Forecasting Panel Discussion State/Local/EPA Meeting: Year-Round/Mult-Pollutant Air Quality Index (AQI) April 24-25, 2003 USEPA Offices RTP, NC Cary Gentry, Forsyth County EAD Dan Salkovitz (Virginia DEQ) Tim Dye (Sonoma Technology, Inc.) Pat Dolwick, USEPA/OAQPS
Goals: Forecasting Panel Session: Goal / Agenda Identify/Discuss 8 Issues related to year-round AQ Based on the 2 preliminary calls, attempt to bound the issues Encourage active audience participation panel will tee up the issues; summarize call discussions, open for debate Attempt to resolve any outstanding issues w/ regard to forecasting if there are roadblocks to 10/03 target start date, how can they be addressed? Agenda: 0) Opening (5 min) 4) Forecast Accuracy (10) 8) Good Web Sites (5) 1) Good/Moderate (15) 5) Resource needs (5) 9) Next Steps (10) 2) Resource Burden (20) 6) EPA Training (5) 3) Smart enough re: PM2.5 (10) 7) Roles: S/L/EPA (5)
Pros: Q1. Why forecast if all my non-photochemical days are GOOD/MODERATE? Do not want to catch the public/legislators off guard Consistent AQ product Easier access to the media. No ramp up required each year. Prepared for special events (fires, dust, etc). Forecast of GOOD not necessarily bad PR Health message associated with MODERATE levels of PM2.5 Prolonged exposure to MODERATE levels of PM2.5 can be harmful longer-term Default forecast of GOOD & only forecast poorer AQ events (resource savings) Cons: Resource costs (staff/legislative) Special events (fires, dust, etc) are the main cause of poorer days Confusion on the part of the public Outreach program will also be needed
Q2. How can year-round forecasts occur without heavy resource constraints? Considerations: Staff burden will vary as a function of the region being forecast and outreach goals. Resource needs can be minimized if regional forecasts are emphasized as opposed to local forecasts or monitor-specific forecasts. Data completeness, real-time data quality, and spatial representativeness issues need to be resolved. how many continuous (TEOM) and Federal Register Method monitors are in your area? PM2.5 forecasting more difficult than ozone forecasting? Jury is out on this. May vary by area/location. Forecasters need to understand PM2.5 problem in any given area. Knowledge of local PM2.5 climatology can lessen forecast time. Experience and intuition are also key. Forecast tools can simplify the burden. such as regression models and Classification and Regression Tree (CART) analyses
Q2. How can year-round forecasts occur without heavy resource constraints? Summary of PM2.5 Forecasting Requirements: Asked 8 institutions that have experience in forecasting PM2.5 to discuss necessary resources based on their experience. (Forsyth Co., Illinois EPA, Indianapolis OES, Puget Sound Clean Air Agency, San Joaquin Valley AQMD, South Coast AQMD, Sonoma Technology, Texas DEQ) Forecast areas: range from single multi-county MSA to as many as 8 MSAs. Frequency of forecasts: generally daily, although some groups cover weekends on Friday or for USG cases forecasts range from 24 to 72 hours generally issued in early afternoon Number of forecast staff range from 1 full time (w/ 1-2 backups) to a staff of 6.
Q2. How can year-round forecasts occur without heavy resource constraints? Summary of PM2.5 Forecasting Requirements (continued): Type of forecast staff: Range from all meteorologists to technical staff (w/ met. training and/or experience). Staff time for forecasting: various by region, function of: 1) area size, 2) severity of AQ problem, 3) outreach duty in some cases, it is hard to separate the PM forecasting time from the O3 time (CA) ranges from as low as 5-10 hours per week to as high as 20-25 hours per week 5-10: Indy, Seattle/Tacoma, STI experience, Texas 10-25: Forsyth Co., San Joaquin Valley, South Coast, Illinois EPA Necessary equipment some groups use little beyond continuous monitors & Internet others have met. towers, visibility cameras, classification schemes, simplified decision aid tools, radar profilers, real-time speciation monitoring, GOES satellite processing tools, etc.
Q2. How can year-round forecasts occur without heavy resource constraints? Summary of PM2.5 Forecasting Requirements (continued): Training Recommendations: Some found that shifting from O3/CO/etc. to PM2.5 required little additional training Others found that several months ramp up was necessary to build tools, gain experience Several attended EPA or other training sessions. Many mentioned that there was no training equal to actually forecasting Estimated Annual Cost/Benefit of Program: nearly impossible question to answer (too many variables, e.g., what counts toward $?) may want to survey air directors as opposed to forecasters ranged from as low as ~7K per year (Indy) to 50-100K or 1 FTE (TX, IL)
Q3. Are we smart enough about PM2.5 to do forecasting? Yes: Slow, but steady improvement in PM2.5 forecasting accuracy over time Good accuracy on initial forecasts after a couple of months after analyzing the real-time PM2.5 data for trends; relationship to meteorology Maybe Not: Many forecasters need to build a comfort level. Busted forecasts requires considerable damage control No range to submit for the AQI enabling ranges could aid accuracy (e.g., high good / low moderate) Is the public smart enough?
Q4. How reliable have the current PM2.5 forecasts been? Related questions: How do we define reliability and accuracy? What is forecast performance now? What should forecast performance be? How do we get there?
Q4(a). How do we define reliability and accuracy? Forecast performance Part reliability timely, always available Part accuracy closeness to target Many measures of accuracy Need measures that focus on high days Probability of detection (POD) Correctly predicted high day False alarm rate (FAR) Missed forecasted high day Focus on the high days (rarer events) Needs to do better than persistence
Ozone Q4(b). What is forecast performance now? Measure next-day forecasts Performance for Unhealthy for SG and above Performance of Next-day Ozone Forecasts Agency POD FAR Established 75% 25% New agencies 25-50% >75% * Based on 2002 results Ozone forecasts better than persistence
Q4(b). What is forecast performance now? PM 2.5 45 Minneapolis Baltimore 40 35 Obs 30 Next Day Forecast 25 20 15 10 5 0 04-Jan 04-Jan 11-Jan 11-Jan 18-Jan 18-Jan 25-Jan 25-Jan 01-Feb 01-Feb 08-Feb 08-Feb 15-Feb 15-Feb 22-Feb 22-Feb 01-Mar 01-Mar 08-Mar 08-Mar 15-Mar 15-Mar 22-Mar 22-Mar 29-Mar 29-Mar 05-Apr 05-Apr 12-Apr 12-Apr 45 40 35 30 25 Performance of Next-day PM 2.5 Forecasts Agency POD FAR Baltimore 50% 52% Minneapolis 59% >46% 24-hr Averaged PM 2.5 Concentration (ug/m3) 20 15 10 5 0 OBS Next Day Forecasts 24-hr Averaged PM2.5 Concentration (ug/m3)
Q4(c). What should forecast performance be? Least Best PM 2.5 Forecasts Persistence Ozone Forecasts Weather Forecasts Goal for PM 2.5 forecasts performance: As accurate as ozone forecasts issued by experienced forecasters
Q4(d). How do we get there? Takes (1) Time-experience and (2) Understanding Age of Forecasting Program (years) 1 2 3 4 5 6 7 8 9 10+ Data Experience and Understanding Techniques Performance Increasing number of site and historical data record Increasing knowledge of what causes episodes More techniques with greater complexity Greater accuracy and precision
Q4(d). How do we get there? Understanding Conceptual models of PM episodes (different than ozone) Tools to predict Good and Moderate days Improved data sources Continuous PM monitors Real-time speciated PM monitors Satellite measurement Surface visibility measurements What else?
Q5. What resources are needed to put a good PM2.5 forecasting program in place? Guidelines for Developing an Air Quality Forecasting Program EPA guidance on developing AQ forecasting programs contains a list of recommended resources should be ready by early summer Short Course on Air Quality Forecasting in San Antonio on February 2, 2003 contained a discussion (Joe Cassmassi - South Coast) on resource needs available from http://www.epa.gov/airnow/2003conference/aq.html Many S/L agencies are already forecasting AQ year-round and are willing to help..
Q6. What are EPA s Plans for Forecaster Training? 4 Regional Workshops (focusing on individual regions/areas) geographically dispersed across the U.S. (west, central, south, northeast?) approximately 2.5 days in length currently targeting dates in Sept. (avoid peak ozone season, prior to 10/01 Draft Agenda background on PM2.5 (e.g., monitoring, processes that influence, etc.) development of local/regional conceptual models / climatologies development of local/regional statistical tools case studies PM forecasting software and much, much more Looking for R/S/L feedback on most suitable format, locations, dates, agenda items, etc.
Q7. What are the roles and responsibilities of EPA and State/Local agencies in the near-term in forecasting PM2.5? State/Local versus EPA roles will be similar to that for ozone: State/Locals will: have primary responsibility of generating and disseminating forecasts initiate any local action day programs EPA will: maintain the AIRNOW web site to promote national consistency encourage State/Local participation maintain the real-time data work with national media partners where appropriate, develop needed tools and utilities.
Q8. Who operates good PM2.5 forecast web sites? http://www.epa.gov/airnow (US EPA) http://www.valleyair.org (San Joaquin Valley California) http://www.sparetheair.org/air_quality.asp (Bay Area California) http://www.co.forsyth.nc.us/envaffairs/dlyairqualrpt.htm (Triad region of NC) http://www.pscleanair.org/airq/aqi.shtml (Washington State) http://www.deq.state.ut.us/eqamc/api.htm (Utah) http://www.tnrcc.state.tx.us/cgi-bin/monops/psi_rpt (Texas) http://www.co.forsyth.nc.us/envaffairs/pm25forecasting.htm (PM2.5 FAQ) http://www.airnowdata.org/pmfine/latest.html (Sample Real-Time PM2.5 maps)
Any other questions? Bring em on.