Model Uncertainty: How well can we represent the processes that affect these intercontinental or global flows of air pollutants in quantitative models?
Our ability to represent the processes that drive intercontinental or global flows of air pollutants in quantitative, predictive models varies across the different pollutants of interest. In general, our confidence in the predictions of the models decreases from O3 to PM to Hg to POPs, as the complexity of the processes that must be represented increases and as the available observational data base decreases.
We can evaluate our modelling capabilities by comparing model estimates of concentrations and deposition to the observed magnitudes, patterns, and trends for each pollutant. We can also compare the estimates of different models, which give us some sense of the lower bound of the uncertainty in our modelled estimates. Estimating the current level of uncertainty or confidence in modelled estimates of intercontinental source-receptor relationships and identifying major areas of uncertainty have been main objectives of the HTAP multi-model experiments.
Ozone and PM
Our confidence in model estimates of the sensitivity of O3 in Europe and North America to anthropogenic emissions in other regions has not changed since HTAP1. Models are able to reproduce broad spatial and seasonal patterns of observations, but biases can be as large as the estimates of extra-regional contributions. Thus, we have some confidence in the ability of models to qualitatively describe the role of regional and extra-regional sources and processes. However, uncertainty in quantitative estimates of extra-regional contributions generally are the same order of magnitude as the estimates themselves.
The range of estimates of surface O3 concentrations in the HTAP2 ensemble is large and of similar magnitude to that in HTAP1, despite having used the same emissions in all HTAP2 models. Regional models, which have higher spatial resolution, are less biased than global models for surface O3 when compared to observations, which is to be expected. The larger errors in global models are particularly important for threshold-based metrics, such as AOT40. Nevertheless, the best performing global models have less error than the worst performing regional models. The largest sources of error are found in temporal processes acting on longer time scales (weeks or months), including emissions, their interaction with chemistry, and long-range transport processes.
The range of estimates for PM2.5 narrowed between HTAP1 and HTAP2. Global and regional models perform similarly for PM2.5, with a tendency to underestimate PM2.5 concentrations in Europe and North America. Removing this low bias may significantly increase estimates of PM2.5 exposure and health impacts.
Our estimates of emissions sources globally have improved as previously unaccounted sources have been included (e.g. sources of black carbon). However, uncertainties in emissions inventories in some parts of the world remain high. For example, HTAP2 emissions for India were greater by a factor of 1.5 to 2 for NOx, NMVOC, and SO2 as compared to an inventory created at the national scale (Venkataraman 2018). Satellite observations have helped improve estimates of emissions trends in many parts of the world.
Current global atmospheric Hg models reproduce the observed ground-level Hg0 concentrations to within 20% of the sparse observations that are available and reproduce the pronounced inter-hemispheric gradient in baseline Hg0 concentrations that has been observed. The agreement between models and observations for Hg wet deposition is weaker, with differences between observed and modelled values up to 100%, mainly due to uncertainties in Hg emission rates, Hg oxidation chemistry and estimated precipitation rates. There are larger differences across model estimates of wet deposition in areas where there are little observational data. Significant differences exist among models for dry deposition, which is believed to contribute as much as wet deposition to total Hg deposition and for which there is little observational data.
Using identical inputs for new anthropogenic emissions, the four models participating in the HTAP1 multi-model experiments predict comparable Hg deposition levels in Europe, North America and South Asia, but the estimated Hg deposition levels are a factor of 1.5 higher in East Asia. The level of agreement is striking, given the significant differences among models in the assumptions used concerning emission rates for natural or re-emitted Hg and the oxidation pathways for Hg0. For deposition in the Arctic, the differences across models are much higher, with a spread greater than a factor of four, due to differences primarily in the representation of atmospheric Hg depletion events.
The relative contributions of the major source regions to region-wide average Hg deposition in different regions are very similar among the models. The most significant deviations in the modelling results are seen in areas with large anthropogenic and natural and secondary emissions.
This is due to large uncertainties in the natural and secondary emission estimates and to the differences in spatial resolution of the participating models, which varied by a factor of four.
Although there are significant uncertainties in the anthropogenic emissions from some source categories and regions of the world, the magnitude of new anthropogenic emissions is thought to be much less than the emissions of re-emitted Hg that was previously deposited. However, the rates of re-emission from terrestrial and aquatic systems, especially from the ocean which may emit twice as much Hg as anthropogenic sources, are not well characterized.
Persistent Organic Pollutants
As with Hg, POPs models must not only simulate the behaviour of pollutants in the atmosphere, but they must also simulate the exchange between the atmosphere and other environmental media (such as water, soil, snow, ice, and vegetation) and the transport and transformations that occur in those other media. Observational data from these media are limited, making it difficult to evaluate models and characterize uncertainties.
Current POPs models vary widely in the level of detail represented. Model simulations for a subset POPs have been conducted and are typically able to reproduce observed annual concentrations to within a factor of three or four, enabling identification of major transport pathways. The POPs that have been successfully modelled and evaluated include selected PCBs, HCHs, and PAHs. In some cases, however, the differences between model estimates and observed values can be much greater indicating fundamental uncertainties both in emission inventories and in modelling approaches.
For some POPs, there are significant uncertainties associated with the pollutant’s physical- chemical properties, such as Henry’s Law constants, vapour pressures, and octanol-air partition coefficients. These physical-chemical parameters are used in the models to predict how the pollutant will move among media. However, the total environmental lifetime and long-range transport potential of a POP are not intrinsic substance properties. The fate of pollutants also depends on the characteristics of the environment in which it is found. Little is known about how the abundances and biodegradation of POPs vary in heterogeneous media such as soil, snow, oceans, and lakes.