Critical Assessment of Tropospheric Ozone Simulations in CMIP6 Earth System Models

Abstract:

Henry Bowman^1,2, David D. Parrish^1,3

(1) NOAA CSL, Boulder, CO

(2) Carleton College, Northfield, MN

(3) CIRES, University of Colorado Boulder, CO

david.d.parrish@noaa.gov

CMIP6 Earth System Models simulate seasonal tropospheric ozone cycles that change in magnitude over time, growing from the 1950s (as anthropogenic precursor emissions rose), reaching a maximum in the 1980s, then decreasing as emission controls began. In the background troposphere over the 165-year model simulations, a 12-parameter equation consistently captures more than 95% of the variance in monthly mean concentrations:

[O₃] = a + bt + ct² + dt³ + et⁴ + A₁*(1 + r*exp-((t-m)/s)²)*sin(2πt + φ₁) + A₂*sin(4πt + φ₂).

The first five terms of the equation—a fourth-degree polynomial—describe long-term (years to decades) changes. The final two terms describe seasonal variations using a Fourier Series with the fundamental and second harmonic, each with magnitude A and phase f. An additional three-parameter Gaussian function is included in the fundamental to account for the changing amplitude of the seasonal cycle over time. The single independent variable is time in years relative to 2000 (t = year - 2000).

We derived model- and observation-based parameter values at high alpine and marine boundary layer (MBL) surface sites, and in the free troposphere (from sonde and aircraft measurements). These derived values provide metrics for quantitative comparison of measurements with models, and between different models. We have assessed six CMIP6 Earth Systems Model simulations (BCC-ESM1, CESM2-WACCM, GFDL-ESM4, GISS-E2-1-H, MRI-ESM2-0, UKESM1-0-LL) at eight sites in Europe and North America chosen to have minimal regional influence. Preliminary indications are that these CMIP6 models regularly overestimate ozone, especially higher in the troposphere. Additionally, the models typically perform better at high alpine sites than in the MBL. These models generally performed better than previous CMIP5 models.

Figure. 165 years of monthly mean ozone concentrations (blue markers) simulated at a Swiss high alpine site. The red curve gives the fit of the 12-parameter equation to the mean concentrations. Annotations show parameters with 95% confidence intervals. The fit has a RMSD of 2.1 ppb and captures 96.6% of the variance in the original data set.