1. Introduction
The study of atmospheric aerosols has become a very interesting topic in recent years due to evidence showing their impact on climate change [
1]. Aerosols, deserving of the same consideration as greenhouse gases, play important roles in atmospheric chemistry, cloud microphysics, temperature, and radiation balance in the lower atmosphere [
2]. Dust storms are one kind of frequently occurring natural phenomena over the continents, which may have potential impacts on the climate, environment, and air quality. Dust storms, usually occurring in arid and semi-arid regions, can carry large quantities of dust and move forward like an overwhelming tide to destroy crop plants, ruin mining and communication facilities, weather vestiges, damage small villages, reduce visibility, and hinder human daily activities, as well as impact aircraft and road transportation. It pollutes the atmosphere and air quality, influences cloud formation [
3], obscures sunlight, and alters temperature. Some dust storms can remain suspended in the air for several days and travel by wind far from where they originated. Recently, several studies have observed that heavy dust storms can impact the formation and evolution of hurricanes [
4,
5].
The detection of dust storms in a timely manner can serve as warnings for people to avoid economic loss and even loss of life. The detection results are also very useful for atmospheric modeling and simulation. Due to the large coverage of each dust storm, satellite remote sensing technology has been applied to detect and monitor dust storms. Color imagery techniques have been the primary tools used for dust storm monitoring in the past, where early researchers used the visible spectrum to monitor dust outbreaks as well as to estimate dust optical depth over oceanic regions [
6,
7]. Dust was detected and its evolution followed by its yellow color on Sea-viewing Wide Field-of-view Sensor (SeaWiFS) satellite images. Miller [
8] applied color enhancement techniques to differentiate dust, ocean surfaces, and clouds. Several studies have also shown that it is possible to detect Saharan dust over land using brightness temperature (BT) in thermal infrared spectra [
9,
10,
11,
12]. A correlation of the BT between 11 µm and 3.7 µm bands for dust outbreaks was proposed by Ackerman [
13], who developed a further tri-spectral (8, 11, and 12 µm) technique for detecting dust over water and for distinguishing dust plumes from water/ice clouds [
14].
Currently, several approaches have been developed for dust storm detection using MODIS measurements given its good spectral, spatial, and temporal resolutions. Qu et al. [
15] used the Normalized Difference Dust Index (NDDI), a normalized ratio of the 2.1 µm band and blue band, to detect dust storms and monitor the moisture change of the dust storm. NDDI has advantages due to the high sensitivity of the MODIS 2.1 µm band to moisture content. Hao et al. [
16] proposed a thermal infrared index to detect Saharan dust storms by combining four moderate resolution Imaging Spectroradiometer (MODIS) thermal emissive bands (TEBs). An automatic multi-spectral approach for detecting dust storms in Northwest China was developed by Han et al. [
17], where a set of indices was used to separate the dust from cloud, snow, and land with several Reflective Solar Band (RSBs) measurements.
However, most of abovementioned methods detect dust storms only with measurements of either RSBs or TEBs. Furthermore, the cloud mask product [
18] is used directly, which may misclassify partial dust as clouds under some conditions, hence decreasing detection quality. In this paper, an algorithm based on the multi-spectral technique was developed, combining measurements of six MODIS RSBs and TEBs. The bands were selected according to the spectral analysis and the thresholds of each test were decided with the statistical analysis. Several dust storm events were selected as case studies to test the algorithm. Additionally, the results were validated not only with MODIS true color images, but also quantitatively with OMI and CALIOP measurements.
4. Conclusions
In this paper, a multi-spectral algorithm for observing and monitoring dust aerosols over both bright and dark surfaces was developed by combining measurements of the MODIS solar (RSB) and thermal bands (TEB).
The spectral curves of several major scene types, such as dust, cloud, vegetated surface, and non-vegetated surface, were derived statistically from a large quantity of training data. According to spectral analysis, the algorithm was divided into bright surface and dark surface branches for dust detection. In the algorithm, the thermal bands were mainly used for filtering out cloud, as well as a water vapor band, while the RSBs were selected for separating dust aerosols from other scene types. Well-developed indices from other previous studies were adopted directly, and some tests were proposed for the first time, including the reflectance of the red band used for dust detection. By comparing the MODIS true color image, the core part of the dust pixels was correctly identified, except for some missing pixels that had relatively low intensity. Meanwhile, OMI UVAI and CALIOP were selected for quantitative validation. The validation by OMI UVAI showed that most plumes (more than 70%) were detected. One main factor that impacted the accuracy of the algorithm may be attributed to the spatial resolution difference between the two sensors. The validation by CALIOP VFM showed that more than 90% of dust aerosol pixels were identified correctly in the selected case. Although there were some misclassified pixels, most of them were concentrated at the edge of the dust storm with light dust aerosol loading. Statistical analysis of the training data was the primary method used in this research to achieve the desired spectral features of the dust. Although the statistical analysis provided reasonable results, its accuracy was dependent on the number of training data. Moreover, the thresholds used in the algorithm were usually site-specific, which limited the application of the algorithm.
This approach for detecting dust aerosols with satellite remote sensing could be enhanced with further studies such as adding more precise site-specific information. Currently, the surface has only been separated into dark and bright surfaces for dust storm monitoring, thus a stricter classification of surface features and site-specific thresholds could enhance the accuracy of the algorithm significantly. It would be of great value to build a lookup table to store the site-specific thresholds for the entire global dataset. On the other hand, some useful research aims to be implemented based on this dissertation in the near future include: (1) tentative retrieving of the AOD based on the detection results; and (2) gathering dust information over a long time range. With long-term aerosol information, it is feasible to analyze their distribution, as well as their motion, height, width, and even seasonal variation.