Investigating Dry Deposition of Ozone to Vegetation

Abstract

Atmospheric ozone loss through dry deposition to vegetation is a critically important process for both air quality and ecosystem health. The majority of atmospheric chemistry models calculate dry deposition using a resistance‐in‐series parameterization by Wesely (1989), which is dependent on many environmental variables and lookup table values. The uncertainties contained within this parameterization have not been fully explored, ultimately challenging our ability to understand global scale biosphere‐atmosphere interactions. In this work, we evaluate the GEOS‐Chem model simulation of ozone dry deposition using a globally distributed suite of observations. We find that simulated daytime deposition velocities generally reproduce the magnitude of observations to within a factor of 1.4. When correctly accounting for differences in land class between the observations and model, these biases improve, most substantially over the grasses and shrubs land class. These biases do not impact the global ozone burden substantially; however, they do lead to local absolute changes of up to 4 ppbv and relative changes of 15% in summer surface concentrations. We use MERRA meteorology from 1979 to 2008 to assess that the interannual variability in simulated annual mean ozone dry deposition due to model input meteorology is small (generally less than 5% over vegetated surfaces). Sensitivity experiments indicate that the simulation is most sensitive to the stomatal and ground surface resistances, as well as leaf area index. To improve ozone dry deposition models, more measurements are necessary over rainforests and various crop types, alongside constraints on individual depositional pathways and other in‐canopy ozone loss processes.

Dry Deposition of Ozone Over Land: Processes, Measurement, and Modeling

“…In general, differences between site and model environmental variables (e.g., meteorology or soil moisture) and observational uncertainty confound model evaluation (Cooter & Schwede, ; Schwede et al, ; Silva & Heald, ; Tuovinen et al, ; Z. Wu et al, ). To reduce the impact of differences between site and model variables on model evaluation and thus separate input uncertainty from process and parameter uncertainty (e.g., Z. Wu et al, ), we recommend driving standalone schemes with observed variables representative of the flux‐tower footprint.…”

Sensitivity of Ozone Dry Deposition to Ecosystem‐Atmosphere Interactions: A Critical Appraisal of Observations and Simulations

“…Furthermore, the spatial and seasonal variability of stomatal deposition in LM4.0 is dynamic, depending not only on LAI but also on climate conditions via their effects on plant functioning. In contrast, V d,O3 from the Wesely scheme in GEOS‐Chem generally increases from spring to summer, simply scaling with the seasonal changes in LAI (Silva & Heald, ), not accounting for variations during the wet versus dry season that are inferred from observations (section ) and from LM4.0 simulations for Mediterranean Europe, the U.S. Pacific Northwest, Mexico, and South Asia (Figures , , and ). Silva and Heald () showed that the Wesely scheme in GEOS‐Chem generally reproduces the seasonal mean observed V d,O3 averaged across sites globally but has limited skill ( R 2 = 0.04) in simulating the site‐to‐site variations in observations, supporting that the Wesely scheme has a lack of sensitivity to local environmental variables.…”

“…With a resistance‐in‐series framework, the Wesely scheme is well suited for inclusions in global models and has success in some applications when evaluating observed and modeled monthly mean climatology of V d,O3 averaged globally across sites for a particular land cover type (Silva & Heald, ; Val Martin et al, ; Wesely & Hicks, ). However, the substantial interannual variability and site‐to‐site differences in V d,O3 derived from measurements are not simulated by CTMs with the Wesely scheme driven by observed meteorology (Clifton et al, ; Silva & Heald, ). The expression for stomatal resistance used in the Wesely scheme is dependent on only solar radiation, air temperature, and leaf area index (LAI; Wesely, ; Kavassalis & Murphy, ).…”

A Deep Learning Parameterization for Ozone Dry Deposition Velocities

“…The DNN model applied to the Hyytiälä test data has predictive skill, with performance as good or better than many theoretical parameterizations (Silva & Heald, ; Wu et al, ). However, the overall generalization error (error on the test set) of the DNN model as it applies to all mixed forests similar to Hyytiälä is likely underestimated in this situation because all of the data are not i.i.d.…”

“…We explore all possible months as a start date, and range from 1 to 12 months of continuous data throughout the year, with the DNN retrained on each combination of months. Figure shows that after approximately 6 months of continuous data collection (~2,800 observations) a generalization loss of within 50% of the fully trained DNN is reached, a value that should outperform the Weseley parameterization as implemented in GEOS‐Chem (Silva & Heald, ). In this specific case, training on months containing data from summers is more effective at reducing the generalization loss than winter months.…”