AIRNow Mapper: EPA's New Air Quality Mapping System

1
AIRNow Mapper EPA's New Air Quality Mapping
System
Prepared by Patrick Zahn Clinton
MacDonald Sonoma Technology, Inc. Petaluma,
CA Presented to National Air Quality
Conferences Portland, OR April 7, 2008
2
Table of Contents

• Historical Maps
• New Maps
• Development of New Maps
• Summary
• Future Steps

3
Overview of EPAs AIRNow Mapping Program
Agency
AIRNow DMC Quality Control Map Creation
O3, PM, Forecasts
Agency

1. Observations
2. Forecasts

Agency
4
Historical Mapping Methods (1 of 4)

• PM2.5 maps

5
Historical Mapping Methods (2 of 4)

• Ozone maps winter

6
Historical Mapping Methods (3 of 4)

• Ozone maps summer

7
Historical Mapping Methods (4 of 4)

• Issues with current maps
• Bubble maps no information where there are no
  monitors (PM2.5 and wintertime ozone)
• Ozone summer maps
• Inconsistent interpolation techniques for
  different maps
• Accuracy not verified
• Masking technique primitive
• No combined AQI maps

8
New Maps (1 of 2)

• Interpolate between observations for both ozone
  and PM2.5

Old PM2.5 map New PM2.5 map
9
New Maps (2 of 2)

• Use a statistical technique to mask off areas
  with sparse measurements

New PM2.5 map New PM2.5 map with
masking
10
Method for Developing New Maps

• Goal Accurate maps that are easy to read and
  modify
• Accuracy
• Match observations
• Capture the peaks
• Represent AQI in areas void of monitors
• Process
• Identify mapping techniques
• Develop accuracy measures
• Fine-tune each technique
• Compare fine-tuned techniques

11
Interpolation Techniques

• Inverse Distance Weighting (IDW)
• Regularized Spline
• Tension Spline
• Kriging
• Constrained Kriging modified to match (or
  nearly match) observations

12
Evaluating Map Accuracy Point-removal Technique
(1 of 4)

• Prediction error quantifies accuracy of maps in
  areas void of monitors
• Calculate interpolated surface based on
  observations

13
Evaluating Map Accuracy Point-removal Technique
(2 of 4)

• Remove an individual observation from the data
  set

Interpolated surface
Observed AQI
AQI
Distance
14
Evaluating Map Accuracy Point-removal Technique
(3 of 4)

• Re-calculate interpolation with point removed

15
Evaluating Map Accuracy Point-removal Technique
(4 of 4)

• Calculate the difference between the observation
  and the new interpolated surface
• Usually around 5-10 AQI units
• Little variation between interpolation techniques

16
Evaluating Map Accuracy Without Point Removal

• Interpolation Error (IE) Quantifies accuracy of
  maps at monitor locations
• Anywhere from 0-10 AQI units
• Varies drastically between interpolation
  techniques

AQI
Distance
17
Method of Analysis (1 of 2)

• Data selection of days over the past two years
• 30 Cases
• ozone (summer)
• PM2.5 (summer)
• PM2.5 (winter)
• Interpolations using ESRI ArcGIS Geostatistical
  Analyst
• Analysis and production
• Maps look good
• Easy to validate and modify in the future

18
Method of Analysis (2 of 2)

• Fine-tuned each interpolation technique
• Minimizing error values
• Visually appealing maps
• Compared techniques
• Statistical measures
• Bias
• Root-mean-square (RMS) weighted absolute error
• Visual appeal and conceptual consistency

19
Results of Map Analysis Ozone (1 of 4)
June 23, 2006
Maximum 8-hour average ozone
20
Results of Map Analysis Ozone (2 of 4)

• June 23, 2006

21
Results of Map Analysis Ozone (3 of 4)
IDW Kriging Regularized
Spline Observations
Summary for all ozone cases
Method RMS Error Std. Dev.
IDW 1.56 0.28
Kriging 6.56 1.64
Reg. Spline 5.24 2.20
Tension Spline 2.27 0.29
Constr. Kriging 1.77 0.29
Tension Spline Constrained Kriging
Interpolation Error (ppb)
22
Results of Map Analysis Ozone (4 of 4)
Constrained Kriging
Observations
23
Results of Map Analysis PM2.5 Winter (1 of 4)

• December 20, 2007

24
Results of Map Analysis PM2.5 Winter (2 of 4)
IDW Kriging Regularized
Spline Observations
Summary for all PM2.5 winter cases
Method RMS IE Std. Dev.
IDW 0.81 0.35
Kriging 5.18 1.49
Reg. Spline 2.03 2.20
Tension Spline 2.02 1.29
Constr. Kriging 1.42 0.44
Tension Spline Constrained Kriging
Interpolation Error (µg/m3)
25
Results of Map Analysis PM2.5 Winter (3 of 4)
December 20, 2007
26
Results of Map Analysis PM2.5 Winter (4 of 4)
Constrained Kriging
IDW
27
Results of Map Analysis PM2.5 Summer (1 of 4)

• May 31, 2007

28
Results of Map Analysis PM2.5 Summer (2 of 4)

• May 31, 2007

29
Results of Map Analysis PM2.5 Summer (3 of 4)
IDW Kriging Regularized
Spline Observations
Summary for all PM2.5 summer cases
Method RMS IE Std. Dev.
IDW 0.70 0.24
Kriging 2.69 1.33
Reg. Spline 1.13 0.24
Tension Spline 1.16 0.16
Constr. Kriging 0.85 0.22
Tension Spline Constrained Kriging
Interpolation Error (µg/m3)
30
Results of Map Analysis PM2.5 Summer (4 of 4)
Constrained Kriging IDW
31
Masking (1 of 3)

• Q How do we deal with areas that have sparse
  monitoring networks?
• A Historically, we manually blank out entire
  states when monitor data
• are unavailable.
• New method use a statistical measure to
  determine where we
• should not plot AQI using standard error

32
Masking (2 of 3)

• Standard error measure of the uncertainty of
  AQI
• Where observations are sparse, standard error is
  large (uncertainty
• in the interpolated AQI is high)
• Set a threshold for standard error, above which
  we apply a mask to
• prevent AQI from being displayed

33
Masking (3 of 3)

• Advantages of using a standard error mask
• Carried out automatically by the mapping system
• Can be static in time, or updated daily or
  seasonally as monitors come in and out of the
  network

34
Combined AQI Maps PM2.5 and Ozone

• Select maximum AQI for each location
• Produce AQI values for areas that are masked out
  for one parameter, but not for another

Ozone with mask PM2.5
with mask
Combined AQI with mask
35
Summary

• Constrained kriging and IDW had greatest accuracy
  
• Constrained kriging
• Smoother
• Constrained peaks, unlike IDW
• The operational mapping system has great
  flexibility. In real time, it can alter
• Interpolation technique
• Technique parameters
• Masking technique

36
Future Work Data Fusion
Data Fusion using other information to create
AQI maps

• Observations Interpolated
  Observations NOAA 5x Ozone Model Output

BlueSky RAINS Smoke PM2.5 Model Output