2013 Progress Report: Creating Building Blocks for a More Dynamic Air Quality Management Framework

EPA Grant Number: R835215
Title: Creating Building Blocks for a More Dynamic Air Quality Management Framework
Investigators: Demerjian, Kenneth L. , Beauharnois, Mark , Bielawa, Robert , Civerolo, Kevin , Hogrefe, Christian , Ku, Michael , Mao, Huiting , Yun, Jeongran
Institution: The State University of New York at Albany , New York State Department of Environmental Conservation , SUNY College of Environmental Science and Forestry , The State University of New York , The State University of New York at Albany , U.S. EPA
Current Institution: The State University of New York at Albany , New York State Department of Environmental Conservation , SUNY College of Environmental Science and Forestry , U.S. EPA
EPA Project Officer: Chung, Serena
Project Period: June 1, 2012 through May 31, 2015 (Extended to May 31, 2016)
Project Period Covered by this Report: June 1, 2013 through May 31,2014
Project Amount: $499,945
RFA: Dynamic Air Quality Management (2011) RFA Text |  Recipients Lists
Research Category: Air Quality and Air Toxics , Air


The overall objectives of the proposed work are to: 1) develop a prototype system for providing real-time information on the contribution of short-term emission sources to air quality in relation to other source categories and the potential air quality benefits from episodic control measures; and 2) perform a comprehensive multi-pollutant air quality assessment that examines trends in pollutant concentrations versus emission controls and co-pollutant effects, and develop possible indicators that may aid in improved tracking of the effect of emission controls. 

Progress Summary:

Status of the Direct Decoupled Method (DDM) with the Community Multiscale Air Quality (CMAQ) model production runs.

After several weeks of sensitivity runs, using the Intel compiled version of CMAQ-DDM v4.7.1 as described in the first year annual report, analysis of the model output revealed computational anomalies that we had not discovered during our benchmark testing of the Intel version of the CMAQ-DDM code received from the U.S. EPA. Our configuration of the simulations contained 12 DDM species and 3 sensitivity regions (NYC Only, Mid-Atlantic/Northeast Visibility Union (MANE-VU) no NYC, and all grids except MANE-VU).

Detailed analysis of the 8/1/2007 model output, showed unreasonably high (E+15) values within the sensitivity output file. We then analyzed the output from 5/1/2007 to 5/25/2007 and discovered that the sensitivity output became numerically unstable on 5/21/2007. We discussed these anomalous output values with CMAQ-DDM developers at EPA, who indicated this instability was known, and was presumably some sort of memory error that was fixed in the beta version of CAMQ-DDM 5.0.2. Whether this is the case or not, moving to the beta version is not an option given the incompatibility of the changing emission requirements of the DDM beta version and the selected emission factors we have configured for our sensitivity application. Therefore, we continued with a reconfigured CMAQ-DDM v4.7.1 to “work-around” this anomaly. With the new DDM configuration, the Intel compiled CMAQ-DDM model, produced reasonable results for the simulation period 5/1/2007 to 5/31/2007. Simulations were continued for the sensitivity period of interest, but with some reservation as to whether there was still a possibility that similar CMAQ-DDM anomalous “numerical instabilities” might be encountered.

In March 2014, further detailed analysis of the CMAQ-DDM sensitivity output from the Intel compiled 16 core simulations revealed anomalous results within the LADCEN (LADCO + CENRAP) region. We then decided to stop all simulations and conduct more benchmark tests on the Intel and PGI versions of the CMAQ-DDM code. After extensive multi-day benchmarks, we concluded that there was most likely an unsolved memory leak with the Intel compiled code and decided to revert back to the PGI version, and run on only 8 cores (to avoid potential MPI communication glitches across network interconnects). Sensitivity runs, using this new configuration, were rerun for our study period of 5/1/2007 to 9/30/2007.

While running the southeastern U.S. sensitivity area for the all sources emissions case, another numerical instability was encountered. This transient event occurred late in the run (8/4/2007) in the southeast corner of the domain, just off the coast of South Carolina, producing a high (>E+15) right next to an extreme low in sensitivity values. Several days of additional benchmark tests using different domain decomposition settings for the PGI compiled, 8 core model code were performed. The intent was that if there was a memory bug in the model code, then perhaps altering the memory map of the domain decomposition might circumvent the bug.

The domain decomposition tests showed that these (albeit rare) model numerical instabilities could be circumvented by restarting the DDM simulation on the last good day, with the modified domain decomposition setting. The simulation is run past the anomalous event for 2 days at which time, the domain decomposition is returned to its original setting and the model simulations continued without difficulty. While debugging this issue, we also discovered that another “work-around” option is to turn off the compiler optimization feature while compiling CMAQ-DDM model. Results from this approach compared to within 6 significant digits to the control domain decomposition simulations. To be consistent with the domain decomposition settings of the base code, we opted to proceed with the optimization feature change “work-around."  The simulation is run past the anomalous event for 2 days and the model simulations are returned to the original setting.

We believe there is still some unresolved memory bug in CMAQ-DDM v4.7.1. This has most likely been resolved in the most recent beta version of CMAQ-DDM 5.0.2. However, our emissions speciation constraints required that we continue using CMAQ-DDM v4.7.1 for this study. The “work-around” options to address these anomalous numerical instability events have proven successful and the CMAQ-DDM sensitivity production runs now are back online. Unfortunately, identifying and resolving the numerical instability events in the CMAQ-DDM has delayed the sensitivity production runs by several months. The findings of all test simulations were shared with CMAQ-DDM developers at EPA because they may be relevant for future CMAQ-DDM releases.

Preliminary Results CMAQ-DDM Sensitivity Simulations

CMAQ-DDM simulations to compute O3 sensitivity to NOx and VOC precursor emission changes are progressing in the following emission categories: 1) all anthropogenic emission sources; 2) mobile source emissions; 3) combined area and nonroad emissions; 4) “peaking unit” electric generating unit (EGU) point sources emissions; 5) all EGU point sources emissions; 6) other point sources emissions; and 7) biogenic emissions. In addition, the sensitivity to boundary conditions also will be calculated. Spatially, sensitivity fields are calculated separately for emissions from the NYC only area (NYCONLY), the MANE-VU region except NYC (MVNONYC), the southeastern U.S. region (SESARM), and the rest of the modeling domain (LADCEN) to distinguish sensitivities from local vs. regional emissions. Figure 1 shows the progress of CMAQ-DDM model production runs for the matrix of sensitivity fields under consideration.  We expect to finish the simulations by the end of December 2014.

Sensitivity of Ozone to Peaking Units Versus All EGU Point and Mobile Source Emissions

As we described in the previous annual report, there is a robust correlation between ambient temperature, energy load, and EGU point sources emissions. On days of high energy demand, which are associated with high ambient temperatures, additional generators are operated for power generation. These units are referred to as “peaking units."  The “peaking units” are identified by total operating time less than 15% of all hours. The peaking unit NOx emissions can contribute significantly to total EGU NOx emissions and air quality on those high temperature days. For example, on 8/8/2007, the peaking units emitted nearly 350 tons/day of NOx, which accounts for approximately 25% of the total point source NOx emissions within the MANE-VU region. We are characterizing the sensitivity of ozone concentrations to peaking EGU units compared to all EGU units and mobile source emissions in the MANE-VU region. 

Figures 2, 3 and 4 show scatterplots of modeled vs. observed 8 hour maximum O3 concentrations in Holtsville, Queens College (QC), and Pinnacle State Park (PSP) in NY, respectively, during the period of May 4 to September 30, 2007. The model overestimated O3 concentrations in urban areas: Holtsville and QC with R2 of 0.722 and 0.6622, respectively, and underestimated O3 concentrations in rural areas: PSP with R2 of 0.4403.  Overall model predicted O3 concentrations show good agreement with observations in urban areas. 

We currently are analyzing the sensitivities of predicted O3 concentrations with respect to seven specified emission source categories and emission domains outlined in Figure 1, and exploring ways to parse this extensive data set.  We grouped hourly O3 sensitivity data from May 4 to September 30 based on 8 hour maximum O3 into the following categories: 10 worst days, greater than or equal to 75 ppb (>=75 ppb), between 50 ppb and 75 ppb (50-75 ppb), and less than 50 ppb (<50 ppb).  We extracted O3 sensitivity data in the following monitoring sites: Holtsville, QC, White Plains, PSP, and Elmira in New York, and Westport in Connecticut.  Hourly modeled O3 distribution for each category in all areas showed a peak around hour 13 or 14 EST (see Figure 5 as an example).

As an example, extracted sensitivity data from the Holtsville monitoring site are analyzed in Figures 6 to 16 showing hourly distributions of O3 sensitivities to NOx and VOC emissions from each emission category: all sources, mobile sources, all EGU point sources, and peaking unit EGU point sources from MVNONYC and NYCONLY for 10 worst days. As seen in Figures 6-16, O3 sensitivities in Holtsville for the 10 worst days show diurnal patterns for MVNONYC and NYCONLY emissions, while there are no distinct diurnal patterns found in the SESARM and LADCEN emissions cases (not shown). NOx emissions from NYCONLY show negative O3 sensitivity due to NOx titration, while NOx and VOC emissions from MVNONYC and NYCONLY and VOC emissions from NYCONLY show positive O3 sensitivity. 

Comparing O3 sensitivity to mobile sources NOx emissions and O3 sensitivity to all sources NOx emissions, up to 50% of O3 sensitivity to all sources emissions is from mobile sources emissions. Up to 20% of O3 sensitivity to all sources NOx emissions is from all EGU sources NOx emissions. On the other hand, O3 sensitivity to peaking EGU sources emissions is small. Even though the peaking units can contribute up to approximately 25% of the total point source NOx emissions within the MANE-VU region, our preliminary results show that O3 sensitivity to peaking EGU sources seems to be minimal. 

We have started to explore the quantification of the temporal and spatial variations of the sensitivity fields by parsing ozone 8 hour max concentrations greater than or equal to 75 ppb (>=75 ppb), between 50 ppb and 75 ppb (50-75 ppb), and less than 50 ppb (<50 ppb). We have started to calculate predicted ozone concentrations resulting from emissions reductions of each sector in each region based on percent reductions calculated from emissions estimates for years 2011 (EPA), 2018 (EPA), and 2020 (MARAMA). 


Progress in the second objective of this project is proceeding on schedule. We have compiled the emissions and air quality concentration data resources for selected monitoring sites in the Northeast to track the comparability of emissions and concentration trends. We have begun analyses of emission tracers and multi-pollutant relationships including CO, NOx, NOy, SO2, CO vs. O3, NOy vs. O3, CO vs. NOx, Hg0 vs. CO, Hg0 vs. SO2 and SO2 vs. NOx as well as analyses of their annual trends and factors impacting inter-annual variability. The comparison of observed trends in multi-pollutant relationships with DDM sensitivities will be considered in the coming year, when DDM results become available. The results then will be analyzed by air quality planners collaborating on the project to identify avenues for making progress towards a more adaptive, dynamic air quality planning framework. 

Factors Controlling Trends in Background O3 and CO in the Northeastern U.S., 2001 – 2010

Seasonal and interannual variability in baseline O3 mixing ratios were examined at seven rural sites in the northeastern U.S. during the time period of 2001 – 2010 or longer depending on data availability at individual sites. Using carbon monoxide (CO) mixing ratios to filter out the local influence of anthropogenic activities, it was found that there were not consistent significant changes in baseline O3. Solar radiation was found to account for 30% of the total variance of baseline O3. In winter, decreased cyclone activities, which appeared to be related to negative North Atlantic Oscillation (NAO), Arctic Oscillation (AO), and positive Pacific North Atlantic Oscillation (PNA), could potentially increase surface background O3 mixing ratios in the Northeast via reduced cloud cover and precipitation and enhanced solar radiation, which could have counteracted the effect of anthropogenic emission reductions on surface O3 levels. In spring there was a negative correlation between baseline O3 and climate indices including NAO, East Atlantic West Russia (EAWR), and Scandinavia pattern (SCA). It suggested that positive teleconnection patterns were associated with intensified jet streams, which accelerated global mixing and subsequently could decrease surface O3 levels in the northeastern U.S. through increasing continental export flux. It also was found that baseline CO decreased significantly at a rate of 2.25-5.42 ppbv yr-1 during 2001-2010, which appeared to be driven largely by the Siberian wildfire emissions in summer 2003, which were the largest of the decade. The contribution from wildfires to background CO in the northeastern U.S. also was verified using the Global Fire Emission Dataset (GFED) and MOPITT satellite CO column retrievals. No clear relation was found between background CO mixing ratios and any climate index. Hemispheric transport from biomass burning in Siberia and Canada may have led to elevated baseline O3 in the Northeast in summer 2003. These findings are in preparation for publication.

Temporal Variability in Hg° in Bronx, NY 2008 – 2012

Seasonal, annual, and interannual variability of Hgo and the potential mechanisms driving them were investigated at an urban site located in the Bronx borough in New York City using continuous measurements data from 2008 – 2012. The most discernible diurnal cycles occurred during the summer seasons, with a general peak of 193 – 211 ppqv between 2:00 and 5:00 AM and a trough of 134 – 158 ppqv between 12:00 and 16:00 PM, which is consistent with previous studies for urban locations, in large part linked to constant emissions and the diurnal variation of the planetary boundary layer height. Winter 2009, spring 2009, spring 2010, fall 2010, and summer 2011 had more than 10 days with extremely high Hg0 mixing ratios (>1000 ppqv), most likely due to local emissions. Our analysis also suggested large interannual variability in the seasonal averaged diurnal cycles of Hg0. Of the five cold seasons, 2010 exhibited the lowest Hg0 levels, whereas 2012 exhibited the highest Hg0 levels largely associated with the interannual variability in circulation patterns, i.e., more frequent and faster transport of relatively cleaner air from southern Canada via northwesterly flow in winter 2010 compared to more frequent and faster transport of polluted air masses via southerly flow in winter 2012. This was supported by the contrasting correlation between Hgo and CO mixing ratios in winters 2010 and 2012 with r2 values of 0.06 and 0.37, respectively. The Hgo-SO2 and Hgo-NOx correlations between Hgo and SO2 and between Hgo and NOx were consistent with the Hgo-CO correlation and the trajectory analysis, corroborating the important role of regional transport in the interannual variability in winter Hgo concentrations. Spring 2010 appeared to experience higher Hg0 mixing ratios compared to springs of other years. Of the four warm seasons, 2011 experienced the lowest concentrations. Analysis of causes for the warm season interannual variability is ongoing. It was hypothesized that for the urban site Bronx, regional influence can be dominant over local sources on ambient contributions of Hgo. Contribution of regional vs. local emissions in Bronx is being quantified. These findings are in preparation for publication.

Potential Controls on Trends in Background Concentrations of Hg0 in the Northeastern U.S.

An analysis of multi-year data sets suggested a decreasing trend of 3.8±0.9 ppqv yr-1 in background mixing ratios of gaseous elemental mercury (Hgo) at an elevated, rural site, Pack Monadnock, in New Hampshire, USA. It is in close agreement with the declining trends reported from Mace Head (3.1±1.1 ppqv yr-1), Cape Point South Africa (3.8±0.6 ppqv yr-1), and mid-latitude Canadian sites (~2.6-3.9 ppqv yr-1). At a coastal, rural site, Thompson Farm, in southern New Hampshire, it was found that an abrupt increase in the fall of 2006 seemed to lead to no trends there during the period of 2003 – 2010. At another rural site, Huntington Wildlife Forest in upstate New York, no trend was observed over February 2006 – August 2013 despite the three unusually large values in February, March, and December 2007. Further examination suggested a decreasing trend in the anthropogenic component at all three sites, which probably was the mechanism that drove the decline in the background concentrations observed at locations from previous studies. However, near the surface, different from other years, there was abundant precipitation in the senescence months in 2006 followed by a lack of snow in the following winter in the northeastern U.S. An examination of long-term soil moisture data for a northeastern site (Lye Brook, NY) suggested soils to be the driest in the year of 2007 during the decade of 2001–2011. Drier soils had been observed to be closely linked to reduced Hgo evasion rates from soils. It was thus hypothesized that near the surface, Hgo evasion from ecosystems in fall 2006 and the year of 2007 was significant enough to alter the declining trend in background Hgo in the northeastern U.S. that appeared to be controlled primarily by decreasing anthropogenic emissions. These findings are in preparation for publication. 

Future Activities:

We plan to complete DDM sensitivity computations for the period May 2007 to September, 2007, and continue diagnostic analyses of sensitivity runs including the calculation of predicted ozone concentrations resulting from emissions reductions of each sector in each region based on percent reductions calculated from emissions estimates for years 2011 (EPA), 2018 (EPA), and 2020 (MARAMA). We will continue analyses of air quality measurements in the northeast. 

Journal Articles on this Report : 1 Displayed | Download in RIS Format

Other project views: All 13 publications 3 publications in selected types All 3 journal articles
Type Citation Project Document Sources
Journal Article Zhou Y, Mao H, Demerjian K, Hogrefe C, Liu J. Baseline carbon monoxide and ozone in the northeast US over 2001–2010. Atmospheric Chemistry and Physics Discussions 2015;15:27253-27309. R835215 (2013)
R835215 (Final)
  • Full-text: ACPD-Full Text PDF
  • Abstract: ACPD-Abstract
  • Supplemental Keywords:

    episodic emission controls, atmospheric modeling, emission trends, air quality trends, multi-pollutant analysis 

    Relevant Websites:

    Progress and Final Reports:

    Original Abstract
  • 2012 Progress Report
  • 2014 Progress Report
  • Final Report