Current and Future Impacts of Wildfires on PM 2.5, Health, and Policy in the Rocky Mountains Yang Liu, Ph.D. STAR Grants Kick-off Meeting Research Triangle Park, NC April 5, 2017
Motivation The Rocky Mountains Region has seen increased frequency of large wildfires, a longer fire season and dramatic increases in average fire size in recent years. Climate models predict the most increased fire activity to occur in this region Changes in wildfire emissions and their interaction with meteorological drivers of PM pollution is an understudied area Epidemiologic information on the health effects of wildfire smoke is still limited in the literature
Study Objectives Goal: investigate the climate effects on PM 2.5 and O 3, with an emphasis on wildfires in the Rocky Mountains Region Aims 1. Estimate daily PM 2.5 and O 3 levels in Colorado in 2011-2014 by fusing ground observations, satellite data and CMAQ simulations 2. Derive fire-season AQ C-R functions for ED visits and acute hospitalizations
Study Objectives 3. Conduct CESM/WRF/CMAQ simulations for 1995 2004; 2030 2039 (funding permits) and 2050-2059 to evaluate how climate affects the spatiotemporal patterns of PM 2.5 and O 3 4. Using results from Aim 2 and 3, estimate the impacts of future wildfires on air quality, population health, and public health decision making in the Rocky Mountains Region.
Project Team Emory Yang Liu (PI): satellite data applications, exposure assessment Howard Chang: biostatistician, RCM bias correction U. Nevada at Reno Matt Strickland: epidemiologist U. Tennessee at Knoxville Joshua Fu: CESM/WRF/CMAQ modeling, dynamical downscaling US Forest Service Yongqiang Liu: fire potential calculation, climate impact on wildfires
Yr 1 Tasks Aims 1 & 2 Evaluation of the impact of different emissions inventories on CTM predictions in Colorado Preliminary CMAQ runs for 2011 2014 Preliminary exposure model development Preliminary epidemiologic study
Evaluation of Emissions Inventories Three sets of WRF-Chem runs in June 2012 from NCAR FINN v1.5 12 km vs. QFED v2.4 12 km FINN v1.5 12 km vs. FINN v1.5 4 km FINN uses MODIS 1 km active fire spots under relatively cloud-free conditions while QFED uses MODIS fire radiative power (FRP) FINN and QFED also use different land cover type data
Preliminary Results Simulated PM 2.5 levels vs. EPA observations r between QFED results and obs = 0.55 r between FINN 12 km results and obs = 0.27 Odds ratios from conditional logistic regression on respiratory endpoints (i.e., asthma and wheeze) FINN 12 km and QFED 12 km results generated statistically significant and similar results in different age groups FINN 4 km results generated noisier effect estimates and null results for ages 65+ group
Preliminary Findings QFED appears to be the more appropriate fire emissions inventory in our study domain Higher spatial resolution does not necessarily yield more accurate results.
Yr 1 Tasks CMAQ Modeling Colorado State (CO) 2011 2014 Colorado (CO) simulations from UTK Horizontal resolution: 12 km Vertical resolution: 25 layers 70 (x) 55 (y) 25 (z) grid cells 2011 NEI Initial/boundary conditions (IC/BC) from GEOS-Chem outputs 2011 2014 using the same 2011 NEI and GEOS-Chem IC/BC Meteorological inputs (MCIP) 2011: from US EPA directly 2012 2014: WRF 3.8.1
Simulation Period Determination 1980 2013 CONUS monthly distribution of major fires (>50K acres) Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec 1 1 8 10 19 57 103 109 14 19 1 0 Suggested simulation period 1980 2013 excluding CA, OR, and WA monthly distribution of major fires Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec 1 1 8 10 19 48 70 70 9 1 1 0 Suggested simulation period Fire months: April October Data from Dr. Yongqiang Liu, USFS
2011 2014 WRF Evaluations 2011 2014 WRF v3.8.1 Criteria: mean bias (MB) between ±0.5 (Zhang et al., 2014) Procedures: daily MB for each station station-averaged daily MB stationaveraged monthly MB 2011 Monthly MB Month TEMPERATURE ±0.5 K WIND SPEED ±0.5 m/s 4-0.07 0.23 5-0.06 0.11 6 0.05 0.31 7 0.25 0.16 8 0.19 0.28 9 0.38 0.45 10 0.33 0.43 2012 Monthly MB Month TEMPERATURE ±0.5 K WIND SPEED ±0.5 m/s 4-0.11 0.08 5-0.01 0.26 6-0.27 0.20 7-0.13-0.01 8 0.24 0.14 9 0.19 0.18 10 0.49 0.25 2013 Monthly MB Month TEMPERATURE ±0.5 K WIND SPEED ±0.5 m/s 4 0.23 0.04 5 0.21 0.03 6-0.10 0.10 7 0.05-0.10 8 0.07-0.01 9 0.26 0.19 10 0.79 0.47 2014 Monthly MB Month TEMPERATURE ±0.5 K WIND SPEED ±0.5 m/s 4 0.22 0.07 5 0.25 0.17 6-0.08 0.11 7-0.08-0.002 8 0.15 0.27 9-0.07 0.37 10 0.02 0.42
2011 CMAQ Preliminary Runs PM 2.5 Appel et al. (2016) 2011 Summer 2011 Spring 2011 Fall Seasonal MB (μg/m 3 ) CMAQ AQS Spring: April, May Summer: June August Fall: September, October We evaluated AQS daily data for model performance check Results are comparable to Appel et al. (2016)
2012 CMAQ Preliminary Runs PM 2.5 2012 Spring 2012 Summer 2012 Fall Seasonal mean bias (μg/m 3 ): CMAQ AQS We evaluated AQS daily data for model performance check Summer results are significantly low Spring: April, May Summer: June August Fall: September, October
2011 CMAQ Preliminary Runs O 3 Appel et al. (2016) 2011 Summer 2011 Spring 2011 Fall Seasonal MB (ppbv) CMAQ AQS Spring: April, May Summer: June August Fall: September, October We evaluated AQS daily data for model performance check Results are comparable to Appel et al. (2016)
2012 CMAQ Preliminary Runs O3 2012 Spring 2012 Summer 2012 Fall Seasonal mean bias (ppbv): CMAQ AQS We evaluated AQS daily data for model performance check Model performance needs improvement Spring: April, May Summer: June August Fall: September, October
Next Steps Preliminary Bayesian fusion modeling of 2011-2014 fire season PM 2.5 and O 3 SSSSSSSSSS 1: YY ss, tt = αα 0 ss, tt + αα 1 ss, tt XX ss, tt + ββzz ss, tt + εε(ss, tt) SSSSSSSSSS 2: YY ss, tt = kk ww(ss, tt) ff kk [YY ss, tt ] Satellite data input: MAIAC 1 km AOD Expected outcome: 4 km resolution daily PM 2.5 and O 3 exposure estimates with complete coverage in space and time
Next Steps Refine exposure model and finalize epidemiologic study to generate C-R functions Start CESM/WRF/CMAQ climate model runs Develop preliminary bias correction schemes for the downscaled climate model output