- Email: [email protected]
On artificial dilution of point source mercury emissions in a regional atmospheric model
The Science of the Total Environment 259 Ž2000. 159᎐168
On artificial dilution of point source mercury emissions in a regional atmospheric model Prasad Pai 1, Prakash Karamchandani, Christian Seigneur U Atmospheric and En¨ ironmental Research, Inc., 2682 Bishop Dri¨ e, Suite 120, San Ramon, CA 94583, USA Received 27 September 1999; accepted 18 February 2000
Abstract Previously, we developed and applied a regional atmospheric mercury model to a domain covering most of North America at a horizontal grid resolution of 100 km. The implication of using this coarse resolution is that point sources of mercury emissions are instantaneously spread over a grid volume of horizontal dimensions 100 = 100 km2 and a vertical dimension equal to the depth of the grid cell where the point source emissions are released. Since point sources comprise a significant majority of a regional mercury emissions inventory, it is important to understand what effect this artificial dilution may have on calculated mercury concentrations and deposition fluxes. To understand this effect, we conducted model simulations using a finer grid, embedded within the original coarse grid, over a sub-domain that includes over 50% of the largest mercury point sources in the north-eastern United States. The horizontal resolution of the fine grid is 20 km, i.e. it is five times smaller than that of the coarse grid. We compared short-term Ždaily. and long-term Žannual. averaged mercury concentrations, and deposition Žwet and dry. fluxes on the coarse and fine grids. As expected, the effect of grid resolution is more clearly seen in close proximity to point sources than at remote locations. For short-term averages near major point sources, the peak concentrations and dry deposition fluxes of mercury from the fine grid are almost a factor of two greater than the corresponding estimates from the coarse grid. At remote locations, however, the concentrations and dry deposition peaks estimated by the two model grid resolutions are more comparable. For total wet deposition of mercury, the distinction between the fine and the coarse grid model results is less significant, regardless of the location. This could be due to the redistribution of precipitation fields or the effect of mercury aqueous chemistry. The effect of grid resolution is more important when model estimates are averaged over short time periods, e.g. daily, as opposed to over long periods, e.g. seasonally and annually. 䊚 2000 Elsevier Science B.V. All rights reserved. Keywords: Mercury; Regional atmospheric model; Grid
U
Corresponding author. Tel.: q-1-925-244-7121; fax: q1-925-244-7129. E-mail address: [email protected] ŽC. Seigneur.. 1 Present address. Sun Microsystems, Inc., 901 San Antonio Road, Palo Alto, CA 94303, USA. 0048-9697r00r$ - see front matter 䊚 2000 Elsevier Science B.V. All rights reserved. PII: S 0 0 4 8 - 9 6 9 7 Ž 0 0 . 0 0 5 7 9 - 9
160
P. Pai et al. r The Science of the Total En¨ ironment 259 (2000) 159᎐168
1. Introduction The 1990 Clean Air Act Amendments ŽCAAA. required the US Environmental Protection Agency ŽEPA. to conduct an assessment of the potential health and ecological risks associated with mercury ŽEPA, 1997.. The most important form of mercury from the point of view of risk assessment is methylmercury that accumulates in fish, which in turn is consumed by humans, other mammals and birds. Methylmercury is a neurotoxin that is produced by biotic and abiotic methylation of inorganic mercury. The discovery of high levels of mercury in fish in lakes remote from human activities implicated atmospheric deposition as a source of mercury ŽHuckabee, 1973; Fitzgerald, 1986.. The atmospheric deposition has complex origins, coming from a variety of natural and anthropogenic sources, and from different spatially important scales ŽSchroeder and Munthe, 1998.. Because atmospheric cycling integrates local, regional and global sources of mercury, understanding mercury source᎐receptor relationships ŽSRRs. is not straightforward. Mathematical models are used to develop SRRs by simulating the transport, formation and destruction processes that govern the pollutant of interest ŽVenkatram et al., 1997.. Comprehensive models are also being developed to understand SRRs for mercury ŽPetersen et al., 1995, 1998; Pai et al., 1997; Xu et al., 1999.. In a previous study ŽPai et al., 1997., we used an Eulerian model, TEAM, to simulate the transport and fate of mercury emissions in the contiguous US and portions of Canada and Mexico. TEAM was evaluated against observations ŽPai et al., 1997; Lindberg et al., 2000. and tolerances in the model estimates were quantified using sensitivity simulations ŽPai et al., 1999.. In TEAM we have included a comprehensive treatment of mercury transport, chemistry and deposition processes. The model was exercised at a 100-km horizontal resolution. The choice of the 100-km resolution was a compromise between relevant spatial scales of interest and computational practicality; a finer resolution for the large modeling domain would have been prohibitively
expensive. In this study, we examine the effect of grid resolution on results from TEAM. The simulated sources of mercury emissions included combustion of mercury-containing fossil fuels, municipal and medical waste incineration, and manufacturing processes ŽChu and Porcella, 1995; Pai et al., 1998, 2000.. The sources of mercury emissions used as model inputs could be grouped into two categories ᎏ point sources and area sources ᎏ depending on their treatment in the Eulerian framework. Point sources are identified by their exact location, i.e. latitude and longitude, by stack parameters Že.g. height, exit velocity. that determine whether material is emitted into the surface layer air or into air aloft, and by the emission rate Žmassrtime.. Area sources are identified by the emission rate Žmassrtime. and are aggregated in 100 = 100-km2 grid cells. The finite spatial resolution of the modeling domain places limitations on the treatment of emissions. Emissions from a point source are immediately spread over a grid volume corresponding to the location of the source. This means that every source has horizontal dimensions of 100 = 100 km2. Furthermore, the artificial spreading of sources leads to an excessive dilution of emissions and must be taken into account when comparing point measurements against grid-cell averaged model estimates so that the model-estimated SRRs can be viewed with confidence. By conducting a simulation with fine grid resolution, we have quantified the effect of the artificial dilution on mercury concentrations and deposition fluxes at various spatial and temporal scales. The fine grid simulation will also help understand if a regional model such as TEAM could be used to capture some of the spatial gradients near major point sources.
2. Fine grid simulation As stated earlier, choosing a fine resolution for the entire modeling domain was not practical from a computational viewpoint. Therefore, we selected a subdomain for the fine grid simulation. The selection of the subdomain was primarily
P. Pai et al. r The Science of the Total En¨ ironment 259 (2000) 159᎐168
guided by the distribution of mercury emissions Žparticularly from large point sources. within the larger domain. Fig. 1 shows the location of the fine grid domain relative to the larger domain. We will hereafter refer to the larger domain of the previous study ŽPai et al., 1997. as the coarse grid domain. The fine grid domain includes Washington, DC and the states of New York, Vermont, Pennsylvania, New Jersey, Delaware, Maryland, Connecticut, Massachusetts, and Rhode Island. In addition, it includes most of New Hampshire, and portions of Maine, Ohio, West Virginia, and Virginia. The fine grid domain includes 50% of the largest mercury point sources in the entire regional Hg inventory. Fig. 2 depicts the locations of the major point sources in the fine grid do-
161
main. The size of the point source circles in Fig. 2 is an indicator of the emission strength of the point source. The horizontal resolution of the fine grid is 20 km, i.e. there are 25 fine grid cells within each 100 = 100-km2 coarse grid cell. The vertical resolution is identical to that of the coarse grid configuration described in Pai et al. Ž1997.. The size of the fine grid domain is 1000 km on each side, resulting in 50 = 50 fine grid cells Žequivalent to 10 = 10 coarse grid cells.. The fine grid simulations were conducted using ‘one-way’ nesting, i.e. the fine grid boundary conditions were derived using the results of the coarse grid simulations ŽPleim et al., 1991.. Thus, time-dependent concentrations at each coarse grid cell adjacent to
Fig. 1. Location of the fine grid subdomain relative to the original coarse grid domain. The circles represent major point sources of Hg emissions.
162
P. Pai et al. r The Science of the Total En¨ ironment 259 (2000) 159᎐168
Fig. 2. Locations of major point sources in the fine grid domain.
the boundary grid cell of the fine grid domain serve as boundary conditions of the fine grid simulation. In addition to the boundary conditions, emissions, meteorology, and geophysical data files were developed for the fine grid simulation. The development of these files is described in Section 3.
3. Preparation of input files The fine grid simulation requires the preparation of spatially and temporally varying emissions, meteorology, cloud and precipitation fields at the 20 km resolution. In addition, a chemical concentrations file is required to provide O3 , HCl, SO2 , Cl2 and H2 O2 concentrations for simulating mercury chemistry ŽSchroeder et al., 1991; Seigneur et al., 1994; Hall, 1995.. This file was identical to that used in the coarse grid simulation. The emission input file consists of grid-resolved
estimates of point and area sources. The point source input file was quite straightforward to generate since it required only a re-specification of the grid cell location based on the stack location of individual point sources. The area source emission estimates are available at a county-level geographic resolution. These emissions were reallocated to the fine grid cells in proportion to the fraction of the county area that falls in each 20-km grid cell. The procedure used to develop emission input files to TEAM is described by Pai et al. Ž1998.. The input files for meteorology, cloud and precipitation were generated using a diagnostic meteorological model. The diagnostic model is similar to that described by Ross et al. Ž1988.. We first extracted the Nested Grid Model ŽNGM. estimates for the fine grid subdomain shown in Fig. 1. The NGM is described by Hoke et al. Ž1989.. We also extracted the cloud and precipitation fields from the National Weather Service ŽNWS. stations in the fine grid subdomain. The NGM and NWS data for the fine grid subdomain were then provided as input to the diagnostic meteorological model to generate the required input fields as described by Pai et al. Ž1997.. Therefore, the meteorological fields used by the fine grid simulation differed from those used by the coarse grid simulation because each used its own processing of the meteorological data with the diagnostic meteorological model. If meteorological data were available at a spatial resolution commensurate with that of the fine grid, we would anticipate larger differences between the fine grid and coarse grid simulations. In any case, the objective of this work pertains to the artificial dilution of mercury emissions from point sources rather than to the effect of meteorology.
4. Results Fig. 3a,b show the effect of enhancing the grid resolution on the cumulative frequency distributions of estimated annually-averaged total gaseous mercury ŽTGM. concentrations and particulate mercury wHgŽp.x concentrations in the subdomain shown in Fig. 1. As expected, the fine grid cap-
P. Pai et al. r The Science of the Total En¨ ironment 259 (2000) 159᎐168
163
Fig. 3. Cumulative frequency distributions of annually-averaged model estimates of: Ža. TGM concentrations Žngrm3 .; Žb. HgŽp. concentrations Žngrm3 .; Žc. dry deposition Žgrm2 per year.; and Žd. wet deposition Žgrm2 per year..
tures a much larger range of values, even for long-term averages. The tails of the distribution are longer at both ends for the fine grid simulation. Similar results are obtained for annual dry and wet deposition fluxes of total mercury, shown in Fig. 3c,d. To see how the grid resolution can affect short-term average concentrations and deposition fluxes, we present results for daily average values at two sites: the first site is in New York, in the vicinity of a large point source, and the second site is in Vermont, located relatively far away from major sources. Simulation results for the coarse grid cell in which each site is located are presented; the fine grid results are presented as a range Žminimum and maximum modeled values for the 25 fine grid cells within the coarse grid cell.. Fig. 4a shows the time series of simulated TGM concentrations for the New York grid cell, while Fig. 4b shows the corresponding values for the Vermont grid cell. We see that for the New York grid cell, the coarse grid value is usually greater than the minimum fine grid value and
lower than the maximum fine grid value. On several days, the maximum fine grid concentration is much greater than the coarse grid concentration, sometimes by a factor of two. On the other hand, the distinction between fine and coarse grid concentrations is not as discernible for the Vermont grid cell. This tendency of the model is better illustrated by sorting the daily average coarse grid model results as shown in Fig. 5a᎐h for the four model output variables, i.e. TGM and HgŽp. concentrations and dry and wet deposition fluxes. We see from Fig. 5a that, for the New York grid cell, the coarse grid TGM concentration is well within the bounds of the fine grid cell concentrations for all but a few days. There is also a discernible gap between the minimum and maximum fine grid concentrations. For the Vermont grid cell, shown in Fig. 5b, it is not as easy to discern the differences between coarse and fine grid concentrations, and the ranges of the fine grid concentrations are considerably narrower than for the New York grid cell. Note that the maximum daily fine
164
P. Pai et al. r The Science of the Total En¨ ironment 259 (2000) 159᎐168
Fig. 4. Time series of daily average TGM concentrations at :Ža. New York grid cell; and Žb. Vermont grid cell for the entire 1990 simulation year.
grid TGM concentration in the New York cell over the 1-year simulation period is approximately a factor of 3 greater than the corresponding concentration in the Vermont cell. The results for daily average HgŽp. concentrations, shown in Fig. 5c,d, are qualitatively similar to the TGM results. The effect of using the fine grid is more apparent for the New York grid cell ŽFig. 5c. than for the Vermont grid cell ŽFig. 5d.. Except for a few days, the ranges of fine grid concentrations are generally wider for the New York grid cell than for the Vermont grid cell. Furthermore, the coarse grid concentration for the New York cell usually lies within the fine grid range, while for the Vermont cell there are several days on which the coarse grid daily average value is outside the fine grid range. As in the case of TGM, the maximum daily fine grid HgŽp. concentration over the simulation period is much greater for the New York cell than for the Vermont cell Žby approx. a factor of 8.. Similar results are also obtained for the daily
dry deposition flux of total mercury, shown in Fig. 5e,f. However, the results for wet deposition, shown in Fig. 5g,h, are somewhat different. The distinction between the coarse and fine grid wet deposition results for the New York cell ŽFig. 5g. is not as clear as for the ambient concentrations and dry deposition flux. Also, most of the maximum fine grid wet deposition fluxes are comparable to the coarse grid fluxes, except for a few values at the upper end of the distribution where the maximum fine grid value is larger than the coarse grid value. There are also a few occasions where the coarse grid value is actually larger than the maximum fine grid value. One possible reason is due to the redistribution of precipitation in the fine grid cells. It should be recalled that the precipitation fields for the fine grid cells were derived using the NWS data and interpolated using a diagnostic meteorological model. Therefore, the precipitation amounts in the fine grid cells are not equal to the precipitation amount in the original coarse grid cell and it is likely that the increases in ambient concentrations in the fine grid are partially offset by decreases in precipitation. In addition, it is possible that the decrease in wet deposition is due to increased reduction of HgŽII. to HgŽ0. in clouds. A third explanation could be due to the fact that there are over 200 days in the year with zero or little precipitation resulting in negligible wet deposition for both the coarse and fine grid simulations. Similar results are noted for the Vermont grid cell ŽFig. 5h.. For this location also, the range of fine grid values of daily average mercury wet deposition is narrow and maximum fine grid wet deposition fluxes are generally comparable to the coarse grid fluxes. There are a few days when the coarse grid value lies in the range of the fine grid values, and a few days when the coarse grid value is greater than the fine grid value. In fact, the highest coarse grid value Žapprox. 2.5 grm2 per day. at the end of the distribution is more than a factor of two greater than the corresponding maximum fine grid value of 1 grm2 per day. As in the case of the New York grid cell, there are over 200 days in the year with negligible precipitation and mercury wet deposition in the Vermont grid cell.
P. Pai et al. r The Science of the Total En¨ ironment 259 (2000) 159᎐168
We now look at the effect of using the fine grid on annually averaged concentrations and deposition fluxes of mercury in each of the states within the fine grid domain. We compare the ranges obtained from the fine grid simulation with those from the coarse grid simulation. Recall that some states ŽOhio, Virginia, West Virginia, Maine, and New Hampshire. are only partially contained in the fine grid domain. The results shown for the individual states are representative only of the portion of each state contained in the domain. Fig. 6a,b compare the coarse and fine grid ranges for annual TGM and HgŽp. concentrations in the different states. The solid line represents the fine grid region, while the open and closed circles represent the minimum and maximum coarse grid values. For almost all the states, the coarse grid ranges are encompassed by the fine grid ranges. The largest differences between the coarse and fine grid ranges are seen for states with major point sources in the fine grid domain, such as
165
New York and Virginia. For a few states without major point sources in the fine grid domain, such as Massachusetts, Vermont, Rhode Island, and Maine, the differences between the fine and coarse grid ranges are negligible or small. Note that statewide average TGM and HgŽp. concentrations Žnot shown here. from the coarse and fine grid simulations are almost identical for all the states, including those with major point sources, suggesting that grid resolution has less effect when domainwide long-term averages are considered. Corresponding results for annual dry and wet deposition fluxes of total mercury are shown in Fig. 6c,d. In general, the results are qualitatively similar to the TGM and HgŽp. results, although there are some small differences, particularly for mercury wet deposition, where the lower bound value with the coarse grid is slightly smaller than that with the fine grid for a few states Žnotably New Hampshire and Virginia.. However, these
Fig. 5. Cumulative frequency distributions of daily-averaged model estimates of TGM concentrations Žngrm3 . at Ža. New York grid cell and Žb. Vermont grid cell. HgŽp. concentrations Žngrm3 . at Žc. New York grid cell and Žd. Vermont grid cell. Dry deposition Žgrm2 per day. at Že. New York grid cell and Žf. Vermont grid cell. Wet deposition Žgrm2 per day. at Žg. New York grid cell and Žh. Vermont grid cell.
166
P. Pai et al. r The Science of the Total En¨ ironment 259 (2000) 159᎐168
Fig. 5. Ž Continued.
differences are too small to change our overall conclusion that the fine grid captures a wider range of concentrations and deposition fluxes than the coarse grid. As in the case of TGM and HgŽp. concentrations, the statewide average annual dry and wet deposition fluxes from the coarse and fine grid simulations are almost identical. We have also examined the spatial patterns in annually-averaged mercury concentrations and deposition fluxes from the coarse and fine grid simulations. The spatial distribution of estimated TGM concentrations from the coarse and fine grid simulations Žnot shown here. indicate that the largest concentrations appear in the source clusters of southern New York, eastern Pennsylvania, and northern Virginia. However, the concentration gradients for the fine grid simulation are tighter than the coarse grid gradients, and the peak concentrations are also greater for the fine grid than for the coarse grid. The same is true for spatial distributions of annual mercury dry deposition from the coarse and fine grid simulations. Peak dry deposition fluxes with the fine grid are a
factor of two greater than peak estimates with the coarse grid. However, mercury wet deposition spatial patterns for the coarse and fine grid simulations are quite different. Although the fine grid pattern exhibits more spatial variation than the coarse grid distribution, peak wet deposition amounts are comparable. The reasons for wet deposition fluxes to exhibit different trends than concentrations and dry deposition fluxes were discussed earlier.
5. Summary and conclusions A comprehensive Eulerian model, developed in a previous study, was used to understand modeling uncertainties associated with a coarse horizontal resolution. This understanding is important before the model can be used to develop SRRs. We have examined the effect of decreasing the horizontal grid resolution from 100 to 20 km on model simulation results that are averaged over short- and long-term periods and those in close
P. Pai et al. r The Science of the Total En¨ ironment 259 (2000) 159᎐168
proximity and far away from major point sources. The simulation using 20 km resolution was conducted on a subdomain of the original coarse grid domain of 100 km resolution. The coarse grid results are then compared to the corresponding fine grid results for TGM and HgŽp. concentrations and dry and wet deposition fluxes. Ranges of fine grid simulation results are wider than ranges of coarse grid simulation results for both short-term Ždaily. and long-term Žannually . concentrations and dry and wet deposition fluxes of mercury in its various forms. The cumulative frequency distributions of annually-averaged concentrations and deposition fluxes over the fine grid domain are longer at both ends of the distribution for the fine grid simulation than for the coarse grid simulation. The effect of grid resolution is larger near point sources than at remote locations. Near point sources, the ranges of fine grid estimates of daily average concentrations and
167
dry deposition, and to a lesser extent, wet deposition, are generally larger than the corresponding ranges at remote locations. Furthermore, peak concentrations and dry deposition fluxes of mercury from the fine grid simulation are as much as a factor of two greater than peak coarse grid values near major point sources. At remote locations, the concentrations and dry deposition peaks from the fine and coarse grid simulations are more comparable. The distinction between coarse and fine grid results is less apparent for both daily and annual total mercury wet deposition, near point sources as well as at remote locations. Peak daily wet deposition estimates are mostly comparable for both coarse and fine grid simulations, primarily because ambient concentration increases are compensated by a better resolved precipitation field and the effect of cloud droplet chemistry Žaqueous reactions tend to minimize the amount
Fig. 6. Annual average coarse grid and fine grid estimates for individual states in the fine grid domain of: Ža. TGM concentrations Žngrm3 .; Žb. HgŽp. concentrations Žngrm3 .; Žc. dry deposition Žgrm2 per year.; and Žd. wet deposition Žgrm2 per year.. The solid line represents the fine grid region and the open and closed circles represent the minimum and maximum coarse grid estimates, respectively.
168
P. Pai et al. r The Science of the Total En¨ ironment 259 (2000) 159᎐168
of Hg removed by precipitation by reducing HgŽII. to HgŽ0.. Also, because of days with negligible precipitation over half of the simulation period in the fine grid domain, mercury wet deposition is negligible for both coarse and fine grid simulations during those days. The effect of grid resolution becomes less important when concentrations and deposition fluxes are averaged over long periods and over large domains. Annually and state averaged concentrations and dry and wet deposition fluxes were almost identical over all the states in the fine grid domain for both the coarse and fine grid simulations.
Acknowledgements The work described in this document was conducted under contract WO3218-08 with EPRI, Palo Alto, California. We thank Ms Mary Ann Allan of EPRI for her support. References Chu P, Porcella DB. Mercury stack emissions from U.S. electric utility power plants. Water Air Soil Pollut 1995;80:135᎐144. Environmental Protection Agency ŽEPA.. Mercury study report to Congress, vol. II, An inventory of anthropogenic mercury emissions in the United States. EPA-452rR-97004, Research Triangle Park, N.C., 1997. Fitzgerald WF. Cycling of mercury between the atmosphere and oceans. In: Buat-Menard P, editor. The role of air᎐sea exchange in geochemical cycling. NATO Advanced Science Institutes Series, Dordrecht, Netherlands: Reidel, 1986:363᎐408. Hall B. The gas phase oxidation of elemental mercury by ozone. Water Air Soil Pollut 1995;80:301᎐315. Hoke JE, Phillips NA, DiMego GJ, Tuccillo J, Sela J. The regional analysis and forecast system of the National Meteorological Center. Weather Forecast 1989;4:323᎐334. Huckabee JW. Mosses: sensitive indicators of airborne mercury pollution. Atmos Environ 1973;7:747᎐754. Lindberg S, Stratton W, Pai P, Allan MA. Measuring and modeling concentrations of water soluble gas-phase mercury species in ambient air. Fuel Process Technol Žin press..
Pai P, Karamchandani P, Seigneur C. Simulation of the regional atmospheric transport and fate of mercury using a comprehensive Eulerian model. Atmos Environ 1997;31: 2717᎐2732. Pai P, Heisler S, Joshi A. An emissions inventory for regional atmospheric modeling of mercury. Water Air Soil Pollut 1998;101:289᎐308. Pai P, Karamchandani P, Seigneur C, Allan M. Sensitivity of simulated atmospheric mercury concentrations and deposition to model input parameters. J Geophys Res 1999;104:13855᎐13868. Pai P, Niemi D, Powers B. A North American inventory of anthropogenic mercury emissions. Fuel Process Technol Žin press.. ˚ Munthe J. Atmospheric mercury Petersen G, Iverfeldt A, species over central and northern Europe: model calculation and comparison with observations from the Nordic air and precipitation network for 1987 and 1988. Atmos Environ 1995;29:47᎐67. Petersen G, Munthe J, Pleijel K, Bloxam R, Kumar AV. A comprehensive Eulerian modeling framework for airborne mercury species: development and testing of the tropospheric chemistry module ŽTCM .. Atmos Environ 1998;32:829᎐843. Pleim J, Chang J, Zhang K. A nested grid mesoscale atmospheric chemistry model. J Geophys Res 1991;90: 3065᎐3084. Ross DG, Smith IN, Manins PC, Fox DG. Diagnostic wind field modeling for complex terrain: model development and testing. J Appl Met 1988;27:785᎐795. Schroeder WH, Yarwood G, Niki H. Transformation processes involving mercury species in the atmosphere ᎏ results from a literature survey. Water Air Soil Pollut 1991;56:653᎐666. Schroeder WH, Munthe J. Atmospheric mercury ᎏ an overview. Atmos Environ 1998;32:809᎐822. Seigneur C, Wrobel J, Constantinou E. A chemical kinetic mechanism for atmospheric inorganic mercury. Environ Sci Technol 1994;28:1589᎐1597. Venkatram A, Karamchandani P, Pai P, Sloane C, Saxena P, Goldstein R. The development of a model to examine source᎐receptor relationships for visibility on the Colorado Plateau. J Air Waste Manage Assoc 1997;47:286᎐301. Xu X, Miller DR, Yang X, Thomas H. A regional-scale modeling study of atmospheric transport, transformation and deposition of mercury in the Northeastern United States. 5th International Conference on Mercury as a Global Pollutant, Rio de Janeiro, May 1999.
On artificial dilution of point source mercury emissions in a regional atmospheric model Prasad Pai 1, Prakash Karamchandani, Christian Seigneur U Atmospheric and En¨ ironmental Research, Inc., 2682 Bishop Dri¨ e, Suite 120, San Ramon, CA 94583, USA Received 27 September 1999; accepted 18 February 2000
Abstract Previously, we developed and applied a regional atmospheric mercury model to a domain covering most of North America at a horizontal grid resolution of 100 km. The implication of using this coarse resolution is that point sources of mercury emissions are instantaneously spread over a grid volume of horizontal dimensions 100 = 100 km2 and a vertical dimension equal to the depth of the grid cell where the point source emissions are released. Since point sources comprise a significant majority of a regional mercury emissions inventory, it is important to understand what effect this artificial dilution may have on calculated mercury concentrations and deposition fluxes. To understand this effect, we conducted model simulations using a finer grid, embedded within the original coarse grid, over a sub-domain that includes over 50% of the largest mercury point sources in the north-eastern United States. The horizontal resolution of the fine grid is 20 km, i.e. it is five times smaller than that of the coarse grid. We compared short-term Ždaily. and long-term Žannual. averaged mercury concentrations, and deposition Žwet and dry. fluxes on the coarse and fine grids. As expected, the effect of grid resolution is more clearly seen in close proximity to point sources than at remote locations. For short-term averages near major point sources, the peak concentrations and dry deposition fluxes of mercury from the fine grid are almost a factor of two greater than the corresponding estimates from the coarse grid. At remote locations, however, the concentrations and dry deposition peaks estimated by the two model grid resolutions are more comparable. For total wet deposition of mercury, the distinction between the fine and the coarse grid model results is less significant, regardless of the location. This could be due to the redistribution of precipitation fields or the effect of mercury aqueous chemistry. The effect of grid resolution is more important when model estimates are averaged over short time periods, e.g. daily, as opposed to over long periods, e.g. seasonally and annually. 䊚 2000 Elsevier Science B.V. All rights reserved. Keywords: Mercury; Regional atmospheric model; Grid
U
Corresponding author. Tel.: q-1-925-244-7121; fax: q1-925-244-7129. E-mail address: [email protected] ŽC. Seigneur.. 1 Present address. Sun Microsystems, Inc., 901 San Antonio Road, Palo Alto, CA 94303, USA. 0048-9697r00r$ - see front matter 䊚 2000 Elsevier Science B.V. All rights reserved. PII: S 0 0 4 8 - 9 6 9 7 Ž 0 0 . 0 0 5 7 9 - 9
160
P. Pai et al. r The Science of the Total En¨ ironment 259 (2000) 159᎐168
1. Introduction The 1990 Clean Air Act Amendments ŽCAAA. required the US Environmental Protection Agency ŽEPA. to conduct an assessment of the potential health and ecological risks associated with mercury ŽEPA, 1997.. The most important form of mercury from the point of view of risk assessment is methylmercury that accumulates in fish, which in turn is consumed by humans, other mammals and birds. Methylmercury is a neurotoxin that is produced by biotic and abiotic methylation of inorganic mercury. The discovery of high levels of mercury in fish in lakes remote from human activities implicated atmospheric deposition as a source of mercury ŽHuckabee, 1973; Fitzgerald, 1986.. The atmospheric deposition has complex origins, coming from a variety of natural and anthropogenic sources, and from different spatially important scales ŽSchroeder and Munthe, 1998.. Because atmospheric cycling integrates local, regional and global sources of mercury, understanding mercury source᎐receptor relationships ŽSRRs. is not straightforward. Mathematical models are used to develop SRRs by simulating the transport, formation and destruction processes that govern the pollutant of interest ŽVenkatram et al., 1997.. Comprehensive models are also being developed to understand SRRs for mercury ŽPetersen et al., 1995, 1998; Pai et al., 1997; Xu et al., 1999.. In a previous study ŽPai et al., 1997., we used an Eulerian model, TEAM, to simulate the transport and fate of mercury emissions in the contiguous US and portions of Canada and Mexico. TEAM was evaluated against observations ŽPai et al., 1997; Lindberg et al., 2000. and tolerances in the model estimates were quantified using sensitivity simulations ŽPai et al., 1999.. In TEAM we have included a comprehensive treatment of mercury transport, chemistry and deposition processes. The model was exercised at a 100-km horizontal resolution. The choice of the 100-km resolution was a compromise between relevant spatial scales of interest and computational practicality; a finer resolution for the large modeling domain would have been prohibitively
expensive. In this study, we examine the effect of grid resolution on results from TEAM. The simulated sources of mercury emissions included combustion of mercury-containing fossil fuels, municipal and medical waste incineration, and manufacturing processes ŽChu and Porcella, 1995; Pai et al., 1998, 2000.. The sources of mercury emissions used as model inputs could be grouped into two categories ᎏ point sources and area sources ᎏ depending on their treatment in the Eulerian framework. Point sources are identified by their exact location, i.e. latitude and longitude, by stack parameters Že.g. height, exit velocity. that determine whether material is emitted into the surface layer air or into air aloft, and by the emission rate Žmassrtime.. Area sources are identified by the emission rate Žmassrtime. and are aggregated in 100 = 100-km2 grid cells. The finite spatial resolution of the modeling domain places limitations on the treatment of emissions. Emissions from a point source are immediately spread over a grid volume corresponding to the location of the source. This means that every source has horizontal dimensions of 100 = 100 km2. Furthermore, the artificial spreading of sources leads to an excessive dilution of emissions and must be taken into account when comparing point measurements against grid-cell averaged model estimates so that the model-estimated SRRs can be viewed with confidence. By conducting a simulation with fine grid resolution, we have quantified the effect of the artificial dilution on mercury concentrations and deposition fluxes at various spatial and temporal scales. The fine grid simulation will also help understand if a regional model such as TEAM could be used to capture some of the spatial gradients near major point sources.
2. Fine grid simulation As stated earlier, choosing a fine resolution for the entire modeling domain was not practical from a computational viewpoint. Therefore, we selected a subdomain for the fine grid simulation. The selection of the subdomain was primarily
P. Pai et al. r The Science of the Total En¨ ironment 259 (2000) 159᎐168
guided by the distribution of mercury emissions Žparticularly from large point sources. within the larger domain. Fig. 1 shows the location of the fine grid domain relative to the larger domain. We will hereafter refer to the larger domain of the previous study ŽPai et al., 1997. as the coarse grid domain. The fine grid domain includes Washington, DC and the states of New York, Vermont, Pennsylvania, New Jersey, Delaware, Maryland, Connecticut, Massachusetts, and Rhode Island. In addition, it includes most of New Hampshire, and portions of Maine, Ohio, West Virginia, and Virginia. The fine grid domain includes 50% of the largest mercury point sources in the entire regional Hg inventory. Fig. 2 depicts the locations of the major point sources in the fine grid do-
161
main. The size of the point source circles in Fig. 2 is an indicator of the emission strength of the point source. The horizontal resolution of the fine grid is 20 km, i.e. there are 25 fine grid cells within each 100 = 100-km2 coarse grid cell. The vertical resolution is identical to that of the coarse grid configuration described in Pai et al. Ž1997.. The size of the fine grid domain is 1000 km on each side, resulting in 50 = 50 fine grid cells Žequivalent to 10 = 10 coarse grid cells.. The fine grid simulations were conducted using ‘one-way’ nesting, i.e. the fine grid boundary conditions were derived using the results of the coarse grid simulations ŽPleim et al., 1991.. Thus, time-dependent concentrations at each coarse grid cell adjacent to
Fig. 1. Location of the fine grid subdomain relative to the original coarse grid domain. The circles represent major point sources of Hg emissions.
162
P. Pai et al. r The Science of the Total En¨ ironment 259 (2000) 159᎐168
Fig. 2. Locations of major point sources in the fine grid domain.
the boundary grid cell of the fine grid domain serve as boundary conditions of the fine grid simulation. In addition to the boundary conditions, emissions, meteorology, and geophysical data files were developed for the fine grid simulation. The development of these files is described in Section 3.
3. Preparation of input files The fine grid simulation requires the preparation of spatially and temporally varying emissions, meteorology, cloud and precipitation fields at the 20 km resolution. In addition, a chemical concentrations file is required to provide O3 , HCl, SO2 , Cl2 and H2 O2 concentrations for simulating mercury chemistry ŽSchroeder et al., 1991; Seigneur et al., 1994; Hall, 1995.. This file was identical to that used in the coarse grid simulation. The emission input file consists of grid-resolved
estimates of point and area sources. The point source input file was quite straightforward to generate since it required only a re-specification of the grid cell location based on the stack location of individual point sources. The area source emission estimates are available at a county-level geographic resolution. These emissions were reallocated to the fine grid cells in proportion to the fraction of the county area that falls in each 20-km grid cell. The procedure used to develop emission input files to TEAM is described by Pai et al. Ž1998.. The input files for meteorology, cloud and precipitation were generated using a diagnostic meteorological model. The diagnostic model is similar to that described by Ross et al. Ž1988.. We first extracted the Nested Grid Model ŽNGM. estimates for the fine grid subdomain shown in Fig. 1. The NGM is described by Hoke et al. Ž1989.. We also extracted the cloud and precipitation fields from the National Weather Service ŽNWS. stations in the fine grid subdomain. The NGM and NWS data for the fine grid subdomain were then provided as input to the diagnostic meteorological model to generate the required input fields as described by Pai et al. Ž1997.. Therefore, the meteorological fields used by the fine grid simulation differed from those used by the coarse grid simulation because each used its own processing of the meteorological data with the diagnostic meteorological model. If meteorological data were available at a spatial resolution commensurate with that of the fine grid, we would anticipate larger differences between the fine grid and coarse grid simulations. In any case, the objective of this work pertains to the artificial dilution of mercury emissions from point sources rather than to the effect of meteorology.
4. Results Fig. 3a,b show the effect of enhancing the grid resolution on the cumulative frequency distributions of estimated annually-averaged total gaseous mercury ŽTGM. concentrations and particulate mercury wHgŽp.x concentrations in the subdomain shown in Fig. 1. As expected, the fine grid cap-
P. Pai et al. r The Science of the Total En¨ ironment 259 (2000) 159᎐168
163
Fig. 3. Cumulative frequency distributions of annually-averaged model estimates of: Ža. TGM concentrations Žngrm3 .; Žb. HgŽp. concentrations Žngrm3 .; Žc. dry deposition Žgrm2 per year.; and Žd. wet deposition Žgrm2 per year..
tures a much larger range of values, even for long-term averages. The tails of the distribution are longer at both ends for the fine grid simulation. Similar results are obtained for annual dry and wet deposition fluxes of total mercury, shown in Fig. 3c,d. To see how the grid resolution can affect short-term average concentrations and deposition fluxes, we present results for daily average values at two sites: the first site is in New York, in the vicinity of a large point source, and the second site is in Vermont, located relatively far away from major sources. Simulation results for the coarse grid cell in which each site is located are presented; the fine grid results are presented as a range Žminimum and maximum modeled values for the 25 fine grid cells within the coarse grid cell.. Fig. 4a shows the time series of simulated TGM concentrations for the New York grid cell, while Fig. 4b shows the corresponding values for the Vermont grid cell. We see that for the New York grid cell, the coarse grid value is usually greater than the minimum fine grid value and
lower than the maximum fine grid value. On several days, the maximum fine grid concentration is much greater than the coarse grid concentration, sometimes by a factor of two. On the other hand, the distinction between fine and coarse grid concentrations is not as discernible for the Vermont grid cell. This tendency of the model is better illustrated by sorting the daily average coarse grid model results as shown in Fig. 5a᎐h for the four model output variables, i.e. TGM and HgŽp. concentrations and dry and wet deposition fluxes. We see from Fig. 5a that, for the New York grid cell, the coarse grid TGM concentration is well within the bounds of the fine grid cell concentrations for all but a few days. There is also a discernible gap between the minimum and maximum fine grid concentrations. For the Vermont grid cell, shown in Fig. 5b, it is not as easy to discern the differences between coarse and fine grid concentrations, and the ranges of the fine grid concentrations are considerably narrower than for the New York grid cell. Note that the maximum daily fine
164
P. Pai et al. r The Science of the Total En¨ ironment 259 (2000) 159᎐168
Fig. 4. Time series of daily average TGM concentrations at :Ža. New York grid cell; and Žb. Vermont grid cell for the entire 1990 simulation year.
grid TGM concentration in the New York cell over the 1-year simulation period is approximately a factor of 3 greater than the corresponding concentration in the Vermont cell. The results for daily average HgŽp. concentrations, shown in Fig. 5c,d, are qualitatively similar to the TGM results. The effect of using the fine grid is more apparent for the New York grid cell ŽFig. 5c. than for the Vermont grid cell ŽFig. 5d.. Except for a few days, the ranges of fine grid concentrations are generally wider for the New York grid cell than for the Vermont grid cell. Furthermore, the coarse grid concentration for the New York cell usually lies within the fine grid range, while for the Vermont cell there are several days on which the coarse grid daily average value is outside the fine grid range. As in the case of TGM, the maximum daily fine grid HgŽp. concentration over the simulation period is much greater for the New York cell than for the Vermont cell Žby approx. a factor of 8.. Similar results are also obtained for the daily
dry deposition flux of total mercury, shown in Fig. 5e,f. However, the results for wet deposition, shown in Fig. 5g,h, are somewhat different. The distinction between the coarse and fine grid wet deposition results for the New York cell ŽFig. 5g. is not as clear as for the ambient concentrations and dry deposition flux. Also, most of the maximum fine grid wet deposition fluxes are comparable to the coarse grid fluxes, except for a few values at the upper end of the distribution where the maximum fine grid value is larger than the coarse grid value. There are also a few occasions where the coarse grid value is actually larger than the maximum fine grid value. One possible reason is due to the redistribution of precipitation in the fine grid cells. It should be recalled that the precipitation fields for the fine grid cells were derived using the NWS data and interpolated using a diagnostic meteorological model. Therefore, the precipitation amounts in the fine grid cells are not equal to the precipitation amount in the original coarse grid cell and it is likely that the increases in ambient concentrations in the fine grid are partially offset by decreases in precipitation. In addition, it is possible that the decrease in wet deposition is due to increased reduction of HgŽII. to HgŽ0. in clouds. A third explanation could be due to the fact that there are over 200 days in the year with zero or little precipitation resulting in negligible wet deposition for both the coarse and fine grid simulations. Similar results are noted for the Vermont grid cell ŽFig. 5h.. For this location also, the range of fine grid values of daily average mercury wet deposition is narrow and maximum fine grid wet deposition fluxes are generally comparable to the coarse grid fluxes. There are a few days when the coarse grid value lies in the range of the fine grid values, and a few days when the coarse grid value is greater than the fine grid value. In fact, the highest coarse grid value Žapprox. 2.5 grm2 per day. at the end of the distribution is more than a factor of two greater than the corresponding maximum fine grid value of 1 grm2 per day. As in the case of the New York grid cell, there are over 200 days in the year with negligible precipitation and mercury wet deposition in the Vermont grid cell.
P. Pai et al. r The Science of the Total En¨ ironment 259 (2000) 159᎐168
We now look at the effect of using the fine grid on annually averaged concentrations and deposition fluxes of mercury in each of the states within the fine grid domain. We compare the ranges obtained from the fine grid simulation with those from the coarse grid simulation. Recall that some states ŽOhio, Virginia, West Virginia, Maine, and New Hampshire. are only partially contained in the fine grid domain. The results shown for the individual states are representative only of the portion of each state contained in the domain. Fig. 6a,b compare the coarse and fine grid ranges for annual TGM and HgŽp. concentrations in the different states. The solid line represents the fine grid region, while the open and closed circles represent the minimum and maximum coarse grid values. For almost all the states, the coarse grid ranges are encompassed by the fine grid ranges. The largest differences between the coarse and fine grid ranges are seen for states with major point sources in the fine grid domain, such as
165
New York and Virginia. For a few states without major point sources in the fine grid domain, such as Massachusetts, Vermont, Rhode Island, and Maine, the differences between the fine and coarse grid ranges are negligible or small. Note that statewide average TGM and HgŽp. concentrations Žnot shown here. from the coarse and fine grid simulations are almost identical for all the states, including those with major point sources, suggesting that grid resolution has less effect when domainwide long-term averages are considered. Corresponding results for annual dry and wet deposition fluxes of total mercury are shown in Fig. 6c,d. In general, the results are qualitatively similar to the TGM and HgŽp. results, although there are some small differences, particularly for mercury wet deposition, where the lower bound value with the coarse grid is slightly smaller than that with the fine grid for a few states Žnotably New Hampshire and Virginia.. However, these
Fig. 5. Cumulative frequency distributions of daily-averaged model estimates of TGM concentrations Žngrm3 . at Ža. New York grid cell and Žb. Vermont grid cell. HgŽp. concentrations Žngrm3 . at Žc. New York grid cell and Žd. Vermont grid cell. Dry deposition Žgrm2 per day. at Že. New York grid cell and Žf. Vermont grid cell. Wet deposition Žgrm2 per day. at Žg. New York grid cell and Žh. Vermont grid cell.
166
P. Pai et al. r The Science of the Total En¨ ironment 259 (2000) 159᎐168
Fig. 5. Ž Continued.
differences are too small to change our overall conclusion that the fine grid captures a wider range of concentrations and deposition fluxes than the coarse grid. As in the case of TGM and HgŽp. concentrations, the statewide average annual dry and wet deposition fluxes from the coarse and fine grid simulations are almost identical. We have also examined the spatial patterns in annually-averaged mercury concentrations and deposition fluxes from the coarse and fine grid simulations. The spatial distribution of estimated TGM concentrations from the coarse and fine grid simulations Žnot shown here. indicate that the largest concentrations appear in the source clusters of southern New York, eastern Pennsylvania, and northern Virginia. However, the concentration gradients for the fine grid simulation are tighter than the coarse grid gradients, and the peak concentrations are also greater for the fine grid than for the coarse grid. The same is true for spatial distributions of annual mercury dry deposition from the coarse and fine grid simulations. Peak dry deposition fluxes with the fine grid are a
factor of two greater than peak estimates with the coarse grid. However, mercury wet deposition spatial patterns for the coarse and fine grid simulations are quite different. Although the fine grid pattern exhibits more spatial variation than the coarse grid distribution, peak wet deposition amounts are comparable. The reasons for wet deposition fluxes to exhibit different trends than concentrations and dry deposition fluxes were discussed earlier.
5. Summary and conclusions A comprehensive Eulerian model, developed in a previous study, was used to understand modeling uncertainties associated with a coarse horizontal resolution. This understanding is important before the model can be used to develop SRRs. We have examined the effect of decreasing the horizontal grid resolution from 100 to 20 km on model simulation results that are averaged over short- and long-term periods and those in close
P. Pai et al. r The Science of the Total En¨ ironment 259 (2000) 159᎐168
proximity and far away from major point sources. The simulation using 20 km resolution was conducted on a subdomain of the original coarse grid domain of 100 km resolution. The coarse grid results are then compared to the corresponding fine grid results for TGM and HgŽp. concentrations and dry and wet deposition fluxes. Ranges of fine grid simulation results are wider than ranges of coarse grid simulation results for both short-term Ždaily. and long-term Žannually . concentrations and dry and wet deposition fluxes of mercury in its various forms. The cumulative frequency distributions of annually-averaged concentrations and deposition fluxes over the fine grid domain are longer at both ends of the distribution for the fine grid simulation than for the coarse grid simulation. The effect of grid resolution is larger near point sources than at remote locations. Near point sources, the ranges of fine grid estimates of daily average concentrations and
167
dry deposition, and to a lesser extent, wet deposition, are generally larger than the corresponding ranges at remote locations. Furthermore, peak concentrations and dry deposition fluxes of mercury from the fine grid simulation are as much as a factor of two greater than peak coarse grid values near major point sources. At remote locations, the concentrations and dry deposition peaks from the fine and coarse grid simulations are more comparable. The distinction between coarse and fine grid results is less apparent for both daily and annual total mercury wet deposition, near point sources as well as at remote locations. Peak daily wet deposition estimates are mostly comparable for both coarse and fine grid simulations, primarily because ambient concentration increases are compensated by a better resolved precipitation field and the effect of cloud droplet chemistry Žaqueous reactions tend to minimize the amount
Fig. 6. Annual average coarse grid and fine grid estimates for individual states in the fine grid domain of: Ža. TGM concentrations Žngrm3 .; Žb. HgŽp. concentrations Žngrm3 .; Žc. dry deposition Žgrm2 per year.; and Žd. wet deposition Žgrm2 per year.. The solid line represents the fine grid region and the open and closed circles represent the minimum and maximum coarse grid estimates, respectively.
168
P. Pai et al. r The Science of the Total En¨ ironment 259 (2000) 159᎐168
of Hg removed by precipitation by reducing HgŽII. to HgŽ0.. Also, because of days with negligible precipitation over half of the simulation period in the fine grid domain, mercury wet deposition is negligible for both coarse and fine grid simulations during those days. The effect of grid resolution becomes less important when concentrations and deposition fluxes are averaged over long periods and over large domains. Annually and state averaged concentrations and dry and wet deposition fluxes were almost identical over all the states in the fine grid domain for both the coarse and fine grid simulations.
Acknowledgements The work described in this document was conducted under contract WO3218-08 with EPRI, Palo Alto, California. We thank Ms Mary Ann Allan of EPRI for her support. References Chu P, Porcella DB. Mercury stack emissions from U.S. electric utility power plants. Water Air Soil Pollut 1995;80:135᎐144. Environmental Protection Agency ŽEPA.. Mercury study report to Congress, vol. II, An inventory of anthropogenic mercury emissions in the United States. EPA-452rR-97004, Research Triangle Park, N.C., 1997. Fitzgerald WF. Cycling of mercury between the atmosphere and oceans. In: Buat-Menard P, editor. The role of air᎐sea exchange in geochemical cycling. NATO Advanced Science Institutes Series, Dordrecht, Netherlands: Reidel, 1986:363᎐408. Hall B. The gas phase oxidation of elemental mercury by ozone. Water Air Soil Pollut 1995;80:301᎐315. Hoke JE, Phillips NA, DiMego GJ, Tuccillo J, Sela J. The regional analysis and forecast system of the National Meteorological Center. Weather Forecast 1989;4:323᎐334. Huckabee JW. Mosses: sensitive indicators of airborne mercury pollution. Atmos Environ 1973;7:747᎐754. Lindberg S, Stratton W, Pai P, Allan MA. Measuring and modeling concentrations of water soluble gas-phase mercury species in ambient air. Fuel Process Technol Žin press..
Pai P, Karamchandani P, Seigneur C. Simulation of the regional atmospheric transport and fate of mercury using a comprehensive Eulerian model. Atmos Environ 1997;31: 2717᎐2732. Pai P, Heisler S, Joshi A. An emissions inventory for regional atmospheric modeling of mercury. Water Air Soil Pollut 1998;101:289᎐308. Pai P, Karamchandani P, Seigneur C, Allan M. Sensitivity of simulated atmospheric mercury concentrations and deposition to model input parameters. J Geophys Res 1999;104:13855᎐13868. Pai P, Niemi D, Powers B. A North American inventory of anthropogenic mercury emissions. Fuel Process Technol Žin press.. ˚ Munthe J. Atmospheric mercury Petersen G, Iverfeldt A, species over central and northern Europe: model calculation and comparison with observations from the Nordic air and precipitation network for 1987 and 1988. Atmos Environ 1995;29:47᎐67. Petersen G, Munthe J, Pleijel K, Bloxam R, Kumar AV. A comprehensive Eulerian modeling framework for airborne mercury species: development and testing of the tropospheric chemistry module ŽTCM .. Atmos Environ 1998;32:829᎐843. Pleim J, Chang J, Zhang K. A nested grid mesoscale atmospheric chemistry model. J Geophys Res 1991;90: 3065᎐3084. Ross DG, Smith IN, Manins PC, Fox DG. Diagnostic wind field modeling for complex terrain: model development and testing. J Appl Met 1988;27:785᎐795. Schroeder WH, Yarwood G, Niki H. Transformation processes involving mercury species in the atmosphere ᎏ results from a literature survey. Water Air Soil Pollut 1991;56:653᎐666. Schroeder WH, Munthe J. Atmospheric mercury ᎏ an overview. Atmos Environ 1998;32:809᎐822. Seigneur C, Wrobel J, Constantinou E. A chemical kinetic mechanism for atmospheric inorganic mercury. Environ Sci Technol 1994;28:1589᎐1597. Venkatram A, Karamchandani P, Pai P, Sloane C, Saxena P, Goldstein R. The development of a model to examine source᎐receptor relationships for visibility on the Colorado Plateau. J Air Waste Manage Assoc 1997;47:286᎐301. Xu X, Miller DR, Yang X, Thomas H. A regional-scale modeling study of atmospheric transport, transformation and deposition of mercury in the Northeastern United States. 5th International Conference on Mercury as a Global Pollutant, Rio de Janeiro, May 1999.