Recent and future changes in high surface ozone events over the Eastern United States Harald E. Rieder 1,2* and Arlene M. Fiore 1,3 1 Lamont-Doherty Earth Observatory, Columbia University, NY, USA 2 Applied Physics and Applied Mathematics, Columbia University, NY, USA 3 Earth and Environmental Sciences, Columbia University, NY, USA Atmospheric Chemistry Meeting 16 JULY 2013 Lamont-Doherty Earth Observatory Columbia University, New York, USA
Changes in high ozone events over the observational time period 1988-2009
MOTIVATION 1) National Ambient Air Quality Standards have tightened over recent decades 2) Improved air quality policy provides ancillary benefits for human health and reduces health system costs [e.g., Bell et al., 2008] 3) changes in climate, as well as regional and global emissions are expected to influence the frequency, duration, and intensity of air pollution events above the NAAQS thresholds [e.g. Turner et al., 2013] 4) observations demonstrate that high air pollution (O3 and PM2.5) is typically associated with stagnation events (heat waves) (e.g., Logan, 1989, Vieno et al., 2010) 5) inter-annual variability of extreme air pollution events may increase, along with their persistence [e.g. Meleux et al., 2007] 6) rich observational data base (e.g., CASTNET, AQS, EMEP) 7) recent development in EVT models 8) continuous improvement in atmospheric chemistry models
CASTNET SITES & SITE SELECTION - Clean Air Status and Trends Network (CASTNET) data for the Eastern US - in operation since 1987-31 operational sites SELECTION CRITERION: at least 20 years of data in the 1988-2009 time period 23 out of 31 stations we analyze three regions of interest following Rasmussen et al. (2011) Data available at: http://epa.gov/castnet/javaweb/index.html
EVT MODELLING We use a Peak-Over Threshold (POT) approach to model days above the NAAQS MDA8 O 3 standard of 75 ppb using the Generalized Pareto Distribution (GPD) F( x) ( x u) 1 1 1/ u the threshold value (75 ppb) σ > 0 is the scale parameter ξ ϵ R is the shape parameter Sample site: Penn-State (PSU106) 1988-1998 1999-2009 Gaussian vs. Observations GPD vs. Observations [Rieder et al. 2013]
EVT MODELLING From the fitted GPD we can derive the T-year return level x T, defined through the relation F(x T )=1-1/T, which describes the probability to observe a value x within a time window T. To analyze if changes in the NAAQS MDA8 O 3 threshold had a significant influence on NE US air quality we calculate the return levels of x T for two 11-year time periods 1988-1998 and 1999-2009 at each station. Return Level within a given time period (T) [Rieder et al. 2013]
MDA8 O 3 RETURN LEVELS ABOVE THE NAAQS 1-year return levels 1988-1998 1-year return levels 1999-2009 -1988-1998 to 1999-2009: 1-yr return level decreases by 2-16 ppb avg. 8 ppb -Large parts of the NE US show 1-yr return levels below 100 ppb in 1999-2009
Comparing CASTNET and CM3 for the historical time period 1988-2005
CASTNET vs. CM3 - Offset between CM3 PDF and CASTNET PDF HOWEVER: CM3 shows very similar response to NOx-SIPCALL as CASTNET SITES i.e, larger reductions for higher quantiles IDEA: Use quantile-mapping to correct CM3 PDF for MDA8-O3
Model Bias Correction through Quantile Mapping GOAL: establish a statistical relationship or transfer function between model output and observations based on historical data Quantile-based mapping approach maps the distribution of model variables on the observed data (first introduced by Panofsky and Brier, 1968) P O = h(p M )
Quantile Mapping Results
Quantile Mapping Results - PDF of QM-corrected CM3 matches CASTNET PDF - Mean-Bias correction improves agreement with the observations though various biases remain QM-correction superior to mean-bias correction ALSO NOTE: QM-correction preserves local differences (in time) on grid-cell basis
Quantile Mapping Results Mean MDA8O3 ORIGINAL-CM3 QMcorrected-CM3 OCM3: overestimates observed Mean MDA8-O3 by 5-20ppb on grid cell basis QMCM3: matches observed Mean MDA8-O3 very well
Quantile Mapping Results - # days above the NAAQS OCM3: largely overestimates #days above the NAAQS (~ factor of 2) QMCM3: matches #days above the NAAQS and provides realistic spatial representation ORIGINAL-CM3 QMcorrected-CM3
Changes in Eastern US high ozone events under different representative concentration pathways
GFDL-CM3 Model & RCPs GFDL-CM3 with coupled chemistry for IPCC AR5 RCP-scenarios [2.5 x 2 grid] RCP4.5: increasing CO2 & decreasing methane (global T ~2 C by 2100 - flattens) RCP4.5X: as RCP4.5 but with constant NOx emissions (year 2000 level) RCP8.5: increasing CO2 & increasing methane (global T ~4 C by 2100 - rises) RCP8.5 RCP4.5 Source: Van Vuuren et al., 2011 [modified]
climate change + decreasing O3 precursors TIME - Strong decline in MDA8-O3 with thime - Stronger decline in high quantiles than low quantiles - QMcorrected-CM3 preserves rate of decline over quantile range AND corrects overall pdf for high biases in CM3
climate change + decreasing O3 precursors MEAN change 2046-55 vs 2006-15 MEAN change 2091-100 vs 2006-15 Q90 Q10 change change 2046-55 2046-55 vs vs 2006-15 2006-15 Q90 change 2091-100 vs 2006-15
climate change only scenario - Very stable PDF over the 21 st century (constant emissions!) - Though small increase in MDA8-O3 over quantile range (~1-2ppb) climate penalty - Not much difference between the low and high tails and bulk of the PDF MEAN change 2046-55 vs 2006-15 MEAN change 2091-100 vs 2006-15
climate change only scenario - HOWEVER spatial differences in MDA8-O3 - increases in MDA8-O3 along the coast and towards higher latitudes - Stronger climate penalty as time evolves MEAN change 2046-55 vs 2006-15 MEAN change 2091-100 vs 2006-15
Changes in the number of days above the NAAQS RCP4.5 2006-15 change 2046-55 vs 2006-15 RCP4.5* 0 50 change 2091-100 vs 2006-15 2006-15 change 2046-55 vs 2006-15 change 2091-100 vs 2006-15 RCP4.5: strong decline in #days above the NAAQS over 21 st century no exeedance by mid century RCP4.5*: larger #days above NAAQS at beginning of 21 st century than RCP4.5 increase in the number of non-compliance days (up to +10) along the coast (region with current highest non compliance rate)
Changes in 1-year return level estimates RCP4.5 100 100 2006-15 50 2046-55 2091-2100 50 RCP4.5: strong decline in the 1 year return level (RL) 1-yr RL well below 75pbb for the entire eastern US by mid 21 st century by 2100 1-yr RL at most by 60 ppb (!sic NAAQS results)
Changes in 1-year return level estimates RCP4.5 100 100 2006-15 50 2046-55 2091-2100 50 RCP4.5* 110 110 2006-15 50 2046-55 2091-2100 50 RCP4.5: strong decline in the 1 year return level (RL) 1-yr RL well below 75pbb for the entire eastern US by mid 21 st century by 2100 1-yr RL at most by 60 ppb (!sic NAAQS results) RCP4.5*: 1 year return level (RL) way higher than under RCP4.5 as time evolves increasing RLs with time climate penalty
Changes in 1-year return level estimates RCP4.5*: 1 year return level (RL) way higher than under RCP4.5 increasing RLs with time about 2-4 ppb increase in 1-yr RLs strongest increase in 1-yr RL along the coast and at higher latitudes?? changes in atmospheric dynamics?? RCP4.5* 110 110 2006-15 50 2046-55 2091-2100 change 2046-55 vs 2006-15 change 2091-100 vs 2006-15 change 2091-100 vs 2046-55 4 50 4-4 -4
CONCLUSIONS improvement in Eastern US air quality following the NOx SIP Call Return level analysis is a useful tool to illustrate these changes GFDL CM3 is found to be biased high compared to observations Quantile Mapping is a useful approach to correct model bias Quantile Mapping preserves spatio-temporal structure of CM3 Quantile Mapping allows to estimate changes in MDA8O3 on local/regional scale Future development of US air quality depends strongly on the RCP followed optimistic scenario (RCP4.5) strong improvement in O 3 pollution (RCP4.5) Strong decline in non-compliance days and probabilistic return levels pessimistic scenario (RCP4.5X) almost no changes to 2000 levels slight increases in # non compliance days and RL along the Coast and higher latitudes
METHOD Goal is to correct the distribution of CM3 (M) to CASTNET (O) P O = h(p M ) first we derive selected quantiles (i) from both M and O i = min, P(0.01, 0.05, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 0.95, 0.99), max next we calculate the difference (D) at step i between M and O D i = M i O i next we calculate the difference (x,y) between i+1 and i for M and O x i = M i+1 M i y i = O i+1 O i next we calculate the ratio (R) of x and y at step i R i = x i y i next we perform a stepwise correction (Z j ) Z j = (M i D i ) + (R i j) for(m i,, M i+xi ), j={0,, x i 1 }
Climate Change only scenario Q10 change 2046-55 vs 2006-15 Q10 change 2091-100 vs 2006-15 Q75 change 2046-55 vs 2006-15 Q75 change 2091-100 vs 2006-15
Changes in the number of days above the NAAQS RCP4.5 2006-15 change 2046-55 vs 2006-15 change 2091-100 vs 2006-15 RCP4.5* 2006-15 change 2046-55 vs 2006-15 change 2091-100 vs 2006-15 HYPOTHETICAL Lower NAAQS of 60 ppb RCP4.5: strong decline in #days above NAAQS over 21 st century only limited number of exceedances by 2100 RCP4.5*: larger #days above NAAQS at beginning of 21 st century than RCP4.5 further increase in non compliance days (up to +6 days) as climate warms