NRL Monterey, Marine Meteorology Division
|MODIS True Color||MODIS Dust Enhancment: Dust Appears Pink|
|The consensus of numerous Navy Meteorology/Oceanography (METOC)
post-deployment reports from Operation Enduring Freedom (OEF) holds desert
dust accountable for the most common and significant adverse impacts to
operations. Specifically, these storms impaired visibility, obscured
targets, and rendered laser-guided weaponry ineffective. Satellite-based
detection of dust is a difficult problem due in part to observing-system
limitations. An unprecedented interagency coordination in support of the
War on Terror mitigated this problem by making available a global, near
real-time dataset from the Moderate Resolution Imaging Spectroradiometer
(MODIS). NRL designed a novel technique for enhancing dust over water and
land, leveraging previous capabilities with these new high spectral/spatial
resolution MODIS data. |
Satellite-based dust detection over the desert terrains characterizing much of the OEF domain requires an observing system capable of extracting as much dust-specific information from the scene as possible. The current approach improves upon previous methods in this regard through use of high spatial and spectral information available from MODIS, a state of the art instrument flown aboard the recently launched NASA Earth Observing System (EOS) Terra and Aqua satellites. This training begins with an overview of why dust detection is possible, followed by a brief description of MODIS and the algorithm designed to exploit the advantages of this sensor. Several examples illustrate dust storm enhancement capabilities, with additional applications to smoke and volcanic plume detection.
|Conceptual Diagram of Cloud/Dust Scattering of Sunlight|
|Physical Basis |
The fundamental principles of dust detection are spatially, spectrally, and temporally based. The idiosyncrasies of dust within these paradigms are what allow for its decoupling from other components of the complex scene. We discuss some of these elements here.
1) Spatial Information
Moderate levels of dust over land cause visual blurring of otherwise sharply definable surface features. This so-called " adjacency effect " has been used with success in dust detection, although the method faces inherent challenges over laminar desert backgrounds where the propensity for contrast reduction is reduced. At visible (VIS) wavelengths, highly reflective dust contrasts against darker backgrounds such as vegetation or bodies of water. At infrared (IR) wavelengths, depressed brightness temperatures from elevated (and cooler) dust over hot surfaces yield similar contrast information. Researchers have demonstrated enhancements based on these VIS/IR spatial contrast principles. Because the corresponding signatures of clouds are similar to dust, an implicit requirement for distinction is recognition of differences between the cloud and dust spatial structures. This inspection is not always straightforward, particularly in the case of thin cirrus clouds that spatially appear very similar to dust.
2) Spectral Information
An important spectral property allowing for dust detection in the IR is the " split-window " difference. The signature arises from the higher spectral absorption of dust sensed at 11.0 micrometers (mm) compared to measurements at 12.0mm. This spectral behavior contrasts liquid and ice clouds, where the opposite absorptive properties hold. The split-window signature is less pronounced for thick dust very close to the surface, where transmission effects are smaller. Additionally, some land surfaces produce ambiguous signatures in the absence of dust.
Research has shown that mineral dust becomes increasingly absorptive with decreasing VIS wavelength (i.e., progressing from red toward blue light), corresponding to a monotonic increase in the complex part of the imaginary index of refraction (ni). These properties result in dust coloration, as illustrated conceptually in figure above. Dust appears as shades of yellow/orange due to its preferential absorption of blue light, whereas clouds (having small and relatively invariant ni across the VIS) yield shades of gray to white depending on the strength of illumination. The reasoning behind why the color yellow is the outcome of preferential blue light absorption is explained in the context of red/green/blue composites further along. Exploiting the dust coloration properties for the purpose of enhancement requires a radiometer with spectral resolution sufficient to partition the VIS spectrum into the required components.
Miller (2002) applies the above principles to the Sea-viewing Wide Field-of-view Sensor (SeaWiFS), which features eight narrowband channels across the VIS spectrum. The technique uses a “normalized dust difference index˙ (NDDI) between the red (670 nanometer (nm)) and blue (443nm) channels of SeaWiFS, defined such that dust produces large positive differences while clouds produce relatively smaller differences. Normalization ensures enhancement of weak dust signals over dark backgrounds. The fundamental shortcoming of this method is that land areas are also enhanced, since their spectral properties at VIS wavelengths are similar to those of dust. Hence, it is a method limited almost exclusively to over-water applications.
3) Temporal Information
The ability to observe and track feature motion by looping consecutive satellite images is very useful for dust detection. For this reason, many applications consider only geostationary satellite imagery (typically having 30-minute refresh). Unfortunately, given current instrumentation limitations on the geostationary platform, the trade-off for this high temporal resolution is reduced spatial and spectral information. The featured work concentrates on the superior dust detection capabilities of low-earth-orbit (LEO) high spatial/spectral resolution sensors, with multiple overpasses from various LEO satellites serving as a possible temporal surrogate.
The Current Approach
The new NRL dust enhancement combines elements of previous methods with the spatial/spectral advantages of MODIS to form a novel, unified product. Owing to the very different restrictions between water and land backgrounds, two distinct algorithms operate on their respective backgrounds as determined by a land/sea database. Careful scaling minimizes discontinuities of enhanced dust crossing coastal (algorithmic) zones.
The algorithm uses 7 of the 36 available MODIS channels to exploit the spatial and spectral contrast features of dust. Listed in terms of [channel index; central wavelength; native spatial resolution; description] and in order of increasing wavelength, they are as follows: (3; 469nm; 500m; blue), (4; 555nm; 500m; green), (1; 645nm; 250m; red), (2; 853nm; 250m; reflective IR), (26; 1.38mm; 1km; short-wave water vapor), (31; 11.0mm; 1km; IR clean window), and (32, 12.0mm; 1km; IR dirty window). The over-water algorithm uses MODIS channels 2, 3, and 4, and the over-land algorithm enlists all seven channels listed above.
The relative ease of dust detection over water (aside from conditions of shallow water or heavy alluvial/biological material suspension) using VIS spatial/spectral contrasts allows for use of a more aggressive method geared toward enhancing fine details of tenuous dust features. Accordingly, the over-water algorithm adopts the NDDI technique of Miller (2002) described previously. An important part of this processing is the molecular scatter correction applied to the VIS channels, based on radiative transfer simulations computed offline and stored in look-up tables. The correction reduces limb-brightening effects that would otherwise wash out the imagery on the swath edges.
Detection of dust over bright land backgrounds such as deserts is far more difficult than over water, and requires additional IR information to separate the dust signal from the other components of the scene. The premise for the over-land enhancement is threefold: i) elevated dust, having cooled to its environmental temperature, produces depressed IR brightness temperatures against the hot skin temperature of the land background; ii) this cool layer of dust is distinguishable from a cloud with the same radiometric temperature using the NDDI technique; and iii) split-window differencing reveals dust. In the three-color composite dust enhancement, the blue and green color guns contain information from the corresponding blue/green channels of MODIS, and the red gun contains a weighted combination of items (i) (iii). Statistical composites from many dust storm cases provided the optimal scaling and weighting coefficients. The short-wave water vapor channel (1.38mm) provides filtering of cirrus clouds whose cold IR temperatures sometimes cause ambiguity.
|The Advantage of Color Composites|
|An advantage of this product over panchromatic (black/white) imagery is the
ability to highlight dust in colors that contrast strongly against other
components of the scene that are not of interest. This is accomplished by
way of feeding the enhanced information into a red/green/blue color
Therefore, a prerequisite for understanding the appearance of the dust enhancement is a familiarity with the three-color composite technique. The concept is as follows: three primary colors (red, green, and blue) form the axes of a " color cube". In the example of an 8-bit computer display, brightness magnitudes for each primary color range from 0 (black) to 255 (full red, green, or blue saturation). This forms a cube of dimension 256 by 256 by 256 elements whose indices map to a discrete representation of all possible colors, based on varying combinations of the primary color brightness values.
The figure above depicts the appearance of the outer surfaces of such a cube, showing how the primaries combine and transition across the three-dimensional color space. This provides a suitable framework for visualizing how the various components of VIS light combine to form all the colors we see. White light is the combination of full red/green/blue saturation, while yellow tonalities arise from high values of green and red with low amounts of blue. Extending this to the real-world example of dust illuminated by a source of white light (the sun), the removal of blue light (via dust absorption) results in the preferential scattering of yellow light, explaining the observed color of dust.
Creation of a " true color " composite from satellite imagery requires sufficient spectral resolution to separate the red, green, and blue components of visible light. After applying atmospheric corrections and appropriate scaling, these channels become the respective red, green, and blue indices (sometimes called ôcolor guns¤) of the color cube described above. The result is an image with an appearance similar to what we would observe with our own eyes. For the dust enhancement, a multi-spectral channel combination designed to enhance dust replaces the red gun of the true color product. The effect is for dust to possess relatively brighter red tonality throughout the enhanced imagery. The non-dust components appear as red-depleted (e.g., cyan clouds, green land) and generally darker tones, so as to focus attention toward the dust features of interest.
|Shallow and/or Sediment-Rich Water Effects||Dry Lakebed and Cold Topography Effects|
|The daytime-only MODIS dust enhancement is not without its own assortment
of interpretive caveats. |
1) The over-water component has the undesirable effect of enhancing the region of sun glint (near-specular reflection of the solar disk off water surfaces) as dust. The current product flags a glint zone based on a prescribed minimum glint angle. This region is replaced by true color imagery (example and discussion in the examples section of this training).
2) High concentrations of sediments suspended in water (e.g., associated with river runoff) occasionally also give rise to false dust enhancements. In other cases, these regions will also show up as shades of cyan or green, similar to the true color product.
3) Over land, cold surfaces may appear falsely enhanced due to the similar VIS/IR properties of thick dust. This problem is particularly severe during the winter months, when extremely cold surfaces may show up as red. Included in the over-land algorithm is a terrain elevation database to filter out a subset of these effects.
4) Depending on surface temperature, an optically thick layer of dust (producing small split-window differences and IR contrast signatures) near the ground may go undetected, a characteristic observed most commonly in close proximity to strong point-sources of dust. The scaling thresholds and weighting coefficients of the IR enhancement components require seasonal tuning, mainly to reduce false enhancements emerging during the cold winter months. Inspection of imagery can more often than not distinguish false land signals from dust plumes.
5) The stronger thermal contrasts between dust and the background observed by Aqua (1330 overpass) compared to Terra (1030 overpass) translates to generally better performance of the enhancement for Aqua. Future modifications to the product may attempt to normalize this difference.
6) Due to increased optical paths at higher sensor zenith angles, dust on the edge of the satellite swath exhibits a stronger (brighter) enhancement than the same dust observed near satellite nadir. While brighter pink tonalities do correspond to higher dust opacity over local regions (e.g., of several hundred km), the current product does not quantify this opacity in terms of optical depth, particle size, slant range visibility, or laser attenuation.
7) This is a qualitative dust identifier, and a first logical step toward physical retrievals on the subset of pixels thought to be dust.
With these limitations in mind, we proceed to highlight the strengths of the enhancement through a series of examples.
|Dust Impacting Aviation: Egypt Air Case Study|
|On May 7th, 2002, an EgyptAir passenger aircraft crashed into a hillside
during an emergency landing attempt near Tunis, Tunisia, killing 18 of the
60 people on board. Local weather at the time of the accident included
fog, rain, and blowing sand from the Saharan Desert interior. NRL dust
products supported the National Transportation Safety Board during its
investigation of this accident. |
Near the time of the crash, Terra MODIS collected the imagery shown in the above figure. The true color product (left panel) shows the city of Tunis obscured by a squall line. A veil of dust fans into the Mediterannean Sea, carried northward by strong southerly winds associated with the advancing storm system. Subtle adjacency effects indicate additional inland dust westward of the squall line. The corresponding enhancement (right panel) reveals considerable inland dust throughout the cloud-free regions of the storm. Dust areas appear as shades of pink, clouds are cyan, and land areas free of overlying dust are green. Since the enhancement cannot detect dust obscured by cloud (although dust-over-cloud generally is detectable) only the inference of its presence near the Tunis crash site was possible, based on its observance on either side of the squall line.
|Distinguishing Between True and False Dust Fronts|
|Included among the many caveats to dust imagery analysis are sudden transitions in vegetation regimes that sometimes masquerade as false dust fronts. The Indus River valley of Pakistan features many such –desert-meets-oasis± discontinuities. Annotated in the true color image of the figure above are two possible dust fronts in advance of a storm system crossing central Pakistan. The dust enhancement reveals that front A (yellow dotted line) is in fact a vegetation boundary, while front B (red dotted line) is a true dust front. Also evident in the enhancement are additional fronts not obvious in the true color imagery.|
|Detecting Thin Dust Over Bright Land Backgrounds|
|This example highlights another Southwest Asia domain application of the dust product. Strong northerly winds in the wake of a passing storm system carry dust from the deserts and dry lakebeds of eastern Iran, southern Afghanistan, and Pakistan. Winds channeled through coastal valleys spawn numerous dense dust plumes, clearly visible in the imagery as they stream into the NAS. Brought to our attention by the corresponding dust enhancement are several additional plumes over land, previously washed out by the bright topography. The solid, sharply defined pink object near the inland plumes is a dry lakebed, enhanced for reasons explained below.|
|Analysis of Dust Source Regions|
|Knowledge of the likely source regions for dust under certain environmental forcing conditions is critical to improving dust-forecasting models. Strong northerly winds channeled into the Margow Desert basin of southwestern Afghanistan produce the recurring dust pattern observed above. Several plumes originate from dry lakebeds noted in the imagery. With high concentrations of fine (and easily lifted) silt deposits, these areas often serve as point sources for dust. The lakebeds are enhanced in the dust product by applying the over-water (NDDI) algorithm to the dry land pixels.|
|Volcanic Ash Plume Enhancement|
|Volcanic aerosol poses an underemphasized and poorly understood hazard to
turbine engine-powered aircraft. The danger lies in the very small
glass-like particles constituting volcanic ash, which melt readily as they
come into contact with certain parts of the hot engine interior.
Subsequently they re-solidify, clogging air intakes and damaging turbine
blades to the point of engine failure. Commercial airline pilot accounts
from near-tragic encounters with volcanic ash (steep-dive recoveries from
complete engine stalls) noted difficulties in visually discerning the ash
haze from the benign cirrus clouds commonly encountered at those same
flight levels. |
The same physics that allows for mineral dust detection in the current product applies to some varieties of volcanic ash plumes. The recent eruptions of Mt. Etna, on the island of Sicily, provided a glimpse into these capabilities. The example above illustrates the detailed structure of the volcanic plume as upper-level winds carry it southward across the Mediterranean Sea. The level of detail, combined with the high sensitivity of the NDDI to low concentrations, suggests some utility for the dust enhancement in this regard.
|Oil Fire Smoke Plume Detection|
|During the Gulf War, Iraqi troops set fire to numerous oil wells in Kuwait. Pitch-black smoke issuing from these fires cloaked the Arabian Gulf skies, impacting Naval flight operations. The difficulty in satellite monitoring of such plumes over dark (low VIS contrast) water backgrounds warrants examination of the dust enhancement¸s utility in this capacity. An opportune MODIS overpass captured a small and short-lived plume emerging from an oil fire in southern Kuwait. Pending a detailed analysis, the preliminary results shown in this example indicate that properties of these plumes are conducive to their enhancement both over land and water under the existing method.|
|Fine Detail Revealed in Dust Plume|
|The MODIS dust enhancement allows for interpretation of fine-scale features of dust over bright land backgrounds. In the example above, meso-scale wave structure within the dust plume is revealed.|
|Treatment of the Sunglint Zone|
|Since the dust enhancement experiences difficulties in the region of potential sunglint (where the sun reflects directly to the satellite off the water surface), true color imagery is replaced in this region. The glint zone is demarcated by a two-toned (yellow/red) line, with the red-side of the line denoting the region of potential glint contamination. The actual expanse of the glint zone is dependent on surface roughness (rougher surface = larger glint zone). However, the current glint zone is fixed. Therefore regions inside the glint zone may actually be free of glint, and at other times the glint zone may extend beyond this zone.|
|Model Wind Overlays|
|It is sometimes very difficult to determine from single-shot imagery what
the motion of the features in the image are. This is especially true for
dust in suspension, where the air flow may vary significantly in both speed
and direction over the region spanned by the dust. Knowledge of wind speed
and direction can also assist in now-casting the arrival of a dust front
over a given location. In the wind-version of the MODIS dust product,
surface winds from the Coupled Ocean Atmosphere Mesoscale Prediction System
(COAMPS (TM)), are interpolated to the closest satellite time and overlayed
on the enhancement imagery.
One should be careful to note that while surface winds are most important for the initial lifting of dust into the atmosphere, they may not be representative of the steering winds for the dust as it advects downwind and potentially reaches higher levels in the atmosphere. Linear interpolation of model winds over time is only as valid as the linear assumption itself. Accelerations/decelerations of atmospheric flow between the analysis time and the interpolation time are not accounted for. Also, the interpolation of wind fields over time can be mis-leading in regions of strong wind shift. For example, consider the case of frontal passage over a given location. At time 0, winds are strong southeasterly, while at time 2 the front has passed and they are strong northwesterly (a 180 degree shift). Linear interpolation at time 1 would interpolate the wind to zero speed (and no associated direction). As a general rule, the longer the interpolation (tau) value, the more prone to uncertainty the wind fields are.
Author: Steve Miller
Last Updated: Fri Mar 28 15:27:20 2003
Produced by: The Composer (Ver: 1.4 )