Characterization of Aerosol Events
R.B. HusarWashington University in St. Louis
Presented at
EPA – OAQPS Seminar
Research Triangle Park, NC, November 1, 2005
NAAMS: National Ambient Air Monitoring Strategy and NCore
Applications
Long-Term Monitoring: Fine
Mass, SO4, K
• Long-term speciated monitoring begun in 1988 with the IMPROVE network
• Starting in 2000, the IMPROVE and EPA networks have expanded
• By 2003, the IMPROVE + EPA species are sampled at 350 sites
• In 2003, the FRM/IMPROVE PM25 network is reporting data from over 1200 sites
Fine Mass
Sulfate
Potassium
Evolution of Spatial Data Coverage: Fine Sulfate, 1998-2003
1998 1999 2000
200320022001
• Before 1998, IMPROVE provided much of the PM2.5 sulfate• In the 1990s, the mid-section of the US was not covered • By 2003, the IMPROVE and EPA sulfate sites (350+) covered most of the US
AIRNOW PM25 - ASOS RH- Corrected Bext
July 21, 2004 July 22, 2004 July 23, 2004
ARINOW PM25 ARINOW PM25
ARINOW PM25
ASOS RHBext
ASOS RHBext
ASOS RHBext
Quebec Smoke July 7, 2002Satellite Optical Depth & Surface ASOS RHBext
Regional Haze Rule: Natural Aerosol
The goal is to attain natural conditions by 2064;Baseline during 2000-2004, first Natural Cond. SIP in 2008;SIP & Natural Condition Revisions every 10 yrs
Natural and Exceptional Event Rule (Making)
Smoke EventJuly 4 2004
July 4 2003
• What is a legitimate Natural or Exceptional event?
• How does one document & quantify the N/E events?
• How is an event treated in NAAQS
Aerosol Event Characterization
• In the past, the definition and documentation of events has been subjective, dependent on the analyst, the is event type etc.
• The routine overall characterization of detected events is accomplished by the rich real-time data through delivered through the Analysts Consoles
• Objective event definition is now possible through spatio-temporal statistical parameters derivable from routine monitoring data
Temporal Analysis • The time series for typical monitoring data are ‘messy’; the signal variation
occurs at various scales and the time pattern at each scale is different
• Inherently, aerosol events are spikes in the time series of monitoring data but extracting the spikes from the noisy data is a challenging endeavor
The temporal signal can be meaningfully decomposed into a
1. Seasonal component with stable periodic pattern
2. Random variation with ‘white noise’ pattern
3. Spikes or events that are more random in frequency and magnitude
Each signal component is caused by different combination of the key processes: emission, transport, transformations and removal
Typical time series of daily AIRNOW PM25 over the Northeastern US
Temporal Signal Decomposition and
Event Detection
• First, the median and average is obtained over a region for each hour/day (thin blue line)
• Next, the data are temporally smoothed by a 30 day moving window (spatial median - red line; spatial mean – heavy blue line). These determine the seasonal pattern.
EUS Daily Average 50%-ile, 30 day 50%-ile smoothing
Deviation from %-ile
Event : Deviation > x*percentile
Median Seasonal Conc.
Mean Seasonal Conc.
Average
Median
• Finally, the hourly/daily deviation from the the smooth median is used to determine the noise (blue) and event (red) components
Seasonal PM25 by Region
The 30-day smoothing average shows the seasonality by region
The Feb/Mar PM25 peak is evident for the Northeast, Great Lakes and Great Plains
This secondary peak is absent in the South and West
Bext Distribution Function
Albany Sigma g = 3.75 Charlotte Sigma g = 1.56
Upper 20 percentile contribution:
Notheast > 45% of dosage Southeast < 30% of dosage
1979
Causes of Temporal Variation by Region
The temporal signal variation is decomposable into seasonal, meteorological noise and events
Assuming statistical independence, the three components are additive:
V2Total = V2
Season + V2MetNoise + V2
Event
The signal components have been determined for each region to assess the differences
Northeast exhibits the largest coeff. variation (56%); seasonal, noise and events each at 30%Southeast is the least variable region (35%), with virtually no contribution from eventsSouthwest, Northwest, S. Cal. and Great Lakes/Plains show 40-50% coeff. variation mostly, due to seasonal and meteorological noise.Interestingly, the noise is about 30% in all regions, while the events vary much more, 5-30%
‘Composition’ of Eastern US Events
• The bar-graph shows the various combinations of species-events that produce Reconstructed Fine Mass (RCFM) events
• ‘Composition’ is defined in terms of co-occurrence of multi-species events (not by average mass composition)
• The largest EUS RCFM events are simultaneously ‘events’ (spikes) in sulfate, organics and soil!
• Some EUS RCFM events are events in single species, e.g. 7-Jul-97 (OC), 21-Jun-97 (Soil)
Based on VIEWS data
The largest EUS Regional PM Event: Nov
15, 2005
Aerosol Event Catalog: Web pages
• Catalog of generic ‘web objects’ – pages, images, animations that relate to aerosol events
• Each ‘web object’ is cataloged by location, time and aerosol type.
Some of the Tools Used in FASTNET
– Data Catalog– Data Browser– PlumeSim, Animator– Combined Aerosol Trajectory Tool (CATT)
Consoles: Data from diverse sources are displayed to create a rich context for exploration and analysis
CATT: Combined Aerosol Trajectory Tool for the browsing backtrajectories for specified chemical conditions
Viewer: General purpose spatio-temporal data browser and view editor applicable for all DataFed datasets
Feb 19 2004: • Isolated high PM25 occurs over the Midwest, Northeast and Texas• The aerosol patches are evident in AIRNOWPM25, ASOS and Fbext maps• The absence of TOMS signal indicates the lack of smoke or dust at high elevation• The high surface wind speed over Texas, hints on possible dust storm activity
• The NAAPS model shows high sulfate over the Great Lakes, but no biomass smoke
• Possible event causes: nitrate in the Upper Midwest and Northeast, sulfate around the Great Lakes and dust over Texas
Jun 6-8• This intensive 3-day episode covers much of the Eastern US• The AIRNOW, ASOS and Visibility FBext are all elevated• TOMS shows smoke(?) over the Gulf and Mexico; MODIS AOT over the Northeast• The surface winds indicate stagnation over the EUS
• NAAPS model shows intense sulfate accumulation over the industrial Illinois-New York .
• Possible causes: sulfate episode
Application of Automatic Event Detection:A Trigger and Screening Tool
• The algorithmic aerosol detection and characterization provides only partial information about events
• However, it can trigger further action during real-time monitoring
• Also, it can be used as a screening tool for the further analysis
FASTNET:
Inter-RPO pilot project, through NESCAUM, 2004
Web-based data, tools for community use
Built on DataFed infra-structure, NSF, NASA
Project fate depends on sponsor, user evaluation
Analysts Consoles for Event Characterization
• Analysts consoles deliver the state of the aerosol, meteorology etc., automatically from real-time monitoring data
• Dozens of maps depict the spatial pattern using dozens of surface and satellite-detected parameters
• The temporal pattern are presented on time series for the regional average and for individual stations
• The following pages illustrate the 2004 EUS events, through a subset of the monitored parameters.
• The event-presentation includes limited interpretative comments; the full interpretation of this rich context is left to subsequent communal analysis
Spatial Console
Temporal Console
Average and 98 Percentile Pattern
SO4
PM2.5 Mass
PM2.5 Mass OC
OC SO4PM2.5 Mass
A V E R A G E
98 Percentile
Exceptional Event Analysis for Regulatory Processes:
Biomass Smoke Aerosol
August-October 2005
Exception Flaggic Waivers
Estimation of Smoke Mass
• The estimation of smoke mass from speciated aerosol data has eluded full quantification for many years
• CIRA, Poirot and others have • While full quantification is still not in hand, a proposed
approximate approach yields reasonably consistent results
• The smoke quantification consists of two steps:– Step 1: Carbon apportionment into Smoke and NonSmoke parts– Step 2: Applying factors to turn OCSmoke and OCNonSmoke into
Mass
Smoke Quantification using Chemical Data
– Step 1: Carbon apportionment into Smoke and NonSmoke partsCarbon (OC & EC) is assumed to have only two forms: smoke and non-smoke
OC = OCS (Smoke) + OCNS (NonSmoke)
EC = ECS (Smoke) + ECNS (NonSmoke)
In each form, the EC/OC ratio is assumed to be constant
ECS/OCS = rs (In smoke, EC/OC ratio rs =0.08)
ECNS/OCNS = rns (In non-smoke, EC/OC ratio rns = 0.4)
With thes four equations, the value of the four unknowns can be calcualted
OCS = (rns*OC –EC)/(rns-rs) = (0.4*OC – EC)/0.32
OCNS = OC-OCS
ECS = 0.08*OCS
ECNS = 0.4*OCNS
– Step2: Apply a factor to turn OC into MassThe smoke and non-smoke OC is scaled by a factor to estimate the mass
OCSmokeMass = OCS*1.5
OCNonSmokeMass = OCNS*2.4
OC – EC Smoke Calibration
PM25
ECOC
Smoke:EC/OC = 0.08PM25/OC = 1.5
OC–EC Non-Smoke Calibration
EC/OC Non-Smoke = 0.15 EC/OC Non-Smoke = 0.2
EC/OC Non-Smoke = 1EC/OC Non-Smoke = 0.4
Negative Smoke – not Possible Maybe??
Maybe?? Too little non-smoke too much smoke
Smoke OC
Non Smoke OC
OCS, OCNS and PM25 Seasonal PatternAverage over 2000-2004 period
PM25Mass
OCS Smoke
OCNS NonSmoke
Day of Year
Mexican Smoke
Agricultural Smoke
Urban NonSmoke Carbon
OC Smoke Spatial Pattern
Dec Jan Feb
Sep Oct Nov
Mar Apr May
Jun Jul Aug
EC NonSmoke
Dec Jan Feb
Sep Oct Nov
Mar Apr May
Jun Jul Aug
PM2.5 (blue) and SmokeMass (red)
Example Smoke Events
Seasonality of OC Percentiles
Great Smoky Mtn:
Episodic OC in the Fall season
Chattanooga::
Elevated and Persistent OC
Monthly Maps of Fire PixelsNOAA HMS – S. Falke
Jan Feb Mar Apr
AugJun JulMay
Sep Oct Nov Dec
Measured and Reconstructed PM25 Mass
• Regional ‘calibration’ constants we applied to OC and Soil
Measured and Reconstructed PM25 Mass, SE
Conclusion
• OC and EC can be apportioned between Smoke and NonSmoke parts
• The reconstructed mass can be matched to the measured PM25
Problems:
• OC Biogenic needs to be separated from OC Smoke
• Scaling OCSmoke and OCNonSmoke to mass needs more calibration
Monthly Maps of Fire PixelsNOAA HMS – S. Falke
Jan Mar
Jul
May
Sep Nov
FASTNET Fast Aerosol Sensing Tools for Natural Event Tracking
DataFed Data Federation
FASTNET is an open communal facility to study non-industrial (e.g. dust and smoke) aerosol events, including detection, tracking and impact on PM and haze.
FASTNET output will be directly applicable, to public health protection, Regional Haze rule, SIP and model development as well as toward stimulating the scientific community.
The main asset of FASTNET is the community of data analysts, modelers, managers and others participating in the production of actionable knowledge from observations, models and human
reasoning
The FASTNET community will be supported by a networking infrastructure based on open Internet standards (web services) and a set of web-tools evolving under the federated data system, DataFed.
DataFed itself is under the umbrella of the interagency Earth Science Information Partners (ESIP) which includes NASA, NOAA and EPA (soon)
Co-retrieval of Aerosol and Surface Reflectance:Analysis of Daily US SeaWiFS Data for 2000-2002
Sean Raffuse, Erin Robinson and Rudolf B. Husar CAPITA, Washington University
Presented at A&WMA’s 97th Annual Conference and ExhibitionJune 22-27, Indianapolis, IN
SeaWiFS Satellite Platform and Sensors
• Satellite maps the world daily in 24 polar swaths
• The 8 sensors are in the transmission windows in the visible & near IR
• Designed for ocean color but also suitable for land color detection, particularly of vegetation
Swath
2300 KM
24/day
Polar Orbit: ~ 1000 km, 100 min.
Equator Crossing: Local NoonChlorophyll Absorption
Designed for Vegetation Detection
Satellite Aerosol Optical Thickness Climatology
SeaWiFS Satellite, Summer 2000 - 2003
20 Percentile
99 Percentile90 Percentile
60 Percentile
Satellite AOT – Time Fraction (0-100%)SeaWiFS Satellite, Summer 2000 - 2003
Dec, Jan Feb
Sep, Oct, NovJun, Jul, Aug
Mar, Apr, May
SeaWiFS AOT – Summer 60 Percentile1 km Resolution
Near Real Time Public Satellite Data Delivery
Early Satellite Detection of Manmade Haze, 1976
Regional Haze
Low Visibility Hazy ‘Blobs’Lyons W.A., Husar R.B. Mon. Weather Rev. 1976
SMS GOES June 30 1975
Temporal Scales of Aerosol Events• A goal of the FASTNET project is to detect and document natural aerosol events in the
context of the overall PM pattern• Inherently, aerosol events are spikes in the time series of monitoring but the definition
and documentation of events has been highly subjective
• Temporal variation occurs at many scales from micro scale (minutes) to secular scale (decades)
• At each scale the variation is dominated different combination of the key processes: emission, transport, transformations and removal
• Natural aerosol events occur mostly at synoptic scale of 3-5 days
Discussion: The Role of Averaging Region
• The size and location of the region strongly influences the event-detection; e.g. events in the Northeast occur at different times than Southwestern events.
• ‘EUS events’ can occur either from a single contiguous ‘haze blob’ or from multiple smaller aerosol patches at different parts of the Eastern US
• It would be desirable to develop a detection scheme that can identify events as they occur at different time and spatial scales
What kind of neighborhood is this anyway?
May 9, 1998 A Really Bad Aerosol Day for N. America
Asian Smoke
C. American Smoke
Canada Smoke