Georgia Institute of Technology Evaluation of the 2006 Air Quality Forecasting Operation in Georgia Talat Odman, Yongtao Hu, Ted Russell School of Civil & Environmental Engineering Michael Chang, Carlos Cardelino School of Earth and Atmospheric Sciences Georgia Institute of Technology 5 th Annual CMAS Conference October 17, 2006

Georgia Institute of Technology Air Quality Forecasting There is an increasing interest in day-to-day variation of air quality –Public becoming more health conscious –Local authorities looking for short-term management strategies Forecasts are produced using various techniques –Persistence –Climatology –Statistical Regression –Close Neighbor –Decision Tree –3-D Air Quality Models

Georgia Institute of Technology Air Quality Forecasting in Atlanta Ozone forecasting since 1996 Olympic Games Panel of experts gets together and issues a forecast for next day –Ozone Alerts One of the methods used is 3-D AQM –Urban Airshed Model (UAM) –Diagnostic Meteorology –Constant Emissions –Arguably first in the U.S. but now mostly outdated Last year, PM 2.5 forecasting started Forecasts being extended to other cities in Georgia –Macon (~135 km South-Southeast of Atlanta)

Georgia Institute of Technology Goal of our Operation To provide accurate, “fine-scale”, local forecasts sufficiently in advance that –Local controls can be triggered to avoid bad episodes –Personal exposure can be minimized NOAA/EPA’s national forecast currently provides 12-km resolution over the Southeast We are currently at 4-km resolution aiming at 1-km or “finer” resolution locally. Our goal is to forecast not just air quality but effectiveness of predetermined local control strategies

Georgia Institute of Technology WRF for meteorology –Driven by NAM (formerly Eta) –3 ½ -day NAM forecasts available every 6 hours (00, 06, 12, 18Z) SMOKE for emissions –Forecasting of EGU, mobile, biogenic emissions CMAQ for chemistry and transport –Currently using standard version 4.5 –Will start using our contributions soon: Direct Decoupled Method (DDM-3D) for emission sensitivities Variable Time Step Algorithm (VARTSTEP) for speed Dynamic Solution Adaptive Grid Algorithm (DSAGA) for high resolution Our Modeling System

Georgia Institute of Technology Modeling Domain and Grids Three grids: –36-km (72x72) –12-km (72x72) –4-km (99x78) Horizontal domains are slightly larger for WRF 34 vertical layers used in WRF 13 layers in CMAQ

Georgia Institute of Technology Our 2006 Operation Started May 1, 2006 (through September 30 th ) Tomorrow’s forecast is due by 10 a.m. today –Processing takes 1 ½ days –Wednesday’s forecasting starts by Sunday night We simulate: –3 days over the 36-km grid using 00Z NAM, IC from previous cycle (warm start) and “clean” BC –2 ½ days over the 12-km grid using 12Z NAM and IC/BC from 36-km –24 hours over the 4-km using 12Z NAM and IC/BC from 12-km Mostly automated –1 hr/day of human interaction –6 CPUs The product is a 24-hr ozone and PM 2.5 forecast once per day

Georgia Institute of Technology Evaluation 11 stations measuring O 3 every hour 6 stations measuring PM 2.5 mass every hour

Georgia Institute of Technology O 3 in Metro Atlanta: Summer of 2006

Georgia Institute of Technology Performance Metrics False AlarmsHits Correct Nonevents Missed Exceedences Forecast Observations

Georgia Institute of Technology O 3 Performance: 4-km vs. EPD’s Our 4-km ForecastEPD Ensemble Forecast MNB11% MNE29% MNB6.2% MNE15%

Georgia Institute of Technology O 3 Performance: 4-km vs. 12-km Our 4-km ForecastOur 12-km Forecast MNB10.9% MNE28.6% MNB11.1% MNE28.0%

Georgia Institute of Technology O 3 Bias & Error by Site MNB15% MNE31%

Georgia Institute of Technology Forecasted vs. Observed O 3

Georgia Institute of Technology O 3 Performance until July 20 MNB-0.4% MNE23% MNB3.3% MNE14% Our 4-km ForecastEPD Ensemble Forecast

Georgia Institute of Technology O 3 Performance: Parts I and II MNB-0.4% MNE23% MNB32% MNE40%

Georgia Institute of Technology O 3 at Gwinnett on July 5, 2006 Observed 8-hr: 91 ppb Forecasted 8-hr : 89 ppb

Georgia Institute of Technology PM 2.5 in Metro Atlanta: Summer of 2006

Georgia Institute of Technology PM 2.5 Bias & Error by Site MNB-31% MNE38%

Georgia Institute of Technology Forecasted vs. Observed PM 2.5

Georgia Institute of Technology PM 2.5 Performance: May-Aug. and Sept. MNB-38% MNE41% MNB-4% MNE26%

Georgia Institute of Technology PM 2.5 at South Dekalb on Sep. 11, 2006 Obs. 24-hr: 32.6  g/m 3 4-km 24-hr : 28.7  g/m 3

Georgia Institute of Technology Conclusion We started a “fine-scale” forecasting operation in GA using 3-D models Spatial variability of O 3 and PM 2.5 indicates the need for finer scales The 4-km forecast is slightly more accurate than the 12-km forecast Predictions at some sites are better than others, especially for PM 2.5 Ozone forecasts were generally accurate until mid-July (no bias, 20% error) but overestimations dominated afterwards (30% bias, 40% error) –Diurnal changes are somewhat captured; daily peaks are generally underestimated PM 2.5 was generally underestimated May-August (40% bias, 40% error) but more accurate in September (-5% bias, 25% error) –Afternoon peaks were generally missed Spatial variability was underestimated both for O 3 and PM 2.5

Georgia Institute of Technology Next Steps Continue the operation in 2007 –Improve accuracy –Extend the domain of coverage –Increase the resolution –Elongate the forecasting period –Issue daily updates Link the forecast to health-effects studies: –Study the impacts on asthmatic children –Build a data archive for long-term exposure studies Forecast the effectiveness of short-term, local control strategies –Predict the impacts of predetermined strategies

Georgia Institute of Technology Acknowledgement This research is supported by: Georgia Department of Natural Resources Air Resources Engineering Center at Georgia Tech