Modelling of Heavy Metal Aairborne Pollution in Europe:
Evaluation of the Model Performance

EMEP/MSC-E Technical Report 8/2005

Ilyin I. and Travnikov O.

ABSTRACT

In accordance with the TFMM work plan [EB.AIR/GE.1/2004/3] MSC-E has started the preparation to the EMEP/TFMM Workshop on the review of MSC-E models. The aim of the Workshop is to identify whether the models are state-of-the-art and fit for purposes declared in the Protocols on heavy metals and POPs. The evaluation procedure consists of three main components: sensitivity study, intercomparison with other models and verification against measurements. Detailed description of the model was published in the special Technical Report [Travnikov and Ilyin, 2005] and presented at the 6th session of TFMM (Croatia, April 2005) along with the model sensitivity study and uncertainty analyses.

The model sensitivity study demonstrated that the modelling results for such heavy metals as lead and cadmium are the most sensitive to anthropogenic emissions, natural emission and re-emission and to removal parameters. Sensitivity of mercury modelling results is the highest to boundary concentration of gaseous elemental mercury. The intrinsic model uncertainty of lead and cadmium concentration in air, concentration in precipitation and total deposition are estimated as 43%, 40% and 33% respectively; the appropriate uncertainties for mercury are 19%, 53% and 39%.

Besides, the model has participated in a series of the intercomparison studies aimed to evaluate ability of the participated models to assess pollution levels in Europe with such pollutants as lead, cadmium and mercury [Sofiev et al., 1996; Gusev et al., 2000; Ryaboshapko et al., 2005]. The lead and cadmium intercomparison studies (1996, 1998-99) have shown that the discrepancy of the modelling results did not exceed a factor of two for all the models and less than 50% for models of similar approaches (Eulerian or Lagrangian). The multi-stage mercury intercomparison study was finished this year. Comparison of the modelling results with observations and between each other demonstrated that the participated models could predict the observed concentrations of elemental mercury with accuracy within ±20%. Accuracy of wet deposition modeling versus measurements is within a factor of two, whereas discrepancy between the models did not exceed 40%. Besides, the models can predict mercury atmospheric budgets for individual countries with accuracy within a factor of 2.
This report is focused on the verification of the model performance via comparison of modelling results with available observations for the period 1990-2003. Consistency of emission and monitoring data is also analysed as an integral part of the model performance analysis.

As it was demonstrated by the sensitivity study emission data are among the most important input parameters for the modelling. The overall completeness of national data on annual total emissions officially submitted to UN/ECE Secretariat for the period 1990-2003 is about 60%. Significant part of the European territory is not covered with national emissions data. Expert estimates of anthropogenic emissions are used for countries not submitted their national data. Analysis of national data consistency has shown that anthropogenic emissions data officially submitted to the Convention cannot explain observed levels of lead and cadmium wet depositions in Europe. Based on the observations of these metals in the EMEP monitoring network one can expect 2-3-fold underestimation of the emission data in Europe. In order to obtain consistent emission datasets for lead and cadmium, which conform modelling results to observed pollution levels in Europe, adjusted anthropogenic emission scenarios were developed for the period 1990-2003 based on available measurements of these metals in air and precipitation. According to the adjusted scenarios emissions of lead and cadmium in some European countries are 2-3 times or even more higher those from official data. The adjusted scenarios were applied for calculations of lead and cadmium pollution levels in Europe used in the model validation.

Measurements of lead, cadmium and mercury concentrations in air and precipitation available from the EMEP monitoring network for the period 1990-2003 were used for the model evaluation. A critical analysis of the data was performed in cooperation with CCC and national experts and inconsistent measurement data were removed from the consideration. Totally data from 81 stations measuring lead and cadmium and from 19 stations measuring mercury were involved into the model verification. Correct evaluation of the model performance via comparison with observations requires information on uncertainty of the measurement data themselves. However, this kind of information is very limited. Only uncertainties of analytical methods are available for lead and cadmium. The effects of other possible sources of the uncertainty (e.g. sampling) are not evaluated. The quantitative characteristics of the uncertainties of mercury measurements are not available at all. Measurement information from other monitoring programmes (ICP-Forests and ICP-Vegetation) is used as the supplementary data source for the model verification. Information on concentrations in precipitation of lead from 26 ICP-Forests stations in Germany and on concentrations of lead in mosses from 24 countries is involved.

The evaluation of the model performance demonstrates that the modeled concentrations of lead in air and in precipitation well agree with the available measurements. For the totality of the stations involved in the model verification for 1990-2003 the correlation coefficient for annual lead concentrations in air is as much as 0.89, and for concentrations in precipitation 0.74. Mean levels of the concentrations are also well reproduced by the model: the coefficients of regression are 0.99 and 0.80 for concentrations in air and in precipitation, respectively. As much as 87% of modeled lead concentrations in air and 71% of concentrations in precipitation agree with measurements with accuracy better than ±50% of measured value. The best model performance is obtained for Central and Western Europe. Significant discrepancies between the modeled and observed lead concentrations in air are obtained for the high Arctic stations and for the high mountain station SK2 in Central Europe. Besides, lead concentrations in precipitation are often underestimated in Scandinavia. Comparison of modelling results with lead concentrations in precipitation measured in forests (ICP-Forests) shows satisfactory agreement at least in a number of German Lands. Significant correlation is also obtained between modelled total deposition fluxes and lead concentrations in mosses (ICP-Vegetation).

Evaluation of modeling results for cadmium is more challenging task comparing to that for lead. Concentrations of cadmium in air and in precipitation are normally much smaller those of lead. That is why number of cadmium samples, which are below the detection limit, is much higher. Besides, results of the regular analytical intercomparison studies indicate that uncertainties of the laboratory analysis are larger for cadmium. Therefore, the results of the comparison of modelled and measured pollution levels for cadmium are more uncertain and difficult for interpretation compared to lead.

Cadmium concentrations in air predicted by the model are in good agreement with available monitoring data in aggregate. The correlation is high (0.78) and the regression coefficient is equal to unity. Besides, for more than 90% of compared values the difference does not exceed ±50% of measured value. The observed cadmium concentrations in precipitation are commonly underestimated by the model (the regression coefficient is 0.6). However, correlation between modeled and measured values is high (0.77) and more than 76% of model/measurement pairs agree within the range ±50% of measured value. The comparison of modeled and measured values for individual monitoring stations demonstrates that the agreement is better for Central and Western Europe, whereas for some Scandinavian and Baltic stations the model performance is poorer. Besides, similar to lead, significant discrepancies are obtained for high Arctic and mountain stations.

Mercury differs from lead and cadmium by complex physical and chemical transformations in the atmosphere. Besides, the global character of mercury dispersion leads to significant contribution of anthropogenic and natural sources from other continents and the oceans to mercury pollution in Europe. This makes mercury modelling to be the most complicated task among other heavy metals.

Concentration levels of mercury in air and in precipitation are well reproduced by the model. Modelled air concentrations of total gaseous mercury (TGM) are in good agreement with observed ones at annual base: the regression coefficient is 0.92. Only few mean annual values of modeled TGM concentrations differ from the observed ones more than by 30%. The correlation coefficient is not high (0.38) because of low spatial and temporal variability of modeled and measured TGM concentrations. Correlation between modelled and observed mercury concentrations in precipitation is significant (the correlation coefficient is 0.5) and almost 80% of modelled values agree with measurements within the range ±50% of measured value. However, the model tends to some overestimation of mercury concentrations in precipitation. Temporal variations of mercury pollution levels are simulated by the model reasonably well. At most monitoring stations the model successfully captures both long-term change and the seasonal cycle of mercury concentrations in air and in precipitation.