• Angal, A., X. J. Xiong, J. Sun, and X. Geng, 2015: On-orbit noise characterization of MODIS reflective solar bands. J. Appl. Remote Sens., 9, 094092, https://doi.org/10.1117/1.JRS.9.094092.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Bhatt, R., D. R. Doelling, A. Wu, X. Xiong, B. R. Scarino, C. O. Haney, and A. Gopalan, 2014: Initial stability assessment of S-NPP VIIRS reflective solar band calibration using invariant desert and deep convective cloud targets. Remote Sens., 6, 28092826, https://doi.org/10.3390/rs6042809.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Cao, C., and A. K. Heidinger, 2002: Inter-comparison of the longwave infrared channels of MODIS and AVHRR/NOAA-16 using simultaneous nadir observations at orbit intersections. Earth Observing Systems VII, W. L. Barnes, Ed., Society of Photo-Optical Instrumentation Engineers (SPIE Proceedings, Vol. 4814), 306–316, https://doi.org/10.1117/12.451690.

    • Crossref
    • Export Citation
  • Cao, C., M. Weinreb, and H. Xu, 2004: Predicting simultaneous nadir overpasses among polar-orbiting meteorological satellites for the intersatellite calibration of radiometers. J. Atmos. Oceanic Technol., 21, 537542, https://doi.org/10.1175/1520-0426(2004)021<0537:PSNOAP>2.0.CO;2.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Cao, C., J. Xiong, S. Blonski, Q. Liu, S. Uprety, X. Shao, Y. Bai, and F. Weng, 2013: Suomi NPP VIIRS sensor data record verification, validation, and long-term performance monitoring. J. Geophys. Res. Atmos., 118, 11 66411 678, https://doi.org/10.1002/2013JD020418.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Cao, C., F. Deluccia, X. Xiong, R. Wolfe, and F. Weng, 2014: 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., 52, 11421156, https://doi.org/10.1109/TGRS.2013.2247768.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Choi, T. J., J. Sun, B. Zhang, Z. Wang, C. Cao, F. Weng, and M. Wang, 2017: Suomi-NPP VIIRS initial reprocessing improvements and validations in the reflective solar bands (RSBS). Earth Observing Systems XXII, J. J. Butler, X. Xiong, and X. Gu Eds., Society of Photo-Optical Instrumentation Engineers (SPIE Proceedings, Vol. 10402), 104021V, https://doi.org/10.1117/12.2274081.

    • Crossref
    • Export Citation
  • Chu, M., J. Sun, and M. Wang, 2016: Radiometric evaluation of the SNPP VIIRS reflective solar band sensor data records via inter-sensor comparison with Aqua MODIS. Earth Observing Systems XXI, J. J. Butler, X. Xiong, and X. Gu Eds., Society of Photo-Optical Instrumentation Engineers (SPIE Proceedings, Vol. 9972), 99721R, https://doi.org/10.1117/12.2236942.

    • Crossref
    • Export Citation
  • Doelling, D. R., C. Lukashin, P. Minnis, B. Scarino, and D. Morstad, 2012: Spectral reflectance corrections for satellite intercalibrations using SCIAMACHY data. IEEE Geosci. Remote Sens. Lett., 9, 119123, https://doi.org/10.1109/LGRS.2011.2161751.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Gordon, H. R., and M. Wang, 1994: Retrieval of water-leaving radiance and aerosol optical thickness over the oceans with SeaWiFS: A preliminary algorithm. Appl. Opt., 33, 443452, https://doi.org/10.1364/AO.33.000443.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Heidinger, A. K., C. Cao, and J. T. Sullivan, 2002: Using Moderate Resolution Imaging Spectroradiometer (MODIS) to calibrate advanced very high resolution radiometer reflectance channels. J. Geophys. Res., 107, 4702, https://doi.org/10.1029/2001JD002035.

    • Search Google Scholar
    • Export Citation
  • NOAA National Calibration Center, 2016: SNPP SNOs with other satellites. Accessed July 2016, http://ncc.nesdis.noaa.gov/VIIRS/SNOPredictions/index.php.

  • NOAA Products Validation System, 2016: NOAA-19 sounding products. Subset used: 1200:00 UTC 14 January 2016, accessed 29 October 2016. http://www.ospo.noaa.gov/Products/atmosphere/soundings/index.html.

  • NOAA/STAR VIIRS SDR Team, 2013: Joint Polar Satellite System (JPSS) Visible Infrared Imaging Radiometer Suite (VIIRS) sensor data records (SDR) algorithm theoretical basis document (ATBD). Rev. C, NOAA Doc. E/RA-00003, 161 pp., http://ncc.nesdis.noaa.gov/documents/documentation/ATBD-VIIRS-RadiometricCal_20131212.pdf.

  • Scarino, B. R., 2016: Spectral band adjustment factor. Accessed July 2016, http://angler.larc.nasa.gov/SBAF.

  • Scarino, B. R., D. R. Doelling, P. Minnis, A. Gopalan, T. Chee, R. Bhatt, C. Lukashin, and C. Haney, 2016: A web-based tool for calculating spectral band difference adjustment factors derived from SCIAMACHY hyperspectral data. IEEE Trans. Geosci. Remote Sens., 54, 25292542, https://doi.org/10.1109/TGRS.2015.2502904.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Sun, J., and M. Wang, 2014: Visible Infrared Imaging Radiometer Suite solar diffuser calibration and its challenges using solar diffuser stability monitor. Appl. Opt., 53, 85718584, https://doi.org/10.1364/AO.53.008571.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Sun, J., and M. Wang, 2015a: On-orbit characterization of the VIIRS solar diffuser and solar diffuser screen. Appl. Opt., 54, 236252, https://doi.org/10.1364/AO.54.000236.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Sun, J., and M. Wang, 2015b: On-orbit calibration of the Visible Infrared Imaging Radiometer Suite reflective solar bands and its challenges using a solar diffuser. Appl. Opt., 54, 72107223, https://doi.org/10.1364/AO.54.007210.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Sun, J., and M. Wang, 2015c: Radiometric calibration of the Visible Infrared Imaging Radiometer Suite reflective solar bands with robust characterizations and hybrid calibration coefficients. Appl. Opt., 54, 93319342, https://doi.org/10.1364/AO.54.009331.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Sun, J., and M. Wang, 2016: VIIRS reflective solar bands calibration progress and its impact on ocean color products. Remote Sens., 8, 194, https://doi.org/10.3390/rs8030194.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Sun, J., and M. Wang, 2017: Reflective solar bands calibration improvements and look up tables for SNPP VIIRS operational mission-long SDR reprocessing. Earth Observing Systems XXII, J. J. Butler, X. Xiong, and X. Gu Eds., Society of Photo-Optical Instrumentation Engineers (SPIE Proceedings, Vol. 10402), 104021W, https://doi.org/10.1117/12.2271636.

    • Crossref
    • Export Citation
  • Sun, J., A. Angal, X. Xiong, H. Chen, X. Geng, A. Wu, T. Choi, and M. Chu, 2012: MODIS reflective solar bands calibration improvements in Collection 6. Earth Observing Missions and Sensors: Development, Implementation, and Characterization II, H. Shimoda et al., Eds., Society of Photo-Optical Instrumentation Engineers (SPIE Proceedings, Vol. 8528), 85280N, https://doi.org/10.1117/12.979733.

    • Crossref
    • Export Citation
  • Sun, J., X. Xiong, A. Angal, H. Chen, A. Wu, and X. Geng, 2014: Time-dependent response versus scan angle for MODIS reflective solar bands. IEEE Trans. Geosci. Remote Sens., 52, 31593174, https://doi.org/10.1109/TGRS.2013.2271448.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Sun, J., M. Chu, and M. Wang, 2016: Degradation nonuniformity in the solar diffuser bidirectional reflectance distribution function. Appl. Opt., 55, 60016016, https://doi.org/10.1364/AO.55.006001.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Uprety, S., and C. Cao, 2015: Suomi NPP VIIRS reflective solar band on-orbit radiometric stability and accuracy assessment using desert and Antarctica Dome C sites. Remote Sens. Environ., 166, 106115, https://doi.org/10.1016/j.rse.2015.05.021.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Uprety, S., C. Cao, X. Xiong, S. Blonski, A. Wu, and X. Shao, 2013: Radiometric intercomparison between Suomi-NPP VIIRS and Aqua MODIS reflective solar bands using simultaneous nadir overpass in the low latitudes. J. Atmos. Oceanic Technol., 30, 27202736, https://doi.org/10.1175/JTECH-D-13-00071.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Uprety, S., S. Blonski, and C. Cao, 2016: On-orbit radiometric performance characterization of S-NPP VIIRS reflective solar bands. Earth Observing Missions and Sensor: Development, Implementation, and Characterization IV, X. J. Xiong, S. A. Kuriakose, and T. Kimura, Eds., Society of Photo-Optical Instrumentation Engineers (SPIE Proceedings, Vol. 9881), 98811H, https://doi.org/10.1117/12.2223788.

    • Crossref
    • Export Citation
  • Vallado, D. A., P. Crawford, R. Hujsak, and T. S. Kelso, 2006: Revisiting Spacetrack Report #3. AIAA/AAS Astrodynamics Specialist Conf. and Exhibit, Keystone, CO, AIAA, AIAA 2006-6753, https://doi.org/10.2514/6.2006-6753.

    • Crossref
    • Export Citation
  • Wang, M., 2007: Remote sensing of the ocean contributions from ultraviolet to near-infrared using the shortwave infrared bands: Simulations. Appl. Opt., 46, 15351547, https://doi.org/10.1364/AO.46.001535.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Wang, M., 2016: Rayleigh radiance computations for satellite remote sensing: Accounting for the effect of sensor spectral response function. Opt. Express, 24, 12 41412 429, https://doi.org/10.1364/OE.24.012414.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Wang, M., and W. Shi, 2007: The NIR-SWIR combined atmospheric correction approach for MODIS ocean color data processing. Opt. Express, 15, 15 72215 733, https://doi.org/10.1364/OE.15.015722.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Wang, M., X. Liu, L. Tan, L. Jiang, S. Son, W. Shi, K. Rausch, and K. Voss, 2013: Impacts of VIIRS SDR performance on ocean color products. J. Geophys. Res. Atmos., 118, 10 34710 360, https://doi.org/10.1002/jgrd.50793.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Wang, M., and et al. , 2014: Evaluation of VIIRS ocean color products. Ocean Remote Sensing and Monitoring from Space, R. J. Frouin, D. Pan, and H. Murakami, Eds., Society of Photo-Optical Instrumentation Engineers (SPIE Proceedings, Vol. 9261), 92610E, https://doi.org/10.1117/12.2069251.

    • Crossref
    • Export Citation
  • Wang, M., W. Shi, L. Jiang, X. Liu, and K. Voss, 2015a: Technique for monitoring performance of VIIRS reflective solar bands for ocean color data processing. Opt. Express, 23, 14 44614 460, https://doi.org/10.1364/OE.23.014446.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Wang, M., and et al. , 2015b: VIIRS ocean color research and applications. 2015 IEEE International Geoscience and Remote Sensing Symposium: Proceedings, IEEE, 29112914, https://doi.org/10.1109/IGARSS.2015.7326424.

    • Crossref
    • Export Citation
  • Wolf, R. E., S. Devadiga, G. Ye, E. J. Masuoka, and R. J. Schweiss, 2010: Evaluation of the VIIRS land algorithms at Land PEATE. 2010 IEEE International and Remote Sensing Symposium: Proceedings, IEEE, 304307, https://doi.org/10.1109/IGARSS.2010.5652831.

    • Crossref
    • Export Citation
  • Wu, A., X. Xiong, C. Cao, and K. Chiang, 2016: Assessment of SNPP VIIRS VIS/NIR radiometric calibration stability using Aqua MODIS and invariant surface targets. IEEE Trans. Geosci. Remote Sens., 54, 29182924, https://doi.org/10.1109/TGRS.2015.2508379.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Xiong, X., J. Sun, X. Xie, W. Barnes, and V. Salomonson, 2010: On-orbit calibration and performance of Aqua MODIS reflective solar bands. IEEE Trans. Geosci. Remote Sens., 48, 535546, https://doi.org/10.1109/TGRS.2009.2024307.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • View in gallery

    Matching SNPP VIIRS and Aqua MODIS granules in an SNO over Antarctica on 21 Mar 2016.

  • View in gallery

    Error bar vs sample size for Aqua MODIS band 8 vs SNPP VIIRS M1 intercomparison: (a) the scatterplot of the comparison results for the four spatial scales ranging from 36 to 160 km, (b) the scatterplot for the same four spatial scales with extended sample size display range, (c) the scatterplot with two results from the two smaller spatial scales rescaled in sample size to match the 80-km square region result, and (d) results with a threshold of 0.02 in error bar and 70% of the maximum sample size applied. Each point represents one SNO event.

  • View in gallery

    Four different radiance comparison charts for Aqua MODIS band 4 vs SNPP VIIRS M4, using IDPS SDRs, with the lower 20% of radiances shown (green diamond symbols): (a) Aqua MODIS radiance vs SNPP VIIRS radiance, (b) absolute radiance difference of SNPP VIIRS and Aqua MODIS with respect to SNPP VIIRS radiance, (c) relative radiance difference (%) with respect to SNPP VIIRS radiance, and (d) time series displaying the outlying features caused by low-radiance discrepancy.

  • View in gallery

    Comparison times series for SNPP VIIRS M1–M4 vs the matching Aqua MODIS bands. Time series mean for OC-based time series (solid light-blue line), and the 2% marks above and below the time series mean (dashed blue lines). SBAF is indicated (dotted line).

  • View in gallery

    Comparison time series for SNPP VIIRS M5–M8 vs the matching Aqua MODIS bands. Time series mean for OC-based time series (solid light-blue line), and the 2% marks above and below the time series mean (dashed blue lines). SBAF is indicated (dotted line).

  • View in gallery

    NOAA-19 ATOVS retrieval for 1200–1300 UTC 14 Jan 2016, shown for (top left) brightness temperature, (top right) water vapor content, (bottom left) cloud-top pressure, and (bottom right) effective cloud amount. Each window shows the location of the scene of comparison for the SNO that occurred at around UTC 1225 14 Jan 2016 (white arrow).

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 72 72 14
PDF Downloads 54 54 11

Performance Evaluation of On-Orbit Calibration of SNPP VIIRS Reflective Solar Bands via Intersensor Comparison with Aqua MODIS

View More View Less
  • 1 NOAA/NESDIS/Center for Satellite Applications and Research, College Park, Maryland, and Cooperative Institute for Research in the Atmosphere, Colorado State University, Fort Collins, Colorado
  • | 2 NOAA/NESDIS/Center for Satellite Applications and Research, College Park, and Global Science and Technology, Greenbelt, Maryland
  • | 3 NOAA/NESDIS/Center for Satellite Applications and Research, College Park, Maryland
© Get Permissions
Full access

Abstract

An intersensor comparison is carried out to evaluate the radiometric performance of the reflective solar bands (RSBs) of the first Visible Infrared Imaging Radiometer Suite (VIIRS) on board the Suomi National Polar-Orbiting Partnership (SNPP) satellite. Two versions of sensor data records (SDRs) for moderate-resolution RSBs M1–M8 (410–1238 nm)—one version from the NOAA Ocean Color (OC) Team and the operational version from the Interface Data Processing Segment (IDPS)—are compared against the well-calibrated Moderate Resolution Imaging Spectroradiometer (MODIS) on the Aqua satellite. This comparison fully exploits the moderate resolution of the sensors and a precise simultaneous nadir overpass (SNO) analysis in a “nadir only” approach to achieve a precision better than 1%. The key issues found to impact the SNO analysis are 1) an underlying bias beyond the 80-km spatial scale, 2) a scene-based sporadic variability of about 2% affecting the sample size selection criteria, and 3) large relative deviations at low radiances. It is shown that the OC SDRs achieve significantly better agreement with Aqua MODIS, such as smaller temporal variation, improved agreement in the early mission, and no observable long-term drift. The lone exception is the downward drift of about 1% in the Aqua MODIS band 8 (412 nm) versus SNPP VIIRS band M1 time series that possibly started in late 2013, which is ultimately attributed to errors in Aqua MODIS band 8. Finally, the long-term drift in the IDPS SDRs further illustrates the consequence of the worsening bias within the standard RSB calibration that will infect any versions of the VIIRS SDRs not mitigated for this error.

© 2018 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).

Corresponding author: Mike Chu, mike.chu@noaa.gov

Abstract

An intersensor comparison is carried out to evaluate the radiometric performance of the reflective solar bands (RSBs) of the first Visible Infrared Imaging Radiometer Suite (VIIRS) on board the Suomi National Polar-Orbiting Partnership (SNPP) satellite. Two versions of sensor data records (SDRs) for moderate-resolution RSBs M1–M8 (410–1238 nm)—one version from the NOAA Ocean Color (OC) Team and the operational version from the Interface Data Processing Segment (IDPS)—are compared against the well-calibrated Moderate Resolution Imaging Spectroradiometer (MODIS) on the Aqua satellite. This comparison fully exploits the moderate resolution of the sensors and a precise simultaneous nadir overpass (SNO) analysis in a “nadir only” approach to achieve a precision better than 1%. The key issues found to impact the SNO analysis are 1) an underlying bias beyond the 80-km spatial scale, 2) a scene-based sporadic variability of about 2% affecting the sample size selection criteria, and 3) large relative deviations at low radiances. It is shown that the OC SDRs achieve significantly better agreement with Aqua MODIS, such as smaller temporal variation, improved agreement in the early mission, and no observable long-term drift. The lone exception is the downward drift of about 1% in the Aqua MODIS band 8 (412 nm) versus SNPP VIIRS band M1 time series that possibly started in late 2013, which is ultimately attributed to errors in Aqua MODIS band 8. Finally, the long-term drift in the IDPS SDRs further illustrates the consequence of the worsening bias within the standard RSB calibration that will infect any versions of the VIIRS SDRs not mitigated for this error.

© 2018 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).

Corresponding author: Mike Chu, mike.chu@noaa.gov

1. Introduction

Ever since the launch of the first Visible Infrared Imaging Radiometer Suite (VIIRS) on board the Suomi National Polar-Orbiting Partnership (SNPP) satellite on 28 October 2011, efforts to evaluate the performance of its sensor data records (SDRs) have been ongoing (Bhatt et al. 2014; Cao et al. 2013, 2014; Uprety et al. 2013, 2016; Uprety and Cao 2015; Wu et al. 2016). The Interface Data Processing Segment (IDPS) generates the official SDRs, which contain the crucial first-level data output, that are then used for the further generation of the environmental data records (EDRs) and the higher-level science products. The correctness of the calibration of the reflective solar bands (RSBs), for which the calibrated top-of-the-atmosphere (TOA) radiance is a primary SDR, is particularly important for the accuracy of the ocean color products (Wang et al. 2013, 2015a). Indeed, the VIIRS-derived ocean color EDRs generated using the IDPS SDRs by the NOAA Ocean Color (OC) Team show long-term drift in normalized water-leaving radiance spectra and chlorophyll-a concentration and diverging discrepancy against key in situ data, as well as other artificial features (Wang et al. 2013, 2014, 2015b), thus showing that the IDPS SDRs need further improvements to its on-orbit calibration.

In a series of independent investigations, Sun and Wang (2014, 2015a,b) have achieved for the NOAA OC Team a robust on-orbit calibration of the short-wavelength RSBs through rigorous examinations of the core on-orbit calibration components to arrive at much-improved SDRs. Significant improvements have been made to each of the core components, including the bidirectional reflectance factor (BRF) of the solar diffuser (SD), the vignetting function (VF) characterizing the transmission of the attenuation screen, the H-factor characterizing the SD degradation as measured by the SD stability monitor (SDSM) and, finally, the RSB calibration coefficients or the F-factors. The aforementioned anomalous features, noise, and artificial yearly variations have been dramatically reduced to achieve an impressive stability of about 0.2% or better in the newly derived standard calibration results. Arguably the most critical progress made outside the standard procedure is the discovery of the “SD degradation nonuniformity effect,” or the anisotropic behavior in the degradation of the SD reflectance, found to be a key contributor to the worsening RSB calibration error and the long-term drift in science products (Sun and Wang 2014, 2015b,c; Sun et al. 2016). This problem, which stems from errors in the standard RSB calibration methodology, has been resolved via a new hybrid method that incorporates lunar calibration into the standard SD-based result to mitigate the long-term drift (Sun and Wang 2015c). As the ocean color EDRs are highly sensitive to the quality of the input SDRs, their performance also serves as a stringent check on the quality of the input SDRs. The subsequently reprocessed ocean color EDRs built from the improved SDRs show a clear upgrade in accuracy to meet the very demanding ocean color product requirements (Wang et al. 2014, 2015a,b; Sun and Wang 2016). In the context of this paper, the SDRs for SNPP VIIRS generated by the NOAA OC Team will be designated as the OC SDRs to distinguish them from the IDPS SDRs. This investigation will put OC SDRs through a direct radiometric performance evaluation via an intersensor comparison with another well-calibrated sensor—the Moderate Resolution Imaging Spectroradiometer (MODIS) instrument aboard the Aqua satellite. The MODIS Collection 6 (C6) release (Sun et al. 2012, 2014) for Aqua MODIS will be used, and a comparison will also be made against IDPS SDRs. The goal is to demonstrate that the OC SDRs indeed are long-term radiometrically stable and that their underlying on-orbit RSB calibration in the final form of hybrid method (Sun and Wang 2015c) is correct. An intersensor radiometric comparison, especially when carried out against another well-calibrated sensor, can precisely evaluate the long-term radiometric stability of a sensor that reveals the correctness of the on-orbit calibration methodology. Additionally, an intersensor comparison is a useful tool to reveal any issues or effects, manifesting as unstable features in the comparison time series, not correctly captured by on-orbit calibration.

SNPP VIIRS and Aqua MODIS are both cross-track whisk broom scanning radiometers with many spectrally matching bands, leading to a very straightforward comparison between the two instruments for the matching bands. The SNPP and Aqua satellites also fly at nearly similar sun-synchronous polar orbits, making simultaneous nadir overpasses (SNOs) comparison studies (Cao and Heidinger 2002; Heidinger et al. 2002; Cao et al. 2004) highly suitable. As Aqua MODIS has long been established to be among the most stable and well-calibrated sensors (Xiong et al. 2010; Angal et al. 2015), a comparison study is expected to generate a definitive and clear result. In contrast to other vicarious comparison methods, including using nonsimultaneous observations and well-studied specified land sites such as Libyan deserts (Bhatt et al. 2014; Uprety and Cao 2015; Uprety et al. 2016; Wu et al. 2016), a strict SNO approach minimizes variability coming from scene variations or different viewing conditions. Scene variations caused by significant time differences can obfuscate the intercomparison result, but the matching conditions for SNPP VIIRS and Aqua MODIS are such that many precise nadir-matching events with very short time differences exist. These exact nadir-matching near-simultaneous events form the basis data of this study. Moreover, the moderate spatial resolutions of the two sensors allow the comparison to be confined to a sufficiently small spatial extent and still contain enough collocated pixels for statistically robust outcomes. Confining the comparison to be within a small area is another key restriction to maintain fidelity to the original purpose of the precise nadir-matching approach, which is to minimize variability associated with a large spatial extent by a “nadir only” restriction. For example, if an observation extends over a large angular range, then instrument effects such as the response-versus-scan angle (RVS) effect of the scan-mirror or scene-based effects such as the bidirectional reflectance distribution function (BRDF) may begin to affect the accuracy in the comparison analysis. To demonstrate, a 36-km square area small enough to minimize variability or bias, including the RVS and scene BRDF effects, is just large enough to generate up to 1296 matching pixel pairs between MODIS and VIIRS, sufficient for good statistics. The pixel-based analysis adopted for this study fully exploits this capability to achieve reliable comparison time series with an overall precision of about 1% or better by eventually using an area just small enough to achieve a nadir-only condition to avoid unwanted variability and bias but large enough to contain enough pixels. The caveat is that the nontrivial interplay between data statistics and the selection process must first be carefully examined to establish both clarity of interpretation and reliability of the result, and this is simply because the current level of precision, limited by the spatial resolution at 1 km or so, remains susceptible to numerous subtleties. The investigation will examine the key statistics and data property issues to lay down some necessary conditions, which will further serve as useful groundwork for future studies. Earlier notable studies of SNPP VIIRS versus Aqua MODIS include an extended-SNO (SNOx) version of the comparison (Uprety et al. 2013, 2016) using the IDPS SDRs. Other variants using fixed Earth targets or a different version of the calibrated SDRs also exist (Bhatt et al. 2014; Uprety and Cao 2015; Wu et al. 2016). Information gained from these earlier studies will help to strengthen the conclusions reached in this work. This evaluation, powered by the overall precision in the comparison time series to better than ~1%, will show that the OC SDRs achieve superior radiometric performance, are the first set of SDRs to illustrate successful VIIRS RSB calibration capturing the correct SD degradation performance, and furthermore validate the successful performance of the derived ocean color EDRs (Wang et al. 2014, 2015b; Sun and Wang 2016). A preliminary result stemming from this analysis using the 80-km comparison areas was reported earlier (Chu et al. 2016). The updated and complete results using the 36-km comparison areas are presented here along with more stringent statistical examinations and the new findings. One key update from the earlier report is that the hybrid method mitigation of the RSB calibration error has been applied also for SNPP VIIRS M5 in this work, whereas in the preliminary examination only RSBs M1–M4 have been mitigated, thus the unmitigated M5 result shows a drift since 2013.

It can be argued that the intersensor comparison has already entered a precision era given the subpercent capability resulting from spatial resolutions reaching the moderate scale of 1 km and smaller. From this perspective, the success of the OC SDRs also previews the future of intersensor comparisons with VIIRS anchoring the next-generational comparison standards to advance deeper into the precision era. With the OC SDRs in place, SNPP VIIRS now fully demonstrates its qualifications to take over Aqua MODIS as the multispectral suite sensor reference. Aqua MODIS remains a quality sensor but an aging one, showing more calibration issues in recent years; it is also an older design with greater performance limitations, including smaller dynamic ranges. This radiometric comparison is made against Aqua MODIS not only for the technical evaluation but also to showcase SNPP VIIRS as the follow-up reference suite with superior capability.

The rest of the paper is organized as follows. In section 2, the SNO methodology and the comparison analysis procedure for SNPP VIIRS versus Aqua MODIS are presented. In section 3, the key postprocessing, statistical analysis, and selection criteria are described. In section 4, comparison time series using both IDPS SDRs and OC SDRs are shown. The results will demonstrate that the OC SDRs achieve overall better performance and that they are free of the long-term drift that impacts the IDPS SDRs. In section 5, key discussions are provided to make explicit some subtle but impactful issues. In section 6, a summary and conclusions are provided for key findings from this comparison study.

2. Procedure

This investigation seeks to address the on-orbit RSB calibration methodology that impacts the performance of SDR radiometric data. Because the on-orbit calibration methodology computes the calibration coefficient on a relative basis with the overall result normalized to an initial point of the mission, the evaluation of the long-term stability in the mission-long radiometric data, which is postlaunch, indeed evaluates the correctness of the on-orbit RSB calibration methodology. Any anomalous features in the radiometric performance, such as any irregular variation or long-term drift, necessarily reveal some inadequacies of the on-orbit calibration procedure. An intersensor comparison analysis, when correctly done using another high-quality sensor, is capable of generating a precise result on the order of 1% or better to reveal these features.

The spectrally matching RSBs of Aqua MODIS and SNPP VIIRS compared in this analysis and their specifications are shown in Table 1. The nominal central wavelength of each band is taken from the work of Wang et al. (2013) and Wang (2016) and differs slightly from the official specifications. VIIRS RSBs M1–M8 (410–1238 nm) have moderate resolution at 750 m, and the corresponding moderate-resolution Aqua MODIS bands are at 1-km resolution. Aqua MODIS also has higher spatial resolution imaging bands, listed in Table 1, that match VIIRS bands spectrally, but for these imaging bands the aggregated 1-km-resolution data will be used instead. Except for VIIRS RSBs M6 and M8, which are single-gain bands, the maximum scene radiance Lmax for other RSBs is shown for both high-gain and low-gain settings. However, it is Aqua MODIS with a lower Lmax that sets the response-saturation limit for the comparison.

Table 1.

Specifications for SNPP VIIRS and Aqua MODIS bands used in the intersensor comparison; Lmax and spectral radiance have units of W m−2 μm−1 sr−1.

Table 1.

The SNPP and Aqua satellites follow very similar sun-synchronous polar-orbiting orbits with the main difference being that the former flies at a mean altitude of 840 km with an orbital period of 101 min, while the latter is at 705 km with an orbital period of 99 min. Because of its slightly faster pace, the Aqua satellite will have multiple passes under the SNPP satellite daily. A small subset of the SNOs will meet the exact one-pixel nadir-matching condition with a very short time difference between the nadir crossings suitable for high-quality comparison analysis. The NOAA Calibration Center (NCC) publishes daily information of SNPP SNOs with other satellites, including ones for the Aqua satellite (NOAA National Calibration Center 2016), that is computed from the Simplified General Perturbation (SGP) model’s algorithm (Vallado et al. 2006). The Aqua MODIS and SNPP VIIRS geolocation data from the beginning of the VIIRS mission up to 15 July 2016 are searched through to find those SNOs meeting the precise nadir-match condition within a single pixel. Figure 1 illustrates a precise SNO over Antarctica, as is common, that occurred on 21 March 2016. This is a typical event with a time difference of 41 s between nadir crossings. The dashed boundary lines map out the 5-min Aqua MODIS granule coverage, whereas the solid boundary lines outline the five matching SNPP VIIRS granules, each with 86 s of coverage. The dashed lines running along the granule center trace the two flight tracks, and the small red square near the middle shows an 80-km square region centered at the matching nadir where the two flight tracks cross. It can be seen that the 80-km square region covers a small spatial extent of the scan swath coverage and that the comparison analysis will be confined to regions no larger than this size. However, larger areas or areas of different shapes can be used to examine statistics, variability, and data selection requirements.

Fig. 1.
Fig. 1.

Matching SNPP VIIRS and Aqua MODIS granules in an SNO over Antarctica on 21 Mar 2016.

Citation: Journal of Atmospheric and Oceanic Technology 35, 2; 10.1175/JTECH-D-17-0008.1

Between 28 October 2011 (the launch of the SNPP VIIRS mission) and 15 June 2016, 732 precise nadir matches have occurred. This averages to about one match per 2.2 days (55 h). The number of usable matches is further reduced by limitations, such as band response saturation in MODIS bands as a result of their smaller dynamic range. In the final analysis, the spatial difference of the selected matches averages close to 0.006°, confirming that nadir matches for the entire dataset are highly precise. The average time difference is about 44 s with no time difference greater than 2 min. In hindsight, it has become clear that the exact nadir-match condition and the orbit geometry of the two sensors are already highly restrictive, thus no additional time-difference threshold is needed.

The radiometry data of the IDPS SDRs and OC SDRs corresponding to each precise SNO are used to compare against the Aqua MODIS C6 radiometry data. Different scene sizes ranging from 30- to 160-km square regions are used to analyze the collocated pixels of Aqua MODIS and SNPP VIIRS and to make a radiometry comparison. In addition to a precise nadir match and a short time threshold, homogeneity of the scenes is another primary and physical requirement to ensure stable scene conditions. The homogeneity of each pixel is represented by the percentage standard deviation (STD) of the radiance of the pixel, which is calculated using the pixel itself and its eight neighbors. Pixels not satisfying the 4.5% homogeneity requirement are excluded in the analyses. In summary, pixels within the small regions centering on the precisely matched nadirs with a short time difference and high homogeneity are selected for radiometric comparison. In each matching nadir event, each pair of collocated pixels, matched using latitude and longitude information, is used to calculate a ratio of the radiance of Aqua MODIS band over the corresponding radiance of the SNPP VIIRS band, and the collection of all calculated ratios form the population sample upon which statistics will be derived for the event. The most important primary statistics are the average and the STD, or the error bar, of the population sample, which will represent the radiometric comparison ratio for the nadir-matching scene. The mean ratios and error bars from all nadir-matching events then form the time series as a representation of the underlying comparison trend.

Additional statistical analyses of the time series outcomes and the final selection criteria are described in the following section. The relevant issues include scene variability, bias, discrepancy at low radiance, and the proper determination of the selection criteria. For the final time series, each qualified SNO event is to have a relative STD below 2%, a sample size above 70% of the maximum possible size, and radiances excluding the lowest 20%. The time series built from the 36-km square comparison areas are used for the performance evaluation of the IDPS SDRs and the OC SDRs. This area size choice is effectively the minimum area size necessary to generate enough collocated pixels to provide good statistical precision, which for this study is at the level of 1%, while maintaining an approximate nadir-only condition. This level of precision can readily expose any multiyear radiometric drift on the order of 1% to achieve the main goal of demonstrating the diverging difference between the calibration of the IDPS and OC SDRs.

On the other hand, this work does not pursue issues of absolute calibration or other inputs outside the on-orbit RSB calibration procedure that characterizes the performance of the SD and RSBs. These additional and nontrivial issues deserve their own thorough and dedicated study, and likely require returning to prelaunch analyses. It is important to point out that if the on-orbit calibration methodology is incorrect, then any radiometric evaluations of the absolute accuracy of the subsequently incorrectly calibrated SDRs cannot generate correct conclusions. Thus, showing the correctness of the on-orbit RSB calibration methodology and its result is in fact a necessary step before evaluating the overall accuracy of the SDRs in the postlaunch phase. One such issue of absolute accuracy is the difference in solar spectra input of each sensor, which can cause a constant bias in the intersensor comparison. Issues of this kind may be important for absolute calibration, but they do not play a role in the on-orbit RSB calibration methodology and cannot cause any anomalous features in the SDR, as previously mentioned. Another consideration is the spectral band adjustment factor (SBAF), a constant factor adjusting for the difference between the relative spectral responses (RSRs) of the two matching bands. The SBAF values are based on the work of Doelling et al. (2012) and Scarino et al. (2016), and they are directly obtainable from the NASA Langley Cloud and Radiation Group site (Scarino 2016). For this paper these SBAF values are presented as is for the purpose of illustration without any adjustment to the intersensor comparison result. In reality, the full impact of the spectral coverage difference between Aqua MODIS and SNPP VIIRS has not been well studied, and it is not a trivial task, given current knowledge, to isolate its impact among the seasonal patterns. This analysis overcomes this shortcoming by examining the multiyear trend to identify the long-term or deviating features that are due to the on-orbit calibration methodology not coming from difference in spectral coverage.

3. Data analysis

Issues ranging from calibration inadequacy to statistical anomalies can create unusual or unwanted time series features in addition to expected short-term yearly patterns. The precise nadir match, the short time constraint, and the homogeneity threshold are not sufficient to filter out all biases or anomalies not relating to the error from the standard RSB calibration result. A simple example is the potential RVS effect in the Aqua MODIS scan mirror that may not be properly characterized. Each pixel representing a homogeneous scene can still pass the homogeneity threshold while a systematic buildup of inaccuracy caused by RVS exists along scan. Such effects are not desirable for this analysis, and careful examinations are carried out to ensure these effects can be minimized. The standard expectation is that the statistics should improve with a larger area size and sample population but variability or bias, such as RVS, can also be enhanced to counter the improvement. Nevertheless, a full exhaustive study of variability or bias is neither necessary nor of general use outside of this particular pairing of Aqua MODIS and SNPP VIIRS. A minimalistic and practicable strategy is simply to quantify the overall effect and to utilize the result to guide a balanced choice of selection requirements to achieve the needed precision to overcome the bias or the variability, if at all possible. The two moderate-resolution sensors are not yet at a performance level powerful enough to permit simpler statistical analyses to readily pass through, and hence the intercomparison analysis still requires some efforts to examine the selection criteria. To avoid side effects or any undesirable increase in complexity to the data, no additional adjustment of any kind, such as resulting from SBAF adjustment, are made other than applying cuts or statistical filters. The adopted approach is to directly examine radiance data and comparison outcomes to ascertain a data subset of sufficient statistical clarity to establish multiyear performance. The careful examinations into the data and the statistical properties for this study are extensive, and this study will present only the relevant results that support the final dataset. The key is to ensure that each selection criterion is established on a sound basis and that the final time series demonstrating a multiyear pattern are reliable on the 1% precision level. The new statistical examinations and findings are presented below.

a. Variability and bias study

The simple strategy to examine variability or bias is to generate different radiometric comparison time series corresponding to different area sizes and to characterize the differences. The result can then directly clarify which time series, along with their selection criteria, are appropriate. In Fig. 2a, the scatterplot of error bar values versus sample size is shown for outcomes corresponding to the 48-, 54-, and 80-km spatial scales for Aqua MODIS band 8 versus SNPP VIIRS M1. The outcome of the 160-km spatial scale is also shown for contrast. The IDPS-based comparison time series are used for illustration, but the OC-based time series will demonstrate the same patterns. It is seen that the larger spatial scales, as expected, can generate much larger samples. For example, the 80-km square comparison areas can generate as much as 6400 pixel-to-pixel matching samples for some SNO events. One remarkable and surprising commonality for all three spatial scales to note is the very similar scatter pattern of having only higher-valued error bars at lower sample size but having many more lower-valued error bars at larger sample size. Although not shown, the scatter pattern for all other MODIS-to-VIIRS matching bands is practically identical. This pattern of error bar versus sample size is thoroughly inconsistent with a normal univariate or any normal multivariate distribution, which will exhibit convergence and a stabilizing pattern with increasing sample size. The result reveals this to be absolutely not the case, and it is clear that the 80-km spatial scale, with samples per each SNO event up to 5 times more than the 36-km scale, does not achieve better precision, and in fact it shows nearly the same pattern. It is immediately obvious that the homogeneity filter, while certainly having cleansed out many pixels that are noisy on the pixel level, does not filter out all inhomogeneity, for example, an overhanging cloud system that can appear homogeneous over many pixels. The scatter pattern, along with its band independence, indicates the existence of a sporadic scene-based variation that is not consistent with an instrument effect and that the homogeneity filter fails to remove. Within the result of each spatial scale, as the sample size increases, the sheer number of cleaner homogeneous pixels not affected by the sporadic variation begins to take over the statistics to generate more precise error bars. This behavior is true for all three spatial scales shown, that the more precise error bars begin to dominate when the sample size, empirically, reaches about 70% of the maximum number of possible collocated pixels. A sporadic source, like moving clouds or aerosols that spread over a region, can impact the statistics in precisely this fashion. At the highest end of the sample size for each of the three results, the error bars scatter across the entire range from about 0.003 to 0.045 but with more concentration toward the lower end of the error bar range. This is consistent with the chancy outcome that at large samples some SNO events can certainly be affected by the sporadic source to end up with lower precision, but many more SNO events will be dominated by the sheer number of available cleaner homogeneous pixels, resulting in higher precision. The fact that error bars bottom out at about 0.003 is both interesting and pertinent, and it clearly marks the best possible precision when captured by clean collocated pixels representative of very homogeneous subscenes. This much smaller underlying variability at the 0.3% level indeed is fully consistent with the combined detector noise of Aqua MODIS and SNPP VIIRS detectors, with the detector noise level in each sensor at 0.1%–0.2% of the response (Angal et al. 2015; NOAA/STAR VIIRS SDR Team 2013). This finding is very encouraging, supporting the expectation that the two instruments are performing well and that the comparison methodology is accurate. Overall, it is rather evident that two separate distributions of variability are at play: a scene-based effect with a larger variability at about 2% and the effect of the combined instrument noise that is an order of magnitude weaker. Most of the error bars seen in Fig. 2a, between 0.003 and 0.02 (~0.3% to ~2%), then reflect the various levels of the impact of the sporadic scene-based variability on the radiometry comparison as a result of scene changes over short time differences. It can also be seen that the more precise error bar values congregate between 0.003 and 0.02, and that above 0.02 the number of outcomes steeply drops off. This interesting statistical behavior provides the needed clarity that an error bar threshold of 0.02 can be safely applied to remove the less precise outcomes while still keeping a sufficient number of outcomes that are better than 2% precision.

Fig. 2.
Fig. 2.

Error bar vs sample size for Aqua MODIS band 8 vs SNPP VIIRS M1 intercomparison: (a) the scatterplot of the comparison results for the four spatial scales ranging from 36 to 160 km, (b) the scatterplot for the same four spatial scales with extended sample size display range, (c) the scatterplot with two results from the two smaller spatial scales rescaled in sample size to match the 80-km square region result, and (d) results with a threshold of 0.02 in error bar and 70% of the maximum sample size applied. Each point represents one SNO event.

Citation: Journal of Atmospheric and Oceanic Technology 35, 2; 10.1175/JTECH-D-17-0008.1

The outcome for the 160-km spatial scale is shown in Figs. 2a and 2b in dark-red diamond symbols. At this wider scale, the error bar values remain above 0.01, or about 1%, for all sample sizes, marking the presence of a pervasive underlying bias preventing error bar values to fall below 1%. Many error bar values are above 4.5% and are not shown. The comparison ratios have also been checked to consistently exhibit slightly higher values, a 0.5%–1.5% upward bias, in comparison to those derived from the smaller spatial scales, thus consistent with the presence of an underlying bias of about 1%. The cause of the larger-scale bias is unknown, but no further examination on this aspect will be pursued. The straightforward but important conclusion is to keep the comparison areas for this intercomparison study to be under the 80-km scale. This study will settle on a 36 km × 36 km comparison area size.

The sample size selection threshold is a particularly tricky one among the additional selection criteria to be examined. One may be tempted to simply apply a hard cutoff without prejudice. This practice is not necessarily illegitimate, since the final statistics can still be made sufficiently robust. However, the more correct and proper approach consistent with the conditions present in this intersensor comparison is to apply the cut on a relative scale, such as keeping all sample size outcomes that are above 70% of the maximum possible sample size. This is because the outcomes of the three spatial scales, as shown in Fig. 2a, demonstrate a scaling phenomenon such that each set of outcomes is a scaled up or down version of the others. Simply stated, at the spatial scale of 80 km or less, a larger area is an extended version of its smaller subareas with the same scene properties and variability. Because the sporadic variability is ubiquitously present throughout this spatial scale, the statistics are significantly affected by this variability until a greater percentage of cleaner pixels can take over the sample population. This is another argument against the distribution being a simple normal univariate distribution, in which case the larger areas certainly would show improved statistics, including tighter error bars, and hence would not exhibit the scaling phenomenon. An explicit demonstration of the scaling phenomenon is shown in Fig. 2c, where the sample size of the 36- and 54-km scale results is rescaled by the factors of 5.10 and 2.28, which are the two ratios of the 80-km square area to the two respective smaller areas. The two rescaled results practically match that of the 80-km scale, thus proving the scaling effect and that the final statistics is neither improved nor determined by the sample size. The application of the sample size selection criterion therefore for evaluating the quality of the final time series and for selecting the optimal comparison area size should be made on a relative or scaling basis, such as applying a threshold of 70% of the maximum sample size. Because the sample size for different time series is different, the sample size threshold needs to be individually applied. A hard cut with a fixed sample size threshold, such as a 300 threshold for all, is not necessarily erroneous, but it will result in an inconsistent statistical quality across different time series, which will make evaluation more challenging.

Results from other different spatial scales and even different shapes have also been observed to exhibit the same scatter pattern, confirming further that the variability and the bias have no directional or viewing-angle dependence within these spatial scales. At the 80-km spatial scale and below, the RVS effect in either the sensor or scene BRDF is decisively eliminated as influencing factors. Removing these two well-known primary effects as possible factors, achieved by the nadir-only approach through the use of a small comparison area size, is important for this study in addressing RSB calibration issues.

The final determination for the two selection criteria is an error bar threshold of 0.02 and a sample size threshold at 70% of the maximum sample. The two applied cuts will remove the less precise outcomes and keep more of those cleaner comparison ratios less affected by the sporadic variability. It is equally important that the two cuts still keep sufficient time series outcomes, and this reflects the fact that this intercomparison generates enough good-quality SNO events. The cleaned time series outcomes are shown in the Fig. 2d scatterplot. All three cleaned-up time series have a comparable number of successful outcomes and achieve an average precision better than 1%. It may still seem surprising that all three spatial scales give practically the same statistics, countering the lingering expectation that at least the larger area should generate more successful outcomes. It is ultimately the case that the sporadic scene-based variability spreads across these spatial scales and affects the statistics in an identical fashion, as discussed above. Starting at around the 36-km scale to about the 80-km scale, the underlying bias is not a factor and clean pixels less contaminated by the larger variability can provide a clean radiometry comparison. Too far beyond the 80-km scale the bias takes over, while too far below the 36-km scale the pixel samples are simply lacking. Further examination does suggest that the 36-km scale provides marginally better statistics and that the 80-km scale may already be infected by a bias at the 0.3% level. It can also be seen in the result of the 80-km spatial scale that the lower bound of its error bars is slightly higher than that in the 36-km spatial scale result, and this indicates that a very slight bias has already crept into the 80-km comparison result. In the early report, the underlying bias in the preliminary result using the 80-km spatial scale was estimated at ~0.5% (Chu et al. 2016). Given the selection criteria and the outcomes for this study, the small bias present at these spatial scales is not a concern for the use of the 36-km comparison areas.

b. Discrepancy artifact at low radiance

The filtered time series encounter an additional complication at low radiances. In Fig. 3a, a common radiance-to-radiance comparison plot is shown for Aqua MODIS band 4 (555 nm) versus SNPP VIIRS M4 (551 nm) for IDPS SDRs. The lowest 20% of the radiances are shown in green diamond symbols and the higher radiances are shown in red cross symbols. The visually perfect linear relationship between the responses of the two sensors easily leads to a quick all-is-well conclusion. This radiance-to-radiance examination is in fact the most standard practice in radiometry comparison. In Fig. 3b, the absolute radiance difference, the VIIRS radiance minus the Aqua MODIS radiance, is shown, and the appearance again shows a very reasonable converging pattern at low radiances. However, in Fig. 3c, the relative radiance difference, defined by the difference between the VIIRS and the MODIS radiance as a percentage relative to the VIIRS radiance, distinctively shows the large relative discrepancy between MODIS and VIIRS at the lower end of the dynamic range. The lowest 20% of the radiances are again marked by the green diamond symbols. It can be seen that despite minute absolute difference in radiance, the relative difference at low radiances can be significantly amplified. For example, in the units W m−2 μm−1 sr−1, a difference of 5 out of 250 amounts to 2%, but even a minute difference of 0.2 at a low radiance of 5 results in a much larger discrepancy percentagewise at 4%. At low signals, a minute bias can generate a large relative deviation, and small noise can generate large scatter on a relative scale. The pattern in Fig. 3c clearly exhibits both a larger deviation as a result of bias and larger scatter as a result of noise. This is not surprising given the low signal-to-noise ratio (SNR) at low signals, but a standard intercomparison makes no distinction between the different radiometry levels and therefore the result can contain many deviating ratios caused by low-signal differences. Figure 3d shows the actual time series result along with the many offending outliers, in green diamond symbols, as a result of the low-radiance discrepancy. Comparison results from other RSBs also display the same low-radiance deviations. Because these offending cases at low signals are all sensitively affected by the minute bias or noise, these results should be excluded altogether. However, an indiscriminate application of statistical filters, such as a two-standard-deviation cut away from the time series mean, constructed to remove outliers, can misguidedly remove other legitimate features without first identifying any systematic problems. Additionally, the remaining low-signal outcomes still contain the larger bias to affect the final statistics. The consistent approach is to remove the lowest 20% of the samples in radiance.

Fig. 3.
Fig. 3.

Four different radiance comparison charts for Aqua MODIS band 4 vs SNPP VIIRS M4, using IDPS SDRs, with the lower 20% of radiances shown (green diamond symbols): (a) Aqua MODIS radiance vs SNPP VIIRS radiance, (b) absolute radiance difference of SNPP VIIRS and Aqua MODIS with respect to SNPP VIIRS radiance, (c) relative radiance difference (%) with respect to SNPP VIIRS radiance, and (d) time series displaying the outlying features caused by low-radiance discrepancy.

Citation: Journal of Atmospheric and Oceanic Technology 35, 2; 10.1175/JTECH-D-17-0008.1

c. Other issues

The examination of the remaining miscellaneous issues finds them to be of negligible or no impact. One example is the occasional saturation of some pixels or related issues near the high end of the dynamic range that can generate rare outliers. A simple and a consistent approach is to apply a cut to remove all radiances within 10% of the saturation limit to ensure no hidden issue remains near the high end of the dynamic range. In practice, unlike the previously discussed low-radiance cut, this cut does not have a statistically significant impact on the final time series. The examination of the distribution of data over the dynamic range also reveals no need for any fix. The intercomparison between Aqua MODIS and SNPP VIIRS occurs exclusively over the two polar regions, and the expectation may lean toward having more high radiances because of highly reflective icy surfaces. The pattern of radiances shown in Fig. 3 demonstrates this not to be the case—low radiances are actually more frequent. The distribution of radiances in all other bands is also well spread over the dynamic range. Furthermore, there are no unusual missing gaps or unbalanced representation other than the fact that the radiances may not fill the high end of the dynamic range. The final time series effectively shows no more dependence on various statistical or physical thresholds, such as population sample, time difference, etc., and contains strong statistics deemed proper for comparison study. It will later be seen that the IDPS-based comparison time series shows a large comparison discrepancy during the early mission but that is a legitimate result in the time series independent of any known numerical artifacts. A systemwide procedure or a pervasive algorithm issue affecting all IDPS SDRs as a whole is likely the cause, and a recommendation not to use the IDPS SDRs before 10 February 2012 indeed has been issued (Cao et al. 2014).

The impact of the solar zenith angle (SZA) is also quickly examined. Although the SZA correction can add to sensor radiance a strong and distinctive seasonal pattern, for a well-controlled SNO analysis this effect is expected to be minimized. This result is confirmed by a quick examination of the SZA correction factor, the ratio of the cosine of SZA, for Aqua MODID and SNPP VIIRS. Because this SNO analysis adopts a nadir-only restriction by employing only a small area size, all pixels within the small area have nearly identical SZA—in fact, for all viewing parameters as well—leading to the cancellation of the SZA correction factor in the comparison ratio of the sensor radiance data; that is, the ratio of the SZA corrections effect is long-term stable at 1 with negligible variation, and it does not generate those discernable seasonal patterns that appear in the comparison time series to be shown. The SZA correction factor is a viewing geometry effect identical for all band pairs and necessarily affects all band pairs identically. Different seasonal patterns in the time series of different matching band pairs, or the lack of it in the result of even just one band pair, effectively eliminate SZA as an influencing factor. The elimination of the impact of the SZA is another strength in this precise nadir-only SNO approach.

4. Results and discussion

Figures 4 and 5 show the comparison time series for SNPP VIIRS RSBs M1–M8 against the corresponding Aqua MODIS bands. The IDPS-based time series are shown in red cross symbols and the OC-based time series are shown in blue square symbols. The time series mean for OC-based time series is represented by the solid light-blue line, and the 2% marks above and below the time series mean are represented by the dashed blue lines. A separate set for each of the IDPS- and the OC-based time series, with a more relaxed set statistic criteria applied, is also displayed for the purpose of demonstrating a fuller trend and showing that many more SNO outcomes of lower statistical quality exist. The relaxed set corresponding to each of the IDPS- and OC-based time series is shown with magenta cross symbols and green triangle symbols, respectively. These relaxed sets have an error bar range between 0.02 and 0.03, a sample size range of 300 to 70% of the maximum size, and a radiance range of the bottom 15%–20% of radiances. However, they are not used in any analyses. There are also many more outcomes of even lower statistical quality not displayed that will fill the entire time range. The SBAF accounting for the RSR difference between the two sensors is represented by a black dotted line. As stated in section 2, this presentation will directly show the comparison time series along with the SBAF.

Fig. 4.
Fig. 4.

Comparison times series for SNPP VIIRS M1–M4 vs the matching Aqua MODIS bands. Time series mean for OC-based time series (solid light-blue line), and the 2% marks above and below the time series mean (dashed blue lines). SBAF is indicated (dotted line).

Citation: Journal of Atmospheric and Oceanic Technology 35, 2; 10.1175/JTECH-D-17-0008.1

Fig. 5.
Fig. 5.

Comparison time series for SNPP VIIRS M5–M8 vs the matching Aqua MODIS bands. Time series mean for OC-based time series (solid light-blue line), and the 2% marks above and below the time series mean (dashed blue lines). SBAF is indicated (dotted line).

Citation: Journal of Atmospheric and Oceanic Technology 35, 2; 10.1175/JTECH-D-17-0008.1

a. Comparison results for bands M1–M4

The main result of this work and the critical differences between the IDPS and the OC datasets for short-wavelength RSBs are in full display in Fig. 4. It is readily seen that the OC-based time series achieve overall superior performance; better agreement with Aqua MODIS, including no discrepancy during the early mission; and more stable trending with no relative long-term drift. For example, the comparison between Aqua MODIS band 4 and SNPP VIIRS M4 in Fig. 4f shows a clear upward drift for the IDPS-based time series (red stars) as opposed to a stable multiyear trend for the OC-based time series (blue squares). This is the key demonstration that the correct RSB calibration indeed has been achieved in the OC dataset. The lone exception is the downward drift of about 1% in the OC-based time series of Aqua MODIS band 8 versus SNPP VIIRS M1 shown in Fig. 4a, which will later be demonstrated to come from errors in Aqua MODIS band 8. The result of this IDPS-based time series very well matches the earlier comparison result from the SNO-x method (Uprety et al. 2013, 2016), which also uses the IDPS SDRs, although the more gentle temporal variation in the SNO-x time series likely comes from the SNO-x methodology being fundamentally different from the strict SNO analysis adopted for this work. Further investigation into the difference between the two intercomparison methodologies is beyond the scope of this study, but the main gist is that the pattern and the large variation are true of the comparison result between Aqua MODIS and SNPP VIIRS. Other similarities and differences are also noted. One important example is that the comparison result between Aqua MODIS band 9 and SNPP VIIRS M2 shows the largest and a very significant bias among the first four VIIRS RSBs in both studies. This analysis shows a near-4% bias in the IDPS-based trend against the SBAF, while the earlier Uprety et al. (2013, 2016) result quantifies the bias at nearly 3% using reflectance difference referencing to Aqua MODIS. However, as the overall temporal pattern of bias comprises a drift on top of a constant offset, the drift likely signals a continual change of instrument performance not properly calibrated; hence, it is the more serious issue to be addressed, as intended in this investigation.

Despite the interesting seasonal variations and perhaps various geolocational effects affecting the time series, the OC-based time series can be observed to be more stable than the IDPS-based time series. Arguably, the most important improved feature in the OC-based time series is the flat long-term trend, as the sensor data have been made accurate by the hybrid method that corrects the RSB calibration error. The is especially evident for the Aqua MODIS band 4 versus SNPP VIIRS M4 time series shown in Fig. 4f, in which the IDPS-based time series shows an upward drift of almost 2% in the past four years, whereas the OC-based time series remains flat. The worsening bias in the context of the standard RSB calibration procedure has already been analyzed, demonstrated, and discussed (Sun et al. 2012, 2014, 2016; Sun and Wang 2014, 2015a,b,c). Most notably, the recently discovered SD degradation nonuniformity effect, mentioned previously, showcases an instance of the built-in error in the standard RSB calibration methodology regardless of how well the measurements can be made (Sun et al. 2016). The OC SDRs have been corrected for the long-term drift by the application of the hybrid method mitigation (Sun and Wang 2015c, 2016) and therefore the OC-based comparison time series are flat. The drift as shown in the IDPS-based trend illustrates the consequence of any SDR versions, including also the Land Product Evaluation and Algorithm Test Element (PEATE) version (Wolf et al. 2010) used in Wu et al. (2016), to contain the worsening error if the calibration baseline using SD results is not properly mitigated.

The downward drift of about 1% in the OC-based trend of Aqua MODIS band 8 versus SNPP VIIRS M1, shown in Fig. 4a, is the unique exception deserving more attention. As the IDPS-based trend has an upward drift of ~1% relative to the OC-based trend, it can be concluded that the IDPS SDR radiance is ~1% lower relative to the OC SDR radiance. The earlier Wu et al. (2016) result, using the Land PEATE version of the VIIRS SDRs (Wolf et al. 2010), has already shown a 1% downward drift in the radiometry of SNPP VIIRS M1 using Earth targets, including desert sites and Dome Concordia in Antarctica. This earlier finding supports the conclusion that the M1 radiance from the IDPS SDRs contains the 1% downward drift but that the OC SDRs are correct. The Land PEATE SDRs, not having been mitigated for same inherent RSB calibration error as in the IDPS SDRs, necessarily will have the same long-term drift, which for M1 radiance has been shown to be a near-1% downward drift. On the other hand, the same Wu et al. (2016) study also shows that the intercomparison between Aqua MODIS band 8 and SNPP VIIRS M1 generates an effectively flat trend, and this result cannot be made consistent with the 1% downward drift observed in the earth site monitoring unless the Aqua MODIS band 8 itself also contains a similar 1% downward drift. In reality, the Wu et al. (2016) methodology may not be sufficiently strong to pin down the magnitude of the drift to be 1%, but is sufficiently consistent to establish the existence of a downward drift. On the other hand, the downward drift in the OC-based trend, shown in Fig. 4a, demonstrates that the correctly mitigated OC SDRs have captured the near-1% downward drift in Aqua MODIS band 8. An estimate of the level of the downward drift in the radiance of SNPP VIIRS M1 for both the IDPS and the Land PEATE version will be given later.

One more aspect to draw attention to is the yearly modulation pattern visible in the time series of Fig. 4 and in many subsequent plots. The yearly oscillation pattern could be induced by the imperfect match in the spectral coverage between each respective band pairs. When the RSRs of the two bands are not identically matched, the different scene types can lead to different relative sensor responses that may change with seasons. Nevertheless, other unknown physical factors remain possible causes. Here, an intersensor comparison analysis demonstrates its capability in revealing some underlying effects clearly not captured by the on-orbit calibration methodology. It may be worthy of future efforts to examine some additional physical behaviors of Aqua MODIS or SNPP VIIRS.

b. Comparison results for bands M5–M8

Aqua MODIS bands 13 and 14 are easily saturated; thus, their comparison with SNPP VIIRS M5 generates few successful ratios. Fortunately, Aqua MODIS band 1 has a much wider dynamic range to permit successful comparisons against SNPP VIIRS M5. The result in Fig. 5e demonstrates that band M5 too suffers from the similar inherent RSB calibration error that had been plaguing bands M1–M4. The IDPS-based time series exhibits a very recent upward drift of about 1%, and it can be argued that the drift started back in 2013. On the other hand, the OC-based time series, resulting from the application of the hybrid method, shows that the drift has been significantly ameliorated. The dramatic yearly variation in Aqua MODIS band 1 versus SNPP VIIRS M5 makes it more challenging for trend analysis; nevertheless, a simple evaluation shows that the OC-based time series for M5 has a drift no worse than 0.5% in the past four years. It is entirely possible that Aqua MODIS band 1 itself has contributed to some of the drift. The seasonal modulation pattern in the comparison trend is also particularly pronounced, indicating possibly that either SNPP VIIRS M5 or Aqua MODIS band 1 can be more strongly infected by the same underlying effect shown in the SNPP VIIRS M1–M4 result.

The comparison between Aqua MODIS band 15 (748 nm) and SNPP VIIRS M6 (745 nm), however, cannot escape the limitation set by the saturation of Aqua MODIS band 15, and also no alternative band exists as a valid substitute to generate a viable comparison. Nevertheless, the available result at least does not illustrate any obvious deviating trend.

The comparison between SNPP VIIRS M7 and M8 (862 and 1238 nm) demonstrates very stable trends, showing that both Aqua MODIS and SNPP VIIRS have been performing reasonably well in these channels and that they are also well calibrated. In both cases, the IDPS-based trend well matches the OC-based trend within statistics, demonstrating that the inherent RSB calibration error has weakened considerably into the near-infrared (NIR) and the shortwave infrared (SWIR) spectral range. The hybrid method mitigation is not needed for these two bands, as supported by the comparison result.

c. Comparison for bands M10 and M11

The successful comparison of SNPP VIIRS M10 (1601 nm) unfortunately cannot be achieved because of the strong limitations of Aqua MODIS band 6 (1640 nm) performance. The data of Aqua MODIS band 6 over the SNO scenes suffer from a combination of bad detectors and constant saturation. In fact, nearly all Aqua MODIS band 6 data over the SNO scenes are found to be invalid.

For Aqua MODIS band 7 (2105–2155 nm) versus SNPP VIIRS M11 (2225–2275 nm), the intersensor comparison outcomes reveal dramatic scene variations that result in highly noisy and unreliable time series unsuitable for further analyses. The combination of the mismatching RSRs, whose range is given above, and the complex polar scene conditions causes the SNO comparison ratio time series to vary wildly from 0.4 to 1.4, depending on how the different reflectance property of the common polar scenes, such as icy surfaces or clouds, affects the radiometry of the individual bands. However, the robust and reliable comparison ratio outcomes, only about 10 events or so out of the entire five years, are all consistent with ~0.4. Although not sufficiently strong to demonstrate the good radiometric performance of SNPP VIIRS M11, this result does reveal a consistency between Aqua MODIS band 7 and SNPP VIIRS M11 when SNO data are robust. While this precise nadir-centered SNO study cannot generate sufficient robust matches for Aqua MODIS band 7 versus SNPP VIIRS M11, future studies might examine different methodologies targeted for this particular pair.

It is important to point out that both the NIR and SWIR bands (~750 to ~2250 nm) are important bands for satellite ocean color remote sensing used for atmospheric correction (Gordon and Wang 1994; Wang 2007; Wang and Shi 2007), a key procedure for the ocean color data processing. Finding a viable approach, such as using a different satellite sensor, to make a robust evaluation of the performance of these key bands will be an important future task.

d. Summary statistics

Table 2 summarizes some key statistics of the comparison time series for the 4-yr interval from June 2012 to July 2016. This 4-yr choice excludes the early significant IDPS calibration inaccuracy and also ensures that seasonal effects are properly averaged over. The statistics are shown for both the IDPS- and OC-based time series side by side for comparison. These numerical results capture the essence of this investigation. It is noted here again that the relaxed sets shown in Figs. 4 and 5 are not used in the summary statistics. In addition, no statistics are shown for those comparisons, such as Aqua MODIS band 16 and SNPP VIIRS M7, which do not have sufficient robust outcomes. The trend scatter in the seventh column is a measure of the multiyear temporal variation of the time series as a whole, and it is represented by the STD computed from all ratios within each time series. It can be seen that the OC-based time series are more temporally stable for all cases. Even for SNPP VIIRS M7 and M8, which appear visually identical for both SDR sets as shown in Figs. 5d and 5f, the statistics still reveal the OC-based time series fluctuate less. The average precision in the next column to the right is computed from the error bar values of all ratios, which differs from the STD of the time series ratios just discussed above. Each individual error bar comes from the pixel-level statistics of each SNO event as explained in section 2, and it is the precision of each ratio representing each SNO event. It can be seen that all time series achieve 1% precision or better on the average, and that the average precision well represents the statistical quality of the time series. It is seen that the overall precision of the OC-based results is better, revealing that the OC SDRs used in the OC-based time series have less scatter or variability within each SNO event. Finally, the 4-yr drift in the rightmost column, except for Aqua MODIS band 8 versus SNPP VIIRS M1, shows that the OC-based time series to be long-term flat with no obvious drift. For SNPP VIIRS M2–M4, the results show that IDPS-based time series to have a drift anywhere between 0.68% and 1.66%, but that the OC-based time series have significantly ameliorated the drift to below 0.31% over the past four years and to as low as 0.01%. For SNPP VIIRS M5, despite the strong seasonal pattern, the 4-yr trend fit nevertheless reveals that the OC-based result has been reduced to 0.48% from the 0.94% of IDPS-based result. For SNPP VIIRS M7 and M8, it is pointed out again that the hybrid method’s mitigation for drift has not been applied, since no drift has yet been seen, and that the IDPS- and OC-based time series appear identical. Nevertheless, the summary statistics point toward the OC-based time series being marginally more consistent with being flat and have slightly better overall statistics.

Table 2.

Summary statistics of the RSB radiometric comparison between Aqua MODIS and SNPP VIIRS for the 4-yr period between June 2012 and July 2016. SBAF is provided next to the trend mean.

Table 2.

The magnitude of the drift in the IDPS version of the SNPP VIIRS M1 radiance can now be estimated from the difference between the 4-yr drifts in the IDPS- and OC-based time series, which are at −0.79% and −1.15%, respectively. The difference amounts to −0.36%. Although the difference displayed in Fig. 4a can be as much as ~1% at times, for example, during early 2016, the greater yearly fluctuation in the IDPS SDRs can enhance the differences at these times. The underlying difference between the IDPS and OC SDRs for M1 radiance, on the whole, is not much more than 0.4%. If Aqua MODIS band 8 has been correctly calibrated and is without any drift, then the 4-yr drift for the IDPS- and OC-based comparison time series would instead show an upward 0.4% drift and a flat trend, respectively. Those Land PEATE versions of the SNPP VIIRS SDRs used by Wu et al. (2016) are expected to contain the same drift as the IDPS SDRs. For Aqua MODIS band 8, its downward drift in radiance is quantified by the OC-based comparison result to be at 1.15%.

In comparison with the results presented in the preliminary report (Chu et al. 2016), the time series here can be seen to have tighter error bars, owing to smaller comparison areas and superior statistical treatments. The differences between IDPS- and OC-based time series are therefore made more explicit. Stronger and more reliable inferences can be drawn because of these better results.

5. General discussion

The connection between the nominal 2% variability and various earth scene phenomena can be quickly inspected. The SNO at 1225 UTC 14 January 2016, occurring over Antarctica at 44.866°W longitude and 72.896°S latitude, with an error bar of 0.38%, is expected to have occurred under near-ideal clear-day conditions. A check using the NOAA-19 Advanced TIROS Operational Vertical Sounder (ATOVS) retrieval system (NOAA Products Validation System 2016) confirms an overall highly clear Antarctica summer day—a low 10% effective cloud amount, a very dry water vapor content of 0.17 g kg−1 at 620 mb, and no evidence of other complex atmospheric conditions, such as blowing snow. The retrieval is shown in Fig. 6, and the location of the SNO orbit crossing is marked by the white arrow in each display window. The brightness temperature is also shown, and indicates a rather uniform and cold condition. This finding well supports the connection between a clear Earth scene and a precise radiometric comparison result. Several other days, with error bars between 0.75% and 2.0%, are also examined. For these days, which are infected with greater variability, ATOVS confirms the presence of more complex conditions, including high cloud content ranging from 60% to 90% and higher water vapor content typically around 2.0 g kg−1 at 620 mb. Aerosol retrievals from remote sensor data over the polar regions are not available because of difficulty in retrieval over these regions; thus, no definitive conclusion about the impact of aerosols can be drawn. In any case, for these days, the facts are consistent with the presence of contaminants that are weakly affecting the RSB data, whether they are clouds, water vapor, aerosol, blowing snow, or other possibilities.

Fig. 6.
Fig. 6.

NOAA-19 ATOVS retrieval for 1200–1300 UTC 14 Jan 2016, shown for (top left) brightness temperature, (top right) water vapor content, (bottom left) cloud-top pressure, and (bottom right) effective cloud amount. Each window shows the location of the scene of comparison for the SNO that occurred at around UTC 1225 14 Jan 2016 (white arrow).

Citation: Journal of Atmospheric and Oceanic Technology 35, 2; 10.1175/JTECH-D-17-0008.1

It is worth reiterating that intersensor comparison is an exercise in statistics with complex interplay among various influencing factors. This examination has been conducted under numerous well-controlled conditions specific to this study; therefore, many inferences presented here are not sufficiently general to be applicable to other studies. In addition to the atmospheric conditions specific to scenes associated with the orbits of these two polar-orbiting sensors, one other example of the complexity is the 4.5% homogeneity requirement that is sufficient to achieve 1% precision on the whole, but the full impact of the homogeneity threshold has not been examined. It is simply the case that the end result, with most error bar values at 2% precision or better, indicates that the 4.5% pixel-level homogeneity filter has not significantly interfered with the final overall precision. A lower homogeneity threshold, while it may initially be perceived to be more selective and can improve the result, can actually interfere and unnecessarily eliminate some clean pixels.

In addition, this study has demonstrated the success of the ~1% precision to be near the optimal result. If the two sensors have less than their current respective resolutions, such as having 5-km resolutions instead, or that the statistical treatments have not been robust, then the reliable ~1% precision result cannot be obtained. In the context of this comparison methodology, the result shown in Fig. 2d establishes that because of the scaling effect, increasing the comparison area size will not increase the number of successful time series comparison ratios or improve the final precision. For example, to achieve a tighter precision to 0.005 (~0.5%) simply requires a more stringent cut to exclude all error bars above 0.01 (~1%); consequently, more time series ratios will be eliminated, and this loss cannot be compensated by resorting to a larger area. It remains possible that some improvements in the treatment of data can tighten the precision further, or generate a more dense time series. It certainly is the case that the MODIS–VIIRS comparison has an upper limit of how many reliable and precise ratios can be generated for the comparison time series. From this perspective, the MODIS–VIIRS comparison heralds the beginning of the precision intercomparison era, at least for polar-orbiting sensors. When future major sensors with finer spatial resolutions become operable, the higher density of pixels per unit area then can permit a greater number of ratios that are also of cleaner comparison, with the obvious caveat that the comparison methodology is also properly tuned to exploit the higher capability. The 1% or better precision achieved in this study is expected to be close to the optimal precision for MODIS versus VIIRS utilizing moderate-resolutions bands. Certainly some smoothing schemes can first be applied to the pixel-level data before comparison statistics are derived, but approaches of this kind actually obfuscate the results, and hence this study does not employ such schemes.

The variability under the 80-km spatial scale and the bias at the 160-km spatial scale and beyond underpin another analytic challenge to be carefully considered. Even with close-to-ideal matching conditions, this study already encounters some significant effects and nontrivial statistics that set a limit on the best possible outcome. Considering that typical standard comparisons or vicarious calibration studies are often carried out under more relaxed conditions, including greater viewing geometry differences or greater time differences, questions on the reliability of these results can be raised. At the least, the interpretation of accuracy and precision is nontrivial. As already stated numerous times, the near-1% precision achieved in this study requires careful analysis to tune to the near-optimal result. Dependence on various conditions and the complex interplay among the resulting statistics must first be carefully examined to ensure that inferences are both reliable and meaningful. Any underlying bias, inaccuracy, or artifacts can generate skewed or erroneous results.

Overall, this radiometric comparison analysis owes its success to the statistical clarity that enables the correct build of selection filters and the eventual inferences. One particularly important and nontrivial point to elucidate is that the quantification of precision, or uncertainty per each SNO event, turns out to have some connection to scene types. As an empirical finding of this analysis, the snowy or icy scenes in the two polar regions are the dominant scenes that generate the best precision outcomes used in the final analysis. This is not surprising, since concurrent SNOs of SNPP versus Aqua occur over the polar regions as shown in Fig. 1. Various other outcomes, such as many in the relaxed sets (in Figs. 4 and 5) and others of even worse quality not displayed, do show occasional deviations that indicate the presence of different surface or scene types. These can be cloudy scenes or ocean surface. In general, different scenes or surface conditions are found to have a different level of statistical precision from the intersensor radiometric comparison. Thus, the quantification of the error bars to enable selection is a key strength of this analysis. In this context, some of data gaps, such as in mid-2014, representing the lack of the most precise outcomes but nevertheless are still with valid data, reveal some underlying unfavorable season-long weather or surface conditions. While this analysis at the current stage makes cuts without exception, future studies may examine the connection to the icy surface condition in order to make a meaningful restoration of some data points in these gaps based on an additional strong criterion or analysis.

Finally, it is here noted that the OC SDRs reported above have been reprocessed and validated only internally by the NOAA OC Team. However, the RSB calibration result, in the form of lookup tables (LUTs), achieved by the team has been adopted for the SDR reprocessing of the official institutional version (Choi et al. 2017; Sun and Wang 2017) that is to be made available to all EDRs and to all users.

6. Conclusions

A precise and “nadir only” intersensor comparison has been carried out to evaluate the multiyear calibration performance of the RSBs of Aqua MODIS against SNPP VIIRS using the SDR versions from the NOAA OC Team and the operational IDPS release. The precise SNO analysis under the stringent one-pixel nadir match along with strong selection criteria has succeeded in achieving unambiguous and precise comparison results with 1% precision or better. Effects such as BRDF of the scene and the scan-mirror RVS effect are shown not to be the influencing factors under the 80-km scale in this nadir-only approach, but a persistent and nominal 2% scene-based variability significantly impacts the statistics and the sample size selection criterion. The result of the comparison study using the 36-km spatial extent shows that the robust SDRs built by the NOAA OC Team demonstrate overall superior performance and good agreement with Aqua MODIS, such as smaller temporal variation; significantly better agreement during early mission; and, most importantly, a flat long-term trend. As the difference between the IDPS-based and OC-based time series, with all other factors being equal, comes only from the difference in on-orbit RSB calibration for SNPP VIIRS, together with strong science quality performance already established by earlier ocean color science tests, the result readily leads to the conclusion that improvements to on-orbit RSB calibration methodology made by Sun and Wang indeed have achieved robust on-orbit calibration performance. The result further demonstrates that the OC SDRs are primed to outfit SNPP VIIRS RSBs for serious science research and applications. In an interesting twist, the comparison originally intended to evaluate the radiometric performance of SNPP VIIRS has instead revealed a downward drift of 1.15% in Aqua MODIS band 8, the 412-nm channel, up to mid-2016. Finally, the stability of retrievals based on the IDPS SDRs, which have not mitigated the embedded long-term drifts highlighted in this study, maybe compromised, thus making them unsuitable for climate analysis.

Acknowledgments

This work was supported by funding from Joint Polar Satellite System (JPSS). The authors thank Michael Chalfant for providing assistance with the ATOVS system and the retrieval images, and Dr. Istvan Laszlo for the expert insights into aerosols. The authors also thank the anonymous reviewer who helped to shape this paper. The views, opinions, and findings contained in this paper are those of the authors and should not be construed as an official NOAA or U.S. government position, policy, or decision.

REFERENCES

  • Angal, A., X. J. Xiong, J. Sun, and X. Geng, 2015: On-orbit noise characterization of MODIS reflective solar bands. J. Appl. Remote Sens., 9, 094092, https://doi.org/10.1117/1.JRS.9.094092.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Bhatt, R., D. R. Doelling, A. Wu, X. Xiong, B. R. Scarino, C. O. Haney, and A. Gopalan, 2014: Initial stability assessment of S-NPP VIIRS reflective solar band calibration using invariant desert and deep convective cloud targets. Remote Sens., 6, 28092826, https://doi.org/10.3390/rs6042809.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Cao, C., and A. K. Heidinger, 2002: Inter-comparison of the longwave infrared channels of MODIS and AVHRR/NOAA-16 using simultaneous nadir observations at orbit intersections. Earth Observing Systems VII, W. L. Barnes, Ed., Society of Photo-Optical Instrumentation Engineers (SPIE Proceedings, Vol. 4814), 306–316, https://doi.org/10.1117/12.451690.

    • Crossref
    • Export Citation
  • Cao, C., M. Weinreb, and H. Xu, 2004: Predicting simultaneous nadir overpasses among polar-orbiting meteorological satellites for the intersatellite calibration of radiometers. J. Atmos. Oceanic Technol., 21, 537542, https://doi.org/10.1175/1520-0426(2004)021<0537:PSNOAP>2.0.CO;2.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Cao, C., J. Xiong, S. Blonski, Q. Liu, S. Uprety, X. Shao, Y. Bai, and F. Weng, 2013: Suomi NPP VIIRS sensor data record verification, validation, and long-term performance monitoring. J. Geophys. Res. Atmos., 118, 11 66411 678, https://doi.org/10.1002/2013JD020418.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Cao, C., F. Deluccia, X. Xiong, R. Wolfe, and F. Weng, 2014: 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., 52, 11421156, https://doi.org/10.1109/TGRS.2013.2247768.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Choi, T. J., J. Sun, B. Zhang, Z. Wang, C. Cao, F. Weng, and M. Wang, 2017: Suomi-NPP VIIRS initial reprocessing improvements and validations in the reflective solar bands (RSBS). Earth Observing Systems XXII, J. J. Butler, X. Xiong, and X. Gu Eds., Society of Photo-Optical Instrumentation Engineers (SPIE Proceedings, Vol. 10402), 104021V, https://doi.org/10.1117/12.2274081.

    • Crossref
    • Export Citation
  • Chu, M., J. Sun, and M. Wang, 2016: Radiometric evaluation of the SNPP VIIRS reflective solar band sensor data records via inter-sensor comparison with Aqua MODIS. Earth Observing Systems XXI, J. J. Butler, X. Xiong, and X. Gu Eds., Society of Photo-Optical Instrumentation Engineers (SPIE Proceedings, Vol. 9972), 99721R, https://doi.org/10.1117/12.2236942.

    • Crossref
    • Export Citation
  • Doelling, D. R., C. Lukashin, P. Minnis, B. Scarino, and D. Morstad, 2012: Spectral reflectance corrections for satellite intercalibrations using SCIAMACHY data. IEEE Geosci. Remote Sens. Lett., 9, 119123, https://doi.org/10.1109/LGRS.2011.2161751.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Gordon, H. R., and M. Wang, 1994: Retrieval of water-leaving radiance and aerosol optical thickness over the oceans with SeaWiFS: A preliminary algorithm. Appl. Opt., 33, 443452, https://doi.org/10.1364/AO.33.000443.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Heidinger, A. K., C. Cao, and J. T. Sullivan, 2002: Using Moderate Resolution Imaging Spectroradiometer (MODIS) to calibrate advanced very high resolution radiometer reflectance channels. J. Geophys. Res., 107, 4702, https://doi.org/10.1029/2001JD002035.

    • Search Google Scholar
    • Export Citation
  • NOAA National Calibration Center, 2016: SNPP SNOs with other satellites. Accessed July 2016, http://ncc.nesdis.noaa.gov/VIIRS/SNOPredictions/index.php.

  • NOAA Products Validation System, 2016: NOAA-19 sounding products. Subset used: 1200:00 UTC 14 January 2016, accessed 29 October 2016. http://www.ospo.noaa.gov/Products/atmosphere/soundings/index.html.

  • NOAA/STAR VIIRS SDR Team, 2013: Joint Polar Satellite System (JPSS) Visible Infrared Imaging Radiometer Suite (VIIRS) sensor data records (SDR) algorithm theoretical basis document (ATBD). Rev. C, NOAA Doc. E/RA-00003, 161 pp., http://ncc.nesdis.noaa.gov/documents/documentation/ATBD-VIIRS-RadiometricCal_20131212.pdf.

  • Scarino, B. R., 2016: Spectral band adjustment factor. Accessed July 2016, http://angler.larc.nasa.gov/SBAF.

  • Scarino, B. R., D. R. Doelling, P. Minnis, A. Gopalan, T. Chee, R. Bhatt, C. Lukashin, and C. Haney, 2016: A web-based tool for calculating spectral band difference adjustment factors derived from SCIAMACHY hyperspectral data. IEEE Trans. Geosci. Remote Sens., 54, 25292542, https://doi.org/10.1109/TGRS.2015.2502904.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Sun, J., and M. Wang, 2014: Visible Infrared Imaging Radiometer Suite solar diffuser calibration and its challenges using solar diffuser stability monitor. Appl. Opt., 53, 85718584, https://doi.org/10.1364/AO.53.008571.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Sun, J., and M. Wang, 2015a: On-orbit characterization of the VIIRS solar diffuser and solar diffuser screen. Appl. Opt., 54, 236252, https://doi.org/10.1364/AO.54.000236.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Sun, J., and M. Wang, 2015b: On-orbit calibration of the Visible Infrared Imaging Radiometer Suite reflective solar bands and its challenges using a solar diffuser. Appl. Opt., 54, 72107223, https://doi.org/10.1364/AO.54.007210.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Sun, J., and M. Wang, 2015c: Radiometric calibration of the Visible Infrared Imaging Radiometer Suite reflective solar bands with robust characterizations and hybrid calibration coefficients. Appl. Opt., 54, 93319342, https://doi.org/10.1364/AO.54.009331.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Sun, J., and M. Wang, 2016: VIIRS reflective solar bands calibration progress and its impact on ocean color products. Remote Sens., 8, 194, https://doi.org/10.3390/rs8030194.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Sun, J., and M. Wang, 2017: Reflective solar bands calibration improvements and look up tables for SNPP VIIRS operational mission-long SDR reprocessing. Earth Observing Systems XXII, J. J. Butler, X. Xiong, and X. Gu Eds., Society of Photo-Optical Instrumentation Engineers (SPIE Proceedings, Vol. 10402), 104021W, https://doi.org/10.1117/12.2271636.

    • Crossref
    • Export Citation
  • Sun, J., A. Angal, X. Xiong, H. Chen, X. Geng, A. Wu, T. Choi, and M. Chu, 2012: MODIS reflective solar bands calibration improvements in Collection 6. Earth Observing Missions and Sensors: Development, Implementation, and Characterization II, H. Shimoda et al., Eds., Society of Photo-Optical Instrumentation Engineers (SPIE Proceedings, Vol. 8528), 85280N, https://doi.org/10.1117/12.979733.

    • Crossref
    • Export Citation
  • Sun, J., X. Xiong, A. Angal, H. Chen, A. Wu, and X. Geng, 2014: Time-dependent response versus scan angle for MODIS reflective solar bands. IEEE Trans. Geosci. Remote Sens., 52, 31593174, https://doi.org/10.1109/TGRS.2013.2271448.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Sun, J., M. Chu, and M. Wang, 2016: Degradation nonuniformity in the solar diffuser bidirectional reflectance distribution function. Appl. Opt., 55, 60016016, https://doi.org/10.1364/AO.55.006001.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Uprety, S., and C. Cao, 2015: Suomi NPP VIIRS reflective solar band on-orbit radiometric stability and accuracy assessment using desert and Antarctica Dome C sites. Remote Sens. Environ., 166, 106115, https://doi.org/10.1016/j.rse.2015.05.021.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Uprety, S., C. Cao, X. Xiong, S. Blonski, A. Wu, and X. Shao, 2013: Radiometric intercomparison between Suomi-NPP VIIRS and Aqua MODIS reflective solar bands using simultaneous nadir overpass in the low latitudes. J. Atmos. Oceanic Technol., 30, 27202736, https://doi.org/10.1175/JTECH-D-13-00071.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Uprety, S., S. Blonski, and C. Cao, 2016: On-orbit radiometric performance characterization of S-NPP VIIRS reflective solar bands. Earth Observing Missions and Sensor: Development, Implementation, and Characterization IV, X. J. Xiong, S. A. Kuriakose, and T. Kimura, Eds., Society of Photo-Optical Instrumentation Engineers (SPIE Proceedings, Vol. 9881), 98811H, https://doi.org/10.1117/12.2223788.

    • Crossref
    • Export Citation
  • Vallado, D. A., P. Crawford, R. Hujsak, and T. S. Kelso, 2006: Revisiting Spacetrack Report #3. AIAA/AAS Astrodynamics Specialist Conf. and Exhibit, Keystone, CO, AIAA, AIAA 2006-6753, https://doi.org/10.2514/6.2006-6753.

    • Crossref
    • Export Citation
  • Wang, M., 2007: Remote sensing of the ocean contributions from ultraviolet to near-infrared using the shortwave infrared bands: Simulations. Appl. Opt., 46, 15351547, https://doi.org/10.1364/AO.46.001535.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Wang, M., 2016: Rayleigh radiance computations for satellite remote sensing: Accounting for the effect of sensor spectral response function. Opt. Express, 24, 12 41412 429, https://doi.org/10.1364/OE.24.012414.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Wang, M., and W. Shi, 2007: The NIR-SWIR combined atmospheric correction approach for MODIS ocean color data processing. Opt. Express, 15, 15 72215 733, https://doi.org/10.1364/OE.15.015722.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Wang, M., X. Liu, L. Tan, L. Jiang, S. Son, W. Shi, K. Rausch, and K. Voss, 2013: Impacts of VIIRS SDR performance on ocean color products. J. Geophys. Res. Atmos., 118, 10 34710 360, https://doi.org/10.1002/jgrd.50793.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Wang, M., and et al. , 2014: Evaluation of VIIRS ocean color products. Ocean Remote Sensing and Monitoring from Space, R. J. Frouin, D. Pan, and H. Murakami, Eds., Society of Photo-Optical Instrumentation Engineers (SPIE Proceedings, Vol. 9261), 92610E, https://doi.org/10.1117/12.2069251.

    • Crossref
    • Export Citation
  • Wang, M., W. Shi, L. Jiang, X. Liu, and K. Voss, 2015a: Technique for monitoring performance of VIIRS reflective solar bands for ocean color data processing. Opt. Express, 23, 14 44614 460, https://doi.org/10.1364/OE.23.014446.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Wang, M., and et al. , 2015b: VIIRS ocean color research and applications. 2015 IEEE International Geoscience and Remote Sensing Symposium: Proceedings, IEEE, 29112914, https://doi.org/10.1109/IGARSS.2015.7326424.

    • Crossref
    • Export Citation
  • Wolf, R. E., S. Devadiga, G. Ye, E. J. Masuoka, and R. J. Schweiss, 2010: Evaluation of the VIIRS land algorithms at Land PEATE. 2010 IEEE International and Remote Sensing Symposium: Proceedings, IEEE, 304307, https://doi.org/10.1109/IGARSS.2010.5652831.

    • Crossref
    • Export Citation
  • Wu, A., X. Xiong, C. Cao, and K. Chiang, 2016: Assessment of SNPP VIIRS VIS/NIR radiometric calibration stability using Aqua MODIS and invariant surface targets. IEEE Trans. Geosci. Remote Sens., 54, 29182924, https://doi.org/10.1109/TGRS.2015.2508379.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Xiong, X., J. Sun, X. Xie, W. Barnes, and V. Salomonson, 2010: On-orbit calibration and performance of Aqua MODIS reflective solar bands. IEEE Trans. Geosci. Remote Sens., 48, 535546, https://doi.org/10.1109/TGRS.2009.2024307.

    • Crossref
    • Search Google Scholar
    • Export Citation
Save