• Ackerman, S A., , K I. Strabala, , W P. Menzel, , R A. Frey, , C C. Moeller, , and L E. Gumley. 1998. Discriminating clear sky from clouds with MODIS. J. Geophys. Res. 103:3214132157.

    • Search Google Scholar
    • Export Citation
  • Allen, R C., , P A. Durkee, , and C H. Wash. 1990. Snow/cloud discrimination with multispectral satellite measurements. J. Appl. Meteor. 29:9941004.

    • Search Google Scholar
    • Export Citation
  • Basist, A., , D. Garrett, , R. Ferraro, , N. Grody, , and K. Mitchell. 1996. A comparison between snow cover products derived from visible and microwave observations. J. Appl. Meteor. 35:163177.

    • Search Google Scholar
    • Export Citation
  • d’Entremont, R P., and L W. Thomason. 1987. Interpreting meteorological satellite images using a color composite technique. Bull. Amer. Meteor. Soc. 68:762768.

    • Search Google Scholar
    • Export Citation
  • Dirmhirn, I., and F D. Eaton. 1975. Some characteristics of the albedo of snow. J. Appl. Meteor. 14:375379.

  • Dozier, J. 1989. Remote sensing of snow in visible and near-infrared wavelengths. Theory and Applications of Optical Remote Sensing, G. Asrar, Ed., John Wiley and Sons, 527–547.

    • Search Google Scholar
    • Export Citation
  • Gao, B-C., , Y J. Kaufman, , W. Han, , and W. Wiscombe. 1998. Correction of thin cirrus path radiances in the 0.4-1.0 μm spectral region using the sensitive 1.375 μm cirrus detecting channel. J. Geophys. Res. 103:3216932176.

    • Search Google Scholar
    • Export Citation
  • Hall, D K., , G A. Riggs, , V V. Salomonson, , N E. DiGirolamo, , and K J. Bayr. 2002. MODIS snow-cover products. Remote Sens. Environ. 83:181194.

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

    • Search Google Scholar
    • Export Citation
  • Kidder, S Q., and H-T. Wu. 1984. Dramatic contrast between low clouds and snow cover in daytime 3.7 μm imagery. Mon. Wea. Rev. 112:23452346.

    • Search Google Scholar
    • Export Citation
  • Lee, T F., , F J. Turk, , and K. Richardson. 1997. Stratus and fog products using GOES-8–9 3.9-μm data. Wea. Forecasting 12:664677.

    • Search Google Scholar
    • Export Citation
  • McCrone, P J. 2003. Theater snow depth estimates for Department of Defense applications. Preprints, 12th Conf. on Satellite Meteorology and Oceanography, Long Beach, CA, Amer. Meteor. Soc., CD-ROM, JP1.16.

  • Miller, S D., , F J. Turk, , T F. Lee, , K. Richardson, , and J D. Hawkins. 2002. The enhanced role of the polar orbiter constellation in tropical system monitoring in the wake of a geostationary platform failure. Preprints, 25th Conf. on Hurricanes and Tropical Meteorology, San Diego, CA, Amer. Meteor. Soc., CD-ROM, P1.22.

  • Pilewskie, P., and S. Twomey. 1987. Cloud phase discrimination by reflectance measurements near 1.6 and 2.2 μm. J. Atmos. Sci. 44:34193420.

    • Search Google Scholar
    • Export Citation
  • Ramsay, B. 1998. The interactive multisensor snow and ice mapping system. Hydrol. Processes 12:15371546.

  • Robinson, D A., , K F. Dewey, , and R R. Heim. 1993. Global snow cover monitoring: An update. Bull. Amer. Meteor. Soc. 74:16891696.

    • Search Google Scholar
    • Export Citation
  • Robock, A., and D. Kaiser. 1985. Satellite-observed reflectance of snow and clouds. Mon. Wea. Rev. 113:20232029.

  • Romanov, P., , G. Gutman, , and I. Csiszar. 2000. Automated monitoring of snow cover over North America with multispectral data. J. Appl. Meteor. 39:18661880.

    • Search Google Scholar
    • Export Citation
  • Schmetz, J., , S A. Tjemkes, , M. Gube, , and L. van de Berg. 1997. Monitoring deep convection and convective overshooting tops with Meteosat. Adv. Space Res. 19:433441.

    • Search Google Scholar
    • Export Citation
  • Schmetz, J., , P. Pili, , S. Tjemkes, , D. Just, , J. Kerkmann, , S. Rota, , and A. Ratier. 2002. An introduction to Meteosat Second Generation (MSG). Bull. Amer. Meteor. Soc. 83:977992.

    • Search Google Scholar
    • Export Citation
  • Turk, F J., and S D. Miller. 2005. Towards improved characterization of remotely sensed precipitation regimes with MODIS/AMSR-E blended data techniques. IEEE Trans. Geosci. Remote Sens. 43:10591069.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    Normalized red/green/blue color-gun contributions to the version-2 snow/cloud algorithm and resultant composite tonality as a function of scene type. Points in the right panel are mean values of corresponding boxes in left panel; whiskers denote 1-sigma standard deviation.

  • View in gallery

    VIS, SIR, CIR, and TIR imagery for a snow/cloud scene over Colorado and Wyoming as observed by Terra MODIS on 1800 UTC 22 Mar 2003.

  • View in gallery

    Terra MODIS (left) true color, (middle) version-1, and (right) version-2 cloud/snow enhancements depict rapid snowmelt across Colorado and Wyoming snowmelt over the period 20–25 Mar 2003.

  • View in gallery

    (left) Visible image, (middle) version-1 snow/cloud enhancement, and (right) corresponding NOAA/NESDIS snow product over the Great Lakes region depicting snow, cloud, and lake ice as observed by Terra MODIS on 1640 UTC 2 Mar 2003.

  • View in gallery

    (left) Visible image and (right) version-2 snow/cloud enhancement over the Hindu Kush mountain range of northeastern Afghanistan as observed by Aqua MODIS on 0845 UTC 19 Feb 2003.

  • View in gallery

    (left) True color, (middle) version-1 cloud/snow enhanced, and (right) version-2 cloud/snow enhancement over the northwestern Zagros mountains within Iraq, Iran, and Turkey as observed by Terra MODIS on 0825 UTC 10 Feb 2004.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 138 138 35
PDF Downloads 87 87 28

Satellite-Based Imagery Techniques for Daytime Cloud/Snow Delineation from MODIS

View More View Less
  • a Naval Research Laboratory, Monterey, California
  • b National Geospatial-Intelligence Agency, Bethesda, Maryland
© Get Permissions
Full access

Abstract

This paper presents two multispectral enhancement techniques for distinguishing between regions of cloud and snow cover using optical spectrum passive radiometer satellite observations from the Moderate Resolution Imaging Spectroradiometer (MODIS). Fundamental to the techniques are the 1.6- and 2.2-μm shortwave infrared bands that are useful in distinguishing between absorbing snow cover (having low reflectance) and less absorbing liquid-phase clouds (higher reflectance). The 1.38-μm band helps to overcome ambiguities that arise in the case of optically thin cirrus. Designed to provide straightforward, stand-alone environmental characterization for operational forecasters (e.g., military weather forecasters in the context of mission planning), these products portray the information that is contained within complex scenes as value-added, readily interpretable imagery at the highest available spatial resolution. Their utility in scene characterization and quality control of digital snow maps is demonstrated.

Corresponding author address: Steven D. Miller, Naval Research Laboratory, 7 Grace Hopper Ave., MS #2, Monterey, CA 93943-5502. miller@nrlmry.navy.mil

Abstract

This paper presents two multispectral enhancement techniques for distinguishing between regions of cloud and snow cover using optical spectrum passive radiometer satellite observations from the Moderate Resolution Imaging Spectroradiometer (MODIS). Fundamental to the techniques are the 1.6- and 2.2-μm shortwave infrared bands that are useful in distinguishing between absorbing snow cover (having low reflectance) and less absorbing liquid-phase clouds (higher reflectance). The 1.38-μm band helps to overcome ambiguities that arise in the case of optically thin cirrus. Designed to provide straightforward, stand-alone environmental characterization for operational forecasters (e.g., military weather forecasters in the context of mission planning), these products portray the information that is contained within complex scenes as value-added, readily interpretable imagery at the highest available spatial resolution. Their utility in scene characterization and quality control of digital snow maps is demonstrated.

Corresponding author address: Steven D. Miller, Naval Research Laboratory, 7 Grace Hopper Ave., MS #2, Monterey, CA 93943-5502. miller@nrlmry.navy.mil

Introduction

Real-time maps of snow cover are extremely useful to weather forecasters, planners, and resource managers. There are numerous applications requiring knowledge of the cloud obscuration of snow-covered land or ice (especially sea ice), including search and rescue missions, aircraft icing, and military surveillance (McCrone 2003) in support of ground forces. Satellite data have accounted for major improvements in the production of reliable global snow cover maps. Often, however, weather forecasters require more detailed information than a simple snow cover mask can offer—they also need products that can put this information in the context of the local environment and meteorological situation (e.g., information on cloud cover and topography in the vicinity of the snow field). This need is particularly strong for operational military forecasters, who may be concerned with the finescale cloud/snow cover details within a narrow data-sparse/-denied location. Over remote, mountainous terrain, where measurements of snow and cloud cover are sparse or nonexistent, satellite observations play a particularly important role.

Traditional cloud/snow analyses have concentrated mainly on digitized masking, separating snow-covered from non-snow-covered regions in a discrete fashion and over relatively coarse grids relative to the native resolution of satellite data contributing to the mask. For example, the National Oceanic and Atmospheric Administration (NOAA) produces comprehensive maps of global snow cover, sea ice cover, and snow-/ice-free conditions produced at a 25-km grid cell resolution on a daily basis. Here, clouds and other nonsnow features are considered as “clutter” to be removed, so that snow cover areas may be identified unambiguously (Ackerman et al. 1998). Thus, the resulting products are usually partitioned into a few broad categories, such as snow cover, snow-free land, inland water bodies, and cloud obscured (e.g., Robinson et al. 1993; Basist et al. 1996; Hall et al. 2002). These categories are invaluable as global and regional summaries of snow cover conditions and in studies of climate, climate change, and water resource management. Because image detail is lost as a result of the classification procedure, however, military weather analysts who are concerned with the finescale environment surrounding an area of interest have difficulty using these kinds of products. Furthermore, because these analysts need to assess snow cover in the context of the overlying and/or adjacent cloud cover, products that retain as much cloud detail (e.g., structure, shadowing, meteorological context) as possible are very useful.

To identify cloud/snow regions forecasters sometimes rely upon comparisons between visible, middle infrared (3.7 μm), and thermal infrared satellite imagery. When snow cover, sea/lake ice, and clouds of various types appear in the scene at the same time, interpretation via single-channel imagery comparisons becomes extremely difficult. On infrared images, for example, low cloud (e.g., stratus/stratocumulus) decks are often not detectable because of poor thermal contrast with the underlying snow-covered surface of the earth. High cirrus, although usually of a much lower thermodynamic temperature than a snow-covered background, may produce radiometric temperatures that are similar to snow cover owing to transmission effects. Middle infrared imagery has the capability of distinguishing snow cover (highly absorbing) from low clouds (highly reflective), but has difficulties dealing with optically thick cirrus, which may contain large ice crystals with absorption properties that are comparable to snow cover. In some instances thermal infrared channels can distinguish this type of cirrus from snow, but even this property fails in the case of strong temperature inversions at the surface. In the visible (0.6 μm) region, highly reflective snow cover contrasts poorly with all varieties of clouds, making the unambiguous identification of either feature difficult. Owing to these problems, the interpretation of clouds over snow-covered regions from single-channel imagery is an inherently difficult challenge. However, as shown here, combining the three bands intelligently results in a more definitive characterization of the cloud/snow distribution.

The current effort seeks to develop imagery-based enhancements of snow cover, clouds, and background at the full spatial resolution capabilities of the sensors. This is accomplished by using color-composite techniques (e.g., d’Entremont and Thomason 1987) to analyze cloud/snow fields using the multispectral Earth Observation System (EOS) Moderate Resolution Imaging Spectroradiometer (MODIS) daytime observations. MODIS features shortwave infrared window channels (e.g., 1.6 and 2.2 μm), a key measurement owing to differential absorption properties between clouds and snow cover at this wavelength. Prior to the availability of these bands on environmental satellites, efforts to discriminate between clouds and snow cover in support of forecasters and other applications relied upon the 3.7–3.9-μm middle infrared channels (Kidder and Wu 1984; Allen et al. 1990; Lee et al. 1997; Romanov et al. 2000) aboard the Geostationary Operational Environmental Satellite (GOES) and Television Infrared Observation Satellite (TIROS). However, stronger ice absorption in the 3.7-μm window occasionally gives rise to ambiguities between cirrus and snow cover. Through the selection 1.6-/2.2- instead of 3.7-μm observations, the techniques here yield a product with less potential for confusion.

Technique descriptions

Outlined here are two versions of cloud-versus-snow enhancement; the first depicts snow cover as white and all clouds as yellow, and the second further partitions low and high clouds into yellow and magenta sections, respectively. Although neither method can be considered as a digital mask, extension to this capability is readily achieved through appropriate thresholds applied to each color gun, as explained below. Both techniques enlist several narrowband channels in the optical portion of the electromagnetic spectrum—visible (VIS; ∼0.65 μm), near-infrared (NIR; ∼0.86 μm), shortwave infrared (SIR; ∼1.6 or 2.2 μm), and thermal infrared (TIR; ∼11 μm). On MODIS, the corresponding channel indices are 1, 2, 7 (2.2 μm), and 31, with additional information from the 1.38-μm “cirrus channel” (“CIR”; channel 26) that is used as a screen for thin cirrus clouds. For Aqua, the MODIS 2.2-μm channel was used instead of the 1.6-μm channel because of faulty detectors (several channel-6 detectors of the 10-element array dislodged during vibrational testing, resulting in severe imagery striping) on the latter.

In contrast to conventional imaging radiometers, which feature a single detector scanning each line of the scene, MODIS employs a focal plane array (FPA) with 10 detectors stacked in the along-orbit-track dimension for the 1-km spatial resolution channels (i.e., a 10-km band is imaged simultaneously). A single physical scan from the MODIS FPA results in the imaging of a 10-km cross-track strip; the subsequent scan occurs 10 km down track, producing no pixel overlap at instrument nadir. Stacking the detectors in this way allows for a slower scan rate than that of a single detector and, therefore, increases the dwell time (building up signal-to-noise ratio) for a given pixel. The motivation for this design was the need to meet both high signal-to-noise ratio and high spatial resolution requirements simultaneously.

Unfortunately, two artifacts emerge as a trade-off to this multidetector arrangement that must be dealt with as a preprocessing step: the so-called “bow tie” effect and imagery striping. The bow-tie effect increases toward the swath edges where the projection upon the earth’s surface of adjacent FPA scans begins to overlap considerably. Because each pixel is correctly geolocated, registration of the data to a map projection remedies this artifact. Imagery striping arises from nonnegligible response function differences between the nominally identical detectors in the FPA. This results in a periodic banding effect in imagery that varies from channel to channel. An additional correction to channel 26 (1.38 μm) to reduce a photon leakage error from an adjacent band has also been applied. The destriping software, applied here as preprocessing, was developed by the Space Science and Engineering Center (SSEC) at the University of Wisconsin—Madison (L. Gumley and C. Moeller 2003, personal communication). The removal of bow-tie and striping effects are important components to ensure the highest quality product.

Snow/cloud algorithm, version 1: A binary discriminator

A useful approach for snow/cloud discrimination entails the spectral difference between visible and shortwave infrared reflection reflectance (e.g., Pilewskie and Twomey 1987; Dozier 1989), formulated here as follows:
i1520-0450-44-7-987-e1
where VIS and SIR are MODIS visible and shortwave infrared albedos (expressed in units of percent), and μo is the cosine of the solar zenith angle. The solar zenith angle corrections improve the contrast near the terminator. The addition of the constant shifts the dynamic range of information to roughly [0, 100], over which the Dsnow variable is then truncated. High values of Dsnow generally indicate snow cover; low values are associated with cloudy or cloud-free regions. However, there are some pixels in the Dsnow field that have relatively high values but do not represent snow cover at all (e.g., over desert regions). We screen out most of these anomalous values from Dsnow by applying stringent thresholds to low-elevation regions but more relaxed thresholds to mountains, where snow cover is more likely. Specifically, we apply the following elevation-parameterized reflectance threshold:
i1520-0450-44-7-987-e2
where Zt is the surface elevation, truncated at 3000 m (i.e., the maximum value assumed by R1, is 30.0), as obtained from a National Geophysical Data Center 2-min-resolution (roughly 4 km at the equator) topographical database that is registered to the satellite map projection. Applying R1(units of percent) to the SIR channel increases the likelihood of assigning snow cover to higher-elevation pixels. We then enforce the following criteria at each pixel for identification as a possible snow-covered pixel:
i1520-0450-44-7-987-e3
i1520-0450-44-7-987-e4
An additional screen for cirrus clouds, using a normalized difference of the VIS and CIR channel albedos, is included,
i1520-0450-44-7-987-e5
such that any significant contributions from CIR (indicative of cirrus being present in the scene) result in no snow cover being analyzed for these pixels. The CIR channel is situated in a part of the spectrum where tropospheric water vapor absorbs sunlight strongly. This absorption shields much of the reflection from low clouds, effectively masking them from the scene and leaving only upper-level clouds (located above most tropospheric water vapor). This spectral band has been utilized both for cirrus detection (e.g., Hutchison and Choe 1996) and filtering (Gao et al. 1998; Turk and Miller 2005) applications. All of the pixels satisfying the screening criteria [Eqs. (3)(5)] contribute to the snow cover term of the version 1 enhancement, computed as the logarithm of Eq. (1),
i1520-0450-44-7-987-e6
Here, logarithmic scaling helps to amplify the relative contributions of patchy snow cover in dark forested regions (Robock and Kaiser 1985). If any of the aforementioned screening criteria are not met, Ds is set to zero. In forming the final snow/cloud imagery, three variables feed into the red (R), green (G), and blue (B) channels (referred to as “color guns”) of a false color composite image as follows:
i1520-0450-44-7-987-e7
i1520-0450-44-7-987-e8
i1520-0450-44-7-987-e9

The resulting enhancement depicts snow cover as shades of white or cyan, reflecting a strong contribution from all three of the color guns. In the case of clouds, Dsnow is zero (no blue color-gun contribution), and the strong red and green color-gun contributions combine in the absence of blue to produce a yellow tonality. For snow- and cloud-free land regions, VIS reflectance dominates slightly over NIR in nonvegetated scenes, making these areas appear green. For more vegetated regions, NIR has a relatively stronger contribution, which results in a reddish-brown tonality. The strong color contrast between cloud and snow areas facilitates delineation among the various features of the scene. In this paper we present the version 1 enhancement with MODIS data. Similar processing (not discussed here) is possible, based on measurements from the Advanced Very High Resolution Radiometers (AVHRR) aboard the NOAA Polar Orbiting Environmental Satellites, but with alternate cirrus screens [Eqs. (4)(5) above], owing to the absence of the 1.38-μm channel on AVHRR.

Snow/cloud algorithm, version 2: Adding high/low cloud discrimination

The version 2 algorithm employs a slightly more sophisticated approach to further decouple elements of the scene. Because high, optically thick clouds produce lower TIR brightness temperatures than low-level clouds, we include TIR data explicitly in the R/G/B color guns for additional cloud discrimination. To reduce the ambiguity associated with a simple TIR temperature criterion, we again enlist the CIR channel (decoupling low/high clouds in the scene) on MODIS. In this way, semitransparent cirrus are depicted correctly as high cloud in the enhancement, and are not misclassified as snow. Last, we introduce a scaled version of the optically thick high cloud indicator (e.g., Schmetz et al. 1997), formed by a difference between the vibrational water vapor absorption band near 6.7 μm (WV) and the TIR window channel:
i1520-0450-44-7-987-e10

For clear-sky scenes or lower-level clouds WV_TIR is typically negative, owing to the presence of absorbing water vapor above cloud top that depresses the WV channel brightness temperatures. The exception to this rule is at high latitudes where the surface TIR brightness temperature is similar to, if not lower than, the water vapor channel value. As such, the WV_TIR test is omitted in regions poleward of 50°N/50°S latitude. Elsewhere, when optically thick clouds reside in the upper part of the troposphere (and above most of the water vapor), the WV and TIR brightness temperatures converge, leading to small absolute values in Eq. (10), and making the test a useful flag for deep clouds. The advantage is that the screen is independent of absolute temperature thresholds.

In formulating the version 2 algorithm, the VIS, CIR, SIR, WV_TIR, and TIR channels first are truncated, normalized, and byte scaled over the following ranges:
i1520-0450-44-7-987-e11
i1520-0450-44-7-987-e12
i1520-0450-44-7-987-e13
i1520-0450-44-7-987-e14
i1520-0450-44-7-987-e15

These scaling bounds were determined experimentally and encompass all seasons (with bias toward the Northern Hemisphere winter season at midlatitudes). The bounds on Eq. (15) were chosen conservatively such that clear-sky high-terrain features [residing above much of the tropospheric water vapor and, hence, producing smaller differences in Eq. (10)] would not contribute erroneously to the enhancement.

Using the above scalings, the R/G/B color-gun components of the version 2 enhancement are then defined as follows:
i1520-0450-44-7-987-e16
i1520-0450-44-7-987-e17
and
i1520-0450-44-7-987-e18
where the elevation scaling applied to CIRnorm, specified as
i1520-0450-44-7-987-e19
reduces the effects of surface reflection contributions to CIRnorm from very high terrain (i.e., less intervening water vapor between the surface and the satellite). In kilometers, Z is the untruncated above mean sea level elevation. As elevation increases, the CIRnorm contributions present in Eqs. (16) and (18) are reduced because of the increased likelihood of the surface reflection contaminating the signal. Each color gun is gamma corrected by a factor of 2.5 prior to forming the composite image. The rationale behind the channels selected for Eqs. (16)(18) is explained below. All multiplicative constants were determined through experimental tuning, based on 1) the known signs of contribution for each scaled/normalized component as a function of scene type, and 2) the relative color-gun magnitudes required for visual distinction of each scene constituent in the enhanced image.

Combining multispectral information into each gun produces intermediate R/G/B image components that yield discriminatory information in color space. Figure 1 provides an example of the behavior of Eqs. (16)(18) in the version-2 enhancement. Here, the mean and standard deviations of R/G/B-normalized color-gun contributions (denoting relative strength) are plotted for the four main categories (land, snow cover, low cloud, and high cloud) present in this scene. The range of values present in any given scene and color gun represents the inherent variability of real-world data (surface/cloud/snow properties and the thermal environment), and, therefore, the exact appearance of a given class, will also vary over a range of optical properties and sun/sensor geometry. While the truncation ranges and scaling coefficients of Eqs. (11)(19) are derived from the case studies that are examined, they are not optimized for any given case (a trade-off made in favor of minimizing enhancement variability).

Referring back to Eqs. (16)(18), and presented in the context of Fig. 1, snow-free land scenes (relatively warm in the infrared and less reflective in the solar channels, in comparison with most clouds) are dominated by the TIRnorm term of Eq. (17), and, thus, appear in varying shades of green. Snow cover appears bright white owing to strong VISnorm contributions in each of the color guns. Low clouds appear yellow owing to high VISnorm and SIRnorm reflectance, but with a slightly reduced contribution in the red gun on account of their relatively warm IR temperatures [TIRnorm is subtracted in Eq. (16)]. For midtroposphere clouds, the terms containing CIRnorm [i.e., Eqs. (16) and (18)] become increasingly important and contribute preferentially to the red color gun, leading to an orange tonality. Still higher clouds begin to demonstrate depressed SIRnorm values because of absorption by ice crystals, and, correspondingly, the magnitude of Eq. (18) (blue) increases. This phase-dependent effect combines with increased WV_TIRnorm contributions [and a decreasing contribution from the green gun given by Eq. (17)] to produce bright magenta tonalities for optically thick high clouds (e.g., thunderstorm tops). Optically thin high clouds lack a strong green gun contribution [weak VISnorm, WV_TIRnorm signals in Eq. (17)] and are dominated by the strong CIRnorm contribution of Eq. (16), resulting a deep magenta color over water. Over land, the transparency of thin cirrus leads to significant green contributions as well [TIRnorm term of Eq. (17)], leading to a deep orange tonality that is generally darker than the orange tonality produced by midlevel clouds.

Examples and applications

The following examples demonstrate applications of both cloud/snow delineation techniques developed above. Although variations on the version-1 and -2 techniques are applicable to both AVHRR and MODIS, here we present only MODIS (a spectral superset of AVHRR, with the added benefits of the cirrus-screening capability) examples in order to compare the two methods directly, thereby avoiding interpretive complications arising from differing satellite-/solar-viewing geometries, spatial resolution, or temporal mismatch.

Conventional single-channel imagery capabilities

During mid-March 2003, a powerful late-season winter storm deposited record-breaking snowfall across Colorado, particularly in areas along the Front Range (e.g., Cheyenne Ridge, Palmer Divide, and Raton Mesa, as noted) where orographic forcing dynamics came into play. By 20 March, when the storm finally left the area, much of the Front Range in central Colorado and eastern Wyoming was blanketed with snow. The MODIS VIS image (Fig. 2, upper left) depicts a complicated scene of low and high clouds and snow cover. Interpretation of the image is difficult because both clouds and snow cover are reflective at visible wavelengths and, thus, resemble one another. In contrast, on the SIR image (Fig. 2, upper right) the snow cover becomes dark, representing significant solar absorption. However, other absorbing features of the SIR scene (e.g., Great Salt Lake in Utah and other surface types) also appear dark. Hence, both the SIR and VIS images can give rise to ambiguity if used independently. Low clouds appear bright in both scenes, demonstrating strong solar reflection by water droplets in both bands. The CIR image (Fig. 2, lower left) shows only upper-level cirrus clouds, because most low clouds and surface features are screened out by intervening water vapor. Thus, it is an effective tool to help to distinguish cirrus from land features. Unfortunately, it is not always effective. At times, in extremely dry air masses, there is insufficient water vapor to screen out some land surfaces. A similar difficulty arises in the case of elevated topography (e.g., mountain peaks) that resides above the bulk of the tropospheric vapor. Conservative reflectivity thresholds can usually be selected to filter out a majority of these ambiguous contributions, albeit at the expense of some sensitivity to thin cirrus. Last, the TIR image (Fig. 2, lower right) is characteristically ineffective in distinguishing cloud from ground snow cover. At this wavelength, optically thin cirrus clouds often blend in with the underlying features.

Snowmelt monitoring over the Colorado Front Range

Figure 3 demonstrates the utility of the enhancements in monitoring the evolution of snow cover over the course of several days. True color (no cloud/snow enhancement) is shown in the left panels, the version 1 enhancement in the middle panels, and the version 2 enhancement in the right panels. Commencing on 20 March 2003, the sequence depicts widespread melting (particularly over the Colorado Front Range) over the period ending on 25 March. Shifting patterns of low and high clouds obscure different areas of snow cover from image to image, but the enhancements allow for consistent interpretation. In comparison with Fig. 2, both snow cover and cloud cover regions are revealed simultaneously, and in vivid contrast. In the version 1 examples, all clouds are enhanced as yellow, and snow cover appears bluish white. Optically thin cirrus clouds falling below the CIR threshold [Eq. (4)] are not enhanced as yellow cloud, such that the bright snow cover enhancement can show through. The version 2 enhancement further decouples high clouds from low clouds, and enables specification of thin cirrus over snow cover.

At these synoptic observing scales, temporal resolution on the order of 24 h was sufficient for monitoring the snowmelt. To take full advantage of snowmelt monitoring for mesoscale scenes, an increased observation frequency is required. Fortunately, the aggregate coverage of several polar-orbiting platforms with similar observing capabilities can provide updates on the order of hours (e.g., Miller et al. 2002). Still higher temporal resolution (e.g., on the order of minutes) monitoring of snowmelt is possible from geostationary platform sensors at the expense of coarser spatial resolution.

Snow/ice coverage over the Midwestern United States

Figure 4 shows a MODIS (Terra) case study over the Great Lakes region on 2 March 2003, with visible channel imagery in the left panel, the cloud/snow version-1 enhancement in the middle panel, and a collocated snow/ice analysis produced operationally by NOAA/National Environmental Satellite, Data, and Information Service (NESDIS) in the right panel (Ramsay 1998). Both Lake Michigan and Lake Superior possess a mixture of clouds, lake ice, and snow over lake ice, but it is difficult to distinguish these features from one another in the visible image alone. Detection of snow cover is straightforward over some land regions in the scene, but is more difficult in other areas. For example, bright gray shades over northern Wisconsin and Michigan strongly suggest snow cover to an experienced analyst, but other darker areas (e.g., near and north of the Canadian border) are easy to misinterpret as being snow free. The low visible reflectance of snow cover in these northern regions is due to several factors. First, forests often exhibit lower reflectance values relative to tree-sparse regions, owing to dark coniferous foliage cloaking much of the surface snow cover. Second, the microphysics associated with old snow cover (snow that has undergone some melting and subsequent refreezing, reducing its polycrystalline structure) is a less efficient reflector of sunlight, relative to freshly fallen snow cover (Dirmhirn and Eaton 1975).

Normalization of the solar-reflecting channels by the cosine of the solar zenith angle enhances the signal in poorly illuminated polar regions, enabling the snow/cloud enhancement to confirm snow cover here. In addition, the enhancement distinguishes between lake ice over Lake Superior and cloud streets flowing southward from polynyas (areas of open water surrounded by ice that behave as point sources of heat/moisture and incipient cloud formation). The enhancement depicts lake ice in a white/blue tonality over parts of the Great Lakes, and the northern quarter of Lake Michigan is seen to be ice-covered and relatively cloud-free. The southern three-quarters of Lake Michigan is mostly ice-free, with additional lake-effect cloud streets oriented along the direction of the low-level winds. The distribution of snow and ice corresponds closely to that depicted in the corresponding NOAA/NESDIS snow map product. The NESDIS product represents inputs from visible, infrared, and microwave satellite data, surface observations, and some numerical model information. However, because the NESDIS product does not show cloud cover, the comparison should not be considered as being one to one. The tactical relevance of simultaneous high-resolution cloud and snow/ice information to military planning is a key motivator for the current products.

Department of Defense applications over southwest Asia

The tactical relevance of simultaneous high-resolution cloud and snow/ice information to military planning is a key motivator for the current products. The U.S. military made use of snow/cloud products in and around Afghanistan in the winter of 2001/02 during Operation Enduring Freedom (OEF), as well as Operation Iraqi Freedom (OIF) in Iraq in the winter of 2002/03. Near real time (within 2–3 h of data collection) MODIS digital data were made available to the Naval Research Laboratory (NRL) through coordination between NOAA, the National Aeronautics and Space Administration (NASA), and Department of Defense agencies. During these periods, NRL assumed the role of a semioperational production hub. The derived snow/cloud products were served to forward-deployed users through a customized Web interface (“Satellite Focus”; Miller et al. 2004a, manuscript submitted to Int. J. Remote Sens.) in cooperation with Fleet Numerical Meteorology and Oceanography Center (FNMOC).

Northeastern Afghanistan, and particularly the rugged terrain of the Central Highlands, Hindu Kush, and Tora Bora regions, presented harsh operating conditions during the winter months of OEF that included heavy snow cover, valley fogs, and a variety of topographically forced clouds. The challenges were mitigated in part by the near-real-time availability of cloud/snow discrimination products, as illustrated in Fig. 5 (version-2 example). MODIS products provided U.S. Navy and Marine Corps analysts with high-resolution detail over an array of complex topographical regions. The availability of such detailed information in this geographical setting was unprecedented in military operations and was used extensively in mission support.

During the OIF conflict these cloud/snow products provided near-real-time characterization of the northern border of Iraq (a region closely monitored during the period leading up to the invasion, when a northern front was still being considered). A comparison between the version-1 and version-2 methods in reference to a true color rendition for this domain is shown in Fig. 6. The portability of these cloud/snow products to domains worldwide makes them an important component in support of environmental characterization and situational awareness. While the products supporting the Middle East conflicts were made available exclusively to U.S. military users, the Naval Research Laboratory (NRL) demonstrates similar products in near–real time over the continental United States (CONUS) on the “NexSat” Web site (see information online at http://www.nrlmry.navy.mil/nexsat_pages/nexsat_home.html; see Miller et al. 2004b, manuscript submitted to Bull. Amer. Meteor. Soc.), in connection with the National Polar-Orbiting Operational Environmental Satellite System (NPOESS).

Limitations

The snow/cloud techniques that are presented here are best described as multispectral enhancements of clouds and snow cover, as opposed to quantified masks. The intended end user of these products is, therefore, the human analyst. The imagery encodes value-added information into distinct colors, while retaining the spatial resolution detail of the original data, enabling the analyst to distinguish important features, such as snow cover, lake ice, and cloud, reliably within a complicated scene. These physically based false-colored composite techniques combine scene constituents in an intelligent way to reveal the features of interest, while eliminating the tedious task of comparing images from different channels.

The version-1 enhancement does not detect patchy snow cover or trace amounts well, partially explaining its somewhat reduced snow cover relative to the coincident NESDIS products. The problem is that enhancement can incorrectly analyze the borders of snowfields as clouds. The NESDIS products perform better in this regard, supplemented by passive microwave satellite data, surface data, station data, and the previous day’s snow cover map (Ramsay 1998).

The version-2 enhancement, which does not enlist snow/no-snow criteria, produces far fewer artifacts of this nature. However, very high elevation snow-covered terrain will sometimes appear anomalously as light magenta. Here, the CIR channel produces an enhanced contribution from these bright targets because of reduced water vapor paths and incomplete solar attenuation despite the threshold corrections invoked in Eqs. (16) and (18). Some very bright land surfaces (e.g., deserts) may appear erroneously as low clouds, although the high skin temperatures that are often associated with these regions generally compensate for this effect in the red color-gun term [Eq. (16)]. For lower-/midtropospheric clouds that are still in the water phase, the differential between the red and green gun magnitudes is smaller than for very low clouds (because of relatively smaller TIRnorm contributions), resulting in a stronger yellow tonality.

Whereas geostationary imagery improves the ability to distinguish between cloud and snow cover through high temporal resolution (e.g., snow fields are relatively stationary in loops in comparison with most clouds, aside from cases of rapid snowmelt), the low-earth-orbiting (LEO) satellites currently possess sensors having typically higher spatial and spectral resolution than their geostationary sensor counterparts—attributes that are critical to detailed imagery applications. However, next-generation geostationary satellites will have the necessary spectral resolution (and increased spatial resolution) to employ variants of the techniques described here. The improved ability to characterize the full extent of a snowfield via temporal cloud-clearing techniques promises to be a boon for applications of this kind.

Conclusions

Feedback from military users confirms the usefulness of these products in remote forecasting regions. In maintaining the original spatial resolution of the imagery, these enhancements are most useful in situations of finescale cloud/snow variability (e.g., in the narrow mountain valleys of Afghanistan). The ability to distinguish snow cover from clouds on a single image allows military analysts to better answer such key environmental characterization questions as the following:

  1. Is a particular area in a data-sparse/-denied region snow covered, snow free, or obscured by clouds? (This is a question that is often impossible to answer by way of conventional visible and infrared single-channel imagery, and difficult is to ascertain at reduced spatial resolution snow masks).
  2. Where has snow recently fallen, and where has it melted?
  3. In conjunction with topographical maps, what is the approximate snow line elevation in mountainous regions?
  4. In nonmountainous regions, where are the geographical boundaries of the snowfield?
  5. Are the highest peaks of mountain ranges cloud covered?
  6. Is a particular surface target located in a snowfield obscured by low-level and/or high-level clouds?
  7. Is an overwater region ice covered, ice free, or obscured by high/low clouds?
  8. How might melting over an observed snowfield progress over time given the observed cloud cover?

These techniques are readily applicable to any sensors possessing the requisite spectral information. The 1.38-, 1.6-, and 2.2-μm channels (enlisted by the current work) will be included on the GOES Advanced Baseline Imager (ABI) on board the GOES-R series, scheduled for first launch in 2012. Loops made from sequences of geostationary-based snow/cloud products improve the monitoring, for example, of rapid snowmelt conditions, and the identification of snow boundaries under partly cloudy conditions. Over Europe, similar products are possible from the Meteosat Second Generation (commissioned as Meteosat-8; see Schmetz et al. 2002), although it does not carry the 1.38-μm “cirrus” channel that is enlisted in the MODIS-equivalent versions of the current enhancements. Numerous forthcoming low-earth-orbiting satellites will be capable of producing similar enhancements over the next decade, including two European Space Agency (ESA) Meteorological Operational Satellite (METOP) satellites, two NOAA AVHRR satellites, and four NPOESS Visible/Infrared Imager Radiometer Suite (VIIRS) satellites [including the NPOESS Preparatory Project (NPP), scheduled for launch in 2006].

Acknowledgments

The support of the research sponsors, the Oceanographer of the Navy through the program office Space and Naval Warfare Systems Command, PMW-150 under program element PE-0603207N, and the Office of Naval Research under program element PE-0602435N is gratefully acknowledged. We also thank our NOAA–NASA partners in the Near Real Time Processing Effort for providing Terra and Aqua datasets that are critical to the operational support of the DoD, as well as the Space, Science, and Engineering Center (SSEC) at the University of Madison—Wisconsin for providing direct-broadcast MODIS supporting this research.

REFERENCES

  • Ackerman, S A., , K I. Strabala, , W P. Menzel, , R A. Frey, , C C. Moeller, , and L E. Gumley. 1998. Discriminating clear sky from clouds with MODIS. J. Geophys. Res. 103:3214132157.

    • Search Google Scholar
    • Export Citation
  • Allen, R C., , P A. Durkee, , and C H. Wash. 1990. Snow/cloud discrimination with multispectral satellite measurements. J. Appl. Meteor. 29:9941004.

    • Search Google Scholar
    • Export Citation
  • Basist, A., , D. Garrett, , R. Ferraro, , N. Grody, , and K. Mitchell. 1996. A comparison between snow cover products derived from visible and microwave observations. J. Appl. Meteor. 35:163177.

    • Search Google Scholar
    • Export Citation
  • d’Entremont, R P., and L W. Thomason. 1987. Interpreting meteorological satellite images using a color composite technique. Bull. Amer. Meteor. Soc. 68:762768.

    • Search Google Scholar
    • Export Citation
  • Dirmhirn, I., and F D. Eaton. 1975. Some characteristics of the albedo of snow. J. Appl. Meteor. 14:375379.

  • Dozier, J. 1989. Remote sensing of snow in visible and near-infrared wavelengths. Theory and Applications of Optical Remote Sensing, G. Asrar, Ed., John Wiley and Sons, 527–547.

    • Search Google Scholar
    • Export Citation
  • Gao, B-C., , Y J. Kaufman, , W. Han, , and W. Wiscombe. 1998. Correction of thin cirrus path radiances in the 0.4-1.0 μm spectral region using the sensitive 1.375 μm cirrus detecting channel. J. Geophys. Res. 103:3216932176.

    • Search Google Scholar
    • Export Citation
  • Hall, D K., , G A. Riggs, , V V. Salomonson, , N E. DiGirolamo, , and K J. Bayr. 2002. MODIS snow-cover products. Remote Sens. Environ. 83:181194.

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

    • Search Google Scholar
    • Export Citation
  • Kidder, S Q., and H-T. Wu. 1984. Dramatic contrast between low clouds and snow cover in daytime 3.7 μm imagery. Mon. Wea. Rev. 112:23452346.

    • Search Google Scholar
    • Export Citation
  • Lee, T F., , F J. Turk, , and K. Richardson. 1997. Stratus and fog products using GOES-8–9 3.9-μm data. Wea. Forecasting 12:664677.

    • Search Google Scholar
    • Export Citation
  • McCrone, P J. 2003. Theater snow depth estimates for Department of Defense applications. Preprints, 12th Conf. on Satellite Meteorology and Oceanography, Long Beach, CA, Amer. Meteor. Soc., CD-ROM, JP1.16.

  • Miller, S D., , F J. Turk, , T F. Lee, , K. Richardson, , and J D. Hawkins. 2002. The enhanced role of the polar orbiter constellation in tropical system monitoring in the wake of a geostationary platform failure. Preprints, 25th Conf. on Hurricanes and Tropical Meteorology, San Diego, CA, Amer. Meteor. Soc., CD-ROM, P1.22.

  • Pilewskie, P., and S. Twomey. 1987. Cloud phase discrimination by reflectance measurements near 1.6 and 2.2 μm. J. Atmos. Sci. 44:34193420.

    • Search Google Scholar
    • Export Citation
  • Ramsay, B. 1998. The interactive multisensor snow and ice mapping system. Hydrol. Processes 12:15371546.

  • Robinson, D A., , K F. Dewey, , and R R. Heim. 1993. Global snow cover monitoring: An update. Bull. Amer. Meteor. Soc. 74:16891696.

    • Search Google Scholar
    • Export Citation
  • Robock, A., and D. Kaiser. 1985. Satellite-observed reflectance of snow and clouds. Mon. Wea. Rev. 113:20232029.

  • Romanov, P., , G. Gutman, , and I. Csiszar. 2000. Automated monitoring of snow cover over North America with multispectral data. J. Appl. Meteor. 39:18661880.

    • Search Google Scholar
    • Export Citation
  • Schmetz, J., , S A. Tjemkes, , M. Gube, , and L. van de Berg. 1997. Monitoring deep convection and convective overshooting tops with Meteosat. Adv. Space Res. 19:433441.

    • Search Google Scholar
    • Export Citation
  • Schmetz, J., , P. Pili, , S. Tjemkes, , D. Just, , J. Kerkmann, , S. Rota, , and A. Ratier. 2002. An introduction to Meteosat Second Generation (MSG). Bull. Amer. Meteor. Soc. 83:977992.

    • Search Google Scholar
    • Export Citation
  • Turk, F J., and S D. Miller. 2005. Towards improved characterization of remotely sensed precipitation regimes with MODIS/AMSR-E blended data techniques. IEEE Trans. Geosci. Remote Sens. 43:10591069.

    • Search Google Scholar
    • Export Citation
Fig. 1.
Fig. 1.

Normalized red/green/blue color-gun contributions to the version-2 snow/cloud algorithm and resultant composite tonality as a function of scene type. Points in the right panel are mean values of corresponding boxes in left panel; whiskers denote 1-sigma standard deviation.

Citation: Journal of Applied Meteorology 44, 7; 10.1175/JAM2252.1

Fig. 2.
Fig. 2.

VIS, SIR, CIR, and TIR imagery for a snow/cloud scene over Colorado and Wyoming as observed by Terra MODIS on 1800 UTC 22 Mar 2003.

Citation: Journal of Applied Meteorology 44, 7; 10.1175/JAM2252.1

Fig. 3.
Fig. 3.

Terra MODIS (left) true color, (middle) version-1, and (right) version-2 cloud/snow enhancements depict rapid snowmelt across Colorado and Wyoming snowmelt over the period 20–25 Mar 2003.

Citation: Journal of Applied Meteorology 44, 7; 10.1175/JAM2252.1

Fig. 4.
Fig. 4.

(left) Visible image, (middle) version-1 snow/cloud enhancement, and (right) corresponding NOAA/NESDIS snow product over the Great Lakes region depicting snow, cloud, and lake ice as observed by Terra MODIS on 1640 UTC 2 Mar 2003.

Citation: Journal of Applied Meteorology 44, 7; 10.1175/JAM2252.1

Fig. 5.
Fig. 5.

(left) Visible image and (right) version-2 snow/cloud enhancement over the Hindu Kush mountain range of northeastern Afghanistan as observed by Aqua MODIS on 0845 UTC 19 Feb 2003.

Citation: Journal of Applied Meteorology 44, 7; 10.1175/JAM2252.1

Fig. 6.
Fig. 6.

(left) True color, (middle) version-1 cloud/snow enhanced, and (right) version-2 cloud/snow enhancement over the northwestern Zagros mountains within Iraq, Iran, and Turkey as observed by Terra MODIS on 0825 UTC 10 Feb 2004.

Citation: Journal of Applied Meteorology 44, 7; 10.1175/JAM2252.1

Save