For several decades, satellite-based microwave sensors have provided valuable soil moisture monitoring in various surface conditions. We have first developed a modeled aerosol optical depth (AOD) dataset by utilizing Soil Moisture and Ocean Salinity (SMOS), Advanced Microwave Scanning Radiometer 2 (AMSR2), and the Global Land Data Assimilation System (GLDAS) soil moisture datasets in order to estimate dust outbreaks over desert areas of East Asia. Moderate Resolution Imaging Spectroradiometer- (MODIS-) based AOD products were used as reference datasets to validate the modeled AOD (MA). The SMOS-based MA (SMOS-MA) dataset showed good correspondence with observed AOD (-value: 0.56) compared to AMSR2- and GLDAS-based MA datasets, and it overestimated AOD compared to observed AOD. The AMSR2-based MA dataset was found to underestimate AOD, and it showed a relatively low -value (0.35) with respect to observed AOD. Furthermore, SMOS-MA products were able to simulate the short-term AOD trends, having a high -value (0.65). The results of this study may allow us to acknowledge the utilization of microwave-based soil moisture datasets for investigation of near-real time dust outbreak predictions and short-term dust outbreak trend analysis.

1. Introduction

Mineral dust plays a critical role in climate forcing and in the atmospheric radiation budget by scattering, absorbing, and reemitting longwave and shortwave radiation [14]. Mineral dust modifies the microphysical and optical properties of clouds, thereby indirectly affecting climate [5, 6]. It also supplies nutrient transfer to oceans and terrestrial ecosystems and neutralizes acid precipitation [79]. However, dust storms can also harm human lives by negatively affecting public health, the solar economy, agriculture, and transportation [10, 11]. The Korean peninsula and Japan are located to the east of China and are annually damaged by the Asian dust storms, called “Hwangsa” in Korean and “Kosa” in Japanese. These dust storms are mostly generated from deserts in Northern China, such as the Taklimakan and Gobi Deserts. The dust then blows into neighboring nations on the westerlies during the spring season from March to May [12]. Hwangsa can cause severe cardiovascular and respiratory diseases [13, 14]. Moreover, the frequency of dust events has been steadily increasing annually in Eastern Asia [15, 16]. Therefore, the ability to identify dust sources and to detect dust events is essential to predict dust outbreaks precisely in response to climate change and to plan prevention measures against dust disasters.

The mechanism through which dust is emitted should be further clarified in order to quantitatively understand and predict dust outbreaks. Many factors determine the mobilization of sand in dust source regions, including erosivity factors (i.e., strong wind speed) and erodibility factors (e.g., soil properties, soil moisture, and type of land cover). Many of the mineral dust source regions in East Asia are located in large bare deserts, and therefore, the surface wind speed and soil moisture (SM) are considered to be the primary factors that control interannual variation in the frequency of dust outbreaks [15, 17]. Various studies were previously conducted using wind tunnel experiments in laboratories or in fields to examine the effect of SM on wind erosion [1823]. However, most of these studies were not able to fully reflect large-scale dust outbreak phenomena under natural conditions. In order to overcome this challenge, the synergetic use of remote sensing datasets and global assimilated datasets is invaluable. This enables global-scale dust emission studies to be conducted beyond observatory-scale dust research [24, 25].

In recent years, many efforts have been made to identify dust source regions and to detect dust events based on satellite remote sensing techniques [10, 2634]. However, the previous studies hardly explain the conditions in the dust source regions that produce dust outbreaks because most analyses are based on data obtained from visible and infrared sensed datasets which have limitations under cloudy and dusty weather conditions. Thus, this study investigates dust outbreak phenomena using microwave-based satellite remote sensing techniques to estimate the surface conditions (i.e., surface SM) when dust events occur. Several studies have investigated surface SM from satellite-based microwave instruments [3539]. Satellite-based SM measurements were required during the initial states of weather forecasting and climate modeling over vast scales [4043]. Recently, satellite missions specializing in global SM measurements have been conducted. The National Aeronautics and Space Administration’s (NASA) Delta II rocket was launched in January 2015 carrying the Soil Moisture Active Passive (SMAP) to monitor SM and to detect the frozen or thawed state of soils [39]. Many other promising satellites have been launched to acquire global SM contents, such as the Advanced Scatterometer (ASCAT), Advanced Microwave Scanning Radiometer 2 (AMSR2), and Fengyun [4048]. Of these, certain satellites have shown better performance in determining the SM products in specific regions because of their specific characteristics, such as the sensor type (e.g., active and passive), platform overpass time, band frequency, polarization, retrieval algorithm, and calibration systems [42, 49, 50].

The Soil Moisture and Ocean Salinity (SMOS) was launched by the European Space Agency (ESA) and the Centre National d’Etudes Spatiales (CNES) of France in November 2009 as the second Earth Explorer opportunity mission [36, 38]. As the name implies, SMOS has been particularly designed to provide global SM observations using the L-band, an optimal wavelength for retrieving SM from space [36, 38]. In particular, SMOS-derived SM contents have proven to be advantageous in desert areas because the L-band signal exhibits a distinct behavior over extremely dry surface conditions [42, 50, 51]; as such, it is hardly influenced by polluted atmospheric conditions [36]. Hence, SMOS measurements have the potential for use in mineral dust research over desert regions. However, no such research has been conducted to date.

The AMSR2, a follow-up microwave sensor to the AMSR for Earth Observing System (AMSR-E), is a passive microwave instrument that has been providing global near-surface SM data since its launch in May 2012. The AMSR2 SM product has been widely validated by many studies [41, 52, 53].

The present study uses SMOS- and AMSR2-derived SM measurements and the Global Land Assimilation Dataset System (GLDAS) to investigate dust outbreaks over the desert regions of East Asia. The relationships among the SM, wind speed, and aerosol optical depth (AOD) are utilized to estimate dust outbreak phenomena. The main objectives of this study are (1) to estimate the modeled AOD (hereafter MA) datasets based on different SM and GLDAS wind speed datasets, (2) to validate MAs with the observed MODIS-AOD products and short-term trend analysis, and (3) to analyze the spatial distribution of MA products and observed AOD datasets in different seasons. To the best of our knowledge, this study is the first to use SMOS and AMSR2 products to predict dust outbreaks at the global-scale.

2. Datasets and Study Areas

In this study, the erosivity factor (i.e., wind speed), erodibility factors (i.e., soil properties, sand fractions, and SM), surface conditions (i.e., surface temperature), and precipitation were considered in order to analyze the effect of SM on dust outbreaks over deserts in East Asia.

The US Department of Agriculture (USDA) topsoil texture classification was used with the Harmonized World Soil Database (HWSD) to define the desert regions and to obtain information on the land and the atmosphere over the land. The SM contents were retrieved from SMOS and AMSR2; SM, wind speed (WS), surface temperature, and precipitation datasets from GLDAS were utilized for generating MA products. Atmospheric properties were obtained using the Moderate Resolution Imaging Spectroradiometer (MODIS) sensors on the Aqua satellite (Refer to the following sections for further details regarding those datasets). All of the datasets were gathered during 2015 and were resampled to a spatial resolution of 0.25°  × 0.25°.

2.1. Land Cover and Land Properties Dataset

Land cover and land properties datasets are used in this study to identify dust source regions and investigate dust outbreak conditions. The land cover dataset is obtained from Biosphere Atmosphere Transfer Scheme (BATS). BATS landform categories were created by Dickinson et al. [54] and modified by Olson [55] to support land-atmosphere modeling. Similarly, the land properties dataset is obtained from the Harmonized World Soil Database (HWSD), which is a raster database that contains more than 16,000 different soil mapping units at a spatial resolution of 1 km × 1 km. This database combines regional and national updates of soil information worldwide. It consists of 221 million grid cells and is contained within the 1 : 5,000,000 scale FAO-UNESCO Soil Map of the World. In particular, the HWSD has been applied in dust research to identify dust sources and investigate dust outbreak conditions as well as in several other research fields [5659]. The study area was masked based on the BATS desert land cover class and an HWSD sand fraction greater than 80% to define the desert regions.

2.2. Study Areas

Figure 1 shows the spatial description of the study areas in terms of the sand fraction. Boxes 1 to 7 show the famous desert areas that have been regarded as the major mineral dust source regions in East Asia [12, 15, 6062]. In this study, these bare desert areas have been classified using HWSD sand fractions greater than 80% and desert regions from the BATS model. Through this masking process, all major deserts in East Asia except the Gobi Desert were extracted. About 1,012,500 km2 (1,620 pixels) have been identified as desert areas, and the average sand, clay, and silt fractions were 87.6, 5.7, and 7.0% respectively. In these regions, especially during spring, frequent dust outbreaks result in Asian dust storms in neighboring countries [63, 64].

2.3. SMOS for Soil Moisture Contents

SMOS provides global near-surface SM (SSM) and covers the globe in three days, with a ground resolution ranging from 27 to 55 km (the average spatial resolution is 43 km), and its mission is to retrieve SSM at an accuracy of 0.04 m3/m3 at a depth of 3 to 5 cm. SMOS has been particularly designed to be on a sun-synchronous orbit, ascending in the morning at 0600 local time (LT) and descending at 1800 LT [3638]. The SMOS instrument is an L-band (1.41 GHz, λ = 21 cm) 2D interferometer radiometer that is considered to have the most suitable frequency for SM retrieval. In particular, the L-band frequency offers the added advantage of taking measurements in all weather conditions [36, 65]. In addition, radiometers are expected to perform better than scatterometers in dry areas [49, 50, 66, 67]. However, the L-band is negatively impacted by radio frequency interference (RFI) from illegal emissions in the protected passive band and unwanted emissions from active services operating in neighboring bands. Thus, several studies have been conducted since the SMOS mission began to mitigate the effects of RFI contamination [68]. The daily SSM products (L3) that were considered in this study were provided by the Centre Aval de Traitement des Données (CATDS Centre) with a spatial sampling of 25 km × 25 km in NetCDF format. We calculated the preceding one- to five-day composite SM products (ascending). This past-compositing SM process is considered because the antecedent moisture conditions that are detected by the SMOS may have the capability to predict dust outbreaks in the near future. Additionally, the FL_RAIN science flag is used for filtering the heavy rainfall events.

2.4. AMSR2 for Soil Moisture Contents

The Japan Aerospace Exploration Agency (JAXA), with the cooperation of NASA, launched AMSR2 sensor on board the Global Change Observation Mission 1-Water (GCOM-W1) platform in May 2012. It is the successor to JAXA’s AMSR-E on board NASA’s Aqua satellite. It utilizes microwave frequencies at C1- (6.9 GHz), C2- (7.3 GHz), and X- (10.65 GHz) bands for SM measurements and provides a 1,450 km swath width and 25 km spatial resolution associated with a revisit time of one to two days. The AMSR2 crosses the equator at 01:30 LT and 13:30 LT for descending and ascending orbits, respectively. Primarily, two algorithms are used to derive AMSR2-based SM products from measured brightness temperature: JAXA and NASA-VUA (VU University Amsterdam) Land Parameter Retrieval Model (LPRM) products [46]. Originally, both of these algorithms utilized a simple radiative transfer model [69]. Moreover, LPRM provides AMSR2 SM retrievals for both C- and X-band microwave frequencies, whereas the JAXA algorithm provides SM retrievals only for the X-band. The LPRM algorithm, however, is expected to be more accurate, which enables the retrievals of the SM and vegetation optical depth, simultaneously, from horizontal and vertical polarized measurements [70]. The AMSR2 SM products used in this study were cross-validated with the AMSR-E slow rotation dataset (http://global.jaxa.jp/press/2015/12/20151207_amsr-e.html). In this study, we only utilized the LPRM X-band SM dataset for conciseness, because SM from the C1- and C2-bands showed almost the same temporal patterns as the SM from the X-band over the study area.

2.5. MODIS for Atmosphere Products

The dust outbreak states that are characterized by the AOD and Angstrom exponent (AE) values were used in this study and were collected by the MODIS platform aboard NASA’s Earth Observing System- (EOS-) Terra and Aqua polar orbiter satellites. Aqua was launched on May 4, 2002, into a sun-synchronous orbit with an ascending orbit at 1330 LT and a descending orbit at 0130 LT. The Naval Research Laboratory (NRL) receives the global digital data from Aqua in near-real time. The MODIS data are distributed by NASA’s Goddard Earth Sciences Data and Information Services Centre (GES DISC) and are provided not only for the atmosphere but also for the land, cryosphere, and oceans [71, 72].

The MODIS sensor has 36 narrow spectral bands and measures in the 0.4–14.4 μm range. Specific channels, including visible and infrared bands (e.g., bands 1, 2, 3, 4, 26, 31, and 32), have been effectively used to detect mineral particles [30, 32, 34]. In particular, deep blue products (deep blue uses information from blue channels such as bands 3 and 8) are generally used over bright land surfaces to detect mineral dust. Deep blue was developed because the standard MODIS aerosol LAND measurements could not be used to retrieve data over bright surfaces, such as desert areas. For more details on the deep blue algorithm, refer to Hsu et al. [33]. In this study, the MYD04_L2 Aqua deep blue AOD (hereafter, MODIS-AOD) and AE datasets (550 nm) were employed because these datasets have already been extensively validated and have been successfully utilized in several studies [10, 25, 29, 30, 73, 74]. To ensure consistency in the analyses, the MODIS-AOD products were upscaled to 0.25°.

2.6. GLDAS for Land Surface Products

GLDAS is a result of datasets assimilated through cooperation of the National Centres for Environmental Prediction (NCEP), NASA, the National Oceanic and Atmosphere Administration (NOAA), and the Goddard Space Flight Centre (GSFC). GLDAS is a global-scale assimilation and modeling system that has been developed using several satellite- and ground-based datasets to provide the best estimation of the land surface conditions in near-real time [7577]. The analyses were based on the GLDAS/Noah products because these provide various land surface products with a high spatial resolution (0.25°) for every 3-hour period. In this study, WS (m/s, height of 10.0 m), soil temperature (K, depth of 0–10.0 cm), and precipitation (mm/3 h) products were also obtained from GLDAS. These products have been extensively used in hydrometeorological applications and validation studies [45, 78, 79].

3. Methodology

The SMOS, AMSR2, and MODIS overpass datasets are available at local times, whereas the GLDAS/Noah products are based on the Universal Time Coordinate (UTC) system. It is of utmost importance that the UTC time of GLDAS products must match the satellite local overpass time before they are used for dust outbreak prediction. The SMOS overpass time (ascending, 0600 LT) precedes the MODIS and AMSR2 overpass time (ascending, 1330 LT). AOD from MODIS, which detects the dust outbreak state, is assumed as the reference dataset in this study, whereas SMOS and AMSR2 which retrieved SM are used alongside the GLDAS wind speed product to obtain the modeled AOD. The SMOS overpasses the study areas from 2100 to 0100 UTC (0600 LT, ascending), and MODIS overpasses the study areas from 0430 to 0830 UTC (1330 LT, ascending). Theoretically, the overpass times have a lag of about 7.5 h for SMOS-based AOD prediction (Figure 2(a)). GLDAS products are available at 3-hour intervals (0000, 0300, 0600, 0900, 1200, 1500, 1800, and 2100 UTC). Because the study area has five different time zones, we matched the local time with the UTC for GLDAS products. Considering the SMOS and AMSR2-based dust model for zones 1 and 2, GLDAS products at 0000 UTC and 0300 UTC were used. For zones 3, 4, and 5, however, GLDAS products at 2100 UTC of the previous day and 0600 UTC of the same day were used (Figure 2(b)). Furthermore, fine mode aerosols, such as transported dust and biomass burning aerosols, were eliminated by setting an AE threshold greater than 0.35 [10, 24, 80].

3.1. Developing Modeled AOD Products from Satellites-Based SM and GLDAS Dataset

In this study, due to the shorter period of data availability, we assumed that SMOS-based and AMSR2-based SM retrievals have a negative relationship with AOD, as previously proposed by Kim and Choi [24] (Figure 3). Since AMSR2 and SMOS datasets are only available every four years and seven years, respectively, deriving the same number of equations as in the study by Kim and Choi [24] was not possible. Their study utilized an 11-year dataset (containing 27.261 million data points), which provided sufficient information to make feasible assumptions regarding the derived equations [24].

The WS magnitude is divided into 10 groups (3 to 12 m/s of WS) at 1 m/s intervals, as in Kim and Choi [24]. Alternatively, WS lower than 4 m/s were classified as 3 m/s, and WS higher than 12 m/s were classified as 12 m/s. Figure 3 shows the previous research finding that, generally, the averaged-AOD values decreased exponentially according to the various conditions of the WS and SM contents [24]. The exponential decrease in the averaged-AOD patterns with the increase in SM and decrease in WS is similar to trends presented in prior studies. The physical meaning of the patterns in Figure 3 is that a stronger WS (i.e., 6.5 m/s) is required to generate dust events under wet soil conditions because highly moistened sand particles have a more sturdy and cohesive strength among soil particles [81, 82]. Thus, a stronger WS and drier surface conditions have a greater propensity to produce dusty atmospheric conditions. Furthermore, the AOD value may be almost unaffected by the SM dynamics in a weaker WS or in damped conditions (about 16% of the SM). Moreover, under a stronger WS (higher than 6.0 m/s) and drier land surface conditions (less than about 16% of the volumetric SM), SM plays an important role in dust outbreak processes.

Considering the equation in Figure 3, the AOD values could be estimated using the following equation with different constants for and that are contingent on the magnitude of WS:From (1), we can establish an MA by using SM (%) and WS (m/s) datasets. In this study, we considered four different MAs: the SMOS SM and GLDAS WS-based MA (hereafter, SMOS-MA), the AMSR2 SM and GLDAS WS-based MA (hereafter, AMSR2-MA), the MA based on GLDAS dataset at 0600 (hereafter, GLDAS-MA_6), and the MA based on GLDAS dataset at 1200 (hereafter, GLDAS MA_12). The GLDAS-MA_6 and GLDAS-MA_12 enable us to make a fair comparison between satellite and GLDAS SM products for their dust outbreak prediction.

3.2. Soil Moisture Composition and Moving Average of MA Products

We used five different composite SM products (composited day: one- to five-day) to investigate the impact of antecedent SM on dust outbreaks. We composited SMOS, AMSR2, and GLDAS SM using following equation:where doy is day of year.

To calculate the time series of the MA dataset of the desert regions, we consolidated the daily available MA data over the desert regions and averaged them by using following equation:where is the number of available data over the desert pixels at doy and doy is day of year.

We set the composited day from one to five days. Moreover, the 11-day moving averaged MA was calculated as follows:

3.3. Comparison Metrics

The Pearson correlation coefficient () between MA and MODIS-AOD was calculated using (5). The bias (see (6)) was also considered. All metrics were used when the value was less than 0.05:

4. Results and Discussion

In East Asian deserts, dust outbreaks usually occur from early March to late May. Figure 4(h) shows the time series for the MODIS-AOD satellite data in 2015. The seasonal patterns of dust outbreaks are consistent with the results of previous research [12, 16]. The patterns can be well accounted for by simultaneously considering the temporal changes of SM and WS conditions (Figures 4(b) and 4(d)). The MODIS-AOD values (Figure 4(h)) were usually high under low SM and high WS conditions. However, this is not easily explained because the threshold value of WS at which dust outbreaks occur is greatly influenced by the SM conditions over the source areas [8184]. Moreover, the threshold values for WS and SM shifted according to the AOD setting for dust outbreak states [24].

4.1. Validation of MAs with AOD Observations

Validation of the MAs with an assumed reference dataset from the MODIS-AOD was performed to check the reliability of the models for future prediction. Figure 5 shows the temporal pattern of the MODIS-AOD and different MAs in the desert regions of East Asia for 2015. A temporal comparison between the SMOS-MA and MODIS-AOD shows a good correlation (-value: 0.56), but mostly higher than MODIS-AOD (Figure 5(a)). In the case of the AMSR-MA, a low correlation coefficient indicates that the temporal pattern of the MA dataset does not follow the reference AOD well, as compared to the SMOS-MA, and has an overall negative bias demonstrating that it usually underestimates the AOD with respect to the MODIS-AOD (Figure 5(b)). When the MA is based on the GLDAS SM product matched with the SMOS (i.e., the GLDAS-MA_6) and AMSR2 (i.e., the GLDAS-MA_12) overpass LT, the correlation coefficient is still lower than the SMOS-MA and underestimates the AOD; however, both of the GLDAS-based models showed very similar temporal patterns (Figure 5(c)). The analysis was further extended to evaluate the prediction based on one- to five-day composite SM products to consider the effect of antecedent conditions for all MAs (Table in Figure 5). The results indicate that different antecedent soil moisture conditions do not have much influence on dust outbreaks, which can be explained by the moisture memory concept [85, 86]. Specifically, desert regions have persistently very low SM conditions without dramatic daily fluctuations; thus, the WS, which is the major factor in bringing about dust outbreaks in bare desert areas, would affect more the timing of dust outbreaks. This conclusion is supported by the time series results shown on Figures 4(b) and 4(d). However, additional studies should be conducted that simultaneously consider the specific amount of variability of the antecedent WS and SM conditions before a dust outbreak event. Investigating these aspects is beyond the scope of this study.

4.2. Short-Term AOD Trends Analysis

In Figure 6, the 11-day moving average MA products are depicted using (3) for the short-term dust trends analysis. The solid red, light blue, orange, gray, and purple lines indicate moving averages of the MODIS-AOD, SMOS-MA, AMSR2-MA, GLDAS-MA_6, and GLDAS-MA_12, respectively. The table in Figure 6 includes the statistical results of each MA, including correlation and bias values, by comparing them with moving averaged AOD from MODIS-AOD. The SMOS-MA products showed the best -value (0.65) among other MAs. However, it showed a reversed trend and overestimated AOD notably in the fall season. This might be explained by unforeseen rainfall events that were not been considered during the 7.5 h time interval between the overpass time of SMOS and MODIS (Figure 2). Particularly, in East Asia, these regions experience monsoon season from September to October; thus, aerosols in the atmosphere might suddenly decrease because of frequent rainfall events. These abrupt AOD abatements would have not been predicted based on the previous information of SM and WS. During the spring and the summer seasons, SMOS better simulated AOD trends, with an -value of 0.70. Similarly, the GLDAS-MA_12 showed an -value of 0.62, but the GLDAS-MA_12 also presented reversed AOD trends similar to the SMOS-MA. The AMSR2-MA and the GLDAS-MA_6 products exhibited -values of 0.46 and 0.45, respectively, and underestimated AOD in the spring and the summer seasons and overestimated AOD in the fall season.

4.3. Limitations and Advantages of MA Products

Figures 7(a), 7(b), and 7(c) depict exemplary maps of the SMOS-MA, masked MODIS-AOD dataset (i.e., eliminating AOD values when AE > 0.35), and original AOD measurements from MODIS-AOD (without AE masking) for investigating a feature of MA products. Box (a)-1 predicts AOD to be between 0.3 and 2.5 in the box area; however, box (b)-1 and box (c)-1 do not show any aerosol observations. There are two feasible reasons that might support this mismatch. First, the AOD data might not have been observed using the deep blue algorithms, mostly because of cloud masking. The visible bands 3 (459 to 479 nm) and 8 (405 to 420 nm) cannot pass through clouds, preventing the use of these data [33]. However, in high cloud coverage season we can still predict the data with a microwave-based SM model. Second, most of the dust particles were likely blown away before observation by MODIS because the SMOS-MA in box (a)-1 was predicted 7.5 h before MODIS overpassed those areas. This is supported by the fact that particles in box (c)-1 were mostly masked out, as shown in box (b)-1, because they had high AE values, which means that most of the aerosols in box (c)-1 would have been susceptible to being blown away. Boxes (a)-2, (b)-2, and (c)-2 indicate the limitation of MA datasets because they can only possibly calculate AOD values over desert areas; thus, MAs are unable to estimate aerosols in nondesert regions. In addition, the MA concept is unreasonable to account for anthropogenic aerosols that were developed from urban areas such as Beijing, China (boxes of (a)-2, (b)-2, and (c)-2), and Seoul, South Korea ((a)-4, (b)-4, and (c)-4).

4.4. Spatial Analysis of MA and AOD from MODIS

We explored the spatial patterns of seasonally averaged AOD calculated from MAs and MODIS-AOD (Figure 8). Generally, the SMOS-MA simulates AOD well in the Taklimakan, Badain Jaran, Tengger, Ulan Buh, and Hobq Deserts; however, it overestimates in the Gurbantunggut, Hunshandake, and Horqin Deserts (Figure 1). Moreover, the prediction in the spring season occurs more frequently than the summer and the fall season. The observed AOD in the fall season is mostly missing in a large portion of the Taklimakan and Gurbantunggut Deserts. This might be due to many rainy days (i.e., monsoon season in the fall) during which clouds could obstruct the MODIS measurements. The AMSR2-MA underestimated the AOD in most of the study regions but captures AOD well in the Gurbantunggut and Gobi Desert regions. Nonetheless, the AMSR2-MA overestimated the AOD in the region (40–45°N and 90–100°E), which may be a result of the underestimation of SM, as the model is calculated by SM. Similarly, the GLDAS-MA also captures the AOD well but underestimates it in most of the study areas. The SMOS-MA map clearly determines severe dust source regions, such as the Taklimakan Desert, Badain Jaran Desert, and Hushandake Desert, and Horqin Desert. These regions are also highly related to sources of traveling dust under the prevailing westerlies. Particularly in northeast Asia, Asian dust storms during the most hazardous period might be generated from these areas. These results are consistent with those from numerous prior studies [60, 87, 88], and these are the first time to be identified using only remotely sensed SM analysis. Other areas that have been known to be dust source regions, such as the Gobi Desert (box 7 in Figure 1), are not included in this study because the sand fraction in that region is low, and the main factors contributing to dust emission in that region are dependent not only on SM and WS but also on vegetation conditions (e.g., dead leaves and NDVI variation) [15, 89]. In future studies, the SM contents retrieved with other satellite-based sensors (i.e., ASCAT and SMAP) and the corresponding land properties will be considered for specific land surface conditions (i.e., vegetated areas) and other factors (i.e., NDVI and clay and silt fraction) in dynamic environments.

5. Summary and Conclusions

In this study, SMOS, AMSR2, and GLDAS/Noah SM products in combination with GLDAS WS product were used to predict the AOD in 2015 over East Asian desert areas. SMOS-MA better estimated the AOD compared to the AMSR2-MA and GLDAS-MA in desert regions with highly polluted atmospheric conditions. The aerosol products, observed from the MODIS sensor, were used as a reference dataset to validate the MA. The validation results showed that the SMOS-MA captured the temporal patterns well, with a -value 0.56; however, it overestimated the reference AOD. A short-term trend analysis was conducted by calculating the 11-day moving average of MAs, the result of which showed that the SMOS-MA dataset demonstrated superior performance in capturing the short-term trends when compared to other MAs ( = 0.65). In addition, the results from investigating the impact of the antecedent SM on dust outbreaks showed that -values have not changed much with respect to different antecedent SM conditions. In this regard, we assume that the antecedent SM condition could not be the major factor that governs the timing of dust outbreaks. This is supported by the fact that bare desert regions have persistently very low SM conditions so that the WS, which fluctuates relatively more severely than SM (Figures 4(b) and 4(d)) over desert regions, would play a more pivotal role in triggering dust outbreaks. Moreover, spatial maps in different seasons also demonstrated that the SMOS-MA showed better consistency with the reference dataset compared to other MAs. The findings of this study reveal the application of microwave-based SM retrievals for near-real time dust outbreak predictions and short-term dust outbreak trend analysis. Future research will focus on the most robust modeling approach, based on machine learning, to improve MA products by utilizing the most recent and accurate available SM dataset from SMAP. In addition, Essential Climate Variable SM datasets are considered to be strong candidate products for the long-term AOD trend analysis, and the Representative Concentration Pathways dataset may be utilized to investigate AOD trends in the near future.

Competing Interests

The authors declare that there is no conflict of interests regarding the publication of this paper.


This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIP) (NRF-2016R1A2B4008312). This research was supported by Space Core Technology Development Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Science, ICT, and Future Planning (NRF-2014M1A3A3A02034789). The SMOS data were provided by the Centre Aval de Traitement des Données SMOS (CNES-IFREMER) and through the ESA-Supported SMOS Validation and Retrieval Team program. The aerosol optical depth data that were used in this study were obtained from https://ladsweb.modaps.eosdis.nasa.gov/data/. The Global Land Data Assimilation System (GLDAS) data used in this study were acquired as part of the mission of NASA’s Earth Science Division; these data are archived and distributed by the Goddard Earth Sciences (GES) Data and Information Services Centre (DISC) (https://disc.sci.gsfc.nasa.gov/). The most recent AMSR2 LPRM SM datasets were provided by Dr. Parinussa.