Real-time PM2.5 Reporting

1
Real-time PM2.5 Reporting

• Prepared by
• Dianne S. Miller
• Adam N. Pasch
• Alan C. Chan
• Timothy S. Dye
• Sonoma Technology, Inc.
• Petaluma, CA
• Presented to the National Air Quality Conference
• Dallas, TX
• March 3, 2009

STI-3549
2
Real-time Hourly AQI Values for PM2.5
3
Why is it Needed?
We have
Step 1
Real-time hourly PM2.5 concentrations
24-hr avg. PM2.5 concentrations
Surrogate
We need real-time PM2.5 AQI
Step 2
Convert to AQI
Insert Map
4
Goal of the Presentation

• Provide an overview of the PM2.5 surrogate
• What is it?
• How does it work?
• Why is it important?
• Present performance expectations
• Discuss its performance
• Suggest possible methods for improving it

5
What is the Surrogate? (1 of 2)

• The surrogate is an hourly estimation of the
  midpoint 24-hr average PM2.5 concentration for a
  given site.
• The Conroy Method is currently being used (See
  McMillan and Hamilton, 2003).

Unknown
4-hr average
Have 12 hours of data
12-hr average

Need 24 hours of data to compute the AQI
6
What is the Surrogate? (2 of 2)

• Surrogate ((1212-hr average)12(4-hr adjusted
  average))/24
• Where the
• 12-hr PM2.5 average represents what has already
  happened
• Adjusted 4-hr PM2.5 average captures short-term
  changes to better estimate the future 12 hours.
  The adjustment to the average depends on
• How high or low the concentration is
• How rapidly the concentration is changing

7
Why is an Accurate Surrogate Important?

• Protect public health
• Avoid false alarms
• Provide real-time information about PM2.5

8
Performance Expectations

• The surrogate should
• Work for all regions
• Be easily implemented
• Be within 80 of the observed PM2.5
  concentrations 80 of the time
• Identify PM2.5 spikes and recover rapidly when
  concentrations drop
• Not be overly sensitive to local sources

9
Performance Evaluation - Methods

• Data used
• 2006-2007 daily and hourly PM2.5 data from AIRNow
• Six cities
• Bakersfield ? Sacramento
• Chicago ? Atlanta
• Liberty, PA ? Baltimore
• Test metrics
• Metric 1 - How accurate is the surrogate by time
  of day?
• Metric 2 - How accurate is the surrogate within a
  four-hour window?
• Metric 3 - How does the peak daily surrogate AQI
  compare with the observed daily maximum AQI?

10
Metric 1
USG hours

• Percentage of hours by time of day the surrogate
  correctly predicted the observed USG AQI category

12 am
Percent of Hours Correct () Percent of Hours Correct () Percent of Hours Correct ()
City All AQI USG
Baltimore 91 4
Liberty 80 53
Atlanta 89 47
Chicago 90 42
Bakersfield 87 69
Sacramento 88 42
Good 80 100
Fair 51 79
Poor 0 - 50
11
Metric 2
USG hours

• Percentage of hours the surrogate AQI category
  matches within four hours of occurrence of the
  observed AQI

Underestimate
Overestimate
Correct
Percent of Hours Correct () Percent of Hours Correct () Percent of Hours Correct ()
City All AQI USG
Baltimore 92 16
Liberty 79 56
Atlanta 89 55
Chicago 90 47
Bakersfield 88 78
Sacramento 87 45
Good 80 100
Fair 51 79
Poor 0 - 50
12
Metric 3

• Air quality agencies often show both the
  surrogate AQI and forecasted AQI on their
  websites.
• The surrogate AQI and the forecasted AQI are
  often perceived as the same metric, which can
  lead to reduced public confidence in the forecast
  when they differ.

Percent of Days Matching () Percent of Days Matching () Percent of Days Matching ()
City All AQI USG
Baltimore 78 100
Liberty 50 33
Atlanta 77 71
Chicago 75 67
Bakersfield 70 84
Sacramento 67 37
Percent of time the peak surrogate matched the
daily observed AQI category
13
Performance Summary
Metric All Days USG Days
1 All hours Good Poor
2 Four-hour window Good Poor
3 Peak daily AQI Fair Fair
14
Alternative Methods

• Re-work the current surrogate method.
• Moderate level of effort, low likelihood of
  improvement
• Use characteristics of PM2.5 at each site to
  determine surrogate. Taking into consideration
  diurnal variability, geographic region, season,
  and time of day.
• Moderate to high level of effort, moderate
  likelihood of improvement
• Use PM2.5 model predictions to estimate missing
  12 hours of data.
• Larger level of effort, higher likelihood of
  improvement

15
Challenge!

• A challenge to you experiment with new
  surrogate formulations
• Contact airnowdmc_at_sonomatech.com for a dataset to
  use
• Thanks!