Adjusting Numerical Model Data in Real Time: AQMOS Prepared by Clinton P. MacDonald, Dianne S. Miller, Timothy S. Dye, Kenneth J. Craig, Daniel M. Alrick Sonoma Technology, Inc. Petaluma, CA Presented at the 2010 National Air Quality Conferences Raleigh, NC March 15-18,

2 Introduction What: The Air Quality Model Output Statistics (AQMOS) forecasting tool provides air quality model predictions adjusted in real time. Why: Gridded numerical weather and air quality models have some errors in their predictions. How: Like model output statistics for weather models, AQMOS uses regression equations to correct for air quality model errors. AQMOS is Automated Dynamically updated after each model run City-, pollutant-, and model-specific

3 How AQMOS Works AQMOS calculates regression equations from recent air quality observations and numerical model predictions (the critical component). The forecasting tool applies these equations to the current numerical model predictions. AQMOS prediction = numerical model prediction x correlation factor + constant + Many days of observed and predicted data

4 AQMOS Data Sources Air quality observations from AIRNow Gateway –Peak daily 8-hr ozone –Daily 24-hr PM 2.5 Model data –NOAA Air Quality Forecast Guidance (6Z and 12Z) –BlueSky Gateway Experimental CMAQ (0Z) The system’s flexibility allows new models and parameters to be added as data become available.

5 Acquire Data Extract the concentration at the geographic centroid of each ZIP Code and calculate the maximum concentration in the area (in this example, 53 ppb for Des Moines). Hourly NOAA grib data provides peak next-day 8-hr ozone forecast. Store daily predictions for each area and model run in AQMOS database. AQMOS AIRNow Gateway files provide peak 8-hr ozone concentrations for each forecast area. AIRNow Gateway Data File Store daily peak concentration data in AQMOS database.

6 Calculate Regression Match recent forecasts (within the last year) with observations Calculate regression for each city using up to six months of data Observation = recent numerical model prediction x correlation factor + constant Top 15% threshold (88 ppb for Dallas) High Low

7 AQMOS Website (1 of 2) 62.8 ppb = 89.0 ppb x (-2.36 ppb) Original NOAA forecast in AQI NOAA prediction: 89.0 ppb AQMOS prediction: 62.8 ppb Verification: 60.0 ppb AQMOS prediction current model prediction x slope + constant =

8 AQMOS Website (2 of 2)

9 AQMOS Performance (1 of 6) NOAA 6Z Next-Day Ozone April 1-October 31, 2009 >-4 and ≤-2 >-2 and ≤0 >0 and ≤2 >2 and ≤4 >4 and ≤6 >6 and ≤8 >8 and ≤10 >10 and ≤12 >12 and ≤14 >14 and ≤16 >16 and ≤18 Improvement = avg (abs (raw model – observed)) – avg (abs (AQMOS – observed)) AQMOS Improvement (ppb)

10 AQMOS Performance (2 of 6) BlueSky Gateway 0Z Next-Day PM 2.5 April 1-October 31, 2009 >-2 and ≤0 >0 and ≤2 >2 and ≤4 >4 and ≤6 >6 and ≤8 Improvement = avg (abs (raw model – observed)) – avg (abs (AQMOS – observed)) AQMOS Improvement (µg/m 3 )

11 AQMOS Performance (3 of 6) Between April 1-October 31, 2009, AQMOS improved predictions in City AQMOS improvement, 6Z NOAA Ozone model (ppb) Yuba City/Marysville, CA17.3 York/Chester/Lancaster, PA15.0 Charleston, SC13.5 Rock Island-Moline, IL13.4 East San Gabriel, CA Banning, CA-0.8 Imperial Valley, CA-0.9 Provo, UT-1.2 Washakie Reservation, UT-1.7 Ogden, UT-3.1 Improvement = avg (abs (raw model – observed)) – avg (abs (AQMOS – observed)) 312 of 326 forecast cities for the next-day NOAA 6Z ozone model 239 of 252 forecast cities for the next-day BlueSky Gateway PM 2.5 model

12 AQMOS Performance (4 of 6) Days on which either observed or model-predicted air quality was USG or higher GreenGreen – forecast improved on at least 75% of these days YellowYellow – forecast improved on 50-75% of these days RedRed – forecast improved on fewer than 50% of these days City NOAA 6Z Ozone NOAA 12Z Ozone BlueSky 0Z PM 2.5 Atlanta92%90%33% Houston50% 67% Philadelphia82%67%100% Sacramento76%80%100% St. Louis63%75%100% Salt Lake City60%38%----

13 AQMOS Performance (5 of 6) Atlanta Next-Day 6Z NOAA Ozone Model April 1-October 31, 2009

14 AQMOS Performance (6 of 6) Salt Lake City Next-Day 6Z NOAA Ozone Model April 1-October 31, 2009

15 Summary AQMOS is available to air quality agencies. AQMOS should be used in conjunction with other guidance. On most days, AQMOS shows improvement over the raw model output in predicting air quality for specific cities. AQMOS showed improvement on at least 50% of critical days in the six cities evaluated. For more information, please visit or contact Dianne Miller