Gong S. L. (2003) A parameterization of sea-salt aerosol source function for sub- and super-micron particles, Global Biogeochem. Cycles, 17, 1097.
Granier C., Bessagnet B., Bond T., et al. (2011) Evolution of anthropogenic and biomass burning emissions at global and regional scales during the 1980–2010 period, Climatic Change, doi 10.1007/s10584-011-0154-1.
Guenther A., Karl T., Harley P., et al. (2006) Estimates of global terrestrial isoprene emissions using MEGAN (Model of Emissions of Gases and Aerosols from Nature), Atmos. Chem. Phys., 6, 3181–3210.
Guenther A. B., Jiang X., Heald C. L., et al. (2012) The Model of Emissions of Gases and Aerosols from Nature Version 2.1 (MEGAN 2.1): An extended and updated framework for modeling biogenic emissions, Geosci. Model Dev., 5, 1471–1492.
Hicks B. B., Baldocchi D. D., Meyers T. P., Hosker Jr. R. P., and Matt D. R. (1987) A preliminary multiple resistance routine for deriving dry deposition velocities from measured quantities, Water, Air and Soil Pollution, 36, 311–330.
Hobbs P. V. (2000) Introduction to Atmospheric Chemistry, Cambridge University Press, Cambridge.
Hudman R. C., Moore N. E., Mebust A. K., et al. (2012) Steps towards a mechanistic model of global soil nitric oxide emissions: Implementation and space-based constraints, Atmos. Chem. Phys., 12, 7779–7795.
Jaeglé L., Quinn P. K., Bates T. S., Alexander B., and Lin J. T. (2011) Global distribution of sea salt aerosols: New constraints from in situ and remote sensing observations, Atmos. Chem. Phys., 11, 3137–3157.
Johnson M. T. (2010) A numerical scheme to calculate temperature and salinity dependent air–water transfer velocities for any gas, Ocean Sci., 6, 913–932.
Kaplan J. O. (2002) Wetlands at the Last Glacial Maximum: Distribution and methane emissions, Geophys. Res. Lett., 29(6), 1079.
Lamarque J.-F., Bond T. C., Eyring V., et al. (2010) Historical (1850–2000) gridded anthropogenic and biomass burning emissions of reactive gases and aerosols: Methodology and application, Atmos. Chem. Phys., 10, 7017–7039, doi: 10.5194/acp-10-7017-2010.
Liss P. (1973) Processes of gas exchange across an air–water interface, Deep Sea Res., 20, 221–238.
Monahan E. C., Spiel D. E., and Davidson K. L. (1986) A model of marine aerosol generation via whitecaps and wave disruption. In Oceanic Whitecaps (Monahan E. and Niocaill G. M.), D. Reidel, Norwell, MA.
Nightingale P. D., Malin G., Law C. S., et al. (2000) In situ evaluation of air–sea gas exchange parameterization using novel conservative and volatile tracers, Global. Biogeochem. Cycles, 14, 373–387.
Riley W. G., Subin Z. M., Lawrence D. M., et al. (2011) Barriers to predicting change in global terrestrial methane fluxes: Analyses using CLM4Me, a methane biogeochemistry model integrated in CESM, Biogeosciences, 8, 1925–1953.
Schnetzler C. C., Bluth G. J. S., Krueger A. J., and Walter L. S. (2007) A proposed volcanic sulfur dioxide index (VSI), J. Geophys. Res., 102, 20087–20091.
Seinfeld J. H. and Pandis S. N. (2006) Atmospheric Chemistry and Physics: From Air Pollution to Climate Change, 2nd edition, Wiley, New York.
Shao, Yaping (2004) Simplification of a dust emission scheme and comparison with data, J. Geophys. Res., D10202, doi:10.1029/2003JD004372.
Slinn W. G. N. (1982) Predictions for particle deposition to vegetative canopies, Atmos. Env., 16, 1785–1794, doi:10.1016/0004-6981(82)90271-2.
Val Martin M., Kahn R. A., Logan J. A., et al. (2012) Space-based observational constraints for 1-D fire smoke plume-rise models, J. Geophys. Res., 117, D22204.
Vignati E., de Leeuw G., and Berkowicz R. (2001) Modeling coastal aerosol transport and effects of surf-produced aerosols on processes in the marine atmospheric boundary layer, J. Geophys. Res., 106, 20225–20238.
Wang J., Park S., Zeng J., et al. (2013) Modeling of 2008 Kasatochi volcanic sulfate direct radiative forcing: Assimilation of OMI SO2 plume height data and comparison with MODIS and CALIOP observations, Atmos. Chem. Phys., 13, 1895–1912.
Wesely M. L. and Hicks B. B. (2000) A review of the current status of knowledge on dry-deposition, Atmos. Environ., 34, 2261–2282.
Xi X. and Sokolik I. N. (2015) Seasonal dynamics of threshold friction velocity and dust emission in Central Asia, J. Geophys. Res., 120, 1536–1564, doi:10.1002/2014JD022471.
Zender C. S., Bian H., and Newman D. (2003) Mineral Dust Entrainment and Deposition (DEAD) model: Description and 1990s dust climatology, J. Geophys. Res., 108(D14), 4416, doi: 10.1029/2002JD002775.
Zhang L., Gong S., Padro J., and Barrie L. (2001) A size-segregated particle dry deposition scheme for an atmospheric aerosol module, Atmos. Environ., 35, 549–560.
10
Atmospheric Observations and Model Evaluation
10.1 Introduction
Atmospheric chemistry models try to provide a physically based approximation to real-world behavior that serves to understand the real world and from there to predict future changes. The approximation comes with some error – by definition, a model is not perfect. As the saying goes, “All models are wrong, but some are useful.” To make a model useful, it is critical to quantify its error. From there we may find that the error is acceptably small for the application of interest. Alternatively we may find that the error is too large and this then provides motivation for improving the model and often advancing scientific knowledge.
Quantifying model error requires reference to truth. Truth is elusive. Observations of atmospheric composition are our best resource. But they are sparse and have their own errors. Model error can never be fully characterized, but it can be estimated through statistical comparisons to observations. This chapter reviews simple metrics for this purpose, and also discusses the use of models as tools to interpret atmospheric observations in terms of processes. Formal approaches for error characterization and model optimization are presented in Chapter 11.
Different terms are used in the literature to describe the testing of models by comparison to observations. The word validation is often used but implies an exercise in legitimation to demonstrate that the model is true (valid) within certain error bounds. This may be appropriate terminology for regulatory models, where conclusions from the model have to hold in a court of law, but less so for research models. The term verification is sometimes used for operational applications (such as to verify a model forecast), but is inadequate for research applications where we may be more interested in falsifying the model, i.e., find out where the model is wrong so that we may improve it. We prefer here to adopt the term evaluation, which implies a broad assessment of model results, considering possible positive and negative outcomes, to understand the value of the model. Model evaluation offers the possibility of identifying unexplained behavior and from there advancing knowledge.
There are four types of model error. The first is error in our understanding of the physics as expressed by the model equations. The second is error in model parameters such as reaction rate constants or emissions that are input to the model equations. The third is numerical error in our approximate methods for solving the equations. The fourth is error in model implementation due to incorrect coding (bugs!). From an atmospheric chemist’s perspective, the first two errors are the most interesting because addressing them deepens our understanding of the physical system. But the other two are important to recognize. Numerical error can be estimated by conducting simulations for different grid resolutions and time steps, by using different numerical solvers, or by comparing to analytical solutions for simple ideal cases (see Chapters 6 and 7). Bugs should of course be hunted down, and are often revealed by comparisons to other models. A complex 3-D model is probably never bug-free, but over time we can hope that the bugs that remain have little impact (and are therefore hard to detect!).
Consider a situation where the model departs from observations more than we deem acceptable, and we have established that this is not due to numerical, implementation, or measurement errors. We are then left with the task of improving the model physics or improving
model parameters. Usually the first reach (because it is easiest) is to adjust the model parameters. These parameters have error ranges that can generally be estimated from the literature, such as uncertainty in rate constants. Adjusting model parameters within their error ranges is a perfectly legitimate exercise, and in fact the optimization of selected model parameters (called state variables) is the objective of inverse modeling, described in Chapter 11.
Adjustment of model parameters is often done in a simple way by constraining the model to match observations. This is called model calibration or tuning. A danger is that by ascribing all model error to the choice of some parameters we may be missing the opportunity to diagnose error in other parameters or in model physics – the familiar story of the drunk at night who looks for his missing keys under the lamppost because that’s where the light is. Model tuning may lead to the model getting the right result for the wrong reasons. To avoid this situation it is important to evaluate the model for a wide range of species, conditions, and statistics. Ad hoc model tuning of multiple parameters by trial and error to fit a limited number of observations is poor practice and may lead to the model behaving like a house of cards – precariously fitting the observations available (“don’t change a thing!”) but ready to collapse when new observations or objective improvements to model parameters are brought in.
This brings up the importance of using a large ensemble of observations for model evaluation. Using observations taken in a wide range of dynamical and chemical environments can test model behavior over different conditions, building confidence in the capability of the model to simulate changes and make predictions. Using observations of chemically coupled species is particularly useful for revealing errors in the model chemistry. For example, a model that simulates sulfate aerosol with no bias but overestimates the precursor SO2 may be producing sulfate with incorrect kinetics. Examining the relationship between two species with common emissions can help to separate emission errors from dynamical errors, as the latter will tend to affect both species similarly.
Research models used in atmospheric chemistry are generally versatile – they are intended to be applicable to a wide range of problems. The choice of application dictates such things as model domain and resolution, chemical mechanism, emission inventories, etc. It also defines the error tolerance. For some applications, we may be satisfied with a factor of 2 uncertainty; for other applications the tolerance may be much less. It is important to establish the error tolerance as it will affect the conclusions to be drawn from model evaluation. It is also important to identify what ensemble of observations can best evaluate the model for the particular application. These observations may not have been taken yet, which then calls for an experimental program as companion to the model study. The experimental program may take the form of a field campaign targeted at providing the observations needed for model evaluation. Such field campaigns involve tight partnership between experimenters and modelers, including, for example, the use of model forecasts to guide the day-to-day collection of observations in a way that can best test the model.
The concept of partnership between model and observations can be expanded by viewing the model as an integral part of the atmospheric observing system needed to answer a particular question. The observing system may include measurements from diverse platforms including ground-based sites, aircraft, and satellites. The model provides a common platform to integrate information from instruments measuring different species and operating on platforms with different measurement locations and schedules. Model evaluation with the ensemble of observations provides a check on the consistency of observations and enables constraints from multiple platforms. This can be formally done through data assimilation, as discussed in Chapter 11.
This chapter presents basic elements for carrying out model evaluation. Section 10.2 gives a primer on experimental methods and platforms. Error characterization for measurements and models is presented in Section 10.3, followed by general approaches to model evaluation in Section 10.4. Section 10.5 gives elementary statistical metrics. Statistical significance of differences is covered in Section 10.6. Section 10.7 discusses the use of models as tools to interpret atmospheric observations.
10.2 Atmospheric Observations
Measurements of atmospheric concentrations and fluxes are the main sources of data used to evaluate atmospheric chemistry models. Measurements are made in situ, when the instrument probes air from its vicinity, or remotely, when the instrument records a spectroscopic signal integrated over an atmospheric line of sight. Measurements are made routinely as part of long-term monitoring programs or intensively as part of field campaigns. Long-term monitoring programs may involve surface networks, sondes, commercial aircraft, or satellites. They are typically for a limited suite of species and provide information on short-term variability (events), long-term trends, and spatial patterns. They are particularly useful for long-term statistics and can be compared to the corresponding model statistics. Field campaigns typically provide a broader array of measurements deployed at specific locations of interest and for limited time. They generally focus on improving understanding of specific processes and are often geared to test model simulations of these processes. In such cases the models play a critical role in designing the field campaign and in interpreting the observations.
General methods for measuring concentrations include spectroscopy, mass spectrometry, chromatography, wet chemistry, and filters. Spectroscopic methods observe the interaction of atmospheric gases or particles with electromagnetic radiation. This radiation may be generated with a laser (active methods) or originate naturally from solar or terrestrial emission (passive methods). Mass spectrometry involves the ionization of an atmospheric sample followed by deflection of the ions in an imposed electromagnetic field. The angle of deflection is determined by the ratio of the electric charge to the mass of the ion. Chromatography involves the flow of an atmospheric sample through a narrow retention column in which individual species are separated by their different flow rates. Individual species are identified by their retention time in the column and their concentrations are measured by a detector at the exit of the column. Wet chemistry methods involve the capture of atmospheric gases and particles in a liquid sample, either by bubbling or spraying, followed by chemical analysis of the sample. Filter methods collect atmospheric samples through a porous filter, sometimes chemically treated. The filter is then analyzed by optical methods, gravimetric methods, or liquid-phase extraction followed by wet chemistry methods. Table 10.1 gives an overview of widely used measurement methods for different atmospheric species, and the following subsections provide additional information on specific methods and measurement platforms. More detailed information can be found, for example, in Finlayson-Pitts and Pitts (2000), Baron and Willeke (2005), Farmer and Jimenez (2010) and Burrows et al. (2011).
Table 10.1 In-situ and remote sensing methods for measurements of atmospheric composition
Species In-situ methods Remote sensing methods
H2O Frost point hygrometer
Lyman alpha absorption
Tunable diode laser IR spectroscopy
Microwave spectroscopy
Raman lidar
Filter radiometry
CO2 Gas chromatography
IR gas correlation IR spectroscopy
Filter radiometry
CO Gas correlation
Chemical conversion
Differential absorption IR spectroscopy
Gas correlation radiometry
CH4 Gas chromatography
Tunable diode laser
Differential absorption
Gas correlation IR spectroscopy
Filter radiometry
VOCs Gas chromatography
PTR-MS
Chemical ionization mass spectrometry IR spectroscopy
O3 UV absorption
Chemiluminescence
Electrochemical sondes UV/Vis spectroscopy
IR spectroscopy
Microwave spectroscopy
Lidar
N2O Gas chromatography
Tunable diode laser
Differential absorption IR spectroscopy
Radiometry
NO Chemiluminescence IR spectroscopy
NO2 Photolysis and chemiluminescence
Laser-induced fluorescence UV/Vis spectroscopy
IR spectroscopy
HNO3 Tunable diode laser
Ion chromatography
Filter and wet chemistry IR spectroscopy
Filter radiometry
N2O5 Cavity ringdown IR spectroscopy
HCl, HF Tunable diode laser IR spectroscopy
Cl, ClO Resonance fluorescence Microwave spectroscopy
OCS Tunable diode laser IR spectroscopy
SO2 Ion chromatography
Chemiluminescence UV spectroscopy
IR spectroscopy
DMS, CS2, H2S Gas chromatography
OH Resonance fluorescence
Laser-induced fluorescence
Chemical ionization mass spectrometry
Radioisotope chemistry Lidar
UV spectroscopy
DOAS
Far-IR spectroscopy
HO2, RO2 Radical amplifier
Laser-induced fluorescence Far-IR spectroscopy
CH2O Gas chromatography
Tunable diode laser
Wet chemical methods
Laser-induced fluorescence UV and IR spectroscopy
H2O2 High-performance liquid chromatography
Chemical ionization mass spectrometry Far-IR spectroscopy
O2, N2, H2,
Ar, Ne, He Mass spectroscopy
Aerosol Filters
Optical particle counters
Condensation nuclei counters
Cascade impactors
Differential mobility analyzers
Mass spectrometry
Electron microscopy Lidar
UV/Vis/IR spectroscopy
Modeling of Atmospheric Chemistry Page 53