A cloud identification algorithm used for cloud masking, which is based on the spatial variability of reflectances at the top of
the atmosphere in visible wavelengths, has been
developed for the retrieval of aerosol properties by MODIS. It is shown that the spatial pattern of cloud reflectance, as observed
from space, is very different from that of aerosols. Clouds show a high spatial variability in the scale of a hundred metres to a
few kilometres, whereas aerosols in general are homogeneous. The concept of spatial variability of reflectances at the top of
the atmosphere is mainly applicable over the ocean, where the surface background is sufficiently homogeneous for the separation
between aerosols and clouds. Aerosol retrievals require a sufficiently accurate cloud identification to be able to mask these ground scenes. However, a conservative mask will
exclude strong aerosol episodes and a less conservative mask could introduce cloud contamination that biases the retrieved
aerosol optical properties (e.g. aerosol optical depth and effective radii). A detailed study on the effect of cloud contamination on
aerosol retrievals has been performed and parameters are established determining the threshold value for the MODIS aerosol cloud mask
(
A prolonged pollution haze event occurred in the northeast part of China during the period 16–21 December 2016. To assess the impact of such events, the amounts and distribution of aerosol particles, formed in such events, need to be quantified. The newly launched Ocean Land Colour Instrument (OLCI) onboard Sentinel-3 is the successor of the MEdium Resolution Imaging Spectrometer (MERIS). It provides measurements of the radiance and reflectance at the top of the atmosphere, which can be used to retrieve the aerosol optical thickness (AOT) from synoptic to global scales. In this study, the recently developed AOT retrieval algorithm eXtensible Bremen AErosol Retrieval (XBAER) has been applied to data from the OLCI instrument for the first time to illustrate the feasibility of applying XBAER to the data from this new instrument. The first global retrieval results show similar patterns of aerosol optical thickness, AOT, to those from MODIS and MISR aerosol products. The AOT retrieved from OLCI is validated by comparison with AERONET observations and a correlation coefficient of 0.819 and bias (root mean square) of 0.115 is obtained. The haze episode is well captured by the OLCI-derived AOT product. XBAER is shown to retrieve AOT well from the observations of MERIS and OLCI.
Haze is an atmospheric phenomenon which is associated with horizontal visibilities of less than l0
A thick smoke haze enveloped the eastern and northern part of China in December 2016. Pictures taken by cameras onboard the satellite TERRA/AQUA show that the area of China affected by haze exceeded about 1.5 million square kilometres. The poor visibility resulted in several highways and regional airports being closed for extended periods. The situation deteriorated significantly during the haze event and became a matter of public concern.
Satellite observations of the reflectance of solar radiation at the top of the atmosphere are used to determine aerosol optical thickness (AOT), which is used as an indicator of air quality (Kaufman et al., 2002). There are numerous attempts for the retrieval of aerosol properties from satellite observations. AOT retrieval algorithms have been developed for use with the measurements of Moderate Resolution Imaging Spectroradiometer (MODIS) (e.g. Dark-Target (Levy et al., 2013), DeepBlue (Hsu et al., 2013), the Multiangle implementation of atmospheric correction (MAIAC) (Lyapustin et al., 2011)), Advanced Along-Track Scanning Radiometer (AATSR) (e.g. AATSR Dual-Viewing (ADV) (Kolmonen et al., 2016; Sogacheva et al., 2017), Oxford-RAL Aerosol and Cloud (ORAC) (Thomas et al., 2009), and Swansea University (SU) (North et al., 1999) algorithms). AOT is also derived from observations of the Multi-angle Imaging SpectroRadiometer (MISR) (Diner et al., 2005), PARASOL's Polarization and Directionality of the Earth's Reflectances (POLDER) (Dubovik et al., 2014), Sea-Viewing Wide Field-of-View Sensor (SeaWiFS) (Sayer et al., 2012) etc.
One challenge for the derivation of AOT long-term datasets from satellite observation is to generate comparable AOT data products from the different instruments, which have limited lifetimes. Consequently, mature aerosol algorithms, which can be applied to data from instruments on different platforms, are required. For example, the three MODIS aerosol algorithms have been applied to the Visible Infrared Imaging Radiometer Suite (VIIRS) instrument and the three AATSR algorithms have been proposed to be applied to the observations of the Sea and Land Surface Temperature Radiometer (SLSTR) instrument (Popp et al., 2016).
The MERIS instrument onboard Environmental Satellite (Envisat) provided valuable information for different applications (Verstraete et al., 1999). There are several previous attempts to develop AOT retrieval algorithms for MERIS, e.g. the Bremen AErosol Retrieval (BAER; von Hoyningen-Huene et al., 2003, 2011), and the European Space Agency (ESA) standard aerosol retrieval (Santer et al., 2007). These had mixed success (Mei et al., 2017a). BAER has limited accuracy away from dark-vegetated surfaces and primarily for non-absorbing aerosols (de Leeuw et al., 2015; Holzer-Popp et al., 2013), while the ESA standard AOT retrieval tends to overestimate AOT (de Leeuw et al., 2015). The recently developed eXtensible Bremen AErosol (XBAER) algorithm (Mei et al., 2017a, b) has been internally validated in the Aerosol-Climate Change Initiative (Aerosol-CCI) project (Popp et al., 2016), and shows very promising results.
The newly launched (on 16 February, 2016) instrument Ocean Land Colour Instrument (OLCI) continues the work of MERIS as it contains all MERIS channels. Theoretically it is possible to transfer the mature MERIS retrieval algorithms to the OLCI instrument. In this paper, the XBAER algorithm has been applied to OLCI instrument for the first time. To our best knowledge, this is the first publication of AOT retrieved from OLCI. Although Sentinel-3 has only recently been launched, applying XBAER to OLCI data we have identified a haze event over Beijing, China, during December 2016. We use observations by OLCI during this episode to test our retrieval of AOT. This study is a necessary first step to observing the aerosol in the Arctic, which is an overarching long-term objective.
In this paper, the characteristics of OLCI and MERIS instruments are presented and compared in Sect. 2. The XBAER algorithm is briefly explained in Sect. 3. Section 4 shows the comparison between OLCI and MERIS instruments – first XBAER OLCI-derived AOT results and a comparison with AOT from MODIS/MISR and AERONET observations is shown and discussed from a global point of view. The AOT retrieved during the regional haze event is also presented and discussed in Sect. 4. Conclusions are given in Sect. 5.
The European Space Agency Sentinel-3 satellite was successfully launched on 16 February 2016. It is one element of the EU Copernicus
system previously known as the Global Monitoring for Environment and Security (GMES) system
(
Spectral channels for MERIS and OLCI instruments.
The primary objective of OLCI is to observe the ocean and land surface in the solar spectral region and thereby to harvest information
related to biology. OLCI also provides information on the atmosphere and contributes to climate studies. OLCI is a push-broom imaging
spectrometer that measures solar radiation reflected by the Earth, at a ground spatial resolution of 300
The XBAER algorithm was designed for the retrieval of AOT from MERIS and similar observations. It has its own cloud-screening approach, aerosol type selection and surface parameterization (Mei et al., 2017a, b). The cloud-screening algorithm minimizes cloud contamination for aerosol retrieval in XBAER. The XBAER cloud-masking algorithm determines the presence of cloud by using (i) the brightness of the scene, (ii) the homogeneity or variability of the top of the atmosphere reflectance, and (iii) cloud height information (Mei et al., 2017b). The threshold values in the XBAER cloud-masking algorithm are selected by a two-step process. The ranges for the thresholds were determined by using accurate radiative transfer modelling with different surface and atmospheric scenarios. A histogram analysis has been used for different cloud, aerosol, and surface scenarios to estimate the optimal threshold values for each criterion.
The XBAER algorithm uses a generic one-parametric surface parameterization for both land and ocean. XBAER uses a set of space–time-dependent spectral coefficients to describe surface properties. The spatial and temporal resolutions are 10
XBAER uses MODIS Dark-Target aerosol type assumptions and the expected aerosol type for a given region and season is taken from an analysis of Aerosol Robotic Network (AERONET) and Maritime Aerosol Network (MAN) observations for both land and ocean. AOT and surface reflectance are retrieved by minimizing the difference between simulated and measured top-of-the-atmosphere (TOA) reflectance using a look-up table (LUT), created by the radiative transfer software package SCIATRAN (Rozanov et al., 2014). Details of the XBAER algorithm can be found in Mei et al. (2017a, b). A post-processing technique used in Aerosol-CCI project and the MODIS monthly snow fraction dataset have been additionally applied to avoid unresolved clouds/snow (Popp et al., 2016).
Spectral response function of MERIS (dashed lines) and OLCI (solid lines) for overlap channels.
Global comparison of OLCI XBAER AOT with AERONET observations for 2016 December.
Comparison of the retrieved global monthly mean AOT at 0.55
Time series of meteorological parameters and pollutants during December 2016.
Daily MODIS RGB and AOT for East China (100–125
One important characteristic investigated is the instrument spectral response function (SRF) because it is the major difference between
MERIS and OLCI for overlap channels. Figure 1 shows the SRF for the MERIS and OLCI overlap channels. The OLCI SRF mean dataset
(
In order to quantitatively investigate the impact of different SRFs, the TOA reflectances have been simulated with and without taking
SRF into account. The simulations have been determined by undertaking radiative transfer simulations using SCIATRAN for atmospheric
and surface conditions (Rozanov et al., 2014). The MERIS observation geometry for 2 July 2009 over Paris was used to perform
a forward simulation. In particular, the solar zenith angle, viewing angle, and relative azimuth were set to (32.32
In order to design representative simulated scenarios, we define a comprehensive set of aerosol optical parameters, surface spectral reflectances, and other atmospheric properties comprising temperature and pressure profiles, the profiles of the concentration of gaseous absorbers and scattering. Suitable ranges of values for all relevant inputs for the radiative transfer model are obtained by statistical analysis of corresponding global products (Mei et al., 2016a). For this purpose, we use the following parameters.
Surface reflectance: three typical surface types representing vegetation, soil and water, i.e. relatively dark land (vegetation-covered city), bright land (desert), and water surface (ocean surface), were used. The typical vegetation and soil spectra are adapted from von Hoyningen-Huene et al. (2011), the liquid water spectrum comes from the SCIATRAN database (see references in Rozanov et al., 2014). Figure 2 shows the corresponding surface reflectance spectra for selected surface types.
Aerosol scenarios: within the ESA Aerosol-CCI project, a representative value for global mean AOT of 0.25 has been selected
(Holzer-Popp et al., 2013; de Leeuw et al., 2015). Thus an AOT of 0.25 was selected for the simulation of “vegetation” and “water”
cases. An AOT value of 0.5 was used for the “soil” scenario to represent a “real” case for the Sahara region. Moderately absorbing
(fine-mode radius
Other atmospheric parameters: the profiles of temperature, pressure, and concentration of the gases ozone,
In Table 1 the spectral channels of OLCI and MERIS are given. Figure 2a presents the surface spectral reflectance for the three surface types selected. Figure 2b presents the simulated TOA differences for the above scenarios. The differences for all surface/atmospheric conditions are less than 1.5 %. These are similar to the simulation with and without convolution for MERIS with the exception of the O2A and water vapour channels. However, the potential impacts of different SRFs may also introduce some uncertainties to the XBAER cloud mask due to the relatively strong impact of SRF to the O2A channels (about 20 % difference).
AERONET observations are considered to be the “ground truth” for satellite validation (Holben et al., 1998). Here, we collocate the
XBAER OLCI aerosol retrievals with the AERONET Version 3.0
(
Figure 3 is a plot which compares XBAER-derived and AERONET-observed AOT at 0.55
Figure 4 shows the global monthly AOT of December 2016 for MODIS collection 6 (Levy et al., 2013), MISR (Diner et al., 2005) and OLCI
(XBAER) algorithm. In order to identify biomass burning events, the active fire points of MODIS
(
Same as Fig. 6 but for MERRA AOT.
Same as Fig. 6 but for OLCI.
Same as Fig. 6 but for
In the following we show the ability of the retrievals of XBAER used with OLCI data to resolve spatial aerosol patterns on a synoptic
scale. A prolonged haze event was observed over Beijing during the period 16–21 December 2016. The intention of applying XBAER to
this event is to show the potential of the retrieval to resolve aerosol patterns at a local level and thus being able to support future
studies analysing such events. This event is investigated by both ground-based measurements and satellite observations. Figure 5a shows
that winds at the surface were weak, with a daily averaged wind speed lower than 3.5
Figure 5d shows the time series of concentration of
Figure 6 shows the MODIS/Terra-derived AOT for the haze period. According to Fig. 6, this intense part of the haze episode has been
partly observed by MODIS. However, a large part of it (under cloud-free conditions) during the first 3 days is missing, mainly due
to cloud masking applied in the MODIS aerosol retrieval. Figure 7 shows the AOT from Modern-Era Retrospective analysis for Research and
Applications Version 2 (MERRA-2) simulation (Rienecker et al., 2011) in order to exclude the impact from cloud screening. According to
Fig. 7, the shape of the area covered by high AOT in MERRA remains stable except for 20 December, indicating the relative stable
meteorological condition during the haze period. Due to the narrower swath width of OLCI compared to that of MODIS (1270
In this study, we have applied XBAER to data from the OLCI instrument onboard Sentinel-3 for the first time on both synoptic and global
scale. The potential differences caused by different spectral response functions for OLCI and MERIS have been investigated by using
SCIATRAN to generate representative simulated scenarios for dust aerosol type over desert, moderately absorbing aerosol over vegetation
regions, and maritime aerosol over water. The overall differences for all selected channels for XBAER are smaller than 1.5 %. This
implies that XBAER can be used to retrieve AOT from OLCI. Although relatively large differences caused by SRFs (approximately 20 %)
have been found for the
The global monthly mean XBAER AOT maps for December 2016 show good agreement with those by MODIS and MISR. The comparison with AERONET
measurements reveals that XBAER can provide promising results over both dark and bright surface. The first comparison with AERONET
shows acceptable agreement between the two data sets, with a regression yielding
A significant haze event during December 2016 over Beijing has been analysed in this paper based on ground-based and satellite observations to show the potential of the retrieval to resolve aerosol patterns at a local level and thus being able to support future studies analysing such events. This large haze event has been attributed to the large local emissions under unfavourable meteorological conditions (temperature inversion in vertical direction and no advection). The MODIS/Terra- and OLCI-derived AOT both detect the haze event. However, due to cloud screening, the MODIS AOT partly misses it while the OLCI AOT is able to detect the main pattern of haze for clear conditions. The overlap retrieval for both MODIS and OLCI has similar values, indicating that OLCI provides another useful data source for air pollution monitoring.
Although the study shows that XBAER can be applied to OLCI observations for synoptic to global applications, several important issues
need to be addressed in the future work. Potential cloud contamination due to both the relative large calibration uncertainty of OLCI
compared to MERIS as well as the impact of SRF on
The XBAER-derived OLCI product is available upon request to the corresponding author.
The authors declare that they have no conflict of interest.
The authors would like to express their appreciation to Andreas Heckel from Swansea University, Bahjat Alhammoud/Manuel Arias from
ARGANS Ltd, and Debbie Richards from EUMETSAT for very valuable and detailed discussion about the OLCI instrument. The discussion
of model simulations with Anne Blechschmidt and Abram Sanders from the University of Bremen is highly appreciated. We would also like to
express our gratitude to the AERONET PIs for establishing and maintaining the long-term AERONET stations used for the validation. The
atmospheric components and meteorological data are from