Asian Dust Detection from the Satellite Observations of Moderate Resolution Imaging Spectroradiometer ( MODIS )

Asian dusts exert significant influences on regional air quality, weather, and climate. Detection of these highly variable aerosol events is challenging due to several factors, such as short lifetime, multiple scales, and strong interactions with local and regional surface and meteorological conditions. Since dust particles can directly alter solar and earth radiation in both visible (VS) and infrared (IR) spectral regions through scattering and absorption processes, both VS and IR remote sensing techniques can be used to detect dust plumes in the atmosphere. A dust detection system for multi-channel satellite imagers was applied in this study. The detection is based on the analysis of reflectance (or radiance) in VS channels or brightness temperature (BT) in IR channels. The magnitude of the difference in reflectance and/or BTs in selected channels due to dust is used to infer the signature of the dust particles. Descriptions of the detection system and its application for Asian dust using the Moderate-resolution Imaging Spectroradiometer (MODIS) satellite measurements are provided. The performance of the algorithm for Asian dust detection and its usefulness for monitoring the outbreaks and dispersion of Asian dust events were emphasized in the current study.


INTRODUCTION
The mineral dusts in Asia exert large influences on regional air quality, hydrological and energy cycles, and ecosystems.Detection of these highly variable aerosol events is challenging because of: episodic features, short lifetimes, multiple-scales, and strong interactions with local surface and meteorological conditions.Dust particles can directly alter solar and Earth radiation in both visible (VS) and infrared (IR) spectral regions through scattering and absorption processes.Due to specific optical properties of dust particles, satellite observed radiances carry the spectral signatures of dust particles that are different from air molecular, cloud, and underlying surface.Based on the differences in the spectral signatures, various detection schemes have been developed to distinguish dust in different desert regions on the Earth surface.For examples, brightness temperature difference (BTD) between 10.5 μm and 12.5 μm channels from Temperature Humidity Infrared Radiometers (THIR) had been used for dust storm detection over land in the early seventeenth of 20th century (e.g., Shenk and Curran, 1974).Thereafter, various mineral dust detection algorithms have been developed using VS and/or near IR channels (Tanré and Legrand, 1991;Kaufman et al., 2000;Miller, 2003;Qu et al., 2006) and thermal IR wavelengths (Ackerman, 1989;Legrand et al., 1989;Ackerman, 1997;Wald et al., 1998;Legrand et al., 2001;Darmenov and Sokolik, 2005;Evan et al., 2006;Verge-Depre et al., 2006;Hao and Qu, 2007) or their combinations (Roskovensky and Liou, 2005;Hansell et al., 2007;Zhao et al., 2010).Uniformity test of the scene has also been shown to be useful for discriminating homogeneous dust layers from clouds (Martins et al., 2002), especially over ocean.
In practice, image-based detection is based on the analysis of reflectance (or radiance) in VS bands or brightness temperature (BT) in thermal IR bands.The magnitude of the difference in reflectance or BT in selected bands (or channels) due to dust effect can be used to infer the signature of dust particles.This is the essence of aerosol imagery detection algorithms.However, it is impossible to distinguish cloud from dust without ambiguity in a global application for image-based detection algorithms with fixed detection thresholds as demonstrated by the validation using active lidar measurement (e.g., Cho et al., 2012).Therefore, validation and evaluation of the dust detection criteria over different continents and dust scenarios should be pursued continually during the application and operation of the algorithms to eventually identify a set of optimal threshold values suitable for global applications.This paper is an effort along this line by introducing an image-based dust detection system for dust storm detection specifically in Asian regions using the Moderate-resolution Imaging Spectroradiometer (MODIS) satellite measurement in VS and IR channels.The rest of the paper is arranged as follows: Section 2 introduces the dust detection algorithm, detection results for several typical Asian dust cases are presented in Section 3, and summary and conclusions are given in the closing section.

DUST DETECTION SYSTEM
The image-based dust detection system used in this paper was initially developed by Zhao et al. (2010) for global dust and smoke detection from multi-channel satellite imagers.In this paper, we will apply the dust detection modules to Asian regions for two major purposes.The first is to check the performance of the detection system for Asian dust specifically and the second is to demonstrate the usefulness of the system in monitoring the outbreak and dispersion of dust storms in Asian regions.The detection decisions are summarized in a flow diagram in Fig. 1.Different detecting schemes are used for land and ocean (or water) and the detection is performed only for daytime (defined as solar zenith angle < 80 degree).

Satellite Observations
Calibrated/navigated pixel level MODIS satellite reflectance (or radiance) and brightness temperature in 0.47, 0.64, 0.86, 1.38, 2.26, 3.9, 11.0, and 12.0 μm channels, observational geolocation (latitude/longitude) and geometric information, and sensor quality flags (together named as level-1B data) are used as the input of our detection system.Some ancillary information, including day/night flag, surface snow/ice flag, and sun glint flag contained in the standard MODIS cloud mask product, are also needed as input.MODIS cloud mask flags are not used due to that MODIS cloud mask has the tendency to misclassify thick dust as cloud (Martins et al., 2002;Frey et al., 2008).The detection tests are implemented sequentially.To balance the efficiency and memory requirement, a block of scanning lines are read into a RAM buffer together instead of reading data pixel by pixel.The output of the detection system for valid pixels is a single digital index (or flag): 0 (no dust and smoke), 1 (dust), 2 (smoke).For invalid pixels, such as those with missing observational geolocation or geometry information in the level-1B data, a specific digital index (3) is set in the output to indicate no detection is performed for the pixels.This paper discusses solely the dust detection algorithm (module).

Detection Criteria over Land
The detection criteria are a series of tests in the reflectances (R) or their ratios (Rat) in VS channels and brightness temperature (BT) differences (BTD) in IR bands, which are used to separate the spectral and spatial signatures of dust from surrounding cloud and underlying surfaces.For examples, the detection tests of 1) and 2) over land (see Fig. 1) are used to remove bad observations and cloudy pixels, respectively.For the subsequent test 3), if BT 3.9µm -BT 11µm ≥ 25 K is satisfied, then flag the pixel as dust laden.This is because dust reflects solar energy at 3.9 µm so it increases the brightness temperature difference with BT 11μm .Moreover, MNDVI < 0.08 (MNDVI = NDVI 2 /(R 0.64µm × R 0.64µm ); NDVI = (R 0.86µm -R 0.64µm )/(R 0.86µm + R 0.64µm )) and Rat 2 > 0.005 (Rat 2 = (Rat 1 × Rat 1 )/(R 0.47µm × R 0.47µm ); Rat 1 = (R 0.64µm -R 0.47µm )/(R 0.64µm + R 0.47µm )) are added to identify dust over some semi-arid surfaces with sparse vegetation.This is because dust absorbs at blue wavelengths and appears visually to be brownish in color.Clouds are spectrally neutral and appear white to our eyes.For this reason, the reflectances at 0.86, 0.64 and 0.47 µm have been used to identify dust.This is often done in a ratio of one to another (e.g., Roskovensky and Liou, 2005) or as a normalized difference index (such as MNDVI or Rat 2 in Zhao et al. (2010)).In the ratio tests, we square the reflectances trying to take advantage of enhanced non-linear behavior.
Two tests are further used to confirm the presence of optically thick dust in test 4).The first is BT 11µm -BT 12µm ≤ -0.5 K and BT 3.9μm -BT 11μm ≥ 25 K.The bulk transmittance of many aerosols displays a strong spectral variation in the 10-12 μm window regions.Thus, IR split window techniques have been developed at 11 and 12 μm to detect volcanic aerosols, particularly those from sulfur-rich eruptions (Prata, 1989;Barton et al., 1992), and dust outbreaks (Legrand et al., 1992(Legrand et al., , 2001;;Evan et al., 2006).Dust has a larger absorption at 12 µm than at 11 µm, so that dust plumes generally have a higher emissivity and lower transmissivity in the 12 µm channel (Ackerman, 1997;Dunion and Velden, 2004), which causes BT 11µm -BT 12µm to become negative.On the other hand, there is absorption and emission of water vapor in the 11 and 12 µm channels and the water vapor weighting function for the 11 µm channel peaks lower in the atmosphere (higher in temperature) than the 12 µm channel does, which results in a positive BT 11µm -BT 12µm for clear-sky atmosphere.However, the presence of a dry air mass, often associated with dust storm events, tends to reduce the positive BT 11μm -BT 12μm values in clearsky atmospheres due to the altitude coincidence of dry air with the peak of the 11 µm water vapor weighting function.For elevated thick dust plumes, the negative BT 11µm -BT 12µm values due to strong dust absorption/emission effect surpass the corresponding water vapor related positive values weaken by the elevated dry/dusty air layer.As a result, BT 11μm -BT 12μm difference has a negative value for thick dust plumes.The second is R 1.38μm < 0.035 and MNDVI < 0.2.Since low level clouds (often towering cumulus) can also have a negative split window brightness temperature difference.These two additional tests are used to separate dust from towering cumulus according to our analysis (Zhao et al., 2010).

Detection Criteria over Ocean
Similar to over land, four sets of detection criteria are used for over ocean.Since dust over ocean is more uniformly distributed than cloud and less reflective than cloud, uniformity examination based on standard deviation of reflectance (StdR) on 3 × 3 pixels (StdR 0.86µm ≤ 0.005) and reflectance check (R 0.47μm ≤ 0.3) are added to BTD test (4 K < BT 3.9µm -BT 11µm ≤ 20 K) in criterion 2) to better separate dust from cloud over ocean.
There are three separate tests for dust over water in criterion 3).Any pixel that passes any of the three tests is flagged as dusty, although some of the tests have multiple conditions that must be satisfied.The first test is BT 11μm -BT 12μm < 0.1 K and -0.3 ≤ NDVI ≤ 0. Since the split window difference BT 11μm -BT 12μm < 0.1 K can also lead to false test so it is coupled with a NDVI-type condition to identify dusty pixel.Dust particles exhibit more uniform scattering across 0.64-0.86μm spectral region than smaller aerosols, like smoke, due to their relatively large particle size.Thus R 0.86μm /R 0.64μm ratio tends to be reduced for dusty pixels and has been found useful in discriminating pixels containing smoke from those with dust (Zhao et al., 2010).Rather than directly using the ratio, a modified version (or NDVI type test) is employed here.Another test is the requirement of R 0.47μm /R 0.64μm < 1.2 since clear-sky ocean surface tends to reflect much more in 0.47 μm than 0.64 μm compared with dust particles.Similar to the dust detection over land, low level clouds (often towering cumulus) can also have a negative split window brightness temperature difference.Therefore, the third test (BT 3.9μm -BT 11μm > 10 K and BT 11μm -BT 12μm < -0.1 K) are applied to attempt reducing cloud contaminated pixels.
Two sets of conditions are also used to identify the presence of optically thick dust over ocean as over land.The first set BT 3.9µm -BT 11µm > 20 K is used to define thick dust regime by separating from thin dust.The second set includes BT 11µm -BT 12µm ≤ 0 K, which is the ocean version of the split window IR detection technique for heavy dust, and -0.3 ≤ NDVI ≤ 0.05, which is used to reduce the false detection in the split window technique as for the detection of non-thick dust.
The threshold values for the tests discussed above were obtained through analysis of many training images over the globe (Zhao et al., 2010).Validation of the detection criteria using CALIPSO (Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations) lidar measurement has been performed and results have been submitted in a separate paper (see Cho et al., 2012).The validation results indicate image-based detection techniques have relatively poor performance in separating thin dust layers from both clear and cloudy pixels comparing to CALIPSO active remote sensing technique.Our multi-channel detection criteria tend to be conservative for dust detection so that it works better for thick dust plumes.This paper further examines the validity of the detection criteria for Asian dust scenarios through the comparison of dust features in images.

RESULTS
The Gobi desert is located in north-, northwestern China and southern Mongolia.The desert is also far less sandy than other deserts.Instead, the desert floor is mostly bare rock, due in most part to the high winds that whip across the plateau.Gobi desert is expanding in a process known as desertification.It is expanding into China's grassland.Dust storms have increased heavily damaging agriculture and degrading air quality.Our first detection example is shown in Fig. 2 for a dust storm over the Gobi desert on May 11, 2011.The dust plumes in red-green-blue (RGB) false color images from MODIS Terra (top left) and Aqua (bottom left) are clearly seen in brownish color over the desert areas.The corresponding dust plumes detected by our system are shown in red color in the two panels on the right hand side.The dispersion features of dust storm displayed in the RGB images have been captured very well by our dust index.
Fig. 3 shows a dust storm over the east coast of China on May 1, 2011.MODIS RGB images are displayed on top three panels and our corresponding dust detection results are shown on bottom three panels.The light brownish color widely spread over the Yellow Sea in the MODIS RGB images are well captured by red color dust index in our detection plots.
Degraded visibility and poor air quality near the surface are common phenomena associated with the outbreak of Asian dust storms.This is also the reason that dusty weather has caught equal public concerns as polluted haze weather in Asia.

SUMMARY AND CONCLUSIONS
The satellite image-based dust detection system developed for global dust detection and monitoring has been applied to Asian regions for detecting Asian dust storms from MODIS satellite observation in this paper.The detection relies on spectral and spatial threshold tests along with some uniformity texture examinations by using MODIS level-1B reflectances (or radiances) and brightness temperatures.The detection product (dust index) produced for individual satellite observational pixels can reproduce and capture the heavy elevated dust plumes very well over both land and coast ocean.The results suggest the detection criteria determined from global dust analysis for both land and ocean can also be applied for Asian dust detection.The algorithm is useful for monitoring the outbreaks and dispersion of Asian dust storms.Due to the relatively weak aerosol signal and large uncertainties associated with a bright surface, current threshold based detection algorithm can miss the detection of thinner outer edges of widely dispersed dust plumes.It also does not work for the snow/ice surface due to associated strong surface perturbation on the dust signal.The algorithm has also been evaluated and validated by comparing with the CALIPSO lidar measurement (Cho et al., 2012), which reveals the detection algorithm is conservative for dust detection -tend to have less misclassification but easy to miss the detection of a thin dust layer.This conclusion is consistent with the result from the current study based on image comparisons.Nevertheless, the detection algorithm can be applied to any satellite multi-channel imagers with proper detection channels at (or close to) 0.47, 0.64, 0.86, 1.38, 3.9, 11.0, 12.0 μm.
There are some studies and improvement of the algorithm that will be pursued in the future, including 1) applying the detection system to other dust active continents/scenarios and other multi-channel satellite imagers for further evaluation from more applications, 2) improving the detection of thinner dust layer by developing alternative detection algorithm, 3) exploring dust detection at night using brightness temperature only.

Fig. 1 .
Fig. 1.Flow chart of dust detection system for land (left) and ocean (right).

Fig. 3 .Fig. 4 .
Fig. 2. MODIS RGB images (left) and the corresponding dust index plots (right) for a dust storm over Gobi desert on May 11, 2011.Brownish color in the RGB images and red color in the index plots indicate dusty area.