Fusion of SeaWIFS and TOMS Satellite Data with Surface Observations
and Topographic Data During Extreme Aerosol Events
Stefan Falke, Rudolf Husar and Bret Schichtel
Center for Air Pollution Impact and Trend Analysis
CAPITA, Washington University, St. Louis, MO 63130
Abstract
Spaceborne sensors now allow near continuous aerosol monitoring throughout the world. This
paper illustrates the fusion of SeaWiFS and TOMS satellite data with surface observations and
topographic data during four extreme aerosol events, 1) the April 1998 Asian dust impacting the
west coast of North America, 2) the May 1998 Central American forest fire smoke impacting
eastern North America, 3) the fall 1999 Northern California fires, and 4) the February 2000
Sahara dust storm. During these dust and smoke events, the aerosol was visualized on SeaWiFS
images as a distinct yellowish dye. TOMS imagery also indicated elevated aerosols. Surface
monitors in the area of the smoke and dust measured major reductions in visual range. Merging
these data provides support for identifying the spatial extent of the aerosol plumes and provides
verification that what is being seen is actually dust or smoke. The experience during the extreme
aerosol events has demonstrated that currently available satellite sensors can provide valuable
support to science and air quality management.
Introduction
Extreme biogeochemical events such as volcanic eruptions, forest fires, and dust storms provide
unique opportunities to examine the inner workings of the atmospheric system. Such events tend
to produce large quantities of dust, smoke, or haze, which is then dispersed over regional or
global scales. The easily observable atmospheric particles can visualize and quantify the nature
of transport, transformation and removal processes along their path. Satellite images offer higher
spatial resolution than surface observations and are available in areas with few surface
observations are, particularly over oceans. Advances in technology provides with high
resolution, color images that can detect aerosol plums. When fused with surface data they can
better describe the aerosol and provide a greater understanding of the spatial, temporal, and other
characteristics of the aerosol.
Data Fusion
Fusing, or integrating, available data sets generates knowledge of pollutant characteristics and
transport that are not readily discernible from a single sensor. The integration of data from
multiple sources is referred to as data fusion. Data fusion is a relatively new field in its
application to earth remote sensing. However, the field has a rich history in physiology and
neural sciences, as well as their artificial counterpart in robotics where automated mechanisms
combine information from multiple knowledge sources to improve the understanding of a given
scene (Abidi and Gonzalez, 1992). Fusion occurs on different levels of a data and information
processing system: pixel, feature, and object (Luo and Kay 1992; Pohl and van Genderen, 1998;
Wald, 1999). Pixel level fusion merges measured data, such as satellite sensor radiance values,
instrument calibration, geometric correction, and atmospheric information including aerosols.
Different satellite sensors can identify different properties, or features, of aerosols, and when
fused can identify the aerosol with more certainty. The resulting images are analyzed for feature
extraction, such as the spatial extent of the aerosol plume or its size distribution. These features
from multiple data sources are then combined to formulate the characteristics of the aerosol
being measured.
Figure 1. Three levels of data fusion: pixel, feature, and object.
In Figure 1, multiple images from satellite sensors, such as SeaWiFS or TOMS, are fused. This
level of fusion could also include point data such as surface observations. The resulting images
are analyzed for feature extraction, such as the spatial extent of the aerosol plume or its size
distribution. These features from multiple data sources are then combined to formulate an idea
of what type of aerosol is being measured from the multiple sensors or data sources. For this
work, the types of features extracted from the data fusion consist mainly in defining the spatial
characteristics, both horizontal and vertical extent, of aerosols.
Data Sources and Preparation
The data used in the fusion efforts presented in this work consist of satellite images, surface
observations, and digital elevation model.
SeaWiFS
Sea-viewing wide field-of-view sensor (SeaWiFS) was designed as an ocean color sensor that
collected sea surface color and other ocean bio-optical properties. It is used extensively for algae
bloom monitoring, tracking oil spills, monitoring water pollution, among many other uses
(Chavez et al 1999; Woodruff et al, 1999). However, its daily visible color images can provide
some striking images of other non-ocean events, such as dust storms and smoke (Hotz 1999).
Spectral reflectance data from the SeaWiFS (www) sensor [McClain et al., 1998; Barnes et al.,
1999] provided detailed spatial pattern of the smoke at local noon each day. The raw (Level 1A)
Local Area Coverage (LAC), 1 km resolution SeaWiFS data were downloaded from the
SeaWiFS Program [McClain et al., 1998] and processed at Washington University. In the first
stage of processing, the scattering by air molecules was removed from the total reflectance using
a procedure by [Vermote and Tanre, 1992], which also included nominal corrections for ozone
and water vapor absorption. Next, the pixel radiance values were transformed to reflectance. The
calculated spectral reflectance values (fraction of radiation reflected) represented the combined
reflectance from the land, clouds and the ambient aerosol.
TOMS
TOMS has been in use since 1978 on the Nimbus-7 platform, scanning at ultraviolet
wavelengths. TOMS is most well known for mapping ozone, however, TOMS is sensitive to
absorbing aerosols and can be used to monitor the motion of large aerosol clouds. Retrievals of
absorbing aerosol index from the TOMS satellite [Herman et al., 1997] provided useful
information on the daily spatial distribution of dust and smoke. The TOMS absorbing aerosol
signal is a semi-quantitative index of the columnar absorption by aerosols at 0.34 m. The signal
is derived from the absorption of the upwelling Rayleigh scattering in the lower strata of the
atmosphere. The columnar absorbing aerosol index (AAI) was used to locate the smoke sources
and to estimate semi-quantitatively the magnitude and distribution of smoke aerosol plumes. The
daily gridded AAI data were obtained from the NASA TOMS project website (www)
Surface Visibility Observations
The National Weather Surface maintains a surface observation database that contains visibility
observations at airports across the U.S. The U.S. Navy also maintains a database of global
visibility-reducing surface weather reports ( )
The smoke aerosol was also detected using visual range data at over 200 surface meteorological
stations distributed throughout Eastern North America. The visibility data were appropriately
filtered for precipitation and also corrected for relative humidity [Husar et al., 1981]. The hourly
National Weather Service surface observations were downloaded from the National Climatic
Data Center (www).
DMSP Fire Observations
During the 1998 fire season, the fire locations were reported daily using the nighttime visible
images from the OSR sensor, on the DMSP (www) meteorological satellite. Fires were
distinguished from the nighttime city lights by their transient behavior [Elvidge et al.,1996].
Digital Elevation Model
GTOPO30 is a global digital elevation model (DEM) with a horizontal grid spacing of 30 arc
seconds (approximately 1 kilometer). GTOPO30 was derived from several raster and vector
sources of topographic information. For easier distribution, GTOPO30 has been divided into tiles
which can be selected from the map shown above. The tiles were downloaded from the
GTOPO30 website. They were imported into the satellite image analysis software program
ENVI and were merged using ENVI’s mosaic functionality.
Aerosol Event Case Studies
The usefulness of data fusion is illustrated through the examination of extreme aerosol events
including 1) the April 1998 Asian dust impacting the west coast of North America, 2) the May
1998 Central American forest fire smoke impacting eastern North America, 3) the fall 1999
Northern California fires, and 4) the February 2000 Sahara dust storm.
1998 Asian Dust Storm
During April 1998, large dust storms were generated from the Gobi Dessert. On the SeaWiFS
images the dust cloud was recognized by its bright yellow color, partial transparency and smooth
spatial texture. A superposition of the TOMS and SeaWiFS data in Figure 2 indicates that on
April 19 the dust pattern from TOMS and SeaWiFS coincides geographically. This implies that
the dust layer was higher than on April 15, since in this case the TOMS sensor detected it.
On April 19, the surface wind speeds from the Naval Research Laboratory increased to over 20
m/s as shown in the red wind speed contours on Figure 2. This was well above the generally
assumed threshold wind speed (5-6 m/s) for dust suspension [Gillette, 1978]. The region of high
wind speeds coincided with the Gobi Desert. The NRL surface visibility observations of low
visibility also coincide with the area of dense dust. A dense dust front is clearly visible at the
leading edge of the dust cloud. The dust cloud would eventually be transported across the Pacific
and reach the North American continent where it contributed to the exceedance of the PM10
standard.
Figure 2. Integrated image of dust over the Gobi Dessert on April 19, 1998. The underlying color image is the
surface reflectance derived from SeaWiFS. The TOMS absorbing aerosol index (level 2.0) is superimposed as green
contours. The red contours represent the surface wind speed from the NRL surface observation data base. The blue
circles are also from the NRL database and indicate locations where dust was observed. The high wind speeds
generated the large dust front seen in the SeaWifS, TOMS, and surface observation data. (Husar et al. 2000a).
1998 Central American Fires
In 1998, the annual springtime fires were more intense than usual. From May 7-17, 1998, smoke
from numerous widespread fires in Central America drifted northward and caused severe
perturbation of the atmospheric environment over parts of Eastern North America. After a period
of stagnation in the gulf of Mexico, a dense smoke cloud was transported northward into the
central United States. On the SeaWiFS images the smoke cloud was recognized by its yellow
coloration near the source, partial transparency and smooth spatial texture compared to whitegray and highly textured clouds (Figure 3). The color images of SeaWiFS provided a rich visible
context for the radiative effects of the smoke, including the magnitude of spectral backscattering,
relationship to clouds and the spectral albedo of the underlying landmasses. The LAC data from
adjacent spacecraft swaths were then merged and georeferenced to produce contiguous coverage
for Eastern North America shown in Figures 2a-d.
The daily maps represent the spatial superposition of three complimentary aerosol signals,
SeaWiFS, TOMS, and extinction coefficient. The color SeaWiFS images were used as a
substrate for the numerical aerosol data derived from the TOMS sensor (green lines) and the
extinction coefficient derived from surface-based meteorological network (red lines). The
spectral reflectance images from SeaWiFS provide a rich visual context relevant to the smoke,
including surface reflectance, position of cloud systems relative to the smoke, as well as
indications of wind direction.
Figure 3a shows the SeaWiFS data, rendered on top of a three-dimensional elevation map. As
indicated in Figure 2a, the smoke dispersion plumes over Central America appears mostly over
low elevation terrain, while high elevation regions are largely smoke free. This suggests that the
high elevation terrain acts as a barrier to the dispersion of smoke. The daily fire map derived
from the DMSP sensor (Figure 2a) shows that on a specific day (May 15, 1998) fire spots occur
at thousands of locations throughout the Central America. However, based on the currently
available satellite “fire products” it is not possible to estimate the magnitude of the smoke
emissions.
Figure 3. Surface reflectance derived from the SeaWiFS satellite data for May 14, 1998. The spectral reflectance data
were rendered as a "true color" digital image by combining the blue (0.412 m), green (0.550 m), and red (0.670 m)
channels. The TOMS absorbing aerosol index (green, levels 12 and 30) and the visibility-derived extinction coefficients
are superimposed as green contours (red, levels 0.2. and 0.4 m -1).
1999 California Fires
In September and October of 1999, several major forest fires occurred throughout California. In
addition to the direct fire damage, the thick smoke plumes required the evacuation of inhabitants
in some affected downwind areas. From a regulatory standpoint, the smoke plumes were
‘exceptional events’ for particulate matter.
Note the yellow coloration of the smoke plume. Also, in the San Joaquin Valley the smoke is
confined to the low elevations while the mountains are evidently outside the smoke layer.
The SeaWiFS image was “wrapped” onto the surface of a 1 km resolution digital elevation
model. The 3-D rendering highlights the influence of elevation barriers on the transport of
smoke.
Figure 4. SeaWiFS image on October 18, 1999. a) a 2-D view of the San Joaquin Valley with fires on the eastern
side of the Valley and smoke dispered to the west and heading out over the Pacific just north of San Francisco. B)
Fusion of topographical data with the SeaWiFS image to illustrate the
Daily color satellite data can be used to semi-quantitatively detect the extent of smoke and haze
over California. Combining the SeaWiFS image with elevation illustrates the topographic
influences on the dispersion of the smoke within the San Joaquin Valley.
2000 Saharan Dust
Three dimensional rendering of the Sahara dust event aids the visualization and analysis of dust
transport and influence of topography on transport. The following example illustrates how fusing
SeaWiFS imagery and a digital elevation model (DEM) can be applied to generate 3D images of
the Sahara dust event. The 3D images are used to estimate the height of the Sahara dust layer
over the Canary Islands.
GTOPO30 is the source of the digital elevation model (DEM) data. It is available at a 30 arc
second spatial resolution (~1 km) and was downloaded from the ISGS EROS Data Center Distributed Active Archive Center. Both the SeaWiFS image and elevation data are
georeferenced onto a geographic lat/lon projection. The 3D surface plotting facility in the remote
sensing image analysis software package ENVI (ENVI, 1998) was used to generate the 3D
images by draping the SeaWiFS image onto the 3D DEM surface.
Figure 5. The region used for 3D rendering of the Sahara dust event. The dust is transported off the west coast of
Africa and across the Canary Islands. A 3D view of the image indicates that the Fuerteventura and Lanzarote
Islands are fully blanketed by the murky yellow colored dust plume. Gran Canaria and Tenerife are partly covered
by the dust layer but their higher elevations appear to stick out above the dust layer at about 1200m as indicated by
their dark green color.
On February 26, 2000, the Sahara dust cloud passed over the Canary Islands. The islands of
Fuerteventura (peak elev. 800 m) and Lanzarote (peak elev. 650 m) are fully immersed within
the dust layer. The lower elevations of islands of Tenerife (peak elev. 3700 m) and Gran Canaria
(peak elev. 1950 m) are also within the dust layer, but their tops protrude into the dust-free
air. Hence, the dust layer over the Canary Islands is evidently within the lower stratum of the
atmosphere, extending from the surface to about 1000 m. As a consequence, the dust flows
around rather than over the Tenerife and Gran Canaria islands. This illustration suggests that
the in rugged terrain, the fusion of satellite remote sensing and elevation data allows the
estimation of the aerosol layer height.
In this method, the SeaWiFS image was simply draped onto a 3D topographic surface so that
land, water, dust, and clouds were all projected directly onto the topographic surface. A more
realistic and revealing rendering scheme would account for the appropriate heights of each of the
atmospheric components. The elevations were exaggerated in the 3D images to aid the
visualization of the dust layer height dependence. The estimation of the dust layer height was
done using only a visual analysis of the SeaWiFS and DEM images. A more rigorous approach
using pixel spectral analysis may provide more accurate height estimates. This example only
focused on a limited area and cannot be used for synoptic scale analysis.
Summary
The events studied in this paper were used data fusion of satellite, surface, and model results to
define the spatial (horizontal and vertical) characteristics of the aerosol events. Advances in
satellite sensors and expansion of sensor wavelengths and observation angles. More recent
satellite sensors and future sensors are designed specifically for monitoring aerosols and will
provide the opportunity to fuse these newer images to derive other characteristics of aerosols
such as chemical composition and size distribution.
http://edcdaac.usgs.gov/gtopo30/gtopo30.html
http://seawifs.gsfc.nasa.gov/SEAWIFS/HTML/dust.html
http://www.rsinc.com/envi/index.cfm
References