• Coauthors, 1997: Discriminating clear-sky from cloud with MODIS. NASA Algorithm Theoretical Basis Doc. MOD35, version 3.2, 95 pp.

  • Coauthors, 2002: Discriminating clear-sky from cloud with MODIS. MODIS Algorithm Theoretical Basis Doc. MOD35, version 4.0, 115 pp. [Available online at http://modis.gsfc.nasa.gov/data/atbd/atbd_mod06.pdf.].

  • Brennan, J. I., Kaufman Y. J. , Koren I. , and Li R. R. , 2005: Aerosol–cloud interaction—Misclassification of MODIS clouds in heavy aerosol. IEEE Trans. Geosci. Remote Sens., 43 , 911915.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Carlson, T. N., and Prospero J. M. , 1972: The large-scale movement of Saharan air outbreaks over the northern equatorial Atlantic. J. Appl. Meteor., 11 , 283297.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hogan, D., Zaccheo S. , Kennelly T. , and He Y. , 2003: Visible Infrared Imaging Radiometer Suite (VIIRS) active fire mask, active fires. Algorithm Theoretical Basis Doc. P1187-TR-I-001 version 1.0.2, Atmospheric Environmental Research, Inc., 55 pp.

  • Hutchison, K. D., and Choe N. J. , 1996: Application of 1.38-μm imagery for thin cirrus detection in daytime imagery collected over land surfaces. Int. J. Remote Sens., 17 , 33253342.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hutchison, K. D., and Jackson J. M. , 2003: Automated cloud detection over desert regions using the 412 nanometer MODIS channel. Geophys. Res. Lett., 30 .2187, doi:10.1029/2003GL018446.

    • Search Google Scholar
    • Export Citation
  • Hutchison, K. D., and Cracknell A. P. , 2005: Visible Infrared Imager Radiometer Suite: A New Operational Cloud Imager. CRC Press, 256 pp.

    • Search Google Scholar
    • Export Citation
  • Hutchison, K. D., Hardy K. R. , and Gao B-C. , 1995: Improved detection of optically thin cirrus clouds in nighttime multispectral meteorological satellite imagery using total integrated water vapor information. J. Appl. Meteor., 34 , 11611168.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hutchison, K. D., Etherton B. J. , and Topping P. C. , 1997a: Validation of automated cloud top phase algorithms: Distinguishing between cirrus clouds and snow in a priori analyses of AVHRR imagery. Opt. Eng., 36 , 17271737.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hutchison, K. D., Etherton B. J. , Topping P. C. , and Huang H. L. , 1997b: Cloud top phase determination from the fusion of signatures in daytime AVHRR imagery and HIRS data. Int. J. Remote Sens., 18 , 32453262.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hutchison, K. D., Roskovensky J. K. , Jackson J. M. , Heidinger A. K. , Kopp T. J. , Pavolonis M. J. , and Frey R. , 2005: Automated cloud detection and classification of data collected by the Visible Infrared Imager Radiometer Suite (VIIRS). Int. J. Remote Sens., 26 , 46814706.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Inoue, T., 1985: On the temperature and effective emissivity determination of semi-transparent cirrus clouds by bi-spectral measurements in the 10-μm window region. J. Meteor. Soc. Japan, 63 , 8899.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kaufman, Y. A., Tanré D. , and Boucher O. , 2002: A satellite view of aerosols in the climate system. Nature, 419 , 215223.

  • Kaufman, Y. J., Karnieli A. , and Tanré D. , 2000: Detection of dust over deserts using satellite data in the solar wavelengths. IEEE Trans. Geosci. Remote Sens., 38 , 525531.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Martins, J. V., Tanré D. , Remer L. A. , Kaufman Y. J. , Mattoo S. , and Levy R. , 2002: MODIS cloud screening for remote sensing of aerosol over oceans using spatial variability. Geophys. Res. Lett., 29 .8009, doi:10.1029/2001GL013252.

    • Search Google Scholar
    • Export Citation
  • Ou, S. C., Liou K. N. , Takano Y. , Wong E. , Hutchison K. , and Samec T. , 2005: Comparison of the University of California at Los Angeles line-by-line equivalent radiative transfer model and the moderate resolution transmission model for accuracy assessment of the National Polar-orbiting Operational Environmental Satellite System’s Visible–Infrared Imager–Radiometer Suite cloud algorithms. Appl. Opt., 44 , 62746284.

    • Search Google Scholar
    • Export Citation
  • Pavolonis, M. J., and Heidinger A. K. , 2004: Daytime cloud overlap detection from AVHRR and VIIRS. J. Appl. Meteor., 43 , 762778.

  • Pavolonis, M. J., Feltz W. F. , Heidinger A. K. , and Gallina G. M. , 2006: A daytime complement to the reverse absorption technique for improved automated detection of volcanic ash. J. Atmos. Oceanic Technol., 23 , 14221444.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Prata, A. J., 1989a: Observations of volcanic ash clouds in the 10–12-micron window using AVHRR/2 data. Int. J. Remote Sens., 10 , 751761.

  • Prata, A. J., 1989b: Infrared radiative transfer calculations for volcanic ash clouds. Geophys. Res. Lett., 16 , 12931296.

  • Prata, A. J., Bluth G. , Rose B. , Schneider D. , and Tupper A. , 2001: Comments on “Failures in detecting volcanic ash from a satellite-based technique”. Remote Sens. Environ., 78 , 341346.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Reed, B., 2002: VIIRS Cloud Mask (VCM) Algorithm Theoretical Basis Doc., version 5. SBRS Doc. Y2412, Raytheon Systems, 44 pp. [Available online at http://eic.ipo.noaa.gov/IPOarchive/SCI/atbd/mso1AEDF.pdf.].

  • Coauthors, 2005: The MODIS aerosol algorithm, products, and validation. J. Atmos. Sci., 62 , 947973.

  • Salomonson, V. V., Barnes W. L. , Maymon P. W. , Montgomery H. E. , and Ostrow H. , 1989: MODIS: Advanced facility instrument for studies of the earth as a system. IEEE Trans. Geosci. Remote Sens., 27 , 145153.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Shenk, W., and Curran R. , 1974: The detection of dust storms over land and water with satellite visible and infrared measurements. Mon. Wea. Rev., 102 , 830837.

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

    Fig. 1. Analyses of MODIS granule MODA2001.213.1210 based upon the VCM and MODIS collection 5 cloud mask algorithms. (a) An RGB (0.65, 1.6, 0.412 μm) color composite of MODIS imagery and (b) a manually generated cloud mask are shown. (c) VCM and (d) MODIS results for cloud confidence are shown along with (e), (f) heavy aerosol flags.

  • View in gallery

    Fig. 1. (Continued)

  • View in gallery

    Fig. 2. (a) The 0.412-μm/0.65-μm reflectance ratios and (b) the 0.412-μm reflectances are shown. (c) Cloud confidences after applying the dust detection test in the EVCM algorithm to MODIS granule MODA2001.213.1210. (Negative numbers reflect fill values in the M1 band)

  • View in gallery

    Fig. 3. Analyses of MODIS granule MODA2003.299.1840 based upon the VCM and MCM algorithms. (a) An RGB [0.412 μm, 1.6 μm (11–12 μm)] color composite of MODIS imagery. (b) A manually generated cloud mask. (c) VCM and (d) MODIS results for cloud confidence are shown along with (e), (f) heavy aerosol flags.

  • View in gallery

    Fig. 3. (Continued)

  • View in gallery

    Fig. 4. (a) The 2.1-μm/0.412-μm reflectance ratios and (b) the EVCM cloud phase analysis are shown. Heavy aerosol flags in the EVCM algorithm after applying the smoke detection test are shown with thresholds of (c) 0.1 and (d) 0.25 to MODIS granule MODA2003.299.1840.

  • View in gallery

    Fig. 5. (a) The VIIRS AOT product when the VCM aerosol detection tests could not be used because of the large number of false alarms produced by tests 1–3 as shown in Fig. 3e. (b) The MODIS collection 5 AOT results that are generated independently from the MODIS cloud mask and heavy aerosol flags. (c) The cloud confidence from the VIIRS aerosol module, after combining the cloud confidence and the heavy aerosol flags from the EVCM, is shown, and it suggests that the EVCM procedures may allow the VIIRS aerosol product to be created across the full range of NPOESS operational requirements.

  • View in gallery

    Fig. 6. Analyses of heavy aerosols and clouds with the EVCM algorithms for MODIS granule MODA2002.091.0240. (a) A true-color composite of MODIS imagery, (b) the automated cloud confidence, and (c) the cloud phase analyses are shown. (d) Candidate heavy aerosols detected by the smoke test are shown along with (e) dust candidates. (f) The final heavy aerosol results after the spatial test has been applied to candidate heavy aerosols.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 260 97 6
PDF Downloads 302 110 7

Distinguishing Aerosols from Clouds in Global, Multispectral Satellite Data with Automated Cloud Classification Algorithms

View More View Less
  • 1 Center for Space Research, The University of Texas at Austin, Austin, Texas
  • | 2 Northrop Grumman Space Technology, NPOESS System Engineering, Modeling & Simulations, Redondo Beach, California
  • | 3 The Aerospace Corporation, El Segundo, California
  • | 4 Northrop Grumman Space Technology, NPOESS System Engineering, Modeling & Simulations, Redondo Beach, California
Full access

Abstract

A new approach is presented to distinguish between clouds and heavy aerosols with automated cloud classification algorithms developed for the National Polar-orbiting Operational Environmental Satellite System (NPOESS) program. These new procedures exploit differences in both spectral and textural signatures between clouds and aerosols to isolate pixels originally classified as cloudy by the Visible/Infrared Imager/Radiometer Suite (VIIRS) cloud mask algorithm that in reality contains heavy aerosols. The procedures have been tested and found to accurately distinguish clouds from dust, smoke, volcanic ash, and industrial pollution over both land and ocean backgrounds in global datasets collected by NASA’s Moderate Resolution Imaging Spectroradiometer (MODIS) sensor. This new methodology relies strongly upon data collected in the 0.412-μm bandpass, where smoke has a maximum reflectance in the VIIRS bands while dust simultaneously has a minimum reflectance. The procedures benefit from the VIIRS design, which is dual gain in this band, to avoid saturation in cloudy conditions. These new procedures also exploit other information available from the VIIRS cloud mask algorithm in addition to cloud confidence, including the phase of each cloudy pixel, which is critical to identify water clouds and restrict the use of spectral tests that would misclassify ice clouds as heavy aerosols. Comparisons between results from these new procedures, automated cloud analyses from VIIRS heritage algorithms, manually generated analyses, and MODIS imagery show the effectiveness of the new procedures and suggest that it is feasible to identify and distinguish between clouds and heavy aerosols in a single cloud mask algorithm.

Corresponding author address: Keith Hutchison, Center for Space Research, The University of Texas at Austin, 3925 W. Braker Lane, Ste. 200, Austin, TX 78759. Email: keithh@csr.utexas.edu

Abstract

A new approach is presented to distinguish between clouds and heavy aerosols with automated cloud classification algorithms developed for the National Polar-orbiting Operational Environmental Satellite System (NPOESS) program. These new procedures exploit differences in both spectral and textural signatures between clouds and aerosols to isolate pixels originally classified as cloudy by the Visible/Infrared Imager/Radiometer Suite (VIIRS) cloud mask algorithm that in reality contains heavy aerosols. The procedures have been tested and found to accurately distinguish clouds from dust, smoke, volcanic ash, and industrial pollution over both land and ocean backgrounds in global datasets collected by NASA’s Moderate Resolution Imaging Spectroradiometer (MODIS) sensor. This new methodology relies strongly upon data collected in the 0.412-μm bandpass, where smoke has a maximum reflectance in the VIIRS bands while dust simultaneously has a minimum reflectance. The procedures benefit from the VIIRS design, which is dual gain in this band, to avoid saturation in cloudy conditions. These new procedures also exploit other information available from the VIIRS cloud mask algorithm in addition to cloud confidence, including the phase of each cloudy pixel, which is critical to identify water clouds and restrict the use of spectral tests that would misclassify ice clouds as heavy aerosols. Comparisons between results from these new procedures, automated cloud analyses from VIIRS heritage algorithms, manually generated analyses, and MODIS imagery show the effectiveness of the new procedures and suggest that it is feasible to identify and distinguish between clouds and heavy aerosols in a single cloud mask algorithm.

Corresponding author address: Keith Hutchison, Center for Space Research, The University of Texas at Austin, 3925 W. Braker Lane, Ste. 200, Austin, TX 78759. Email: keithh@csr.utexas.edu

1. Introduction

The National Aeronautics and Space Administration’s (NASA) Moderate Resolution Imaging Spectroradiometer (MODIS) is a heritage sensor to the Visible/Infrared Imager/Radiometer Suite (VIIRS) sensor, which first launches on the NASA-sponsored National Polar-orbiting Operational Environmental Satellite System (NPOESS) Preparatory Project (NPP) mission. MODIS collects data in 36 spectral bands (Salomonsen et al. 1989), and these data are used to generate many data products while VIIRS will collect data in 22 bands (Hutchison and Cracknell 2005). A key product created with both MODIS and VIIRS sensors is the cloud mask, which is generated in both algorithms using sophisticated logic that includes a series of cloud detection tests. Although the MODIS cloud mask (MCM) algorithm has evolved since the launch of MODIS on the Terra spacecraft in December 1999 (Ackerman et al. 1997, 2002), the VIIRS cloud mask (VCM) algorithm was based upon the earliest MCM version (Reed 2002). The VCM algorithm has been recently updated to more fully exploit the unique design features of the VIIRS sensor (Hutchison et al. 2005).

The VCM algorithm was designed to form the beginning of the processing chain for other data products. For VIIRS this chain includes seven cloud products (cloud optical thickness and effective particle size; cloud-top pressure, height, and temperature; cloud-base height; and cloud cover/layers), three ocean surface products (sea surface temperature, ocean color, and net heat flux), five land surface products (land surface temperature, normalized difference vegetation index and enhance vegetation index, albedo, soil moisture, and surface type), three snow/ice products (a subpixel snow mask, ice surface temperature, and ice age), and three atmospheric aerosol products (optical thickness, particle size parameter, and suspended matter; Hutchison and Cracknell 2005). The MCM and VCM algorithms output a pixel-level cloud confidence consisting of four possible classes, that is, confidently clear, probably clear, probably cloudy, and confidently cloudy conditions. In addition, these algorithms also generate a cloud phase analysis. The VCM cloud phase includes seven possible classes: water, mixed phase, cirrus, opaque ice, overlap (water and ice clouds in a single pixel), partly cloudy, and cloud free (Hutchison et al. 2005).

For an operational system like NPOESS, it is highly desirable to use a single cloud mask algorithm that is capable of satisfying the requirements for all applications derived from it. Therefore, the VCM algorithm must accurately differentiate between heavy aerosols and clouds in order to support the production of these surface, cloud, and atmospheric products. However, the task of distinguishing between heavy atmospheric aerosols and clouds can be daunting since the signatures of both features are often very similar in spectral data normally used in automated cloud classification algorithms (Martins et al. 2002). While the MCM algorithm includes many cloud tests and procedures to both detect clouds and distinguish them from aerosols (Ackerman et al. 2002), studies have shown that aerosols with optical depths of about 0.6 τ or greater may frequently be misclassified as clouds (Brennan et al. 2005). The inability to accurately discriminate between clouds and cloud-free obscurations in the MODIS cloud mask (MOD35) product led to multiple cloud-screening approaches being used in the MODIS program (Remer et al. 2005), with each approach having limited value to other applications (Martins et al. 2002). As a consequence, the MODIS aerosol (MOD04) product currently employs a separate cloud-clearing algorithm, which is distinct from the MOD35 cloud mask product. Thus, users that require accurate, combined cloud and aerosol climatological datasets for climate change studies may find significant differences between these products. In addition, the use of an internally generated cloud-screening algorithm in the MODIS aerosol module presents challenges for investigators who seek to tailor this algorithm for regional applications.

Therefore, a new approach has been developed to differentiate aerosols from clouds with the VCM algorithm. Section 2 describes the overall VCM and the original method of discriminating aerosols from clouds in it. Section 3 discusses new techniques developed at Northrop Grumman Space Technology (NGST) and presents results through comparisons with MODIS imagery along with manually generated and automated cloud analyses. Conclusions are drawn in section 4. [Note that throughout the text, the terms leakage, false alarms, and probability of correct typing (PCT) are used to assess VCM performance. Leakage is defined as the number of pixels classified to be cloud free in the automated cloud mask but found to contain clouds in the truth data, which is often based on manually generated cloud analyses created using well-documented procedures (Hutchison et al. 1995, 1997a, b; Hutchison and Choe 1996; Hutchison and Cracknell 2005). Thus, leakage rate is the ratio of number of confidently clear pixels in the VCM that are cloudy in the manually generated cloud mask to the total number of pixels in each surface type, for example, land or ocean. False alarms are similarly based on the number of pixels classified to be cloudy in the automated cloud mask product but are cloud free in the truth data. The PCT is based on the probably of confidently clear pixels in the automated cloud mask that are clear in the truth data plus probably or confidently cloudy pixels in the automated cloud mask that are cloudy in the truth data divided by the total number of pixels in each surface type.] The goal for the VCM design is to minimize the leakage rate, which affects land and ocean surface products; minimize the false alarm rate, which impacts all cloud and aerosol products; and maximize PCT for all surface types. System requirements for VCM PCT have been reported elsewhere (Hutchison et al. 2005).

2. Cloud–aerosol discrimination with existing algorithms

Aerosols are defined in the NPOESS program as suspensions of liquid droplets or solid particles in the atmosphere and include smoke, dust, sand, volcanic ash, and smog. The measurement range of the operational NPOESS aerosol product is 0–2 τ while the range for climate applications is 0–5 τ (see section D.15 of the NPOESS technical requirements document, available online at http://eic.ipo.noaa.gov/IPOarchive/SCI/Clean_Final_TRD_Ver7%5B1%5D.pdf). Under the operational concept that a single cloud mask algorithm satisfies the requirements for all products derived from it, it is essential that the VCM algorithm accurately discriminate between clouds and noncloudy obscurations in order to meet the requirements of the NPOESS system.

The VCM algorithm has its heritage in several cloud mask algorithms but most strongly follows the architecture of the MCM (Reed 2002). However, this original VCM algorithm has been substantially modified to more fully exploit the unique data collected by the VIIRS sensor (Hutchison et al. 2005). For example, the VCM algorithm now exploits bands not previously used in automated cloud classification algorithms such as the ocean-color bands (Hutchison and Jackson 2003; Hutchison and Cracknell 2005), which are dual gain in the VIIRS design to avoid saturation in cloudy atmospheres, as may occur with MODIS data. In addition, the VCM algorithm now 1) includes new cloud-top phase logic, based upon the procedures described by Pavolonis and Heidinger (2004); 2) incorporates a top-of-canopy (TOC) normalized difference vegetation index (NDVI) database to aid in cloud detection over diverse land surfaces; and 3) uses spatial tests with imagery resolution (375 m) bands that are nested inside radiometry resolution (750 m) bands to detect cloud edges that can impact sea surface temperature (SST) analyses, as shown in Fig. 1.4 of Hutchison and Cracknell (2005).

In the VCM and MCM architectures, additional tests are applied, after cloud confidence has been determined, in an attempt to identify the presence of heavy aerosols that have been erroneously classified as clouds. In the VCM algorithm, if at least one of the following tests detects a noncloud obscuration, the VCM bit structure is changed to reflect the presence of heavy aerosols; however, the cloud confidence was not updated. The heavy aerosol detection tests in the original VCM algorithm (Reed 2002) included the following:

  • Test 1 examined reflectances in the 2.2-μm (VIIRS M11) band and 0.650-μm (VIIRS M5) band, that is, RM11 and RM5, to detect heavy aerosols, which are transparent in the longer wavelength but more highly reflective in the shorter (Kaufman et al. 2000). If the conditions shown in Eq. (1) were valid, the heavy aerosol flag was set. The test was applied to all pixels regardless of background and cloud confidence.
    i1520-0426-25-4-501-e1
  • Test 2 applied the reverse absorption technique (Prata 1989a,b; Prata et al. 2001) to test for volcanic ash, which is normally more strongly absorbing in the shorter wavelength than the longer in contrast to most cloudy and cloud-free humid atmospheric conditions. This test classified a pixel as heavy aerosol under the conditions that the brightness temperature difference between the 11.0- and 12.0-μm bands was less than a threshold; that is, T11.0 T12.0 < −1.0. The reverse absorption test was applied to all pixels except those classified as confidently cloudy.

  • Test 3 set a flag to heavy aerosol if fire was detected in the pixel by examining the brightness temperature (BT) in the 3.7-μm (M12) band along with the 3.7–10.76-μm brightness temperature difference, that is, BTM12 − BTM15, over all land pixels. If Eq. (2) was satisfied, the heavy aerosol flag was set in the VCM; however, this test has been replaced in the VCM, which now uses the VIIRS fire mask, generated by the active fire algorithm (Hogan et al. 2003), as ancillary data.

i1520-0426-25-4-501-e2

The MODIS algorithm theoretical basis document (Ackerman et al. 2002) does not describe the heavy aerosol detection tests used in the MCM algorithm; however, results that follow will show differences between those obtained from the VCM and MCM algorithms. The MODIS results used in this study were obtained directly from the MOD35 collection 5 data products, which are available through the Earth Observing System (EOS) data gateway.

Results from the tests used to detect heavy aerosol in the original VCM and MCM collection 5 algorithms are shown in Fig. 1, which contains a false-color composite of a MODIS granule MODA2001.213.1210 (Fig. 1a). This composite was constructed by placing the 0.65-, 1.6-, and 0.412-μm MODIS bands into the red–green–blue (RGB) of a color display. (These bands correspond closely to the VIIRS M5, M10, and M1 channels, respectively.) The area shown in the scene covers the eastern Atlantic Ocean centered on a region near the Canary Islands, labeled A. An extensive area of sun glint is evident near the center of the image, just to the left of the heavy aerosol region. Optically thick ice clouds, labeled B, appear purple in the color composite because of their low reflectance in the 1.6-μm band, while water clouds are mostly white since their reflectivity is similar in all three bands. Regions with high aerosol concentrations have a greenish hue over the ocean, especially near Mauritania, between labels A and B, where the signature is similar to the desert surface, due to the relatively large reflectance in the 1.6-μm band. However, the aerosol feature takes on a purplish hue over the ocean regions near Morocco, labeled C, and the Iberian Peninsula, where the aerosol concentration decreases and the reflectance in the 1.6-μm band becomes smaller relative to that of the shorter wavelength bands. A manually generated cloud mask for this scene is shown in Fig. 1b.

Cloud confidences from the VCM and MCM algorithms are shown in Figs. 1c and 1d, respectively. Maroon represents confidently cloudy, gold is probably cloudy, light blue is probably clear, and dark blue is confidently clear. It is evident from these analyses that both algorithms classify the airborne dust over the cloud-free ocean as confidently cloudy and a summary of the cloud detection PCT for each cloud mask is shown in Table 1. The statistics show that, using the 0.412-μm band with the VCM algorithm over desert surfaces, as discussed by Hutchison and Jackson (2003), helps reduce by about 50% the number of false alarms compared to the MCM. The PCT for the VCM and MCM binary cloud masks are 84.9% and 79.4%, respectively. The large errors in these cloud masks, that is, 15.1% and 20.6%, respectively, result mainly from the misclassification of heavy aerosols as clouds. Figures 1e and 1f show, in white, the results for the VIIRS and MODIS heavy aerosol flags, respectively. Comparisons between Figs. 1c and 1e or Figs. 1d and 1f reveal the inadequacy of the heavy aerosol tests in the VCM and MCM collection 5 algorithms. Cirrus clouds are identified as heavy aerosols in the right side of each analysis and in the upper-left corner of the VCM analysis. The optically thin dust extending toward the Iberian Peninsula is classified as heavy aerosol in the VCM analysis; however, the very thick dust off the coast of Africa is not. Finally, numerous pixels toward the center of the VCM analysis are erroneously identified as heavy aerosols.

3. New approaches to cloud–aerosol discrimination

The analysis of hundreds of MODIS granules collected over a global range of conditions has shown that heavy aerosols are frequently misclassified as confidently cloudy by both the VCM and MCM algorithms, as demonstrated in the previous example. Therefore, more accurate procedures are needed to identify pixels that contain heavy aerosols in order to meet the derived requirements from the NPOESS aerosol products. The development and evaluation of these new tests is the focus of this section.

The new approach exploits the differences in spectral and textural signatures between clouds and heavy aerosols to identify “candidate” pixels that might contain heavy aerosols, including dust, smoke, volcanic ash, and industrial pollution. The term candidate is used to emphasize that these highly accurate, new spectral tests developed to detect heavy aerosols over ocean surfaces using the 0.412-μm band, also detect some cloud edges as shown in sections 3a and 3b. Therefore, the heavy aerosol candidates are analyzed using spatial tests to differentiate between heavy aerosols and water clouds. This approach works because heavy aerosols normally have a more homogeneous signature than water clouds. These spectral tests are not applied in sun-glint regions. Over land, variations in the cloud-free reflectance of different surfaces limit the value of these spectral tests; therefore, all water clouds are considered candidate heavy aerosols and are examined with the spatial tests as shown in section 3c. A brief discussion of the procedures used to detect volcanic ash is in section 3d; however, space limitations preclude a demonstration of them in this paper.

The accuracy of the new spectral tests used to detect heavy aerosols over ocean backgrounds is now presented through comparisons with the VCM and MCM algorithms using manually generated cloud analyses as truth data. For simplicity, the VCM algorithm that contains any of these new aerosol tests, and omits tests 1–3 listed in section 2, is referred to herein as the enhanced VCM (EVCM).

a. Identifying airborne sand and dust over oceans

There is a lengthy heritage in using satellite data to monitor airborne soil such as dust and sand (Carlson and Prospero 1972; Shenk and Curran 1974). Early research was directed toward the detection of these aerosols over ocean surfaces and to support estimates of aerosol optical thickness. The importance of mapping airborne sand and dust increased with improved sensor technology, based primarily from the Advanced Very High Resolution Radiometer (AVHRR), which allowed the development of improved SST analyses and global studies on desertification. Today, satellites are used to monitor aerosols on a global scale and to understand the impact of aerosols on cloud and climate feedback mechanisms (Kaufman et al. 2002). Therefore, it is highly desirable to more fully exploit VIIRS data for the global mapping of airborne dust.

The launch of the MODIS sensor in 1999 on the Terra spacecraft provided a greatly improved capability of detecting airborne dust (particles) from satellite data. Kaufman et al. (2000) demonstrated that dust had a small reflectance in the MODIS 2.1-μm band, which allowed its detection using ratios to the 0.65- and 0.47-μm bands. More recent results obtained with the 0.412-μm band on MODIS have proven valuable for cloud detection over desert regions (Hutchison and Jackson 2003). Since the VIIRS sensor will operate in dual-gain mode in this and other ocean-color bands, as previously noted, these data will be useful for automated cloud analysis algorithms. Data collected in the 0.412-μm band are used to help discriminate between heavy dust aerosols and clouds since the contrast in spectral signatures is most strong in this band, as seen in Fig. 4.8 of Hutchison and Cracknell (2005).

The new approach identifies airborne sand and dust using a reflectance ratio test between the 0.412- and 0.65-μm bands (i.e., VIIRS M1 and M5 channels) along with a maximum M1 reflectance, which is key to the successful detection of high concentrations of airborne dust. Pixels are classified as heavy aerosol candidates if the M1/M5 ratio is less than the first threshold, that is, TH1 = 2.0, while simultaneously having an M1 reflectance that does not exceed the second, that is, TH2 = 0.3. The test was evaluated with pixels classified as confidently cloud or probably cloudy in the EVCM, since no aerosol retrieval is attempted under these conditions. In the final implementation, the test is applied only to pixels classified as water clouds by the cloud phase algorithm, since results have shown that these heavy aerosols are classified exclusively as water clouds.

Thresholds for this test were established by applying principles of radiative transfer (Ou et al. 2005). Under cloudy conditions, the ratio between VIIRS M1 and M5 bands is nearly unity for ice clouds. For water clouds, this ratio is roughly proportional to the ratio of extinction coefficients in the M1 and M5 bands, that is, 2.1042/2.1431 or 0.98 for droplets with a particle radius of 6 μm. However, in cloud-free regions, the ratio of the extinction coefficients is inversely proportional to the fourth power of the wavelength ratio, that is, [(0.65/0.412)4] or ∼6.3, as seen by inspection of Fig. 2a. If aerosol particles are also present, the total (M1/M5) ratio must consider the weighted average of the air molecule, cloud, and aerosol optical depths. Since airborne dust and blowing sand may occur at altitudes relatively near the earth’s surface, that is, in the troposphere, it is expected that a significant number of air molecules are present at higher levels in the atmosphere above these noncloud obscurations. Therefore, the upper limit for the TH1 is set to 2.0. The threshold of the M1 reflectance is set to 0.3 since heavy dust typically has a much lower reflectance than clouds in this band, as shown in Fig. 2b.

The results of applying this restoral test in the EVCM to the scene contained in Fig. 1 are shown in Fig. 2c. In this case, the EVCM cloud confidence has been changed to 1) facilitate the visualization of the results even though, in the operational algorithm only the aerosol flag will be changed and not the cloud confidence and 2) understand the impact of the test on pixels classified as probably clear and probably cloudy, as shown below. Thus, Fig. 2c shows the combined results of the EVCM cloud confidences plus the heavy aerosol flags. Comparisons with Fig. 1c show the effectiveness of the heavy aerosol test. These combined results are identical to those that are used in other VIIRS algorithms, such as the aerosol optical thickness algorithm.

By changing the cloud confidences, it becomes apparent that the new restoral test identifies correctly most of the airborne sand and dust. Quantitative results are shown in Table 2 on the performance of the VCM with aerosol restoral tests 1–3 listed in section 2, followed by results with the EVCM after applying the new heavy aerosol tests to pixels with a cloud confidence of 1) confidently cloudy or probably cloudy and 2) only confidently cloudy. The test is only applied to clouds classified as water clouds by the VCM cloud phase algorithm. Applying the test to pixels classified as confidently cloudy or probably cloudy improved the PCT from 84.9% in the VCM to 88.0% in the EVCM. However, the improvement in PCT is accompanied by an increase in the leakage. Prior to applying the test, there were 41 803 pixels classified as probably cloudy and another 73 181 pixels were misclassified as probably clear. After executing the aerosol restoral test, these numbers were found to be 13 446 and 127 663, respectively. The larger number represents a 54 482 pixel and 75% increase in leakage. This increase in leakage implies that the restoral test is best used to set the heavy aerosol flag rather than change the cloud confidence from confidently cloudy to a less-cloudy confidence, for example, confidently clear. Such a change in cloud confidence could adversely impact other products, for example, by introducing cloud edges into the SST field.

Based upon the results shown in rows 1 and 2 of Table 2, the criterion for applying the heavy aerosol tests was changed to only examine only those pixels classified as confidently cloudy, and the results are shown in row 3. Using this criterion, the analyses were repeated and the final results show there were still 41 803 pixels classified as probably cloudy but only 24 935 additional pixels misclassified as probably clear by the restoral test. This means that the restoral test was able to detect the heavy aerosol, but only after increasing the leakage in the confidently clear EVCM category from 73 181 pixels to 98 116 in the population of 2 729 664 pixels that are contained in the scene, so the increase is less than 1% in the leakage rate.

A spatial test is used to further distinguish between the pixels that contain heavy aerosols and the 98 116 pixels that contain the edges of water clouds. This spatial test follows the procedure used in the MODIS aerosol algorithm (Martins et al. 2002) and significantly reduces the 98 116 pixels that are misclassified by the spectral test. In the VCM, this spatial test uses the 0.65-μm imagery (375 m) resolution band, as discussed in section 3d below, and is demonstrated through the analysis of a MODIS granule found to contain dust and industrial pollution over both land and ocean backgrounds.

b. Identifying smoke over oceans

Another spectral test was developed to detect pixels that contain smoke over ocean backgrounds and classify them as heavy aerosols rather than clouds. The approach again uses the 0.412-μm band, but this time in a ratio with the 2.1-μm MODIS band or the 2.2-μm VIIRS (M11) band. As noted by Kaufman et al. (2000), the reflectances of most aerosols are very small in the larger bandpass. On the other hand, smoke is highly reflective in the 0.412-μm band and clouds are highly reflective in both bands. Therefore, a test based upon the 2.1-μm/0.412-μm (M11/M1) reflectance ratio allows heavy aerosols to be readily distinguished from clouds.

MODIS granule MODA2003.299.1840 is used to demonstrate the value of the 2.1-μm/0.412-μm reflectance ratio for discriminating between clouds and smoke. A color composite of these data, shown in Fig. 3a, was constructed to enhance the signature of smoke along with different cloud types by putting the MODIS 0.412-μm, 1.6-μm, and the 11.0-μm − 12.0-μm brightness temperature difference (T11T12) data into the RGB of a color display. (These bands equate to the reflectances in the VIIRS M1 and M10 channels and the brightness temperature differences between the M15–M16 channels, respectively.) The data include much of coastal southern California through the Baja Peninsula and the associated eastern Pacific Ocean region. Smoke has a reddish hue, in regions labeled A, as it streams off the southern California coast in this composite since the main contribution comes from the 0.412-μm band. Water clouds, labeled B, are yellow because of their high reflectivity in both the 0.412- and 1.6-μm bands. Cirrus clouds range in color from blue (optically thin, labeled C) to pink (optically thick, labeled D) showing differences in the 1.6-μm reflectances and (T11T12) thermal contrast. Midlevel water clouds (orange) are also present in the lower-right corner of the image along with cirrus clouds.

A manually generated cloud analysis is contained in Fig. 3b, where white again denotes clouds and black represents cloud-free pixels. In this manually generated cloud analysis, water clouds that underlie the smoke have been classified as clouds, and this was a conscious decision since ocean surface products, for example, SST analyses, are generated when aerosols are present but not for pixels that contain clouds. Therefore, it is desirable that the cloud mask classify as confidently cloudy those pixels that contain both clouds and heavy aerosols to ensure minimal impact upon the SST product, which is one of the two highest-priority NPOESS products generated from VIIRS data.

Cloud confidence analyses from the VCM and MCM algorithms are shown in Figs. 3c and 3d, respectively, and Table 3 summarizes the performance of both cloud masks against the manually generated cloud analysis. Figures 3c,d show cloud confidences that are very similar for both cloud masks; that is, the PCT is 88.7% and 88.8% for the VCM and MCM, respectively. However, results from the heavy aerosol flags, seen in Figs. 3e,f, continue to show larger differences. The aerosol flag in the VCM includes many pixels that contain both ice clouds and water clouds along with the smoke. On the other hand, the MCM aerosol flag identifies very few pixels as heavy aerosol. The MCM collection 5 algorithm does not classify as heavy aerosol any pixels that actually contain the heavy smoke.

Next, the 2.1-μm/0.412-μm (M11/M1) reflectance ratio test was used in the EVCM algorithm, while replacing VCM tests 1–3 described in section 2, to quantify the value of this test for distinguishing between smoke and cloudy features. The threshold used in the test was again based upon radiative transfer theory. The ratio between the 2.1- and 0.412-μm bands for water droplets is about the same in each band, while the reflectance of ice particles is smaller in the 2.1-μm band than the 0.412-μm band, so the ratio can be much smaller than unity. Figure 4a shows larger reflectance ratios in regions that contain water clouds, while smoke regions have much smaller reflectance ratios, for example, ∼0.1, that increase slightly toward the source of the fire, for example, ∼0.25. (Note that Fig. 4a contains only those values for pixels with a valid 0.412-μm value since this MODIS band may saturate when viewing highly reflective clouds. Optically thin cirrus clouds, seen toward the lower-right corner, have small 2.1-μm/0.412-μm ratios. Thus, any attempt to tune the threshold to distinguish between smoke and thin ice clouds with the 2.1-μm/0.412-μm ratio test would prove futile. Consequently, the 2.1-μm/0.412-μm test is only applied to pixels classified as water clouds in the VCM algorithm, as shown in Fig. 4b.)

The 2.1-μm/0.412-μm reflectance ratio restoral test was applied in the EVCM algorithm using two different thresholds of 0.1 and 0.25, and the results are shown in Figs. 4c,d. The performance of the test on pixels classified confidently cloudy or probably cloudy is shown for these thresholds in rows 2 and 3, respectively, in Table 4. Row 4 shows the results after applying the test only to pixels classified as confidently cloudy and water phase with a threshold of 0.25. First, there were 43 571 pixels classified as probably cloudy, and another 77 477 pixels were misclassified as probably clear in the EVCM prior to applying this aerosol detection test. The probability of correct typing was 88.7%. After executing the aerosol detection tests on pixels classified confidently cloudy and probably cloudy using a 2.1-μm/0.412-μm reflectance ratio threshold of 0.1, these numbers became 42 713 and 88 400, respectively. Thus, the PCT increased to 91.7% while leakage increased by 10 923 pixels. After increasing the threshold to 0.25, 31 405 pixels were classified as probably cloudy and 143 856 pixels were misclassified as probably clear, and the PCT increased to 93.9%. Finally, the heavy aerosol detection test was applied only to confidently cloudy pixels in the EVCM algorithm while leaving the threshold for the 2.1-μm/0.412-μm reflectance ratio set at 0.25 (nadir). These final results showed 131 690 pixels misclassified as probably clear with a PCT of 91.0% and most pixels containing smoke identified as heavy aerosol. This threshold was selected to provide maximum detection of the heavy aerosol features, and the threshold varies with viewing geometry to compensate for Rayleigh scattering effects on the M1 band.

Since the 2.1-μm/0.412-μm reflectance ratio test was found to accurately detect smoke but also misclassify water cloud edges, pixels detected by this spectral test are also considered heavy aerosol candidates and further tested with the spatial tests. Again, the spatial test uses the 0.65-μm imagery (375 m) resolution band to eliminate edges of water clouds from heavy aerosols. Note that if the results of the 2.1-μm/0.412-μm reflectance ratio are less than 0.1 (nadir), the pixel is classified as heavy aerosol, and this classification is not changed by results from the spatial test.

Figure 5 demonstrates the impact of different cloud masks on the aerosol optical thickness (AOT) product. Figure 5a shows the retrieved VIIRS AOT product based solely upon the VCM cloud confidence; that is, the VCM aerosol flags were not used because of the large number of false alarms produced by tests 1–3 described in section 2, as shown in Fig. 3e. These results show that AOT values can only be retrieved in the 0–0.6 τ range without the benefit of an accurate heavy aerosol flag. Figure 5b shows a much improved AOT analysis with retrievals in the ∼0–3 τ range. This analysis is the MODIS collection 5 aerosol (MOD04) product, available through the EOS data gateway, which uses the cloud-screening logic described by Martins et al. (2002), not the combination of the MODIS collection 5 cloud confidence plus heavy aerosol results shown in Figs. 3d,f. Note that pixels in the more dense smoke regions (highlighted area) are still classified as cloudy with this algorithm since no retrievals were possible. Finally, Fig. 5c shows the results extracted from the VIIRS aerosol module after the EVCM cloud confidences have been combined with the new heavy aerosol tests. It is noteworthy that the cloud confidences of the heaviest smoke have been modified and retrievals are performed where AOT values exceed 3.0. Thus, these new procedures may allow the VIIRS aerosol product to be created across the full range of NPOESS operational requirements and well into the ranged needed for climate applications.

c. Identifying heavy aerosols over land

While spectral tests can be applied to identify heavy aerosol candidates over water backgrounds; variations in cloud-free surface reflectance negate using similar tests over land backgrounds. Therefore, all water clouds over land surfaces are considered heavy aerosol candidates and are evaluated with the spatial tests alluded to in previous sections. In this section, the spatial test is described in detail and demonstrated through a complex scene that contains industrial pollution along with airborne dust in MODIS data collected over the Yellow Sea that includes much of mainland China, the Korean Peninsula, and parts of the Japanese Islands.

All candidates from spectral tests discussed in sections 3a and 3b are combined with pixels over land identified by the VCM as confidently cloudy with a phase of water. This composite of heavy aerosol candidates is then examined using a spatial test based on the 0.65-μm imagery resolution (375 m) reflectance data contained within a 2 × 2 array of moderate-resolution pixels that extends across a nominal 1.5 km × 1.5 km analysis region. The standard deviation for all candidates within this 2 × 2 pixel array is calculated using a hopping window, as opposed to a sliding window, as follows:

  • If only one candidate exists within the 2 × 2 array, it is assumed to be a cloud edge.

  • Over ocean backgrounds, if at least two candidates exist in the 2 × 2 pixel array, the standard deviation is calculated using all imagery resolution pixels within these candidates. At least 8 but as many as 16 imagery resolution pixels are used in this calculation.

    • If the standard deviation >1%, all candidates are classified as water clouds. No heavy aerosol flag is set.

    • If the standard deviation is ≤1%, all moderate-resolution pixels are defined to contain heavy aerosols.

  • Over land backgrounds, if at least two candidates exist in the 2 × 2 pixel array and the TOC NDVI > 0.3, the standard deviation is calculated.

    • If the standard deviation is >2%, all candidates are classified as water clouds. No heavy aerosol flag is set.

    • If the standard deviation is ≤2%, all moderate-resolution pixels are defined to contain heavy aerosols.

  • If the TOC NDVI ≤ 0.3, all heavy aerosol candidates are assumed to be water clouds. This part of the logic may be changed in the future.

The thresholds used to define whether candidates contain cloud edges or heavy aerosols are larger than those used in the MODIS aerosol algorithm. The algorithm used to create the MOD04 product employs a threshold of 0.25% in a 3 × 3 array of MODIS 500-m pixels collected in the 0.55-μm band (Martins et al. 2002). The threshold used in the VIIRS cloud mask are quadrupled because the VCM provides a cloud mask that accurately defines cloudy pixels, including those with subpixel clouds (Hutchison et al. 2005), while the algorithm used to generate the MOD04 product must differentiate between clouds and cloud-free features including aerosols with this spatial test. There is no requirement for the VIIRS cloud mask to identify aerosol. The requirement is only for it not to classify as cloudy those pixels that contain heavy aerosols. Since candidates composed of water clouds have much large standard deviations across the 2 × 2 array, the heavy aerosol threshold can be larger and remain highly effective as shown in the following example.

Figure 6a shows a true-color composite of NASA’s Terra MODIS data center near the Yellow Sea at 0240 UTC 1 April 2002. The granule identifier is MOD.2002.091.0240. An extensive area of industrial pollution, labeled A, is evident on the left portion of the image extending toward the lower-left corner and also into the East China Sea. At the same time, airborne dust, labeled B, extends across the northern part of the Yellow Sea into the Sea of Japan where it blankets lower-level clouds in the region.

Figure 6b contains the results from the VIIRS cloud mask algorithm and shows the industrial pollution classified as confidently cloudy (dark red) over land and into the East China Sea, along with many regions of the heavy dust. (Dark blue means confidently clear while lighter blue is probably clear. Yellow stands for probably cloudy.) The VIIRS cloud phase algorithm, shown in Fig. 6c, identifies these regions of heavy aerosols to be water cloud phase (light green); thus, they become heavy aerosol candidates when over land. [Cirrus clouds are orange, red, or brown depending upon their classification as opaque, thin, or overlap (cirrus over lower-level water clouds in a single field of view), respectively.] Figure 6d contains results from the spectral test used to identify heavy aerosol candidates that contain smoke over the ocean, and possibly water cloud edges, as discussed in section 3a. Figure 6e contains results from the spectral tests used to identify heavy aerosol candidates that contain dust over the ocean, and possibly edges of water clouds, as discussed in section 3b. All candidates are combined and the standard deviation calculated, as discussed earlier in this section, to produce the final heavy aerosol analyses shown in Fig. 6f. Notice the reduction in heavy aerosols shown in the lower-left part of Fig. 6f, after the spatial test is applied, compared to the candidates detected by the dust test in Fig. 6e, which contained many water cloud edges. A comparison between Figs. 6a and 6f shows that almost all pixels identified as confidently cloudy, but seen to contain heavy aerosols, have been correctly identified by these new procedures. Only a small region of water clouds has been misclassified, as seen in the lower-left corner of the scene.

d. Identifying volcanic ash

Detection of aerosols composed of volcanic ash in the EVCM algorithm follows the approach developed by Pavolonis et al. (2006, hereafter P06) The VCM algorithm applied the “reverse absorption” method (Prata 1989a, b), which is based upon the T11T12 feature in NOAA’s AVHRR sensor. These bands are similar to MODIS channels 31 and 32 or VIIRS channels M15 and M16, respectively. The reverse absorption technique exploits the fact that T11T12 is positive for cirrus clouds and cloud-free atmospheres (Inoue 1985). The same is true for water clouds that typically have atmospheric water vapor between the cloud top and the satellite sensor. However, volcanic ash can produce a T11T12 value that is generally negative; although this signature can be masked, for example, by high levels of atmospheric water vapor (Prata et al. 2001).

Because of the limitations of the reverse absorption method, a more robust algorithm has been developed by P06. This approach combines reflectances in the 0.65- and 3.75-μm bands with the reverse absorption temperature differences through a series of tests, referred to as tiers. Tier I and tier II tests exploit the spectral signatures that are most highly correlated with volcanic ash and were shown to produce no false detections when applied to global data that contained no volcanic ash (P06). Tier III and tier IV tests generate some false detections. Since a false detection in a restoral test is equivalent to leakage in the cloud mask algorithm, it is important that the aerosol detection test accurate identify volcanic ash while minimizing false alarms. In fact, the false alarm rate was reported to be zero when tested against global satellite data known to contain no volcanic ash (P06).

The EVCM procedures to detect volcanic ash employ the reverse absorption technique of Prata over ocean backgrounds to all pixels classified as confidently cloudy. The threshold has been set at −0.25 K to avoid false detections described in Fig. 10 of P06. Testing of scenes known to contain volcanic ash have shown this threshold identifies volcanic ash in proximity to eruptions while generates few false detections of water clouds. Over land, both tier I and tier II tests are used in the EVCM algorithm to detect volcanic ash, again on pixels classified as confidently cloud, since use of the negative absorption technique has been demonstrated to detect many land features, especially pixels that contain sand. Results on the performance of these tests are available in P06.

e. Limitations

Initial testing of the procedures described above to identify heavy aerosols using numerous MODIS granules has demonstrated only two potential difficulties. First, there exists a potential impact on these procedures because of errors in the cloud phase analysis. As noted in section 3b above, the spectral test developed to detect heavy smoke is restricted from ice clouds since these particles have a depressed 2.1-μm/0.412-μm reflectance ratio that causes them to be classified as heavy aerosol candidates. In subsequent testing of many other MODIS granules, it was observed that the VCM cloud phase algorithm occasionally classifies thin cirrus clouds as water clouds, especially in the humid tropics. When this occurs, results from the spatial test, based upon the 0.65-μm imagery band described in section 3d, is unable to distinguish these clouds from aerosols since the standard deviation may be very small (∼0.1%), that is, typically much smaller than the (1%) detection threshold used in these procedures. The cause of this error in the VIIRS cloud phase analyses has been identified and a solution to the problem appears imminent.

A second problem can arise from the implementation of the volcanic ash test. As noted above, this test can detect water cloud edges especially over regions that contain semipermanent fields of stratocumulus, which are associated with cold ocean currents. Use of the current −0.25 K shows very few of these clouds are detected as heavy aerosols. If further testing determines that the number of false detections are too large, results of the volcanic ash test might also be examined with the spatial test to remove unwanted cloud edges. However, this represents a minor modification to the procedures described herein.

4. Conclusions

It is desirable to utilize a single approach to generate automated cloud analyses to meet the system-level performance requirements and minimize latency in operational weather systems, such as NPOESS. However, it becomes necessary that the approach used in generating automated cloud masks be sufficiently robust to satisfy all the derived requirements of the remaining VIIRS data products. The NASA user community attempted to use the results generated by the MODIS cloud mask algorithm before developing alternative approaches to meet individual product quality, which led to multiple automated cloud detection approaches. The use of multiple, application-specific approaches to cloud detection is not conducive to real-time operations and should be avoided if possible for several reasons. First, algorithms tailored to individual applications lack the robustness needed for multiple applications, which increases system latency. Second, users of these data products are left with products that were created by algorithms tuned for a single application (Martins et al. 2002) and must contend with contradictions in the datasets. Therefore, it is preferable that an operational, real-time system utilize a single approach to generate cloud masks.

This research suggests that several types of heavy aerosols (e.g., smoke, dust, volcanic ash, and industrial pollution) can be accurately identified over land and ocean surfaces by an automated cloud mask algorithm that use a combination of spectral and spatial tests. The procedures rely strongly upon new data that will be available in the VIIRS, dual-gain 0.412-μm band, where the reflectance of dust or sand has a minimum and smoke a maximum. This new approach also benefits from cloud confidence and cloud-top phase analyses generated within the VIIRS cloud mask logic. As a result, tests to identify heavy aerosols can be applied selectively to various cloud phase groups, which reduces misclassifications by the heavy aerosol detection tests. The aerosol flags provide users of automated cloud mask algorithms with a superior tool to determine if a given pixel should be considered as cloud or aerosol contaminated. Therefore, users of the VIIRS cloud mask products must review both the cloud confidences and the heavy aerosol flags to fully exploit these data for their applications.

While additional testing is planned to further improve the performance of the VIIRS cloud mask algorithm for the NPOESS user community, it is becoming increasingly evident that sufficiently accurate cloud analyses can be generated from a single approach to meet the data and latency requirements of the NPOESS user community.

Acknowledgments

The views, opinions, and findings contained in this report are those of the authors and should not be construed as an official position of Northrop Grumman, NOAA, The Aerospace Corporation, or the NPOESS program.

REFERENCES

  • Coauthors, 1997: Discriminating clear-sky from cloud with MODIS. NASA Algorithm Theoretical Basis Doc. MOD35, version 3.2, 95 pp.

  • Coauthors, 2002: Discriminating clear-sky from cloud with MODIS. MODIS Algorithm Theoretical Basis Doc. MOD35, version 4.0, 115 pp. [Available online at http://modis.gsfc.nasa.gov/data/atbd/atbd_mod06.pdf.].

  • Brennan, J. I., Kaufman Y. J. , Koren I. , and Li R. R. , 2005: Aerosol–cloud interaction—Misclassification of MODIS clouds in heavy aerosol. IEEE Trans. Geosci. Remote Sens., 43 , 911915.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Carlson, T. N., and Prospero J. M. , 1972: The large-scale movement of Saharan air outbreaks over the northern equatorial Atlantic. J. Appl. Meteor., 11 , 283297.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hogan, D., Zaccheo S. , Kennelly T. , and He Y. , 2003: Visible Infrared Imaging Radiometer Suite (VIIRS) active fire mask, active fires. Algorithm Theoretical Basis Doc. P1187-TR-I-001 version 1.0.2, Atmospheric Environmental Research, Inc., 55 pp.

  • Hutchison, K. D., and Choe N. J. , 1996: Application of 1.38-μm imagery for thin cirrus detection in daytime imagery collected over land surfaces. Int. J. Remote Sens., 17 , 33253342.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hutchison, K. D., and Jackson J. M. , 2003: Automated cloud detection over desert regions using the 412 nanometer MODIS channel. Geophys. Res. Lett., 30 .2187, doi:10.1029/2003GL018446.

    • Search Google Scholar
    • Export Citation
  • Hutchison, K. D., and Cracknell A. P. , 2005: Visible Infrared Imager Radiometer Suite: A New Operational Cloud Imager. CRC Press, 256 pp.

    • Search Google Scholar
    • Export Citation
  • Hutchison, K. D., Hardy K. R. , and Gao B-C. , 1995: Improved detection of optically thin cirrus clouds in nighttime multispectral meteorological satellite imagery using total integrated water vapor information. J. Appl. Meteor., 34 , 11611168.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hutchison, K. D., Etherton B. J. , and Topping P. C. , 1997a: Validation of automated cloud top phase algorithms: Distinguishing between cirrus clouds and snow in a priori analyses of AVHRR imagery. Opt. Eng., 36 , 17271737.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hutchison, K. D., Etherton B. J. , Topping P. C. , and Huang H. L. , 1997b: Cloud top phase determination from the fusion of signatures in daytime AVHRR imagery and HIRS data. Int. J. Remote Sens., 18 , 32453262.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hutchison, K. D., Roskovensky J. K. , Jackson J. M. , Heidinger A. K. , Kopp T. J. , Pavolonis M. J. , and Frey R. , 2005: Automated cloud detection and classification of data collected by the Visible Infrared Imager Radiometer Suite (VIIRS). Int. J. Remote Sens., 26 , 46814706.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Inoue, T., 1985: On the temperature and effective emissivity determination of semi-transparent cirrus clouds by bi-spectral measurements in the 10-μm window region. J. Meteor. Soc. Japan, 63 , 8899.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kaufman, Y. A., Tanré D. , and Boucher O. , 2002: A satellite view of aerosols in the climate system. Nature, 419 , 215223.

  • Kaufman, Y. J., Karnieli A. , and Tanré D. , 2000: Detection of dust over deserts using satellite data in the solar wavelengths. IEEE Trans. Geosci. Remote Sens., 38 , 525531.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Martins, J. V., Tanré D. , Remer L. A. , Kaufman Y. J. , Mattoo S. , and Levy R. , 2002: MODIS cloud screening for remote sensing of aerosol over oceans using spatial variability. Geophys. Res. Lett., 29 .8009, doi:10.1029/2001GL013252.

    • Search Google Scholar
    • Export Citation
  • Ou, S. C., Liou K. N. , Takano Y. , Wong E. , Hutchison K. , and Samec T. , 2005: Comparison of the University of California at Los Angeles line-by-line equivalent radiative transfer model and the moderate resolution transmission model for accuracy assessment of the National Polar-orbiting Operational Environmental Satellite System’s Visible–Infrared Imager–Radiometer Suite cloud algorithms. Appl. Opt., 44 , 62746284.

    • Search Google Scholar
    • Export Citation
  • Pavolonis, M. J., and Heidinger A. K. , 2004: Daytime cloud overlap detection from AVHRR and VIIRS. J. Appl. Meteor., 43 , 762778.

  • Pavolonis, M. J., Feltz W. F. , Heidinger A. K. , and Gallina G. M. , 2006: A daytime complement to the reverse absorption technique for improved automated detection of volcanic ash. J. Atmos. Oceanic Technol., 23 , 14221444.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Prata, A. J., 1989a: Observations of volcanic ash clouds in the 10–12-micron window using AVHRR/2 data. Int. J. Remote Sens., 10 , 751761.

  • Prata, A. J., 1989b: Infrared radiative transfer calculations for volcanic ash clouds. Geophys. Res. Lett., 16 , 12931296.

  • Prata, A. J., Bluth G. , Rose B. , Schneider D. , and Tupper A. , 2001: Comments on “Failures in detecting volcanic ash from a satellite-based technique”. Remote Sens. Environ., 78 , 341346.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Reed, B., 2002: VIIRS Cloud Mask (VCM) Algorithm Theoretical Basis Doc., version 5. SBRS Doc. Y2412, Raytheon Systems, 44 pp. [Available online at http://eic.ipo.noaa.gov/IPOarchive/SCI/atbd/mso1AEDF.pdf.].

  • Coauthors, 2005: The MODIS aerosol algorithm, products, and validation. J. Atmos. Sci., 62 , 947973.

  • Salomonson, V. V., Barnes W. L. , Maymon P. W. , Montgomery H. E. , and Ostrow H. , 1989: MODIS: Advanced facility instrument for studies of the earth as a system. IEEE Trans. Geosci. Remote Sens., 27 , 145153.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Shenk, W., and Curran R. , 1974: The detection of dust storms over land and water with satellite visible and infrared measurements. Mon. Wea. Rev., 102 , 830837.

    • Crossref
    • Search Google Scholar
    • Export Citation

i1520-0426-25-4-501-f101

Fig. 1. Analyses of MODIS granule MODA2001.213.1210 based upon the VCM and MODIS collection 5 cloud mask algorithms. (a) An RGB (0.65, 1.6, 0.412 μm) color composite of MODIS imagery and (b) a manually generated cloud mask are shown. (c) VCM and (d) MODIS results for cloud confidence are shown along with (e), (f) heavy aerosol flags.

Citation: Journal of Atmospheric and Oceanic Technology 25, 4; 10.1175/2007JTECHA1004.1

i1520-0426-25-4-501-f102

Fig. 1. (Continued)

Citation: Journal of Atmospheric and Oceanic Technology 25, 4; 10.1175/2007JTECHA1004.1

i1520-0426-25-4-501-f02

Fig. 2. (a) The 0.412-μm/0.65-μm reflectance ratios and (b) the 0.412-μm reflectances are shown. (c) Cloud confidences after applying the dust detection test in the EVCM algorithm to MODIS granule MODA2001.213.1210. (Negative numbers reflect fill values in the M1 band)

Citation: Journal of Atmospheric and Oceanic Technology 25, 4; 10.1175/2007JTECHA1004.1

i1520-0426-25-4-501-f301

Fig. 3. Analyses of MODIS granule MODA2003.299.1840 based upon the VCM and MCM algorithms. (a) An RGB [0.412 μm, 1.6 μm (11–12 μm)] color composite of MODIS imagery. (b) A manually generated cloud mask. (c) VCM and (d) MODIS results for cloud confidence are shown along with (e), (f) heavy aerosol flags.

Citation: Journal of Atmospheric and Oceanic Technology 25, 4; 10.1175/2007JTECHA1004.1

i1520-0426-25-4-501-f302

Fig. 3. (Continued)

Citation: Journal of Atmospheric and Oceanic Technology 25, 4; 10.1175/2007JTECHA1004.1

i1520-0426-25-4-501-f04

Fig. 4. (a) The 2.1-μm/0.412-μm reflectance ratios and (b) the EVCM cloud phase analysis are shown. Heavy aerosol flags in the EVCM algorithm after applying the smoke detection test are shown with thresholds of (c) 0.1 and (d) 0.25 to MODIS granule MODA2003.299.1840.

Citation: Journal of Atmospheric and Oceanic Technology 25, 4; 10.1175/2007JTECHA1004.1

i1520-0426-25-4-501-f05

Fig. 5. (a) The VIIRS AOT product when the VCM aerosol detection tests could not be used because of the large number of false alarms produced by tests 1–3 as shown in Fig. 3e. (b) The MODIS collection 5 AOT results that are generated independently from the MODIS cloud mask and heavy aerosol flags. (c) The cloud confidence from the VIIRS aerosol module, after combining the cloud confidence and the heavy aerosol flags from the EVCM, is shown, and it suggests that the EVCM procedures may allow the VIIRS aerosol product to be created across the full range of NPOESS operational requirements.

Citation: Journal of Atmospheric and Oceanic Technology 25, 4; 10.1175/2007JTECHA1004.1

i1520-0426-25-4-501-f06

Fig. 6. Analyses of heavy aerosols and clouds with the EVCM algorithms for MODIS granule MODA2002.091.0240. (a) A true-color composite of MODIS imagery, (b) the automated cloud confidence, and (c) the cloud phase analyses are shown. (d) Candidate heavy aerosols detected by the smoke test are shown along with (e) dust candidates. (f) The final heavy aerosol results after the spatial test has been applied to candidate heavy aerosols.

Citation: Journal of Atmospheric and Oceanic Technology 25, 4; 10.1175/2007JTECHA1004.1

Table 1. Performance of the VIIRS and MODIS cloud mask algorithms compared to a manually generated cloud mask for MODIS granule MODA2001.213.1210.

i1520-0426-25-4-501-t01

Table 2. Performance of the VCM algorithm without and with the new heavy aerosol detection tests (EVCM) for MODIS granule MODA2001.213.1210.

i1520-0426-25-4-501-t02

Table 3. Performance of the VIIRS and MODIS cloud mask algorithms compared to a manually generated cloud mask for MODA2003.299.1840.

i1520-0426-25-4-501-t03

Table 4. Performance of the VCM algorithm without and with the new heavy aerosol detection tests (EVCM) for MODIS granule MODA2003.299.1840.

i1520-0426-25-4-501-t04
Save