A Web-Based Tool for Calculating Spectral Band ...

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING

1

A Web-Based Tool for Calculating Spectral Band Difference Adjustment Factors Derived From SCIAMACHY Hyperspectral Data Benjamin R. Scarino, David R. Doelling, Patrick Minnis, Arun Gopalan, Thad Chee, Rajendra Bhatt, Constantine Lukashin, and Conor Haney

Abstract—Monitoring and adjusting calibrations of various satellite imagers is often exacerbated by differences in their spectral response functions (SRFs). To help account for spectral disparities among satellite imagers, a web-based spectral band difference correction calculator has been developed to characterize the relationship between a specified pair of satellite imager channels in the hyperspectral wavelength range of 240–1750 nm. These spectral band adjustment factors (SBAFs) are derived by convolving hyperspectral data from the SCIAMACHY instrument with the SRFs of a reference and target sensor. The SBAF tool can be used for all combinations of instrument/channel pairs over predefined Earth spectra, intercalibration domains, or user-defined spatial domains. Options are available to the user whereby SBAFs can be subsetted by time, angle, and/or precipitable water content. To evaluate the relative spectral calibration of SCIAMACHY, comparisons of SBAFs derived from SCIAMACHY, Hyperion, and Global Ozone Monitoring Experiment-2 (GOME-2) were performed. Using observations over the Libya 4 desert pseudoinvariant calibration site, it is shown that SCIAMACHY-based SBAFs are within 0.1%–0.3% of SBAFs derived from Hyperion or GOME-2. This result implies that spectral calibration differences, i.e., the calibration uncertainties of SCIAMACHY relative to other potential spectral sources, have a minor impact on the SBAF compared with the influence of effective parameter-based subsetting. The SCIAMACHY instrument is most suited for calculating the SBAFs, given its high spectral resolution, broad spectral range, and nearly continuous global availability. The calibration community will find this SBAF tool useful for mitigating the SRF differences that can complicate the comparison and intercalibration of visible and near-infrared sensors. Index Terms—Hyperspectral sensors, spectral band adjustment factor (SBAF), visible imager calibration.

Manuscript received November 19, 2014; revised August 11, 2015 and October 15, 2015; accepted November 11, 2015. This work was supported in part by the NASA Modeling, Analysis, and Prediction Program, by the NASA Satellite Calibration Interconsistency Program, and by the NOAA Climate Data Records Program Agreement IA1-17982. B. R. Scarino, A. Gopalan, T. Chee, R. Bhatt, and C. Haney are with Science Systems and Applications, Inc., Hampton, VA 23666 USA (e-mail: [email protected]; [email protected]; thad.l.chee@nasa. gov; [email protected]; [email protected]). D. R. Doelling, P. Minnis, and C. Lukashin are with the NASA Langley Research Center, Climate Sciences Branch, Hampton, VA 23681 USA (e-mail: [email protected]; [email protected]; [email protected]). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TGRS.2015.2502904

I. I NTRODUCTION

T

HE Global Space-Based Inter-Calibration System (GSICS) is an international organization designed to uniformly calibrate operational satellites to ensure climate-quality-consistent retrievals across all operational sensors [1]. With the advent of well-calibrated infrared (IR) hyperspectral instruments, GSICS has used the Infrared Atmospheric Sounding Interferometer (IASI) and the Atmospheric Infrared Sounder (AIRS) as absolute calibration references to normalize the IR imager radiances from low- and geostationary-Earth-orbit (LEO and GEO) satellites. The calibration difference between IASI and AIRS is less than 0.1 K across most of the hyperspectral channels [1], [2]. The IR hyperspectral sensors are better calibrated than most operational IR imagers. The IASI or AIRS reference calibrations are transferred during Simultaneous Nadir Overpasses (SNOs) by regressing the operational IR imager radiances and IASI hyperspectral radiances convolved with the imager spectral response function (SRF) [2]–[5]. The SRF-convolved hyperspectral radiances eliminate the need to account for spectral band differences. The IASI instrument is onboard both the MetOp-A and MetOp-B satellites and is projected to be launched on MetOp-C [6]. Moreover, there are three IASI nextgeneration instruments planned for the MetOp-Second Generation series [7]. These instruments will allow for a sequence of uninterrupted IASI sensors to be used as a calibration reference and, when properly intercalibrated, will provide a consistent and traceable calibration reference over time. This ensures that all operational IR and sounder sensors calibrated against IASI will yield atmospheric profiles and cloud retrievals that will not differ because of calibration discrepancies. It would be ideal to have a calibration approach, similar to that for IR imagers, for operational imager channels measuring reflected solar radiation. There is, however, no hyperspectral instrument, suitable as a reference calibration, that has the following features: a reflected solar hyperspectral wavelength range of ∼300–2400 nm, continuously scanning and freely available public data, reliable onboard calibration, and is part of a multiple-instrument project designed to provide overlapping temporal coverage. The future CLARREO reflected-solar hyperspectral sensor, having an onboard traceable absolute calibration reference, is perfectly suited to calibrate operational imagers [8]. Currently available, the Envisat Scanning Imaging Absorption Spectrometer for Atmospheric Chartography (SCIAMACHY) visible hyperspectral instrument was designed

0196-2892 © 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. 2

for atmospheric chemistry retrievals and has a spectral range of 240–2380 nm. This single-instrument project provided a limited data record between 2002 and 2012. The Global Ozone Monitoring Experiment-2 (GOME-2) hyperspectral sensor onboard the MetOp series of satellites is similar to SCIAMACHY, with the exception that the spectral range is limited to only 240–790 nm. The Hyperion hyperspectral instrument (400– 2500 nm) was designed for high-spatial-resolution land imagery and was launched in 2000 on Earth Observing-1 (EO-1). Because EO-1 is a tasking satellite, the Hyperion instrument does not scan continuously but rather operates by user request or by preset scheduling for calibration purposes over specific regions. As a result, there may not be Hyperion coverage data for all areas of interest. Most other hyperspectral instruments, such as the Earth Observing System-Aura Ozone Monitoring Instrument, are designed for ozone retrievals and are spectrally limited to wavelengths shorter than 500 nm. Because the current suite of hyperspectral instruments does not have adequate calibration accuracy or traceability, the GSICS community has chosen the Moderate Resolution Imaging Spectroradiometer (MODIS) 0.65-μm channel as the visible reference calibration. Aqua-MODIS has an absolute calibration accuracy of 1.64% and is stable to within 1% per decade [4], [9]. It is anticipated that the Visible Infrared Imaging Radiometer Suite (VIIRS) instruments will have calibration accuracy comparable to that of MODIS [10]. Intercalibration of the MODIS and VIIRS records will provide a visible calibration reference over time. To transfer the MODIS calibration reference to other operational imagers, many strategies are used, as summarized in the work of Chander et al. [11]. These techniques include intercalibration using bore-sighted radiance pairs, as well as invariant Earth targets, including deserts, polar ice, deep convective clouds (DCCs), and clear-sky ocean using either Rayleigh scattering or sunglint. Because MODIS, rather than a hyperspectral sensor, is used as the calibration reference, a spectral band adjustment factor (SBAF) is needed to account for the spectral band differences between the target and reference sensor radiances or reflectances using any of the previously named techniques. These adjustment factors will vary by calibration target based on the reflected scene spectra. The concept of the SBAF is well established in the actively monitored and invariant surface site calibration communities. A surface site is typically characterized using hyperspectral reflectances while the effects of the atmosphere on spectral radiances at the top of the atmosphere (TOA) are taken into account using radiative transfer model (RTM) calculations [12]–[15]. Bright high clouds, with tops in the vicinity of the tropopause, have the smallest and most consistent SBAF adjustments of all Earth scenes for wavelengths less than 1 μm, based on RTM radiances [16]–[18]. As such, Ham and Sohn [19] and Okuyama [20] calibrated the MTSAT-1R instrument using RTM radiances based on coincident MODIS-retrieved cloud properties over bright clouds to accommodate spectral band differences. Doelling et al. [21] intercalibrated GOES-12 and AquaMODIS using ray-matched radiance pairs and land and ocean SBAFs previously derived from SCIAMACHY under shared conditions and over geographical domains similar to those

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING

during intercalibration events and concluded that the GOES-12 calibration gain difference obtained over land and ocean is less than 0.3%, whereas without an SBAF, the difference is 4.7%. To test the effectiveness of the a priori SBAF for raymatching domains and invariant targets, Doelling et al. derived SCIAMACHY-based SBAFs over DCCs and desert scenes as well as over MODIS/GEO ray-matching domains [22], [23]. The SCIAMACHY hyperspectral radiances were then intercalibrated with the Aqua-MODIS 0.65-μm band and were found to be stable to within 0.6% per decade. The DCC, desert, and ray-matched GEO calibration gains were then directly compared with the SCIAMACHY/GEO gains. All gains were tied to the MODIS absolute calibration and were found to be within 1% of each other for the Meteosat-9 0.65-μm channel, i.e., high consistency was achieved among three independent calibration methods owing to the application of the separate SBAFs. This study also concludes that directly transferring the SCIAMACHY calibration using ray matching provided smaller uncertainties than using the Aqua-MODIS-based calibration methods proposed by GSICS or CLARREO. Scarino et al. [24] showed that the difference between DCC and marine-stratuscloud SBAFs for the MODIS 0.65-μm narrowband imager channel and the Meteosat-9 High-Resolution Visible broadband imager channel can be as large as 6%. The spectrally flat DCCs reside near the tropopause above most water absorption, whereas the TOA radiances reflected by low stratus clouds are attenuated by strong near-IR (NIR) water vapor absorption bands. The study emphasized that SBAFs need to be specific for each Earth-target-and-sensor pair because atmospheric absorption and spectral reflectance signatures are unique and scene dependent. These results support the need for a flexible means of calculating SBAFs over various Earth targets. The previously mentioned SBAF studies were based on SCIAMACHY. The SBAFs do not rely on the absolute calibration of SCIAMACHY but rather on the relative spectral calibration between hyperspectral bands. Chander et al. [25] found that Landsat-7 Enhanced Thematic Mapper Plus (ETM+) and MODIS reflectances over the Libya 4 calibration site agreed within 5.5% and 3.3% using SBAFs based on Hyperion and SCIAMACHY, respectively. Those results are based on the assumption that the ETM+ and MODIS sensors are accurately calibrated. It is difficult to conclude which is the more accurate SBAF based on this one example. A more comprehensive case study is warranted in which SBAFs are carefully compared using three independent hyperspectral radiance sources, e.g., SCIAMACHY, Hyperion, and GOME-2. Consistency of SBAFs based on three hyperspectral sensors would validate the relative spectral calibration of SCIAMACHY and the resulting SBAF computations. An online SBAF calculator, based on SCIAMACHY, would then meet the needs of the calibration community. The ideal scenario, however, is to offer users the choice of preferred hyperspectral radiance source. Doing so would allow users to take advantage of the smaller footprint size and extensive global coverage offered by GOME-2 or utilize the 1750–2500-nm spectral range available from Hyperion (see Section II-A). The goal of this paper is to describe an online userinterfaced SBAF tool that can be used to compute SBAFs

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. SCARINO et al.: WEB-BASED TOOL FOR CALCULATING SPECTRAL BAND DIFFERENCE ADJUSTMENT FACTORS

between all available instrument/channel pairs, based on a global SCIAMACHY archive. Section II-A and B describe the SCIAMACHY instrument and Earth spectra scene selection from the SCIAMACHY database. The web tool functionality, SBAF calculation, and a demonstration of parameter subsetting, which can be used to improve the SBAF accuracy, are detailed in Sections II-C and III-A. In Sections II-D and III-B, the consistency of the SCIAMACHY-based SBAFs relative to GOME-2- and Hyperion-based spectral corrections is introduced and evaluated using an August 2007 case study over the Libyan Desert. The conclusions are contained in Section IV. The objective is that this SBAF tool, described as follows, will be an important asset for the calibration community.

II. DATA AND M ETHODOLOGY A. SCIAMACHY Data The European Space Agency developed the Environment Satellite (Envisat; 1 March 2002–8 April 2012) with the primary intention of improving the continuous Earth-observing measurements initiated with the European Remote-Sensing Satellites (ERS and ERS-2). Envisat operates in a 10:00 A . M . sun-synchronous 35-day repeating orbit at an altitude of 790 km. The platform houses an array of nine Earth-observing instruments, including SCIAMACHY—an image spectrometer primarily designed for measuring trace gases and aerosols in the lower atmosphere. SCIAMACHY has a fine spectral resolution across eight channels covering 240–2380 nm. Radiances at most of the visible (VIS) and NIR wavelengths pass through Channels 3–6, which have a spectral resolution of 0.44–1.48 nm covering the wavelength range of 394–1750 nm. Specifically, Channels 1 (240–314 nm), 2 (309–405 nm), 3 (394–620 nm), 4 (604–805 nm), and 5 (785–1050 nm) have spectral resolutions of 0.24, 0.26, 0.44, 0.48, and 0.54 nm, respectively, whereas Channel 6 (1000–1750 nm) has a spectral resolution of 1.48 nm [26]–[28]. To best preserve relative interchannel calibration, overlapping channel spectra are averaged when computing an SBAF. Channels 7 and 8 cover limited spectral ranges and are subject of icing and stray light issues, as well as poor spectral calibration quality [29]. Thus, the tool developed here operates only over the continuous spectral range of Channels 1–6 (240–1750 nm). The broad spectral width and fine resolution in these channels are suitable for constructing SBAFs for most VIS and NIR channel sensors. The SBAF calculation tool described in this paper was compiled using SCIAMACHY Level-1b Version-7.03 radiances from August 2002 through December 2010. In keeping with its mission design, SCIAMACHY alternated its scanning mode between nadir and limb measurements every 7 min [26]. For the purpose of deriving scene-dependent SBAFs, only nadirmode scans can be used. The total nadir scan width of 960 km is divided into four nadir-like rectangular fields of view (FOVs), defined by a 30-km length in the along-track direction and a 240-km across-track width, and comprised of 3200 hyperspectral measurements. Level-1b processing allows for higher spatial resolution (30 km × 60 km) at the cost of reduced channel portions, but the combined continuous Channel 1–6

3

spectra can only be achieved when using the complete 30 km × 240 km footprint observed over a 1-s integration time. These footprints are situated end to end with two on either side of the ground track within a 30◦ view angle—two interior near-nadir FOVs and two exterior off-nadir FOVs. The center viewing zenith angles (VZAs) are ∼7.5◦ and ∼9.7◦ for the interior FOVs and ∼24.9◦ and ∼27.1◦ for the exterior FOVs. The SCIAMACHY instrument also performed operational direct solar measurements for degradation monitoring and has absolute on-orbit reflectance calibration accuracy of 2% for Channel 4 and 6% for Channel 5. The calibration accuracy of Channels 3 and 6 is between 2% and 6% [29]. Doelling et al. [22] found that the SCIAMACHY hyperspectral radiances convolved with the Aqua-MODIS 0.65-μm SRF were consistent within 0.44% of Aqua-MODIS over seven years. A selection of ancillary data is available for each SCIAMACHY FOV. The UTC scan-line time is provided for each footprint, along with the VZA, the solar zenith angle (SZA), and the solar azimuth angle (SAA) at the center of the FOV. The Clouds and the Earth’s Radiant Energy System (CERES) 20-km Level-2 Single Scanner Footprint (SSF) Edition 2 cloud retrievals from Terra-MODIS data help define the composition of the scene in each FOV [30]. The cloud properties (i.e., cloud amount, optical depth, phase, cloud top temperature, etc.) were matched with each SCIAMACHY FOV. The MODIS data were nominally acquired ∼30 min after a given Envisat overpass. Precipitable water (PW) for the entire atmospheric column was taken from the Goddard Earth Observing System Model Version 4 (GEOS-4) analyses. The land coverage within each CERES footprint was determined from the International Geosphere-Biosphere Programme. Land coverage and all cloud and column properties, along with latitude/longitude coordinates, are available at the center and four corners of each 30 km × 240 km FOV, thereby allowing SCIAMACHY footprint data to be easily organized into appropriate subsets, such as desert, DCC, clear-sky ocean, and all-sky tropical land or ocean. B. SCIAMACHY FOV Selection Obtaining the correct Earth target site or scene spectra for intercalibration is critical. The importance of proper scene identification is highlighted by Scarino et al. [24], who, for a narrowband-to-broadband imager SBAF example, showed that the calibration correction needed to account for absorption discrepancies alone is −4.5% over the Libya 4 pseudoinvariant calibration sites (PICS) compared with +3.2% over DCC scenes. Chander et al. [25] noted the importance of deliberate scene selection when computing desert SBAFs as the adjustments are not transferrable from one PICS to another, therefore suggesting that spectral corrections are highly dependent on the unique spectral signature of each target. The advantage of the quasi-continuously scanning SCIAMACHY instrument is that spectra over the entire Earth are available. Site-specific spectra can be easily identified using geographic coordinates and ancillary data provided in the matched CERES SSF product. 1) Desert and Polar Ice PICS FOVs: Ten commonly examined PICS have been predefined for the SBAF tool. If the

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. 4

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING

TABLE I D OMAIN B OUNDARY C OORDINATES , A PPROXIMATE FOV C OUNT, AND S TANDARD D EVIATION OF SCIAMACHY-C ONVOLVED -W ITH -AQUA -MODIS 0.65-μm S CALED R ADIANCE FOR THE T EN PICS C URRENTLY AVAILABLE FOR S ELECTION IN THE O NLINE SBAF T OOL . S NOW /I CE PICS A RE N OT C LOUD S CREENED . FOV C OUNT CAN C HANGE AS THE T OOL I S U PDATED

center coordinate of a SCIAMACHY FOV is within the PICS domain boundaries, then that footprint is considered for the PICS SBAF. Table I provides the domain boundaries of each PICS. Additional details of the predefined PICS are provided by Bhatt et al. [31]. The use of small-area PICS for SBAFs that are based on relatively large hyperspectral footprints presents several challenges. Although the Saharan and Arabian Deserts, as well as the polar ice sheets, are fairly well represented by the SCIAMACHY FOV, other sites, such as the Badain Jaran and Sonoran Deserts, are surrounded by vegetation or mountainous terrain. These nondesert spectra are part of the overall SCIAMACHY FOV measurement. Their areal coverage, however, is small compared with the desert area of interest, and therefore, their impact on the overall PICS spectra should at least be reduced [31]. These land coverage criteria mean that for certain PICS, e.g., Sonoran Desert and Greenland South, only SCIAMACHY footprints with specific orientations that allow all four corners of the FOV to be over land are used to determine the SBAF. That is, a SCIAMACHY footprint must be adequately centered over the land or ice sheets such that there is no chance for spectral contributions from open water. For further quality control, the desert PICS are screened for clear-sky conditions. The CERES-based cloud matching as ap plied to SCIAMACHY footprints does not reliably filter all cloudy footprints over the warm-target PICS. Therefore, clearsky conditions were determined by applying a spatial homogeneity threshold to the standard deviation of the visible-channel reflectance measured using all ∼4-km pixels within the PICS domain from the appropriate GEO satellite image nearest the Enivsat overpass time. This method is applicable because the daily local noon clear-sky spatial standard deviations are typically small, whereas on cloudy days, the standard deviation is large [31]. Owing to the difficulty in detecting clear-sky conditions over snow/ice PICS, these PICS spectra represent all-sky conditions. Fig. 1 displays the mean of the combined SCIAMACHY footprint reflectance spectra for each PICS. Reflectance P is derived from SCIAMACHY radiance L and solar irradiance E as follows:   L × π × d2 (1) P = E × cos(SZA) where d is the Earth–Sun distance factor in astronomical unit. Table I provides the number of SCIAMACHY footprints and

Fig. 1. Spectral signatures, based on SCIAMACHY reflectance, of the ten predefined PICS described in Table I. The SCIAMACHY spectra are subsampled at five-bin intervals to improve figure clarity.

the standard deviation of the SCIAMACHY 0.65-μm scaled radiance based on a convolution with the Aqua-MODIS Band 1 SRF. Note that output from the tool is expressed in terms of either radiance or scaled radiance, but not true reflectance. Using a true reflectance that is regulated by the cosine-SZA term, as in (1), for computing SBAFs would limit the dynamic range of measurements gathered for any scene. One strength of this methodology is allowing users the option to define their SBAF in terms of higher-order regressions rather than only a simple ratio of mean values; thus, a large dynamic range of measurements is desirable (see Section II-C1 and C2). Removing the cosine-SZA term from (1) therefore maximizes the dynamic range, and rather than reflectance P , the resultant variables are referred to as scaled radiance P  . It is important to note that an SBAF in terms of a simple ratio of the mean true reflectance will be identical to that of scaled radiance. Scaled radiance, however, offers the benefit of separation of Sun angle from the magnitude of incident energy on the detector, which may be a nonlinear relationship depending on SRF disparity. That is, high Sun angles yield lower reflected energy, which may require a different SBAF compared with low-SZA but high-energy cases. Using scaled radiance terms, rather than reflectance, gives users more options as to what range of reflectance values they wish to apply an SBAF. The footprint reflectance spectra in Fig. 1 represent the TOA spectral variability as observed from SCIAMACHY and include the intra-annual variability of the atmospheric column and seasonal progression of the solar viewing geometry. Users

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. SCARINO et al.: WEB-BASED TOOL FOR CALCULATING SPECTRAL BAND DIFFERENCE ADJUSTMENT FACTORS

Fig. 2. Spectral signatures, based on SCIAMACHY reflectance, of clear-sky ocean, approximate DCC, and precise DCC scenes. Thick lines are the mean FOV spectra, and thin lines are the standard deviation. The SCIAMACHY spectra are subsampled at five-bin intervals to improve figure clarity.

can further filter the SCIAMACHY footprints by PW, viewing geometry, month, season, etc., to meet their requirements during a calibration event, or they may specify certain parameter ranges to reduce the overall uncertainty of the SBAF (see Section III-A). Therefore, no standard deviation thresholdbased filtering of footprints is employed because there is a risk of losing relevant data, and the users have the means to subset their own spectral data sets and thereby reduce SBAF uncertainty. 2) DCC and Clear-Sky Ocean FOV Selection: DCC footprints are selected from all tropical land and ocean SCIAMACHY FOVs provided the center coordinate is within ±15◦ latitude of the equator, the FOV-centered SZA is less than 45◦ , the center cloud top temperature is less than 205 K, the center cloud fraction is greater than 99%, and the center cloud optical depth is greater than 70. These criteria result in ∼3485 SCIAMACHY FOVs. The scaled radiance standard deviation of the ∼3485-FOV spectra is ∼0.08 at 0.65 μm (see Fig. 2). Further investigation reveals that the channel spectra are inconsistent, particularly in Channel 5, owing to the FOV size, long dwell time, and cloud advection. This data set represents a collection of bright tropospheric clouds and is therefore referred to, within the tool, as approximate DCC. To provide a data set that more accurately represents DCC, a subset of ∼388 bright 0.65-μm FOVs was found by using the same criteria described above, with the additional condition that all four corners must also be associated with a cloud top temperature less than 205 K. The FOVs in this subset have rather similar channel spectra. The mean scaled radiance at 0.65 μm from the ∼388-FOV spectra is 0.05 greater than that from the ∼3485-FOV spectra. The associated standard deviation of the ∼388 FOVs is ∼0.03 at 0.65 μm (Fig. 2, precise DCC). Although it is effectively impossible to find a DCC cell encompassing an entire SCIAMACHY FOV, this DCC data set contains the brightest equatorial high cloud footprints found on the planet and should provide the most accurate DCC SBAF based on SCIAMACHY. The DCC FOVs offer perhaps the brightest calibration reference on Earth, whereas clear-sky ocean scenes provide perhaps the darkest. Similar to DCC selection, it is rather difficult to

5

consistently locate a 30 km × 240 km area that is completely cloud free. Clear-sky ocean footprints are found by identifying ocean FOVs for which the center coordinate is within ±15◦ latitude of the equator and the cloud fraction at all four corners and the center is 0%. These criteria do not, however, guarantee that the footprint is entirely free of clouds because the CERES cloud fractions are valid only for a 20-km FOV, thus allowing for the possibility of clouds existing elsewhere within the much larger SCIAMACHY footprint. A simple radiance threshold filter, therefore, accounts for cloud-contaminated footprints by allowing only those FOVs with radiance values less than 100 Wm−2 μm−1 sr−1 at 0.65 μm. The clear ocean spectrum in Fig. 2 represents ∼1785 SCIAMACHY FOVs (standard deviation of ∼0.03) and contains the darkest equatorial ocean footprints. 3) Other Predefined Footprint Selection Options: Several SBAF scene choices are available to the user that are not specific to any one PICS or narrow spectral signature, but rather cover larger general domains or represent boundaryindependent scene locations designed to account for the spectral band difference when performing GEO/LEO ray-matching calibration or LEO/LEO SNO calibration. A completely unconstrained global SBAF option is also available. For GEO/LEO calibration purposes, tropical SBAF scene choices are offered based on SCIAMACHY footprints taken over the entire equatorial domain from 15.0◦ N to 15.0◦ S. These footprints are divided into all-sky ocean and all-sky land domains. To ensure no significant bodies of water are present within a SCIAMACHY land FOV, the center and all four corner-coordinates of any SCIAMACHY land footprint must be classified as more than 90% land. Similarly, all ocean footprint coordinates must be classified as more than 90% ocean. If both a land and ocean tropical all-sky domain is desired, then the user must choose the global data set and specify coordinates for the tropical domain, as the tropical domain data sets are not direct subsets of the global data set and do not contain mixed land/ocean footprints. Also available are north (south) pole domain footprints, which are constrained to only those FOVs that are north (south) of 60◦ N (60◦ S). These two subsets are useful for LEO/LEO SNO calibration algorithms. The global option should be used when none of the other predefined Earth spectra meet the user’s needs, as its sampling is divided evenly across the globe, not concentrated in a targeted domain or averaged over a specific spectral signature. Therefore, because of the completely unconstrained nature of the global option, it is the user’s responsibility to ensure accurate subsetting for the particular application. This caveat applies to all other Earth spectra options as well, because any predefined data set described in this section and the previous sections can be further subsetted by the user. C. Online SBAF Tool Design 1) Web Tool Options: To illustrate the web tool options as presented to a user, Fig. 3 displays a screen capture of the initial design of the online SBAF tool, publicly available at http://cloudsgate2.larc.nasa.gov/SBAF (alternatively: http:// www-angler.larc.nasa.gov/SBAF). Shown are the five basic options that the user must select: Earth spectra, reference

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. 6

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING

Fig. 3. Screen capture of the initial SBAF selection tool web interface.

Fig. 4. SRFs for common Aqua-MODIS (blue) and NOAA-14-AVHRR (red) channels.

instrument SRF and central wavelength, target instrument SRF and central wavelength, units, and regression type. The user first selects the desired Earth spectra, including the predefined or global cases. Then, the reference and target satellite instrument and associated band are chosen. Assistance in identifying the desired SRF/band pairing is available at http://cloudsgate2.larc. nasa.gov/SPECTRAL-RESPONSE (alternatively: http://wwwangler.larc.nasa.gov/SPECTRAL-RESPONSE). As an example, Fig. 4 shows two pairs of SRFs corresponding to Channels 1 [see Fig. 4(a)] and 2 [see Fig. 4(b)], respectively, for AquaMODIS and NOAA-14-AVHRR. With the web tool, the SBAFs for any pair of SRFs can be computed in terms of radiance or scaled radiance using regression options that include linear, quadratic, cubic, and force fits (linear fit through the zero offset). The user may also select any number of advanced options, which consist of threshold specifications for year, month, day, min/max latitude and longitude bounds, SAA, SZA, VZA, and PW content. Such advanced options are useful, for instance, for determining whether standard error about the regression (StdRegErr) is owed to atmospheric or angular conditions. Users can select the seasonal option to exclude days that, intraannually, fall outside the specified month/day limits. Regardless of the data set chosen, proper subsetting as described here can yield highly accurate spectral corrections for specific calibration studies. Additional explanation and application of the subsetting options are provided in Section III-A. 2) SBAF Calculation: The SCIAMACHY-based scenedependent SBAF derivation methodology has been extensively described by Doelling et al. [21]–[23] and Scarino et al. [24].

The spectra from each appropriate SCIAMACHY FOV must be carefully convolved with the SRFs of the reference and target imager channels to compute imager-equivalent radiance or scaled radiance values, i.e., pseudo radiances or pseudo scaled radiances, specific to the scene of interest. That is, a pseudo radiance or pseudo scaled radiance reference and target pair is convolved for each footprint. This calculation is performed as the SBAF request is made, following the user input, for any of the aforementioned predefined or user-specified scenes and subsetting options. Equation (2) shows the derivation of a target–sensor pseudo radiance (Lp,tar ), i.e., nbins 

Lp,tar =

Ls,n × RSRtar,n × (λn − λn−1 )

n nbins 

(2) RSRtar,n × (λn − λn−1 )

n

where Ls,n is the SCIAMACHY at-sensor radiance for SCIAMACHY wavelength bin n; nbins is the total number of wavelength bins used to divide the SCIAMACHY spectral range, which, in this case, is 3200; and RSRtar,n is the target relative spectral response at that same wavelength (λ). Similarly, the reference pseudo radiance Lp,ref is calculated by (2) except with the reference relative spectral response. Target or reference pseudo scaled radiance is calculated by replacing Lp with Pp in (2). A simple mean ratio, linear, or nonlinear regression of the pseudo radiance target and reference pairs, with Lp,tar as the ordinate and Lp,ref as the abscissa, will yield an SBAF for the target–reference pair over the selected Earth spectra. The user can select the regression type that most effectively captures the variability of the radiance pairs. Applying this SBAF to the reference sensor radiance values will yield the expected target sensor radiance values. That is, the SBAF is applied to the true reference sensor radiance (Lref ) to arrive at a predicted target sensor radiance (Ltar,predict), as shown in the following equation: Ltar,predict = SBAF × Lref .

(3)

In this example, the SBAF term represents a simple ratio of mean Lp,tar to mean Lp,ref . The SBAF accounts for disparate SRFs, thereby the true target sensor radiance (Ltar ) is spectrally consistent with Ltar,predict.

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. SCARINO et al.: WEB-BASED TOOL FOR CALCULATING SPECTRAL BAND DIFFERENCE ADJUSTMENT FACTORS

7

FOV centers are indicated by “X.” Concerns over whether the large size of the SCIAMACHY footprint means it cannot accurately represent the spectra of the Libya 4 PICS and, thus, would introduce error into the SBAF, were addressed by Bhatt et al. [31]. They compared the influence of an SBAF determined using the Libya 4 PICS boundaries listed in Table I and for a Libyan Desert PICS that is one-third the original Libya 4 size. The spectral radiance differences between the two Libya 4 domains are less than 0.6%, indicating that the impact of spatial mismatch on the SBAF is minimal for that domain. Libya-4-specific SBAFs were determined using the signatures from 15 SCIAMACHY FOVs, 17 GOME-2 FOVs, and 5 Hyperion overpass swaths. For all measurements, the FOVcentered VZA was limited to less than 10◦ . In this experiment, radiance values were normalized to the SCIAMACHY SZA of 29.0◦ as follows: Fig. 5. Libya 4 PICS domain (green box) relative to consecutive SCIAMACHY (dark blue) and GOME-2 (cyan) footprints, along their respective orbits, and a Hyperion (yellow) swath. The X’s indicate the centers of each SCIAMACHY or GOME-2 footprint. Solid dark-blue or cyan rectangles indicate that the SCIAMACHY or GOME-2 footprint is included in the SBAF calculation, whereas dotted dark-blue or cyan rectangles indicate that the footprint is excluded. These example overpasses are from August 2007.

L=

L × cos(29◦ ). cos(SZA)

Furthermore, an Earth–Sun distance correction of 1.0258, which is valid for 15 August, was employed, i.e., L=

D. SBAF Consistency Experiment Setup The accuracy of a SCIAMACHY-based spectral correction is analyzed by comparing SCIAMACHY SBAFs to SBAFs derived using the Hyperion and GOME-2 hyperspectral instruments. Hyperion is mounted onboard the Earth Observing-1 Mission satellite, which was launched on 21 November 2000. Specifications of the Hyperion instrument are described in depth by Chander et al. [11]. The sensor has a spectral resolution of 10 nm, covering the spectral range of 400–2500 nm. It has a spatial resolution of 30 m with a swath width of 7.7 km and is radiometrically stable to within 5% [32]. The GOME-2 sensor orbits onboard the MetOp-A satellite, which launched on 19 October 2006. It has a spectral resolution effectively matching that of SCIAMACHY, but its spectral range is limited to 240–790 nm. The GOME-2 footprint size is 40 km × 80 km, which is slightly smaller than half of a SCIAMACHY FOV. The SBAF consistency analysis is conducted for the Libya 4 PICS during one month, i.e., August 2007, to best ensure consistency of the target as observed from three independent sensors. Libya 4 is considered one of the most stable calibration targets by the Centre National d’Etudes Spatiales [33]. It is an ideal choice given that its relatively broad domain is easily represented by all three sensors and the fact that Hyperion, which does not continuously operate, often sampled Libya 4 to validate the Hyperion calibration stability [25]. Hyperion measured Libya 4 five times during August 2007. Data from GOME-2 and SCIAMACHY were subsampled within the Libya 4 PICS domain boundaries, as listed in Table I. Specifically, if the center of the SCIAMACHY or GOME-2 footprint is within the PICS boundaries, then those spectra are included in the SBAF analysis. (See Fig. 5 for an illustration of the Hyperion, SCIAMACHY, and GOME-2 FOVs overlaid on a Google Earth image of Libya 4). The Hyperion FOV represents the entire overpass swath. The SCIAMACHY and GOME-2

(4)

L . 1.0258

(5)

III. R ESULTS AND D ISCUSSION A. Online SBAF Tool Applications 1) SBAF Web Tool Output: Once the user has selected the input for the SBAF web tool as described in Section II-C1, all footprint pseudo radiance or pseudo scaled radiance pairs for the given SRF selections are plotted and regressed, and the coefficients of that regression constitute the SBAF as outlined in Section II-C2. Fig. 6 shows examples of SBAF regressions for the Libya 4 PICS using the Aqua-MODIS and NOAA-14-AVHRR SRFs for the 0.65-μm [see Fig. 4(a)] and 0.86-μm bands [see Fig. 4(b)]. For the 0.65-μm band SBAF [see Fig. 6(a)], the linear fit is sufficient, and the offset is nearly zero, suggesting that the force fit, or linear regression with a zero intercept, could also be utilized. In the case of the 0.86-μm pseudo radiance pair regression [see Fig. 6(b)], the SBAF substantially deviates from the 1:1 line because there are more gaseous absorption lines (see Fig. 1) within the AVHRR SRF range that are not found in the narrower MODIS Channel-2 band (see Fig. 4). Nevertheless, the quadratic SBAF encompasses much of the radiance variability. Use of the quadratic SBAF is particularly justifiable for cases when the transition from the darkest to the brightest points along the dynamic range is noticeably nonlinear; however, users are cautioned to not apply any SBAF beyond the range of the measured data, where nonlinearities may be unrealistically large. The standard error about the regression (StdRegErr) describes the uncertainty of the SBAF. That is, the uncertainty is defined by the random error in matched pairings about the regression curve caused by the small spectral variations among the individual SCIAMACHY footprint measurements and how each SRF convolution is influenced by those variations. Uncertainty does

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. 8

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING

Fig. 6. (a) Linear and (b) second-order quadratic SBAF tool output examples for the Libya 4 PICS in terms of pseudo radiance between (a) Aqua-MODIS Channel 1 and NOAA-14-AVHRR Channel 1 and (b) Aqua-MODIS Channel 2 and NOAA-14-AVHRR Channel 2 (see Fig. 4).

Fig. 7. SBAF tool output examples for the Libya 1 PICS in terms of pseudo scaled radiance, subsetted by PW intervals of (a) 0.0–3.0 g cm−3 , (b) 0.0–0.75 g cm−3 , (c) 0.75–1.5 g cm−3 , (d) 1.5–2.0 g cm−3 , and (e) 2.0–3.0 g cm−3 . The latter four SBAFs correspond to the spectral signatures in Fig. 8(b).

not relate to an absolute error or bias, as SCIAMACHY is considered truth for the purposes of these SBAFs. For Fig. 6, only the scene spectra, instrument SRFs, units, and regression options were selected (see Fig. 3), and no further subsetting options were employed. Therefore, it is possible to reduce the SBAF uncertainty via further filtering by viewing geometry, time, or PW, which is discussed in the following section. 2) SBAF Dependence Relations: As discussed for the case in Fig. 6(b), water vapor absorption not common to both SRFs will cause fluctuations between the instrument radiances, particularly in the NIR. This differential dependence will be the case for all clear-sky and low-cloud conditions, and thus, the SBAF uncertainty estimated by the standard error of the radiance pair regression can be quite large. These fluctuations can be attributed to a number of variables, such as sun angle, aerosols,

and atmospheric water vapor. If each of the parameter-subsetted SBAF standard errors is less than the overall standard error, then the parameter-dependent SBAF optimization has been properly implemented. The Libya 1 PICS is used to demonstrate the value of parameter-based subsetting. Unfortunately, PW information is not available for every SCIAMACHY footprint. Ideally in future version updates, a more complete PW data set can be matched with the SCIAMACHY data. All other parameters are available given that they are instrument dependent, rather than model dependent. Libya 1 is chosen because it has a larger number of footprints with valid PW values. Footprints with no associated PW use a default PW fill value of −1. Fig. 7(a) shows the 526-footprint Libya 1 PICS NOAA-14-AVHRR and AquaMODIS 0.86-μm SCIAMACHY pseudo radiance pairs that

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. SCARINO et al.: WEB-BASED TOOL FOR CALCULATING SPECTRAL BAND DIFFERENCE ADJUSTMENT FACTORS

9

Fig. 8. (a) SCIAMACHY reflectance mean spectra (thick lines) and their standard deviations (thin lines) for the Libya 1 PICS, subsetted by season. (b) Same as (a), except subsetted by PW content of the atmospheric column. (c) Difference between winter and summer mean reflectance spectra. (d) Mean scaled reflectance difference for PW content of 0.0–0.75 g cm−3 and 2.0–3.0 g cm−3 . The SCIAMACHY spectra are subsampled at five-bin intervals to improve figure clarity.

have an associated PW value (note that specific footprint counts can change over time as the tool is updated). The overall uncertainty of the force fit linear regression is 2.47%. Fig. 7(b)–(e) shows the footprint radiance pairs for four PW intervals. The range limits are based on user trial and error to achieve sufficient sampling and reduced uncertainties across each of the PW ranges. Each PW interval has an uncertainty that is less than 1.5%, which is below the overall standard error of ∼2.5%. The corresponding force fit slopes show a systematic decrease of 0.885, 0.862, 0.846, and 0.835 as the PW increases. Therefore, the SBAF with PW subsetting is an improvement on the general SBAF. For the parameter subsetting to be successful, the parameter interval uncertainties should be less than the overall uncertainty, and a systematic change in the SBAF slope should be observed. Fig. 8(b) shows the Libya 1 SCIAMACHY spectral reflectance for each of the PW intervals in Fig. 7. Fig. 8(d) illustrates the spectral reflectance difference between the lowest and highest PW intervals, which can be as great as 0.1 in the NIR (0.8–1.0 μm). This difference is due to the narrower AquaMODIS Band-2 SRF falling between two strong water vapor absorption lines that are encompassed by the corresponding AVHRR Band-2 SRF [see Section III-B, Fig. 9(a)]. Another means of reducing SBAF uncertainty is seasonal subsetting, as shown in Fig. 8(a), with Fig. 8(c) illustrating the reflectance difference between summer and winter. It is not surprising to see similar seasonal and PW reflectance variations given that the amount of PW in the atmosphere depends on temperature. The seasonal variation could also be owed to aerosols, ozone absorption, and sun angle effects. An increase in SZA corresponds to an increase in the atmospheric path length, which

enhances the Rayleigh scattering effect and ozone absorption. Surface albedo and bidirectional reflectance distribution function (BRDF) effects also increase at higher sun angles, which are seasonally dependent. For example, Fig. 8(c) clearly shows an increase in reflectivity in winter over the blue wavelengths. Blue-band SBAFs over clear-sky ocean will need to account for Rayleigh scattering variations. Accounting for viewingangle-dependence effects in this method is severely limited because the SCIAMACHY data are taken at four fixed nearand off-nadir views. Thus, it is important that the user confines application of these SBAFs to data measured within 30◦ VZA, unless an adequate BRDF model is employed to normalize the geometric conditions. Subsetting by date can be useful in instances when the user may have interest in a specific local short-term occurrence, such as during a single-satellite ground-track intersect event. In such a case, a specific set of angular, PW, and date range options can be selected. Generally, however, it is encouraged that users utilize the complete available data record to increase sample size and capture the long-term parameter variability over a given site and also because of the excellent stability of the SCIAMACHY hyperspectral radiances relative to AquaMODIS. Given the fact that most SBAFs can be described as a linear function indicates that the SBAF is not contingent on the absolute calibration of SCIAMACHY. This statement implies that the SCIAMACHY method of SBAF calculation is applicable to instrument pairings that precede and follow the SCIAMACHY mission. Except for certain VZA dependence relations noted above, the necessary tools are available to the user to properly identify systematic dependence relations in the SBAFs and perhaps

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. 10

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING

Fig. 9. (a) August 2007 Libya 4 radiance mean spectra (thick lines) from SCIAMACHY (SCIA), GOME-2, and Hyperion (HYP) and their standard deviations (SCIAs, GOME-2s, HYPs; thin lines). Aqua-MODIS channel radiance mean (magenta circles) and standard deviation (magenta X’s) values are shown for the same time period and location. Aqua-MODIS (black lines) and NOAA-14-AVHRR (cyan lines) Channel 1 (solid lines) and Channel 2 (dotted lines) SRFs are overlayed. SCIAMACHY-based SBAF examples of this case are shown in Fig. 6. (b) Hyperspectral relative radiance difference between SCIA and GOME-2 and SCIA and HYP. The horizontal solid lines are the mean relative differences. (c) Channel relative radiance differences between SCIA and Aqua-MODIS, GOME-2 and Aqua-MODIS, and HYP and Aqua-MODIS. The solid blue, green, and red lines are the respective mean differences.

relate those dependence relations to angular, seasonal, or atmospheric causes. To aid future development of this tool, the web interface includes an area for feedback submission, thereby allowing users to request additional subsetting options and parameters that may be useful for isolating specific SBAFs. B. SBAF Consistency Experiment Results As introduced in Sections I and II-D, SCIAMACHY is merely one instrument option available for deriving SBAFs using hyperspectral visible data. It is therefore important to compare the SCIAMACHY-derived SBAFs with those generated from concurrent GOME-2 and Hyperion data. If the SCIAMACHY hyperspectral radiances are more variable than those from similar instruments, then the convolved pseudo radiance pairs may have a larger standard error than that associated with the true Earth reflected spectra. Furthermore, if the spectral calibration varies by wavelength, biases will be manifested in the resulting SBAF. The hyperspectral radiance signatures from all three instruments over the Libya 4 PICS during August 2007 are plotted in Fig. 9(a). Qualitatively, there is good agreement among the three instruments—remembering that one must consider the lower spectral resolution of Hyperion. The

SCIAMACHY and GOME-2 signatures correlate rather well with each other given their equal resolution. Hyperion is unable to capture the fine spectral intricacies of the other two sensors. All three sensors qualitatively match well with individual channel radiance values measured by Aqua-MODIS. Fig. 9(b) provides a more quantitative view of the relative radiance differences between SCIAMACHY and GOME-2 and between SCIAMACHY and Hyperion. In Fig. 9(c), the relative difference is computed for all three sensors with respect to Aqua-MODIS. In Fig. 9(b), percent differences are determined at the Hyperion spectral band interval resolution. The prominent inflection at the start of the spectrum (also seen in Fig. 9(a) and noted in the EUMETSAT GOME-2 Product Guide) and the two large positive maxima at 400 and 600 nm in the SCIAMACHY-minus-GOME-2 differences are associated with uncertainties in the SCIAMACHY and GOME-2 radiance data near the interface regions of the main optical channels [34]. The artifacts that induce these interchannel uncertainties are removed for the MODIS SRF GOME-2 convolved radiances such that they do not add a false bias to the correction. The mean relative differences between SCIAMACHY and GOME-2 radiances (horizontal green line) and between the SCIAMACHY and Hyperion radiances (horizontal red line) are

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. SCARINO et al.: WEB-BASED TOOL FOR CALCULATING SPECTRAL BAND DIFFERENCE ADJUSTMENT FACTORS

TABLE II SBAF S F ROM SCIAMACHY, GOME-2, AND H YPERION C ALCULATED OVER THE L IBYA 4 PICS, AUGUST 2007. T HE L IMITED GOME-2 S PECTRAL R ANGE D OES N OT A LLOW FOR SBAF D ERIVATION FOR BAND 2. T HE SBAF S A RE D ETERMINED F ROM AQUA -MODIS AND NOAA-14-AVHRR M EAN P SEUDO R ADIANCE R ATIOS , W HICH A RE D ERIVED F ROM THE SRF C ONVOLUTIONS W ITH M EAN I NSTRUMENT-M EASURED H YPERSPECTRAL R ADIANCE VALUES

2.4% and −3.4%, respectively. In Fig. 9(c), relative differences are calculated using MODIS SRF hyperspectral convolved radiance for each MODIS band. The mean relative differences between SCIAMACHY, GOME-2, and Hyperion relative to Aqua-MODIS channel radiances are 13.5%, −2.1%, and 15.1%, as marked by the horizontal blue, green, and red lines, respectively. Note that the relatively high percent differences of SCIAMACHY and Hyperion compared with Aqua-MODIS are owed to the sensitivity of conducting relative difference calculations for the near-zero radiance values associated with the Aqua-MODIS 1.38-μm channel. Here, the absolute radiance differences for both hyperspectral instruments with respect to MODIS are close to 0.3 Wm−2 μm−1 sr−1 . When excluding the 1.38-μm channel, the overall mean relative differences reduce to −2.7% and −0.5% for SCIAMACHY and Hyperion, respectively. The absolute and relative radiance differences have a minor impact on the comparisons of the respective SBAFs from each hyperspectral sensor. As such, SBAFs are calculated between Aqua-MODIS and NOAA-14-AVHRR Channels 1 and between Aqua-MODIS and NOAA-14-AVHRR Channels 2. These sensors were chosen to not only show both 0.65-μm channel and 0.86-μm channel SBAFs, but also to demonstrate that SBAFs can effectively provide correction for cases of a minimal significant overlap between the two spectral bands [compare SRFs for Channels 1 and 2 in Fig. 4 and Fig. 9(a)]. Note that no SBAF can be calculated from GOME-2 for the 0.86-μm channel, which is one of the weaknesses of that instrument for this application. Because of the various instrument spatial resolutions, wavelength resolutions, and the limited number of footprints, the SBAFs were computed using the mean spectra. Given that all FOVs were observed in August 2007, it is assumed that the atmospheric and desert conditions were uniform throughout the month [note that the standard deviations of the mean spectra in Fig. 9(a) are small], and a single factor is therefore justified to represent the SBAF. That is, for each hyperspectral sensor, there is one pair of pseudo radiances, which are based on the mean of the combined radiance spectra measured by the instrument. An SBAF for each sensor is then simply the ratio of the pseudo radiance pair. Table II lists the Aqua-MODIS and NOAA-14-AVHRR Channels 1 and 2 SBAFs derived from SCIAMACHY, GOME-2, and Hyperion. For Channel 1, the SCIAMACHY-based SBAF is within 0.3% and 0.2%, respectively, of the GOME-2 and Hyperion values. The SCIAMACHY Channel-2 SBAF is within 0.1% of its Hyperion counterpart. Overall, the SCIAMACHYderived SBAFs compare well with those from GOME-2 and

11

Hyperion, thereby confirming their relative spectral accuracy. These results indicate that SCIAMACHY-based SBAFs are within 0.1%–0.3% of SBAFs derived from Hyperion or GOME-2, for the 0.65-μm and 0.86-μm (SCIAMACHY and Hyperion only) spectral bands. The small differences in SBAF suggest that absolute radiance variance among the three instruments is unimportant compared with atmospheric/view angle considerations. That is, the mean differences plotted in Fig. 9(b) and (c) have little impact on the SBAF, whereas PW and seasonality have significant influences on the SBAFs, as previously summarized in the examples in Figs. 7 and 8. With SBAFs from all three sensors being comparable, one must look at the other advantages of SCIAMACHY to justify its initial sole use in this tool. Although GOME-2 has a smaller footprint size, better global coverage (larger swath and continuous nadir observation), and similar spectral resolution relative to SCIAMACHY, its spectral range ends at only 790 nm. Hyperion, on the other hand, covers most of the relevant spectral range (even for wavelengths beyond that of reliable SCIAMACHY data), but lacks the high spectral resolution. More importantly, Hyperion, being onboard a tasking satellite, may not have coverage data for all areas of interest. The NIR coverage and continuous global sampling of SCIAMACHY are key features for its use in the online tool. Therefore, it is concluded that SCIAMACHY is currently the most capable hyperspectral sensor available for routine determination of SBAFs. (See Table III for a comparative summary of the key instrument features.) The ideal solution, and a near-term future goal, is to offer the choice of all three instruments such that users would also be able to take advantage of the strengths offered by GOME-2 (smaller FOV and broader coverage) and Hyperion (completes the 1750–2500-nm spectral gap). IV. C ONCLUSION An online SBAF tool (http://cloudsgate2.larc.nasa.gov/SBAF or http://www-angler.larc.nasa.gov/SBAF), based on global SCIAMACHY spectra, will improve interinstrument comparisons, validations, and transfers of calibration by providing adjustments for spectral band differences over pseudoinvariant Earth targets and intercalibration domains. The web-based SBAF tool and accompanying instrument- and band-specific SRF comparison plotting tool (http://cloudsgate2.larc.nasa.gov/ SPECTRAL-RESPONSE or http://www-angler.larc.nasa.gov/ SPECTRAL-RESPONSE) allow the calibration community to investigate how instrument-pair-observed reflected radiances over specific Earth scenes are affected by spectral band differences. Nearly a decade of SCIAMACHY data are available to derive SBAFs by means of the web interface—a design that grants users unprecedented control over exactly where, when, and with what variable considerations they want spectral corrections computed, using any available instrument/band pairing they choose. Users have the option to decide from the 17 predefined spectral scenes and may further subdivide via eight-parameter subsetting options. Users can subset based on date range, season, latitude, longitude, SCIAMACHY FOVcentered VZA, SAA, SZA, and PW content of the atmospheric column. All results are user driven and can be applied toward

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. 12

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING

TABLE III S UMMARY C OMPARISON OF THE FOV, S WATH W IDTH , AND H OST S ATELLITE W ITH L OCAL E QUATOR C ROSSING T IME , A LONG W ITH S PECTRAL R ESOLUTION , S PECTRAL C OVERAGE , AND D ATA AVAILABILITY OF THE T HREE H YPERSPECTRAL I NSTRUMENTS U SED FOR THE SBAF C ONSISTENCY E XPERIMENT. T HE SCIAMACHY C HANNELS 7 AND 8 A RE N OT U TILIZED IN THE SBAF T OOL , W HICH I S W HY T HEIR I NFORMATION I S E XCLUDED F ROM T HIS TABLE (S EE S ECTION II-A)

established and future calibration techniques, thereby establishing the usefulness of this tool for the calibration community. Spectral radiances, and therefore SBAFs, from a specific geographic location are highly dependent on atmospheric and surface reflectance conditions. It was shown for the Libya 1 calibration site that the spectral signature can significantly vary over the same PICS as seasons and atmospheric PW content change. Therefore, PW, angular geometry, and time options are available to the user so that they may either derive multiple SBAFs or fine-tune an SBAF to a singular condition in time. Successful parameter subsetting is realized when the individual subsetted SBAF uncertainties are less then the overall uncertainty and when the resulting SBAFs systematically change with respect to the parameter interval. The variety of useful combinations of instrument/channel pairs, spectral scenes, and subsetting options for the available SCIAMACHY data record is extensive, thus necessitating a web-input-driven selection tool that allows users to design SBAFs for their specific applications. A user-feedback section is available in the event that additional subsetting options, SRFs, or spectral scenes are necessary. A comparison study of Aqua-MODIS and NOAA-14AVHRR 0.65- and 0.86-μm-band SBAFs derived from SCIAMACHY, Hyperion, and GOME-2 was performed over the Libya 4 PICS during August 2007. The results show that all SBAFs are within 0.1%–0.3% of one another, despite variance in the interinstrument radiance spectra. The results suggest that spectral calibration differences are unimportant compared with effective atmospheric/angular-based subsetting. That is, no one instrument has an advantage over another with regard to relative spectral calibration, and given the fact that viewing and atmospheric conditions cause greater SBAF variation than the relative spectral calibration, all three instruments are suitable for SBAF computations. However, of the three sensors, SCIAMACHY is the best choice for the routine calculation of SBAFs because of its high spectral resolution and nearly continuous global availability compared with Hyperion, and its significantly broader spectral range compared with GOME-2. Nevertheless, offering the user a choice of source instrument for the SBAF is the ideal scenario and is deemed future work. Version updates will be made to the web interface as new instruments, Earth spectra options, land masking options, and

subsetting parameters, e.g., relative azimuth angle, are needed or requested. That is, future satellite instrument SRFs will be added, and user demand may necessitate the addition of new and/or frequently used, although not predefined, PICS. A nearterm goal is to allow future versions of the tool to output the scene-specific SCIAMACHY spectra, similar to Figs. 1 and 2. Other auxiliary data sets containing cloud or atmospheric retrievals, e.g., Envisat MERIS retrievals, offered by users for subsetting purposes will be considered for implementation. Regardless, in its current version, this tool allows unprecedented flexibility and utility in SBAF calculation and should prove to be of significant benefit to the calibration community. ACKNOWLEDGMENT The authors would like to thank J. Burrows, S. Noel, and K. Bramstedt at Bremen University and R. Snel at the Netherlands Institute for Space Research for the assistance with the SCIAMACHY data. They would also like to thank the European Space Agency Envisat program for providing the SCIAMACHY data. R EFERENCES [1] M. Goldberg et al., “The Global Space-based Inter-Calibration System (GSICS),” Bull. Amer. Meteorol. Soc., vol. 92, no. 4, pp. 467–475, Apr. 2011. [2] L. Wang et al., “Comparison of AIRS and IASI radiances using GOES imagers as transfer radiometers toward climate data records,” J. Appl. Meteorol. Climatol., vol. 49, no. 3, pp. 478–492, Mar. 2010. [3] M. M. Gunshor, T. J. Schmit, W. P. Menzel, and D. C. Tobin, “Intercalibration of broadband geostationary imagers using AIRS,” J. Atmos. Ocean. Technol., vol. 26, no. 4, pp. 746–758, Apr. 2009. [4] A. Wu et al., “Characterization of Terra and Aqua MODIS VIS, NIR, and SWIR spectral band calibration stability,” IEEE Trans. Geosci. Remote Sens., vol. 51, no. 7, pp. 4330–4338, Jul. 2013. [5] D. C. Tobin, H. E. Revercomb, C. C. Moeller, and T. S. Pagana, “Use of atmospheric infrared sounder high-spectral resolution spectra to assess the calibration of moderate resolution imaging spectroradiometer on EOS aqua,” J. Geophys. Res., vol. 111, no. D9, Mar. 2006, Art. ID D09S05. [6] “Instrument: IASI,” World Meteorol. Org., Geneva, Switzerland, Observing Systems Capability Analysis and Review Tool. [Online]. Available: http://www.wmo-sat.info/oscar/instruments/view/207 [7] “Instrument: IASI-NG,” World Meteorol. Org., Geneva, Switzerland, Observing Systems Capability Analysis and Review Tool. [Online]. Available: http://www.wmo-sat.info/oscar/instruments/view/206 [8] B. A. Wielicki et al., “Achieving climate change absolute accuracy in orbit,” Bull. Amer. Meteorol. Soc., vol. 94, no. 10, pp. 1519–1539, Oct. 2013.

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. SCARINO et al.: WEB-BASED TOOL FOR CALCULATING SPECTRAL BAND DIFFERENCE ADJUSTMENT FACTORS

[9] X. Xiong et al., “Terra and Aqua MODIS calibration algorithms and uncertainty analysis,” in Proc. SPIE, 4th Sens., Syst., Next-Gen. Satellites, Bruges, Belgium, Oct. 2005, vol. 5978, pp. 59780V-1–59780V-10. [10] C. Cao, F. J. De Luccia, X. Xiong, R. Wolfe, and F. Weng, “Early on-orbit performance of the visible infrared imaging radiometer suite onboard the Suomi National Polar-Orbiting Partnership (S-NPP) satellite,” IEEE Trans. Geosci. Remote Sens., vol. 52, no. 2, pp. 1142–1156, Feb. 2014. [11] G. Chander et al., “Overview of intercalibration of satellite instruments,” IEEE Trans. Geosci. Remote Sens., vol. 51, no. 3, pp. 1–25, Mar. 2013. [12] P. N. Slater et al., “Reflectance- and radiance-based methods for the in-flight absolute calibration of multispectral sensors,” Remote Sens. Environ., vol. 22, no. 11, pp. 11–37, 1987. [13] P. N. Slater, S. F. Biggar, K. J. Thome, D. I. Gellman, and P. R. Spyak, “Vicarious radiometric calibrations of EOS sensors,” J. Atmos. Ocean. Technol., vol. 13, no. 2, pp. 349–359, Apr. 1996. [14] P. M. Teillet et al., “A generalized approach to the vicarious calibration of multiple Earth observation sensors using hyperspectral data,” Remote Sens. Environ., vol. 77, no. 3, pp. 304–327, Sep. 2001. [15] P. Henry, G. Chander, B. Fougnie, C. Thomas, and X. Xiong, “Assessment of spectral band impact on intercalibration over desert sites using simulation based on EO-1 Hyperion data,” IEEE Trans. Geosci. Remote Sens., vol. 51, no. 3, pp. 1297–1308, Mar. 2013. [16] E. F. Vermote and Y. J. Kaufman, “Absolute calibration of AVHRR visible and near infrared channels using ocean and cloud views,” Int. J. Remote Sens., vol. 16, no. 13, pp. 2317–2340, 1995. [17] J. F. Le Marshal, J. J. Simpson, and Z. Jin, “Satellite calibration using a collocated nadir observation technique: Theoretical basis and application to the GMS-5 pathfinder benchmark period,” IEEE Trans. Geosci. Remote Sens., vol. 37, no. 1, pp. 499–507, Jan. 1999. [18] B. Lafrance, O. Hagolle, B. Bonnel, Y. Fouquart, and G. Brogniez, “Interband calibration over clouds for POLDER space sensor,” IEEE Trans. Geosci. Remote Sens., vol. 40, no. 1, pp. 131–142, Jan. 2002. [19] S. H. Ham and B. J. Sohn, “Assessment of the calibration performance of satellite visible channels using cloud targets: Application to Meteosat-8/9 and MTSAT-1R,” Atmos. Chem. Phys., vol. 10, no. 5, pp. 11 131–11 149, 2010. [20] A. Okuyama, “Outlined algorithm theoretical basis for MTSAT visible calibration for GSICS (liquid cloud method),” GSICS ATBD. [Online]. Available: https://gsics.nesdis.noaa.gov/pub/Development/AtbdCentral/ MTSAT_vis_vicarious_calibration_outline.pdf [21] D. R. Doelling, C. Lukashin, P. Minnis, B. Scarino, and D. Morstad, “Spectral reflectance corrections for satellite intercalibrations using SCIAMACHY data,” IEEE Geosci. Remote Sens. Lett., vol. 9, no. 1, pp. 119–123, Jan. 2012. [22] D. R. Doelling et al., “The intercalibration of geostationary visible imagers using operational hyper-spectral SCIAMACHY radiances,” IEEE Trans. Geosci. Remote Sens., vol. 51, no. 3, pp. 1245–1254, Mar. 2013. [23] D. R. Doelling, D. Morstad, B. R. Scarino, R. Bhatt, and A. Gopalan, “The characterization of deep convective clouds as an invariant calibration target and as a visible calibration technique,” IEEE Trans. Geosci. Remote Sens., vol. 51, no. 3, pp. 1147–1159, Mar. 2013. [24] B. R. Scarino et al., “Using SCIAMACHY to improve corrections for spectral band differences when transferring calibration between visible sensors,” in Proc. SPIE, Oct. 2012, vol. 8510, pp. 1–13. [25] G. N. Chander et al., “Applications of spectral band adjustment factors (SBAF) for cross-calibration,” IEEE Trans. Geosci. Remote Sens., vol. 51, no. 3, pp. 1267–1281, Mar. 2013. [26] H. Bovensmann et al., “SCIAMACHY: Mission objectives and measurement modes,” J. Atmos. Sci., vol. 56, no. 2, pp. 127–150, Jan. 1999. [27] J. Skupin et al., “SCIAMACHY solar irradiance observation in the spectral range from 240 to 2380 nm,” Adv. Space Res., vol. 35, no. 3, pp. 370–375, 2005. [28] G. Lichtenberg et al., “SCIAMACHY Level 1 data: Calibration concept and in-flight calibration,” Atmos. Chem. Phys., vol. 6, no. 12, pp. 5347–5367, Nov. 2006. [29] Disclaimer for SCIAMACHY IPF 7.03 Level 1b Products (SCI_NL_1P) PO-RS-MDA-GS-2009, vol. 15, no. 3L, 2011. [30] P. Minnis et al., “CERES Edition-2 cloud property retrievals using TRMM VIRS and Terra and Aqua MODIS data, Part I: Algorithms,” IEEE Trans. Geosci. Remote Sens., vol. 49, no. 11, pp. 4374–4400, Nov. 2011. [31] R. Bhatt, D. R. Doelling, D. Morstad, B. R. Scarino, and A. Gopalan, “Desert-based absolute calibration of successive geostationary visible sensors using a daily exoatmospheric radiance model,” IEEE Trans. Geosci. Remote Sens., vol. 52, no. 6, pp. 3670–3682, Jun. 2014.

13

[32] S. G. Ungar, E. M. Middleton, L. Ong, and P. K. E. Campbell, “EO-1 Hyperion onboard performance over eight years: Hyperion calibration,” in Proc. EARSel, 2009, pp. 1–6. [33] H. Cosnefroy, M. Leroy, and X. Briottet, “Selection and characterization of Saharan and Arabian desert sites for the calibration of optical satellite sensors,” Remote Sens. Environ., vol. 58, no. 1, pp. 101–114, Oct. 1996. [34] EUMETSAT GOME-2 Product Guide. [Online]. Available: http:// oiswww.eumetsat.org/WEBOPS/eps-pg/GOME-2/GOME2-PG4ProdOverview.htm

Benjamin R. Scarino received the B.S. and M.S. degrees in meteorology from the Pennsylvania State University, University Park, PA, USA, in 2008 and 2010, respectively. He is currently a Senior Research Scientist with Science Systems and Applications, Inc. (SSAI), Hampton, VA, USA, where he supports work performed in the NASA Langley Research Center. Specifically, he supports the improvement of the Clouds and the Earth’s Radiant Energy System (CERES) global multispectral cloud and surface property retrieval algorithms and satellite calibration/radiation property data set development projects. He is a specialist on intercalibration and trend analysis techniques for geostationary and polar-orbiting satellite instruments, developing signal-correction algorithms for interinstrument spectral band differences using hyperspectral detectors and absolute calibrations for satellite sensors. He also has specialized skill in using satellite instrument retrievals to derive global high-resolution clear-sky surface skin temperature for use in data assimilation and Earth radiation budget analysis. Mr. Scarino received the NASA Group Achievement Award and the SSAI Performance Award in 2014, for sustained excellence and innovation in developing and validating the CERES Editions 2 and 4 Climate Data Records, and the NASA Group Achievement Award in 2015, for outstanding accomplishments related to the Studies of Emissions and Atmospheric Composition, Clouds, and Climate Coupling by Regional Surveys (SEAC4RS) field campaign.

David R. Doelling received the B.S. degree in meteorology with The University of Utah, Salt Lake City, UT, USA, in 1985 and the M.S. degree in atmospheric science with the University of Washington, Seattle, WA, USA, in 1991. He is currently a Senior Research Scientist with the NASA Langley Research Center, Hampton, VA, USA, where he is the Time Interpolation and Spatial Averaging sublead for the Clouds and Earth’s Radiant Energy System (CERES) project and responsible for the diurnal averaging and spatial gridding of CERES footprint cloud and radiative flux parameters. He is also a member of the Geostationary Earth Radiation Budget Experiment and Megha-Tropiques Science Teams, which are projects similar to CERES that measure broadband fluxes. He has studied the orbital sampling errors for the proposed satellites as a member of the Climate Absolute Radiance and Refractivity Observatory (CLARREO) science definition team. His research interests include geostationary imager calibration, diurnal averaging techniques of satellite observations, and satellite sampling studies. Mr. Doelling is the Global Space-based Inter-Calibration System NASA representative and the lead for geostationary/moderate-resolution imaging spectroradiometer ray-matching and deep-convective-cloud visible calibration methods.

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. 14

Patrick Minnis was born in Oklahoma City, OK, USA. He received the B.E. degree in materials science and metallurgical engineering from Vanderbilt University, Nashville, TN, USA, in 1972; the M.S. degree in atmospheric science from Colorado State University, Fort Collins, CO, USA, in 1978; and the Ph.D. degree in meteorology from The University of Utah, Salt Lake City, UT, USA, in 1991. He is currently a Senior Research Scientist with the Climate Sciences Branch, NASA Langley Research Center, Hampton, VA, USA, where he has been serving for more than 35 years. He is the author/coauthor of over 250 peer-reviewed publications. His research is focused on the remote sensing of clouds and surface properties from satellite imagery for weather and climate investigations. He is a member of the Clouds and Earth’s Radiant Energy System, Atmospheric Radiation Measurement, and Aviation Climate Change Research Initiative Science Teams and leads a research group conducting analyses of polar-orbiting and geostationary satellite data for field missions, aircraft safety, and weather and climate research. Dr. Minnis is a Fellow of the American Geophysical Union and the American Meteorological Society and a member of the American Institute of Aeronautics and Astronautics (AIAA). He was a recipient of NASA medals for Exceptional Scientific Achievement in 1993, Exceptional Achievement in 2005, Exceptional Service in 2008, and Distinguished Service in 2011; the AMS Henry G. Houghton Award for Atmospheric Physics in 1998; and the 2011 AIAA Losey Atmospheric Sciences Award.

Arun Gopalan received the B.S. degree in mechanical engineering from Victoria Jubilee Technical Institute, University of Bombay, Mumbai, India, in 1991 and the M.S. degree in mechanical engineering from the State University of New York at Stony Brook, Stony Brook, NY, USA, in 1993, where he is currently working toward the Ph.D. degree and has completed all Ph.D. requirements excluding dissertation defense as of 1997. His dissertation focused on the research and development of an optimal retrieval algorithm for simultaneous extraction of diurnal stratospheric nitrogen dioxide and aerosol extinction profiles using Earth-limb Infrared Emission measurements from the National Aeronautics and Space Administration (NASA) Upper Atmospheric Research Satellite Cryogenic Limb Array Etalon Spectrometer instrument. He is currently a Senior Research Scientist with Science Systems and Applications, Inc., Hampton, VA, USA, in support of NASA Langley Research Center’s Cloud and Earth’s Radiant Energy System and Climate Absolute Radiance and Refractivity Observatory projects. He has previously coauthored a number of journal papers and worked in the area of satellite remote sensing for over 15 years with earth science research groups at the NASA Goddard Space Flight Center, Greenbelt, MD, USA, and the National Center for Atmospheric Research, Boulder, CO, USA. His research interests include studies of postlaunch calibration techniques to understand satellite sensor performance, science retrieval algorithm development, remotely sensed data processing methods, aerosol-cloud-climate feedback mechanisms, visualization of largescale earth science data sets, and data mining.

Thad Chee received the B.A. degree in computer science and the B.S. degree in computer engineering from Iowa State University, Ames, IA, USA, in 1991 and 1992, respectively, and the M.B.A. degree in information technology from Old Dominion University, Norfolk, VA, USA, in 2011. He is currently working toward the Ph.D. degree in business management specializing in information technology. He is currently a Senior Web Developer with Science Systems and Applications, Inc., Hampton, VA, USA, in support of the Clouds and Earth’s Radiant Energy System project at the National Aeronautics and Space Administration Langley Research Center, Hampton. His primary roles include developing web applications and supercomputing solutions. His research interests include social media, international business, and finance.

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING

Rajendra Bhatt received the B.S. degree in electronics engineering from Tribhuvan University, Kirtipur, Nepal, in 2002 and the M.S. degree in electrical engineering from South Dakota State University, Brookings, SD, USA, in 2009. He is currently a Senior Research Scientist with Science Systems and Applications, Inc., Hampton, VA, USA, in support of the Clouds and Earth’s Radiant Energy System project at the NASA Langley Research Center, Hampton. His research is focused on developing vicarious techniques for in-flight calibration of satellite visible and infrared observations to assess the sensor’s performance on-orbit and ensure the usability of climate data.

Constantine Lukashin received the M.S. degree in nuclear physics and engineering from Saint Petersburg State Polytechnic University, Saint Petersburg, Russia, in 1990 and the Ph.D. degree in physics from Virginia Polytechnic Institute and State University, Blacksburg, VA, USA, in 2000. He participated in nuclear and high-energy particle physics experiments from 1990 to 2001, working at national accelerator facilities in Russia, Italy, and the U.S. In 2001, he joined the NASA Clouds and Earth’s Radiant Energy System (CERES) team as a Research Scientist for the development and validation of angular distribution models (ADMs). As a member of the CERES team, he developed an artificial neural network approach for reproducing ADMs in the absence of coincident imager observations. He is currently a Research Physical Scientist with the NASA Langley Research Center, Hampton, VA, USA, where he is responsible for developing the Climate Absolute Radiance and Refractivity Observatory reference intercalibration methodology in the reflected solar wavelength range.

Conor Haney received the B.S. and M.S. degrees in atmospheric sciences from the University of Illinois at Urbana–Champaign, Champaign, IL, USA, in 2010 and 2013, respectively. He is currently a Research Scientist with Science Systems and Applications, Inc., Hampton, VA, USA, supporting NASA Langley Research Center contracts. He is primarily responsible for maintaining the real-time visible sensor calibration for geostationary satellites for the Clouds and Earth’s Radiant Energy System project. He is currently developing and improving geostationary visible imager calibration techniques and monitoring their performance.