer in this paper to evaluate VSWIR remote sensing measurement performance is both forward thinking and advances atmospheric measurements and modeling for the Arctic region in particular. Two non-imaging airborne spectrometers were flown as a part of the Slope Imaging Multi-polarization Photon-Counting Lidar (SIMPL)/Advanced Visible Infrared Imaging Spectrometer-Next Generation (AVIRIS-NG) 2015 airborne campaign to northwest Greenland in July and August 2015 (Brunt et al., 2015). The nadir-viewing spectrometer's objective was to acquire non-imaging profile measurements of snow, ice, and liquid water radiance, and the zenith-viewing spectrometer's objective was to characterize sky conditions during nine science flights. Non-imaging profile measurements are defined as along-track radiance spectra of the surface directly below the aircraft within the airborne spectrometer's instantaneous field of view (IFOV). The campaign was conducted in support of the Ice, Cloud, and land Elevation Satellite 2 (ICESat-2) mission launched on 15 September 2018. ICESat-2, a follow-on laser altimeter mission to ICESat (Schutz et al., 2005; Zwally, 2002), will continue measurements of ice sheet elevation and change, sea ice thickness, ocean surface height, land topography, vegetation height and structure, and atmospheric clouds and aerosols. The Geoscience Laser Altimeter System (GLAS) (Abshire et al., 2005) on the ICESat mission used a traditional single-beam, near-infrared (NIR) (1064 nm), analog waveform method for the surface altimetry measurements. The Advanced Topographic Laser Altimeter System (ATLAS) (Abdalati and Zwally, 2010; Markus et al., 2017) on ICESat-2 will use a more efficient measurement producing multiple beams using a green (532 nm) micropulse, photon-counting approach.

In order to prepare for the ICESat-2 mission, the Greenland campaign was conducted to better understand how ATLAS will represent the height, roughness, and topography of snow and ice surfaces to determine the spatial extent, and potentially the depth, of meltwater on the ice sheet and sea ice surface. Four instruments were flown, two of which included non-imaging airborne spectrometers. While the dual non-imaging airborne spectrometer integration was considered experimental to the Greenland campaign's overall mission objective, this calibrated instrument configuration did provide a low-cost airborne payload because of its size, weight, and power to support ICESat-2's efforts to characterize green laser pulse penetration into snow and ice based on the known reliability to retrieve information on surface contaminants, grain size, and liquid water from VSWIR spectra. Furthermore, the radiometric calibration and traceability of VSWIR measurements acquired by the nadir-viewing spectrometer during airborne remote sensing flights reflect a unique contribution to Arctic and Greenland ice sheet remote sensing science by establishing standard calibration/validation practices for future airborne polar region campaigns.

The non-imaging airborne spectrometers and SIMPL (Dabney et al., 2010; Harding et al., 2011) were flown together on a National Aeronautics and Space Administration (NASA) Langley Research Center King Air (hereinafter UC-12B). SIMPL uses a micropulse, photon-counting, multi-beam measurement like that of ATLAS but provides added information about light scattering by using co-aligned green and NIR laser pulses and a measure of pulse depolarization. AVIRIS-NG (Hamlin et al., 2010) was flown on a King Air (C-12) operated by Dynamic Aviation. Snow radiative transfer modeling (Aoki et al., 2000; Bohren and Barkstrom, 1974; Libois et al., 2013, 2014; Painter and Dozier, 2004a; Picard et al., 2009; Warren, 1982; Wiscombe and Warren, 1980; Kokhanovsky and Zege, 2004) and VSWIR spectroscopy studies have shown that optical snow surface reflectivity is most sensitive to concentrations of light-absorbing particles (e.g., dust, soot, and black carbon containments) at visible wavelengths (Aoki et al., 2000; Dozier et al., 2009; Painter et al., 2007, 2009, 2013; Warren, 2013; Warren and Wiscombe, 1980), whereas effective snow surface grain size is a measure of melt state, which can be quantified by exploiting the position, depth and shape of spectral absorption by liquid water within near-infrared wavelengths (Clark and Roush, 1984; Dang et al., 2016; Dozier and Painter, 2004; Gardner and Sharp, 2010; Green et al., 2006; Libois et al., 2013, 2014; Nolin and Dozier, 2000; Painter et al., 1998, 2009; Warren et al., 2006; Wiscombe and Warren, 1980).

Because the ATLAS green laser pulses may penetrate into snow and ice, to a significant depth to cause surface height measurements to be biased low, the primary objective of the SIMPL/AVIRIS-NG 2015 Greenland campaign was to obtain the necessary remote sensing measurements to enable the ICESat-2 project to determine if green light depth of penetration, measured by SIMPL, is correlated with surface grain size, contaminant, and/or wetness properties determined using VSWIR spectra. A comparison of green laser pulse shape broadening caused by volume scattering in snow, ice, and liquid water, as compared to NIR pulses that only undergo surface scattering, provides the measurement of penetration depth. If that depth is correlated with any particular surface property, changes in those properties seasonally and/or interannually could potentially cause bias in rates of ice sheet elevation change from ICESat-2 retrievals. The nadir-viewing spectrometer optical head was mounted inside SIMPL and their IFOVs were aligned to ensure the spectroscopic and altimetry profile measurements were coincident, observing the same surface location at the same time through the same atmospheric column. AVIRIS-NG followed the SIMPL flight path at a higher altitude and trailing by about 15 min. Flying with AVIRIS-NG was important because its estimations of grain size, contaminant concentrations, and wetness are relatively mature and by imaging a swath, it provides information about the spatial variability of these surface properties.

The non-imaging airborne spectrometer integration on the UC-12B included a nadir-viewing spectrometer measuring upwelling spectral radiance (W m-2 sr-1 nm-1, where sr is the FOV full angle), and a zenith-viewing spectrometer measuring downwelling spectral irradiance (W m-2 nm-1). We predicted spectral radiance for the nadir-viewing spectrometer over bright Greenland ice and dark bare rock/soil targets using MODTRAN to determine whether airborne measurement performance was within the targeted 5 % requirement. MODTRAN inputs included a sub-Arctic summer (geographical–seasonal) model, Navy maritime aerosol profile, top-of-atmosphere (TOA) solar irradiance spectrum, CIMEL Sun photometer atmospheric measurements of aerosol optical depth and columnar water vapor as part of the AErosol RObotic Network (AERONET) (Holben et al., 1998), nadir-viewing spectrometer spectral response functions, and line-of-sight (LOS) geometries. For the MODTRAN-predicted measurement comparison, we selected flight segments from the 29 July in-flight radiometric calibration experiment that was intended to optimize the nadir-viewing spectrometer's visible–near-infrared (VNIR) integration time and shortwave infrared (SWIR) gains across the full solar spectrum dynamic range. Along the northern portion of the UC-12B 29 July flight line over the Greenland ice sheet interior, the Landsat 8 Operational Land Imager (OLI) acquired a coincident multispectral image.

We exploited this Landsat 8 OLI image acquisition by predicting TOA spectral radiance for OLI using identical MODTRAN parameterization as constructed for the nadir-viewing spectrometer. Because Landsat is a well-regarded standard for optical satellite remote sensing calibration/validation (Markham and Helder, 2012), we felt it was important to evaluate the nadir-viewing spectrometer's bright Greenland ice measurement performance along with Landsat 8 OLI as an additional comparison step. Landsat's capabilities to measure polar regions since the launch of Landsat 8 in February 2013 have been unprecedented because of onboard instrument performance and changes to its long-term acquisition plan that include imaging of all sunlit land and nearshore coastal regions greater than 5∘ solar elevation. Imaging higher latitudes and polar ice sheets in solar-reflected wavelengths is complicated by low solar illumination angles, surface bidirectional reflectance distribution function (BRDF) effects (Aoki et al., 2000; Hudson et al., 2006), and persistent cloudiness with cloud shadows cast on the ice sheet (Choi and Bindschadler, 2004; Hudson and Warren, 2007). Yet, because Landsat's orbital tracks converge at the poles, swath imaging side lap results in much higher temporal imaging frequency than tropical and middle latitude regions. Up until this Greenland campaign, Landsat 8 OLI's capability to measure Greenland ice sheet upwelling radiance had not been fully assessed, because for the first time in the Landsat mission's measurement history, Landsat 8 OLI does not saturate over homogenous bright targets at high latitudes due to substantially improved radiometry over prior instruments. Therefore, an important outcome of this paper is our ability to establish Landsat 8 OLI's radiometric performance over Greenland by comparing to MODTRAN-predicted upwelling radiance using the nadir-viewing airborne spectrometer's ice sheet apparent reflectance as the reference spectra.

The specific objectives of this paper are to (1) describe the non-imaging airborne spectrometer integration and NIST-traceable radiometric calibration procedures for pre-flight, in-flight, and post-flight time frames; (2) describe the equations necessary to calculate the nadir-viewing spectrometer ground IFOV footprint; (3) characterize downwelling spectral irradiance measurements to screen for cloud-contaminated data to support atmospheric compensation modeling for clear-sky observational conditions; and (4) compare the nadir-viewing spectrometer's measurement performance over bright Greenland ice and dark bare rock/soil targets against MODTRAN and a coincident Landsat 8 OLI image acquisition.

A method to radiometrically calibrate, deploy, and assess measurement performance of a non-imaging airborne spectrometer to measure the Greenland ice sheet surface has been presented. This NIST-traceable calibration included traceable laboratory, in-flight, and field procedures to fully characterize spectrometers, their foreoptics, and their measurements. The nadir-viewing spectrometer's stability was determined to be within 2 % using a NIST-traceable source, and well within the targeted 5 % spectral radiance requirement for the airborne mission. The point spread function and IFOV footprint of the nadir-viewing spectrometer's 1∘ foreoptic was measured to enable direct comparison to SIMPL's green and NIR polarimetric lidar measurements, AVIRIS-NG's VSWIR measurements, and other on-orbit satellite measurements such as Landsat, for example. The 29 July in-flight radiometric calibration experiment over Greenland bright and dark targets proved to be invaluable for optimizing the nadir-viewing spectrometer's measurement capabilities during the airborne campaign, as well as evaluating in-flight measurement performance across the full solar spectrum dynamic range using MODTRAN and atmospheric measurements from both ground and satellite instruments. The main objective of measuring spectral irradiance with a zenith-viewing spectrometer and remote cosine receptor optic was to characterize in-flight sky conditions. Even though the zenith-mounted OrangeCan on top of the UC-12B aircraft limited the hemispherical IFOV, these measurements are useful for screening out cloud-contaminated flight data that will expedite identification of clear-sky VSWIR data that can be used to address airborne mission objectives.

With no ground calibration/validation in situ measurements on the Greenland ice sheet, or ship campaign on the open ocean, we had to develop an alternative approach to compare the nadir-viewing spectrometer's measurement performance against an atmospheric radiative transfer model. By identifying homogenous bright Greenland ice and dark bare rock/soil flight segments on 29 July, we were able to assess airborne measurement performance with MODTRAN over both low- and high-radiance targets (e.g., Moran et al., 1995) under similar atmospheric and solar illumination conditions. We used apparent airborne reflectance spectra for both bright and dark targets to predict spectral radiance for the nadir-viewing spectrometer (Fig. 12a and b), and then compared predictions with measured spectral radiance (e.g., Green, 2001; Slater et al., 1987; Thome, 2001; Vane et al., 1993). Our MODTRAN predictions indicate that the nadir-viewing spectrometer VNIR and SWIR1 detectors measured bright Greenland ice with an average uncertainty between 2.5 % and 4.7 % for VSWIR wavelengths with greater than 80 % atmospheric transmittance (Fig. 12c). For dark bare rock/soil, the nadir-viewing spectrometer VNIR and SWIR1 detectors' measurement uncertainty was between 0.6 % and 1.2 % on average (Fig. 12d). As stated earlier, UC-12B optical window transmission beyond 2.0 µm was more uncertain and was evident when evaluating the SWIR2 detector data. For bright Greenland ice and dark bare rock/soil, the nadir-viewing spectrometer's measurement uncertainty for the SWIR2 detector was on average 4.3 % and 19.7 %, respectively (Fig. 12c and d).

MODTRAN predictions for assessing airborne spectrometer measurement performance are, in part, dependent on the quality of the surface reflectance spectra and availability of atmospheric measurements near the target measurement performance location. Fortunately, for this airborne campaign, baseline atmospheric measurements were accessible via the Thule Air Base CIMEL as part of AERONET. It is clear that spatial proximity to a CIMEL matters in terms of in-flight atmospheric aerosols and columnar water vapor concentrations because we observed less measurement uncertainty for the closer dark bare rock/soil target when compared to the bright Greenland ice target much further away. Interestingly, we found that the nadir-viewing VSWIR spectra for bright Greenland ice in the interior were much more sensitive to columnar water vapor concentrations than aerosols. This result caused us to evaluate the nadir-viewing spectrometer's measurement sensitivities to a variety of input satellite atmospheric water vapor products. Narrowing in on 0.94 and 1.13 µm water vapor absorption lines uncovered the spread in satellite-retrieved daily atmospheric water vapor over the Greenland interior. We were able to identify that the MODIS Aqua low mean atmospheric water vapor product is most suitable to ingest when processing the UC-12B science flight data for MODTRAN-based atmospheric compensation. The daily MODIS Aqua overpass times generally align well with UC-12B flight times during airborne science flights. The MODIS Aqua low mean atmospheric water vapor retrievals are designed to partition columnar water vapor concentrations between the surface and 680 mbar (see details at https://modis-atmosphere.gsfc.nasa.gov/documentation/collection-6.1, last access: 12 February 2018), which is within the atmosphere boundary layer.

As an additional airborne spectrometer performance comparison over the Greenland ice sheet, we used a Landsat 8 OLI coincident image acquired within ∼3 min of the UC-12B bright Greenland ice target flight segment (Fig. 13a). We predicted Landsat 8 OLI TOA spectral radiance using MODTRAN with the following parameters: solar illumination geometry, OLI viewing geometry, the same atmospheric inputs used for the airborne nadir-viewing spectrometer assessment, OLI relative spectral response functions (Fig. 13b), and the apparent airborne reflectance spectrum for bright Greenland ice (Fig. 13b). By comparing MODTRAN-predicted and measured Landsat 8 OLI TOA spectral radiance (Fig. 13c), we found that Landsat 8 OLI is measuring between 6 % and 16 % more TOA spectral radiance from the Greenland ice sheet with VNIR and SWIR1 spectral bands than was predicted with the nadir-viewing spectrometer's apparent airborne reflectance spectrum (Fig. 13d). It is important to note that Landsat 8 OLI's pixel-level LOS imaging is highly accurate over Greenland due to spacecraft geolocation (Storey et al., 2014), and that we accounted for cross-track imaging effects in MODTRAN using NIR spectral band LOS geometry.

Landsat 8 OLI is a well-characterized instrument on both pre- and post-launch timescales with exceptional on-orbit performance since 2013 (Markham et al., 2014; Morfitt et al., 2015). Routine onboard diffuser, lunar, and vicarious calibrations, over midlatitude pseudo-invariant calibration sites in particular, are conducted to track OLI's instrument performance and degradation while in orbit (Helder et al., 2010, 2013; Mishra et al., 2014). We speculate that differences between predicted and measured Landsat 8 OLI TOA spectral radiance over the Greenland ice sheet presented in this paper are possibly a byproduct of both techniques used to derive OLI gain coefficients over midlatitude desert sites with stable dry atmospheres, and VNIR differences between the Kurucz and Chance Kurucz (ChKur) reference TOA solar irradiance spectrums (Chance and Spurr, 1997; Kurucz, 2005) used for airborne spectrometer and Landsat 8 OLI radiometric calibration/validation. Nevertheless, more investigation is required, and looking ahead, Greenland and Antarctic ice sheets need to receive expanded calibration/validation consideration when characterizing and monitoring on-orbit satellite instrument performance, as has been attempted for other Earth observing systems (Cao et al., 2010; Six et al., 2004). The airborne method of calibration/validation presented here, including the rigorous laboratory NIST-traceable radiometric calibration, is put forth as an option to augment polar ice sheet calibration/validation.

Landsat 8 OLI's capabilities to measure Greenland and Antarctic ice sheets has advanced since 2013 thanks to revisions in its higher latitude and polar image frequency (Fahnestock et al., 2016). While Landsat 8 OLI measurements are providing new insights and applications for polar ice sheet science, specifically superglacial lake and ice velocity mapping (Alley et al., 2018; Gardner et al., 2018; Pope et al., 2016), results from this study suggest that the Greenland ice sheet surface may be less reflective than what is currently being measured by Landsat 8 OLI at TOA. Thus, Landsat 8 OLI reflectance-based interpretations of ice sheet surface properties and change should remain cautious until additional measurement validation is undertaken. We anticipate that airborne VSWIR measurements acquired over Arctic and Antarctic domains, as demonstrated during this airborne campaign, offer a cost-effective approach to validate medium-resolution atmospherically corrected products from Landsat class instruments well above ±60∘ latitude while also supporting algorithm development to derive higher-level information products on cryospheric surface properties and conditions.

It has been suggested that optical remote sensing instruments must be able to measure the ice sheet surface at an uncertainty of 2 % or less to distinguish between the presence of light-absorbing constituents and other factors controlling VSWIR ice sheet albedo (Warren, 2013). For airborne and on-orbit satellite instruments, this stringent of a measurement requirement demands careful instrument radiometric calibration and characterization and could remain difficult to achieve for polar atmospheres because of atmospheric measurement uncertainty and the ability to compensate for such effects. One strategy for future optical satellite instrument design is to address aerosol and columnar water vapor effects along with VSWIR surface retrieval uncertainties over polar regions by requiring an expanded set of measurements in spectral regions where atmospheric scattering and absorption dominate the VSWIR remote sensing signal. Requiring these atmospheric measurements would offer the capability to retrieve atmospheric parameters directly from the remote sensing measurement itself, rather than ingesting ancillary data from other sources with different spatial resolutions and temporal sampling frequencies.

This initial effort to describe and document the traceable laboratory radiometric calibration and in-flight measurement performance of the non-imaging airborne spectrometer configuration flown as part of the SIMPL/AVIRIS-NG 2015 Greenland campaign indicates that the nadir-viewing spectrometer was able to achieve its targeted VSWIR measurement requirement for the airborne mission when compared against MODTRAN. Compared to instruments with a spectral resolution greater than or equal to 10 nm, the 1 nm spectral resolution of these airborne VSWIR snow and ice measurements may more effectively isolate the exact wavelength center where green light depth of penetration and surface contaminants interact, and where the depth, area, and asymmetry of near-infrared liquid water absorption can be inverted into a measure of surface grain size while also detecting the presence of liquid water. As a result, airborne VSWIR data products from UC-12B science flights are of sufficient radiometric traceability and quality to evaluate green laser pulse penetration into Greenland snow and ice, and to compare with other VSWIR remote sensing measurements acquired during the airborne mission time frame.