Recent Developments in Air Quality Forecasting in Chile. Marcelo Mena-Carrasco 1, Pablo Saide 2, Gregory Carmichael 2 Scott Spak 3,Luisa Molina 4, Center for Sustainability Research, Universidad Andrés Bello. Center for Global and Regional Environmental Research, University of Iowa. Public Policy Center, The University of Iowa, Molina Center for Energy and Environment. mmena@unab.cl Fondecyt project 11090084: EVALUATING REGIONAL INFLUENCE OF MEGACITY EMISSIONS ON AIR QUALITY, METEOROLOGY AND CLIMATE.
Santiago, Chile Population of ~6 million people, 1.3 million cars. Surrounded by two mountain ranges, 2000 and 4000m. Exceeds PM 10, PM 2.5, NO 2, CO, and O 3 standards
Models used to determine day to day pollution control measures Level ICAP number PM10 (µg/m 3 ), 24h mean Restrictions Good 0-100 0-150 40% non catalytic cars Regular 100-200 150-195 40% non catalytic cars Alert 200-300 195-240 Chimneys are banned Preemergency (Critical) Preemergency (Dangerous) 300-400 240-285 60% non catalytic cars, 20% of catalytic cars, 100% of domestic chimneys, and the 798 largest point sources of particulate matter 400-500 285-330 Emergency 500-330- 80% of non catalytic cars, 40% catalytic cars, bans on domestic chimneys, and 2603 point sources Wood burning 14% Other 21% Industry 30% Trucks 16% Light and medium cars 14% Buses, 5% Contribution of economic sectors to ambient concentrations of PM2.5 in Santiago, 2005 (based on DICTUC, 2007 and USACH, 2005). Results: cars are 80% catalytic. Industry works at 30mg/m3 of PM emission standard, lower wood burning use (8% of homes)
Recent trends in annual PM2.5 concentrations 100 90 80 70 69 61 60 55 56 52 50 40 30 47 42 43 39 38 36 35 35 34 34 30 29 Annual Standard for PM2.5 = 20 ug/m3 31 33 31 28 27 20 10 0 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
Car use bans are highly controversial (probably too blunt of an instrument that does not discriminate based on emissions, fuels) General situation Official model under scrutiny (linear regression model). Less than 50% correct forecast New PM2.5 standard (2012) requires new forecast model. Need to correctly communicate that new standard leads to more frequent AQ episodes.
WMO and UIowa led technology transfer University of Iowa, UNAB, and Chilean Meteorological Office sign MOU to transfer model to the Met Office. Master s students training for implementation(rodrigo Delgado, Pablo Hernandez) Run on Chilean Met Office s 100+ CPU HPC system. Web development by UNAB grad students.
WRF-Iowa model (Saide et al., 2011) PM obs vs CO obs, 1hr data Builds on strong correlation between PM10 vs CO and PM2.5 vs CO concentrations during episodes. Then, by forecasting CO properly, a PM10 and PM2.5 forecast can be made by multiplying by a factor per station. 39 horizontal levels. Nested domain 36 to 2km (97x97),MYNN3 PBL, WSM3 microphysics, RRTM LW radiation, Dudhia SW radiation PM obs vs CO obs, 24hr mean 1 day hindcast (FNL), 4 day forecast (GFS).
Process flow for forecasting system
Each day is forecasted 5 times.
Also are capable of full chemistry runs (Saide et al., 2012) Figure 10: Horizontal plots of cloud effective radius (µm, a and b) and first level, second bin SO 4 concentration (µgr/m3, c).(a) shows MODIS-AQUA cloud effective radius for October 16 th 17UTC overpass while (b) and (c) shows NW model results for the same time. Model cloud effective radius is computed for the cloud top.
Residential wood burning (Mena-Carrasco et al.,2012) Bad air days per year, vs month. Heating degree days correlated to bad air quality. Emissions distributed based on monthly HDD. Monthly PM2.5 vs HDD Monthly PM2.5 emissions basedon monthy HDD
Applied science informing policy: Communicating that episodes are due to accumulation of multiple days of pollution.
(Saide et al., 2011) 0% contribution from same day emissions
Air quality forecasting system Authorities realized that alert days (wood burning bands) were almost as effective as pre-emergency days in reducing pollution, but politically easier. Started preventive focus, alerts were decreed two or three days before pre-emergency was predicted.
Time Lapse Video
Strong media outreach is important for community support Featured in El Mercurio, June 5 th, 2011 Sunday edition.
Government attributes air pollution episode reduction to preventive approach (made possible by WRF-Iowa model)
Correlation coefficients from D1 0.59, D2 0.49, D3, 0,61. Fractionalized bias: D1 0.04, D2 0.09, D3 0.09 Model Performance 1 day forecast Good Regular Alert Pre-emergency Emergency Total Correct Good 36 10 2 Regular 3 18 6 3 Alert 8 9 Pre-emergency 1 1 1 Emergency Total forecasted values 40 37 17 4 0 % correct 90% 49% 53% 25% 0% 65% 2 day forecast Good Regular Alert Pre-emergency Emergency Good 35 9 1 Regular 4 17 8 3 Alert 9 8 Pre-emergency 1 2 1 Emergency Total forecasted values 40 37 17 4 062% % correct 88% 46% 47% 25% 0% 3 day forecast Good Regular Alert Pre-emergency Emergency Good 32 11 Regular 5 17 9 2 Alert 2 9 8 1 Pre-emergency 1 1 Emergency Total forecasted values 40 37 17 4 059% % correct 80% 46% 47% 25% 0%
Conclusion. Today we have a PM2.5 forecast model that performs better than the official model, but with much more information. Tool allows preventive approach, forecasts days in advance. Did we prevent air pollution episodes? Future work will tell but we certainly reduced pre-emergency days. Need to work on better pollution abatement tools Obviously model performance cannot be cleanly evaluated, since emissions have been modified (future work). Should model objective be to hit thresholds? Or should it be used to prevent exposure? Next steps: deploy model for rest of country.
RESEARCH LINES SURFACE WATER PROCESSES AND ASSOCIATED HAZARDS Tsunamis, Floods & Landslides Frequency analysis Land use, DEM-DSM GIS analysis Numerical Modeling Tsunami modeling Wave-structure interaction Spatial and Temporal Characterization Anthropogenic factors Real Time Monitoring In situ & Remote sensing - WSN Stochastic/Det erministic input scenarios Forecasting Data assimilation Inverse modeling Built Environment Propagation Natural Environme nt Chemical weather forecasting P r o b a b i l i s t i c R i s k A n a l y s i s Hazard field Coupled hydrometeorological and hydraulic modeling Forecast (Dynamic Hazard characterization) Stochastic and Deterministic Hazard Scenarios (demand parameters) Researchers Dr. R. Cienfuegos (Nonlinear Wave Propagation) Dr. P. Catalán (Coastal Engineering) Dr. M. Mena (Chemical Weather Forecasting) Dr. J. Gironás (Hydrology) Dr. C. Escauriaza (River Hydraulics) Dr. C. Ledezma (Geotechnical Engineering)
Acknowledgments Rodrigo Delgado, Ricardo Alcafuz (DMC) Scott Spak, Pablo Saide (UIowa) Estefanía Oliva, Pablo Hernandez, Pablo Moreno (UNAB) Luisa Molina (MCE2) Liisa Jalkanen (WMO) Elliott Campbell, Chi-Chung Tsao (UC Merced).