Long-term measurements of global aerosol loading and optical
properties are essential for assessing climate-related questions. Using
observations of spectral reflectance and radiance, the dark-target (DT)
aerosol retrieval algorithm is applied to Moderate Resolution Imaging
Spectroradiometer sensors on both Terra (MODIS-T) and Aqua (MODIS-A)
satellites, deriving products (known as MOD04 and MYD04, respectively) of
global aerosol optical depth (AOD at 0.55
Measurements of aerosol loading and optical properties are essential for many applications, including quantifying global direct aerosol radiative forcing for climate studies (e.g., Belloiun et al., 2005; Chung et al., 2005; Yu et al., 2006; Kahn, 2011; Boucher et al., 2013), investigating the effect of aerosols on cloud microphysical properties and lifetimes (Nakajima et al., 2001; Lohmann and Feichter, 2005; Koren et al., 2008, 2012; Niu and Li, 2012) and estimating global exposure to air pollution (van Donkelaar et al., 2010; Evans et al., 2012; Boys et al., 2014; Ma et al., 2016). Because aerosols vary significantly by region (Kaufman et al., 2002) and have a lifetime on the order of days (Haywood and Boucher, 2000; Croft et al., 2014), near-daily observations over the entire globe are necessary to characterize the global aerosol system. The Global Climate Observing System (GCOS, 2011, 2016) has designated particular aerosol parameters as essential climate variables (ECVs) for quantifying Earth's climate system and change. To be considered a viable Climate Data Record (CDR), an ECV must be measured globally, with specified accuracies, precisions, spatial and temporal resolution. The ECV also must be measured over the long-term (e.g., multiple decades).
Aerosol optical depth (AOD – a measure of column-integrated aerosol loading)
is a designated ECV. To meet requirements as a CDR (Popp et al., 2016; GCOS,
2016), AOD must be measured globally, with a spatial resolution of 10 km or
finer and accuracy better than 0.03 or 10 %. In addition, this AOD record
must be multi-decadal and drift less than 0.01 decade
Using remote sensing to detect changes or trends in the physical world (e.g., the ambient aerosol), however, requires confidence that the algorithms and retrievals are consistently applied and that the sensors themselves (e.g., calibration, sampling and orbital characteristics) are also consistent. Creation of long-term climate data records often requires the observations from different instruments and platforms to be combined, because a single instrument may not provide sufficient spatial, temporal, or long-term coverage (e.g., the Global Precipitation Climatology Project; Adler et al., 2018). As the community moves towards creating aerosol CDRs that span the lifetime of more than one sensor, we need to pay even more attention to systematic biases and offsets.
In this study, we compare the aerosol climatology from the two identically designed MODIS sensors that fly simultaneously for over 15 years. The specifications of the instruments are essentially identical (sensor characteristics, calibration methods), and the retrieval algorithms are identical. In Sect. 2, we show that there are systematic differences in the derived global aerosol products for Collection 6. Although each sensor shows an insignificant global drift, their differences appear as a small but statistically significant trend. More alarming is that the two data sets are offset from each other, on average by 13 % of their global mean. This is larger than the GCOS requirements for accuracy (GCOS, 2016) and will introduce greater uncertainty than needed in narrowing error bars on current estimates of global aerosol radiative forcing (Boucher et al., 2013). In Sect. 3 we sample model output data to show that the differences between Terra and Aqua aerosol climatology are most likely unphysical, and in Sect. 4 we tested two methods of calibration correction to reduce the problem. Section 5 offers a discussion and conclusion, including suggestions for future calibration efforts from a product-based perspective.
Terra and Aqua were launched in late 1999 and early 2002, and are polar-orbiting
sun-synchronous satellites. Terra (Aqua) has a 10:30 (13:30) local
equator crossing time and descends (ascends) on the sunlit part of
the Earth. From each satellite, MODIS observes top-of-atmosphere (TOA)
reflectance (solar origin) and radiance (terrestrial origin) in 36
wavelength bands (0.41 <
In terms of sensor specifications, including spectral wavelength characteristics, calibration methods and presentation of data and file formats, the two MODIS instruments (MODIS-Terra or MODIS-T and MODIS-Aqua or MODIS-A) are twins. MODIS data products (from raw data through high-level aggregations), MODIS-T and MODIS-A, are known as MOD and MYD, respectively.
As the flagship sensor aboard two high-profile satellites, MODIS is a complex
enterprise. Though the scientific literature is immense, most relevant
information can be gleaned across the myriad of NASA websites. The general
sensor concept and design are presented at
As briefly described on the MODIS design page
(
MODIS products are grouped together in collections, in that a consistent protocol is used to derive L1B data, and then consistent algorithms are used to derive the L2 products and L3 aggregations. The same combination of L1B, L2 and L3 production rules are maintained so that all data in a collection are created the same way. This includes production of new data (collected forward in time) known as “forward processing” and archived data known as “reprocessing”. Under a collection, the entire time series of derived parameters (e.g., AOD) from MODIS-T (2000–present; MOD04 product) and MODIS-A (2002–present; MYD04 product) should be consistent with each other and presumably provide consistent global climatologies of the parameters. The most recent complete collection is known as Collection 6 (C6), and it encompasses time series from both MODIS sensors. Collection 6.1, discussed in Sect. 5, began processing in late October 2017.
Aerosols are ubiquitous in the atmosphere, and there are multiple algorithms
for removing the aerosol effect (known as atmospheric correction, or AC) when
retrieving properties of land (e.g., Vermote and Kotchenova, 2008; Lyapustin
et al., 2011) or ocean (e.g., Ahmad et al., 2010) surfaces. While these AC
algorithms report the aerosol information, they are not necessarily focused
on providing a global (land
The MODIS DT aerosol retrieval operates primarily by using observations from
the seven RSBs with spatial resolutions of 0.5 km or finer. These bands
are known as B3, B4, B1, B2, B5, B6 and B7; near
The DT algorithm (both land and ocean) follows a lookup table (LUT)
approach. This means that prior to retrieval, TOA
spectral reflectance (in a subset of the seven bands depending on the surface)
is simulated using scattering and radiative transfer codes (Wiscombe, 1980;
Dubovik et al., 2002; Evans and Stephens, 1991; Ahmad and Fraser, 1982).
These LUTs represent realistic combinations of aerosol, molecular and
surface reflectance, which during the retrieval are compared with the
observations. The solution is the LUT scenario (or multiple scenarios), which
minimizes the cost function. From the LUT, one infers the total column loading
(the aerosol optical depth or AOD or
The details of the retrieval algorithm have evolved over time. However, for
each collection, the same retrieval algorithms are applied to both MODIS-T
and MODIS-A. For Collection 4 (C4), Remer et al. (2006) compared the two data
sets and showed that they derived essentially the same monthly mean AOD over
ocean. For C5 data, however, Levy et al. (2010) noted that there were
discrepancies between the two data sets and that the MOD04 product appeared
to have a high bias compared to ground-based AERONET data in 2003 and a low
bias by 2008. There was no apparent overall bias to the MYD04 data. By 2013,
the C5 Aqua products continued to show little or no apparent AOD trend over
either ocean or land. However, Terra showed a
By the late 2000s, it was increasingly clear that, in addition to aerosols, other C5-derived data records were showing signs of non-physical trends (e.g., Lyapustin et al., 2014; Wang et al., 2012). The redundant on-board calibration protocol appeared to be insufficient for capturing degradation of the MODIS sensors, leading to artificial drifts in observed reflectance and subsequent derived geophysical parameters. To mitigate these drifts, MCST embarked on a new calibration protocol for Collection 6 (C6). In addition to regular observations of the moon and the on-board solar diffuser, MCST began monitoring observations over quasi-stable calibration desert targets, presumed to be nearly invariant (no rain, no changes in vegetation, etc.). Over such invariant Earth View (EV) targets, by compiling statistics of observed reflectance one could monitor long-term drifts in MODIS-observed reflectance. At the same time, the bidirectional reflectance function (BRDF) of such surfaces should be quasi-stable over time, so that, in addition to overall trending, MCST could characterize any trending response vs. scan angle (RVS) (Lyapustin et al., 2014; Toller et al., 2013; Sun et al., 2012; Wu et al., 2011). Corrections would be applied to any MODIS wavelength bands and scan angles that appeared to be drifting by more than 2 % since the first year of each mission. For MODIS-T, nearly all visible bands were drifting, with the shortest wavelengths drifting more rapidly. By the early 2010s, the shortest wavelength (e.g., blue bands) for MODIS-A also required correction.
In synergy with the overhaul of the upstream calibration method, the aerosol
retrieval was updated for C6. Levy et al. (2013) introduced changes to the
land/sea masking, the upstream cloud mask (e.g., MxD35; Frey et al., 2008) as
well as the ancillary data inputs. There were also changes to the aerosol
retrieval algorithm: some were made in response to upstream changes but
others improved the physical aerosol retrieval. However, while there may be
differences in the detrending coefficients applied to each sensor and a
particular band may require detrending, the C6 aerosol algorithm is applied
independently of upstream processing. The aerosol retrieval for C6, while
different from C5, is applied in the same way to both sensors. This results
in two sets of level 2 aerosol products (MxD04_L2) which are aggregated into
daily (MxD08_D3) and monthly (MxD08_M3) gridded products, following the
level 3 protocol for aerosol (L2
Yet, large positive offsets (MOD04 – MYD04) remain in C6. From Fig. 1, we
see that over land this offset averages about
Time series of Collection 6, monthly global mean AOD (at
0.55
The offsets to AOD and spectral AOD appear to be pervasive globally and have a significant enough magnitude to be of concern in the creation of climate data
records. The accuracy of satellite retrievals is generally assessed through
comparison with the ground-based sun photometer aerosol measurements from the
AErosol RObotic NEtwork (AERONET; Holben et al., 1998), where the AERONET
measurements are considered the “truth” with an uncertainty of
Summary of scatterplots (not shown) of collocated MODIS and AERONET
measurements of AOD at 0.55
While some AERONET sites exhibit a diurnal cycle, Kaufman et al. (2000) show that using AERONET data sampled at the MODIS passing time, the global AOD diurnal cycle is within 2 % of the daily mean AOD. This difference is at the same magnitude or much smaller than the discrepancies between Terra and Aqua retrievals that we discovered in this study, depending on the time span of AERONET data used. Although aerosol regional diurnal cycles may range more widely depending on location and/or season (Smirnov et al., 2002; Zhang et al., 2012), we expect global differences between morning and afternoon to be less than the offsets to MODIS seen here. To test this, we separately compare each MODIS data set to AERONET.
Table 1 summarizes the statistics of scatterplots (and linear regressions),
comparing each MODIS C6 data set with appropriate AERONET data from the period
2003–2014. Obviously inappropriate collocation sites are excluded (e.g., Mauna Loa at
elevation 3397 m being compared with sea-level retrievals), and
data are filtered by QAC recommended by Levy et al. (2013). From these
statistics, we see that, while both Terra- and Aqua-retrieved data sets
perform similarly (
We note that there are more collocations (
In coordination with Table 1, Fig. 2 provides the MODIS–AERONET differences as a function of AERONET-measured AOD. The bins of AERONET AOD are set so that there are a nearly equal number of points in each bin. Over both land (panel a) and ocean (panel b), the biases for MOD04 (red dots/shaded envelopes) are larger than that of MYD04 (blue dots/shaded envelopes). Over ocean, both products appear to have positive biases at low AOD over ocean, which is due to DT-O not being allowed to retrieve zero or negative AOD (leading to automatic positive bias). Over land, while both products have a median positive bias, it is larger for MOD04. The difference in the median bias (blue and red dots) is relatively constant across all AOD bins over land but increases with AOD over ocean. Overall, the difference in median bias (each MxD04 collocated with AERONET) is roughly equal to the overall offset in AOD between the two MxD04 time series. The statistics of each MODIS-retrieved AE compared to AERONET are very similar, as shown in panel c, except that AE from MOD04 is lower (by about 0.05) compared to MYD04.
Statistics of MxD04–AERONET difference as a function of AERONET
values for AOD (0.55
It is still possible that offsets between the two MODIS time series are tied
to unequal sampling of heavy aerosol events across the globe. To compare
offsets as a function of location, we focus on 2008 data. Here, we derive a
yearly mean AOD (per grid box) from monthly mean data, assuming valid data in
at least 2 months (e.g., L2
Except for the well-known aerosol hotspots (African dust/smoke, Asian
pollution/dust, etc.), Fig. 3a shows that most of the globe experiences annual
mean low AOD (
Gridded (0.625
As mentioned in Sect. 2.1, the DT retrieval over ocean reports spectral AOD.
Using annual mean AOD at 0.86 and at 0.55
Figures 3 and 4 demonstrate that the AOD from MODIS-T is consistently higher than MODIS-A and AE is consistently lower, and the uniformity of the offsets is suspicious. However, it is interesting that the southeastern Atlantic downwind of the southern African savanna (Meyer et al., 2013) shows opposite AOD and AE offsets.
Gridded (0.625
Section 2 identified significant differences between the aerosol products
derived from Terra and Aqua. There could be many causes for these
discrepancies, from instrument calibration and sampling to physical causes.
While MODIS-T and MODIS-A processing are identical, differences could arise
from differences in orbits and satellite overpass times. Terra is in
descending orbit with a daytime equator crossing (southward) at 10:30 local
solar time and Aqua is in ascending orbit (northward) at 13:30 LST. Because
of the different headings, although the local overpass time difference is
3 h at the equator, it is closer to 1.5 h in the Northern Hemisphere (NH)
midlatitudes and 4.5 h in the SH (Fig. 5). Therefore, it is possible that
different aerosol statistics might arise, whether due to diurnal cycles of
aerosol or clouds (leading to different sampling). Interestingly, because of
the symmetry of the orbits (
Gridded average MODIS local observation time (LST) for
Aqua
Cloud types and cloud properties show significant diurnal variation (e.g.,
Eastman and Warren, 2014). In fact, King et al. (2013) catalogued differences
between cloud statistics from the two MODIS sensors, which can be repeated
using C6 MODIS data. To explore the differences identified in Sect. 2 and to
discount the possibility of diurnal sampling (related to cloud fraction
differences) being the root cause of the Terra–Aqua offset, we use results
from aerosol simulations performed with the NASA Goddard Earth Observing
System, version 5 (GEOS-5) Earth system model (Molod et al., 2015). GEOS-5 is
run here in a replay mode, using winds, temperature and pressure fields from
the recent Modern-Era Retrospective analysis for Research and Applications,
version 2 (MERRA-2) joint aerosol and meteorological reanalysis (Gelaro et
al., 2017; Randles et al., 2017). Using MERRA-2 meteorological constraints
ensures simulation of real weather events and realistic cloud fields. The
model is run globally at a c180 horizontal resolution
(
Consider the partial orbits of Aqua and Terra around 12:00 UTC, specifically, the samplings of each MODIS between 11:30 and 12:30 UTC (Fig. 6). The light colors represent the swath of the MODIS track, and the dark colors represent where AOD was retrieved. Not only do the two tracks cover different parts of the world, each DT product retrieves less than 10 % (due to clouds, glint, bright surfaces, etc.) of the possible opportunities along the swath. Repeating the analysis of Fig. 6 at each hour for the entire year 2008 leads to two aggregations of the model for each satellite: the first represents the full MODIS swaths and the second represents the retrieval of the MODIS-DT products.
Figure 7 is analogous to Figs. 3 and 4, except that instead of aggregating
MYD04 versus MOD04 products, we have aggregated MERRA-2 outputs along the
full MODIS swaths. Figure 7a and c show global AOD (at 0.55
From Fig. 7, we can make some generalizations. First, the general patterns of the afternoon AOD (Fig. 7a) are similar to the aggregated MYD04 DT data (Fig. 3a). However, there are no gaps, because there are no DT retrieval decisions (e.g., masking). Most importantly, unlike the MODIS retrieval product (Fig. 6b), there is no overall AM–PM offset to the AOD (Fig. 3b). There are, however, regional differences to the offsets. Morning AOD tends to be lower for the biomass-burning regions over land, which is expected due to diurnal cycle of fire emissions (Boschetti and Roy, 2009). Over the ocean, there is even less variability from zero offset. For over-ocean AE (Fig. 7c), although the general patterns are similar to the MODIS retrieval products (Fig. 4a), the model outputs show lower AE, suggesting that the model has simulates larger particles than the retrievals. The outflow from equatorial Africa is one exception; the model reports much higher AE than the MODIS product, suggesting finer-sized particles. In addition, there are small positive and negative AM–PM differences, with no apparent systematic pattern. Generally, by comparing the model sampled in the morning to the afternoon sample, we see little evidence of global offsets to either AOD or AE.
Map of hourly swath and retrieval aggregation during
Gridded (0.625
Gridded (0.625
However, due to clouds, glint and bright surfaces, less than 10 % of the area sampled is actually retrieved by the DT algorithm. Due to differences in cloud fraction between morning and afternoon orbits (e.g., King et al., 2013), there may be systematic differences in the aerosol sampling. For example, while heavy smoke conditions (high AOD, high AE) may be present throughout the day, preferentially cloudy conditions (AM or PM) would affect the sampling of these aerosol events and thus the AM–PM offsets. Figure 8 is analogous to Fig. 7 but represents the model being sampled where and when there is AOD reported in the MYD04 or MOD04 products. The overall AOD and AE patterns (e.g., Fig. 8a and c) are much like those from the entire swath (Fig. 7a and c) but with gaps exactly like the satellite retrievals (Figs. 3a and 4a). Comparing Fig. 8b and d with their counterparts in Fig. 7 shows that, by imposing satellite sampling, the variability of both AOD and AE offsets more resembles the satellite regional distributions. However, the mean offsets to both AOD and AE have not increased due to the imposition of satellite sampling. Sampling alone cannot explain the overall offset seen in the satellite data products.
Except for the Amazon region, where both show negative offsets, there is not much resemblance between the AOD differences shown in Figs. 8c and 3b. Since Fig. 8c represents the expected offsets, the overall positive offset in Fig. 3b has probably masked some of the diurnal cycles expected in that figure. The overwhelming positive offset in Fig. 3b, especially over the oceans where the model shows very little difference, indicates there is a systematic difference in the two retrievals that could only be attributed to instrument calibration.
The C6 MODIS products report persistent systematic offsets in the AOD and AE that cannot be explained by diurnal sampling differences, as was explored in the modeling exercise of Sect. 3. The next possible explanation for the offset is calibration. As explained by Lyapustin et al. (2014), although the MODIS Characterization Support Team (MCST) updated MODIS calibration to account for the severe trending observed in C5 data, there still may be offsets in C6.
The MODIS DT retrieval algorithm is an inversion on multi-spectral data. The
reality is that, due to the retrieval being a multi-channel inversion, changing
one wavelength at a time leads to nonlinear changes in retrieved AOD and AE.
However, the over-ocean-retrieved AOD is most sensitive to changes in the
0.86
The spectral channels used in the retrieval algorithm are calibrated
independently for each sensor and may drift differently over time. Based on
monitoring bidirectional-reflectance function (BRDF) over the same
pseudo-invariant (remote desert) surfaces as used by MCST, Lyapustin et al. (2014) devised a method for correcting the L1B reflectances. This method,
known as C6
Impact of applying C6
The C6
Lyapustin et al. (2014) presented formulas (polarization
Impact of applying C6
Figures 9 and 10 show absolute difference (
Impact of applying C6
Histograms of the C6
The C6
These correction characteristics clearly should have implications for the
retrieval of global AOD and AE during 2008. We apply the C6
Because the C6
Applying the C6
For example, Doelling et al. (2015) compared the two MODIS sensors using
observations from nearly simultaneous nadir overpasses (NSNOs). Although
Terra and Aqua have orbits in opposite directions, their orbits nearly cross
each other 14 times per day. Nadir crosses only happen at
Impact of the C6
Overall, the mean scaling providing Table III of Doelling et al. (2015)
appears to be similar in polarity to the cross-calibration factors provided by
Table 3 of Lyapustin et al. (2014). For example, both studies suggest that
in B3 (0.47
We used Table III of Doelling et al. (2015) to estimate scaling coefficients
appropriate for 2008, and we tested them by applying them directly within the aerosol
retrieval (rather than the upstream C6
Differences between the C6.1 and C6 MOD04 (Terra) product for
AOD
The radiometric calibration for C6 was based on the combination of prelaunch, solar diffuser, moon observations and selected targets on Earth. When C6 processing began in 2012, calibration coefficients were derived in order to smoothly connect the beginning of the mission through to 2012 (reprocessing). As long as the instrument performance did not change too quickly, the C6 methodology could be used for forward processing. However, by early 2016, some of the TEBs for MODIS-T were becoming unusable.
Since the beginning of our analysis of the C6 differences between Terra and
Aqua aerosol products, the MODIS team has released an updated collection
denoted by C6.1
(
Figure 14 shows the differences (panel a: AOD and panel b: AE) between C6.1 and C6 for MOD04 (Terra) during 2008, showing only small changes in global AOD and AE. On average, AOD increased over ocean by about 0.001 and decreased over land by similar magnitude, which are much less than the desired changes (e.g., Figs. 3b and 4b). The changes from C6 to C6.1 may alleviate some of the AOD offsets over land in 2008 (maybe as much as 10–20 % of the bias in some places) but will exacerbate the bias over ocean by about the same percentage. Likewise changes from C6 to C6.1 will only affect the annual mean biases in AE by 10–20 %, both positively and negatively. The changes introduced by C6.1 are just too small to eliminate the Terra–Aqua differences that were identified and explored in the analysis presented above.
We note that the changes to C6.1 L1B products are temporally dependent, so we might expect larger differences between C6 and C6.1 in later years (especially after 2015). We also have not yet analyzed C6.1 MYD04 data (Aqua C6.1 reprocessing began on 28 December 2017). Thus, there may be slightly different consequences to the aerosol products than are shown here for 2008. However, given the small magnitude seen in 2008 and expected through the entire time series, it is unlikely that the C6.1 changes will provide the fix necessary to bring the Terra and Aqua aerosol products into agreement.
The DT aerosol retrieval has been applied to MODIS-T data since 2000 and
MODIS-A data since 2002. Time series of the C6 products (MOD04 and MYD04)
are almost in lockstep (Fig. 1). However, compared to MODIS-A (afternoon
overpass), the global mean MOD04 (morning overpass) shows consistently
higher AOD at 0.55
Focusing on 2008, we studied the AOD offset. Over ocean, the offset appears everywhere, regardless of the overpass time difference (4.5 h in SH midlatitudes, 1.5 h in NH). Over land, there is more variability in the offset, but only known biomass-burning regions display a negative offset (morning AOD is lower than afternoon). Over ocean, we also see that there are consistent offsets in the spectral AOD, as demonstrated by the Ångström exponent (AE), showing that MODIS-T reports lower AE globally by about 0.05.
We used the GEOS-5 replay model output to question the observed global offsets in AOD and AE. When we sample the model along the MODIS swaths (Terra and Aqua separately, then take the differences), most of the globe appears to have no AOD or AE offsets. However, we might expect to observe negative offsets (morning AOD lower than afternoon) in the biomass-burning regions. As we sample the model only for the MxD04 retrievals (MOD04 and MYD04 separately and then take the differences), we see increasing variability in both AOD and AE offsets. This is due to differences in cloudiness between the morning and afternoon, which is convolved into the MODIS data. There is the suggestion of a more generalized offset to AE but not approaching the magnitudes seen from the satellite retrieval products.
Of course, we cannot yet rule out other physical reasons for the offsets. For example, although the retrieval algorithm corrects for gas absorptions (column water vapor, ozone, etc.; Patadia et al., 2018), unaccounted differences between morning and afternoon (for example if 12:00 UTC water vapor was assumed for both 10:30 and 13:30 overpasses) could lead to systematic biases in retrieved AOD. We should consider that the aerosol optical properties themselves (e.g., refractive index, size/shape distribution) could be wrong and also lead to generalized AOD bias compared to AERONET (e.g., Ichoku et al., 2003; Eck et al., 2013). Additionally, if there were differences in optical properties which were not accounted for and due to very late-morning cloud processing (e.g., Eck et al., 2012), one might see offsets between AM and PM, and different offsets between AM and PM versus AERONET. Although the current modeling framework (e.g., our MERRA-2 sampling) does not suggest that it leads to a global offset, this is definitely a topic for further study.
Since the Terra–Aqua bias is so similar to the difference between
Terra–AERONET and Aqua–AERONET (Terra–Aqua
The MODIS Atmosphere Science Team recently began processing the C6.1 family
of products, primarily to address issues related to thermal infrared bands
and impacts on the standard cloud mask for MODIS-T. There was no major change
to the methodology of the MCST reflective band calibration for MODIS-T.
Except for improvements over urban regions, the C6.1 aerosol retrieval is
also nearly unchanged. Thus, based on comparing the C6.1 aerosol product with
C6 during 2008, we expect there to be no change to the overall offsets to
both AOD and AE. However, since 2011 (beginning of C6 processing), additional
reflective bands (on both MODIS-T and MODIS-A) have strayed more than
2 %, so there are revisions to overall calibration that may show apparent
effects of the two time series in later years (well after 2008)
(
In the meantime, users of the products should not interpret differences
between Terra and Aqua aerosol products as representing a true diurnal
signal in the aerosol, unless magnitudes of the observed signal greatly
exceed the biases described here. However, because collocated comparisons
between MODIS aerosol retrievals and AERONET observations show Terra with a
larger high bias, the recommendation is to rely more on Aqua retrievals for
quantitative long-term climate-related applications. On the other hand, we
note that the bias in AOD is only
For accessing information (including doi information and
links for downloading) on the MODIS aerosol product (MxD04) and gridded
product (MxD08), please use
RCL is the PI of the MODIS DT project and led the research effort.
SM is the lead programmer for the MODIS-DT project; she maintained the
C6/C6.1 products and ran the experiments with the different calibrations. VS
and YS performed the analyses and prepared most of the figures. PRC provided
the MERRA/GEOS-5 model output and interpretation. AIL and YW are responsible
for the C6
The authors declare that they have no conflict of interest.
This work was supported by the NASA ROSES program NNH13ZDA001N-TERAQEA: Terra and Aqua – Algorithms – Existing Data Products and NASA's EOS program managed by Hal Maring. We thank MCST for their efforts in maintaining and improving the radiometric quality of MODIS data, and LAADS/MODAPS for the continued processing of the MODIS products. We are grateful to Yaping Zhou and Falguni Patadia (Morgan State University/GSFC) and Pawan Gupta (USRA/GSFC) for reviewing early drafts of this paper. Edited by: Marloes Gutenstein-Penning de Vries Reviewed by: Pekka Kolmonen and two anonymous referees