Inversion Approach to Validate Mercury Emissions ... - ACS Publications

Article pubs.acs.org/est

Inversion Approach to Validate Mercury Emissions Based on Background Air Monitoring at the High Altitude Research Station Jungfraujoch (3580 m) Basil Denzler,† Christian Bogdal,*,† Stephan Henne,‡ Daniel Obrist,§ Martin Steinbacher,‡ and Konrad Hungerbühler† †

Institute for Chemical and Bioengineering, ETH Zurich, Vladimir-Prelog-Weg 1, CH-8093 Zürich, Switzerland Empa, Swiss Federal Laboratories for Materials Science and Technology, Ü berlandstrasse 129, CH-8600 Dübendorf, Switzerland § Desert Research Institute, Division of Atmospheric Sciences, 2215 Raggio Parkway, Reno, Nevada 89512, United States ‡

S Supporting Information *

ABSTRACT: The reduction of emissions of mercury is a declared aim of the Minamata Convention, a UN treaty designed to protect human health and the environment from adverse effects of mercury. To assess the effectiveness of the convention in the future, better constraints about the current mercury emissions is a premise. In our study, we applied a topdown approach to quantify mercury emissions on the basis of atmospheric mercury measurements conducted at the remote high altitude monitoring station Jungfraujoch, Switzerland. We established the source-receptor relationships and by the means of atmospheric inversion we were able to quantify spatially resolved European emissions of 89 ± 14 t/a for elemental mercury. Our European emission estimate is 17% higher than the bottom-up emission inventory, which is within stated uncertainties. However, some regions with unexpectedly high emissions were identified. Stationary combustion, in particular in coal-fired power plants, is found to be the main responsible sector for increased emission estimates. Our top-down approach, based on measurements, provides an independent constraint on mercury emissions, helps to improve and refine reported emission inventories, and can serve for continued assessment of future changes in emissions independent from bottom-up inventories.

■

INTRODUCTION Mercury is a heavy metal of particular concern due to its ability to accumulate in ecosystems in its organic forms, and its significant adverse effects on human health and the environment. Major anthropogenic releases of mercury to the environment result from atmospheric emissions by artisanal and small scale gold mining and combustion processes; mainly coal burning, cement production, and metallurgic processes.1 Three forms of mercury have to be considered for atmospheric emissions: gaseous elemental mercury (GEM), gaseous oxidized mercury, and particle bound mercury.2 GEM is the most common form of mercury in the atmosphere.3 Due to its high vapor pressure and its long residence time, GEM undergoes long-range atmospheric transport and is thus distributed globally. In 2011, the United Nations adopted the UNEP Minamata Convention on Mercury, with the objective to protect human health and the environment from anthropogenic emissions of mercury. So far, multiple studies have been carried out to quantify the anthropogenic contribution to mercury emissions on a global scale.4−9 All of them share the methodological basis of raising an inventory of © XXXX American Chemical Society

mercury-emitting activities, which are coupled with activityspecific emission factors to estimate the release of mercury from various sources, as for example coal fired power plants. We refer to this method as the bottom-up approach. Such inventories are intelligible and their universal applicability enables a worldwide coverage. However, appreciable uncertainty in the emission inventories is introduced by various assumptions on the activity data and emission factors. Since the bottom-up approach relies on officially reported emission data, a further bias and differences in the accuracy between countries and regions is not inconceivable. The alternative is a top-down approach, where emissions are calculated by combining atmospheric transport and chemistry models with atmospheric concentration measurements. This powerful technique, often referred to as atmospheric inversion, has the advantage of a more objective view on the emission Received: November 8, 2016 Revised: January 19, 2017 Accepted: January 20, 2017

A

DOI: 10.1021/acs.est.6b05630 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology

masses arriving at Jungfraujoch are backtracked in time for 10 days by FLEXPART using a dispersing plume of 50 000 infinitesimally small tracer particles. These undergo transport by the mean, turbulent and convective flow. This allows to establish the source-receptor relationship (SRR, unit: [s m3kg−1]) that gives the relation of a mass emission at a source onto the atmospheric mass mixing ratio at the receptor. In other words, the multiplication of the SRR with a spatial mass emission m of a tracer will result in the mixing ratio of the tracer at the receptor site. The SRR were calculated for the lowest 100 m above model ground, representing the influence of surface emissions. SRR output was stored on a regular longitude/latitude grid of 0.5° × 0.5° horizontal resolution. The model region covered by this grid ranges from −100° to 60.5° longitude and 15.5° to 75.5° latitude. Baseline Estimate. Simply put, the atmospheric GEM concentrations of our measurement locations are composed of three different components. First, the contribution by GEM emissions of inventoried regional anthropogenic sources inside the model region, whose fate is simulated by our transport simulation for the last 10 days prior to arrival at Jungfraujoch. As described, their contribution to the concentration at JFJ is calculated using the SRR. Second, there exists an additional term of uninventorised natural emissions and other unknown sources. And the third share of the measured GEM concentrations originates from distant sources and past emissions, not covered by the FLEXPART model region. We describe the latter contribution as baseline concentration, bl. To determine the contribution of the baseline concentration blprior to the GEM measurement series, a statistical method based on robust local regression was applied.22 The approach fits a smooth baseline curve to the observations, iteratively rejecting observations outside an uncertainty band and giving asymmetric weights to the remaining data. A smoothing window of 30 days and a number of 10 iterations were used here. Inversion Method. To deduce an emission inventory based on the measurement series via the top-down method, a Bayesian Inversion approach was used as introduced by Stohl et al.10 The application here is based on Henne et al.14 This technique provides a solution to the inverse of the following linear problem of estimating the spatial emissions m from the observations given the transport simulation as expressed in the sensitivity matrix G, which is obtained from the FLEXPART simulations, dobs = Gm + bl + error. bl thereby corrects for the error of the direct SSR. The problem is expressed in the following way:

associated side of the problem. Top-down approaches have been successfully applied over the past decade in the field of greenhouse gases10−12 and recently also in the field of mercury emissions.13 In such studies, uncertainties arise from simplifications in the transport model and depend on the quality of the atmospheric observations. Here, we present high-resolution measurements of atmospheric GEM concentrations taken at the remote high altitude monitoring station Jungfraujoch (JFJ), Switzerland. We used a dispersion model to track the origin of the sampled air masses and applied a Bayesian inversion approach to derive mercury emission on a European scale, based on our measured GEM data series. The distinction between different emission sectors allows the identification of the most prominent industrial sectors. With this study, we are able to evaluate existing mercury emission inventories and derive spatially resolved emission estimates based on atmospheric GEM measurements. Furthermore, indications on relevant single point sources become available. Thus, our data provide an important contribution to the ongoing discussions about the magnitude, the share, and the distribution of anthropogenic mercury emissions especially, when assessing the effect of the Minamata Convention on mercury emissions.

■

MATERIALS AND METHODS Measurement Site and Data. The measurements were conducted at the high altitude research station Jungfraujoch (JFJ) located in the Swiss Alps (7°59′ E, 46°32′ N) at 3580 m above sea level. It is situated mostly in the free troposphere but intermittently receives air masses that were in recent contact with the planetary boundary layer, increasing loads of anthropogenic pollutants toward the site.14−16 As such, the site is ideally placed to, on the one hand, monitor background air concentrations and, on the other hand, investigate regional and long-range transport of air pollutants.17 From April 2011 to April 2012, GEM concentrations were measured using a Tekran 2537A vapor phase mercury analyzer with 5 min temporal resolution. It uses active gold cartridge sampling and cold vapor atomic fluorescence spectroscopy (CVAFS) to quantify the GEM concentrations with a high sensitivity (0.1 ng m−3). Outside air was sampled from a heated and well ventilated inlet by PTFE tubing and drawn through a 0.2 μm particulate filter. Data quality was assured by manual calibration prior to the measurement campaign. Automated internal calibration was carried out every 25 h. No manual onsite calibration was conducted during the measurement period. The relative combined measurement uncertainty is estimated to be 10%.18−20 This uncertainty is an upper bound, especially since it was not reduced for 3 hly means. The achieved temporal coverage for the yearlong campaign is 74%. Additionally to GEM, an extended set of trace gases is continuously measured at JFJ as part of the operation of the Swiss National Air Pollution Monitoring Network (NABEL). Meteorological parameters are recorded by the Swiss Federal Office for Meteorology and Climatology (MeteoSwiss) within their SwissMetNet program. Backward Modeling. A Lagrangian particle dispersion model, FLEXPART, version 9.01,21 was used to investigate the atmospheric transport of GEM. The model is driven by windfield data of the European Centre for Medium-Range Weather Forecasts (ECMWF) with 91 vertical levels and a horizontal resolution of 1° × 1° globally and a resolution of 0.2° × 0.2° for the Alpine area. For every 3 hly interval, air

χ = Gx

(1)

where χ is the simulated mole fractions at the receptor and x is the state vector that combines the gridded emissions m and the baseline bl. The baseline blprior is thereby included in the optimization as a part of the state vector. As a constraint to the ill-conditioned inverse problem of guessing the emission pattern m, and a priori estimate mprior for the source vector is introduced. The a priori emissions were once compiled from the AMAP bottom-up emission inventory of 201023 and once from the EDGARv4.tox1 inventory of 2008.9 The emissions inventory Figure 2E was therefore reduced to the inversion grid. Only GEM emissions over all height classes (0 to >150 m) were used and GEM is treated as a conservative tracer undergoing no oxidation or deposition processes. A second guidance is given by the a priori knowledge about the baseline B


Article


Figure 1. GEM time series dobs (gray) in 3 hly resolution in comparison with a posteriori dapost (green) and a priori dprior (red) model results. blprior (blue) shows the baseline concentration prior to the inversion, blapost (violet) corresponds to the a posteriori baseline.

longitude and 28° to 62.5° latitude (Supporting Information (SI) Figure S1). To prevent any invalid results for the inversion, cells with negative emissions are eliminated in an iterative procedure as suggested by Stohl et al.,10 since no mercury sink regions were considered.

concentrations. The prior knowledge of the baseline concentration and the mercury emissions is expressed in the a priori state vector xb. The Bayesian least-squares estimate of the state vector x is then obtained as the vector that minimizes the objective cost function Jbls, penalizing deviations of x from xb and of simulated and observed mole fractions χobs: Jbls

■

RESULTS AND DISCUSSION Measurement Series. GEM concentrations at JFJ were measured with 5 min temporal resolution between April 2011 and April 2012. Observed 3 hly aggregates range from 0.9 and 1.9 ng/m3 with a median concentration of 1.30 ng/m3 and 10% and 90% quantiles of 1.15 and 1.48 ng/m3, respectively (Figure 1 and SI Figure S2). In comparison to other remote atmospheric measurement locations in Europe, the concentrations at JFJ resided in the lower range, lower than measurements from Mace Head, Ireland,24 but comparable to remote measurements at high latitude in Finland and Sweden.25 Higher concentrations have been measured at Pic du Midi Observatory, a measurement site at similar altitude in the Pyrenees at the border between France and Spain.26 Numerous events showing high GEM concentrations, exceeding the 90% quantile can be seen throughout the measurement period, mainly in spring 2011, September 2011, and in the winter months of 2012. The diel cycle does not show any pronounced pattern for GEM concentrations. However, the seasonal variability based on monthly medians reveals a late summer minimum in resemblance to carbon monoxide (CO), which is also an anthropogenic tracer as it is primarily emitted by the combustion of carbonaceous fuels (SI Figure S3). The seasonal correlation of GEM with CO are shown in SI Figure S4. The GEM pattern does not correlate with the seasonal cycle of ozone (O3), which is photochemically produced in the troposphere (SI Figure S3). Hereafter, we only focus on observed peak events (i.e., positive deviation from a time averaged baseline GEM concentration) and explain their relation to the baseline GEM concentration. Footprint. Modeled footprints, acquired by the Lagrangian particle dispersion model FLEXPART, show the residence time of the air masses above ground before arriving at the

1 = ((x − xb)T B−1(x − xb) + (Gx − χobs )T R−1 2 (Gx − χobs ))

(2)

the minimization leads to x = xb + BGT (GBGT + R )−1 ·(χobs − Gxb)

(3)

From x we obtain the gridded a posteriori emissions mapost. The covariance matrices B and R thereby represent the uncertainty for the a priori emissions and for the combined model and measurement uncertainty. Since mercury is predominantly emitted by large point sources, no off diagonal elements were considered for the a priori covariance matrix B to allow an independent adjustment of each grid cell. An uncertainty relative to the emission strength was assigned to each cell to meet a total uncertainty for the inversion domain of 20%. This relatively small uncertainty was chosen to closely guide the inversion and to avoid an overfitting of the observations and connected emission attribution errors. R was built as described by Stohl et al.,10 by iteratively using the root−mean−square error (RMSE) of the model simulations from preliminary inversions, again no off diagonal elements were considered. For the inversion, the number of cells covered needs to be reduced for two reasons: (i) increasing computational costs to solve eq 1 and (ii) the limited amount of independent information in the observations does not allow to constrain a large number of cells. Therefore, the model region used for the FLEXPART simulations is reduced to the area of interest with high SRR. Cells outside the given area are not inverted, however their emissions are included in the a priori simulations and subtracted from the observations before optimization. The inversion domain covers the area from −12.5° to 32.5° C


Article


Figure 2. A: a posteriori GEM emission map obtained by the Bayesian optimization from AMAP a priori with a total emission of 89 ± 14 t/a. B: a posteriori GEM emission map obtained by the Bayesian optimization from homogeneous a priori emissions with a total emission of 177 ± 31 t/a. C: difference between a priori and a posteriori emissions. Emission increase from a priori to a posteriori shown in red decrease in blue. D: difference between homogeneous a priori and a posteriori emissions. Emission increase from a priori to a posteriori shown in red decrease in blue. E: a priori emissions based on the GEM emissions of the AMAP mercury emission inventory over all height levels shown only for the inversion domain. For the inversion the emissions for the whole model region were used. F: Source receptor relationship (SRR) normalized by area of grid cell for inversion domain.

measurement location. It reveals over which geographical regions air masses were potentially affected by ground-based mercury emissions. A dominance of western wind inflow (SI Figure S1) is visible from the composite footprint of air mass trajectories based on the entire measurement series. Since all approaching air masses focus to the measurement location, highest residence times are obtained in the vicinity of JFJ and decline with increasing distance. Regions with longer residence time allow conclusions regarding the source-receptor relationship (SRR) (Figure 2D) with greater confidence. Therefore, the model region of the total FLEXPART footprint was cropped to an area of higher SRR as seen in SI (Figure S1). We refer to this

area as the inversion domain. To identify the source regions of GEM specific for JFJ, the footprints (H) for the events resulting in the maximum 10% of the measured 3 hly GEM aggregates were compared to the total footprint (T). Thereby, potential source regions for the measurement location were identified in northern Italy and southeastern Europe (SI Figure S5). Inversion Performance. When using the 2010 global AMAP/UNEP inventory23 prepared for the UNEP 2013 Global Mercury Assessment27 as a priori (mprior) to derive a calculated GEM time series (dprior) an astonishingly close match (R = 0.67, normalized root-mean-square NRMSE = 6.79) between dprior and the measurement series (dobs) is observed D


Article

Environmental Science & Technology (Figure 1 and SI Figure S6). Especially, the fluctuations of the positive peaks above the baseline of the measurement series are well represented by the calculated series. This finding substantiates the assumption that positive fluctuations in the measurement series are mainly caused by anthropogenic sources located in the catchment area of the Jungfraujoch monitoring station. However, the peak concentrations of dprior do not reach the measured levels. This leads to the hypothesis that the mercury emission fluxes in some areas of the spatial emission pattern of the AMAP inventory are underestimated. The information on the AMAP emission inventory was, therefore, used as the a priori set for a Bayesian Inversion to find an emission map (mapost) that produces a time series (dapost) with an even better match to the measured GEM series. For the Bayesian optimization, a 20% uncertainty for the total emission of the a priori domain was chosen. Hence, the match between (dapost) and the measurement series (dobs) is further improved (R = 0.73 and NRMSE = 6.32), (Figure 1). For points above the baseline an even better fit is observed (R = 0.76). The total GEM emissions of the inversion area now are 89 ± 14 t/a. This corresponds to an increase of 13 t/a, or 17% in GEM emissions, therefore, suggesting an underestimation of the emission strength of Europe by the AMAP inventory, which has a GEM emission strength of 76 t/a in the chosen area. The uncertainty for the a posteriori emissions was reduced by the inversion to 15%. This analytical uncertainty, however, has to be regarded as a minimum base value, which may be exceeded by the structural uncertainty of the inversion system. The uncertainty reduction could certainly be enhanced, provided a longer time series is available. The baseline determination does influence the outcome of the inversion and affects mainly the total amount of the emissions. Changing the smoothing span from 30 to 60 days only had a minor impact on the total emission estimate (+4%). Single cells, however, especially with low SRR can be affected to a bigger extent. Running the inversion with greater a priori uncertainty, results in higher emission estimate that converges to a maximum of 147 ± 42 t/a for 400% a priori uncertainty. As an alternative bottomup inventory, the EDGAR emission inventory9 was used in a second inversion. Applying the EDGAR a priori set of GEM emissions resulted in a very similar outcome (R = 0.73 and NRMSE = 6.29) with emissions changing from a priori 65 ± 13 t/a to 76 ± 12 t/a for the a posteriori (SI Figure S7). In the following two sections we discuss patterns in the emissions maps and single point sources. We thereby comment on the extremes of our findings. The SRR and the uncertainty range serve as guideline for the robustness in the predications. These sections highlight the power a tool like atmospheric inversion has and show an exemplary way of retrieving information from the top-down results. Emission Patterns. Both a priori (Figure 2E) and a posteriori (Figure 2A) emission patterns show large variations in emission strength between grid-cells. Few cells dominate the emissions; the 10% strongest emitting cells represent over 90% of the total emissions of the displayed area and also show the biggest absolute difference after the inversion. The highest emitting cells, which for the a priori emission inventory are located in Germany and Poland, are now found in Romania and Greece. This change is most clearly shown in Figure 2C, which displays the difference between the a priori and the a posteriori emission map. An increase in emissions for a cell from a priori to a posteriori is shown in red, decreases in blue. It reveals that according to our findings in Table 1, on a country level

Table 1. Annual GEM Emissions with Uncertainties (σ) by Countries in Order of Their Highest a Priori Emission Strength, Only Countries Covered Entirely by the Scope Are Considered emissions in [t/a] country Germany Poland Bulgaria Greece Spain France Italy Czech Republic United Kingdom Romania

a priori 11.0 6.3 4.5 4.4 4.2 3.7 3.3 3.1 3.0 3.0

a posteriori ±7.7 ±4.4 ±4.4 ±5.9 ±2.8 ±1.7 ±1.4 ±2.5 ±2.1 ±2.6

10.0 5.4 5.7 8.3 6.7 4.3 4.5 3.2 3.7 4.4

±6.9 ±4.2 ±4.3 ±5.3 ±2.6 ±1.6 ±1.1 ±2.4 ±2.1 ±2.5

Germany and Poland emit less mercury than initially predicted by the AMAP emission inventory, whereas the other countries emit more. In particular the emissions of Greece, Spain, and Romania were strongly adjusted upward by the inversion. Greece shows the highest difference in emissions. Emissions increased from 4.4 to 8.3 t/a (+89%). We hypothesize that this rise is in parts due to an underestimation of the emission strength of certain point sources in Greece and may also be caused by unknown sources, previously overlooked by the AMAP emission inventory. Natural sources contributing to the higher emission are an other possibility. Since Greece is situated in an area with reduced emission sensitivity and, therefore, transport uncertainties may be relatively large, further studies are needed for more confident conclusions. To test for the influence of the a priori knowledge on the distribution of the a posteriori emissions, we conducted an inversion assuming homogeneous distribution for the a priori emissions with an equal total amount to the AMAP inventory. Each grid cell was attributed the same emission rate for the a priori map. Total a posteriori emissions reach values of 177 ± 31 t/a when considering a 200% a priori uncertainty (Figure 2B). The performance of this inversion even exceeds the previous ones (R = 0.77 and NRMSE = 5.88). Despite a more diffuse distribution of emission (Figure 2B), areas of peaking emissions have arisen, which in many cases resemble the hotspots in the previous emission map in Figure 2A. Though, also other areas emerge, which are not considered as source regions in the AMAP emission inventory. These new source areas could either represent areas with anthropogenic emission not appropriately quantified in emission inventories or be of natural origin. Point Sources and Emission Sectors. The AMAP emission inventory distinguishes for each grid cell between three emission sectors: (i) stationary combustion, (ii) industrial sources, and (iii) intentional use and product waste associated processes. Since the inversion does not distinguish between different emission categories, their relative contribution in each individual grid cells remains unchanged by the inversion. However, applying the prior values of the relative contributions to the posterior inversion results, country specific contributions of each emission category may differ from the prior values. From the changes in the contribution pattern between the a priori and the a posteriori emissions, the sectors most affected by the inversion become evident (Figure 3). The share of Greek and Romanian emissions by combustion sources for E


Article


Figure 3. A posteriori (dark gray) and a priori mercury emissions (light gray) displayed by country in order of the a posteriori emission strength. The error bars show the uncertainty (σ) allocated to the emission inventory before and after inversion. The subdivision of the total a priori and a posteriori emissions into three different emission sectors according to the AMAP inventory is shown by color coding: stationary combustion (red), industrial emissions (green), intentional use and waste associated sectors (blue).

Table 2. Biggest GEM Point Sources According to E-PRTR for Cells with the Highest a Posteriori Emission Strength in Comparison to a Priori Value (± uncertinty σ); Activities: Thermal Power Plant (PP), Metallurgic Processes (M) and Chemical Industry (CI) center of grid cell longitude (°E)

latitude (°N)

21.75 21.25 13.25 −5.75 24.25 24.75 24.75 22.75 −2.75 23.25 −2.75

40.25 42.75 52.25 43.75 42.75 41.25 42.25 45.75 53.25 44.75 43.25

emissions in [t/a] a priori 1.9 1.8 2.5 0.7 1.0 0.9 0.9 0.6 0.6 0.4 0.3

±5.1 ±5.0 ±6.7 ±1.9 ±2.8 ±2.4 ±2.3 ±1.7 ±1.7 ±1.2 ±0.9

country

biggest point source listed

activity

Greece Kosovo Germany Spain Bulgaria Greece Bulgaria Romania United Kingdom Romania Spain

Agioy Dhimhtrioy Kosovo (not in E-PRTR) Berlin Arcelormittal

PP PP

Mintia (not in E-PRTR) Runcorn Halochemicals Turceni Nervacero S.A.

PP CI PP M

a post. 5.0 3.5 2.1 2.1 1.5 1.4 1.3 1.3 1.1 0.8 0.7

±4.6 ±4.7 ±5.9 ±1.7 ±2.7 ±2.2 ±2.3 ±1.6 ±1.6 ±1.1 ±0.8

example increased, whereas for Germany and Poland they decreased. An explanation for this could be found in the emission factors applied in the UNEP toolkit approach, used to establish the bottom-up inventory. To allow for regional differences in technology profiles associated with mercury emissions, five different regions were defined to construct the AMAP inventory. Germany and Poland, as well as Greece, are all allocated to group 1, the most state-of-the-art group, and are assumed to have similar characteristics. This assumption is reflected in our findings for Germany and Poland, which show a similar decline in the sector of stationary combustion. Greece on the other hand does not follow the same pattern, possibly due to differences in the technology profile regarding mercury emissions in this sector. For Spain, the increase of emissions for the a posteriori calculations is allocated primarily to cells where intentional use and waste incineration dominate. To identify single GEM point sources on the factory level contributing to the total emission of an inversion grid cell, we used total gaseous mercury (TGM) emission data provided by the European Pollutant Release and Transfer Register (E-

M

PRTR). The change in emissions from a priori to a posteriori of a cell is related to these single point sources. When looking at the cells with highest a posteriori emission (Table 2), some cells clearly stand out. For most of them coal fired power plants seem to be the biggest contributors. The situation in Greece, which was addressed before, is remarkable. Most of the increase of the Greek emissions are allocated to one cell (21.75° E, 40.25° N), where the coal fired power plant Agioy Dhimitrioy is the largest point source among two other power plants. On the basis of our inversion results, we conclude that the reported emissions of 1.01 t for the year 2011 may be considerably larger, + 163% at most (Table 2). The same applies for the other point sources, always given the inversion results are accurate and on the condition that no other unknown source is causing the discrepancy between a priori and a posteriori emissions. For a detailed understanding of its GEM emissions, Germany was chosen, since all major industrial sources are present in Germany and the strong SRR allows for more reliable conclusions than for more distant source areas. Additionally, comprehensive national reporting of mercury F



■

ACKNOWLEDGMENTS We acknowledge that the International Foundation High Altitude Research Stations Jungfraujoch and Gornergrat (HFSJG), 3012 Bern, Switzerland, made it possible for us to carry out our experiments at the High Altitude Research Station at Jungfraujoch. We also thank Joan and Martin Fischer and Maria and Urs Otz (HFSJG) for their support. We thank the Jungfraubahn for medial contributions. Matthew MacLeod (Stockholm University), Asif Qureshi (IIT Hyderabad) and Sandy Ubl (ETH Zurich) are acknowledged for initiating and supporting the GEM monitoring and modeling. Furthermore, we acknowledge the Swiss National Air Pollution Monitoring Network (NABEL) and the Swiss Federal Office for Meteorology and Climatology (MeteoSwiss) for providing CO measurements and meteorological data. At last, we thank the Swiss Federal Office for the Environment (FOEN) for the project funding (grant numbers 00.0248.P2/M371-4632, 14.0039.KP/N412-1043).

emission sources is available. Overall, we obtain slightly lower GEM emissions for Germany (−9%). In SI (Table S1) the grid cells with largest a posteriori emissions in Germany are compared to E-PRTR emission data for individual facilities. The sum of E-PRTR point sources for the corresponding grid cells show similar emissions compared to the emissions from the whole cell reported by the AMAP inventory. For power plants both, increasing and decreasing trends are observable. Some a posteriori results exceed the a priori emissions and the E-PRTR emissions, while others are below. In some cases these trends can be narrowed down to single plants as for example Kraftwerk Niederaussem (SI Table S1), which by far makes up for the biggest share in the corresponding cell and, therefore, likely accounts for the difference. Our results would imply emissions to be lower by 15% for this facility. For Akzo Nobel Ind. Chemicals on the contrary, emissions may be underestimated by 10%. Furthermore, large power plants in Germany mostly show lower emissions in the a posteriori results, cement plants show an opposite trend. This suggests an underestimation of emissions of combustion processes for cement production. Over the whole inversion domain the a posteriori emission estimates show higher overall emissions, which are caused by an increase in some specific areas, mainly in Greece, Spain, and Romania. Primarily combustion sources, particularly from coal combustion facilities, are found responsible for the higher emissions. However, discrepancies in this regard between countries of different industrial standards, which were treated identically for the bottom-up emission inventories, were found. These findings highlight some weaknesses of the UNEP tool-kit approach. A more detailed classification regarding technology profiles of countries for bottom-up inventories may be necessary. On the point source level, our results suggest under-reporting for some facilities for different industrial sectors. Our findings can be of help in setting up new focus points to investigate for bottom-up emissions and eventually doing on site measurements to verify these findings. Additional observations, especially in eastern Europe,28 would be helpful to obtain a better SRR coverage. Further investigations, comprising a combination of different measurement sites and longer measurement periods, may enable to draw final conclusions on the facility level. The future inclusion of seasonal influences, mercury sink processes, and a distinction between natural and anthropogenic sources, could contribute to an even more in depth understanding of anthropogenic mercury emissions and natural exchange processes.

■

■

REFERENCES

(1) Driscoll, C. T.; Mason, R. P.; Chan, H. M.; Jacob, D. J.; Pirrone, N. Mercury as a global pollutant: Sources, pathways, and effects. Environ. Sci. Technol. 2013, 47, 4967−4983. (2) Schroeder, W. H.; Munthe, J. Atmospheric mercury - An overview. Atmos. Environ. 1998, 32, 809−822. (3) Gay, D. A.; Schmeltz, D.; Prestbo, E.; Olson, M.; Sharac, T.; Tordon, R. The Atmospheric Mercury Network: measurement and initial examination of an ongoing atmospheric mercury record across North America. Atmos. Chem. Phys. 2013, 13, 11339−11349. (4) Pacyna, E. G.; Pacyna, J. M.; Steenhuisen, F.; Wilson, S. Global anthropogenic mercury emission inventory for 2000. Atmos. Environ. 2006, 40, 4048−4063. (5) Pacyna, E. G.; Pacyna, J. M.; Sundseth, K.; Munthe, J.; Kindbom, K.; Wilson, S.; Steenhuisen, F.; Maxson, P. Global emission of mercury to the atmosphere from anthropogenic sources in 2005 and projections to 2020. Atmos. Environ. 2010, 44, 2487−2499. (6) Pirrone, N.; Cinnirella, S.; Feng, X.; Finkelman, R. B.; Friedli, H. R.; Leaner, J.; Mason, R.; Mukherjee, A. B.; Stracher, G. B.; Streets, D. G.; Telmer, K. Global mercury emissions to the atmosphere from anthropogenic and natural sources. Atmos. Chem. Phys. 2010, 10, 5951−5964. (7) Streets, D. G.; Devane, M. K.; Lu, Z.; Bond, T. C.; Sunderland, E. M.; Jacob, D. J. All-time releases of mercury to the atmosphere from human activities. Environ. Sci. Technol. 2011, 45, 10485−91. (8) Rafaj, P.; Bertok, I.; Cofala, J.; Schöpp, W. Scenarios of global mercury emissions from anthropogenic sources. Atmos. Environ. 2013, 79, 472−479. (9) Muntean, M.; Janssens-Maenhout, G.; Song, S.; Selin, N. E.; Olivier, J. G. J.; Guizzardi, D.; Maas, R.; Dentener, F. Trend analysis from 1970 to 2008 and model evaluation of EDGARv4 global gridded anthropogenic mercury emissions. Sci. Total Environ. 2014, 494−495, 337−50. (10) Stohl, A.; et al. An analytical inversion method for determining regional and global emissions of greenhouse gases: Sensitivity studies and application to halocarbons. Atmos. Chem. Phys. 2009, 9, 1597− 1620. (11) Kirschke, S.; et al. Three decades of global methane sources and sinks. Nat. Geosci. 2013, 6, 813−823. (12) Le Quéré, C.; et al. Global Carbon Budget 2015. Earth Syst. Sci. Data 2015, 7, 349−396. (13) Song, S.; et al. Top-down constraints on atmospheric mercury emissions and implications for global biogeochemical cycling. Atmos. Chem. Phys. 2015, 15, 7103−7125. (14) Henne, S.; Brunner, D.; Oney, B.; Leuenberger, M.; Eugster, W.; Bamberger, I.; Meinhardt, F.; Steinbacher, M.; Emmenegger, L. Validation of the Swiss methane emission inventory by atmospheric

ASSOCIATED CONTENT

S Supporting Information *

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.est.6b05630. Additional tables and figures as mentioned in the text (PDF)

■

Article

AUTHOR INFORMATION

Corresponding Author

*Phone: +41 (0)44 632 5951; e-mail: christian.bogdal@chem. ethz.ch. ORCID

Christian Bogdal: 0000-0001-9137-8663 Notes

The authors declare no competing financial interest. G


Article

Environmental Science & Technology observations and inverse modelling. Atmos. Chem. Phys. 2016, 16, 3683−3710. (15) Zellweger, C.; Forrer, J.; Hofer, P.; Nyeki, S.; Schwarzenbach, B.; Weingartner, E.; Ammann, M.; Baltensperger, U. Partitioning of reactive nitrogen (NOy) and dependence on meteorological conditions in the lower free troposphere. Atmos. Chem. Phys. 2003, 3, 779−796. (16) Herrmann, E.; Weingartner, E.; Henne, S.; Vuilleumier, L.; Bukowiecki, N. Analysis of long-term aerosol size distribution data from Jungfraujoch with emphasis on free tropospheric conditions, cloud influence, and air mass transport. J. Geophys. Res. Atmos. 2015, 120, 9459−9480. (17) Brunner, D.; Henne, S.; Keller, C. A.; Vollmer, M. K.; Reimann, S.; Buchmann, B. Estimating European halocarbon emissions using lagrangian backward transport modeling and in situ measurements at the jungfraujoch high-alpine site. Geophys. Monogr. Ser. 2013, 200, 207−221. (18) Ebinghaus, R.; et al. International field intercomparison measurements of atmospheric mercury species at Mace Head, Ireland. Atmos. Environ. 1999, 33, 3063−3073. (19) Brown, R. J. C.; Brown, A. S.; Yardley, R. E.; Corns, W. T.; Stockwell, P. B. A practical uncertainty budget for ambient mercury vapour measurement. Atmos. Environ. 2008, 42, 2504−2517. (20) Gustin, M.; Jaffe, D. Reducing the uncertainty in measurement and understanding of mercury in the atmosphere. Environ. Sci. Technol. 2010, 44, 2222−2227. (21) Stohl, A.; Forster, C.; Frank, A.; Seibert, P.; Wotawa, G. Technical note: The Lagrangian particle dispersion model FLEXPART version 6.2. Atmos. Chem. Phys. 2005, 5, 2461−2474. (22) Ruckstuhl, A. F.; Henne, S.; Reimann, S.; Steinbacher, M.; Vollmer, M. K.; O’Doherty, S.; Buchmann, B.; Hueglin, C. Robust extraction of baseline signal of atmospheric trace species using local regression. Atmos. Meas. Tech. 2012, 5, 2613−2624. (23) AMAP/UNEP. Geospatially Distributed Mercury Emissions Dataset 2010v1; AMAP/UNEP: Geneva; Oslo, 2013. (24) Weigelt, A.; Ebinghaus, R.; Manning, A. J.; Derwent, R. G.; Simmonds, P. G.; Spain, T. G.; Jennings, S. G.; Slemr, F. Analysis and interpretation of 18 years of mercury observations since 1996 at Mace Head, Ireland. Atmos. Environ. 2015, 100, 85−93. (25) Wängberg, I.; Munthe, J. Atmospheric Mercury in Sweden, Northern Finland and Northern Europe; Results from National Monitoring and European Research, 2001. (26) Fu, X.; Marusczak, N.; Heimbürger, L.-E.; Sauvage, B.; Gheusi, F.; Prestbo, E. M.; Sonke, J. E. Atmospheric mercury speciation dynamics at the high-altitude Pic du Midi Observatory, southern France. Atmos. Chem. Phys. 2016, 16, 5623−5639. (27) UNEP. Global Mercury Assessment 2013: Sources, Emissions, Releases, and Environmental Transport; UNEP Chem. Branch: Geneva, 2013. (28) Keller, C. A.; Hill, M.; Vollmer, M. K.; Henne, S.; Brunner, D.; Reimann, S.; O’Doherty, S.; Arduini, J.; Maione, M.; Ferenczi, Z.; Haszpra, L.; Manning, A. J.; Peter, T. European emissions of halogenated greenhouse gases inferred from atmospheric measurements. Environ. Sci. Technol. 2012, 46, 217−25.

H


Inversion Approach to Validate Mercury Emissions ... - ACS Publications

Recommend Documents