Recently launched multichannel geostationary Earth orbit (GEO) satellite
sensors, such as the Geostationary Ocean Color Imager (GOCI) and the Advanced
Himawari Imager (AHI), provide aerosol products over East Asia with high
accuracy, which enables the monitoring of rapid diurnal variations and the
transboundary transport of aerosols. Most aerosol studies to date have used
low Earth orbit (LEO) satellite sensors, such as the Moderate Resolution
Imaging Spectroradiometer (MODIS) and the Multi-angle Imaging
Spectroradiometer (MISR), with a maximum of one or two overpass daylight
times per day from midlatitudes to low latitudes. Thus, the demand for new GEO
observations with high temporal resolution and improved accuracy has been
significant. In this study the latest versions of aerosol optical depth
(AOD) products from three LEO sensors – MODIS (Dark Target, Deep Blue, and
MAIAC), MISR, and the Visible/Infrared Imager Radiometer Suite
(VIIRS), along with two GEO sensors (GOCI and AHI), are validated,
compared, and integrated for a period during the Korea–United States Air
Quality Study (KORUS-AQ) field campaign from 1 May to 12 June 2016 over East
Asia. The AOD products analyzed here generally have high accuracy with high
Atmospheric aerosol particles are composed of solid and liquid matter and have diameters of a few nanometers up to several micrometers and lifetimes from a single day to tens of days. Aerosol particles affect the atmospheric radiation balance by scattering and absorbing incident top-of-atmosphere (TOA) sunlight and light scattered from the surface, as well as by interacting with clouds (e.g., by changing cloud distributions, optical properties, and precipitation by acting as cloud condensation nuclei) with global climate effects (IPCC, 2013). Global net radiative cooling or heating is determined partially by interactions for which the level of understanding is still low and varies significantly with geographic region. Additionally, ambient particulate matter (PM) at the ground level adversely affects human health through pulmonary and respiratory transport, resulting in heart disease, stroke, and lung cancer (Gao et al., 2015; Lim et al., 2012). Many developing countries in East Asia have both large anthropogenic emission sources and natural aerosol sources, such as the Taklamakan and Gobi deserts and wildfire regions. For this reason, East Asia currently has one of the most heavily polluted atmospheres in the world (S. W. Kim et al., 2007; Mehta et al., 2016; Yoon et al., 2014; Zhao et al., 2017).
Aerosol measurements are routinely conducted at diverse scales by laboratory experiments, in situ measurements, and remote sensing, and from various platforms including ground-based, airborne, shipborne, and satellite sensors. Accurate microphysical and chemical properties of aerosols can be obtained from laboratory experiments or ground-based and airborne measurements, but their spatial coverage is limited. Satellite-based remote-sensing measurements provide aerosol optical properties, including aerosol optical depth (AOD), over much broader areas. Currently operating low Earth orbit (LEO) satellite sensors, such as the Moderate Resolution Imaging Spectroradiometer (MODIS), the Multi-angle Imaging Spectroradiometer (MISR), and the Visible/Infrared Imager Radiometer Suite (VIIRS), provide global aerosol information but at a temporal resolution that is limited to once per day at least and, typically, to once every 2–3 d due to cloud cover (Garay et al., 2017; Hsu et al., 2013; Jackson et al., 2013; Levy et al., 2013). Most satellite-based aerosol retrieval techniques and algorithms have been developed for these LEO sensors (Diner et al., 1998; Higurashi and Nakajima, 1999; Hsu et al., 2004; Kaufman et al., 1997; J. Kim et al., 2007; Remer et al., 2005; Torres et al., 1998). To overcome temporal resolution limitations, there were several attempts to retrieve AOD using first-generation meteorological geostationary satellites such as the Geostationary Operational Environmental Satellite (GOES), the Geostationary Meteorological Satellite (GMS), and the Multifunction Transport Satellite (MTSAT), but they showed worse accuracy than those of LEO sensors due to the wider and fewer visible channels with coarser spatial resolution, which make it difficult to distinguish aerosol types (Kim et al., 2008; Knapp et al., 2002; Urm and Sohn, 2005; Wang et al., 2003; Yoon et al., 2007). As the specifications of recently launched geostationary Earth orbit (GEO) sensors, such as the Geostationary Ocean Color Imager (GOCI) and the Advanced Himawari Imager (AHI) over East Asia and the Advanced Baseline Imager (ABI) over the United States, are approaching those of current LEO sensors, aerosol optical properties can be retrieved with an accuracy as high as that of LEO sensors, and at much higher temporal resolutions, from a few minutes to an hour during daylight hours (Chen et al., 2018; Choi et al., 2016, 2018; Daisaku, 2016; Kikuchi et al., 2018; Laszlo and Liu, 2016; Lee et al., 2010; Lim et al., 2018; Zhang et al., 2018). This breakthrough in temporal resolution of GEO aerosol data enables us to monitor highly variable aerosol conditions and improve air quality forecasting, particularly for PM, with data assimilation (Jeon et al., 2016; Lee et al., 2016; Pang et al., 2018; Park et al., 2014; Saide et al., 2014) or machine learning (Park et al., 2019). To improve air quality model accuracy through satellite AOD retrieval, the satellite AOD should have broader coverage, high spatiotemporal resolution, and high accuracy. Most AOD data assimilation systems have been developed by using LEO satellite products such as MODIS because they have global coverage and high accuracy through the continuous retrieval algorithm improvement. The GEO satellite can provide more frequent AOD, but its spatial coverage can be limited to a specific area, especially in the case of GOCI. The period of AOD retrieval algorithm development and investigation using GEO is relatively shorter than LEO. Also, LEO sensors generally have more suitable channels with a high resolution and advanced measurement characteristics such as multi-angle and/or polarization for aerosol retrievals, which could result in higher accuracy of AOD from LEO than GEO generally (Jiang et al., 2019). Therefore, both accuracy and spatiotemporal coverage can be obtained simultaneously by using combined GEO and LEO AODs. For these reasons, the demand for GEO aerosol measurements is high.
Satellite aerosol retrieval algorithms have been improved toward higher spatial resolution (e.g., 5–10 km or finer) and higher temporal resolution (e.g., from daily to hourly and few minutes resolution) with higher accuracy (e.g., AOD uncertainty less than 0.03 or 10 %) to fulfill the requirement for an understanding of long-term climatological changes (GCOS, 2016).
Several field campaigns have been performed over East Asia to investigate aerosol chemical, microphysical, and optical properties based on in situ and remote-sensing measurements. These include the Transport and Chemical Evolution over the Pacific (TRACE-P) aircraft campaign in 2001 (Jacob et al., 2003), the Atmospheric Brown Cloud–East Asia Regional Experiment (ABC–EARES) in 2005 (Nakajima et al., 2007), the Distributed Regional Aerosol Gridded Observation Networks (DRAGON)–Asia campaign in 2012 (Holben et al., 2018), and the Megacity Air Pollution Studies (MAPS) in 2015 (Kim et al., 2018). Aerosol retrieval algorithms have been developed, improved, and validated using the extensive measurement datasets obtained from these field campaign studies (Garay et al., 2017; Jeong et al., 2016; S. W. Kim et al., 2007, 2016; Lee et al., 2018; Xiao et al., 2016).
The Korea–United States Air Quality Study (KORUS-AQ;
Here, in addition to the GOCI AOD dataset, the latest versions of AOD datasets from LEO sensors (MODIS, MISR, and VIIRS) and another GEO sensor (AHI) are validated, compared, and integrated for the period of the field campaign. The latest version of the Aerosol Robotic Network (AERONET) ground-based sun photometer dataset (version 3) over East Asia is used as a reference for the campaign period (Eck et al., 2018; Giles et al., 2019). Characteristics of the various AOD products are analyzed for specific transport cases with a high temporal resolution at the daily scale over the campaign period.
The remainder of this paper is organized as follows. In Sect. 2, satellite and ground-based remote-sensing data used in this study are summarized. In Sect. 3, the various AOD products are validated and compared using ground-based AERONET observations separately over ocean and land. The specific aerosol loading cases during the campaign are analyzed in Sect. 4. In Sect. 5, daily representative AOD is generated on a common spatial grid for each product and used to calculate the mean AOD distribution during the campaign. The daily AOD integration is tested using multiple AOD products at the daily scale. Finally, a discussion and conclusions are presented in Sect. 6.
The GOCI is a unique ocean color sensor in GEO (128.2
The case of 22 May 2016, 13:30 LT:
Comparison of AERONET AOD and
According to Choi et al. (2018), the GOCI YAER version 2 AOD shows
increased errors when the geometrical cloud fraction within AOD pixel
increases (particularly near cloud edges) and the remaining cloud
contamination was largely due to the absence of infrared (IR) measurement in
GOCI. Thus, in this study, additional cloud masking is applied to the GOCI
cloud-masking procedure, using Himawari-8 IR data. The IR cloud masking
processes in the AHI YAER algorithm is summarized as in Table 2, which are
based on several previous studies (Iwabuchi et al., 2014; Kim et al.,
2014). It consists of four tests using the brightness temperature (BT)
difference (BTD) of a different IR channel pair to detect high-level cloud,
low-level cloud, and cirrus cloud, which are difficult to detect or
classify using only visible channels of GOCI. The AHI IR cloud masking
information has 2 km spatial resolution (at the Equator) every 10 min for
the full-disk area, thus the co-location processing is required to match with
GOCI of 500 m spatial resolution and 1 h temporal resolution. The
spatially closest AHI IR pixel to each GOCI pixel is co-located. The GOCI
observation takes from 15 to 45 min of each hour, thus the pixels are
flagged as cloud if at least one of 10, 20, 30, 40, and 50 min of each
hour of AHI measurements determines the pixels as cloud. Then, the
co-located AHI IR cloud information is applied to the GOCI at 500 m spatial
resolution between the original cloud masking step and the pixel aggregation
step to 6 km. Details of the full AHI IR cloud masking procedure for aerosol
retrieval are described by Lim et al. (2018). On 22 May 2016, thin
cloud was detected over Manchuria (44–48
The AHI, on board the Himawari-8 and Himawari-9 satellites, is part of a new
generation of meteorological satellite sensors. Compared with previous
meteorological sensors, such as the Japanese Advanced Meteorological Imager
(JAMI) on board the Japanese Multifunction Transport Satellite–1R
(MTSAT–1R, also referred to as Himawari-6) or the Meteorological Imager
(MI) on board the Korean COMS satellite, the AHI has more channels (16)
including three visible channels (0.47, 0.51, and 0.64
MODIS is one of the most widely used instruments for global aerosol
measurements. It has been in operation on board NASA Terra (10:30 LT
descending) satellite since 1999 and the Aqua (13:30 LT ascending) satellite
since 2002. In general, MODIS measurements employ single-angle viewing,
multiple channels (36 channels), high spatial resolution (0.25 to 1.00 km
according to channel), and a wide swath (2330 km) enabling daily global
coverage for shortwave channels. The MODIS Dark Target (DT) aerosol
retrieval algorithm uses the broader-bandwidth MODIS channels (> 20 nm) in the visible to SWIR range. The DT algorithm assumes that
land surface reflectance in the visible range has a linear relationship with
SWIR (2.1
The MODIS Deep Blue (DB) aerosol algorithm uses ocean color channels and IR
channels to retrieve aerosol optical properties over bright land surfaces.
Using the enhanced DB algorithm, MODIS DB AOD is retrieved over
arid and semiarid surfaces; natural vegetation areas; and urban, built-up, and
transitional regions using several surface-reflectance calculations. These
calculations use a pre-calculated surface reflectance database with the
minimum reflectance technique, a DT-like approach, and a hybrid method over
arid and semiarid surfaces, vegetation, and urban, built-up, and transition
surfaces. The MODIS DB algorithm calculates AOD for each of the original level 1B (L1B)
1 km pixels, and aggregates and averages retrieved AOD pixels to 10 km
The MODIS multiangle implementation of atmospheric correction (MAIAC) aerosol algorithm performs aerosol retrievals and atmospheric correction over both dark vegetated surfaces and bright desert land surfaces (Lyapustin et al., 2011a, b). Compared to each scene and pixel-based approach of MODIS DT and DB algorithm, the MAIAC algorithm has a time series analysis and image-based processing. Maximum 16 d data that have multi-viewing angle are used to obtain surface bidirectional reflectance distribution function (BRDF) characteristics providing three parameters of the Ross-thick Li-sparse BRDF model. Recent MODIS Collection 6 MAIAC aerosol algorithm was improved in terms of higher spatial resolution of surface characterization from 25 to 1 km, cloud detection, aerosol model, optimization of LUT-based radiative transfer calculation, and others (Lyapustin et al., 2018). Also, an over-water retrieval process based on Fresnel reflectance model with the Cox–Munk assumption was added to provide ocean AOD. The MAIAC algorithm uses eight different aerosol models and the same channels as the MODIS DT algorithm for AOD inversion. The latest version C6 MODIS MAIAC 550 nm AOD pixels with the “best quality” are used in this study.
The VIIRS is a sensor on board the Suomi-NPP satellite, which was launched in
October 2011. The general characteristics of VIIRS are similar to those of
MODIS, and include single-angle viewing, multiple channels (22 channels),
high spatial resolution (375–750 m), and a wide swath (3040 km) that
results in no gaps between adjacent swaths near the Equator. Recent VIIRS
aerosol products provided by NOAA were updated from the previous
Environmental Data Record (EDR) and the Intermediate Product (IP) to the new
Enterprise Processing System (EPS) product. The previous VIIRS EDR and IP
aerosol retrieval algorithm was similar to the DT algorithm in terms of the
coupling of land surface reflectance in the visible range using the SWIR
channel (2.25
The MISR is one of the sensors on board the Terra satellite along with MODIS.
Unique characteristics of MISR include multichannel (four wavelengths,
centered at 446, 558, 672, and 866 nm) and multi-angle measurements (nine
cameras; nadir,
To evaluate the various satellite AOD products during the 2016 KORUS-AQ
campaign (1 May to 12 June 2016), extensive data from ground-based
remote-sensing AERONET sun–sky radiometers were collected from total 33
sites over East Asia, including 19 South Korean sites (Holben et al., 1998, 2018). Detailed site information, including locations, is
available at the AERONET home page
(
Characteristics of multi-sensor aerosol products.
The Sun–Sky Radiometer Observation Network (SONET) operated by the Institute of Remote Sensing and Digital Earth, Chinese Academy of Sciences also provides aerosol optical and microphysical data from ground-based CIMEL sun–sky radiometer measurement using their own retrieval algorithm (Li et al., 2018). A total of five SONET sites' data (Harbin, Hefei, Nanjing, Shanghai, and Zhoushan) are used to evaluate satellite land AOD products. The SONET aerosol data and site information during the campaign are available from the AERONET home page.
The statistical metrics as used by Sayer et al. (2014) were also applied
here for comparison of satellite AOD measurements over land and ocean using
AERONET and SONET and are summarized in Tables 3, 4, and 5. Because the
distribution of AOD is non-Gaussian and skewed towards low values, AOD
evaluation is difficult using simple statistical techniques. Thus, the
metrics applied here consist of the number of matched and co-located data points
(
The results of land AOD show increased
Each satellite measures the area within its swath at different times during
daylight hours, as listed in Table 1. In contrast to the hourly and 10 min
interval measurements of GOCI and AHI, respectively, the LEO satellites
observe East Asia only once per day. The overpass time for Terra is at 10:30 LT and those for Aqua
and Suomi-NPP are at 13:30 and 13:25 LT, respectively. When measurement
times are similar,
Comparison of observed
All land AOD products show high
IR cloud masking processes of the AHI YAER algorithm. Note that “high latitude” in step 1 corresponds to the 1st, 2nd, 9th, and 10th segments from the north of the AHI observation segments (a total of 10 segments for the full disk area) and “midlatitude to low latitude” in step 1 corresponds to the other 6 segments.
In contrast, GOCI and AHI MRM AOD shows negative MB of
MISR and MAIAC land AOD shows highest
Because of the limited number of SONET sites, the
Validation statistics for land AOD products using AERONET.
Validation statistics for land AOD products using SONET.
The target area for ocean aerosol retrievals differs among the various
algorithms. The MODIS DT, MISR, MAIAC, and VIIRS algorithms retrieve aerosol
properties only for dark ocean pixels, which means that surface pixels that
are not completely dark, such as those containing shallow or turbid water,
are masked. The GOCI/AHI Yonsei aerosol algorithms are also designed to
retrieve aerosols over dark pixels, but they include moderately turbid water
pixels by considering the climatological ocean surface reflectance based on
minimum-reflectance techniques in the GOCI and AHI MRM algorithms and by
considering chlorophyll
According to most validation metrics, ocean AOD products are more accurate
than those over land. This difference leads to generally lower errors for
ocean AOD compared to their respective over land retrievals, based on
AERONET AOD measurements (Fig. 3b). The sign of the ocean AOD error in the
low AOD range is the same as that of the land AOD error for all products,
i.e., negative in GOCI and AHI MRM and positive in DT and AHI ESR. The
MAIAC, VIIRS, GOCI, and AHI products have high accuracy, as evidenced by a
low RMSE (0.12–0.13) and a near-zero MB (
In summary, most LEO and GEO aerosol products over East Asia are highly
accurate and based on a comparison with AERONET with high
Validation statistics for ocean AOD products using AERONET.
Noticeable aerosol transport was observed over Hokkaido, Japan, during the
period 18–21 May 2016. Although GOCI and AHI AODs were retrieved at 1 h and
10 min temporal resolutions, respectively, only data for 09:30 and 13:30 LT
are presented in Fig. 4 for comparison with MODIS, MISR, and VIIRS
distributions. A time series of satellite AODs co-located with AERONET AOD
from Hokkaido University, located at 43.08
AOD distributions from GOCI, AHI MRM, AHI ESR, MODIS DT03K, MODIS DB, MODIS MAIAC, MISR, and VIIRS over the Hokkaido region during 18–21 May 2016. Note that morning and afternoon AODs for GOCI and AHI refer to 10:30 and 13:30 LT, respectively, and for MODIS these refer to the Terra and Aqua measurements, respectively. MISR only has a morning measurement and VIIRS only has an afternoon measurement. The pink symbol in a bottom-left panel shows the location of the Hokkaido University AERONET site.
As the dense smoke aerosol plume (AOD > 2.0 at the center)
generated from the Russian forest fires was transported to Hokkaido
continuously from morning to afternoon on 18 May, AERONET AOD at Hokkaido
University increased rapidly from 0.1 to 1.4, and the GOCI and AHI
successfully detected this abrupt increase. The MODIS and VIIRS instruments
also detected increasing AOD accurately, but the first and last AODs during
the day were 0.6 and 1.1 at 10:30 and 13:30 LT, respectively, and therefore
did not capture the full diurnal variation detected by AERONET, GOCI, and
AHI. The increase in AOD at Hokkaido on 18 May was anticipated from the
southward movement of the plume revealed by the GOCI and AHI measurements.
On 19 May, the plume remained over Hokkaido and the spatial distribution
changed little during daylight hours. The AOD observed by AERONET decreased
from 1.3 to 0.9, and the GOCI and AHI instruments detected this change but
with a slight overestimation during the morning. The VIIRS and MODIS DT and DB
AODs are higher by about 1.5 and the MISR AOD is lower than the
AERONET value by 0.9. On 20 May, the AERONET AOD began to increase to 1.0 at 06:00 LT, peaked up to 1.3 at 12:00–13:00 LT, and sharply decreased down to
Time series of multiple satellite AODs and AERONET AOD at the Hokkaido University site during 18–21 May 2016.
It is very hard to figure out the exact reason for overestimation of MODIS DT,
DB, VIIRS, and MAIAC AOD over this plume despite reasonable accuracy from
AERONET validation. The statistical metrics of MODIS DT, DB, MAIAC, and
VIIRS validation at Hokkaido University site during the campaign show very
high
It can be summarized that overall evaluation is not matched with an individual site or case over East Asia because of the complexity of surface conditions and dynamic aerosol types. Additionally, MODIS and VIIRS do not provide spatially continuous AOD distributions because of sun glint masking over ocean areas near Hokkaido, making identification of plume sources and transport patterns difficult. In contrast, GEO can avoid sun glint area over midlatitude areas. Sun glint is a bright ocean surface due to the reflected solar radiance, which is brighter in nadir viewing angles. Due to the measurement geometry, single-angle-viewing LEO sensors such as MODIS and VIIRS generally have the sun glint pixels in the middle of the swath. In contrast, GEO has the sun glint pixels as a circle shape centered at the Equator because GEO sensors are located at the Equator. Because of multi-temporal measurement without sun glint pixels, GEO such as GOCI and AHI can detect these transported aerosol plumes across ocean with more continuous spatiotemporal distribution than LEO.
Next, two heavy aerosol loading cases over the Korean Peninsula are analyzed (as
in Fig. 6). During the campaign, the first noticeable increase in PM above
the South Korean national air quality standard (50
AOD distributions from GOCI, AHI MRM, AHI ESR, MODIS DT03K, MODIS
DB, MODIS MAIAC, MISR, and VIIRS over the Yellow Sea and Korean Peninsula
(35–38
Meridional mean GOCI AOD over the Yellow Sea and the Korean
Peninsula (35–38
Compared with conditions on 25 May 2016, the overall AODs on 5 June 2016
over the Yellow Sea and Korean Peninsula was low (0.1–0.2) and the AOD over
the Seoul Metropolitan Area (SMA) near 37
The two events analyzed in this section involved rapid changes in hourly AOD but have noticeably different spatiotemporal characteristics, leading to high AOD conditions that are attributed to either long-range transboundary transport from China or local emissions in South Korea (Lee et al., 2019). To accurately assess these types of events, spatiotemporally continuous measurements with minimal data gaps are required, which are currently possible only from GEO measurements.
Because the various satellite AOD products were validated using AERONET,
results are only valid for specific ground sites. A comparison between
satellite products can provide the relative difference in AOD for each
pixel, but a direct comparison between satellite products of level 2 (L2)
data is difficult because they differ in spatial coverage, measurement time,
and spatiotemporal resolution. For this reason, each L2 AOD product was
regenerated as a daily average value on the spatial grid of the level 3 (L3)
products. Although some products are available in the L3 format, the methods
and criteria used in their L3 calculation differ considerably. Thus, a
simple and commonly used method is applied here to generate daily L3 AOD.
The spatial domain is set to 20–50
The number of L2 AOD pixel samples within each
Relative sampling frequency of L2 AOD pixels used to calculate
mean AOD of
Mean of daily fusion AOD (
The number of L2 pixels for each grid can be normalized by
Retrieved satellite AOD errors can be classified into two types: random
error and bias. Although some algorithms, such as the optimal estimation
method, can provide an estimated random error or uncertainty quantitatively
(e.g., Jeong et al., 2016), the random error and bias of retrieved AOD can
be assessed only over AERONET sites, making it difficult to quantify and
validate uncertainties for all pixels. As errors were found to be
distributed equally around zero for land and ocean surfaces during the
validation using AERONET data, the combined AOD is calculated by selecting
the median value from the daily
Validation of daily mean AOD using AERONET daily mean AOD during
the KORUS-AQ campaign period (1 May to 12 June 2016) for the
Difference between campaign period mean gridded AOD of
Evaluation of the daily average AOD for each product and the combined AOD
using daily AERONET AOD is presented in Fig. 10. The closest grid point to
each AERONET site is selected for the comparison. The number of selected
grid points is 870 (AHI), 768 (GOCI), 677 (MODIS MAIAC), 658 (VIIRS), 436
(MODIS DT), 303 (MODIS DB), and 106 (MISR). Most products show similar bias
patterns to the level 2 pixel-level validation. For instance, GOCI is
negatively biased and MODIS DT is positively biased. However, when we
combined all of these products for the fusion AOD, it has a higher
The validation using AERONET is only available over a few specific grids. Thus, the difference between each product AOD and fusion AOD of Fig. 9 is calculated and compared (as in Fig. 11). GOCI shows relatively low AOD compared to fusion AOD for land pixels over the Korean Peninsula and Japan by about 0.2 and over southeastern China by up to 0.4 and higher AOD over Manchuria by about 0.3. AHI MRM and ESR show the least difference overall over most areas and this is related to the highest selection ratio of AHI products for fusion AOD. An interesting feature is that positive–negative pattern is the opposite between MRM and ESR over most grids, which was found to agree with improved accuracy in AHI when these two products are merged in Lim et al. (2018). The narrow swath of MISR leads to a broad gap between paths, and the discontinuity of the MISR L2 AOD data is noticeable along the swath boundary. More period averaging seems to be required to analyze MISR AOD characteristics compared to others. The difference between VIIRS AOD and fusion AOD is quite similar to that of AHI ESR, as there is higher AOD over southeastern China (0.3) and lower AOD over northeastern China (0.2). MODIS DT shows high noise patterns over the Yellow Sea compared to others, which can be related to the lower sampling frequency due to sun glint and turbid or shallow water masking with coarse pixel resolution (10 km). MODIS DB shows similar pattern to GOCI over land except for Manchuria. Because Manchuria has a bright surface, MODIS DB, MODIS MAIAC, VIIRS, and MISR can have better accuracy than others. The MAIAC generally shows less difference with fusion AOD except for higher AOD over ocean grids near the Chinese coast. Lyapustin et al. (2018) also notes that current masking of MAIAC misses several coastal waters with high sediments where AOD retrievals often show a high bias.
In this study, we compare spatiotemporal characteristics of three GEO AOD
products (GOCI, AHI MRM, and AHI ESR) and five LEO AOD products (MODIS DT,
MODIS DB, MODIS MAIAC, MISR, and VIIRS) and validate each product using the
AERONET version 3 and SONET dataset for the 2016 KORUS-AQ campaign. Most AOD
products have high accuracy and wide coverage over East Asia, but each have
individual unique characteristics (e.g., detailed accuracy and sampling
frequency). Although Choi et al. (2018) showed that GOCI AOD is
reliably accurate for the period 2011–2015, it is negatively biased during
the 2016 campaign period. This difference in accuracy may be attributable to
changes in climatological surface reflectance or calibration drift.
Improvement of surface reflectance including these calibration drifts or
surface reflectance changes is required. The DT method used in AHI ESR and
MODIS DT AOD retrievals results in a positive bias and higher AOD over East
Asia compared to other products. The MISR AOD has smaller coverage than
MODIS and VIIRS, but the AOD accuracy is higher than for the other products
because of an improved surface-reflectance treatment that takes advantage of
multi-angle measurements. However, it also seems that the MISR retrievals
often screen out the highest AOD events, thereby biasing the sampling in
this region. MISR uses neither SWIR channels nor pre-calculated surface
reflectance; the algorithm does not retrieve AODs if aerosol signal is too
high to get surface signals consistently. The range of MISR AOD product is
set to be from 0.0 to 3.0 according to Witek et al. (2018).
The maximum value is lower than others, such as 3.6 for GOCI and 5.0 for MODIS.
These dynamic range and accuracy differences are due primarily to algorithm
design, which is optimized for particular sensor specifications, such as the
available channels, and are not related to orbit types. The MAIAC AOD shows
high accuracy (
As GOCI and AHI AOD can be retrieved with high accuracy at near real time,
the highly variable AOD conditions over East Asia, including transport from
Russia to Japan, transport from China to South Korea, and local emissions in the
SMA and subsequent transport to the Yellow Sea, can be successfully
detected. This results in more representative daily AOD values. A combined
AOD using GEO and LEO data is also tested using a median value selection at
the daily scale with a
Although the validation using AERONET data reveal relative characteristics
among the various AOD products in terms of accuracy, it is insufficient to
thoroughly investigate these characteristics. Each algorithm includes
subjective criteria, such as those used in cloud masking,
surface-reflectance determination, aerosol model selection, inversion
methods, and quality control. For example, the possible AOD range that can
be retrieved and provided as the final AOD product varies among GOCI
(
This study focuses only on the spring season of 2016, when the KORUS-AQ campaign was conducted. An extended long-term study will be required to evaluate monthly or seasonal mean AOD trends of GEO and LEO measurements and combined AOD products. Additionally, the integration of multiple datasets may be improved by a consideration of pixel-level uncertainties; varying error characteristics, pixel size, and pixel shape; and the application of more advanced statistical techniques. Other optical properties, such as the Ångström exponent and single-scattering albedo, should also be investigated along with AOD in future studies.
The GOCI and AHI Yonsei aerosol retrieval data during the KORUS-AQ campaign
are available from
MC, HyL, SL, JK, TE, BH, MG, EH, PS, HoL MC, and JK designed the data analysis. MC, HyL, SL, JK, EH, and PS carried out the GOCI and AHI data production, distribution, and analysis. MC, HyL, SL, JK, TE, and BH carried out the installation, maintenance, and data analysis of the AERONET measurements during the 2016 KORUS-AQ campaign. MG provided the MISR AOD data and contributed to the data analysis. HoL provided the VIIRS AOD data and contributed to the data analysis. MC and JK wrote the manuscript with comments from all coauthors.
The authors declare that they have no conflicts of interest.
We thank all principal investigators and their staff for establishing and maintaining the AERONET and SONET sites used in this investigation. We also thank the MODIS, MISR, and VIIRS science teams for providing valuable data for this research. This research was supported by the National Strategic Project-Fine Particle of the National Research Foundation of Korea (NRF) funded by the Ministry of Science and ICT (MSIT), the Ministry of Environment (ME), and the Ministry of Health and Welfare (MOHW; NRF-2017M3D8A1092021). Some research tasks were supported by the NASA ROSES-2013 Atmospheric Composition: Aura Science Team program and NASA Headquarter Directed Research and Technology Development Task (grant number: NNN13D455T, manager: Kenneth W. Jucks and Richard S. Eckman). A portion of this research was carried out at the Jet Propulsion Laboratory, California Institute of Technology, under a contract with the National Aeronautics and Space Administration. The editor and two anonymous reviewers are thanked for numerous useful comments, which improved the content and clarity of the manuscript.
This research has been supported by the National Strategic Project-Fine Particle of the National Research Foundation of Korea (NRF) funded by the Ministry of Science and ICT (MSIT), the Ministry of Environment (ME), and the Ministry of Health and Welfare (MOHW) (grant no. NRF-2017M3D8A1092021) and the NASA ROSES-2013 Atmospheric Composition: Aura Science Team program and NASA Headquarter Directed Research and Technology Development Task (grant no. NNN13D455T).
This paper was edited by Cheng Liu and reviewed by two anonymous referees.