Integration of Satellite and Surface Aerosol Observations with Modeling
Introduction [fragment, NEW]
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.
Surface PM 2.5 mass concentration and extinction coefficient data. The mass
concentration of fine particles, PM2.5, is being monitored in real time at about 300 sites
over the US. The sites are operated by variety of organizations. The hourly- federated
data are collected into a database by the highly successful EPA AIRNow program. While
the primary goal of this program is to deliver air quality information to the public in form
of Air Quality Indices, the collected PM2.5 data are also available to the air quality
research, air quality forecasting, and for other AQ management activities. A good
example of the AIRNOW PM2.5 data usage is the NASA Langley IDEA project in which
the operational MODIS AOT data are compared to the PM2.5 Baron Advanced
Meteorological Systems NASA Decisions/05-2-0000-00498 concentrations in real time
and the transport direction and speed of observed hazy air masses is predicted using
forecast trajectories.
A snapshot of the AIRNOW PM2.5 concentration is shown in Figure 6a. The circle
diameter at each PM2.5 monitoring site is proportional to the PM2.5 concentration at that
location and hour. Spatial extrapolation to a continuous grid can be accomplished using
inverse distance weighing, Kriging, or by more elaborate physically based extrapolation
schemes. These operations are now routine features of CAPITA DataFed and other data
analysis software packages. The real-time PM2.5 concentration map clearly indicates that
the station density is non-uniform and multi-state regions of the country have no real time
PM2.5 monitoring sites within 300 km. Hence, the PM2.5 estimation error in those areas
is significant.
Fortunately, the spatial texture of PM2.5 concentration can be significantly improved by
the use of the surface visibility data obtained by the ASOS forward scattering aerosol
instrument. Since for ambient aerosol scattering comprises over 95% of the aerosol
extinction, the ASOS data can be expressed as extinction coefficients. The ASOS
visibility data are reported in realtime (within an hour or two) by the global METAR
surface weather reporting system. At any given hour there are about 1200 monitoring
sites reporting visibility over the conterminous US. As seen in Figure 6b, the spatial
density allows detailed mapping of the surface extinction coefficient and indirectly the
PM2.5 concentration.
The ASOS data represent the in-situ forward light scattering that is caused by fog,
clouds,mist, as well as by ambient aerosols arising from urban industrial emissions, windblown dust, biomass smoke, and other sources. In order to use the in-situ visibility data as
a surrogate for dry aerosol, the contribution of cloud, fog, and mist needs to be identified
and eliminated. The filtering and adjustments (e.g. RH correction) of the visibility data
have been developed at CAPITA over the past three decades and are incorporated in the
DataFed processing stream for the surface meteorological data. An example plot of the
surface aerosol extinction coefficient is shown in Figure 6b. Comparing the spatial
pattern of the AIRNOW PM2.5 concentration and the ASOS Bext shows the improved
spatial texture for ASOS as well as the spatial coverage over areas void of real-time
PM2.5 concentration.
The use of surface visibility as a surrogate for dry PM2.5 concentration is justifiable on
the grounds that the surface light scattering exhibits a strong correlation with fine particle
mass. In fact, over most of the US, the light scattering to fine mass ratio, i.e., the mass
extinction efficiency as of fine mass ratio is in the range of 3-5 m2/g. Notable exception
to this well established light scattering-PM2.5 relationship is wind-blown dust which is
about 0.3 m2/g. Scattering efficiency factors are being developed as part of past and
ongoing research at Washington University. The fusion of the PM2.5 and ASOS data will
explicitly incorporate the geographic and seasonal pattern of the mass extinction
efficiency.
The surface data that will be provided for assimilation into the BAMS aerosol forecast
model will consist of (1) the in-situ extinction coefficient and (2) PM2.5 concentrations.
This may be delivered as a gridded map derived as the “best estimate” fusion of the
AIRNOW and ASOS data, or the data may be provided as raw surface observations. In
the former case, the insitu extinction data will be used to estimate the aerosol scale height
as described below. Baron Advanced Meteorological Systems NASA Decisions/05-20000-00499
Both the PM2.5 concentration and the ASOS data are available in real-time from their
respective providers (AIRNOW, National Weather Service). The real-time access to the
data from their providers and the processing will be performed through the DataFed data
access, transformation, and delivery system, http://datafed.net. The open DataFed data
exchange facility is maintained at Washington University and it is supported by grants
from NSF, NASA, EPA, state-Regional Planning Organizations and other agencies
interested in the open sharing of air quality relevant data. The proposed project will
leverage the existing support in the DataFed system (about $350,000/yr). The data
filtering, fusion, and gridding for this project will be performed by Lantern Corporation,
the consulting arm of the CAPITA group.
[New] Data-Model Integration Framework: The challenges of unpredictable smoke
emission modeling and the opportunities arising from rich real-time fire location and
smoke data suggests an observation-based smoke emission monitoring strategy.
(http://capita.wustl.edu/capita/capitareports/050511EastFire_GMason/1B1Husar.doc).
The approach depicted schematically in Figure 2, uses a smoke dispersion model to
simulate emissions. The model is driven by the best available observed fire locations,
land and fuel conditions and transport winds. However, the pattern of smoke emission
rates is derived from the observed smoke data assimilated into the smoke model. This is
an inverse modeling approach to emission estimation. The smoke observations include
satellite data from MODIS, GOES, AVHRR, TOMS and other emerging sensors. The
surface observations include continuous PM25 (EPA AIRNOW), surface visibility (NWS
ASOS), chemical measurements from IMPROVE and STN networks and miscellaneous
sensors such as sun photometers (AERONET), lidar (e.g. MPLNet) etc. Each dataset is
contributed by different agency. A key aspect of the framework is the information system
that brings together, homogenizes and fuses the numerous observational and model data
on various aspects of smoke.
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.
Assimilation Algorithms. Historically, a variety of algorithms have been used to
assimilate observations for meteorology modeling: a number of gridding and Kriging
techniques culminating in the technique called Optimal Interpolation, variational
techniques (3D-Var and 4D-Var), and various Kalman filter and ensemble Kalman filter
based techniques. The initial assimilation modules in the Assimilator will employ 3D-Var
and OI style techniques. However, we may be able to do this more “intelligently” than
has been done for models like GOCART— a result of having at least some species
differentiation in MODIS combined with the surface observations. We note that the
problem, however, is quite complex. Full characterization of atmospheric aerosols system
requires at least six independent dimensions: x,y,z,t, particle size, composition. The first
four are standard physical dimensions of the atmosphere, while the particle size and
composition are dimensions unique to the aerosols. In a sense, the aerosol system is
equivalent to a mixture of hundreds of gaseous species, each having additional
composition and “size”. Detailed chemical speciation of size segregated samples is now
obtained from over 350 monitoring stations as part of the IMPROVE and the EPA
speciation networks.
Hence, these measurements provide considerable detail along the chemical species and
particle size dimensions. Surface monitoring data (PM2.5 and ASOS) also provide
moderate-resolution surface aerosol characterization in space and time (x,y,t). However,
there are virtually no data on the vertical distribution of aerosols. It is fair to say that at
this time the least understood dimension of the atmospheric aerosol system is the vertical
distribution in the atmosphere Fortunately, MODIS Deep Blue data not only has total
optical depth, it also has Angstrom exponent, which can be used to differentiate between
dust, smoke, and sulfate/nitrate aerosols, giving rise to the possibility of three scaling
factors that would apply to a vertical column, not just one. The surface observations
further may allow us to differentiate upperatmosphere scaling factors from loweratmosphere (mixed-layer) ones. Since the AOT is a
vertical integral of the light extinction coefficient attributed to aerosols. In fact, the AOT
is an integral over other dimensions: aerosol type, particle size, angle, shape, etc. The
surface extinction coefficient Bext, 1/m, on the other hand represents the local aerosol
extinction at the surface. From the two measurements it is possible to estimate aerosol
scale height,H=AOT/Bext, that has a unit of meters. By definition the scale height is the
physical height of Baron Advanced Meteorological Systems NASA Decisions/05-20000-004910 the uniformly distribute aerosol layer that has the same extinction
coefficient as the aerosol at the surface (Figure 7).
Knowing the aerosol scale height will not reveal the detailed vertical texture of the
aerosol. However, it provides an estimate of the ”characteristic” aerosol height that can
be compared to the model derived aerosol scale height. The aerosol scale height can be
derived from the model simulations by vertically integrating the calculated aerosol
scattering and dividing the integral with the values at the surface. A comparison of the
model derived and observed aerosol scale height provides virtually the only opportunity
for model validation and nudging along the vertical dimension. Knowledge of the aerosol
scale height will be particularly useful for identifying thick aerosol layers that are
transported aloft. It is well known from lidar, aircraft and other observations that major
smoke plumes that are transported to long distances reside mostly in elevated layers,
above the reach of the daily mixing layer. The detection of such elevated layers can now
be made routinely by the presence of high vertical AOT (MODIS) and low surface
extinction. An outstanding difficulty arises from the fact that the chemical models need to
be operated for individual aerosol components (sulfate, organics, nitrate, dust) while the
real-time observations are only available for the aggregate sum of all aerosol
components. Speciated chemical data are usually available only months after the
sampling. Hence, in matching the model to the observations it is not obvious which of the
aerosol components needs to be nudged and by how much. From the observations it is
only possible to determine the overall result of the nudging for all components.
For aerosol data assimilation over the eastern US domain, we are proposing a novel
assimilation procedure. Data assimilation for the eastern US domain can benefit from the
known behavior of the aerosol system. Sulfates constitute a major (50+%) of the fine
particle mass. Also, the sources, transport, and inetic processes for sulfates are well
understood. For this reason, sulfate simulation can be performed reliably and therefore it
will not require daily nudging. Given this “known” sulfate component, it is possible to
infer the concentration of organics (the next major aerosol component), from the
difference between the observed total PM2.5 and the simulated sulfate. The nitrate can be
estimated by similar approach, but utilizing the knowledge that nitrates occur mainly
during the winter months over the agricultural Upper Midwest. This proposed excessnon-sulfate aerosol approach to the determination of carbonaceous and nitrate aerosols is
expected to improve the quality of the real-time aerosolforecasts significantly.
Secondly, the observance of large quantities of dust or smoke in the interior of the model
domain that are not reflected in the previous-cycle model forecast may be assumed to be
the result of emissions events that could not have been forecast, such as fires or dust
storms. These events can be assumed to be lower-atmosphere/surface phenomena, and we
will attempt to extrapolate the atmospheric emissions tendencies that give rise to them,
and introduce those emissions tendencies into the forecast system. Finally, the forecasts
are influenced by boundary conditions—where remote events outside the modeling area
introduce long-lived long-rangetransport of aerosol into the modeling domain. The 2002
Alaskan fires, for example, influenced the ability of a number of researchers who were
attempting to validate their aerosol models over Baron Advanced Meteorological
Systems NASA Decisions/05-2-0000-004911 the CONUS. The assimilation system will
attempt to estimate these effects, and thereby forecast their effects upon the interior of the
modeled domain. Since these boundary effects themselves vary over time, it may be
necessary to model a large buffer zone around the actual forecast area (a zone across
which transport time would exceed the time between forecast cycles). Note that under
conditions such as the 2002 Alaskan fires, being able to determine the vertical structure
of the aerosol boundary condition data is a particularly important open question.
During the second and third project years (see below), the aerosol model performance
will be tested with comparison to chemical data (sulfate, organics, nitrate, dust) from the
combined 350 station IMPROVE/EPA Speciation database available through the CIRAVIEWS database. However, such comparison and evaluation will be avaliable only for
data 3-12 months after the observations. The real-time assimilation of monitoring data
also provides an immediate feedback on the model performance. For instance under
conditions of predominantly sulfate aerosols e.g. stagnation over the over the Ohio River
Valley source region, the data-model comparison will yield direct quantification of model
performance. Such evaluation will be performed operationally as part of the daily
monitoring process.
Applications of the Project [NEW]
The haze portion of the proposed work will be incorporated into decision processes of the
Regional Planning Organizations (RPOs) for regional haze. The RPO activity, FASTNET
(Fast Aerosol Sensing and Tools for Natural Event Tracking) (Poirot, et al, 2004) is a
data acquisition and analysis facility for improving the efficiency as air quality analysts,
with particular emphasis on detailed real-time and post-analysis of major aerosol events.
For the RPOs, natural aerosol events from forest fire smoke and windblown dust are
particularly interesting due to their large emission rates over short periods of time,
continental and global-scale impacts, and unpredictable sporadic occurrence. This
proposed work on smoke, dust and regional haze will be directly applicable to the RPOs
and the results will be incorporated into FASTNET which is managed by the Washington
University CAPITA group for the RPOs.
Another new application area for the proposed satellite data assimilation is for
Exceptional Events Rule under development by EPA. The National Ambient Air Quality
Standards for PM2.5 allow flagging certain days with Exceptional Event flags due to
natural and ‘extra-jurisdictional’ events such as smoke, dust or intercontinental pollution
transport. The State agencies submitting the flagged data are required to provide
appropriate justification for the exceptional event flags. For the first time in the
regulatory history, the new EPA Rules will allow the use of satellite data for such
exceptional event documentation. Our project team (Husar and coworkers) has been
developing the key technical components of the Exceptional Event Rule including the use
of satellite-based evidence. The data, tools and experience resulting from this project can
greatly benefit this evolving process and facilitate the further infusion of satellite data
into the AQ decision making processes.
Poirot, R. L. et al. Aerosol and Haze observations during 2004 with the FASTNET Distributed Monitoring System. Presented at the
A&WMA Specialty Conference on Regional and Global Perspectives on Haze: Cause, Consequences and Controversies, Asheville,
NC, 2004; Paper # 93.
Husar, R B..; Poirot, R. DataFed and FASTNET: Tools for Agile Air Quality Analysis; Environmental Manager 2005, September, 3941