Improving Real-Time AIRNow Maps Using Data Fusion

1
Improving Real-Time AIRNow Maps Using Data Fusion

• Prepared by
• Patrick H. Zahn and Clinton P. MacDonald
• Sonoma Technology, Inc.
• Petaluma, CA

Scott A. Jackson U.S. Environmental Protection
Agency Research Triangle Park, NC
Presented at the National Air Quality
Conference Dallas, TX March 2-5, 2009
3548
2
Outline

• Introduction
• Approach for Data Fusion
• Fusion Example with Ozone Observations
• Fusion Example with Ozone Forecasts
• Future Work

3
Introduction (1 of 2)
Spatial coverage of current AIRNow maps is
lacking.
4
Introduction (2 of 2)
Previous work has focused on producing accurate
maps with improved spatial coverage using
interpolation in ArcGIS.
5
Objective
Current work explores techniques to use model
predictions to extend the spatial coverage of air
quality information.
6
Approach for Data Fusion (1 of 2)

• The following cases were used to test the fusion
  of air quality model predictions with
  observations and forecasts
• Fuse ozone observations with NOAA ozone model
  predictions
• Fuse PM2.5 observations with BlueSky Gateway
  predictions
• Fuse agencies ozone forecasts with NOAA ozone
  model predictions
• Caveat this is a pilot project/proof of concept

7
Approach for Data Fusion (1 of 2)

• The following cases were used to test the fusion
  of air quality model predictions with
  observations and forecasts
• Fuse ozone observations with NOAA ozone model
  predictions
• Fuse PM2.5 observations with BlueSky Gateway
  predictions
• Fuse agencies ozone forecasts with NOAA ozone
  model predictions
• Caveat this is a pilot project/proof of concept
• We will focus on these two cases.

8
Fusion of Ozone Observations and NOAA Model
Ozone Predictions

• The current ozone monitor network has large gaps.
• Regions with insufficient monitor coverage can be
  defined using Standard Error a measure of the
  uncertainty in ozone estimates in areas without
  monitors.

April 19, 2008, 8-hr Maximum Ozone Station
Observations with Standard Error Mask
9
Fusion of Ozone Observations and NOAA Model
Ozone Predictions Method (1 of 2)
A new estimation of ozone concentrations was made
by interpolating across both observations (dots)
and the NOAA model predictions.
April 19, 2008, 8-hr Maximum Ozone Station
Observations and NOAA Model Output
10
Fusion of Ozone Observations and NOAA Model
Ozone Predictions Method (2 of 2)

• Issue NOAA model predictions tend to dominate
  the resulting interpolation because of their high
  spatial density.
• Solution NOAA model predictions are randomly
  sampled at a density similar to the monitor
  network density.

8-hr Maximum Ozone Observations and Randomly
Sampled NOAA Model Output
11
Fusion of Ozone Observations and NOAA Model
Ozone Predictions Results

• The fused product
• Provides additional information in monitor-sparse
  areas (black circles)
• Limits the influence of data from monitors in
  remote areas (blue circles)
• Note Only Moderate and above AQI levels are
  shown on green terrain background

Interpolation with fusion (observations and NOAA
model)
Interpolation without fusion (observations only)
12
Fusion of Ozone Forecasts and NOAA Model Ozone
Predictions

• Agencies forecasts are currently represented
  with dots
• The gaps between agency forecasts are too large
  to cover with interpolation
• Current forecasts give no indication of
  hour-to-hour variation in pollutant levels, which
  are provided by the NOAA model

13
Fusion of Ozone Forecasts and NOAA Model Ozone
Predictions Method (1 of 2)

• NOAA Ozone Model predictions (left) have much
  better spatial coverage than agency forecasts
  alone (right)
• NOAA model predictions have hourly time
  resolution
• Agency forecasts remain unchanged

San Joaquin Valley
NOAA model data
Agency forecasts available by ZIP code
14
Fusion of Ozone Forecasts and NOAA Model Ozone
Predictions Method (2 of 2)
1. The hour of maximum 8-hr average ozone is
determined for each grid point in the NOAA
model. 2. Where agency forecasts were available,
the model maximum at that hour is adjusted to
match the agencys forecast maximum. 3. All other
hours of NOAA model data are adjusted according
to the ratio of model max to agency max. 4. Where
no agency forecast is available, raw model output
is used. 5. To eliminate sharp discontinuities,
the adjusted hourly NOAA model data are smoothed
using interpolation.
NOAA model data
Agency forecast
15
Fusion of Ozone Forecasts and NOAA Model Ozone
Predictions Results (1 of 2)
Hourly animation of 8-hour average ozone AQI
whose maxima match with agencies forecasts for
that day at every grid point.
16
Fusion of Ozone Forecasts and NOAA Model Ozone
Predictions Results (2 of 2)
San Joaquin Valley (black circles) peak 8-hour
average ozone predictions in the animation match
the agency forecasts.
Observations of maximum 8-hour ozone AQI
Agency forecasts of maximum 8-hour ozone AQI
Model predictions of maximum 8-hour ozone AQI
Animation of fused model and agency forecasts
17
Future Work

• Investigate more advanced fusion techniques in
  ArcGIS
• Multiple pollutant representation
• Dynamic mask of Standard Error automatically
  updates with changing monitor network
• Co-kriging allows interpolation across two data
  sets (primary and secondary)
• Improve model predictions with bias correction
• Bias correction could be applied globally or
  regionally
• Bias correction could be static or updated
  seasonally or daily