- Email: [email protected]
A comprehensive Eulerian modeling framework for airborne mercury species: comparison of model results with data from measurement campaigns in Europe
ARTICLE IN PRESS
Atmospheric Environment 37 Supplement No. 1 (2003) S51–S62
A comprehensive Eulerian modeling framework for airborne mercury species: comparison of model results with data from measurement campaigns in Europe S.R. Schmolkea, G. Petersenb,* a
Institute of Ecology and Environmental Chemistry, University of Lueneburg, Scharnhorststrasse 1, Lueneburg D-21332, Germany b GKSS Research Centre, Institute of Coastal Research, System Analysis and Modelling, Max-Planck-Strasse, Geesthacht D-21502, Germany
Abstract Previously, we have simulated the atmospheric transport and fate of mercury emissions in Europe and derived estimates of ambient concentrations and dry and wet deposition of mercury species using a mercury version of the Acid Deposition and Oxidants Model (ADOM). In this study, we focus on comparisons of model results against observations from European measuring campaigns performed between 1995 and 1999 in the framework of the Germany–Canada Science and Technology Co-operation Agreement and the Mercury Species over Europe (MOE) project funded by the European Commission. Simulations were conducted using 1997/1998 meteorology and 1990 emission inventories. Model simulations show a good performance in explaining the peak mercury concentrations on a 800 km south to north transect, which were represented by simultaneous measurements at four sampling sites. Regional concentration gradients and differences in the concentration variability are well reproduced by the model even on a time base of 1 h. r 2003 Published by Elsevier Science Ltd. Keywords: Mercury; Numerical modeling; Eulerian models; Model validation; Total gaseous mercury
1. Introduction Efforts have been made to simulate the atmospheric transport and fate of mercury species and to derive estimates of ambient concentrations and dry and wet deposition fluxes of mercury over North America (Shannon and Voldner, 1995; Bullock et al., 1997; Pai et al., 1997; Seigneur et al., 2001) and Europe (Petersen et al., 1995; Ryaboshapko et al., 1998; Pirrone, 1998; Lee et al., 2001) through either relative simple Lagrangian formulations or Eulerian approaches employing extensive gas and aqueous phase chemical mechanisms and tracking explicitly numerous species concentrations. The starting point for this work was the existing Eulerian modeling framework of the Acid Deposition *Corresponding author.. E-mail address: [email protected] (G. Petersen).
and Oxidants Model (ADOM) originally designed for regional-scale acid precipitation and photochemical oxidants studies. Modules of ADOM have been restructured to accommodate most recent developments in atmospheric processes of mercury. These processes include transport by three-dimensional flows over a domain with horizontal scales of a few thousand kilometers and vertical scales of a few kilometers, transformation by gas and aqueous phase chemistry, scavenging by cloud processes and interactions of gaseous and particulate species with the ground. Fig. 1 shows schematically the various components of the ADOM model modified for transport, transformations and deposition of mercury. Two different versions of the model exist: a European 76 76 grid domain and a North American 33 33 grid domain version with a grid cell size of approximately 55 55 and 135 135 km2, respectively. The vertical grid, with 12
1352-2310/03/$ - see front matter r 2003 Published by Elsevier Science Ltd. doi:10.1016/S1352-2310(03)00250-4
ARTICLE IN PRESS S52
S.R. Schmolke, G. Petersen / Atmospheric Environment 37 Supplement No. 1 (2003) S51–S62 METEOROLOGICAL PRE-PROCESSOR EUROPE: HIRLAM High Resolution Limited Area Weather Prediction Model NORTH AMERICA: CMC Canadian Meteorological Center Large-Scale Model
INPUT DATA
POLLUTANT TRANSPORT MODEL (ADOM)
emission data bases
transport and diffusion
upper and surface meteorology
dry deposition
geophysical data
gas-phase chemistry
initial conditions boundary conditions secondary pollutants
wet scavenging
cloud physics
aqueous phase chemistry
Fig. 1. The ADOM model system for mercury. Relation between ADOM modules and input parameters.
unevenly spaced levels between 0 and 10 km, is identical for both versions and is designed to resolve the higher concentration gradients in the boundary layer. The basic model time step is 1 h. Horizontal and vertical wind fields along with the eddy diffusivity, temperature, humidity, surface precipitation, and information about the distribution of clouds make up the meteorological input data set. The data set is derived diagnostically using the weather prediction model HIRLAM for Europe and the Canadian Meteorological Center’s model for a North American version of the model. Other than meteorological data, the input and data requirements for ADOM are emissions, initial and boundary conditions, geophysical data and fields of other air pollutants relevant for chemical reactions with mercury species such as ozone and elemental black carbon (soot). European emission data for mercury were derived from the UBA/TNO inventory for toxic substances in Europe (Berdowski et al., 1997). These inventories were compiled for the reference year 1990. An emission update for 1995 including mercury speciation and emission heights for various source categories has become available now (Pacyna et al., 2001) and comparative model runs using the inventories for 1990 an 1995 are now underway. Wet scavenging involves modules for handling cloud physics and aqueous phase chemistry. Clouds are classified as stratus (layer clouds) or cumulus (convective clouds) according to the diagnostic output from the
weather prediction model. Observations of the fractional coverage and the vertical extent of clouds are combined with output from the diagnostic model to yield the input fields used in this module. A detailed description of the entire ADOM modeling framework can be found in ERT (1984). The development and testing of the cloud physics and mercury chemistry module considered to be the core part of the ADOM model system for mercury and the entire model system and first applications are described in Petersen et al. (1998) and Petersen et al. (2001), respectively.
2. Comparison of model results against observations Model evaluation is a key consideration when developing new and advanced comprehensive models. Since the ADOM mercury version was intended to be applied to simulations with policy implications, thorough validation and verification of the model and its components is a requirement. A model developed or utilized without continual comparison against actual data is less than worthless, it is dangerous. Such a model is nothing more than a collection of mathematical formulae, no matter how elegant its formulation or how efficient its coding. Only with the close integration of state-of-science models and state-of-science experimental measurements real progress can be made towards the solution of the complex problems that are currently
ARTICLE IN PRESS S.R. Schmolke, G. Petersen / Atmospheric Environment 37 Supplement No. 1 (2003) S51–S62
S53
faced, i.e. the continual interplay between conceptual understanding and experimental evidence. High quality monitoring data for mercury air concentrations and deposition in Europe are still limited, although several recent investigations have improved the database (Ebinghaus et al., 1995; Lee et al., 1998; Munthe, 2001; Pirrone, 1998; Schmolke et al., 1999). Fortunately, a few studies of total gaseous mercury (TGM) concentration measured simultaneously in Germany and Sweden are available now for evaluation of model performance. These data, although limited to four sites in Germany and Sweden (Fig. 2) and to a few 1–2 months measurement periods provide an opportunity for comparison with model predicted concentrations. A detailed description and a full documentation of the comparison of model results against observations is provided in the final report of the project ‘‘Mercury Species over Europe’’ funded by the EC (Munthe, 2001). This section is restricted to some illustrative examples concerning episodes of high and low mercury concentrations in Europe.
German atmospheric monitoring stations. The distance between the most northern and most southern site was approximately 800 km (Fig. 2). Details about the applied automated mercury measurement techniques are provided in Schmolke et al. (1999). Measurements were taken during two sampling periods: A 2 weeks early spring campaign during March 1995 (automated measurements at all four sites), and a winter period during November/December 1998 (automated measurements at all sites except Roervik). During the winter measurements at Roervik only a few data were collected by manual enrichment of ambient air mercury on gold traps and later thermodesorption and detection by AFS in the laboratory. The applied measurement techniques are generally accepted to detect TGM, which is an integral parameter over all gaseous mercury species. In addition to the simultaneous data at the four sites a full 2 month (November and December 1998) 0.5 h time resolved TGM data set measured at Zingst was available and used for some additional investigations.
2.1. Measurements in ambient air
2.2. Applications of the model to episodes of high mercury concentrations in Central Europe
From south to north transect atmospheric mercury measurement studies performed in Central and Northern Europe were used for this model evaluation exercise. The sampling was performed at two Swedish and two
In Fig. 3 1 h averages of model predicted mercury concentrations in ambient air are compared with observations at four sampling sites, namely Neuglobsow,
Fig. 2. Location of atmospheric monitoring sites, involved in the evaluation of model performance. The detail map of the model domain displays the major atmospheric mercury source areas. The 1990 UN-ECE/OSPARCOM/HELCOM emission source strength are indicated in tons/year.
ARTICLE IN PRESS S54
S.R. Schmolke, G. Petersen / Atmospheric Environment 37 Supplement No. 1 (2003) S51–S62
Fig. 3. Observed 1 h average total gaseous mercury (TGM) concentration values (black area) superimposed by calculated elemental mercury concentrations time series (gray Line). The displayed time window covers the entire November/December 1998.
Zingst, Aspvreten and Roervik. As already mentioned above, all observed data except Roervik are available with high time resolution during 2 weeks. The 1 h average TGM observations are indicated by black needles, whereas the discontinuously measured Roervik data are represented by shaded bars. Each individual bar is spanning over the sampling time and indicating the average concentration. Superimposed to the measured
data, the model predicted hourly Hg0 concentration time series are plotted as gray lines. We decided to neglect the oxidized gaseous mercury species in our model evaluation exercise due to their minor fraction of only 0.470.3% (maximum 2.3%) of total gaseous species at the southern most site, and even a decreasing trend and variability with increasing distance to the major source regions. Thus the subsequent
ARTICLE IN PRESS S.R. Schmolke, G. Petersen / Atmospheric Environment 37 Supplement No. 1 (2003) S51–S62
discussion will compare modeled Hg0 concentration with measured TGM concentration. Whereas the project related measurements were conducted only during approximately 2 weeks, an entire 2 month TGM data set was available at the Zingst monitoring site operated by the German Environmental Protection Agency. To summarize the ambient air mercury concentrations at the four sites, and for direct comparison of the calculated and observed mercury 1 h averages, basic descriptive statistical indices were calculated and are listed in Table 1. The considered time window is restricted to the periods of available simultaneous measurements during 18 March to 1 April 1997 and 23 November to 7 December. One should note the episodic nature of the observed and model predicted values at the two German sites spanning more than a three-fold range of concentrations from 1.2 to approximately 7 ng m 3. A few observations lower than 1 ng m 3 were made during 2 and 6 December at Neuglobsow. In general the lower end of the range is representative of hemispheric background concentrations of about 1.5 ng m 3 (Ebinghaus et al., 1995; Lee et al., 1998), whereas the concentrations at the upper end of the range are most probably due to longrange transport from anthropogenic sources in Central Europe. The very low concentration events of o1 ng m 3 are possibly affected by long-range transport of arctic air masses, or due to inflow of stratospheric air parcels. Two particular events of simultaneously elevated ambient air mercury concentration in the range between 2.5 to 5.5 ng m 3 at all four sampling sites during 25 November and 4 and 5 December 1998, which were
S55
detected by the measurements and also simulated by the model, are good examples to examine the model performance. Not only the temporal occurrence of the peak events, but also the concentration levels are in a similar range except Aspvreten on 25 November where the measurements do not support the model predicted peak. The wind fields during that day has been examined by means of 72 h back trajectories, which have been derived from the horizontal meteorological input data at approximately 250 m height. These trajectories describe the route taken during the previous 3 days by the air masses arriving at the four sites. Fig. 4 shows 24 trajectories, each with 1 h delay to the next, during 01 a.m. to 12 p.m. on 25 November 1998. Within the graph major source regions are indicated by source strength proportional black shading on the underlying map. Originating in Romania and the Ukraine air masses move along the trajectories into the main emission areas in Southern Poland and Eastern Germany, where they change their directions and move northwards arriving at the four sites. Trajectories which are attached to the modeled peak events at the individual sampling sites during 25 November are plotted in bold. Whereas the air parcels, which were responsible for the significant peak concentrations at the three southern most sites, were clearly influenced by the hot spot emission area in Eastern Germany (Figs. 4b–d), the backward trajectories reaching Aspvreten only pass the north eastern vicinity of this area. Thus, some uncertainties in the location/strength of the touched emission area or in the wind fields may be responsible for the model/measurement inconsistency at the northern most site during the discussed event.
Table 1 Summary statistics of observed and calculated ambient air mercury concentration during Spring 1997 and Winter 1998 at four atmospheric monitoring stations Neuglobsow Calc.
Zingst Obs.
Roervik
Aspvreten
Calc.
Obs.
Calc.
Obs.
Calc.
Obs.
18 March, 01:00–1 April, 12:00 1997 Arit. Mean 2.02 2.14 Median 1.76 2.03 S.D. 0.63 0.36 Maximum 4.10 4.03 Minimum 1.42 1.62 Count 324 319
1.76 1.62 0.39 3.45 1.34 324
2.13 2.07 0.31 3.78 1.71 324
1.39 1.35 0.16 2.41 1.23 324
1.93 1.93 0.20 2.87 1.53 321
1.41 1.36 0.24 2.79 1.26 324
1.77 1.78 0.24 4.81 1.38 302
23 November, 13:00–7 December, 09:00 1998 Arit. Mean 2.95 2.06 Median 2.65 1.99 S.D. 1.20 0.87 Maximum 6.04 6.68 Minimum 1.46 0.33 Count 333 333
2.83 2.57 1.11 5.45 1.33 333
2.64 2.43 0.86 7.02 1.53 333
1.86 1.76 0.49 3.45 1.21 333
1.60 1.52 0.23 2.10 1.45 8
1.61 1.56 0.26 2.37 1.30 333
1.69 1.62 0.26 2.69 1.21 333
ARTICLE IN PRESS S56
S.R. Schmolke, G. Petersen / Atmospheric Environment 37 Supplement No. 1 (2003) S51–S62
Fig. 4. 72 h backward trajectories 25 November 1998 calculated for the receptor areas: (a) Aspvreten, (b) Roervik, (c) Zingst, (d) Neuglobsow. Trajectories which are attached to the peak concentration during 25 November at the individual sampling sites are plotted in bold.
Another example of coinciding peak events at the four sampling sites during spring 1997 is depicted in Fig. 5. The different subdivision of the time axis should be noted. In contrast to Fig. 3 major tick marks denote 24 h, and minor 12 h time intervals, that is to say, each vertical line indicates 12 h. Elevated levels during 26 March 1997 are quite well reproduced by the model at the four sites. Maximum concentrations have been observed and calculated around noon at Neuglobsow, Zingst and Roervik. The model predicted a second smaller peak some hours later in the evening. The observation at Neuglobsow, and less pronounced at Zingst and Roervik as well, showed clearly the same pattern. The Aspvreten maximum is occurring a couple of hours later and the slight evening peak is missing at all. These phenomena are supported by the routes of the trajectories on 26 March (Fig. 6). During the first-half of
the day air masses start over Poland and adjacent areas, traverse East Germany picking up high emission rates and eventually arrive at the four sites. In accordance with Fig. 4 this graph also highlights trajectories which are related to the calculated peak concentrations. Except Aspvreten, trajectories are changing their routes significantly during the afternoon originating over the Western Atlantic and subsequently passing over emission areas in Britain and West Germany. Evidently, the routes of these trajectories cause the second smaller concentration peaks. The overall good agreement between observed and model predicted peak concentrations during episodes when trajectories exhibit the above described travel pattern is an indication that the model has capabilities to simulate atmospheric transport over distances of several hundred kilometers and that the emission inventory used with the
ARTICLE IN PRESS S.R. Schmolke, G. Petersen / Atmospheric Environment 37 Supplement No. 1 (2003) S51–S62
S57
Fig. 5. Observed 1 h average TGM concentration values (black area) superimposed by calculated elemental mercury concentrations time series (gray Line). The Time window covers 2 weeks in the end of March 1997.
model is based on realistic emission estimates for Central Europe in 1998. 2.3. Applications of the model to episodes of low mercury concentrations in Northern Europe In terms of comparing measurements with model output at all sites, there is evidence that the model has a tendency to underestimate observations during low
concentration and low variability episodes probably due to missing natural and re-emission processes in the model at present. Both, observed and model predicted time series are characterized by low temporal variability except the high concentration episode described in the previous section and except some low concentration episodes typically occurring over a time period of 12 h or less. Of particular interest in Fig. 5 are the four events of observed low concentration at the northern most site
ARTICLE IN PRESS S58
S.R. Schmolke, G. Petersen / Atmospheric Environment 37 Supplement No. 1 (2003) S51–S62
Fig. 6. 72 h backward trajectories 26 March 1997 for: (a) Aspvreten, (b) Roervik, (c) Zingst, (d) Neuglobsow. Trajectories which are attached to the peak concentration during 26 March are plotted in bold.
Aspvreten on 22, 23, 24 and 28 March 1997, which are not reproduced by the model. The related trajectories in Fig. 7 are either originating from Finland and Northern Russia or they are pointing towards ice-covered regions of the Arctic Ocean as a potential source of mercury depleted air masses during that time of the year. Indeed, mercury depletion phenomenon have been observed during the 3 months period following polar sunrise in March (Schroeder et al., 1998; Schroeder and Barrie, 1998) and the observed episodic minimum concentrations in the Aspvreten time series may be an indication for transport of air masses from polar regions. The current model version is not able to reproduce these effects which are due to particular chemical processes occurring outside of the model domain. However, the database is too scarce to draw any firm conclusions and more work in terms of using a model approach that can take into account the global cycling of mercury is
required to identify this phenomenon with more confidence (Seigneur et al., 2001). On the other hand, these regional simulations must rely on boundary conditions that may influence the model results. It is necessary therefore to develop a multiscale modeling approach, that consists of the existing regional scale ADOM model for mercury and a global chemical transport model, which provides the time-dependent boundary concentrations for the ADOM model.
3. Evaluation of the model performance Subject of the previous section was the discussion of some particularly high and low concentration episodes due to different meteorological conditions. Now some more general aspects of the model capability to reproduce ambient mercury concentrations pattern will
ARTICLE IN PRESS S.R. Schmolke, G. Petersen / Atmospheric Environment 37 Supplement No. 1 (2003) S51–S62
S59
Fig. 7. 72 h backward trajectories in Aspvreten on: (a) 22 March 1997, (b) 23 March 1997, (c) 24 March 1997, and (d) 28 March 1997.
be discussed. The box and whisker plot (Fig. 8) gives an aggregated overview about the modeled and measured 1 h average concentrations at the four selected sampling sites and during the previously discussed time window. Each box is indicating the minimum and maximum concentration (whisker), the inner quartile range (box), the median concentration (central symbol), and the confidence interval of the median (notch). The boxes are arranged in four pairs from the southernmost site at the left end to the northernmost site at the right end of the X-axis. It is important to note that the distance to the major mercury source regions in Eastern Germany increases on the south to north transect. Each pair of boxes represent one site, each left-hand box summarizes the modeled data, the right hand box the observations. The time window of observations is restricted to approximately 2 weeks during both experiments, when measurements at all sites were performed. The model predicts significantly decreasing concentrations of
approximately 0.5 ng m 3 on the south to north transect during the spring period (lower part of the graph) and 1 ng m 3 during the winter period. There is evidence for increasing median concentrations with decreasing distance to the major source regions. This finding is in good agreement with the observations. Except the Neuglobsow winter measurements, which show systematically lower absolute concentrations, these measurements also display a qualitative and absolute similar gradient, however with a tendency of the model to underestimate the atmospheric mercury concentration during springtime. Not only the regional pattern but also the frequency distribution of observed concentrations at each site are well reproduced by the model. Both, the calculated and measured data reflect a tendency of increasing signal variability (box extension) with decreasing distance to the major sources regions. Despite the systematically lower absolute concentration of the Neuglobsow winter data the frequency distribution of
ARTICLE IN PRESS S.R. Schmolke, G. Petersen / Atmospheric Environment 37 Supplement No. 1 (2003) S51–S62
S60 7
calc obs
6
Hg[ng/m³]
5
Nov. 23 - Dec. 07, 1998 4 3 2 1 7 6
March 17 - 31, 1997
Hg[ng/m³]
5
and measured time series. The middle and top part of the graph gives the residuals of the original time series and the 24 h running average curve which show a significant 24 h component. The major characteristics of the measured time series are sufficiently well reproduced by the model. Not only the major peaks but also details in the shape of certain peak events are simulated surprisingly well. However, the model has a tendency to overmodulate the signals. Fig. 10 represents a scatter plot of measured versus simulated data. The slope of 0.59 of the included regression curve supports this finding. Low concentrations tend to be underestimated and conversely, peak events are overestimated by the model. The linear regression analysis gives a coefficient of determination (R2 ) of 0.688 which is an encouraging number with respect to the model performance.
4 3
4. Summary and conclusions
2 1 0 neu
zin
roe
asp
Fig. 8. Campaign separated box statistics. Condensed view on the calculated and observed Hg concentration during the period 23 November–7 December 1998 (top) and 17–31 March 1997 (bottom).
simulated and observed data is similar. Based on the available data material a final judgement whether the effect of unexpected low measured concentration at the southern most site Neuglobsow is due to a calibration/instrumental error of the measurements or an overestimation of the model is not possible. However, the backward trajectory analysis previously discussed indicates that measurements might be too low. Finally, a value by value comparison of the measured and modeled time series is performed on the basis of the 2 winter month 1998 data set at Zingst. The following considerations were made concerning the direct comparability between model simulations and measurements. Highly time resolved field measurements always contain a short time random component which is due to local effects in the vicinity of sampling sites such as irregular car traffic, ship emissions and local heating. A 24 h cycle process, which is driven by diurnal meteorological conditions is often an additional part of atmospheric trace substance concentration time series. Focus of this investigation are transport distribution processes which act on a longer time scale. To filter out short time effects, a 24 h moving average procedure was applied. Fig. 9 (lower part) shows the smoothed curves of the modeled
A comprehensive model for airborne mercury species was tested by applying it to episodes of high and low mercury concentrations in ambient air in Central and Northern Europe, respectively. Time series of Hg0 estimated by the model were compared on an hourly basis with observed concentrations of TGM. The evaluation data material consists of 1 h average total gaseous mercury concentrations time series on a 800 km south to north transect, consisting of simultaneous measurements at four sampling sites. The model evaluation led to the following issues: The model prediction of regional Hg0 concentrations in ambient air is in good agreement with observed time series. Regional gradients, characterized by source strength and density, are well reproduced not only qualitatively but also quantitatively. High Hg0 concentration episodes could be traced back to specific source regions. The pattern of individual peak events originating from different source areas is well reproduced. These findings suggest that the implementation of the transport algorithms and emission estimates are on a satisfying level. However, the model shows a tendency to overmodulate high and low concentration episodes. Moreover, the model underestimates the average concentration level in Northern Europe during spring conditions at continuos northern inflow conditions. The reasons are not yet clear, but there are some indications that these effects could be attributed to overestimated point source strength in the 1990 emission data base and missing diffuse sources like reemission from soil and water surfaces. The application of a recently available Hg emission data base for 1995 may help to clarify this point. More information about air/surface exchange fluxes of gaseous mercury species are needed to
ARTICLE IN PRESS
Hg [ng/m³]
Hg [ng/m³]
S.R. Schmolke, G. Petersen / Atmospheric Environment 37 Supplement No. 1 (2003) S51–S62
S61
1.0 0.0 -1.0 1.0 0.0 -1.0 6 calcualted observed
Hg [ng/m³]
5 4 3 2 1
0
96
192 288 384 480 576 672 768 864 960 1056 1152 1248 1344 1440
hour (data origin Nov.01. 1998 01:00) Fig. 9. Decomposed 60 days Mercury time series, measured during November and December 1998 at the Zingst air quality monitoring site. Bottom: Observed and calculated 24 h moving average concentration. Middle: Short-term component (residuals) of the calculated series. Top: Short-term component of the observed time series.
implement these kind of processes into the model system. Additionally, an improvement of the inflow boundary conditions has to be considered in the future. The systematic underprediction of mercury concentrations at the northern inflow boundaries points to this issue. Also, the measured mercury depletion at the northern most site is thus an issue of hemispheric or global transport. This process, which may be due to atmospheric chemical processes outside of the model domain, cannot be reproduced by the model. Consequently, nesting of the regional ADOM model into a hemispheric model system should be one of the future activities to combine the advantage of high spatial resolution of regional models with a realistic integration of meteorological transport and chemical processes of a global pollutant.
6
Hg observed [ng/m³]
5
4
3
2
obs = 0.59 * calc + 0.76
1
R²=0.688 0 0
1
2
3
4
5
6
Acknowledgements
Hg calculated [ng/m³] Fig. 10. Scatter plot and regression analysis comparing calculated and observed 6 h average Hg concentrations during the November and December 1998 at Zingst.
This work was part of the Environment and Climate project Mercury Species over Europe (MOE) (Contract ENV4-CT97-0595) and of the EU European-LandOcean Interaction Studies (ELOISE). MOE was funded
ARTICLE IN PRESS S62
S.R. Schmolke, G. Petersen / Atmospheric Environment 37 Supplement No. 1 (2003) S51–S62
by the European Commission DG XII. The authors wish to acknowledge the financial support for this work which was provided by the International Bureaus of the GKSS Forschungszentrum Geesthacht (GKSS) and the Deutsche Forschungsanstalt fur . Luft-und Raumfahrt (DLR) in the framework of the Germany/Canada Science and Technology Co-operation Agreement (Project No. CAN 99/002). The authors express their appreciation of the peer reviewers’ suggestions for improving the manuscript.
References Berdowski, J., Baas, J., Bloos, J.J., Visschedijk, A.J.H., Zandveld, P.Y.J., Bousardt, F.L.M., 1997. The European atmospheric emission inventory of heavy metals and persistent organic pollutants. UBA/TNO, June 1997, CD-ROM. Bullock Jr, O.R., Benjey, W.G., Keating, M.H., 1997. The modeling of regional-scale atmospheric mercury transport and deposition using RELMAP. In: Baker, J.E. (Ed.), Atmospheric Deposition of Contaminants to the Great Lakes and Coastal waters. SETAC Press, Pensacola, pp. 323–347. Ebinghaus, R., Kock, H.H., Jennings, S.G., McCartin, P., Orren, M.J., 1995. Measurements of atmospheric mercury concentrations in northwestern and Central Europe-comparison of experimental data and model results. Atmospheric Environment. 29, 3333–3344. ERT, 1984. ADOM/TADAP Model Development Program, Vols. 1–8, ERT No. P-B980-535, July 1984. Environmental Research and Technology, Inc. Newbury Park, CA 91230. Lee, D.S., Dollard, G.J., Pepler, S., 1998. Gas-phase mercury in the atmosphere of the United Kingdom. Atmospheric Environment 32, 855–864. Lee, D.S., Nemitz, E., Fowler, D., Kingdon, R.D., 2001. Modeling atmospheric mercury transport and deposition across Europe and the UK. Atmospheric Environment 35, 5455–5466. Munthe, J., 2001. Mercury species over Europe. Relative importance of depositional methylmercury fluxes to various ecosystems (MOE). Contract ENV4-CT97-0595 of the European Commission, Directorate General XII Science, Research and Development, Environment and Climate Programme 1994–1998. Final report. Pacyna, E.G., Pacyna, J.M., Pirrone, N., 2001. European emissions of atmospheric mercury from anthropogenic sources in 1995. Atmospheric Environment 35, 2987–2996.
Pai, P., Karamchandani, P., Seigneur, C., 1997. Simulation of the regional atmospheric transport and fate of mercury using a comprehensive Eulerian model. Atmospheric Environment 31, 2717–2732. Petersen, G., Iverfeldt, A., Munthe, J., 1995. Atmospheric mercury species over Central and Northern Europe Model calculations and comparison with measurements from the Nordic air and precipitation network for 1987 and 1988. Atmospheric Environment 29, 47–67. Petersen, G., Munthe, J., Pleijel, K., Bloxam, R., Kumar, A.V., 1998. A comprehensive Eulerian modeling framework for airborne mercury species: development and testing of the tropospheric chemistry module (TCM). Atmospheric Environment 32, 829–843. Petersen, G., Bloxam, R., Wong, S., Munthe, J., Kruger, . O., Schmolke, S.R., Kumar, A.V., 2001. A comprehensive Eulerian modeling framework for airborne mercury species: model development and applications in Europe. Atmospheric Environment 32, 829–843. Pirrone, N., 1998. Modelling the dynamics of atmospheric mercury over the Mediterranean Sea: the MAMCS project. Journal of Aerosol Science 29, 1155–1156. Ryaboshapko, A., Ilyin, I., Gusev, A., Afinogenova, O., 1998. Mercury in the atmosphere of Europe: Concentrations, deposition patterns, transboundary fluxes. EMEP Meteorological Synthesizing Center-East, EMEP/MSC-E Report 7/98. Schmolke, S.R., Schroeder, W.H., Kock, H.H., Schneeberger, D., Munthe, J., Ebinghaus, R., 1999. Simultaneous measurements of total gaseous mercury at four sites on a 800 km transect: distribution and short-time variability of total gaseous mercury over central Europe. Atmospheric Environment. 33, 1725–1733. Schroeder, W.H., Barrie, L.A., 1998. Is mercury input to Polar ecosystems enhanced by springtime ozone depletion chemistry? IGACtivities Newsletter of the International Global Atmospheric Chemistry Project, Issue No. 14, September 1998. Schroeder, W.H., Anlauf, K.G., Barrie, L.A., Lu, J.Y., Steffen, A., Schneeberger, D.R., Berg, T., 1998. Arctic springtime depletion of mercury. Nature 394, 331–332. Seigneur, C., Karamchandani, P., Lohman, K., Vijayaraghavan, K., 2001. Multiscale modeling of the atmospheric fate and transport of mercury. Journal of Geophysical Research 106 (D21), 27795–27809. Shannon, J.D., Voldner, E.C., 1995. Modeling atmospheric concentrations of mercury and depositions to the Great Lakes. Atmospheric Environment 29, 1649–1661.
Atmospheric Environment 37 Supplement No. 1 (2003) S51–S62
A comprehensive Eulerian modeling framework for airborne mercury species: comparison of model results with data from measurement campaigns in Europe S.R. Schmolkea, G. Petersenb,* a
Institute of Ecology and Environmental Chemistry, University of Lueneburg, Scharnhorststrasse 1, Lueneburg D-21332, Germany b GKSS Research Centre, Institute of Coastal Research, System Analysis and Modelling, Max-Planck-Strasse, Geesthacht D-21502, Germany
Abstract Previously, we have simulated the atmospheric transport and fate of mercury emissions in Europe and derived estimates of ambient concentrations and dry and wet deposition of mercury species using a mercury version of the Acid Deposition and Oxidants Model (ADOM). In this study, we focus on comparisons of model results against observations from European measuring campaigns performed between 1995 and 1999 in the framework of the Germany–Canada Science and Technology Co-operation Agreement and the Mercury Species over Europe (MOE) project funded by the European Commission. Simulations were conducted using 1997/1998 meteorology and 1990 emission inventories. Model simulations show a good performance in explaining the peak mercury concentrations on a 800 km south to north transect, which were represented by simultaneous measurements at four sampling sites. Regional concentration gradients and differences in the concentration variability are well reproduced by the model even on a time base of 1 h. r 2003 Published by Elsevier Science Ltd. Keywords: Mercury; Numerical modeling; Eulerian models; Model validation; Total gaseous mercury
1. Introduction Efforts have been made to simulate the atmospheric transport and fate of mercury species and to derive estimates of ambient concentrations and dry and wet deposition fluxes of mercury over North America (Shannon and Voldner, 1995; Bullock et al., 1997; Pai et al., 1997; Seigneur et al., 2001) and Europe (Petersen et al., 1995; Ryaboshapko et al., 1998; Pirrone, 1998; Lee et al., 2001) through either relative simple Lagrangian formulations or Eulerian approaches employing extensive gas and aqueous phase chemical mechanisms and tracking explicitly numerous species concentrations. The starting point for this work was the existing Eulerian modeling framework of the Acid Deposition *Corresponding author.. E-mail address: [email protected] (G. Petersen).
and Oxidants Model (ADOM) originally designed for regional-scale acid precipitation and photochemical oxidants studies. Modules of ADOM have been restructured to accommodate most recent developments in atmospheric processes of mercury. These processes include transport by three-dimensional flows over a domain with horizontal scales of a few thousand kilometers and vertical scales of a few kilometers, transformation by gas and aqueous phase chemistry, scavenging by cloud processes and interactions of gaseous and particulate species with the ground. Fig. 1 shows schematically the various components of the ADOM model modified for transport, transformations and deposition of mercury. Two different versions of the model exist: a European 76 76 grid domain and a North American 33 33 grid domain version with a grid cell size of approximately 55 55 and 135 135 km2, respectively. The vertical grid, with 12
1352-2310/03/$ - see front matter r 2003 Published by Elsevier Science Ltd. doi:10.1016/S1352-2310(03)00250-4
ARTICLE IN PRESS S52
S.R. Schmolke, G. Petersen / Atmospheric Environment 37 Supplement No. 1 (2003) S51–S62 METEOROLOGICAL PRE-PROCESSOR EUROPE: HIRLAM High Resolution Limited Area Weather Prediction Model NORTH AMERICA: CMC Canadian Meteorological Center Large-Scale Model
INPUT DATA
POLLUTANT TRANSPORT MODEL (ADOM)
emission data bases
transport and diffusion
upper and surface meteorology
dry deposition
geophysical data
gas-phase chemistry
initial conditions boundary conditions secondary pollutants
wet scavenging
cloud physics
aqueous phase chemistry
Fig. 1. The ADOM model system for mercury. Relation between ADOM modules and input parameters.
unevenly spaced levels between 0 and 10 km, is identical for both versions and is designed to resolve the higher concentration gradients in the boundary layer. The basic model time step is 1 h. Horizontal and vertical wind fields along with the eddy diffusivity, temperature, humidity, surface precipitation, and information about the distribution of clouds make up the meteorological input data set. The data set is derived diagnostically using the weather prediction model HIRLAM for Europe and the Canadian Meteorological Center’s model for a North American version of the model. Other than meteorological data, the input and data requirements for ADOM are emissions, initial and boundary conditions, geophysical data and fields of other air pollutants relevant for chemical reactions with mercury species such as ozone and elemental black carbon (soot). European emission data for mercury were derived from the UBA/TNO inventory for toxic substances in Europe (Berdowski et al., 1997). These inventories were compiled for the reference year 1990. An emission update for 1995 including mercury speciation and emission heights for various source categories has become available now (Pacyna et al., 2001) and comparative model runs using the inventories for 1990 an 1995 are now underway. Wet scavenging involves modules for handling cloud physics and aqueous phase chemistry. Clouds are classified as stratus (layer clouds) or cumulus (convective clouds) according to the diagnostic output from the
weather prediction model. Observations of the fractional coverage and the vertical extent of clouds are combined with output from the diagnostic model to yield the input fields used in this module. A detailed description of the entire ADOM modeling framework can be found in ERT (1984). The development and testing of the cloud physics and mercury chemistry module considered to be the core part of the ADOM model system for mercury and the entire model system and first applications are described in Petersen et al. (1998) and Petersen et al. (2001), respectively.
2. Comparison of model results against observations Model evaluation is a key consideration when developing new and advanced comprehensive models. Since the ADOM mercury version was intended to be applied to simulations with policy implications, thorough validation and verification of the model and its components is a requirement. A model developed or utilized without continual comparison against actual data is less than worthless, it is dangerous. Such a model is nothing more than a collection of mathematical formulae, no matter how elegant its formulation or how efficient its coding. Only with the close integration of state-of-science models and state-of-science experimental measurements real progress can be made towards the solution of the complex problems that are currently
ARTICLE IN PRESS S.R. Schmolke, G. Petersen / Atmospheric Environment 37 Supplement No. 1 (2003) S51–S62
S53
faced, i.e. the continual interplay between conceptual understanding and experimental evidence. High quality monitoring data for mercury air concentrations and deposition in Europe are still limited, although several recent investigations have improved the database (Ebinghaus et al., 1995; Lee et al., 1998; Munthe, 2001; Pirrone, 1998; Schmolke et al., 1999). Fortunately, a few studies of total gaseous mercury (TGM) concentration measured simultaneously in Germany and Sweden are available now for evaluation of model performance. These data, although limited to four sites in Germany and Sweden (Fig. 2) and to a few 1–2 months measurement periods provide an opportunity for comparison with model predicted concentrations. A detailed description and a full documentation of the comparison of model results against observations is provided in the final report of the project ‘‘Mercury Species over Europe’’ funded by the EC (Munthe, 2001). This section is restricted to some illustrative examples concerning episodes of high and low mercury concentrations in Europe.
German atmospheric monitoring stations. The distance between the most northern and most southern site was approximately 800 km (Fig. 2). Details about the applied automated mercury measurement techniques are provided in Schmolke et al. (1999). Measurements were taken during two sampling periods: A 2 weeks early spring campaign during March 1995 (automated measurements at all four sites), and a winter period during November/December 1998 (automated measurements at all sites except Roervik). During the winter measurements at Roervik only a few data were collected by manual enrichment of ambient air mercury on gold traps and later thermodesorption and detection by AFS in the laboratory. The applied measurement techniques are generally accepted to detect TGM, which is an integral parameter over all gaseous mercury species. In addition to the simultaneous data at the four sites a full 2 month (November and December 1998) 0.5 h time resolved TGM data set measured at Zingst was available and used for some additional investigations.
2.1. Measurements in ambient air
2.2. Applications of the model to episodes of high mercury concentrations in Central Europe
From south to north transect atmospheric mercury measurement studies performed in Central and Northern Europe were used for this model evaluation exercise. The sampling was performed at two Swedish and two
In Fig. 3 1 h averages of model predicted mercury concentrations in ambient air are compared with observations at four sampling sites, namely Neuglobsow,
Fig. 2. Location of atmospheric monitoring sites, involved in the evaluation of model performance. The detail map of the model domain displays the major atmospheric mercury source areas. The 1990 UN-ECE/OSPARCOM/HELCOM emission source strength are indicated in tons/year.
ARTICLE IN PRESS S54
S.R. Schmolke, G. Petersen / Atmospheric Environment 37 Supplement No. 1 (2003) S51–S62
Fig. 3. Observed 1 h average total gaseous mercury (TGM) concentration values (black area) superimposed by calculated elemental mercury concentrations time series (gray Line). The displayed time window covers the entire November/December 1998.
Zingst, Aspvreten and Roervik. As already mentioned above, all observed data except Roervik are available with high time resolution during 2 weeks. The 1 h average TGM observations are indicated by black needles, whereas the discontinuously measured Roervik data are represented by shaded bars. Each individual bar is spanning over the sampling time and indicating the average concentration. Superimposed to the measured
data, the model predicted hourly Hg0 concentration time series are plotted as gray lines. We decided to neglect the oxidized gaseous mercury species in our model evaluation exercise due to their minor fraction of only 0.470.3% (maximum 2.3%) of total gaseous species at the southern most site, and even a decreasing trend and variability with increasing distance to the major source regions. Thus the subsequent
ARTICLE IN PRESS S.R. Schmolke, G. Petersen / Atmospheric Environment 37 Supplement No. 1 (2003) S51–S62
discussion will compare modeled Hg0 concentration with measured TGM concentration. Whereas the project related measurements were conducted only during approximately 2 weeks, an entire 2 month TGM data set was available at the Zingst monitoring site operated by the German Environmental Protection Agency. To summarize the ambient air mercury concentrations at the four sites, and for direct comparison of the calculated and observed mercury 1 h averages, basic descriptive statistical indices were calculated and are listed in Table 1. The considered time window is restricted to the periods of available simultaneous measurements during 18 March to 1 April 1997 and 23 November to 7 December. One should note the episodic nature of the observed and model predicted values at the two German sites spanning more than a three-fold range of concentrations from 1.2 to approximately 7 ng m 3. A few observations lower than 1 ng m 3 were made during 2 and 6 December at Neuglobsow. In general the lower end of the range is representative of hemispheric background concentrations of about 1.5 ng m 3 (Ebinghaus et al., 1995; Lee et al., 1998), whereas the concentrations at the upper end of the range are most probably due to longrange transport from anthropogenic sources in Central Europe. The very low concentration events of o1 ng m 3 are possibly affected by long-range transport of arctic air masses, or due to inflow of stratospheric air parcels. Two particular events of simultaneously elevated ambient air mercury concentration in the range between 2.5 to 5.5 ng m 3 at all four sampling sites during 25 November and 4 and 5 December 1998, which were
S55
detected by the measurements and also simulated by the model, are good examples to examine the model performance. Not only the temporal occurrence of the peak events, but also the concentration levels are in a similar range except Aspvreten on 25 November where the measurements do not support the model predicted peak. The wind fields during that day has been examined by means of 72 h back trajectories, which have been derived from the horizontal meteorological input data at approximately 250 m height. These trajectories describe the route taken during the previous 3 days by the air masses arriving at the four sites. Fig. 4 shows 24 trajectories, each with 1 h delay to the next, during 01 a.m. to 12 p.m. on 25 November 1998. Within the graph major source regions are indicated by source strength proportional black shading on the underlying map. Originating in Romania and the Ukraine air masses move along the trajectories into the main emission areas in Southern Poland and Eastern Germany, where they change their directions and move northwards arriving at the four sites. Trajectories which are attached to the modeled peak events at the individual sampling sites during 25 November are plotted in bold. Whereas the air parcels, which were responsible for the significant peak concentrations at the three southern most sites, were clearly influenced by the hot spot emission area in Eastern Germany (Figs. 4b–d), the backward trajectories reaching Aspvreten only pass the north eastern vicinity of this area. Thus, some uncertainties in the location/strength of the touched emission area or in the wind fields may be responsible for the model/measurement inconsistency at the northern most site during the discussed event.
Table 1 Summary statistics of observed and calculated ambient air mercury concentration during Spring 1997 and Winter 1998 at four atmospheric monitoring stations Neuglobsow Calc.
Zingst Obs.
Roervik
Aspvreten
Calc.
Obs.
Calc.
Obs.
Calc.
Obs.
18 March, 01:00–1 April, 12:00 1997 Arit. Mean 2.02 2.14 Median 1.76 2.03 S.D. 0.63 0.36 Maximum 4.10 4.03 Minimum 1.42 1.62 Count 324 319
1.76 1.62 0.39 3.45 1.34 324
2.13 2.07 0.31 3.78 1.71 324
1.39 1.35 0.16 2.41 1.23 324
1.93 1.93 0.20 2.87 1.53 321
1.41 1.36 0.24 2.79 1.26 324
1.77 1.78 0.24 4.81 1.38 302
23 November, 13:00–7 December, 09:00 1998 Arit. Mean 2.95 2.06 Median 2.65 1.99 S.D. 1.20 0.87 Maximum 6.04 6.68 Minimum 1.46 0.33 Count 333 333
2.83 2.57 1.11 5.45 1.33 333
2.64 2.43 0.86 7.02 1.53 333
1.86 1.76 0.49 3.45 1.21 333
1.60 1.52 0.23 2.10 1.45 8
1.61 1.56 0.26 2.37 1.30 333
1.69 1.62 0.26 2.69 1.21 333
ARTICLE IN PRESS S56
S.R. Schmolke, G. Petersen / Atmospheric Environment 37 Supplement No. 1 (2003) S51–S62
Fig. 4. 72 h backward trajectories 25 November 1998 calculated for the receptor areas: (a) Aspvreten, (b) Roervik, (c) Zingst, (d) Neuglobsow. Trajectories which are attached to the peak concentration during 25 November at the individual sampling sites are plotted in bold.
Another example of coinciding peak events at the four sampling sites during spring 1997 is depicted in Fig. 5. The different subdivision of the time axis should be noted. In contrast to Fig. 3 major tick marks denote 24 h, and minor 12 h time intervals, that is to say, each vertical line indicates 12 h. Elevated levels during 26 March 1997 are quite well reproduced by the model at the four sites. Maximum concentrations have been observed and calculated around noon at Neuglobsow, Zingst and Roervik. The model predicted a second smaller peak some hours later in the evening. The observation at Neuglobsow, and less pronounced at Zingst and Roervik as well, showed clearly the same pattern. The Aspvreten maximum is occurring a couple of hours later and the slight evening peak is missing at all. These phenomena are supported by the routes of the trajectories on 26 March (Fig. 6). During the first-half of
the day air masses start over Poland and adjacent areas, traverse East Germany picking up high emission rates and eventually arrive at the four sites. In accordance with Fig. 4 this graph also highlights trajectories which are related to the calculated peak concentrations. Except Aspvreten, trajectories are changing their routes significantly during the afternoon originating over the Western Atlantic and subsequently passing over emission areas in Britain and West Germany. Evidently, the routes of these trajectories cause the second smaller concentration peaks. The overall good agreement between observed and model predicted peak concentrations during episodes when trajectories exhibit the above described travel pattern is an indication that the model has capabilities to simulate atmospheric transport over distances of several hundred kilometers and that the emission inventory used with the
ARTICLE IN PRESS S.R. Schmolke, G. Petersen / Atmospheric Environment 37 Supplement No. 1 (2003) S51–S62
S57
Fig. 5. Observed 1 h average TGM concentration values (black area) superimposed by calculated elemental mercury concentrations time series (gray Line). The Time window covers 2 weeks in the end of March 1997.
model is based on realistic emission estimates for Central Europe in 1998. 2.3. Applications of the model to episodes of low mercury concentrations in Northern Europe In terms of comparing measurements with model output at all sites, there is evidence that the model has a tendency to underestimate observations during low
concentration and low variability episodes probably due to missing natural and re-emission processes in the model at present. Both, observed and model predicted time series are characterized by low temporal variability except the high concentration episode described in the previous section and except some low concentration episodes typically occurring over a time period of 12 h or less. Of particular interest in Fig. 5 are the four events of observed low concentration at the northern most site
ARTICLE IN PRESS S58
S.R. Schmolke, G. Petersen / Atmospheric Environment 37 Supplement No. 1 (2003) S51–S62
Fig. 6. 72 h backward trajectories 26 March 1997 for: (a) Aspvreten, (b) Roervik, (c) Zingst, (d) Neuglobsow. Trajectories which are attached to the peak concentration during 26 March are plotted in bold.
Aspvreten on 22, 23, 24 and 28 March 1997, which are not reproduced by the model. The related trajectories in Fig. 7 are either originating from Finland and Northern Russia or they are pointing towards ice-covered regions of the Arctic Ocean as a potential source of mercury depleted air masses during that time of the year. Indeed, mercury depletion phenomenon have been observed during the 3 months period following polar sunrise in March (Schroeder et al., 1998; Schroeder and Barrie, 1998) and the observed episodic minimum concentrations in the Aspvreten time series may be an indication for transport of air masses from polar regions. The current model version is not able to reproduce these effects which are due to particular chemical processes occurring outside of the model domain. However, the database is too scarce to draw any firm conclusions and more work in terms of using a model approach that can take into account the global cycling of mercury is
required to identify this phenomenon with more confidence (Seigneur et al., 2001). On the other hand, these regional simulations must rely on boundary conditions that may influence the model results. It is necessary therefore to develop a multiscale modeling approach, that consists of the existing regional scale ADOM model for mercury and a global chemical transport model, which provides the time-dependent boundary concentrations for the ADOM model.
3. Evaluation of the model performance Subject of the previous section was the discussion of some particularly high and low concentration episodes due to different meteorological conditions. Now some more general aspects of the model capability to reproduce ambient mercury concentrations pattern will
ARTICLE IN PRESS S.R. Schmolke, G. Petersen / Atmospheric Environment 37 Supplement No. 1 (2003) S51–S62
S59
Fig. 7. 72 h backward trajectories in Aspvreten on: (a) 22 March 1997, (b) 23 March 1997, (c) 24 March 1997, and (d) 28 March 1997.
be discussed. The box and whisker plot (Fig. 8) gives an aggregated overview about the modeled and measured 1 h average concentrations at the four selected sampling sites and during the previously discussed time window. Each box is indicating the minimum and maximum concentration (whisker), the inner quartile range (box), the median concentration (central symbol), and the confidence interval of the median (notch). The boxes are arranged in four pairs from the southernmost site at the left end to the northernmost site at the right end of the X-axis. It is important to note that the distance to the major mercury source regions in Eastern Germany increases on the south to north transect. Each pair of boxes represent one site, each left-hand box summarizes the modeled data, the right hand box the observations. The time window of observations is restricted to approximately 2 weeks during both experiments, when measurements at all sites were performed. The model predicts significantly decreasing concentrations of
approximately 0.5 ng m 3 on the south to north transect during the spring period (lower part of the graph) and 1 ng m 3 during the winter period. There is evidence for increasing median concentrations with decreasing distance to the major source regions. This finding is in good agreement with the observations. Except the Neuglobsow winter measurements, which show systematically lower absolute concentrations, these measurements also display a qualitative and absolute similar gradient, however with a tendency of the model to underestimate the atmospheric mercury concentration during springtime. Not only the regional pattern but also the frequency distribution of observed concentrations at each site are well reproduced by the model. Both, the calculated and measured data reflect a tendency of increasing signal variability (box extension) with decreasing distance to the major sources regions. Despite the systematically lower absolute concentration of the Neuglobsow winter data the frequency distribution of
ARTICLE IN PRESS S.R. Schmolke, G. Petersen / Atmospheric Environment 37 Supplement No. 1 (2003) S51–S62
S60 7
calc obs
6
Hg[ng/m³]
5
Nov. 23 - Dec. 07, 1998 4 3 2 1 7 6
March 17 - 31, 1997
Hg[ng/m³]
5
and measured time series. The middle and top part of the graph gives the residuals of the original time series and the 24 h running average curve which show a significant 24 h component. The major characteristics of the measured time series are sufficiently well reproduced by the model. Not only the major peaks but also details in the shape of certain peak events are simulated surprisingly well. However, the model has a tendency to overmodulate the signals. Fig. 10 represents a scatter plot of measured versus simulated data. The slope of 0.59 of the included regression curve supports this finding. Low concentrations tend to be underestimated and conversely, peak events are overestimated by the model. The linear regression analysis gives a coefficient of determination (R2 ) of 0.688 which is an encouraging number with respect to the model performance.
4 3
4. Summary and conclusions
2 1 0 neu
zin
roe
asp
Fig. 8. Campaign separated box statistics. Condensed view on the calculated and observed Hg concentration during the period 23 November–7 December 1998 (top) and 17–31 March 1997 (bottom).
simulated and observed data is similar. Based on the available data material a final judgement whether the effect of unexpected low measured concentration at the southern most site Neuglobsow is due to a calibration/instrumental error of the measurements or an overestimation of the model is not possible. However, the backward trajectory analysis previously discussed indicates that measurements might be too low. Finally, a value by value comparison of the measured and modeled time series is performed on the basis of the 2 winter month 1998 data set at Zingst. The following considerations were made concerning the direct comparability between model simulations and measurements. Highly time resolved field measurements always contain a short time random component which is due to local effects in the vicinity of sampling sites such as irregular car traffic, ship emissions and local heating. A 24 h cycle process, which is driven by diurnal meteorological conditions is often an additional part of atmospheric trace substance concentration time series. Focus of this investigation are transport distribution processes which act on a longer time scale. To filter out short time effects, a 24 h moving average procedure was applied. Fig. 9 (lower part) shows the smoothed curves of the modeled
A comprehensive model for airborne mercury species was tested by applying it to episodes of high and low mercury concentrations in ambient air in Central and Northern Europe, respectively. Time series of Hg0 estimated by the model were compared on an hourly basis with observed concentrations of TGM. The evaluation data material consists of 1 h average total gaseous mercury concentrations time series on a 800 km south to north transect, consisting of simultaneous measurements at four sampling sites. The model evaluation led to the following issues: The model prediction of regional Hg0 concentrations in ambient air is in good agreement with observed time series. Regional gradients, characterized by source strength and density, are well reproduced not only qualitatively but also quantitatively. High Hg0 concentration episodes could be traced back to specific source regions. The pattern of individual peak events originating from different source areas is well reproduced. These findings suggest that the implementation of the transport algorithms and emission estimates are on a satisfying level. However, the model shows a tendency to overmodulate high and low concentration episodes. Moreover, the model underestimates the average concentration level in Northern Europe during spring conditions at continuos northern inflow conditions. The reasons are not yet clear, but there are some indications that these effects could be attributed to overestimated point source strength in the 1990 emission data base and missing diffuse sources like reemission from soil and water surfaces. The application of a recently available Hg emission data base for 1995 may help to clarify this point. More information about air/surface exchange fluxes of gaseous mercury species are needed to
ARTICLE IN PRESS
Hg [ng/m³]
Hg [ng/m³]
S.R. Schmolke, G. Petersen / Atmospheric Environment 37 Supplement No. 1 (2003) S51–S62
S61
1.0 0.0 -1.0 1.0 0.0 -1.0 6 calcualted observed
Hg [ng/m³]
5 4 3 2 1
0
96
192 288 384 480 576 672 768 864 960 1056 1152 1248 1344 1440
hour (data origin Nov.01. 1998 01:00) Fig. 9. Decomposed 60 days Mercury time series, measured during November and December 1998 at the Zingst air quality monitoring site. Bottom: Observed and calculated 24 h moving average concentration. Middle: Short-term component (residuals) of the calculated series. Top: Short-term component of the observed time series.
implement these kind of processes into the model system. Additionally, an improvement of the inflow boundary conditions has to be considered in the future. The systematic underprediction of mercury concentrations at the northern inflow boundaries points to this issue. Also, the measured mercury depletion at the northern most site is thus an issue of hemispheric or global transport. This process, which may be due to atmospheric chemical processes outside of the model domain, cannot be reproduced by the model. Consequently, nesting of the regional ADOM model into a hemispheric model system should be one of the future activities to combine the advantage of high spatial resolution of regional models with a realistic integration of meteorological transport and chemical processes of a global pollutant.
6
Hg observed [ng/m³]
5
4
3
2
obs = 0.59 * calc + 0.76
1
R²=0.688 0 0
1
2
3
4
5
6
Acknowledgements
Hg calculated [ng/m³] Fig. 10. Scatter plot and regression analysis comparing calculated and observed 6 h average Hg concentrations during the November and December 1998 at Zingst.
This work was part of the Environment and Climate project Mercury Species over Europe (MOE) (Contract ENV4-CT97-0595) and of the EU European-LandOcean Interaction Studies (ELOISE). MOE was funded
ARTICLE IN PRESS S62
S.R. Schmolke, G. Petersen / Atmospheric Environment 37 Supplement No. 1 (2003) S51–S62
by the European Commission DG XII. The authors wish to acknowledge the financial support for this work which was provided by the International Bureaus of the GKSS Forschungszentrum Geesthacht (GKSS) and the Deutsche Forschungsanstalt fur . Luft-und Raumfahrt (DLR) in the framework of the Germany/Canada Science and Technology Co-operation Agreement (Project No. CAN 99/002). The authors express their appreciation of the peer reviewers’ suggestions for improving the manuscript.
References Berdowski, J., Baas, J., Bloos, J.J., Visschedijk, A.J.H., Zandveld, P.Y.J., Bousardt, F.L.M., 1997. The European atmospheric emission inventory of heavy metals and persistent organic pollutants. UBA/TNO, June 1997, CD-ROM. Bullock Jr, O.R., Benjey, W.G., Keating, M.H., 1997. The modeling of regional-scale atmospheric mercury transport and deposition using RELMAP. In: Baker, J.E. (Ed.), Atmospheric Deposition of Contaminants to the Great Lakes and Coastal waters. SETAC Press, Pensacola, pp. 323–347. Ebinghaus, R., Kock, H.H., Jennings, S.G., McCartin, P., Orren, M.J., 1995. Measurements of atmospheric mercury concentrations in northwestern and Central Europe-comparison of experimental data and model results. Atmospheric Environment. 29, 3333–3344. ERT, 1984. ADOM/TADAP Model Development Program, Vols. 1–8, ERT No. P-B980-535, July 1984. Environmental Research and Technology, Inc. Newbury Park, CA 91230. Lee, D.S., Dollard, G.J., Pepler, S., 1998. Gas-phase mercury in the atmosphere of the United Kingdom. Atmospheric Environment 32, 855–864. Lee, D.S., Nemitz, E., Fowler, D., Kingdon, R.D., 2001. Modeling atmospheric mercury transport and deposition across Europe and the UK. Atmospheric Environment 35, 5455–5466. Munthe, J., 2001. Mercury species over Europe. Relative importance of depositional methylmercury fluxes to various ecosystems (MOE). Contract ENV4-CT97-0595 of the European Commission, Directorate General XII Science, Research and Development, Environment and Climate Programme 1994–1998. Final report. Pacyna, E.G., Pacyna, J.M., Pirrone, N., 2001. European emissions of atmospheric mercury from anthropogenic sources in 1995. Atmospheric Environment 35, 2987–2996.
Pai, P., Karamchandani, P., Seigneur, C., 1997. Simulation of the regional atmospheric transport and fate of mercury using a comprehensive Eulerian model. Atmospheric Environment 31, 2717–2732. Petersen, G., Iverfeldt, A., Munthe, J., 1995. Atmospheric mercury species over Central and Northern Europe Model calculations and comparison with measurements from the Nordic air and precipitation network for 1987 and 1988. Atmospheric Environment 29, 47–67. Petersen, G., Munthe, J., Pleijel, K., Bloxam, R., Kumar, A.V., 1998. A comprehensive Eulerian modeling framework for airborne mercury species: development and testing of the tropospheric chemistry module (TCM). Atmospheric Environment 32, 829–843. Petersen, G., Bloxam, R., Wong, S., Munthe, J., Kruger, . O., Schmolke, S.R., Kumar, A.V., 2001. A comprehensive Eulerian modeling framework for airborne mercury species: model development and applications in Europe. Atmospheric Environment 32, 829–843. Pirrone, N., 1998. Modelling the dynamics of atmospheric mercury over the Mediterranean Sea: the MAMCS project. Journal of Aerosol Science 29, 1155–1156. Ryaboshapko, A., Ilyin, I., Gusev, A., Afinogenova, O., 1998. Mercury in the atmosphere of Europe: Concentrations, deposition patterns, transboundary fluxes. EMEP Meteorological Synthesizing Center-East, EMEP/MSC-E Report 7/98. Schmolke, S.R., Schroeder, W.H., Kock, H.H., Schneeberger, D., Munthe, J., Ebinghaus, R., 1999. Simultaneous measurements of total gaseous mercury at four sites on a 800 km transect: distribution and short-time variability of total gaseous mercury over central Europe. Atmospheric Environment. 33, 1725–1733. Schroeder, W.H., Barrie, L.A., 1998. Is mercury input to Polar ecosystems enhanced by springtime ozone depletion chemistry? IGACtivities Newsletter of the International Global Atmospheric Chemistry Project, Issue No. 14, September 1998. Schroeder, W.H., Anlauf, K.G., Barrie, L.A., Lu, J.Y., Steffen, A., Schneeberger, D.R., Berg, T., 1998. Arctic springtime depletion of mercury. Nature 394, 331–332. Seigneur, C., Karamchandani, P., Lohman, K., Vijayaraghavan, K., 2001. Multiscale modeling of the atmospheric fate and transport of mercury. Journal of Geophysical Research 106 (D21), 27795–27809. Shannon, J.D., Voldner, E.C., 1995. Modeling atmospheric concentrations of mercury and depositions to the Great Lakes. Atmospheric Environment 29, 1649–1661.