We invert global black carbon (BC), organic carbon (OC)
and desert dust (DD) aerosol emissions from POLDER/PARASOL spectral aerosol
optical depth (AOD) and aerosol absorption optical depth (AAOD) using the
GEOS-Chem inverse modeling framework. Our inverse modeling framework uses
standard a priori emissions to provide a posteriori emissions that are constrained by
POLDER/PARASOL AODs and AAODs. The following global emission values were
retrieved for the three aerosol components: 18.4 Tg yr
Atmospheric aerosol emission inventories are often used to drive chemical transport model (CTM) simulations of aerosol distributions on regional and global scales (Boucher, 2015; Brasseur and Jacob, 2017; Granier et al., 2011). Satellite-retrieved columnar aerosol optical depth (AOD) is directly related to light extinction due to the presence of aerosols; hence, satellite-retrieved columnar AOD is widely used to evaluate the spatial and temporal variability of aerosols simulated from CTMs (e.g., Chin et al., 2002; Ginoux et al., 2006; Kinne et al., 2006, 2003; Liu et al., 2012; Ocko and Ginoux, 2017; Pozzer et al., 2015; Schulz et al., 2006; Tegen et al., 2019). A general agreement has been shown for columnar AOD between model simulations and satellite observations in the Aerosol Comparisons between Observations and Models (AeroCom) “Experiment A” multimodel assessments (Kinne et al., 2006). However, this study also revealed a large model diversity of species-specific AOD and aerosol absorption optical depth (AAOD), which encourages research to harmonize and improve the emissions of individual aerosol species and aerosol precursors and the representation of aerosol absorption and other elements (Kinne et al., 2006; Samset et al., 2018). Accurate knowledge of spatial and temporal distribution of species-specific aerosol emissions is also useful for numerical weather prediction (NWP) (Benedetti et al., 2018; Xian et al., 2019).
There have been several efforts to improve aerosol emission inventories by using satellite observations and inverse modeling (Dubovik et al., 2008; Huneeus et al., 2012, 2013; Wang et al., 2012; Xu et al., 2013; Zhang et al., 2015, 2005; Escribano et al., 2016, 2017). However, most of them are regional studies that focus on a single aerosol or gas species, or they use predefined regions to reduce the size of state vector (Huneeus et al., 2012; Zhang et al., 2015, 2005). There are only a few studies that use detailed satellite information to simultaneously retrieve emissions of multiple aerosol components at the native spatial resolution of a forward CTM model. For example, we have previously developed an inverse modeling framework for retrieving aerosol emissions of black carbon (BC), organic carbon (OC) and desert dust (DD) components within the GEOS-Chem model (Chen et al., 2018). Specifically, the emissions of BC, OC and DD were simultaneously derived from satellite retrievals of spectral AOD and AAOD that were provided by the PARASOL (Polarization and Anisotropy of Reflectances for Atmospheric Sciences coupled with Observations from a Lidar) products. Chen et al. (2018) successfully used this method at the regional scale over all of Africa and the Arabian Peninsula.
This work expands the inversion algorithm of Chen et al. (2018) to the global scale. We also refine an assumption that defines the observation error covariance matrix for the recently released PARASOL Level 3 AOD and AAOD aerosol products generated by the GRASP (Generalized Retrieval of Atmosphere and Surface Properties) algorithm. The method is applied to derive global BC, OC and DD aerosol emissions for the year 2010, using the updated Hemispheric Transport of Air Pollution (HTAP) Phase 2 emission data (HTAP, 2010; Janssens-Maenhout et al., 2015) for the initial estimate of anthropogenic emissions. Then the satellite-derived global a posteriori aerosol emissions are intensively evaluated in comparison to the a priori emission inventories and other independent measurements.
General concept of satellite remote sensing of global aerosol emissions.
Figure 1 demonstrates the general concept of satellite remote sensing of the global distribution and strength of aerosol emissions. A priori emissions are used with the GEOS-Chem model to create a simulated aerosol loading, which are then compared to observed spectral AOD and AAOD from PARASOL. The inverse modeling based on adjoint GEOS-Chem iteratively optimizes a priori emissions to minimize the differences between observed and modeled AOD and AAOD. A posteriori emissions are equivalent to retrieved or optimized emissions, which are then input to the GEOS-Chem model simulation, and the results are verified with independent AERONET, MODIS and OMI aerosol products.
The GRASP algorithm implements statistically optimized fitting of diverse observations using the multi-term LSM (least square method) (Dubovik, 2004). The basic concept of this approach was introduced and implemented in the AERONET algorithm developed for aerosol characterization from ground-based radiometric observations (Dubovik and King, 2000; Dubovik et al., 2000; King and Dubovik, 2013). Dubovik et al. (2011, 2014) have adapted and extended this concept in GRASP algorithm, which is designed to retrieve aerosol and surface properties from satellite and other observations. As a new inversion development, a multipixel retrieval concept was implemented in the GRASP (Dubovik et al., 2011). Using this concept, the satellite retrieval is implemented as a statistically optimal simultaneous fitting of observations over a large number of pixels. This approach allows for improved accuracy of the retrieval by applying known a priori constraints on temporal or/and spatial variability of the derived parameters. In addition, GRASP is a highly versatile algorithm that has been applied for a large variety of different types of satellite, ground-based and airborne remote-sensing measurements by photometers, lidars, satellite sensors, nephelometers, sky cameras, etc. (Benavent-Oltra et al., 2017; Dubovik et al., 2019; Espinosa et al., 2017; Hu et al., 2019; Li et al., 2019; Lopatin et al., 2013; Román et al., 2017, 2018; Torres et al., 2017; Tsekeri et al., 2017).
The spaceborne multidirectional, multispectral polarized POLDER-3 (Polarization and Directionality of the Earth's Reflectances) imager on board PARASOL can measure the global angular distribution of intensity and polarization of solar radiation reflected to space by the earth–atmosphere system (Deschamps et al., 1994; Deuzé et al., 1999, 2001; Tanré et al., 2011). Throughout this article, we use “PARASOL” to denote POLDER/PARASOL observations. The GRASP algorithm inverts PARASOL comprehensive measurements to derive aerosol properties (e.g., extinction, absorption, size and composition) and surface BRDF (bidirectional reflectance distribution function) and BPDF (bidirectional polarization distribution function) properties. The development of the GRASP algorithm is described in Dubovik et al. (2011, 2014), and a description of some PARASOL/GRASP aerosol products can be found in Chen et al. (2018), Kokhanovsky et al. (2015), Popp et al. (2016) and Sayer et al. (2018). The accuracy of the GRASP algorithm configured for AERONET measurements has been evaluated with a laboratory experiment by Schuster et al. (2019).
We use PARASOL/GRASP Level 3 AOD and AAOD at six wavelengths (443, 490, 565,
670, 865 and 1020 nm) in this study. The Level 3 PARASOL/GRASP aerosol
products are rescaled in
The GEOS-Chem chemical transport model simulates the spatial and temporal
mass distribution of each aerosol species by modeling transport processes
(e.g., advection, convection, diffusion, deposition) and source
injection (Bey et al., 2001; Brasseur and Jacob,
2017; Jacob, 1999). Once the global distribution of aerosol mass is known,
it is generally converted to distributions of aerosol extinction (AOD) and
absorption (AAOD) by modeling the aerosol microphysical and optical
properties (Martin et al., 2003). For this study, we
simulate five major aerosol components, including sulfate (SU), BC, OC, DD
(seven bins, with effective radii of
The inverse modeling framework used here was originally presented by Chen et al. (2018), wherein BC, OC and DD aerosol emissions were estimated simultaneously using spectral AOD and AAOD observations. The framework uses an adjoint of the GEOS-Chem model that was developed by Henze et al. (2007, 2009) and Wang et al. (2012). Here, SU and SS emissions are kept fixed (similar to Chen et al., 2018); in future studies, we plan to retrieve SU and SS emissions together with other emissions by using additional spatial and temporal (smoothness) constraints (Dubovik et al., 2008).
Our inverse modeling method iteratively seeks adjustments of aerosol
emissions that can minimize the cost function
The inversion is initialized using a priori model emissions plus a spatially uniform
value of 10
The inversion system derives daily total BC, OC and DD aerosol emissions for
each grid box. The daily ratio between biomass burning and anthropogenic
contribution for BC and OC and the proportion of DD bins for each grid box
are kept as the a priori GEOS-Chem assumption. Distinguishing the anthropogenic
contribution from total emission is crucial for climate effects evaluation.
Here, we propose a simple method to estimate daily anthropogenic BC
(
The error covariance matrix
The error covariance matrix
We applied our method to retrieve BC, OC and DD daily emissions on a global
basis using PARASOL spectral AODs and AAODs for the year 2010. The retrieved
emissions dataset is publicly available at the LOA website (
The retrieved aerosol emissions are 18.4 Tg yr
Total source strengths (unit: Tg yr
Anthropogenic (AN) and biomass-burning (BB) emissions
(unit: Tg yr
Spatial distributions of the a priori and a posteriori annual emissions of BC, OC and DD, and their differences (a posteriori minus a priori emissions), are shown in Fig. 2a, b and c for the year 2010. The a posteriori BC emissions are generally greater than the a priori BC emissions throughout the globe; the a posteriori increases are particularly significant over certain regions, such as Southeast Asia and central and northwest China. However, there are also notable decreases of BC a posteriori emissions (with respect to a priori emissions) observed in several grid boxes over South America and north China. The largest increase in OC emissions is in southern Africa, where biomass burning is the predominant source. Consistent with BC, there is also a slight decrease in OC emissions over the high-emission grid boxes in South America. In contrast, the a posteriori DD emissions are reduced throughout the global desert regions. As a reference, the seasonal cycle of the a posteriori BC, OC and DD emissions are presented in the Supplement illustrations in Figs. S2–S4.
Notably, the a posteriori BC and OC emission distributions are more
homogenous than the a priori emission inventories. This phenomenon is
probably because the emissions are reported for
Global distribution of emissions for 2010 for BC (left
column), OC (middle column) and DD (right column) based on
Comparison of monthly emissions for a priori and a
posteriori datasets:
Comparisons of monthly global total BC, OC and DD emissions between a priori and a posteriori emission inventories for 2010 are shown in Fig. 3. Both a priori and a posteriori BC and OC emission inventories show a maximum in August and September, while the second peak in March observed in the a priori database shift to April and May in the a posteriori database. However, a posteriori BC and OC emissions are higher than the a priori emissions throughout the year. The a posteriori to a priori ratio for monthly BC emissions is up to 4.4 in April and down to 1.1 in July. Meanwhile, the a posteriori to a priori ratio for monthly OC emissions is up to 4.6 in October and down to 1.4 in July. In contrast, the a posteriori DD emissions capture seasonal variations that are similar to the a priori DD emissions. The a posteriori DD emissions are reduced consistently throughout the year. The posteriori and priori monthly DD emissions ratio slightly varies between 0.51 (December) and 0.78 (June).
In a numerical modeling experiment for the year 2010, we used the a priori and a posteriori emissions inventories as inputs to the GEOS-Chem model. Then we compared the simulated AODs and AAODs that were generated from the a priori emissions inventory to the AODs and AAODs that were likewise generated from the a posteriori emissions inventory. The results are shown in Figs. 4 and 5 for AOD and AAOD at 550 nm.
Both simulations with the a priori (Fig. 4a) and a posteriori (Fig. 4b) emissions show that high values of AOD appear over the Sahara in north Africa and the Taklimakan and Gobi deserts in Asia, which are strong dust source regions. Figure 4 also indicates high values of AOD in East Asia, where anthropogenic aerosols (e.g., BC, OC and SU) are the predominant components.
One of the major differences (Fig. 4c) between a priori and a posteriori AOD is that the a posteriori AOD over desert regions is reduced, especially over the Sahara; meanwhile, the a posteriori AOD increases over industrial and biomass-burning regions. The strongest increase of a posteriori AOD occurs in southern Africa, where it is associated with biomass-burning emissions.
Figure 5 shows the comparison of global distribution of a priori (Fig. 5a) and a posteriori (Fig. 5b) AAOD in 2010. Figure 5c clearly reveals that the a posteriori AAOD is higher than the a priori AAOD throughout the globe, except over the Sahara. The differences are high over southern Africa, India, Southeast Asia and central China, where they are associated with biomass-burning and anthropogenic emissions.
The statistics of global annual mean AOD and AAOD at 550 nm for the five
major aerosol components using a priori and a posteriori simulations are
shown in Table 2. The a priori GEOS-Chem global mean AOD at 550 nm is 0.105,
while the a posteriori simulation showed a slightly increased value of
0.119. The dust AOD decreases from 0.031 to 0.019 and its relative
contribution to total AOD decreases from 29.9 % to 16.1 %, owing to the
reduction of global DD emission from 1269.4 Tg yr
In general, the global mean AAOD at 550 nm significantly increases from
0.0039 (a priori) to 0.0071 (a posteriori), i.e., by a factor of 1.8. In
particular, DD and BC are the two major components to the total aerosol
absorption that collectively account for 90.1 % (a priori) and 88.2 % (a
posteriori) of total AAOD. The a posteriori DD AAOD decreases by 42.8 %
relative to the a priori simulation, from 0.0014 to 0.0008. The BC AAOD
increases from 0.0020 (a priori) to 0.0054 (a posteriori), a factor of
In comparison to the AeroCom Phase II assessments of global AOD and AAOD
(Myhre et al., 2013), our GEOS-Chem a posteriori global mean AOD is
Comparison of GEOS-Chem simulation of
global aerosol optical depth in the year 2010 at 550 nm based upon a priori and
a posteriori emission datasets:
Same as Fig. 4 but for AAOD.
To achieve a more robust evaluation, the a priori and a posteriori aerosol
properties simulated with the GEOS-Chem model are evaluated with other
independent measurements that are not used in our emission inversion. The
definition of the statistics used in the comparison, including correlation
coefficient (
The Aerosol Robotic Network (AERONET;
Holben et al., 1998) has
provided comprehensive and accurate aerosol data from a worldwide
ground-based sun photometer network for more than 2 decades. The data
products include measurements of multiple-wavelength AODs and
Ångström exponents (AExp). Additionally, the AERONET products also
include retrievals of AAOD, single scattering albedo (SSA), absorption
Ångström exponent (AAExp), size distribution, and complex refractive
index (Dubovik and King,
2000, Dubovik et al., 2000, 2002, 2006). The AERONET aerosol dataset has
been commonly used for satellite product validation and model evaluation for
a wide range of aerosol research topics. In this section, the AERONET
version 2 daily level 2.0 (Smirnov et al.,
2000) AOD, AAOD, SSA at 550 nm, AExp (440–870 nm), and AAExp (440–870 nm)
products are used to evaluate the a priori and a posteriori GEOS-Chem model
aerosol simulation. We convert spectral AOD and AAOD to SSA, AExp and AAExp
using the following equations:
Global mean AOD and AAOD at 550 nm of five major aerosol components. Numbers in parenthesis are relative contributions of different components to AOD and AAOD at 550 nm. Statistics are based upon a priori and a posteriori GEOS-Chem simulations for the year 2010.
Comparison of a priori and a posteriori GEOS-Chem
simulated AOD, AAOD, SSA at 550 nm, AExp (440–870 nm), and AAExp (440–870 nm) with AERONET Level 2 daily aerosol products. The data for density plots
are all aggregated into 100 bins for both
Figure 6 shows the comparison of a priori (red) and a posteriori (blue)
GEOS-Chem model simulated AOD, AAOD, SSA, AExp and AAExp with the AERONET
dataset. The a posteriori GEOS-Chem simulation has a much better agreement
with AERONET data for AAOD. The correlation coefficient (
The inversion framework derives aerosol emissions by fitting spectral AOD
and AAOD from PARASOL. The SSA, AExp and AAExp are the derived products from
spectral AOD and AAOD (Eqs. 9–11). The GEOS-Chem model SSA correlations with
AERONET improve slightly from
The differences between a posteriori and a priori
GEOS-Chem simulated AOD, AAOD, SSA, AExp and AAExp correlation coefficients
(
To better understand the regional performance of a priori and a posteriori GEOS-Chem simulations, we conducted the comparison of daily aerosol products of AOD, AAOD, SSA, AExp and AAExp over all AERONET sites with colocated data in 2010. Figure 7 shows the differences in correlation coefficients between the a posteriori and a priori simulations. The red circles indicate sites where the a posteriori simulation has higher correlation with AERONET than the a priori simulation. Alternatively, the blue circles indicate sites where the a priori simulation shows better correlation than the a posteriori simulations. There are 202 out of a total of 282 sites that show improved correlation for AOD using the a posteriori emission data. Additionally, 176 out of 272 sites show improvement for AExp, 97 of 162 sites show improvement for SSA, 89 of 167 sites show improvement for AAOD, and 84 of 167 sites have a better correlation for AAExp. The a posteriori simulation loses correlation with AERONET for AAOD in central Eurasia and the western United States, which needs further investigation in future studies.
Intensive wild fire events over central Russia during the Summer of 2010
have been reported in several studies
(e.g.,
Chubarova et al., 2012; Gorchakova and Mokhov, 2012; Huijnen et al., 2012;
Péré et al., 2014; R'Honi et al., 2013). An increase in daily AOD is
observed at the AERONET site Moscow_MSU_MO
(55.707
Time serial plot of AOD and AAOD at
550 nm from AERONET (blue star), PARASOL/GRASP (red circles), a priori (black
line) and a posteriori (green line) GEOS-Chem simulations at
the Moscow_MSU_MO (55.707
In this section, we evaluate the agreement of a priori and a posteriori
GEOS-Chem simulated AOD at 550 nm with the Moderate Resolution Imaging
Spectroradiometer (MODIS) Collection 6 (C6) Level 3 merged aerosol products
(Sayer et al., 2014). Dark Target (DT) and Deep
Blue (DB) are two well known retrieval algorithms developed for processing
MODIS atmospheric aerosol products
(Hsu
et al., 2004, 2013; Kaufman et al., 1997; Levy et al., 2013; Remer et al.,
2005; Sayer et al., 2013; Tanré et al., 1997). The MODIS/Aqua merged
aerosol products combing the DB with DT land/ocean data provide more
gap-filled retrievals, which are suitable for model evaluation. In order to
colocate with model data, the 1
Comparison of a priori and a posteriori GEOS-Chem simulated seasonal AOD at 550 nm with MODIS C6 Dark Target and Deep Blue merged products.
Differences between a priori and a posteriori GEOS-Chem seasonal AOD with MODIS C6 merged products.
In order to assess the aerosol seasonal cycle and some peaks of aerosol loading, we focus on AOD seasonal pattern from MODIS, a priori and a posteriori GEOS-Chem simulations (Fig. 9), and their differences (Fig. 10). The a priori GEOS-Chem simulation strongly overestimates aerosol loading over the Sahara throughout the year; the aerosol loading over Sahara is reduced in the a posteriori GEOS-Chem simulation, while the simulated AOD is still slightly higher than the MODIS AOD. Over India, the a priori simulation underestimates aerosol loading during December–January–February (DJF), June–July–August (JJA) and September–October–November (SON), and the a posteriori simulation improves the consistency between model and observation especially in DJF and SON. The a posteriori model simulated AOD in JJA is still lower than MODIS aerosol products over India. High aerosol loading occurred over eastern China throughout the year, which can be inferred both from observations and simulations. While the a posteriori simulated AOD over eastern China is lower than MODIS data in JJA, over biomass-burning regions (e.g., southern Africa and South America) the model simulation shows a consistent seasonal variability with MODIS data and the peaks are in JJA and SON. One of the major discrepancies between the a priori model simulation and MODIS over biomass-burning regions is that the a priori model simulated AOD is lower than observations. The biomass-burning peak over South America is observed in SON by MODIS; however, the a priori model simulated AOD in JJA is higher than that in SON. The a posteriori simulation using retrieved emissions reduces this bias. The 1–2 month delayed biomass-burning peak inferred from observations has also been reported over Africa in a recent study by Zheng et al. (2018). Overall, the a posteriori model simulated AOD shows a better agreement with independent MODIS observations over southern Africa and South America, where the aerosol is associated with biomass-burning emissions. Over central Europe, MODIS observed a high aerosol loading event in JJA (wild fires events over central Russia in summer 2010; see Fig. 8), which is not well reproduced by the a priori GEOS-Chem simulation. The retrieved emission data help to improve the a posteriori simulation reporting high aerosol loading there; however, the a posteriori AOD is still somewhat lower than MODIS.
The statistics of a priori and a posteriori GEOS-Chem simulated AOD
evaluated with MODIS AOD are shown in Table 3. The evaluation was conducted
at 550 nm for daily products in the year 2010. The GEOS-Chem simulation shows a
better agreement with independent MODIS AOD from a priori (
Statistics for evaluation of a priori and a posteriori daily GEOS-Chem simulation with MODIS AOD and OMI AAOD in the year 2010.
In this section, we discuss the seasonal variability of AAOD from the Ozone
Monitoring Instrument (OMI) near-UV algorithm (OMAERUV) and the a priori and
a posteriori GEOS-Chem simulation, as well as their differences. The OMAERUV
algorithm uses two UV wavelengths to derive columnar AAOD
(Torres et al., 2007). The climatology of Atmospheric
Infrared Sounder (AIRS) carbon monoxide (CO) observations and aerosol layer
height information from the Cloud-Aerosol LIdar with Orthogonal Polarization
(CALIOP) are adopted in the latest OMAERUV algorithm
(Torres et al., 2013), and the assessment of the
OMI/OMAERUV aerosol products are described in Ahn
et al. (2014) and Jethva et al. (2014). The OMI/OMAERUV Level 3 aerosol
products with 1
Comparison of a priori and a posteriori GEOS-Chem simulated seasonal AAOD at 500 nm with OMI/OMAERUV products.
Differences between a priori and a posteriori GEOS-Chem seasonal AAOD with OMI/OMAERUV products.
Figure 11 shows seasonal variability of AAOD at 500 nm for OMI/OMAERUV (left
column), a priori GEOS-Chem (middle column), and a posteriori GEOS-Chem
(right column). The differences between a priori and a posteriori GEOS-Chem
AAOD with OMI are presented in Fig. 12. Here, the GEOS-Chem 500 nm AAOD is
interpolated using AAOD at two wavelengths based on absorbing
Ångström exponent,
The a posteriori GEOS-Chem simulation indicates moderate aerosol absorption
(AAOD
The statistics of a priori and a posteriori GEOS-Chem daily AAOD evaluated
with OMI AAOD are shown in Table 3. The a priori AAOD shows a similar
correlation coefficient (
This section describes an evaluation that we conducted for simulated surface
concentrations of BC mass. The Interagency Monitoring of Protected Visual
Environments (IMPROVE) is a network of in situ aerosol measurement sites
located in US national parks (Malm et al., 1994,
2003). We obtained the surface concentration measurements of BC at IMPROVE
sites and evaluated the a priori and a posteriori GEOS-Chem simulation of
surface BC at these sites for the year 2010. Results for the annual mean surface
concentration of BC are shown in Fig. 13. Here, we see that the a priori
GEOS-Chem surface BC concentration is lower than the IMPROVE data, with
NMB
Evaluation a priori
In this study, we have used PARASOL spectral AOD and AAOD generated by the GRASP algorithm to retrieve global BC, OC and DD aerosol emissions based upon the development of the GEOS-Chem inverse modeling framework. Specifically, PARASOL/GRASP AOD and AAOD at six wavelengths (443, 490, 565, 670, 865 and 1020 nm) were used to correct the aerosol emission fields using the inverse modeling framework developed by Chen et al. (2018). This resulted in improved global daily aerosol emissions at the spatial resolution of transport models.
The retrieved global annual DD emission of 731.6 Tg yr
We used the GFED v4s and HTAP v2 emission inventories for the a priori
GEOS-Chem simulation, where the global BC emission is 6.9 Tg yr
We introduced a method to separate anthropogenic and biomass-burning BC and
OC from retrieved total BC and OC emissions by using a priori daily
proportion. The retrieved anthropogenic BC emissions are 14.8 and 4.6 Tg yr
The resulting GEOS-Chem a posteriori annual mean AOD and AAOD using the retrieved emission data are 0.119 and 0.0071, respectively, at the 550 nm wavelength. These calculations indicate a decrease of 8 % for AOD and an increase of 69 % for AAOD with respect to the AeroCom Phase II multimodel assessment.
The fidelity of the results is confirmed by evaluating the a posteriori simulations of aerosol properties with independent measurements. Namely, in order to validate the retrieved emissions, the a posteriori model simulations of AOD, AAOD, SSA, AExp and AAExp were compared to independent measurements from AERONET, MODIS and OMI. We also note that the AERONET dataset is temporally more frequent than the PARASOL observations that we used to obtain the a posteriori emissions. The a posteriori GEOS-Chem daily AODs and AAODs show a better agreement (higher correlation coefficients and lower biases) with AERONET values than the a priori simulation. In addition, the a posteriori SSA, AExp and AAExp also show good agreement with AERONET data; this indicates that PARASOL provided sufficient constraints for fitting spectral AOD and AAOD, and that the retrieved emission dataset can provide reliable model simulations. Besides, a posteriori GEOS-Chem AOD and AAOD exhibit a similar seasonal pattern with the MODIS AOD and OMI AAOD, respectively, during all seasons, which indicates that the retrieved emissions are capable of capturing the major events (e.g., dust hot spots, biomass burning, anthropogenic activities). However, the a posteriori simulation overestimates AOD and AAOD over that Sahara dust source region, while underestimating AOD and AAOD over grid boxes located downwind over the Atlantic Ocean. This is probably caused by an overestimation of the retrieved DD emission over the Sahara combined with a GEOS-Chem removal process that may be too rapid (Ridley et al., 2012, 2016). During biomass-burning seasons (e.g., JJA and SON over South America and southern Africa), the a posteriori AOD shows good agreement with MODIS; meanwhile, the a posteriori AAOD is slightly higher than the OMI AAOD. This could be caused by light-absorbing OC that is not included in the simulation or the inversion. Light-absorbing OC is also known as brown carbon (BrC), and it is characterized by absorption that decreases from UV to mid-visible wavelengths (Feng et al., 2013; Lack et al., 2012). The lack of BrC in our framework could cause the retrieval to generate more BC in order to capture the observed aerosol absorption (since BC has strong absorption throughout the visible and near-infrared wavelengths; Sato et al., 2003). As a consequence, the a posteriori AAOD may be overestimated from mid-visible to near-infrared wavelengths when BrC is not included.
Our evaluation of BC surface concentrations with the IMPROVE network over the United States indicates the possibility to improve the aerosol mass simulation based on inversion of satellite-derived columnar spectral aerosol extinction and absorption. However, the a posteriori simulation shows overestimation of surface BC concentration over sites with low levels of BC, which is probably due to an overestimation of a posteriori BC emissions over low-loading regions or a modeled BC lifetime that is too long (Lund et al., 2018; Wang et al., 2014).
The result of this study is the satellite-based aerosol emission database that is commonly used as GEOS-Chem default input and was adjusted to tune the global POLDER/PARASOL observations of spectral AOD and AAOD. The analysis in the paper, as well as previous work by Chen et al. (2018), shows that the aerosol distribution modeled with the satellite-based aerosol emission improves the agreement of modeling results with the independent AERONET, MODIS, OMI and IMPROVE data. These validation results support the validity of the identified corrections of the emissions. Therefore, if this database is used for initialization of not only the GEOS-Chem model but also any other aerosol transport or GCM (global climate model), then the effects of these suggested significant corrections to the amount of mass of DD, BC and OC, and their spatiotemporal effects on the climate and environment, can be studied.
To recapitulate, we derived global BC, OC and DD aerosol emission fields in a GEOS-Chem modeling framework that was constrained with PARASOL/GRASP spectral AODs and AAODs. Our study shows that this method can be useful for improving global aerosol simulations with CTMs. In the future, we plan to use the entire PARASOL dataset to generate a satellite-based aerosol emission database; this is expected to improve multiyear aerosol simulations of AOD, AAOD, SSA, AExp and AAExp in CTMs. In addition, the efforts to better understand the aerosol life cycle at the process level (Textor et al., 2007) are essential to inversion of aerosol emission, aerosol prediction and aerosol climate effect evaluation.
The AERONET version 2 data are available at
The supplement related to this article is available online at:
CC, OD, DKH and TL contributed to the inversion algorithm development. DF, PL and AL prepared the aerosol dataset from POLDER/PARASOL generated by the GRASP algorithm. CC, TL and FD carried out the data processing. CC analyzed the results with contribution from OD, MC, DKH and GLS; LL, QH, PL, AL and BT participated in scientific discussions. CC and OD wrote the paper with inputs from all authors.
The authors declare that they have no conflict of interest.
This work is supported by the Laboratory of Excellence CaPPA – Chemical and Physical Properties of the Atmosphere – project, which is funded by the French National Research Agency (ANR) under contract ANR-ll-LABX-0005-01. We would like to thank the AERONET, ACTRIS project infrastructure, MODIS and IMPROVE teams for sharing the data used in this study.
This research has been supported by the French National Research Agency (grant no. ANR-ll-LABX-0005-01).
This paper was edited by Johannes Quaas and reviewed by two anonymous referees.