1. science
2. ecology
3. pollution

PUBLICATIONS
Geophysical Research Letters
Lifetimes and emissions of SO2 from point sources
estimated from OMI
RESEARCH LETTER
10.1002/2015GL063148
Key Points:
• A new method to estimate SO2
lifetimes and emissions from
point sources
• Sources with annual SO2 emissions as
1
low as 30 kt yr can be detected
from OMI
• There is a high correlation between
reported SO2 emissions and
OMI estimates
Supporting Information:
• Text S1, Table S1, and Figures S1–S9
Correspondence to:
V. E. Fioletov,
vitali.fi[email protected]
Citation:
Fioletov, V. E., C. A. McLinden, N. Krotkov,
and C. Li (2015), Lifetimes and emissions
of SO2 from point sources estimated
from OMI, Geophys. Res. Lett., 42,
doi:10.1002/2015GL063148.
Received 16 JAN 2015
Accepted 29 JAN 2015
Accepted article online 2 FEB 2015
V. E. Fioletov1, C. A. McLinden1, N. Krotkov2, and C. Li2,3
1
Environment Canada, Toronto, Ontario, Canada, 2Atmospheric Chemistry and Dynamics Laboratory, NASA Goddard Space
Flight Center, Greenbelt, Maryland, USA, 3Earth System Science Interdisciplinary Center, University of Maryland, College
Park, Maryland, USA
Abstract A new method to estimate sulfur dioxide (SO2) lifetimes and emissions from point sources using
satellite measurements is described. The method is based on fitting satellite SO2 vertical column density to a
three-dimensional parameterization as a function of the coordinates and wind speed. An effective lifetime (or,
more accurately, decay time) and emission rate are then determined from the parameters of the fit. The method
was applied to measurements from the Ozone Monitoring Instrument (OMI) processed with the new principal
component analysis (PCA) algorithm in the vicinity of approximately 50 large U.S. near-point sources. The obtained
results were then compared with available emission inventories. The correlation between estimated and reported
emissions was about 0.91 with the estimated lifetimes between 4 and 12 h. It is demonstrated that individual
sources with annual SO2 emissions as low as 30 kt yr1 can produce a statistically significant signal in OMI data.
1. Introduction
Sulfur dioxide (SO2), a designated criteria air pollutant in many countries, plays an important role in Earth’s
chemistry and climate. It forms sulfate aerosols that influence weather and climate and lead to acid
deposition through formation of sulfuric acid. Volcanoes and anthropogenic sources such as coal-burning
power plants, smelters, and oil refineries are the primary emitters of SO2 into the atmosphere.
Satellite observations have been used to monitor anthropogenic SO2 emissions since 1995 [Streets et al., 2013,
and references therein]. Inferring the emission strength (E) requires knowledge of the total SO2 mass (m) near
the source and its lifetime, or more accurately, decay time (τ). Assuming a steady state, these quantities are
related through E = m/τ. While m can be directly derived from satellite measurements, a determination of τ is not
as straightforward. Fioletov et al. [2011] related Ozone Monitoring Instrument (OMI) measurements of the SO2
mass over large U.S. power plants to their bottom-up emissions to derive a value of τ. Once τ was known, it was
used to estimate the emission rates for other sources seen by OMI [Lu et al., 2013; Fioletov et al., 2013]. An
alternative approach is a direct estimate of τ based on the rate of decay of the vertical column density with
distance downwind [Carn et al., 2013; Beirle et al., 2014]. Beirle et al. [2014], however, mentioned that it is
challenging to apply this approach based on the mean wind speed for locations with variable wind conditions or
interfering SO2 sources. Herein, a similar approach is adopted but one which is better suited for measurements
at locations experiencing variable wind speeds and directions. Furthermore, it is possible to derive m and τ
for relatively small anthropogenic sources by fitting a three-dimensional function of the coordinates and wind
speed. Model-based comparisons of different methods for E and τ estimates [de Foy et al., 2014] demonstrate
that such methods can produce accurate estimates of τ. As both parameters m and τ are to be derived from OMI
data, the suggested method can be considered as an independent source of information on emissions.
This is an open access article under the
terms of the Creative Commons
Attribution-NonCommercial-NoDerivs
License, which permits use and distribution in any medium, provided the
original work is properly cited, the use is
non-commercial and no modifications
or adaptations are made.
FIOLETOV ET AL.
Measurements from OMI were analyzed in this study. OMI, a Dutch-Finnish UV-visible wide field of view
“pushbroom” spectrometer flying on NASA’s Aura spacecraft [Schoeberl et al., 2006], provides daily global
coverage at high spatial resolution [Levelt et al., 2006]. OMI has the highest spatial resolution and is the most
sensitive to SO2 sources among the satellite instruments of its class [Fioletov et al., 2013].
2. Data Sets
2.1. OMI SO2 Data
The new generation operational OMI planetary boundary layer (PBL) SO2 data produced with the principal
component analysis (PCA) algorithm [Li et al., 2013] were used in this study. The PCA algorithm has been
©2015 Her Majesty the Queen in Right of Canada. Reproduced with the permission of the Minister of the Environment.
1
Geophysical Research Letters
10.1002/2015GL063148
demonstrated to largely eliminate biases and improve the precision by a factor of 2 as compared to the
previous operational OMI band residual difference (BRD) PBL SO2 product, providing a greater sensitivity
to anthropogenic emissions. In particular, the standard deviation of PCA-retrieved SO2 is ~0.5 Dobson unit
(DU, 1 DU = 2.69 · 1026 molecule km2), half that of the operational OMI product (~1.0 DU). The BRD algorithm
uses a constant air mass factor (AMF) of 0.36 that is representative of typical summertime conditions in
the eastern U.S. [Krotkov et al., 2006]. While the PCA algorithm uses spectrally dependent SO2 Jacobians
instead of an AMF, its current version assumes the same fixed conditions as those in the BRD algorithm to
facilitate the comparison between the two algorithms (see Li et al. [2013] for details). The PCA retrievals can
therefore be interpreted as having an effective AMF of 0.36. Retrieved SO2 vertical column density (VCD)
values are given as total column SO2 in Dobson unit (DU).
OMI SO2 data are retrieved for 60 cross-track positions (pixels or rows), and the pixel ground size varies
depending on the track position from 13 × 24 km2 at nadir to about 28 × 150 km2 at the outermost swath angle.
Data from the first and last 10 cross-track positions were excluded from the analysis to limit the across-track
pixel width to about 40 km. Beginning in 2007, some track positions were affected by field of view blockage and
stray light (so-called “row anomaly,” see http://www.knmi.nl/omi/research/product/rowanomaly-background.
php), and the affected pixels were also excluded from the analysis. Only clear-sky data, defined as having a
cloud radiance fraction (across each pixel) less than 20%, were used. To exclude cases of transient volcanic SO2,
the range of analyzed values was limited to a maximum of 15 DU (see supporting information). Measurements
with snow on the ground were also excluded as the algorithm presently does not account for the effects of
snow albedo on SO2 Jacobians. Snow information was obtained from the Interactive Multisensor Snow and Ice
Mapping System [Helfrich et al., 2007]. Furthermore, wintertime measurements were excluded from the analysis
because they exhibit larger uncertainties due to various factors such as larger solar zenith angles and higher
ozone optical depth.
2.2. SO2 Emission Sources
The top 100 largest individual U.S. sources (according to the U.S. National Emissions Inventory for 2005, see
http://www.epa.gov/ttn/chief/eiinformation.html) were examined in this study. The majority of these sites
are coal-burning power plants with emissions in the range between 27 and 170 kt yr1. The inventory data for
these sources were based on direct stack measurements using continuous emission monitoring systems
as mandated by Title IV of the 1990 U.S. Clean Air Act Amendments (Public Law 101-549) (e.g., http://www.
epa.gov/air/caa/title4.html). For the comparison with satellite data, it was assumed that emissions from these
sources are constant throughout the year.
2.3. Wind Data
Wind speed and direction data are required to determine the lifetime. European Centre for Medium-Range
Weather Forecasts reanalysis data (http://data-portal.ecmwf.int/data/d/interim_full_daily) were merged
with OMI observations. Wind profiles are on a 6 h, 0.75° horizontal grid and are interpolated in space and
time to the location of the OMI pixel center. U and V (east-west and north-south, respectively) wind
speed components were averaged through the vertical to account for the vertical distribution of the SO2
profile. For this study, an average at the surface, 950 and 900 hPa levels (roughly over the lowest 800 m),
was used and then spatially and temporally interpolated to the location and overpass time of each OMI
pixel. Additional information on the wind data is available from the supporting information. All sites
analyzed in this study were located below 1000 m.
This merging allowed for the classification of observations according to wind direction, analogous to that
used in other studies [e.g., Beirle et al., 2011]. Instead of using this directional classification to examine the
distribution as a function of downwind distance, the approach adopted here involves a rotation of the
location of each OMI pixel about the source such that after rotation, all have a common wind direction
(see Pommier et al. [2013] for details). The rotation of the OMI pixels effectively redistributes the
observations near the sources along the upwind-downwind direction. In other words, with the rotation
applied, we can analyze the data assuming that the wind always has the same direction (see the
supporting information).
FIOLETOV ET AL.
©2015 Her Majesty the Queen in Right of Canada. Reproduced with the permission of the Minister of the Environment.
2
Geophysical Research Letters
10.1002/2015GL063148
Figure 1. (a) Mean total column SO2 (in DU) near Norilsk (69°N, 88°E) after rotation of all pixels in a upwind-downwind direction for 2005–2013 stratified by the
wind speed. The axis show the distance from the source in kilometers. (b) Mean total column SO2 near Norilsk for different wind speed groups for the area
within ±50 km across the wind direction (the white rectangle in Figure 1a) as a function of the distance from the source (negative for the downwind) after the
rotation of all pixels in a upwind-downwind direction. (c) Exponentially modified Gaussian function g(y,s), where y is the distance from the source and s is the wind
speed that represents the best fit to Norilsk data on the left. (d) The results of the fit by OMISO2 ¼ a f ðx; y Þgðy; sÞ calculated for different wind speeds as indicated
on the plot.
3. The Fitting Algorithm
For the illustration of the suggested approach, OMI measurements of SO2 near the smelters in Norilsk, Russia
(69°N, 88°E), were investigated. Collectively, they are one of the world largest antropogenic SO2 sources with
emission levels close to 2 megaton yr1 [Norilsk Nickel, 2009] and with no other significant sources in the
vicinity. Bauduin et al. [2014] discussed SO2 pollutions in Norilsk in details. They also pointed to some
inhomogeniety of the SO2 distribution due to the orography. We found that this inhomogeneity had a rather
limited impact on results presented here as discussed in the supporting information. As the SO2 signal from
the Norilsk smelters was very strong and the emissions were high during the entire 2004–2013 period
(see also Figure S4 in the supporting information), this source provides a good example to illustrate the SO2
distribution for various wind speed bins. Figure 1a shows the mean SO2 maps near Norilsk for the periods of
FIOLETOV ET AL.
©2015 Her Majesty the Queen in Right of Canada. Reproduced with the permission of the Minister of the Environment.
3
Geophysical Research Letters
10.1002/2015GL063148
2005–2013, stratified by the wind speed after rotation of all pixels as described above. The maps clearly
demonstrate the changes in the SO2 distribution with the wind speed with higher winds leading to less SO2
right over the source but a longer downwind tail. This is further illustrated by Figure 1b, where SO2 values,
averaged over ±50 km in the cross-wind direction, are shown for different wind speeds.
It is assumed that SO2 concentrations (emitted from a point source) decline with time (t) as exp(λt), i.e., with
a constant decay rate (τ = 1/λ). Then, in the absence of diffusion and with a constant wind direction and speed
(s), SO2 is transported downwind (along the y axis) with a concentration that declines exponentially with
the distance from the source. Since t = y/s (note that y is negative in the chosen coordinate system in
Figure 1), this decay is simply exp(λy/s) or exp(λ1y), where λ1 = λ/s. Likewise, in the absence of any wind, the
SO2 distribution near the source is governed by diffusion and can be described by a 2-D Gaussian function of the
distance from the source with one parameter σ. As both diffusion and exponential decay of the concentration
along the y coordinate take place, the overall behavior can be described as a combination of normal and
exponential random variables. The distribution function for such a combination is an exponentially modified
Gaussian function. Figure 1c shows the fit of data used for Figure 1b by an exponentially modified Gaussian
function of y and s with two unknown parameters (σ and λ) multiplied by a scaling factor a. As Figures 1b
and 1c show, the function captures the behavior of the observed SO2 values of the distance and wind speed, in
agreement with de Foy et al. [2014].
Thus, in order to describe the total amount of SO2 near a source, a product of Gaussian function
f(x,y) and exponentially modified Gaussian function g(y,s) was fit to OMI SO2 measurements:
OMISO2 ðx; y; sÞ ¼ a f ðx; y Þ gðy; sÞ, where
1
x2
pffiffiffiffiffiffi exρ 2 ;
2σ 1
σ 1 2π
2
λ1
λ1 ðλ1 σ 2 þ 2y Þ
λ1 σ þ y
pffiffiffi
gðy; sÞ ¼ exρ
erfc
;
2
2
2σ
( pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
σ 2 1:5y ; y < 0
σ1 ¼
;
σ; y ≥ 0
λ1 ¼ λ=s;
f ðx; y Þ ¼
(1)
x and y (in km) refer to the coordinates of the OMI pixel center across and along the wind direction respectively,
2
∞
s is the wind speed (in km h1) at the pixel center, and erfcðx Þ ¼ p2ffiffiπ ∫x et dt. The f (x,y) function represents
the Gaussian distribution across the wind direction line.
The function g(y,s) is essentially a convolution of Gaussian (determined by the parameter σ) and exponential
functions (determined by the parameter λ1) and represents an exponential decay along the y axis smoothed by
a Gaussian function: when σ is close to 0, then g(y,s) ≈ λ1exp(λ1y), where y ≤ 0. Note that the wind speed s is
included in equation (1) only through λ1 = λ/s. We also used σ 1 that increased with the distance from the source
instead of σ in f(x,y) to reflect the change in the winds between the source and the analyzed pixel that yields
an additional spread of the “plume” after the rotation of all pixels in a upwind-downwind direction. The rate
of the increase (1.5 y) in σ 1 was estimated from fits for several large sources. Using σ 1 instead of σ has practically
no impact on the estimated parameters but produces a more realistic fit. Also, it increases the number of
locations where the nonlinear regression estimates based on equation (1) converge.
Parameters σ, λ, and a were estimated from the fit of the observed OMI values by equation (1), i.e., by the nonlinear
regression model, by minimizing the sum of squares between measured and estimated values. In order to
produce a more uniform fit for the whole range of distances and wind speeds, the data were binned with a 3 km
distance increment and a 2 km h1 wind speed increment, and the averages for each bin were used for fitting.
Once parameters a, σ, and λ are estimated, the SO2 distribution as a function of x, y, and s (OMISO2(x,y,s)) can
be reconstructed. Figure 1d shows the result of such reconstruction for Norilsk for different values of the wind
speed s. For Norilsk, the parameter estimation was done using OMI pixels centered within a rectangular
area that spreads ±90 km across the wind directions, 90 km in upwind and 270 km in downwind directions,
and for the wind speed between 0.5 and 45 km h1. To further illustrate the suggested approach, we
applied it to measured SO2 values near the largest U.S. emission sources in 2005–2007. Unlike Norilsk, most of
FIOLETOV ET AL.
©2015 Her Majesty the Queen in Right of Canada. Reproduced with the permission of the Minister of the Environment.
4
Geophysical Research Letters
10.1002/2015GL063148
1
Figure 2. The mean column SO2 values (in DU) for 2005–2007 near the largest U.S. SO2 source (Bowen power plant in Georgia, 170 kt yr in 2005) located in
the center of each plot. The data are stratified by the wind speed, and the three rows represent different wind speed intervals as indicated. (left column) The
original data smoothed by pixel averaging with an 18 km window. (middle column) The same data after the rotation of all pixels in a upwind-downwind direction.
(right column) The fitting results for different wind speeds as indicated on the plot.
them are not isolated and have other sources in the area. The wind rotation approach can additionally be used
to better isolate a signal from a single source. If there is a secondary source located R (km) from the analyzed
one, the wind rotation turns that secondary source into a ring (with the radius R) of elevated SO2 values. As
the total mass is preserved, the amplitude of the SO2 signal would decline proportionally to 1/R (see also the
supporting information). As an illustration, Figure 2 shows the mean column SO2 values for the 2005–2007
period near the largest U.S. SO2 source (Bowen power plant in Georgia, 170 kt yr1) located in the center and
surrounded by other sources. The data are stratified by wind speed, and the three rows represent different
wind speeds bins. Figure 2 (left column) shows the original data smoothed by pixel averaging with an 18 km
window as described by Fioletov et al. [2011]. Figure 2 (middle column) shows the same data after the wind
rotation was applied. The wind rotation procedure makes the source in the center more pronounced. The
values over the other sources diminish substantially as the rotation spreads out the pixels with elevated SO2.
Figure 2 (left column) also shows that simple maps of the mean SO2 distribution can be misleading. The same
sources would produce larger signals in the areas of low wind speed compared to the areas of high wind
speed. The wind does not impact the emission rate, but it does spread SO2 over a larger area and reduces the
peak values.
FIOLETOV ET AL.
©2015 Her Majesty the Queen in Right of Canada. Reproduced with the permission of the Minister of the Environment.
5
Geophysical Research Letters
10.1002/2015GL063148
0.6
40
30
0.4
Sigma(km)
Lambda (1/hour)
0.5
0.3
0.2
0.1
10
0.0
a
-0.1
b
0
0
Estimated Annual Emissions(kT)
20
100
200
300
400
0
400
400
300
300
200
200
100
100
c
0
0
100
200
300
400
4
200
300
400
d
0
0
Reported Annual Emissions (kT)
Decay time (hours)
100
100
200
300
400
Reported Annual Emissions (kT)
6
8
12
Figure 3. Scatterplots of parameters (a) λ and (b) σ estimated from equation (1) as functions of reported annual SO2
emission from the largest U.S. sources in 2005. (c) Annual SO2 emissions calculated from OMI data for 2005–2007 using
equation (1) with the decay time (τ = 6 h) estimated from the best fit of the reported emissions and OMI SO2 integrated
around the source (parameter a). (d) The same as Figure 3c but with emissions calculated from OMI data directly from
1
equation (1) as aλ, i.e., with both parameters estimated from the fit. Emissions are given in kt yr calculated assuming a
constant emission rate. The slopes of the linear regression line for data shown in Figure 3d is 1.13 ± 0.03 (one sigma). The
error bars represent the one sigma confidence intervals. The different colors indicate the estimated decay times (τ = 1/λ)
rounded to the nearest even integer.
Available satellite SO2 data products are affected by local biases [e.g., Fioletov et al., 2013]. While the PCA data
product has much smaller biases than other data sets, a bias correction was useful in removing the impact
of the slightly elevated background SO2 in regions that contain several SO2 sources, particularly along the
Ohio River. To calculate the local biases, SO2 values from the pixels centered between 50 and 150 km away
from the source on the upwind side were averaged. Local biases were calculated for each year and then
removed from the data before the analysis.
The fitting procedure described above was applied, σ, λ, and a were estimated for major U.S. sources, and the
OMISO2(x,y,s) function was reconstructed. Figure 2 (right column) demonstrates the reconstruction results
for Bowen for different wind speeds as indicated on the plot. The comparison of the middle and right
columns of Figure 2 shows that the fitting captured the magnitude and the spread of the SO2 plume
correctly. As individual U.S. sources are about 10 times smaller than Norilsk, the fitting was done using an
area that was smaller than for Norilsk. For all U.S. sources, the fitting was done for the area that spans
±30 km across the wind directions, 30 km in upwind and 100 km in downwind directions. As smaller area
also means less interference from other sources.
Figures 3a and 3b show the estimates for the U.S. source parameters λ and σ, respectively, as a function of the
reported annual emissions in 2005. Note that the approach used here is justified for single sources only. If
there is a secondary source that elevates SO2 within the fitting area, both σ and λ can be affected yielding
unreliable estimates of emissions. Only sites without secondary sources within 50 km were used in Figure 3
FIOLETOV ET AL.
©2015 Her Majesty the Queen in Right of Canada. Reproduced with the permission of the Minister of the Environment.
6
Geophysical Research Letters
10.1002/2015GL063148
with the exception of 6 very large sources that had secondary sources located within 20 km. In these cases,
the primary and secondary sources were combined and treated as one. As a result, 48 sites out of 100 top U.S.
SO2 sources were selected, and 40 of these were able to produce a reliable fit. This includes some sources,
ranked near the bottom of the top 100 list, with emissions less than 30 kt yr1.
Figures 3a and 3b demonstrate that as the emission level declines, the uncertainties of the estimated
parameters increase along with the spread between the values for individual sources. Almost all U.S. sources
discussed here are located in the eastern U.S. and thus experienced very similar conditions. Therefore, the
spread of the σ and λ values likely reflects the uncertainty of the estimates for these relatively small sources
rather than that actual differences in the plume spread and SO2 lifetimes.
4. Emission Inventories and OMI SO2 Values
Once the parameters σ, λ, and a are estimated, they can be used to calculate emissions.
!
∞
Since
∞
∫ ∫
∞ ∞
f ðx; y Þ gðy; sÞdxdy ¼
∞
∞
∞
∞
∫ ∫
f ðx; y Þdx
gðy; sÞdy ¼
∞
∫
∞
gðy; sÞdy ¼1, the parameter a represents
the total observed number of SO2 molecules (or the SO2 mass) near the source. If OMISO2 is in DU, and σ is in
kilometers, then a is in 2.69 · 1026 molecules or 0.029 T(SO2). As shown by Fioletov et al. [2011], the total
mass of SO2 near the source can be related to the reported emissions. By considering many sources, a linear
relationship can be established between a and the reported emissions, with the slope being an estimate of
the average decay time, or it can be estimated directly as aλ, since λ = 1/τ.
Two methods were used to estimate emissions from the largest U.S. sources using the parameterization in
equation (1), summarized in Figures 3c and 3d. Here emissions estimated from OMI data for 2005–2007
(when emissions were quite stable) were compared with the reported emissions for 2005. Figure 3c, similar to
Fioletov et al. [2011], is based on the fitted mass and a specified, constant decay time for all sources, about 6 h.
This value was chosen as it gave the best fit between estimated and reported emissions: a slope of 1 and
a correlation coefficient of 0.93. This decay time is longer than 5 h as reported by Fioletov et al. [2011], but
still about 4 times less than the current estimates of the SO2 chemical conversion to sulfate time for the
eastern U.S. in summer derived from low-resolution model calculations [Lee et al., 2011]. However, a direct
comparison of τ and chemical lifetime is not correct since they represent different characteristics as discussed
by Fioletov et al. [2013]. Estimates of τ from the fits vary between 4 and 12 h but typically are about 5–6 h.
Figure 3d also shows the estimated emissions, but now using the fitted, source-specific decay times. Note
that these uncertainties are larger than in Figure 3a since these also include a contribution from λ. The
correlation coefficient is 0.91, and the slope of the linear regression line is 1.13, suggesting that the method
overestimates emissions by about 13%. There are several possible reasons for this. As noted by de Foy et al.
[2014], the lifetime estimates are sensitive to the accuracy of the plume rotation. This may be contributing
to an underestimate of the lifetime and hence an overestimate of emissions. Also, the PCA VCDs are based
on constant SO2 Jacobians (or a constant effective air mass factor (AMF)) calculated for typical conditions
over the eastern U.S. on polluted summer days. Figure 3b confirms that while the constant Jacobians overall
produces realistic SO2 emission estimates, this approximation may still be contributing to the observed
bias. Furthermore, the eastern U.S. Jacobian may not be appropriate everywhere, and location-specific
calculations may be necessary for more accurate retrievals of SO2 VCD [e.g., McLinden et al., 2014]. A third
potential source of systematic error is the height at which the winds are obtained. An average over the lowest
800 m or so was used here, but this may not always be appropriate.
The energy production from U.S. power plants is seasonally dependent with maximum production (and
presumably SO2 emissions) in summer. To study the seasonal effects, emissions were estimated for summer
(June to August) and for the combined March to May plus September to November seasons. Winter data were
not used due to larger uncertainties in observations. Estimated emissions for the combined spring-autumn
season (converted to the annual rate) were found to be very similar to the reported annual emissions (lower by
1%), while summer emissions were 35% higher (see also the supporting information). The decay time in
summer calculated as an average of the estimates for individual sources is slightly shorter for summer (5.1 h)
than for the combined spring-autumn season (5.4 h).
FIOLETOV ET AL.
©2015 Her Majesty the Queen in Right of Canada. Reproduced with the permission of the Minister of the Environment.
7
Geophysical Research Letters
10.1002/2015GL063148
5. Summary and Discussion
Information about SO2 emissions from point sources can be derived from the total amount of SO2 near the source
and the decay rate, both of which can be obtain from OMI measurements coupled with the wind information.
The estimation algorithm is based on fitting the observed SO2 values aligned along the upwind-downwind
direction by a 3-D function of coordinates and the wind speed. The fitting function is a product of Gaussian and
exponentially modified Gaussian functions with three parameters estimated from the fit. Emission and decay
rates are then derived from the estimated parameters.
The estimated emissions for the top 100 U.S. SO2 sources were compared with the available emission
inventories. About a half of the top 100 sources can be considered as point sources, and the method
produces reliable emission estimates for 40 of them. A high correlation (0.91) was found between the
reported annual emissions from an individual source and SO2 emissions estimated from OMI data by the
suggested method. Sources with emission rates as low as 30 kt yr1 can be detected and analyzed (based on
3 years of OMI measurements). This limit is substantially lower than previously reported (70 kt yr1) for the
previous standard OMI PBL SO2 data product [Fioletov et al., 2011]. This improvement is largely due to a factor
of 2 smaller noise of the PCA OMI SO2 retrievals.
Ultimately, the proposed technique will work best with frequent high spatial resolution measurements
from the future missions such as TROPOspheric Monitoring Instrument (TROPOMI), the next generation of
polar-orbiting atmospheric composition instruments [Veefkind et al., 2012], and Tropospheric Emissions:
Monitoring of Pollution (TEMPO) on a geostationary satellite (http://www.cfa.harvard.edu/atmosphere/
TEMPO/). The method should also work for other species measured by satellite instruments such as NO2.
Acknowledgments
We acknowledge the NASA Earth Science
Division for funding the OMI SO2
product development and analysis. The
Dutch-Finnish-built OMI instrument is
part of the NASA EOS Aura satellite
payload. The OMI project is managed
by KNMI and The Netherlands Agency
for Aerospace Programs (NIVR). The
U.S. Environmental Protection Agency
provided SO2 emission data. OMI PCA
SO2 retrievals used in this study have
been publicly released as part of the
OMI SO2 product and can be obtained
free of charge at the Goddard Earth
Sciences Data and Information Services
Center (http://daac.gsfc.nasa.gov/).
The Editor thanks two anonymous
reviewers for their assistance in
evaluating this paper.
FIOLETOV ET AL.
References
Bauduin, S., L. Clarisse, C. Clerbaux, D. Hurtmans, and P.-F. Coheur (2014), IASI observations of sulfur dioxide (SO2) in the boundary layer of
Norilsk, J. Geophys. Res. Atmos., 119, 4253–4263, doi:10.1002/2013JD021405.
Beirle, S., K. F. Boersma, U. Platt, M. G. Lawrence, and T. Wagner (2011), Megacity emissions and lifetimes of nitrogen oxides probed from
space, Science, 333, 1737–1739, doi:10.1126/science.1207824.
Beirle, S., C. Hörmann, M. Penning de Vries, S. Dörner, C. Kern, and T. Wagner (2014), Estimating the volcanic emission rate and atmospheric
lifetime of SO2 from space: A case study for Kılauea volcano, Hawaii, Atmos. Chem. Phys., 14, 8309–8322, doi:10.5194/acp-14-8309-2014.
Carn, S. A., N. A. Krotkov, K. Yang, and A. J. Krueger (2013), Measuring global volcanic de-gassing with the Ozone Monitoring Instrument (OMI),
in Remote Sensing of Volcanoes and Volcanic Processes: Integrating Observation and Modeling, Geol. Soc. London Spec. Publ., vol. 380,
pp. 229–257, edited by D. M. Pyle, T. A. Mather, and J. Biggs, doi:10.1144/SP380.12.
de Foy, B., J. L. Wilkins, Z. Lu, D. G. Streets, and B. N. Duncan (2014), Model evaluation of methods for estimating surface emissions and
chemical lifetimes from satellite data, Atmos. Environ., 98, 66–77.
Fioletov, V. E., C. A. McLinden, N. Krotkov, M. D. Moran, and K. Yang (2011), Estimation of SO2 emissions using OMI retrievals, Geophys. Res.
Lett., 38, L21811, doi:10.1029/2011GL049402.
Fioletov, V. E., et al. (2013), Application of OMI, SCIAMACHY, and GOME-2 satellite SO2 retrievals for detection of large emission sources,
J. Geophys. Res. Atmos., 118, 11,399–11,418, doi:10.1002/jgrd.50826.
Helfrich, S. R., D. McNamara, B. H. Ramsay, T. Baldwin, and T. Kasheta (2007), Enhancements to, and forthcoming developments in the
Interactive Multisensor Snow and Ice Mapping System (IMS), Hydrol. Processes, 21, 1576–1586, doi:10.1002/hyp.6720.
Krotkov, N. A., S. A. Carn, A. J. Krueger, P. K. Bhartia, and K. Yang (2006), Band residual difference algorithm for retrieval of SO2 from the Aura
Ozone Monitoring Instrument (OMI), IEEE Trans. Geosci. Remote Sens., 44, 1259–1266.
Lee, C., R. V. Martin, A. van Donkelaar, H. Lee, R. R. Dickerson, J. C. Hains, N. Krotkov, A. Richter, K. Vinnikov, and J. J. Schwab (2011), SO2
emissions and lifetimes: Estimates from inverse modeling using in situ and global, space-based (SCIAMACHY and OMI) observations,
J. Geophys. Res., 116, D06304, doi:10.1029/2010JD014758.
Levelt, P. F., G. H. J. van den Oord, M. R. Dobber, A. Mälkki, H. Visser, J. de Vries, P. Stammes, J. Lundell, and H. Saari (2006), The Ozone
Monitoring Instrument, IEEE Trans. Geosci. Remote Sens., 44(5), 1093–1101, doi:10.1109/TGRS.2006.872333.
Li, C., J. Joiner, N. A. Krotkov, and P. K. Bhartia (2013), A fast and sensitive new satellite SO2 retrieval algorithm based on principal component
analysis: Application to the ozone monitoring instrument, Geophys. Res. Lett., 40, 6314–6318, doi:10.1002/2013GL058134.
Lu, Z., D. G. Streets, B. de Foy, and N. A. Krotkov (2013), Ozone Monitoring Instrument observations of interannual increases in SO2 emissions
from Indian coal-fired power plants during 20052012, Environ. Sci. Technol., 47, 13,99314,000.
McLinden, C. A., V. Fioletov, K. F. Boersma, S. K. Kharol, N. Krotkov, L. Lamsal, P. A. Makar, R. V. Martin, J. P. Veefkind, and K. Yang (2014),
Improved satellite retrievals of NO2 and SO2 over the Canadian oil sands and comparisons with surface measurements, Atmos. Chem.
Phys., 13, 21,609–21,664, doi:10.5194/acpd-13-21609-2013.
Norilsk Nickel (2009), Annual Report 2009, OJSC MMC Norilsk Nickel, 27 May 2010 Resolution № ГМК/20-пр-сд. [Available at http://www.
nornik.ru/en/investor/report/annual/.]
Pommier, M., C. A. McLinden, and M. Deeter (2013), Relative changes in CO emissions over megacities based on observations from space,
Geophys. Res. Lett., 40, 3766–3771, doi:10.1002/grl.50704.
Schoeberl, M. R., et al. (2006), Overview of the EOS Aura Mission, IEEE Trans. Geosci. Remote Sens., 44(5), 1066–1074, doi:10.1109/TGRS.2005.861950.
Streets, D. G., et al. (2013), Emissions estimation from satellite retrievals: A review of current capability, Atmos. Environ., 77, 1011–1042,
doi:10.1016/j.atmosenv.2013.05.051.
Veefkind, J. P., et al. (2012), TROPOMI on the ESA Sentinel-5 Precursor: A GMES mission for global observations of the atmospheric
composition for climate, air quality and ozone layer applications, Remote Sens. Environ., 120, 70–83.
©2015 Her Majesty the Queen in Right of Canada. Reproduced with the permission of the Minister of the Environment.
8