• Ackerman, S. A., 1997: Remote sensing aerosols using satellite infrared observations. J. Geophys. Res., 102, 17 06917 079.

  • Ackerman, S. A., , K. Strabala, , W. Menzel, , R. Frey, , C. Moeller, , and L. Gumley, 1998: Discriminating clear sky from clouds with MODIS. J. Geophys. Res., 103, 32 14132 157.

    • Search Google Scholar
    • Export Citation
  • Ackerman, S. A., , R. E. Holz, , R. Frey, , E. W. Eloranta, , B. C. Maddux, , and M. McGill, 2008: Cloud detection with MODIS. Part II: Validation. J. Atmos. Oceanic Technol., 25, 10731086.

    • Search Google Scholar
    • Export Citation
  • Bréon, F.-M., , and M. Doutriaux-Boucher, 2005: A comparison of cloud droplet radii measured from space. IEEE Trans. Geosci. Remote Sens., 43, 17961805.

    • Search Google Scholar
    • Export Citation
  • Buriez, J. C., and Coauthors, 1997: Cloud detection and derivation of cloud properties from POLDER. Int. J. Remote Sens., 18, 27852813.

    • Search Google Scholar
    • Export Citation
  • Deschamps, P. Y., , F. M. Bréon, , M. Leroy, , A. Podaire, , A. Bricaud, , J. C. Buriez, , and G. Sèze, 1994: The POLDER mission: Instrument characteristics and scientific objectives. IEEE Trans. Geosci. Remote Sens., 32, 598615.

    • Search Google Scholar
    • Export Citation
  • Deuzé, J. L., and Coauthors, 2001: Remote sensing of aerosols over land surfaces from POLDER-ADEOS-1 polarized measurements. J. Geophys. Res., 106, 49134926.

    • Search Google Scholar
    • Export Citation
  • Frey, R. A., , S. A. Ackerman, , Y. Liu, , K. I. Strabala, , H. Zhang, , J. R. Key, , and X. Wang, 2008: Cloud detection with MODIS. Part I: Improvements in the MODIS cloud mask for collection 5. J. Atmos. Oceanic Technol., 25, 10571072.

    • Search Google Scholar
    • Export Citation
  • Gao, B.-C., , A. F. H. Goetz, , and W. J. Wiscombe, 1993: Cirrus cloud detection from airborne imaging spectrometer data using the 1.38 micron water vapor band. Geophys. Res. Lett., 20, 301304.

    • Search Google Scholar
    • Export Citation
  • Goloub, P., , H. Chepfer, , M. Herman, , G. Brogniez, , and F. Parol, 1997: Use of polarization for cloud studies. Polarization: Measurement, Analysis, and Remote Sensing, D. H. Goldstein and R. A. Chipman, Eds., International Society for Optical Engineering, (SPIE Proceedings, Vol. 3121), 330–341.

    • Search Google Scholar
    • Export Citation
  • Goloub, P., , M. Herman, , H. Chepfer, , J. Riedi, , G. Brogniez, , P. Couvert, , and G. Sèze, 2000: Cloud thermodynamic phase classification from the POLDER spaceborne instrument. J. Geophys. Res., 105, 14 74714 759.

    • Search Google Scholar
    • Export Citation
  • Hubanks, P., , M. King, , S. Platnick, , and R. Pincus, 2008: MODIS atmosphere L3 gridded product algorithm theoretical basis document. MODIS Doc. ATBD-MOD-30, 96 pp.

    • Search Google Scholar
    • Export Citation
  • Inoue, T., 1987: A cloud type classification with NOAA 7 split-window measurements. J. Geophys. Res., 92, 39914000.

  • Kaufman, Y. J., , C. Ichoku, , L. Giglio, , S. Korontzi, , D. A. Chu, , W. M. Hao, , R.-R. Li, , and C. O. Justice, 2003: Fire and smoke observed from the earth observing system MODIS instrument—Products, validation, and operational use. Int. J.Remote Sens., 24, 17651781.

    • Search Google Scholar
    • Export Citation
  • King, M., , Y. Kaufman, , W. Menzel, , D. Tanré, , N. Center, , and M. Greenbelt, 1992: Remote sensing of cloud, aerosol, and water vapor properties from the moderate resolution imaging spectrometer (MODIS). IEEE Trans. Geosci. Remote Sens., 30, 227.

    • Search Google Scholar
    • Export Citation
  • Leroy, M., and Coauthors, 1997: Retrieval of atmospheric properties and surface bidirectional reflectances over the land from POLDER. J. Geophys. Res., 102, 17 02317 037.

    • Search Google Scholar
    • Export Citation
  • Minnis, P., 1989: Viewing zenith angle dependence of cloudiness determined from coincident GOES East and GOES West data. J. Geophys. Res., 94, 23032320.

    • Search Google Scholar
    • Export Citation
  • Parol, F., , J. Buriez, , C. Vanbauce, , P. Couvert, , G. Seze, , P. Goloub, , and S. Cheinet, 1999: First results of the POLDER “earth radiation budget and clouds” operational algorithm. IEEE Trans. Geosci. Remote Sens., 37, 15971612.

    • Search Google Scholar
    • Export Citation
  • Parol, F., and Coauthors, 2004: Review of capabilities of multi-angle and polarization cloud measurements from POLDER. Adv. Space Res., 33, 10801088.

    • Search Google Scholar
    • Export Citation
  • Parol, F., , C. Vanbauce, , J. Riedi, , F. Thieuleux, , Z. Poussi, , and A. Lifermann, 2007: Comparison and statistical analysis of cloud properties derived from POLDER and MODIS. EUMETSAT Meteorological Satellite Conf. and the 15th AMS Satellite Meteorology and Oceanography Conf., Amsterdam, Netherlands. [Available from http://www.eumetsat.int/home/Main/AboutEUMETSAT/Publications/ConferenceandWorkshopProceedings/2007/groups/cps/documents/document/­pdf_conf_p50_s8_11_parol_p.pdf.]

    • Search Google Scholar
    • Export Citation
  • Platnick, S., , M. King, , S. Ackerman, , W. Menzel, , B. Baum, , J. Riedi, , and R. Frey, 2003: The MODIS cloud products: Algorithms and examples from Terra. IEEE Trans. Geosci. Remote Sens., 41, 459473.

    • Search Google Scholar
    • Export Citation
  • Riedi, J., and Coauthors, 2010: Cloud thermodynamic phase inferred from merged POLDER and MODIS data. Atmos. Chem. Phys., 10, 11 85111 865.

    • Search Google Scholar
    • Export Citation
  • Rossow, W. B., , and R. A. Schiffer, 1991: ISCCP cloud data products. Bull. Amer. Meteor. Soc., 72, 220.

  • Rossow, W. B., , and R. A. Schiffer, 1999: Advances in understanding clouds from ISCCP. Bull. Amer. Meteor. Soc., 80, 22612288.

  • Roy, D. P., , J. S. Borak, , S. Devadiga, , R. E. Wolfe, , M. Zheng, , and J. Descloitres, 2002: The MODIS Land product quality assessment approach. Remote Sens. Environ., 83, 6276.

    • Search Google Scholar
    • Export Citation
  • Sassen, K., , Z. Wang, , and D. Liu, 2008: Global distribution of cirrus clouds from CloudSat/Cloud–Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO) measurements. J. Geophys. Res., 113, D00A12, doi:10.1029/2008JD009972.

    • Search Google Scholar
    • Export Citation
  • Stubenrauch, C. J., , S. Cros, , A. Guignard, , and N. Lamquin, 2010: A 6-year global cloud climatology from the Atmospheric Infrared Sounder (AIRS) and a statistical analysis in synergy with CALIPSO and CloudSat. Atmos. Chem. Phys., 10, 71977214.

    • Search Google Scholar
    • Export Citation
  • Vanbauce, C., , J. C. Buriez, , F. Parol, , B. Bonnel, , G. Sèze, , and P. Couvert, 1998: Apparent pressure derived from ADEOS-POLDER observations in the oxygen A-band over ocean. Geophys. Res. Lett., 25, 31593162.

    • Search Google Scholar
    • Export Citation
  • Warren, S. G., , C. J. Hahn, , J. London, , R. M. Chervin, , and R. L. Jenne, 1986: Global distribution of total cloud cover and cloud type amounts over land. NCAR Tech. Note NCAR/TN-273+STR, 29 pp. + 200 maps.

    • Search Google Scholar
    • Export Citation
  • Warren, S. G., , C. J. Hahn, , J. London, , R. M. Chervin, , and R. L. Jenne, 1988: Global distribution of total cloud cover and cloud type amounts over the ocean. NCAR Tech. Note NCAR/TN-317+STR, 41 pp. + 170 maps.

    • Search Google Scholar
    • Export Citation
  • Wielicki, B., , and L. Parker, 1992: On the determination of cloud cover from satellite sensors: The effect of sensor spatial resolution. J. Geophys. Res., 97, 12 79912 823.

    • Search Google Scholar
    • Export Citation
  • Wylie, D. P., , W. P. Menzel, , H. M. Woolf, , and K. I. Strabala, 1994: Four years of global cirrus cloud statistics using HIRS. J. Climate, 7, 19721986.

    • Search Google Scholar
    • Export Citation
  • Zhang, Z., , P. Yang, , G. Kattawar, , J. Riedi, , L. C. Labonnote, , B. Baum, , S. Platnick, , and H.-L. Huang, 2009: Influence of ice particle model on retrieving cloud optical thickness from satellite measurements: Model comparison and implication for climate study. Atmos. Chem. Phys. Discuss., 9, 17571796.

    • Search Google Scholar
    • Export Citation
  • Zhao, G., , and L. Di Girolamo, 2004: A cloud fraction versus view angle technique for automatic in-scene evaluation of the MISR cloud mask. J. Appl. Meteor., 43, 860869.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    Latitudinal variations of cloud fraction of POLDER and MODIS for the official level-3 dataset and for the dataset created for this study. Error bars present for POLDER the minimal and maximal cloud covers (from level 3) obtained before the reclassification tests. The ISCCP cloud fraction is for the period of 1984–2007.

  • View in gallery

    Latitudinal variations of CFC cloud fraction from wide swath (MODIS swath), narrow swath (POLDER swath), and edge of the large swath (POLDER swath subtracted from MODIS swath), and the differences of cloud fraction between MODIS and POLDER swath.

  • View in gallery

    Latitudinal variations of cloud fraction from POLDER and MODIS.

  • View in gallery

    Latitudinal variations of ice and water cloud fraction from POLDER and MODIS CFC.

  • View in gallery

    Seasonal variations of cloud fraction in the subtropics area (0°–30°) over (a) ocean for the Northern Hemisphere, (b) ocean for the Southern Hemisphere, (c) land for the Northern Hemisphere, (d) land for the Southern Hemisphere; and in midlatitudes regions (30°–60°) over (e) ocean for the Northern Hemisphere; (f) ocean for the Southern Hemisphere; (g) land for the Northern Hemisphere; and (h) land for the Southern Hemisphere.

  • View in gallery

    Histograms of cloud fractions over (a) ocean, (b) land, and (c) snow-covered land; and histograms of cloud fraction differences (POLDER-MODIS CFC) over (d) ocean, (e) land, and (f) snow-covered land.

  • View in gallery

    Geographical distribution of cloud fraction for (first row) POLDER, (second row) MODIS CFC, and (third row) MODIS CFD; geographical distribution of cloud fraction differences for (fourth row) MODIS CFD minus CFC, (fifth row) MODIS CFD minus POLDER, (sixth row) POLDER minus MODIS CFC, and also for (seventh row) water cloud fraction differences (POLDER minus MODIS CFC) and (eighth row) ice cloud fraction differences (POLDER minus MODIS CFC) during the four seasons in 2008.

  • View in gallery

    Cloud fraction as a function of VZA for different solar zenith angular bins for (a) MODIS and (b) POLDER. Negative view zenith angles correspond to relative azimuth less than 90°, namely, backward scattering directions.

  • View in gallery

    Histograms of the POLDER and MODIS cloud fractions for (a) all pixels, (b) pixels in MODIS sun-glint directions, and (c) pixels out of sun-glint regions. Histograms are for data between 30° and 60°N during summer 2008.

  • View in gallery

    Geographical distribution of the active fires detected by MODIS for a period (9 Jul 2008–18 Jul 2008) in the summer 2008 from (a) NASA rapid respond Web site, (b) POLDER Quality index of cloud detection, and (c) POLDER fine mode aerosol optical thickness.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 90 90 10
PDF Downloads 67 67 4

Examination of POLDER/PARASOL and MODIS/Aqua Cloud Fractions and Properties Representativeness

View More View Less
  • 1 Laboratoire Optique d’Atmosphérique, UMR CNRS 8518, Université Lille 1, Villeneuve d’Ascq, France
© Get Permissions
Full access

Abstract

The Polarization and Anisotropy of Reflectances for Atmospheric Sciences Coupled with Observations from a Lidar (PARASOL) and Aqua are two satellites on sun-synchronous orbits in the A-Train constellation. Aboard these two platforms, the Polarization and Directionality of Earth Reflectances (POLDER) and Moderate Resolution Imaging Spectroradiometer (MODIS) provide quasi simultaneous and coincident observations of cloud properties. The similar orbits but different detecting characteristics of these two sensors call for a comparison between the derived datasets to identify and quantify potential uncertainties in retrieved cloud properties.

To focus on the differences due to different sensor spatial resolution and coverage, while minimizing sampling and weighting issues, the authors have recomputed monthly statistics directly from the respective official level-2 products. The authors have developed a joint dataset that contains both POLDER and MODIS level-2 cloud products collocated on a common sinusoidal grid. The authors have then computed and analyzed monthly statistics of cloud fractions corresponding either to the total cloud cover or to the “retrieved” cloud fraction for which cloud optical properties are derived. These simple yet crucial cloud statistics need to be clearly understood to allow further comparison work of the other cloud parameters.

From this study, it is demonstrated that on average POLDER and MODIS datasets capture correctly the main characteristics of global cloud cover and provide similar spatial distributions and temporal variations. However, each sensor has its own advantages and weaknesses in discriminating between clear and cloudy skies in particular situations. Also it is shown that significant differences exist between the MODIS total cloud fraction (day mean) and the “retrieved” cloud fraction (combined mean). This study found a global negative difference of about 10% between POLDER and MODIS day-mean cloud fraction. On the contrary, a global positive difference of about 10% exists between POLDER and MODIS combined-mean cloud fraction. These statistical biases show both global and regional distributions that can be driven by sensors characteristics, environmental factors, and also carry potential information on cloud cover structure. These results provide information on the quality of cloud cover derived from POLDER and MODIS and should be taken into account for the use of other cloud products.

Corresponding author address: Shan Zeng, Laboratoire Optique d’Atmosphérique, UMR CNRS 8518, Université Lille 1, 59655 Villeneuve d’Ascq CEDEX, France. E-mail: shan.zeng@loa.univ-lille1.fr

Abstract

The Polarization and Anisotropy of Reflectances for Atmospheric Sciences Coupled with Observations from a Lidar (PARASOL) and Aqua are two satellites on sun-synchronous orbits in the A-Train constellation. Aboard these two platforms, the Polarization and Directionality of Earth Reflectances (POLDER) and Moderate Resolution Imaging Spectroradiometer (MODIS) provide quasi simultaneous and coincident observations of cloud properties. The similar orbits but different detecting characteristics of these two sensors call for a comparison between the derived datasets to identify and quantify potential uncertainties in retrieved cloud properties.

To focus on the differences due to different sensor spatial resolution and coverage, while minimizing sampling and weighting issues, the authors have recomputed monthly statistics directly from the respective official level-2 products. The authors have developed a joint dataset that contains both POLDER and MODIS level-2 cloud products collocated on a common sinusoidal grid. The authors have then computed and analyzed monthly statistics of cloud fractions corresponding either to the total cloud cover or to the “retrieved” cloud fraction for which cloud optical properties are derived. These simple yet crucial cloud statistics need to be clearly understood to allow further comparison work of the other cloud parameters.

From this study, it is demonstrated that on average POLDER and MODIS datasets capture correctly the main characteristics of global cloud cover and provide similar spatial distributions and temporal variations. However, each sensor has its own advantages and weaknesses in discriminating between clear and cloudy skies in particular situations. Also it is shown that significant differences exist between the MODIS total cloud fraction (day mean) and the “retrieved” cloud fraction (combined mean). This study found a global negative difference of about 10% between POLDER and MODIS day-mean cloud fraction. On the contrary, a global positive difference of about 10% exists between POLDER and MODIS combined-mean cloud fraction. These statistical biases show both global and regional distributions that can be driven by sensors characteristics, environmental factors, and also carry potential information on cloud cover structure. These results provide information on the quality of cloud cover derived from POLDER and MODIS and should be taken into account for the use of other cloud products.

Corresponding author address: Shan Zeng, Laboratoire Optique d’Atmosphérique, UMR CNRS 8518, Université Lille 1, 59655 Villeneuve d’Ascq CEDEX, France. E-mail: shan.zeng@loa.univ-lille1.fr

1. Introduction

Clouds, which cover 50%–70% of the globe (Rossow and Schiffer 1991), are essential components of the atmosphere, and their properties tightly and directly affect the radiation budget and the hydrological cycle of the earth–atmosphere system. Better descriptions of the cloud’s properties are required to correctly understand the climate system and its natural or human-induced variations. Modern satellites have become an essential part in the global effort to monitor the earth–atmosphere system and to understand its various components among which clouds are recognized to be of primary importance. Projects like the First International Satellite Cloud Climatology Project (ISCCP), the Pathfinder extended dataset (PATMOS-X), and the Global Energy and Water Cycle Experiment (GEWEX), combine a series of different geostationary and polar-orbiting satellites to provide long records of cloud properties needed to understand climate changes at both regional and global scales. Accuracy and limitations of these climatological records need to be clearly established to assess potential trends in cloud cover and its associated properties. This calls for the establishment of additional and carefully characterized datasets that can serve as a reference to evaluate longer records derived from series of operational satellites.

The Polarization and Anisotropy of Reflectances for Atmospheric Sciences Coupled with Observations from a Lidar (PARASOL) and Aqua are among the five sun-synchronous satellites queued in line, forming the so-called A-Train, which is an unprecedented effort to study the global atmosphere. This group of satellites provides, within minutes of each other, thorough information about our atmosphere from a set of passive and active sensors with broad performances and application ranges. As parts of the A-Train, Aqua carrying the Moderate Resolution Imaging Spectroradiometer (MODIS) and PARASOL carrying the Polarization and Directionality of the Earth Reflectances (POLDER), fly in ascending orbit and cross the equator within 105 s of one another at around 1330 local time. Among its noticeable instrumental characteristics, MODIS provides a relatively high spatial resolution as well as a wide spectral coverage from solar, near infrared, to thermal infrared spectrum (Platnick et al. 2003) while the virtues of POLDER rely on its multipolarization, multidirectionality, and multispectral capability (Parol et al. 2004). These different characteristics make each sensor able to identify clouds in different ways. Analyses that combine two different sensors can obviously help to obtain a more accurate representation of global cloud cover properties and a better comprehension of cloud properties and their effect on the radiative budget, but they also provide a better understanding of both sensor’s characteristics and abilities to retrieve cloud properties, especially their cloud detection performances. Such recent joint analyses of MODIS and POLDER cloud datasets concerning global cloud properties (Parol et al. 2007) or more specifically cloud phase (Riedi et al. 2010) and cloud microphysical properties (Bréon and Doutriaux-Boucher 2005; Zhang et al. 2009) led to interesting results.

Among all parameters driving radiative impact of clouds, cloud fraction is obviously of primary importance. The identification of cloud-contaminated pixels is crucial since it is a first compulsory step for further retrieval of cloud properties, and any difficulty in establishing cloud detection in turn causes errors in the determination of other cloud properties’ statistics. In this paper, we are thereby focusing on the statistical comparison of this key parameter.

A brief overview of cloud detection algorithms and their theoretical basis is provided individually for both POLDER and MODIS in the second section. A third section is devoted to the description of our new merged dataset prepared for the comparison studies. Results from statistical comparison and analyses of the differences in cloud detection for different situations are presented and discussed in sections 4 and 5. We especially illustrate and discuss the differences that can be observed between the total cloud fraction and the fraction of clouds for which optical properties can be derived. Although the total cloud fraction is of primary importance, it is also critical to understand which part of the cloud cover is actually sampled when other cloud properties are retrieved. A summary of the main findings and outlook is provided in the last section.

2. Main features of cloud detection algorithms

a. POLDER

POLDER is a component of a series of sensors (Deschamps et al. 1994), developed by Centre National d’Études Spatiales (CNES), flying on board PARASOL since 2004. It is a multispectral imaging radiopolarimeter designed to provide global and repetitive observations of the solar radiation and polarized radiance reflected by the earth–atmosphere system. The instrument design consists of a wide field of view (1800 km) telecentric optics, a rotating wheel carrying spectral filters and polarizers, and a charge-coupled device (CDD) array of detectors that induces a moderate spatial resolution of about 6 km at the ground independent of the viewing angle. When it passes over a scene, POLDER acquires up to 16 successive multiangle measurements of both the total and polarized solar radiance in eight narrow bands from 443 to 1020 nm for daytime observations only.

POLDER cloud detection algorithm, a component of the “earth radiation budget (ERB) and clouds” processing line, is based on a series of separated and independent threshold tests applied to each individual pixel for every viewing direction (Buriez et al. 1997; Parol et al. 1999). Four tests aim at detecting clouds and three additional tests are applied to confirm clear pixels. Thus, POLDER cloud detection aims to identify the confident clouds pixels, which are used then to derive the cloud optical properties. The four “cloudy” tests are on the apparent pressure that is obtained from the estimate of oxygen absorption around 763 nm (Vanbauce et al. 1998), on the solar reflectance at 865 nm (490nm) over ocean (land), on the 865-nm-polarized reflectance (Goloub et al. 2000), and on the 490-nm-polarized reflectance (Goloub et al. 1997). Three additional “clear” tests with particular clear thresholds are applied to indicate a confident cloud-free surface if a pixel fails to pass the four cloudy tests above: a low reflectance test, a spectral reflectance variability test (between 865 and 443 nm), and an apparent pressure test.

If a pixel fails to pass all these seven tests, and remains unclassified, angular and spatial variability tests will be used. Afterward, when all of the elementary pixels are identified as either clear sky or cloudy, the cloud fraction is computed at superpixel scale (3 × 3 pixels), direction by direction. The final cloud fraction is then averaged over all the 16 directions.

In summary, POLDER carries forward the strengths of its multipolarization and multidirectional capability to discriminate clouds from clear sky. A confidence index is provided to qualify the cloud detection process. This index ranges from 0 to 1 and is computed based on the number of available useful viewing directions and overall number of tests indicating either cloudy or clear-sky conditions. An index equal to 1 corresponds to high confidence of detection while 0 indicates poor confidence. In the presence of aerosols, thin, broken clouds, or cloud edges, this index tends toward low values.

b. MODIS

MODIS is a 36-band scanning spectroradiometer on board Aqua, launched in May 2002 as a part of National Aeronautics and Space Administration (NASA)’s Earth Observing System (EOS) (King et al. 1992). It provides 29 spectral bands at 1-km resolution, 5 bands at 500-m resolution, and 2 bands at 250-m resolution. Its spectral coverage ranges from visible (VIS) to thermal infrared (IR) (0.415–14.235 μm). Cloud mask is based on a variety of tests using as many as 20 of these 36 spectral bands to maximize reliability of cloud detection.

The MODIS cloud mask algorithm (MOD35) (Ackerman et al. 1998; Platnick et al. 2003) consists of a series of threshold tests based on the contrast between cloudy scene and the background surface in a given target area of 1 × 1 km2 pixel. Once a pixel is identified as one of a particular domain (land, water, desert, etc), each threshold test will be performed to indicate a level of confidence. These series of tests combining the 250-m resolution bands and 1-km cloud mask are assembled into five different groups that indicate similar cloud conditions. They are arranged so that independence between groups is maximized. The minimum confidence is determined for each group from in-group tests, and the final cloud mask confidence (value Q) is then determined from the products of results for each group. After all tests, a summary is provided and a pixel is classified as one of four situations: confident clear (Q > 0.99), probably clear (0.99 > Q > 0.95), uncertain/probably cloudy (0.95 > Q > 0.66), or cloudy (Q < 0.66). For confidence values between 0.66 and 0.95, spatial and temporal continuity tests are further applied in order to determine whether the pixel is confidant clear or confidant cloudy. The final cloud fraction is then calculated from 5 × 5 km2 cloud mask pixel groupings and stored in MODIS cloud product.

Recent modifications to the operational algorithm (Ackerman et al. 2008; Frey et al. 2008) have led to significant progress in cloud detection in the polar region, at night, and in the sun-glint region. Compared to POLDER, MODIS cloud detection greatly benefits from the absorption, emission, and reflectance information provided by near infrared (NIR) and IR channels. Thus, MODIS algorithms will have better skills for detection of thin cirrus and low clouds over snow.

3. Description of POLDER/MODIS joint dataset

In this study, we used cloud parameters from collection 5 of the MODIS level-2 and level-3 products as well as collection 2 of POLDER level-2 and -3 products. The official MODIS cloud products in level 3 correspond to monthly averages stored in latitude versus longitude rectangular grid with a 1° × 1° gridbox resolution. Level-2 products are stored in granule, which covers a five-minute time interval, at either 1- or 5-km pixel resolution. POLDER level-1 products are geolocated on an integrated sinusoidal equal-area grid with pixels of 6 × 6 km2 resolution. The official POLDER cloud products in level 3 (monthly average) and level 2 (daily observation) are averaged at the superpixel scale corresponding to approximately 20 × 20 km2.

To compare accurately and conveniently the cloud parameters from these two satellites, we created a joint dataset containing both POLDER and MODIS level-2 official cloud products collocated and reprojected in a common sinusoidal grid, which is hereafter called POLDER–MODIS (PM) data. POLDER single-orbit files are used as reference for collocation of coincident MODIS granules. For each individual POLDER product orbit file, the sinusoidal grid used for collocation is centered at POLDER ascending node longitude. This procedure completely preserves POLDER cloud products while MODIS products are being collocated with POLDER pixels using a nearest pixel approximation and averaged at the scale of one POLDER superpixel. The final joint products thus provide both POLDER and MODIS cloud products at the same resolution of about 20 × 20 km2.

To compare the cloud products, we used a whole year of data from December 2007 to November 2008 to provide sufficient sampling and therefore ensure that the analyzed statistical results are as representative as possible. In the following, and unless otherwise stated, the results only concern this period.

Concerning the MODIS cloud fraction, two different daytime (solar zenith angle < 81.4°) products exist in the daily level-3 parameters list. One is called cloud fraction day mean (CFD), and the other is cloud fraction combined mean (CFC). CFD is obtained directly from statistics of the official cloud mask product (MOD35). The CFC is computed and associated to the pixels for which cloud optical properties have been successfully retrieved (Hubanks et al. 2008). A “clear sky restoral” algorithm is applied to the initial cloud mask results before retrieving cloud optical properties in an attempt to remove pixels that are initially either falsely detected as cloudy (heavy aerosol events, residual sun-glint contamination) or only partly cloudy (cloud edges). This algorithm mostly cuts down the edge of clouds, especially over the ocean, and reclassifies aerosols like blowing dust and sand around Africa as clear sky. A direct consequence is that the CFC is always smaller than or equal to the CFD. The CFC is theoretically smaller biased to the real cloud fraction than the CFD in regions of intense aerosols transport but will intrinsically underestimate cloud fraction in an area where fractional cloud cover is dominant. It is thus important to stress that the CFC does not represent the total cloud fraction but the cloud fraction for which optical properties have been derived. However, this part of the cloud fraction in absolute value and relative to the total cloud fraction needs to be especially understood to correctly interpret the statistical characteristics of cloud optical properties derived from MODIS. Comprehension of this part of cloud fraction is indeed important when comparing cloud properties derived from the two different sensors hereafter. In the following, it may be considered that the two cloud fractions derived from MODIS actually provide an upper (all clouds) and lower (solid clouds) limit of the global cloud fraction that should encompass POLDER cloud fraction at all times. The relative values of these three cloud fractions will thus provide valuable information on cloud cover characteristics and associated uncertainties, especially the POLDER cloud fraction and MODIS CFC both representing the cloud fraction from which the cloud optical properties are retrieved.

The ice and water cloud fraction used in this article are derived from the cloud fraction and cloud phase flags. Note, in addition, that all differences between POLDER and MODIS CFC (CFD) presented in this article correspond to the difference MODIS CFD minus POLDER and POLDER minus MODIS CFC.

Before using the new dataset to perform our comparisons, we have verified its consistency with respect to the official level-3 products. Figure 1 displays the latitudinal variations of cloud fraction averaged over one year for POLDER level-3 official data; MODIS level-3 official data (CFD and CFC) and our new dataset collected and averaged in the same period. The latitudinal variations of cloud fraction obtained from our joint level-2 dataset show the consistency with the official level-3 data. The small differences observed for CFC between the official and the joint POLDER/MODIS dataset have been attributed to one main difference in computing level-3 statistics: MODIS official level-3 algorithm computes statistics by sampling every other fifth level-2 pixel while we choose for the present analysis to average all level-2 MODIS pixels at POLDER superpixel scale. In this figure, we also plotted POLDER maximum and minimum cloud fraction values that are associated to the difference of cloud identification without applying additional tests (spatial and angular dispersion tests, section 2a). The maximum (minimum) value is computed assuming that all the pixels remaining undetermined after the first phase of the cloud detection algorithm are finally classified as cloudy (clear). These upper and lower cloud fractions are between CFC and CFD except in high latitudes because of a risk of snow cover. Fewer tests can then be executed. In this figure, we also show the ISCCP cloud fraction for a different period currently available that is from 1984 to 2007. ISCCP cloud fraction shows similar variations to MODIS and POLDER cloud fractions with absolute value closer to MODIS CFD.

Fig. 1.
Fig. 1.

Latitudinal variations of cloud fraction of POLDER and MODIS for the official level-3 dataset and for the dataset created for this study. Error bars present for POLDER the minimal and maximal cloud covers (from level 3) obtained before the reclassification tests. The ISCCP cloud fraction is for the period of 1984–2007.

Citation: Journal of Climate 24, 16; 10.1175/2011JCLI3857.1

MODIS swath is wider than the POLDER one, and this can have a significant impact on the overall statistical cloud fraction derived from each. Using the joint POLDER–MODIS dataset, we evaluated the cloud fraction differences due to these different instrument swaths. The latitudinal variations of the cloud fractions (CFCs) derived from the full MODIS swath (large swath), from the full POLDER swath (narrow swath), and from the MODIS observations outside of POLDER swath (MODIS swath minus POLDER swath) are presented in Fig. 2. The cloud fraction is larger for MODIS swath and even larger for the edge of the swath. This is due to the known viewing zenith angle (VZA) dependence of cloudiness derived from single-direction satellites observations (Minnis 1989, section 4.4). Indeed, cloud fraction detection increases with increasing VZA, and as the edges of the swath always correspond to larger VZAs, the cloud fractions appear greater than what are observed at the swath center. An almost constant difference of 2%–3% on CFC cloud fraction is observed between MODIS and POLDER swaths for all latitude. To minimize the impact of this systematic well identified bias and analyze only those differences associated to the cloud detection schemes, results presented in the following correspond to statistics computed for the overlapping zone of MODIS and POLDER swath. Meanwhile, some residual contamination of this effect may remain because the two satellites’ orbits are not exactly inline.

Fig. 2.
Fig. 2.

Latitudinal variations of CFC cloud fraction from wide swath (MODIS swath), narrow swath (POLDER swath), and edge of the large swath (POLDER swath subtracted from MODIS swath), and the differences of cloud fraction between MODIS and POLDER swath.

Citation: Journal of Climate 24, 16; 10.1175/2011JCLI3857.1

4. Statistical cloud fraction comparison

a. Global cloud fraction and latitudinal variations

We now start our comparison work by a glance of the global frequency on cloud detection for both sensors. In Table 1, we report the annual occurrence frequency of the clouds (either cloudy or clear determined by POLDER and MODIS CFC) over the globe. There are 18% (69%) of pixels that both sensors detect consistently clear-sky (cloudy) scene and only 12% of pixels disagree.

Table 1.

Global frequency of cloud detection from POLDER/MODIS CFC.

Table 1.

In Table 2, we show the global annual cloud fraction for POLDER and for the two MODIS products (CFD and CFC) over land and ocean, for ice and water clouds separately. MODIS detects about 67% (CFD) and 47% (CFC) of cloudiness over the whole globe compared to 56% for POLDER. These two instruments detect fewer clouds over land than over ocean, in accordance with established climatologies (Warren et al. 1986, 1988; Rossow and Schiffer 1999; Stubenrauch et al. 2010). The difference between POLDER cloud fraction and MODIS CFC (CFD) is 9% (−11%) on global average. Compared to MODIS CFC, POLDER observes more water clouds and less ice clouds. This will be confirmed and analyzed in the following.

Table 2.

Global-averaged cloud fractions (%) from POLDER and MODIS over land and ocean, and for total, liquid, and ice clouds. Note that the differentiation liquid/ice is not indicated for MODIS CFD as the phase product does not exist at this level.

Table 2.

In Fig. 3, we present latitudinal variations of cloud fraction for the two sensors. They have been restricted to the zone between 60°S and 60°N to discard the complex situations in the polar region due to snow-covered surfaces and day/night transition where POLDER has little skill because of the lack of thermal infrared channels. This figure shows that, following the synoptic atmospheric motion, variations of cloud fraction from both satellites follow a “W” shape, with high values in the intertropical convergence zone (ITCZ) and storm track (ST) areas of each hemisphere, and low values in the subtropical subsidence areas. As explained in section 3, the CFC cloud fraction is much smaller than the CFD one and in agreement with Table 2, POLDER cloud fraction is between the two MODIS cloud fractions. It is about 10% higher than the MODIS CFC and 10% lower than the MODIS CFD. The difference between POLDER and MODIS CFC is larger in the Southern Hemisphere, especially in midlatitude areas (40°S) and tends to decrease at higher latitudes.

Fig. 3.
Fig. 3.

Latitudinal variations of cloud fraction from POLDER and MODIS.

Citation: Journal of Climate 24, 16; 10.1175/2011JCLI3857.1

In Fig. 4, we also plot latitudinal variations of the cloud fraction for water and ice clouds separately. We again obtain consistent results with Table 2, that is, POLDER detects more water clouds with a quasi-constant bias for every latitude and on the contrary MODIS detects more ice clouds. For ice clouds, differences tend to be larger in the ITCZ and STs’ regions and smaller in subtropical subsidence areas.

Fig. 4.
Fig. 4.

Latitudinal variations of ice and water cloud fraction from POLDER and MODIS CFC.

Citation: Journal of Climate 24, 16; 10.1175/2011JCLI3857.1

b. Seasonal variations

In Fig. 5, we present the seasonal variations of MODIS (CFC and CFD) and POLDER cloud fractions. We focus on the seasonal cycle and represent here the differences between monthly averages and annual average. This representation masks out systematic bias (10%) between the two sensors (Fig. 3) and points out the main characteristics of the cloud fraction seasonal variations. Figure 5 displays this seasonal cycle over land and over ocean for four different zones: the tropics and subtropics (0°–30°) and the midlatitude (30°–60°) regions of each hemisphere. Although a systematic and significant bias among MODIS CFC, CFD, and POLDER cloud fraction is observed in Fig. 3, their seasonal cycles follow similar behavior in the four zones. Over land in the subtropics, cloud fraction rises to a maximum value in summer and goes down to a minimum in winter for each hemisphere. Situations are reversed in midlatitude regions where more clouds are found in winter. Over ocean, seasonal variations appear less pronounced. These very similar seasonal variations of cloud fraction confirm that the differences between POLDER and MODIS do not change with seasons. Beyond this generally good agreement, we can still notice some minor inconsistencies existing for certain regions and seasons. For example, over ocean in the Southern Hemisphere, POLDER and MODIS CFD detect a little bit more clouds than MODIS CFC during the summer (June–August) and over land in midlatitudes of Northern Hemisphere POLDER observes less clouds during winter.

Fig. 5.
Fig. 5.

Seasonal variations of cloud fraction in the subtropics area (0°–30°) over (a) ocean for the Northern Hemisphere, (b) ocean for the Southern Hemisphere, (c) land for the Northern Hemisphere, (d) land for the Southern Hemisphere; and in midlatitudes regions (30°–60°) over (e) ocean for the Northern Hemisphere; (f) ocean for the Southern Hemisphere; (g) land for the Northern Hemisphere; and (h) land for the Southern Hemisphere.

Citation: Journal of Climate 24, 16; 10.1175/2011JCLI3857.1

c. Differences between MODIS and POLDER cloud fractions

To better understand the cloud fraction differences observed between the two sensors, we plotted in Fig. 6 the probability density functions (PDFs) of cloud fractions and the difference (POLDER − MODIS CFC) for the full year over three types of surfaces: ocean without snow/ice cover, land without snow/ice cover and land covered by snow/ice. Most of the POLDER and MODIS pixels are labeled as either clear (bin value equal to 0) or overcast (bin value equal to 1) (Figs. 6a, 6b, and 6c). At the resolution of the PM dataset (about 20 × 20 km2), about 30% of the cloud cover is fractional. Over land and ocean, more POLDER pixels are declared overcast compared to MODIS CFC and logically the number of pixel declared overcast by MODIS CFD with value close to POLDER are larger than MODIS CFC. This is the opposite over snow, where more POLDER pixels are labeled as clear. Again in these figures, we can notice the higher cloud fraction over ocean than over land.

Fig. 6.
Fig. 6.

Histograms of cloud fractions over (a) ocean, (b) land, and (c) snow-covered land; and histograms of cloud fraction differences (POLDER-MODIS CFC) over (d) ocean, (e) land, and (f) snow-covered land.

Citation: Journal of Climate 24, 16; 10.1175/2011JCLI3857.1

In Figs. 6d, 6e, and 6f we report the mean values of the cloud fraction differences (POLDER − MODIS CFC) for the whole year. These are 10.6%, 10.8%, and −11.9% over ocean, land, and snow, respectively. Dispersion values represented by their root-mean-square (RMS) are 24%, 26%, and 36% over these three types of surfaces, respectively. Thus, we notice that the lowest mean difference and dispersion are observed over ocean and that a negative mean difference and the larger dispersion are obtained over snow/ice-covered surfaces.

To locate the significant differences observed between the three cloud fractions, Fig. 7 presents geographical distributions of POLDER and MODIS cloud fractions for different seasons. It also reports the differences for the overall clouds among the three cloud fractions. Differences [CF POLDER − CFC MODIS] are also presented for water and ice clouds separately. In general, POLDER and MODIS cloud fractions show similar geographical distributions, as already seen in the zonal variations presented in Fig. 3 with an overall larger CFD than POLDER, which in return is larger than CFC. We note that the difference between MODIS CFD and MODIS CFC is larger in strong aerosol loading regions and in broken cloud areas.

Fig. 7.
Fig. 7.

Geographical distribution of cloud fraction for (first row) POLDER, (second row) MODIS CFC, and (third row) MODIS CFD; geographical distribution of cloud fraction differences for (fourth row) MODIS CFD minus CFC, (fifth row) MODIS CFD minus POLDER, (sixth row) POLDER minus MODIS CFC, and also for (seventh row) water cloud fraction differences (POLDER minus MODIS CFC) and (eighth row) ice cloud fraction differences (POLDER minus MODIS CFC) during the four seasons in 2008.

Citation: Journal of Climate 24, 16; 10.1175/2011JCLI3857.1

For the three cloud fractions, high values are associated with deep convection located in the ITCZ, in STs, and monsoon areas and along the west coast of the continents (stratocumulus areas). Low values are associated to the subtropical ocean and deserts in subsidence regions. As already identified, differences between total cloud fractions are found almost everywhere positive and range from 0% to 20%. Ice cloud fraction differences are mostly negative ranging from −20% and 0. Namely, POLDER detects more clouds than MODIS CFC overall, but MODIS detects more ice clouds than POLDER. Of course these results from a combination of absolute cloud detection sensitivity and cloud phase determination differences are difficult to sort out without going into the details of the respective cloud phase products.

From the geographical distributions in Fig. 7, we can point out and classify some major regional and seasonal features. Related reasons are discussed in the next section:

  • Large positive differences between POLDER and MODIS CFC while negative differences between MODIS CFD and POLDER are observed especially in spring and summer in eastern and southern Africa, the center of South America and Australia, and in the summer of North Asia (section 5a). Note these regions are mostly over land where cloud fractions are smaller suggesting that broken clouds dominate.
  • Large positive differences (POLDER minus MODIS CFC) higher than 30% are observed during summer in the central part of South Africa and during winter in the Gulf of Guinea. We note that ice cloud fraction differences also show large positive values in the same place and time. (section 5c)
  • Negative differences between POLDER and MODIS CFC corresponding to large positive differences between MODIS and POLDER CFD are observed during the whole year over Greenland and Antarctica and during winter above Siberia and North America. (section 5b)
  • Negative differences (POLDER minus MODIS CFC) are observed during spring and summer in the transition zone between the desert and nondesert area (e.g., Sahara or Thar Desert in northwest of Indian; section 5b)
  • Strong negative differences for ice cloud fraction associated with large positive differences for water cloud are observed around the ITCZ (Indonesia, western equatorial Pacific Ocean, and equatorial South America), in the mid–high latitude STs’ regions and in Indian during the monsoon. (section 5d).
  • We also notice small negative differences in the northern Pacific Ocean around (50°N, 180°E/W), during spring and summer (section 5b).

d. Angular variation of the cloud fraction

Figure 8 presents the cloud fraction over ocean as a function of the VZA for solar angular bins of 5°. Negative view zenith angles correspond to backscattering directions with a relative azimuth angle less than 90° and positive values to forward directions with a relative azimuth angle greater than 90°. For MODIS, which obtains the observations by an east–west scanning, the VZA is directly linked to the cross-track distance from the center of orbit, namely, the nadir direction corresponds to the center of the orbit and oblique directions to the edges with the backscattering directions in the eastern part of the swath. For POLDER, which uses a CCD matrix, increasing VZAs correspond to increasing rings from the center of the wide two dimensional instantaneous field of view.

Fig. 8.
Fig. 8.

Cloud fraction as a function of VZA for different solar zenith angular bins for (a) MODIS and (b) POLDER. Negative view zenith angles correspond to relative azimuth less than 90°, namely, backward scattering directions.

Citation: Journal of Climate 24, 16; 10.1175/2011JCLI3857.1

In this figure, at first glance, it seems that cloud fraction increases steadily with the solar zenith angles (SZA) for both sensors. This is an artifact and is mainly due to statistic occurrences of SZA that are linked to specific latitudinal regions. Indeed, SZA tends to increase with latitudes as does cloud fraction for the latitudes between 20° and 60° (Fig. 1).

Besides, we notice that MODIS CFC has a stronger dependence than POLDER cloud fraction on the VZA with an increase more pronounced in the forward directions. The growth of cloud amount with VZA has already been observed and explained as a consequence of an increase of observed cloud sides (Minnis 1989). This effect can be also accentuated in case of thin clouds because they are better detected in oblique views, as the slant path through the cloud is longer. Detection of thin or fractional clouds can also explain the difference in cloud amount between forward and backscatter directions. As more radiation is scattered into forward directions, radiative threshold will lead to detect more clouds in these directions. In addition, spatial contrast used to select cloudy pixel will be enhance by shadow effects (Zhao and Di Girolamo 2004). The increase of cloud fraction with VZA can also come from a resolution effect (Wielicki and Parker 1992) as MODIS pixels near the swath edges cover more than 4 times larger areas than those at the center. Thanks to the POLDER optical design (combination of telecentric optics and aspherical lenses), the instrument does not present this drawback and provides an almost constant spatial resolution throughout its wide field of view, independent of the viewing angle (Deschamps et al. 1994). This explains, in part, why POLDER cloud fraction does not depend as much on VZA value. In addition, MODIS with a higher spatial resolution can certainly better identify broken clouds in near nadir direction comparing to oblique directions while POLDER with a lower spatial resolution is not able to see well the gaps between clouds even in near nadir directions.

5. Discussions

From Fig. 7, it is clear that both satellites are able to correctly capture the main features of global cloud cover and that observed differences are not randomly distributed. This calls for particular attention to identify specific situations where cloud climatologies could present significant uncertainties related to specific cloud macrophysical properties or environmental conditions. In section 4, we listed the most noticeable differences and noted that their sign and amplitude depend on the considered region and/or season. Reasons behind these differences are not obvious at first glance and may be due either to intrinsic instrument characteristics or to cloud detection algorithms. In this section, we try to explain most of the significant differences observed to identify and quantify uncertainties inherent to POLDER and MODIS cloud products, and from there, point out potential biases relevant to general cloud climatologies. This exercise can also provide useful guidance to improve POLDER and MODIS cloud detection algorithms or create an optimal merged cloud fraction dataset.

a. Impact of sensors spatial resolution

Figures 6d, 6e, and 6f, show that differences between POLDER and MODIS CFC cloud fraction are close to zero for more than 50% of the cases independently of surface type. These null differences are associated with either totally overcast or clear scenes, which appear very frequent (close to 60%) at the POLDER superpixel scale (Figs. 6a, 6b, and 6c). In these situations, both instruments determine consistently (and presumably correctly) the cloud coverage. Not surprisingly, differences happen, mostly for partly cloudy pixels (broken clouds or cloud edges). Indeed, as presented in section 2, POLDER and MODIS have different spatial resolutions that are 1 × 1 km2 for MODIS at nadir compared to 6 × 7 km2 for POLDER. Differences in cloud fraction due to spatial resolution have already been observed and studied (Wielicki and Parker 1992). Large effects are found for boundary layer clouds but differences are limited for thin cloud like cirrus. For algorithms using threshold methods, two opposite effects control the errors. With a lower spatial resolution, on one hand, the amount of optically thin clouds will be underestimated whereas thick clouds tend to be overestimated. Indeed, the statistical threshold for a low-resolution sensor needs to be large enough to discriminate a cloud-contaminated scene from a clear scene accounting for the instrument noise, which leads to missing some thin clouds. On the other hand, a low-resolution sensor will usually tend to classify more partly cloudy pixels into overcast situations as spatial resolution degrades, leading to an overestimation of cloud fraction. The happening frequency of the second situation increases as fractional clouds get thicker. According to these two opposite effects, if small clouds cover a POLDER pixel, the average reflectance may be sufficient (insufficient) for the pixel to be declared as cloudy if small clouds are (not) bright enough. In any case, the 6 × 7 km2 area observed by POLDER would be declared as either overcast or clear ignoring the underpixel cloud cover variations. On the other hand, MODIS with a smaller pixel resolution is able to distinguish a clear scene among fractional clouds at scale of POLDER pixel. The statistic results show in Fig. 7 that POLDER cloud fraction is generally larger than MODIS CFC in almost the whole globe. In land areas, where there may exist a lot of fractional clouds over bright surfaces (e.g., eastern and southern Africa, Australia, and south of South America), POLDER cloud fraction is about equal or even larger than CFD (negative differences) and in the meanwhile significant positive differences appear (>30%) between POLDER cloud fraction and MODIS CFC. This is explained by one of the two competitive effects described above. Fractional clouds are probably bright clouds and over bright surface that produce the larger observed differences between the high- and low-resolution instruments, POLDER with its lower resolution detects more clouds compared to any MODIS cloud fraction. For the fractional clouds over the subsidence tropical ocean, we see significant positive cloud cover differences between POLDER and MODIS CFC and small positive differences between MODIS CFD and POLDER. This illustrates the difference between MODIS CFC and CFD that MODIS CFD tends to classify more pixels containing even small clouds or cloud edges as cloudy while MODIS CFC cuts off a great number of them as explained in section 3. The statistic results also show that these systematically positive CF differences between POLDER and CFC are almost not dependent (Fig. 3) on the latitude and the season (Fig. 5). It confirms that this systematic bias (close to 10%) is not related to environmental conditions.

b. Impact of the surface types

According to the surface types, cloud detection algorithms have different efficiencies to detect clouds. In this section, we analyze MODIS and POLDER cloud fraction differences in relation with the ground surface type. In Fig. 6, we have noticed that histograms of cloud fraction differences are slightly different according to the type of surface: the dispersion (RMS) and mean difference are a little smaller over ocean than over land but much more important over snow-covered surfaces. The cloud detection is obviously easier over ocean than over land. Indeed, over ocean, out of the sun-glint region, the dark background is an ideal surface to perform threshold tests for all satellites using solar wavelength whereas over land or in sun-glint region, the surfaces are brilliant and thus make the use of such tests much more complex. In addition, terrestrial surface reflectances are highly variable in space and time.

Over snow-covered surfaces, cloud detection is even more difficult since snow surfaces are often as cold and brilliant as clouds. In Fig. 6c, we see that POLDER comparing to MODIS classify more pixels as clear. This implies a negative mean difference with the largest dispersion values (Fig. 6f). The omission of clouds by POLDER is also observed in Fig. 7 in mid-to-high latitude continents (northern America and northern Asia) during winter and in Greenland where negative differences appear for the whole year. Indeed, MODIS cloud detection algorithm can take advantage of the thermal IR bands information whereas POLDER, which uses mainly a apparent pressure test, detects well middle and high clouds and some thick clouds but miss many low cloud situations.

More specifically, in section 4, we pointed out the negative differences observed in the transition zones between desert and nondesert. Two reasons may explain these differences. One concerns the clear-sky ground reflectance, that is, for POLDER, derived from a time series analyzed by the POLDER “land surfaces” processing team (Leroy et al. 1997) while MODIS algorithm uses statistical desert mask to mark the desert border and may thus falsely identify desert area as cloudy pixels (Roy et al. 2002). This issue has been noticed and improved in MODIS cloud mask algorithm but during particular season when plants grow rapidly, a statistical value may still lead MODIS to label falsely desert pixels. Another reason can lie on the better detection of the very thin cirrus by MODIS (section 5d).

Sun-glint regions are also particular surfaces that make cloud detection a very difficult task when using solar wavelength. Indeed, as the solar reflectance is very high in the sun-glint directions, a choice of threshold between clear and cloudy scenes is unachievable. Consequently, MODIS uses the so-called sea surface temperature (SST) test, to improve the discrimination of low clouds with above-freezing cloud–top-temperatures. And moreover, a low product of the mean and standard deviation of 0.86-μm reflectances calculated over the pixel of interest and eight neighboring pixels is used to restore the clear sky from the uncertain. However, for those low clouds as warm as the sea surface, cloud detection may still be false (Frey et al. 2008). Concerning POLDER, only the apparent pressure test and the low 865-nm polarized reflectance test are used in the sun-glint directions, and the cloud amount is finally averaged over the direction both in and out of sun-glint that will generally minimize the possible bias due to sun glint.

To look for possible differences in cloud detection between sun-glint directions and out of sun-glint directions, we plot in Fig. 9 the histograms of cloud fraction for all pixels, for pixels corresponding to MODIS sun-glint directions (0°–36° from the specular direction), and for pixels out of sun glint. Since the two satellites are almost in line, whenever MODIS is contaminated by sun glint, some directions of POLDER will also be. Note that to avoid geographical or seasonal biases, histograms are limited to regions between 30° and 60°N during summer in 2008. Figure 9b shows less overcast pixels in sun-glint directions for both sensors compared to Fig. 9c out of sun-glint directions. However, again, this appears to be a geographical effect. As A-Train satellites cross the equator at around 1330 local time, sun-glint regions always correspond to smaller viewing angles compared to out of sun-glint angles. According to Fig. 8, this leads to larger cloud fraction outside of sun-glint due to view angle dependence of cloud detection. Another noticeable difference between Figs. 9b and 9c is the smaller difference in quasi-clear pixels detection between the two sensors compared to out of sun-glint directions. The reason for this is not obvious but cloud fraction of POLDER being averaged over all the 16 directions, even if more pixels are falsely classified as clear in sun-glint directions, in average with other directions or by reclassification, the error will be minimized and the cloud fraction will always be greater than zero. However, in sun-glint directions, MODIS uses the SST test and the spatial deviation test from 0.86-μm reflectances, which can falsely classify certain sun-glint as cloudy and then lead MODIS to find less clear pixels compared to out of sun-glint directions. This trends to reduce, in sun-glint directions, the difference between the numbers of clear pixels observed by POLDER and MODIS. Nevertheless, overall, no really significant differences are observed between the three histograms. This is confirmed by Fig. 8 where no unusual angular variations of cloud fraction corresponding to sun-glint directions (observed in forward direction around the solar angle value) can be detected except maybe for small solar angles (between 15° and 20°).

Fig. 9.
Fig. 9.

Histograms of the POLDER and MODIS cloud fractions for (a) all pixels, (b) pixels in MODIS sun-glint directions, and (c) pixels out of sun-glint regions. Histograms are for data between 30° and 60°N during summer 2008.

Citation: Journal of Climate 24, 16; 10.1175/2011JCLI3857.1

In addition during the present study, we identified a default in the POLDER cloud detection scheme. In some specific area, POLDER algorithm detects fractional cloud cover where scene is clearly overcast. This comes from a threshold problem in the sun-glint direction. As explained hereabove, in this particular direction, cloud detection is mainly based on the comparison of the apparent pressure with the ground pressure. Consequently, in regions of low clouds with an apparent pressure too close to the surface pressure, the algorithm classifies pixel as clear in the sun-glint directions and as cloudy in the other directions. The average over all directions results thus in a fractional cloud cover. This issue happens mostly in northern Pacific Ocean around (50°N, 180°E/W), during spring and summer when very low clouds appear.

c. Impact of the aerosols

Inversions of aerosols and clouds are inextricably connected and neglecting any one of them will lead to an inversion error. Certain simple tests taking advantage of IR and NIR channels’ information may be constructed by MODIS to indicate a presence of aerosol contamination such as the heavy dust around Africa (Ackerman 1997) and the biomass burning/smoke over dark vegetated surfaces (Kaufman et al. 2003). Without additional information from these IR and NIR channels POLDER probably classifies, in some directions, the heavy aerosol particles like biomass burning ones as cloud. This causes therefore an important positive difference in cloud fraction between POLDER and MODIS. It clearly appears in Fig. 7 where large positive differences are found in the center of Africa during summer. These large positive differences coincide well with the high accumulation rates of fires detected by MODIS and available from NASA rapid respond Web site (Fig. 10a). They also correspond to a bad cloud detection quality index for POLDER (Fig. 10b), which means that either the first part of the POLDER cloud detection scheme leaves the pixels undetermined or/and POLDER does not detect the same cloud fraction according to the direction. In these regions, the POLDER “Aerosols” processing line (Deuzé et al. 2001) finds a large value of optical thickness for fine mode aerosol (Fig. 10c). Moreover, as these aerosols are often nonspherical they are classified as ice clouds by POLDER (Goloub et al. 2000). Consequently, large positive differences for total clouds as well as for ice clouds cover are associated to the presence of these aerosols as illustrated in Fig. 7.

Fig. 10.
Fig. 10.

Geographical distribution of the active fires detected by MODIS for a period (9 Jul 2008–18 Jul 2008) in the summer 2008 from (a) NASA rapid respond Web site, (b) POLDER Quality index of cloud detection, and (c) POLDER fine mode aerosol optical thickness.

Citation: Journal of Climate 24, 16; 10.1175/2011JCLI3857.1

d. Impact of cirrus

Cirrus detection is another complex situation in satellite remote sensing domain as cirrus can be very optically thin. MODIS developed its own method to identify the very thin cirrus by using the 1.38-μm channel (Gao et al. 1993) and brightness temperature differences tests in split-window channels (Inoue 1987). POLDER polarized radiance is also sensitive to the cirrus but remain insensitive to the very thin cirrus. Again, the lack of additional NIR and IR information together with a larger resolution, leads to relatively poor skill in thin cirrus detection for POLDER. It is, however, worth noting that some cirrus clouds are observed with oblique directions among the total of 16. This is an advantage since the oblique and extended path through the clouds is more favorable for cirrus detection. Statistical results in Fig. 7 show negative differences for ice clouds cover almost all over the whole globe outside of regions with high aerosols loadings. Negative values are even noticeable in the following regions: Indian monsoon area, Indonesia, west-central Pacific Ocean warm pool, equatorial central South America as well as mid–high-latitude STs’ belts where active deep convection seems to generate lots of cirrus (Wylie et al. 1994; Sassen et al. 2008). So, for one part, these negative ice cloud fraction differences between POLDER and MODIS can be attributed to POLDER cloud detection limitations for thin cirrus and for another part to the sensitivity of thermodynamic phase determination for the very thin cirrus overlapping liquid clouds. Indeed, cirrus overlapping water clouds with an optical thickness less than 2 will be classified as liquid as the rainbow signature from the lower water clouds is still visible (Goloub et al. 2000; Riedi et al. 2010). Further studies need to be performed in the very near future with the help of information from active sensors like the Cloud–Aerosol Lidar with Orthogonal Polarization (CALIOP) on the Cloud–Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO) platform.

6. Conclusions

Cloud cover is one of the most important cloud parameters, first because it can be used for climate applications and second because all the other retrieved cloud parameters result on it. In this study, we compare cloud fractions obtained from two passive sensors, POLDER/PARASOL and MODIS/Aqua. To correctly handle the comparisons, we began our work by developing and checking a joint dataset that contains both POLDER and MODIS level-2 cloud products collocated and reprojected on a common sinusoidal grid. The advantage of this dataset lies on the simplification and the accuracy of the comparison work, which is done in the common swath of the two instruments at a scale of the POLDER superpixel (20 × 20 km2).

Although having different characteristics and different spatial resolutions, POLDER and MODIS show an interesting spatial and seasonal consistency in cloud detection. We noticed, however, an almost constant bias between MODIS cloud fraction day mean (CFD) and POLDER of about 10% and between POLDER and MODIS cloud fraction combined mean (CFC) of about 10% again. Compared to MODIS CFD, POLDER detects fewer clouds but more clouds than MODIS CFC. In addition to the removal of lots of fractional clouds and cloud edges in the MODIS CFC, we show that the differences come, in part, from the sensor resolution differences. In addition to these constant biases, we also focused on specific areas showing important positive or negative differences associated to typical cloud detection difficulties. For example, because of a lower resolution, POLDER badly detects the small clouds and tends to classify those as overcast cloud cover. Over snow, both sensors have difficulties in the cloud detection and particularly POLDER, which underestimates the cloud fractions significantly because of the lack of efficient information in the visible bands. We also note important differences over the transition region between desert and nondesert, which can be due to a better cirrus detection or an erroneous desert detection in the MODIS algorithm. In sun glint, besides the threshold problem of POLDER in the northern Pacific during summer, which has to be corrected, we did not notice large discrepancies caused by sun glint for both sensors. We also report that POLDER tends to misidentify the heavy aerosols and the clouds in some directions and that POLDER misses the very thin cirrus because of its lower resolution, its narrow range of detecting bands, and/or its tendency to detect the water phase instead of ice in cases of thin cirrus overlapping low water clouds.

This work allows us to exhibit a geographical localization of some problems in the cloud detection algorithms and demonstrates there are still some improvements that can be done for both instruments in the future. Typically, concerning POLDER, algorithm over snow surface, or in the sun-glint direction, needs to be improved to identify the low clouds. In addition, remedies for the identification of thin cirrus in the POLDER cloud detection need to be sought, as do methods for correctly identifying heavy aerosols loads as aerosols and not clouds. MODIS algorithm shows also some uncertainties, for example, in the transition region between desert and nondesert. To go further and refine the comparison of clouds detection, it can be also very helpful to use other instruments like CALIOP that would allow us to study those specific cases such as the thin cirrus or the multilayer cloud’s systems. Further comparisons on different types of clouds are also of great interest for further studies to gain a better comprehension of limitations and advantages of each cloud detection algorithm specifically for a certain types of clouds.

Finally, this paper provides the basis for further comparison studies of cloud thermodynamic phase, cloud optical properties, and cloud-top pressure products from these two sensors.

Acknowledgments

This work was supported by the French Space Agency (CNES) and the Région Nord-Pas de Calais. The POLDER level-2 and level-3 data were processed and distributed by the French ICARE Data Management and Processing Center (CGTD). The MODIS data were obtained from the NASA/GSFC/LADS. The authors are grateful to CNES and NASA for making the POLDER and MODIS data available.

REFERENCES

  • Ackerman, S. A., 1997: Remote sensing aerosols using satellite infrared observations. J. Geophys. Res., 102, 17 06917 079.

  • Ackerman, S. A., , K. Strabala, , W. Menzel, , R. Frey, , C. Moeller, , and L. Gumley, 1998: Discriminating clear sky from clouds with MODIS. J. Geophys. Res., 103, 32 14132 157.

    • Search Google Scholar
    • Export Citation
  • Ackerman, S. A., , R. E. Holz, , R. Frey, , E. W. Eloranta, , B. C. Maddux, , and M. McGill, 2008: Cloud detection with MODIS. Part II: Validation. J. Atmos. Oceanic Technol., 25, 10731086.

    • Search Google Scholar
    • Export Citation
  • Bréon, F.-M., , and M. Doutriaux-Boucher, 2005: A comparison of cloud droplet radii measured from space. IEEE Trans. Geosci. Remote Sens., 43, 17961805.

    • Search Google Scholar
    • Export Citation
  • Buriez, J. C., and Coauthors, 1997: Cloud detection and derivation of cloud properties from POLDER. Int. J. Remote Sens., 18, 27852813.

    • Search Google Scholar
    • Export Citation
  • Deschamps, P. Y., , F. M. Bréon, , M. Leroy, , A. Podaire, , A. Bricaud, , J. C. Buriez, , and G. Sèze, 1994: The POLDER mission: Instrument characteristics and scientific objectives. IEEE Trans. Geosci. Remote Sens., 32, 598615.

    • Search Google Scholar
    • Export Citation
  • Deuzé, J. L., and Coauthors, 2001: Remote sensing of aerosols over land surfaces from POLDER-ADEOS-1 polarized measurements. J. Geophys. Res., 106, 49134926.

    • Search Google Scholar
    • Export Citation
  • Frey, R. A., , S. A. Ackerman, , Y. Liu, , K. I. Strabala, , H. Zhang, , J. R. Key, , and X. Wang, 2008: Cloud detection with MODIS. Part I: Improvements in the MODIS cloud mask for collection 5. J. Atmos. Oceanic Technol., 25, 10571072.

    • Search Google Scholar
    • Export Citation
  • Gao, B.-C., , A. F. H. Goetz, , and W. J. Wiscombe, 1993: Cirrus cloud detection from airborne imaging spectrometer data using the 1.38 micron water vapor band. Geophys. Res. Lett., 20, 301304.

    • Search Google Scholar
    • Export Citation
  • Goloub, P., , H. Chepfer, , M. Herman, , G. Brogniez, , and F. Parol, 1997: Use of polarization for cloud studies. Polarization: Measurement, Analysis, and Remote Sensing, D. H. Goldstein and R. A. Chipman, Eds., International Society for Optical Engineering, (SPIE Proceedings, Vol. 3121), 330–341.

    • Search Google Scholar
    • Export Citation
  • Goloub, P., , M. Herman, , H. Chepfer, , J. Riedi, , G. Brogniez, , P. Couvert, , and G. Sèze, 2000: Cloud thermodynamic phase classification from the POLDER spaceborne instrument. J. Geophys. Res., 105, 14 74714 759.

    • Search Google Scholar
    • Export Citation
  • Hubanks, P., , M. King, , S. Platnick, , and R. Pincus, 2008: MODIS atmosphere L3 gridded product algorithm theoretical basis document. MODIS Doc. ATBD-MOD-30, 96 pp.

    • Search Google Scholar
    • Export Citation
  • Inoue, T., 1987: A cloud type classification with NOAA 7 split-window measurements. J. Geophys. Res., 92, 39914000.

  • Kaufman, Y. J., , C. Ichoku, , L. Giglio, , S. Korontzi, , D. A. Chu, , W. M. Hao, , R.-R. Li, , and C. O. Justice, 2003: Fire and smoke observed from the earth observing system MODIS instrument—Products, validation, and operational use. Int. J.Remote Sens., 24, 17651781.

    • Search Google Scholar
    • Export Citation
  • King, M., , Y. Kaufman, , W. Menzel, , D. Tanré, , N. Center, , and M. Greenbelt, 1992: Remote sensing of cloud, aerosol, and water vapor properties from the moderate resolution imaging spectrometer (MODIS). IEEE Trans. Geosci. Remote Sens., 30, 227.

    • Search Google Scholar
    • Export Citation
  • Leroy, M., and Coauthors, 1997: Retrieval of atmospheric properties and surface bidirectional reflectances over the land from POLDER. J. Geophys. Res., 102, 17 02317 037.

    • Search Google Scholar
    • Export Citation
  • Minnis, P., 1989: Viewing zenith angle dependence of cloudiness determined from coincident GOES East and GOES West data. J. Geophys. Res., 94, 23032320.

    • Search Google Scholar
    • Export Citation
  • Parol, F., , J. Buriez, , C. Vanbauce, , P. Couvert, , G. Seze, , P. Goloub, , and S. Cheinet, 1999: First results of the POLDER “earth radiation budget and clouds” operational algorithm. IEEE Trans. Geosci. Remote Sens., 37, 15971612.

    • Search Google Scholar
    • Export Citation
  • Parol, F., and Coauthors, 2004: Review of capabilities of multi-angle and polarization cloud measurements from POLDER. Adv. Space Res., 33, 10801088.

    • Search Google Scholar
    • Export Citation
  • Parol, F., , C. Vanbauce, , J. Riedi, , F. Thieuleux, , Z. Poussi, , and A. Lifermann, 2007: Comparison and statistical analysis of cloud properties derived from POLDER and MODIS. EUMETSAT Meteorological Satellite Conf. and the 15th AMS Satellite Meteorology and Oceanography Conf., Amsterdam, Netherlands. [Available from http://www.eumetsat.int/home/Main/AboutEUMETSAT/Publications/ConferenceandWorkshopProceedings/2007/groups/cps/documents/document/­pdf_conf_p50_s8_11_parol_p.pdf.]

    • Search Google Scholar
    • Export Citation
  • Platnick, S., , M. King, , S. Ackerman, , W. Menzel, , B. Baum, , J. Riedi, , and R. Frey, 2003: The MODIS cloud products: Algorithms and examples from Terra. IEEE Trans. Geosci. Remote Sens., 41, 459473.

    • Search Google Scholar
    • Export Citation
  • Riedi, J., and Coauthors, 2010: Cloud thermodynamic phase inferred from merged POLDER and MODIS data. Atmos. Chem. Phys., 10, 11 85111 865.

    • Search Google Scholar
    • Export Citation
  • Rossow, W. B., , and R. A. Schiffer, 1991: ISCCP cloud data products. Bull. Amer. Meteor. Soc., 72, 220.

  • Rossow, W. B., , and R. A. Schiffer, 1999: Advances in understanding clouds from ISCCP. Bull. Amer. Meteor. Soc., 80, 22612288.

  • Roy, D. P., , J. S. Borak, , S. Devadiga, , R. E. Wolfe, , M. Zheng, , and J. Descloitres, 2002: The MODIS Land product quality assessment approach. Remote Sens. Environ., 83, 6276.

    • Search Google Scholar
    • Export Citation
  • Sassen, K., , Z. Wang, , and D. Liu, 2008: Global distribution of cirrus clouds from CloudSat/Cloud–Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO) measurements. J. Geophys. Res., 113, D00A12, doi:10.1029/2008JD009972.

    • Search Google Scholar
    • Export Citation
  • Stubenrauch, C. J., , S. Cros, , A. Guignard, , and N. Lamquin, 2010: A 6-year global cloud climatology from the Atmospheric Infrared Sounder (AIRS) and a statistical analysis in synergy with CALIPSO and CloudSat. Atmos. Chem. Phys., 10, 71977214.

    • Search Google Scholar
    • Export Citation
  • Vanbauce, C., , J. C. Buriez, , F. Parol, , B. Bonnel, , G. Sèze, , and P. Couvert, 1998: Apparent pressure derived from ADEOS-POLDER observations in the oxygen A-band over ocean. Geophys. Res. Lett., 25, 31593162.

    • Search Google Scholar
    • Export Citation
  • Warren, S. G., , C. J. Hahn, , J. London, , R. M. Chervin, , and R. L. Jenne, 1986: Global distribution of total cloud cover and cloud type amounts over land. NCAR Tech. Note NCAR/TN-273+STR, 29 pp. + 200 maps.

    • Search Google Scholar
    • Export Citation
  • Warren, S. G., , C. J. Hahn, , J. London, , R. M. Chervin, , and R. L. Jenne, 1988: Global distribution of total cloud cover and cloud type amounts over the ocean. NCAR Tech. Note NCAR/TN-317+STR, 41 pp. + 170 maps.

    • Search Google Scholar
    • Export Citation
  • Wielicki, B., , and L. Parker, 1992: On the determination of cloud cover from satellite sensors: The effect of sensor spatial resolution. J. Geophys. Res., 97, 12 79912 823.

    • Search Google Scholar
    • Export Citation
  • Wylie, D. P., , W. P. Menzel, , H. M. Woolf, , and K. I. Strabala, 1994: Four years of global cirrus cloud statistics using HIRS. J. Climate, 7, 19721986.

    • Search Google Scholar
    • Export Citation
  • Zhang, Z., , P. Yang, , G. Kattawar, , J. Riedi, , L. C. Labonnote, , B. Baum, , S. Platnick, , and H.-L. Huang, 2009: Influence of ice particle model on retrieving cloud optical thickness from satellite measurements: Model comparison and implication for climate study. Atmos. Chem. Phys. Discuss., 9, 17571796.

    • Search Google Scholar
    • Export Citation
  • Zhao, G., , and L. Di Girolamo, 2004: A cloud fraction versus view angle technique for automatic in-scene evaluation of the MISR cloud mask. J. Appl. Meteor., 43, 860869.

    • Search Google Scholar
    • Export Citation
Save