GFDL - Geophysical Fluid Dynamics Laboratory

Skip to main content

D.J.'s Chemistry-Climate Research Page

This is an overview of my involvement in various research projects focusing on current and future chemistry-climate interaction using the GFDL AM3, a global chemistry-climate model.

Please scroll down to read a synopsis about each research project

Ozone-Temperature relationships in the eastern US

The relationship between temperature and surface O3

Observational studies have shown strong correlation between surface temperature and ozone (O3) concentrations. This relationship is commonly thought to reflect at least three components in the eastern US: (1) association of warm temperatures with stagnant air masses enabling accumulation of local chemistry precursors that feed O3 formation in the planetary boundary layer (Jacob et al., 1993; Olszyna et al., 1997); (2) thermal decomposition of peroxyacetylnitrate (PAN) at high temperatures, thus decreasing NOx and HOx sequestration at low temperatures (Cardelino and Chameides, 1990; Sillman and Samson, 1995); and (3) increasing biogenic emissions of isoprene, a major NMVOC precursor for O3 formation under high-NOx conditions (Guenther et al., 1993; Lamb et al., 1987; Meleux et al., 2007). These effects can be represented aggregately by a total derivative, d[O3]/dT.
Past studies have found d[O3]/dT to be approximately linear over the temperature range of 290-305 K ( Bloomer et al., 2009; Camalier et al., 2007; Sillman and Samson, 1995; Steiner et al., 2006). The slope of this linear relationship, mO3-T, has been referred to as the "climate change penalty factor" (Bloomer et al., 2009) in reference to the increase in O3 associated with increasing temperature.

Climate change impacts on O3
It is widely anticipated that a warming climate will exacerbate O3 pollution in densely populated regions of the US, such as over the Northeast where climate models consistently show annual temperature increases of at least 2 K over the 21st century. Increasing concentrations of surface O3 resulting from climate change are a public health concern. As such, air quality managers seek to be informed as to how surface O3, among other pollutants, will evolve in the future. Chemistry-climate models (CCMs) are increasingly being applied to project air quality under various global change scenarios. These models, however, have known biases in their simulations of present-day meteorology and chemical environments that raise concern as to their ability to project accurately the response of air pollution to changes in climate (Fiore et al., 2009;  Reidmiller et al., 2009)

Methodology
We describe a mechanistic approach to model evaluation that seeks to characterize pollutant sensitivity to year-to-year fluctuations in weather, motivated by the hypothesis that our approach offers a good observational basis for assessing model skill at projecting air quality response to changes in climate.

We first produce a monthly climatology of the surface O3-temperature relationship (d[O3]/dT) using monthly averages of maximum daily temperature and maximum daily 8-hour average (MDA8) O3 from the EPA CASTNet over the eastern US (archives of these relationships from EPA CASTNet sites across the entire US are available for download in text format here). We then evaluate the ability of the GFDL AM3 (Donner et al., 2011), a global CCM, to resolve the observation-derived O3-temperature relationships. Since a global CCM is expected to resolve synoptic, though not local, scales (e.g. Fiore et al., 2003), we construct regional O3-temperature climatologies in the eastern US where pollution episodes are large-scale (Logan, 1989) and observational records are longest.

Fig. Geographic locations of CASTNet observation sites in three defined regions in the eastern US: Great Lakes, Northeast, and the Mid-Atlantic.

Our evaluation of the GFDL AM3 CCM shows modeled MDA8 O3 biases in summer (ranging from +10 to +20 ppb in the Northeast and +10 to +30 ppb in the Great Lakes and Mid-Atlantic). Simulated monthly mean Tmax is at times 5 K too warm with respect to observations in the latter two regions. Despite these biases, GFDL AM3 reproduces the general spatial and temporal characteristics of mO3-T and associated correlation coefficients O3 in the Northeast, although it underestimates mO3-T by 2–4 ppb K-1 in the summer over the Mid-Atlantic, where simulated correlation coefficients are excessively weak.

Fig. Relationships between monthly regional averages of MDA8 O3 (ppb) and of daily Tmax (K) for individual years from 1988 to 2009 (solid black circles) and from the GFDL AM3 model for 1981–2000 (red triangles). For each regional average, we require 75% of all regional sites within a region for the specified month to meet the selection criteria. Scatter plots of July monthly mean MDA8 O3 concentration (ppb) and July daily Tmax (K) with linear regression fits ( mO3-T; left column). Also shown are monthly values of Pearson correlation coefficients (r ; middle column) and mO3-T (right column); mO3-T is not shown for months where r < 0; error bars indicate ± 1 standard error on  mO3-T (ppb K-1). The dashed lines at r = 0.5 indicate where at least 25% of the variance in the surface O3 is associated with temperature variability.

Estimating the impact of model biases in Tmax on simulated O3

Excess summertime surface O3 formation in the eastern US is a pervasive problem in gridded global (Fiore et al., 2009; Murazaki and Hess, 2006; Reidmiller et al., 2009) and regional (Nolte et al., 2008) models and raises questions about the accuracy of their estimates of future O3 concentrations. Here, we use the O3-temperature relationship derived from observations (conservatively estimating a 3 ppb K-1 sensitivity) to investigate the potential contribution of model temperature biases to excess surface O3 over our study domain. 


To evaluate modeled Tmax biases where CASTNet sites do not exist, we utilize 20 years of monthly mean Tmax from two independent gridded datasets: The University of Washington (UW) (Maurer et al., 2002) and the North American Regional Reanalysis (NARR) (Mesinger et al., 2006). The UW data are hourly observations from the National Oceanic and Atmospheric Administration /National Climatic Data Center Co-op stations spatially interpolated and gridded to 1/8o x 1/8o resolution. The NARR data are the product of assimilating the NCEP /Department of Energy global reanalysis (Kanamitsu et al., 2002) with a version of the NCEP Eta model at 32 km x 32 km horizontal resolution.

The estimated simulated summer MDA8 O3 associated with the model Tmax bias is shown in Figure 7. Both the NARR and the UW indicate that temperature is not a factor in excess O3 production in June in the eastern US, but they disagree in July. The UW data suggest the GFDL AM3 CCM is too cool in this month, yet both the NARR and the CASTNet observations suggest the opposite. The NARR output is 3-hourly while the UW data are calculated from hourly data, so NARR should be inherently cooler than UW; we suggest this discrepancy between the datasets requires additional study.

Fig. June, July, August, and September excess modeled O3 (ppb) attributed to eastern US maximum daily surface temperature biases in the GFDL AM3 CCM; left column uses temperature bias of GFDL AM3 versus Maurer et al. (2002); right column uses bias of GFDL AM3 versus NARR; temperature biases are multiplied by a conservative, observationally-derived estimate for mO3-T for 3 ppb K-1 over the eastern US.

Key Findings
  • We produce monthly climatologies (1988-2009) of surface ozone-temperature relationships
    (d[O3]/dT) over the eastern United States, using monthly averages of daily maximum surface
    temperature (Tmax) and of maximum daily 8-hour average (MDA8) O3 calculated from
    hourly observations at the Environmental Protection Agency Clean Air Status and Trends
    Network (CASTNet).
  • Using two different methods to aggregate regional data, we show that the GFDL AM3
    chemistry-climate model reproduces the slope of the ozone-temperature relationship, mO3-T
    (ppb K-1), over the northeastern US, despite substantial biases in surface O3 in this region.
    The model severely underestimates the observationally derived mO3-T, by 4 ppb K-1 in some
    summer months, over the Mid-Atlantic in part due to excessively warm modeled
    temperatures above which ozone production saturates.
  • Combining modeled Tmax biases with a conservative observation-based mO3-T estimate of 3
    ppb K-1 we estimate that excessively warm temperatures may contribute an upper limit of
    10-15 ppb to the pervasive model O3 bias in August and September, but that this factor is not
    the major driver of the O3 bias (up to +30 ppb over parts of the domain from June through
    September).

Characterizing policy-relevant background (PRB) ozone over the US



The US Environmental Protection Agency defines policy-relevant background (PRB) ozone as the surface ozone that would be present over the US in the absence of all North American (US, Canada, and Mexico) anthropogenic ozone precursor emissions. The PRB is used to advise the setting of the US National Ambient Air Quality Standard (NAAQS) to gauge the maximum ozone reduction that would result from North American emissions controls. The current NAAQS for maximum daily 8-hour average surface ozone is 75 ppb. However, the US EPA has recently been making attempts to lower this to a value in the range of 60-70 ppb. As this air quality standard continues to edge closer to PRB levels, greater interest is elicited in characterizing these quantities.

Fig. Concentrations of observed (1988 - 2002) monthly mean MDA8 ozone during the month of April at the locations of EPA CASTNet sites across the continental US.

Fig. Concentrations of monthly mean policy-relevant background MDA8 ozone during the month of April at the locations of EPA CASTNet sites across the continental US from a 20 year (1981 - 2000) GFDL AM3 simulation with zero North American anthropogenic emissions.



Since PRB ozone is not directly observable, estimates must be calculated with chemical transport models. We use the GFDL AM3 chemistry-climate model to calculate PRB ozone at locations over the US with global simulations ran with zero North American anthropogenic emissions. The components of PRB ozone (intercontinental transport of ozone and its precursors, stratospheric ozone entrained into the troposphere, biogenic ozone precursor emissions, wildfires, lightening NOx, methane) are both spatially and temporally variable. As such, we seek to characterize both year-to-year variations and climatological means of PRB ozone at locations across the US.


A time series showing the year-to-year variation in March-April-May averages of policy-relevant background (green), base case modeled (red), and observed (black) MDA8 ozone at the Grand Canyon National Park EPA CASTNet site. The spearman rank correlation coefficient is given for both observations and base case modeled and observations and policy-relevant background conditions. The two-sided significance from zero is given in parenthesis (values closer to zero imply a more significant correlation).