Satellite Detection of Atmospheric Aerosols
Intro 1
Radiation, SatDetection 4
Applications 8
DataTransmission 2
*DataBases 1
SummaryConclusion 1
Introduction 1
Satellite remote sensors constitute a unique new way of characterizing atmospheric
aerosols. Satellites provide a synoptic and global view of particulate air pollution with
spatial resolution of a km or less and temporal resolution of a day or less. Stunning
images …
Aerosol episodes of industrial haze, windblown dust, and forest fire smoke are
particularly well detected by satellites.
Aerosol detection by satellites begun in the mid-1960’s with advent of the Defense
Meteorological Satellite Program (DMSP), which operated a series of meteorological
satellites. Early 70s Anomalous gray shades
Parmenter 1972 forest fire smoke over Central America.
The era of near real-time air quality analysis began in the late 1990s, with the availability
of real-time satellite data over the Internet. High-resolution color satellite images were
uniquely suited for early detection, and tracking of extreme natural or anthropogenic
aerosol events.
Satellite aerosol detection for
1. Quantification of atmospheric aerosols, climate effect air quality health and
visibility, --2. Aerosols as nuisance to remote sensing of surfaces and of atmospheric parameters
These are interrelated
Both class of applications requires the spatiotemporal characterization of aerosols. Hence
this short review chapter will focus on general principles of characterization through
satellites.
Early satellite aerosol sensors were designed for the detection of the Earth surface
(LANDSAT), meteorological parameters (DMSP, GOES, AVHRR), atmospheric
columnar ozone (TOMS) and other applications. However, aerosols were a nuisance for
Atmospheric aerosols constitute nuisance to satellite and other remote sensing of surfaces
for the same reason as clouds and water vapor do: their perturbation of the image is
unpredictably variable in space and time. Cloud scattering and absorption completely
obscures the ground for satellite vision. Water vapor absorption is significant but it is
confined to specific spectral bands which can be avoided. The role of haze aerosols
perturbation stretches over the UV , visible and near IR radiation. but the haze is
partially transparent vertically and for many conditions the perturbation on the image can
be corrected for.
By the 1990s, a new generation of satellites sensors were deployed that were designed
from ground up for aerosol detection. MODIS, OMI …Robert Fraser and Yoram
Kaufman ..NASA
Chapter content: Atmospheric Radiation, Aerosols Physical and optical parameters,
Radiative transfer, aerosol retrieval
Satellite Platforms 2
MODIS 1-2 day global coverage in 36 wavelengths from 250 m to 1 km
resolution
MISR Stereo images at 9 look angles
These use many different aerosol detection principles
Radiation, SatDetection 4
The radiation sensed by SeaWiFS1 can be conceptualized as originating from at least four
discreet radiative ‘streams’ as shown in Figure 1: air (gas) scattering and absorption,
aerosol scattering and absorption, cloud reflectance, and surface reflectance. The air
scattering and absorption stream is a function of sun-target-sensor geometry and
elevation and can be calculated and removed. Clouds completely obscure the surface,
thus pixels containing clouds must be masked.
The solar radiation reflected from the surface is modified by aerosols. By the time the
reflected radiation reaches the top of the atmosphere, the aerosol add reflectance in some
wavelengths and deplete the upwelling radiation others in a highly non-linear fashion.
Consequently, the satellite sensor receives a reflectance that contains both the surface and
aerosol contributions. Conversely, the detected aerosol signal is dependant on the color
and magnitude of the surface reflectance. Thus, to properly retrieve the aerosol stream,
one must also retrieve the surface reflectance stream. The main challenge of aerosol
retrieval over land is to deconvolute the integrated reflectance signal.
Figure 1. Radiation detected by the sensor can be approximated as originating from one
of four "streams."
The eight bands are located in transmission windows in the visible and near infrared as
shown in Figure 2. The bands were chosen for low interference from atmospheric
absorption. The greatest concentration of bands is in the low visible wavelengths to
facilitate the study of ocean color.
Figure 2. The eight SeaWiFS sensors measure wavelengths of low absorption by
atmospheric gases.
The wavelengths sensed by SeaWiFS are also useful for land color detection, particularly
of vegetation. Figure 3 shows a reflectance spectrum for a typical vegetated surface, with
and without atmospheric haze. Chlorophyll absorption is detected by the first six
channels and especially channels 3 and 6 (0.49 m and 0.67 m).
Within the aerosol population, the particles compatible in size to the wavelength of light
are those that dominate the deterioration of optical environment.
Surfaces
Cloud shadows
Aerosol
Apparent Surface Reflectance, R
• The surface reflectance R0 objects viewed from space is modified by aerosol scattering and absorption.
• The apparent reflectance, R, is: R
= (R0 + Ra) Ta
Apparent Reflectance
Surface Reflectance
Aerosol Reflectance
Aer. Transmittance
R may be smaller or larger
then R, depending on
aerosol reflectance and
filtering.
The surface reflectance
R0 is an inherent
characteristic of the
surface
Aerosol scattering acts as
reflectance, Ra adding
‘airlight’ to the surface
reflectance
Both R0 and Ra are
attenuated by aerosol
transmittance Ta which
act as a filter
R
Aerosol as Filter: Ta
P
R0
= e-
Aerosol as Reflector:
Ra = (e-– 1) P
R = (R0 + (e-– 1) P) e-
300
Ocean
Reflectance,x1000
250
200
Surface + Haze
150
100
Haze
50
Surface
0
0.3
0.5
0.7
Wavelength,um
0.9
Figure 3. Spectra of vegetation reflectance with and without atmospheric haze.
The condition for successful aerosol retrieval is that its key aerosol optical properties of
the haze are either known a priori, or they can bee obtained from the image. These are the
wavelength dependent optical depth, τ(tau), the scattering phase function, and the single
scatter albedo. Unfortunately, due to strong and unpredictable spatial-temporal
variability of τ(tau), and to the lesser degree of the scattering phase function and the
single scatter albedo, determining their values, a priori, is not feasible. One is forced,
therefore to seek methods of aerosol optics characterization from the distorted image
itself. While such a ‘bootstrap’ approach may appear hopelessly undefined physically,
one draws encouragement from the fact that if such a scheme is successful then remote
sensing offers means of detecting relevant optical aerosol parameters, which in turn have
application in many branches of atmospheric physics: most notably in radiative climate
and weather modification.
In fact, the primary motivation for this work was drawn from the possible quantitative
detection by remote sensing of aerosol optics in a vertical column and our interest in
haze correction of images was only secondary. In the approach described below, the two
applications are tightly coupled: one needs: one needs a haze perturbed image to estimate
the aerosol optics parameters: conversely, a ‘perfectly’ reconstructed haze aerosol
parameter is achieved only if one is able to reconstruct a haze free image. A major source
of encouragement was the remarkably success if semi-empirical haze correction scheme
of Gordon and Clark (1980), applied to the Coastal Zone Color Scanner data.
Furthermore, the aerosol perturbation of an image is rather subtle: it acts as an attenuator
(filter) of the radiation originating from the surface which is characterized by the aerosol
optical depth ( ??). It also acts, however, as a source of light thus adding light to the
surface reflected radiation (1-??)?? se section ???. Thus a dark section of an image may
become brighter while the brighter sections may darken depending on the relative values
of the haze attenuation of source functions. These subtle features of haze effects on the
transmitted images combined with the data inadequately on the phase function, ??? and to
a lesser degree, ??? hamper the development of haze correction schemes for haze-free
image.
In what follows, the physics of photon transmission through a hazy atmosphere will be
expressed via the radiative transfer equation and leading to the haze image perturbation
as a function of three aerosol parameters: optical depth, ?? scattering phase function, ??
and the single scatter albedo, ??. In Part ?? specific procedures are outlined to estimate
these parameters from hazy LANDSAT images and for the reconstruction of the hazefree image.
Vision through the hazy atmosphere may that be human or electronic, can be subdivided
into four components as illustrated in Figure1. : (1) the light source, usually the sun (2)
the reflecting objects; (3) the radiatively active hazy atmosphere and (4) the image
sensing detector.
In the absence of clouds, solar radiation first passes through the ozone layer (10-40km)
optically defined by its absorption optical depth, ?? (Figure 2) For remote sensing only
the kopius ozone absorption band (0.52-0.74 µm) is of interest with its peak optical depth
of ??-?? at ?? µm. The absorbed shortwave radiation by ozone contributes only to the
??ing of the upper atmosphere but none to the backscattered radiation.
Much of the backscattering or ‘airlight’ arises from the molecular or Rayleigh
atmosphere which also attenuates the incoming solar radiation with its strongly
wavelength dependent optical depth: ?? = 0.00888 ?4?.?0?5? (Robinson 1966)’ The
phase function of Rayleigh scatter ??= ¾(1+cos??2?).
The most severe perturbation to the incoming solar radition occurs in the lowest 3 km of
the atmosphere which contains light scattering/absorbing aerosol and water vapor.
Indirect??
The vertical aerosol optical depth may range from 0.01 to 1.0 and its phase function may
also be variable. Some general features of the haze aerosol were documented by
Barteneva (1960) and are of substantial utility. The aerosol phase function measured
over several locations exhibited well defined continental USSR pattern as a function of
visual range (inversely proportional to the extinction coefficient (///). At low ???=0.02
km-1 the measured phase function closely resembled the Rayleigh phase function, with
the same amount scattered. However, as the haziness (???/) increased the scattering
phase function became systematically skewed toward the forward direction. At visual
range about 50 km, measured light scattering in the forward direction was 3 times that of
the backscattering , and at visual range of 10 km, the forward/backscattering ratio was
about 7.0, i.e. the backscattering vonstituted about 12.5% of the total scattering. For
these conditions, over 50% of the scattered radiation was confined to the forward
scattering angles 0-30 degrees.
AEROSOL SCATTERING AND ABSORPTION
For the visible and near IR of the spectrum the degradation of optical environment arises
from the scattering and absorption of haze particles.
It is worth noting that scattering is not restricted to the visible part of the electromagnetic
radiation and that the scattering laws apply to all wavelengths. The scattering of radio ??
waves by satellites , microwaves by raindrops, thermal radiation by clouds, light by fine
particles and electron scattering by molecules are all similar phenomena. In each case, the
wavelength is of the same magnitude as the scatterer. Thus, a natural scaling factor for
scattering is the wavelength, of the incoming radiation and is used in the dimensionless
optical size parameter.
Surface Reflectance Retrieval
The retrieval of surface reflectance (in the absence of haze and clouds) is based on the
MS Thesis of S. Raffuse2 The generation of daily surface reflectance data involves four
major steps:
1. Preprocessing of raw L1a SeaWiFS data to L1b physical units
2. Rayleigh correction to remove the air scattering
3. Scattering angle normalization of total reflectance (150 deg)
4. Surface reflectance creation
The first three steps involve processing one daily image at a time. The final step relies on
temporal analysis of consecutive daily images. Daily surface images are derived from
Rayleigh and scattering angle corrected data. The goal of the surface reflectance
algorithm is to to select pixels from days that are cloud and haze free. After that the
cloud-free image is further processed to reduce the residual haze using features of the
extracted haze - thus the term co-retrieval.
The two backscattering phenomena that are not part of the surface reflectance are clouds
and atmospheric aerosol. Both tend to increase the reflectance measured by the sensor at
short wavelengths. Aerosols, especially anthropogenic haze, increase the reflectance
most prominently in the shorter wavelengths. Cloud shadows constitute a nagging
problem in aerosol retrieval over land. Raffuse2 has implemented an elaborate time-series
filtering algorithm to minimize the effect of cloud shadows. Figure 4 shows spectra of
surface reflectance, haze contaminated, and cloudy pixels respectively.
Figure 4. Spectra in the first 6 channels for clean vegetation, hazy vegetation, and subpixel cloud pixels.
Given a series of spectral reflectance values for the same pixel on different days (a time
series), the cleanest day is the one with the lowest sum of the first four spectral bands.
This value is taken to be most representative of the haze-free surface color. The choice
of the proper time series length is important. By utilizing a large enough time window, a
clean day will be found for all pixels. However, the surface color itself changes over
time,
especially in the spring and fall. Raffuse2 has devised an adoptive 30-day moving
window approach. Figure 5 shows the surface reflectance image for a sub-region after the
effect of clouds and haze have been removed. This surface reflectance is used to derive
the aerosol optical thickness on any hazy day.
Figure 5. The surface reflectance for sub-region in the Midwest in the absence of clouds
and haze.
AOT Retrieval
The aerosol optical thickness retrieval used in this work is based on the co-retrieval
algorithms developed by Husar2. Below is a brief summary of the method. Remote
sensing provides the combined surface and aerosol reflectance but not the separate
contributions of each. The challenge of co-retrieval is to separate the aerosol and surface
reflectances. Aerosols of different types cause different perturbation of the surface
reflectance.
Fortunately, some of the aerosol optical parameters can be estimated from the distorted
surface reflectance. The condition is that one can derive a high quality and temporally
stable haze-free surface reflectance image with substantial texture containing bright and
dark surfaces. Conversely, the derived aerosol parameters allow the reconstruction of
haze-free surface reflectance. The surface and aerosol retrievals are interdependent and
through iteration provide a mutual quality control: Successful aerosol retrieval depends
on high quality surface reflectance data; conversely, the surface reflectance can only be
retrieved when the aerosol-optical properties are properly estimated.
In this work, the retrieved aerosol parameters are the columnar aerosol optical thickness
at several wavelengths (e.g. 0.416 and 0.67) um and the spectral aerosol reflectance
function. Both the aerosol parameters and the surface reflectance are retrieved over all
cloud-free areas at all wavelengths. However, the quality of the retrievals is low in the
near IR. The major assumptions for the retrieval algorithm are listed below:
 Gaseous scattering and absorption is subtractable from the sensed radiation.
 All the non-absorbed solar radiation reaches the surface directly or diffusely –
small backscattering fraction.
The radiative processes and the resulting radiative transfer equations, based on Husar3 are
shown in Figure 6.
Figure 6: Schematic representation of the radiative transfer processes relevant to the
aerosol retrieval, Husar3
Major Assumptions
• Gaseous scattering and absorption is subtractable from the sensed radiation –
multiple scattering is negligible.
• All the remaining solar radiation reaches the surface directly or diffusely –small
backscattering fraction.
• Backscattering to space is due to incoming solar radiation – low surface
reflectance.
•
•
•
•
•
Rayleigh air scattering and gaseous absorption is removed first by the E. Vermote
algorithm.
Cloudy pixels are masked out since they obscure the surface and aerosol
reflectance
The remaining reflectance over land and water consists of the combined effect of
aerosol scattering/absorption and surface reflectance.
The goal of the co-retrieval is to separate the reflectance due to aerosol from
surface reflectance
The aerosol scattering and absorption has twofold influence on the apparent
reflectance of objects remotely sensed from space:
• It reduces surface reflectance through the transmittance Ta as a filter
• It adds to surface reflectance through backscattering ‘airlight’
• The net aerosol influence depends on the relative contributions of filtering
and aerosol reflectance effects
The critical parameter whether aerosols will increase or decrease the apparent reflectance,
R, is the ratio of aerosol angular reflectance, P, to bi-directional surface reflectance, R0,
P/ R0
Aerosols will increase the apparent surface reflectance, R, if P/R0 < 1. For this reason,
the reflectance of ocean and dark vegetation increases with τ.
When P/R0 > 1, aerosols will decrease the surface reflectance. Accordingly, the
brightness of clouds is reduced by overlying aerosols.
At P~ R0 the reflectance is unchanged by haze aerosols (e.g. soil and vegetation at 0.8
um)..
At large τ (radiation equilibrium), both dark and bright surfaces asymptotically approach
the ‘aerosol reflectance’, P
Whether contrast decays fast or slow with increasing τ depends on the ratio of aerosol to
surface reflectance, P/ R0
Note: For horizontal vision against the horizon sky, P/R0 = 1, contrast decays
exponentially with τ, C/C0=e-τ.
The perturbed surface reflectance, R, can be used to derive the the aerosol optical
thickness, τ , provided that the true surface reflectance R0 and the aerosol reflectance
function, P are known. The excess reflectance due to aerosol is : R- R0 = (P- R0)(1-e- τ)
and the optical depth is:
For a black surface, R0 =0 and optically thin aerosol, τ < 0.1, τ is proportional to excess
radiance, τ =R/P. For τ > 0.1, the full logarithmic expression is needed.
As R0 increases, the same excess reflectance corresponds to increasing values of τ.
When R0 ~P the aerosol τ can not be retrieved since the excess reflectance is zero. For
R0 > P, the surface reflectance actually decreases with τ, so τ could be retrieved from the
loss of reflectance, e.g. over bright clouds.
The value of P is derived from fitting the observed and retrieved surface reflectance
spectra. For summer light haze at 0.412 μm, P=0.38.
Accurate and automatic retrieval of the relevant aerosol P is the most difficult part of the
co-retrieval process. Iteratively calculating P from the estimated τ( λ) is one possibility.
In the blue (λ=0.412) and red (λ=0.67) both the land and the ocean have low surface
reflectance and the excess reflectance is the same. However at green and near IR the
excess reflectance over land is lower then over the ocean as expected from radiative
transfer theory.
•
Comparison of Haze Effects on Land and Ocean
•
•
Summary
The aerosol and the surface parameters over cloud-free land and the ocean can
be retrieved from daily 8 wavelength SeaWiFS LAC satellite data.
•
•
•
•
•
•
•
•
•
•
A 30 day pilot study (July15-Aug 15, 1999) over the NE United States has shown
the existence of large, 1000-scale hazy air masses with strongly varying
spectral extinction, particularly near clouds. A full-scale calibration of the results
has not taken place.
The pilot study also demonstrates that the methodology is potentially applicable
for daily global-scale co-monitoring of aerosols and surface color over cloudfree areas.
It appears that the next level of improvement in co-retrieval would benefit from
the closer interaction of experts in ocean and land surface color and aerosol
characterization
This is work in progress and it will be periodically updated. It is shared here for
purposes of seeking interaction with interested researchers. Interested? Pleas
contact us: rhusar@me.wustl.edu
The goal of the work is to simultaneously obtain (co-retrieve) the columnar
aerosol optical properties as well as the aerosol-free surface reflectance. The
focus of this work-phase is to retrieve the spectral aerosol optical thickness, τ
and spectral surface reflectance.
The work is performed using the 8 wavelength (0.4-0.9 μm) SeaWiFS satellite
data.
The surface-aerosol co-retrieval method is based largely on an unpublished
procedure developed for LANDSAT data in the 70s; the Rayleigh correction
algorithm is by Eric Vermote.
The SeaWifs data pre processing programs were written by Fang Li and R.B.
Husar at CAPITA and the procedures .
The SeaWifs data pre-processing was performed by Fang Li of CAPITA using
the commercial software ENVI by RSI with built-in IDL language support.
This work builds on the rich literature on aerosol retrieval/surface detection work
of Kaufmann, Tanre, King, Vermote…as well as on the remarkably successful
ocean color retrieval work of Gordon. At some later time, proper references will
be made to their extensive work.
•
Co-Retrieval of Aerosol and Surface Reflectance
•
Remote sensing provides the combined surface and aerosol reflectance but not
the separate contributions. The challenge of co-retrieval is to separate the aerosol
and surface reflectances.
The aerosol optical parameters can be estimated from the distorted surface
reflectance. The condition is that one has a high quality and stable haze-free
surface image with substantial texture containing bright and dark surfaces.
Conversely, the derived aerosol parameters allow the reconstruction of haze-free
surface reflectance.
The retrievals are interdependent and through iteration provide a mutual quality
control:
– Successful aerosol retrieval depends on high quality surface reflectance
data.
•
•
–
•
The surface reflectance can only be retrieved for known aerosol-optical
properties.
In this work, the retrieved aerosol parameters are the columnar aerosol optical
thickness at several wavelengths (e.g. 0.416 and 0.67) um and the spectral
aerosol reflectance function. Both the aerosol parameters and the surface
reflectance are retrieved over all cloud-free areas at all wavelengths. However,
the quality of the retrievals is low in the near IR.
The procedure for aerosol retrieval consists of the following steps:
 Remove air scattering and absorption from daily images by the algorithm of
Vermote and Tanre4
 Generate aerosol and cloud-free surface from long-term data by the Raffuse
method
 Estimate the initial aerosol properties, P and τ over non-cloudy areas
 Derive the aerosol τ for each pixel based on an excess reflectance
 Reconstruct daily aerosol-free surface reflectances
 Loop to update aerosol properties in 3 using improved value of P.
The result of the aerosol retrieval algorithm is illustrated in figure 7 for ocean and land. It
is evident, that most of the aerosol perturbation is at shorter wavelengths.
•
•
•
Aerosols add to the reflectance and sometimes reduce the reflectance of surface
objects
Aerosols always diminish the contrast between dark a bright surface objects
Haze and smoke aerosols change the color of surface objects to bluish while
dust adds a yellowish tint. (Click on the Images to View)
Figure 7. Illustration of the aerosol retrieval algorithm over ocean and land.
•
The retrieved aerosol parameters are τ(λ) and P(λ). The τ is a measure of the
columnar aerosol concentration and it has a strong spatial variation. The aerosol
reflectance function P is a measure of characteristic particle size and particle
absorption. P is presumed to vary less spatially. (Actually, the data indicate that in
•
some conditions (e.g. transition between ‘haze’ and ‘mist’ there are strong spatial
gradients in P as well as absorption).
The most reliable wavelengths for aerosol retrieval over land are blue (0.412)
and red (0.67) since both are dark compared to the brighter green. The Angstrom
exponent, b, τ~ λ-b , is derived from the estimated τ at the two ‘dark surface’
wavelengths (0.412 and 0.67):
b
•
•
•
•
log( 0.41 /  0.67 )
log( 0.41 / 0.67)
The spectral aerosol optical thickness, τ (λ ) is retrieved for all pixels, except
over clouds.
The aerosol retrieval does not assume a particular form of aerosol optical
parameters, i.e. there is no a priori ‘aerosol model’. Rather, the spectral τ is
retrieved for each 8 wavelength from the excess reflectance and calibrated P.
At this time the aerosol angular reflectance function wλPλis calibrated using best
fit to bright and dark surfaces. In the future, P will also be estimated iteratively
based on the shape of the spectral extinction curve.
The current aerosol retrieval method has many limitations:
– It is only possible if the aerosol-free reflectance is available
– Estimation of P is tedious; the derivation of Pλfrom τ λis not completed
– The estimation τλ at 0.7 and 0.8 μm over land is uncertain due to low
excess reflectance
– Tests were only done for the hazy Northeastern US; dusty regions will
differ
– Plus many, many other limitations
Remove air scattering and absorption from daily images
1.
2.
3.
4.
5.
Generate aerosol and cloud-free surface from long-term data
Estimate the initial aerosol properties, P and τ over non-cloudy areas
Derive the aerosol τ for each pixel based on an excess reflectance
Reconstruct daily aerosol-free surface reflectances
Loop to update aerosol properties in 3 using improved value of P.
At blue (0.412) wavelength, the haze reflectance dominates over land surface
reflectance. The surface features are obscured by haze. Air scattering (not included)
would add further reflectance in the blue. The blue wavelength is well suited for aerosol
detection over land but surface detection is difficult.
At green (0.555) over land, the haze is reduced and the vegetation reflectance is
increased. The surface features are obscured by haze but discernable. Due to the low
reflectance of the sea, haze reflectance dominates. The green not well suited for haze
detection over land but appropriate for haze detection over the ocean and for the detection
of surface features.
At red (0.67) wavelength over land, dark vegetation is distinctly different from brighter
yellow-gray soil. The surface features, particularly water (R0<0.01), vegetation
(R0<0.04), and soil (R0<0.30) are are easily distinguishable. Haze reflectance dominates
over the ocean. Hence, the red is suitable for haze detection over dark vegetation and
the ocean as well as for surface detection over land.
In the near IR (0.865) over land, the surface reflectance is uniformly high (R0>0.30)
over both vegetation and soil and haze is not discernable. Water is completely dark
(R0<0.01) making land and water clearly distinguishable. The excess haze reflectance
over land is barely perceptible but measurable over water. Hence, the near IR is suitable
for haze detection over water and land-water differentiation.
Key features of aerosol retrieval algorithm are listed below:

The spectral aerosol optical thickness, τ (λ ) is retrieved for all pixels, except
over clouds.
 The aerosol retrieval does not assume a particular form of aerosol optical
parameters, i.e. there is no a priori ‘aerosol model’. Rather, the spectral τ is
retrieved for each 8 wavelength from the excess reflectance and calibrated P.
 At this time the aerosol angular reflectance function λPλis calibrated using best
fit to bright and dark surfaces. In the future, P will also be estimated iteratively
based on the shape of the spectral extinction curve.
The current aerosol retrieval method has many limitations:
 Estimation of the aerosol phase function from the data is tedious; the derivation of
Pλfrom τ λis not completed

The estimation τλ at 0.7 and 0.8 μm over land is uncertain due to low excess
reflectance
Just like the human eye, satellite sensors detect the total amount of solar radiation that is
reflected from the earth’s surface (Ro) and backscattered by the atmosphere from
aerosol, pure air, and clouds. A simplified expression for the relative radiation detected
by a satellite sensor (I/Io) is:
Key aerosol microphysical parameters
Particle size and size distribution: Aerosol particles > 1 mm in size are produced
by windblown dust and sea salt from sea spray and bursting bubbles. Aerosols
smaller than 1 mm are mostly formed by condensation processes such as
conversion of sulfur dioxide (SO2) gas to sulfate particles and by formation of
soot and smoke during burning processes.
Effective radius. Moment of size distribution weighted by particle area and
number density distribution
Complex refractive index. The real part mainly affects scattering and the
imaginary part mainly affects absorption
Particle shape. Aerosol particles can be liquid or solid, and therefore spherical or
nonspherical. The most common nonspherical particles are dust and cirrus
• What is “remote sensing”?
– Using artificial devices, rather than our eyes, to
observe or measure things from a distance without
disturbing the intervening medium
• It enables us to observe & measure things on
spatial, spectral, & temporal scales that
otherwise would not be possible
• It allows us to observe our environment using a
consistent set of measurements throughout the
globe, without prejudice associated with
national boundaries and accuracy of datasets or
timeliness of reporting
• How is remote sensing done?
– Electromagnetic spectrum
• Passive sensors from the ultraviolet to the
microwave
• Active sensors such as radars and lidars
– Satellite, airborne, and surface sensors
– Training and validation sites
• Remote sensing uses the radiant energy that is reflected
and emitted from Earth at various “wavelengths” of the
electromagnetic spectrum
• Our eyes are only sensitive to the “visible light” portion of
the EM spectrum
• Why do we use nonvisible wavelengths?
Applications 4
However, during extreme air quality events like the above-mentioned, air quality
managers need more extensive 'just in time analysis’, not just qualitative air quality
information
In April 1998, for example, a group of analysts keenly followed and documented on the
Web, in real-time the trans-continental transport and impact of Asian dust from the Gobi
desert on the air quality over the Western US. (Husar, et al., 2001,
http://capita.wustl.edu/Asia-FarEast). Soon after, in May 1998, another well-documented
incursion of Central American forest fire smoke caused record PM2.5 concentrations
over much of the Eastern US (e.g. Peppler, et al., 2000; http://capita.wustl.edu/CentralAmerica).
The high value of qualitative real-time air quality information to the public is well
demonstrated through EPA’s successful AIRNOW program (Weyland and Dye, 2006).
In fact, in the1998 Asian Dust Event, local air quality managers in Oregon and
Washington used the real time analysis to issue health advisories . Soon after the
CentralAmerican Smoke Event, the federal EPA granted some states ‘exceptional event’
exemptions from ozone standard violations.
For illustration of the FASTNET analysis concept, tools and application methods, we use
the April 2003 Kansas smoke event. For several days every spring, the rangeland grass in
Kansas-Oklahoma is burned resulting in major smoke plumes that cover multi-state areas
of the Midwest. Figure 3 shows the location of fires over Kansas derived from the
MODIS satellite sensor, the noon aerosol optical thickness derived from the SeaWiFS
satellite sensor and the spatial pattern of organic fine particle mass concentration derived
from speciated aerosol data collected through several surface monitoring networks.
Fig
Fig 3 Multi-sensory characterization of agricultural smoke over the Midwest on April 12, 2003: a.
Fire pixels over Kansas from the MODIS sensor; b. Vertical aerosol optical thickness derived from
the SeaWiFS satellite; c. Concentration pattern of organics from speciated chemical samplers (EPA)-
The spatio-temporal sparseness of routine surface monitoring networks prevents the full
characterization of aerosol events but complementary satellite observations can fill in
many of the missing pieces. Auxiliary data from the 1200 station NOAA surface
visibility network (not shown) is particularly useful for the study of fine-scale aerosol
pattern at the surface. During the event and the days-years after, the data from these realtime monitoring systems were analyzed by several groups using the FASTNET tools and
methods. Based on such exploratory analyses, the federal EPA is considering exceedance
waivers for remote sites that were impacted by the agricultural smoke. The local
managers are also using the dynamic analysis tools to evaluate alternative grass burning
schedules and approaches. Collectively, these types of analyses are used in the new, more
flexible ‘weight of evidence’ approach to air quality compliance analysis.
Retrieved Aerosol Pattern, 2000-2002
This section presents limited results of the aerosol retrieval covering the three year period
2000 – 2002. The results include both spatial and temporal analysis derived from the
daily data. Figure 8 shows the SeaWiFS images for the August 2000 Idaho smoke event.
A (top) Total Reflectance B(middle) Surface Reflectance C(bottom) Aerosol AOT at
0.412 um. Note the yellow coloration of the smoke, clouds and the retrieved smoke
optical thickness. Figure 9 illustrates similar features of the July 2002 Quebec forest
smoke over New England. Finally Figure 10 shows the daily time series for a single 1 km
pixel.
Future analyses will present a more thorough presentation and explanation of these
results.
This project proposes to advance the use of satellite data along two complementary
tracks. In the first track, we will pursue the assimilation of satellite-derived gaseous
pollutants, most notably NO2 into the BAMS model. Since at present the atmospheric
lifetime of nitrogen species is poorly understood, this track will require considerable
diagnostic exploration of procedures for integrating the available satellite and surface
data and their assimilation into the model. However, once the assimilation is competed,
it will yield new scientific insights into this important precursor of ozone, nitric acid and
nitrate aerosols.
The second track will seek to use satellite data to characterize the emissions from nonindustrial emissions, including smoke-organics, agricultural nitrates and windblown dust.
While in the past, biomass smoke and dust emissions were clearly observable from
satellite images, the newly developed, high-resolution (‘Deep Blue’) aerosol retrieval
algorithm now allows the quantitative estimation smoke and dust emissions rates from
satellite sensors such as MODIS and SeaWiFS. This will be unique contribution since the
current smoke and dust emissions rate estimates are poor at best. More importantly, the
results of this track will be of direct and immediate benefit to both air quality and hazard
management communities.
Satellite Smoke Detection: Satellites offer a unique opportunity for spatial (X,Y)
characterization of smoke through high spatial resolution snap shots. The Deep-Blue
aerosol retrieval at 1 km resolution or less provides unprecedented spatial detail,
routinely sensed by the contemporary MODIS or SeaWiFS. However, the satellite data
also indicate that some of the forest fire smoke is bluish in color, while other smoke
plumes are yellow. This clearly indicates that size and chemical composition of these two
smoke types is different. There is also considerable evidence that multiple scattering in a
thick smoke plume is prevalent. Thus, the retrieval of the true vertical optical thickness
for such plumes is problematic.
Examples are given below to provide a realistic expectation on what satellite and surface
smoke sensing can contribute to smoke quantification. Figure 3 illustrates the application
of surface monitoring data (PM2.5 mass and organic composition) during an agricultural
smoke event, April 12, 2003. Figure 4a,b shows the two main real-time surface networks
(1200-station ASOS and 350-station AIRNow). The1 km SeaWiFS-derived AOT clearly
shows the individual smoke plumes emanating from the fires. The ASOS light scattering
network data further reveal that the smoke is at the surface at night but elevated during
the daytime. Early Monte Carlo plume simulations indicate that through a fitting
procedure one can construct a smoke emission field that yields the observed smoke
pattern after dispersion. The chemical composition data confirm that the elevated PM2.5
is due to smoke organics. Figure 5 shows the pattern of the Quebec smoke on July 6,
2002 as observed through the GOES satellite (30 minute intervals, 1 km resolution);
ASOS light scattering (circles on satellite images), AIRNow PM2.5 time series and the
MPLNet lidar at NASA Goddard. The multi-sensory integration documents the
magnitude of the smoke emission pattern (highly diurnal), the smoke impact on surface
PM concentration and light scattering as well as some aspects of the smoke vertical
distribution.
Clearly, this activity will closely linked with parallel work through other smoke-related
projects. The above data-model data integration framework for smoke emission
estimation has been discussed with some of the members of the BlueSkyRAINS
Advisory Group with very favorable response. It is therefore expected that there will be
mutual interests and benefits from close collaboration with the BlueSkyRAINS team,
including Sonoma Tech. Inc through their recently awarded NASA Applications projects.
Event-driven Emissions and System Extensions. Distinct from aerosol observational data
but related to it are data which may be used to infer events which result in aerosol related
emissions (as opposed to aerosol state), particularly data about fire events. Modeling such
events is a topic of current and on-going research. For example, NOAA ARL has
implemented a demonstration project utilizing the Hazard Mapping System (HMS) and
the HYSPLIT model (http://www.arl.noaa.gov/smoke/forecast.htm ,
http://www.ssd.noaa.gov/PS/FIRE/hms.html ) to attempt to model the emissions from fire
events, based upon detection by the GOES Automated Biomass Burning Algorithm
(ABBA) product and subjective input from trained satellite analysts. However, the
current HMS has considerable manual intervention, with a turnaround time on the order
of one day. Unfortunately, this is not acceptable for use in an operational aerosol forecast
system.
For this project, we will adapt the algorithms from the HMS to construct a combined
MODIS/GOES fire emissions module for the SMOKE emissions model. This model will
run during the Corrector phase of the data assimilation step and potentially the next
forecast phase (subject to determination of the mean duration/persistence of the fire
events being modeled). This new event-driven SMOKE component will not have the
extensive human quality assurance steps found in the HMS, but still should result in
improved aerosol forecast. At a later date, it may be possible to incorporate fire-generated
heat fluxes and other (e.g., humidity and wind speed) effects of the fires into the
meteorology forecast system as well. Also, at a later date it may be possible to use land
use datasets to determine fuel loading characteristics of each fire (e.g. type of vegetation
(crops, oak, etc.) in order to obtain better estimates of emissions rates, fire intensity and
corresponding plume rise, and fire duration and spread rates.
Data Transmission Integration 2
These responsive air quality management actions were largely facilitated by the agile
event analyses provided by the ad hoc community of scientist and managers collaborating
through the Internet.
In recent years, air quality management has also changed . The old command and control
style is giving way to a more participatory approach that includes all the key stakeholders
from multiple jurisdictions and application of scientific and ‘weight of evidence’
approaches. The air quality regulations now emphasize short-term monitoring while at
the same time long-term air quality goals are set to glide toward ‘natural background’
levels over the next decades. In response to these and other development, EPA has
undertaken a major redesign of the monitoring system that provides the data input for air
quality management. The new National Ambient Air Monitoring Strategy (NAAMS),
through its multi-tier integrated monitoring system, is geared to provide more relevant
and timely data for these complex management needs. All these changes in management
style place considerable burden on the information system that supports them.
Fortunately, both the air quality monitoring and data dissemination technologies have
also advanced considerably since the 1990s. Recent developments offer outstanding
opportunities to fulfill the information needs for the new agile air quality management
approach. The data from surface-based air pollution monitoring networks now provides
routinely highgrade, spatio-temporal and chemical patterns throughout the US for PM25
and ozone. Satellite sensors with global coverage and kilometer-scale spatial resolution
now provide real-time snapshots which depict the pattern of haze, smoke and dust in
stunning detail. The ‘terabytes’ of data from these surface and remote sensors can now be
stored, processed and delivered in near real time. The instantaneous ‘horizontal’ diffusion
of information via the Internet now permits, in principle, the delivery of the right
information to the right people at the right place and time. Standardized computercomputer communication languages and Service-Oriented Architectures (SOA) now
facilitate the flexible processing of raw data into high-grade ‘actionable’ knowledge.
Last but not least, the World Wide Web has opened the way to generous sharing of data
and tools leading to faster knowledge creation through collaborative analysis and virtual
workgroups. Nevertheless, air quality analysts face significant hurdles.
The new developments introduced a new set of problems. The “data deluge” problem is
especially acute for analysts interest in aerosol pollution, since aerosols are so inherently
complex and since there are so many different kinds of relevant data – from extensive,
new surface-based monitoring networks, meteorological and aerosol forecast models,
satellite imagery and associated data products, etc.
Recent developments offer outstanding opportunities to fulfill the information needs for atmospheric
sciences and air quality management. High-resolution satellite sensors with global coverage now provide
near-real-time snapshots which depict the spatial and temporal pattern of haze, smoke, dust and other
atmospheric constituents in stunning detail. The data from surface-based monitoring networks now
routinely provide detailed chemical composition of the atmosphere. The ‘terabytes’ of data from these
surface and remote sensors can now be stored, processed and delivered in near-real time and the
instantaneous ‘horizontal’ diffusion of information via the Internet now permits, in principle, the delivery
of the right information to the right people at the right place and time. Standardized computer-computer
communication languages and Service-Oriented Architectures (SOA) now facilitate the flexible processing
of raw data into high-grade ‘actionable’ knowledge. Last but not least, the World Wide Web has opened the
way to generous sharing of data and tools leading to faster knowledge creation through collaborative
analysis in real and virtual workgroups.
Nevertheless, atmospheric scientists and air quality managers face significant hurdles. The production of
Earth observations and models are rapidly outpacing the rate at which these observations are assimilated
and metabolized into actionable knowledge that can produce societal benefits. The “data deluge” problem
is especially acute for analysts interested in aerosol pollution, since the aerosol processes are inherently
complex, the numerous relevant data range form detailed surface-based chemical measurements to
extensive satellite remote sensing and the integration of these requires the use of sophisticated models. As a
consequence, Earth Observations (EO) are under-utilized in making societal decisions. A remedy is
anticipated from the Global Earth Observation System of Systems (GEOSS).
This paper is an early report on the application of the GEOSS concepts in the federated data system,
DataFed. The paper focuses on the architectural aspects of the DataFed design, as a user-driven
contribution to the emerging architecture of GEOSS. It is recognized that it represents just one of the many
configurations that is consistent with the spirit of GEOSS. The implementation details and the various
applications of DataFed are reported elsewhere [4]-[6].
GEOSS is an emerging public information infrastructure for finding, accessing and
applying diverse data useful for decision makers [1]. GEOSS is proactively pursuing the
linking together of Earth observing systems so that data can be combined and turned into
actionable knowledge for many societal benefit areas, SBAs. The Group on Earth
Observations (GEO) [2] is the body that coordinates the construction of GEOSS by the
year 2015, it ensures universal access to EO data as public good and helps building trust
and collaboration among the diverse stakeholders in GEOSS.
Earth observations are conducted by all nations, by diverse organizations, and performed through many
independent projects. Hence, centralized design and management of a Global Earth Observation System is
not an option. A unique contribution of GEOSS is the adoption and promotion of the advanced 'system of
systems' (SoS) approach toward the integration of the multiplicity of autonomous Earth observations and
models. System of Systems (SoS) is a fledgling field of science and there are a number of different
definitions and interpretations. According to an authoritative assessment [3], SoS consists of autonomous
constituents that are managed independently, the constituents evolve independently and an SoS composed
of such constituents acquires an emergent behavior. Furthermore, in SoS no stakeholder has a complete
insight and understanding; central control is limited; distributed control is essential and the
users/stakeholders must be involved throughout the life of SoS.
The specific goals of DataFed are: (1) facilitate the access and flow of atmospheric data from provider to
users, (2) support the development of user-driven data processing value chains, and to (3) participate in
specific air quality application projects. The federation currently mediates access to over 100 datasets
which includes both near real-time and historical observations and models. Since 2004 DataFed has
provided IT support to a number of air quality management applications. Virtually all the content and a
record of its evolution is accessible through the community workspace wiki at datafedwiki.wustl.edu.
DataFed is now an applied system used in everyday research by several air quality analysis groups. Hence,
DataFed is an autonomous information system (IS) with a purpose of its own, managed and evolving
It is worth highlighting that key users of air quality decision systems are technical analysts and the IS
needs to be tailored primarily to their needs. Also, much of the communication along the value chain in the
DSS is between the human participants through reports and verbal communication rather than computercomputer interactions.
In summary, the role of DataFed in the GEO enterprise is to provide air quality-related data as services
and to participate in the testing of the GEOSS core architecture. DataFed is also a decision support system
for air quality management and contributes to the refinement of the user requirements in this field. As the
GEOSS public infrastructure evolves, DataFed as a DSS, will seek to access and use the GEOSS-mediated
resources.
The ESIP Air_Quality_Cluster [15] is an activity within the Federation of Earth Science Information
Partners, ESIP [16]. It connects air quality data consumers with the providers of those data. The AQ Cluster
aims to (1) bring people and ideas together (2) facilitates the flow of earth science data to air quality
management and (3) provide a forum for individual AQ projects. The DataFed group is active in the
evolution of the OGC WCS specification to air quality data. A specific goal is to include into OGC WCS
point coverages arising from surface-based monitoring networks.
SummaryConclusion 1
Recent developments in surface and satellite sensing along with new information
technologies now allow real-time, ‘just-in-time’ data analysis for the characterization and
partial explanation of the of major air pollution events as well as more in-depth postanalysis, integration and fusion can also be performed using the federated historical data
resources and tools. By making available many spatio-temporal data sources through a
single web interface and in a consistent format, the DataFed tools allow anyone to view,
process, overlay, and display many types of data to gain insight to atmospheric physical
and chemical processes. A goal of the current effort is to encourage use of these tools by
a broad community of air pollution researchers and analysts, so that a growing group of
empowered analysts may soon enhance the rate at which our collective knowledge of air
pollution evolves. In recent years, such agile analyses have provided occasional real-time
support to the air quality managers and to the public but much more could be done. The
current challenge is to incorporate such support into the air quality management process
in a more regular and robust way.
Unfortunately, retrieving the aerosol properties from the satellite data is rather difficult,
particularly over land with variable surface texture. “Co-retrieval” of aerosol and land
surface reflectance recognizes the fact that land surface reflectance strongly influences
the retrieval of aerosols and vice versa. This work utilizes co-retrieval to characterize the
daily aerosol pattern over the US during the three-year period, 2000-2002.