Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Urban atmospheric mercury contamination from artisanal mining : mapping, modeling, and mitigation Cordy, Paul David 2014

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2014_spring_cordy_paul.pdf [ 47.61MB ]
Metadata
JSON: 24-1.0103432.json
JSON-LD: 24-1.0103432-ld.json
RDF/XML (Pretty): 24-1.0103432-rdf.xml
RDF/JSON: 24-1.0103432-rdf.json
Turtle: 24-1.0103432-turtle.txt
N-Triples: 24-1.0103432-rdf-ntriples.txt
Original Record: 24-1.0103432-source.json
Full Text
24-1.0103432-fulltext.txt
Citation
24-1.0103432.ris

Full Text

    URBAN ATMOSPHERIC MERCURY CONTAMINATION FROM ARTISANAL MINING: MAPPING, MODELING, AND MITIGATION      by    Paul David Cordy      A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF   DOCTOR OF PHILOSOPHY   in   The Faculty of Graduate and Postdoctoral Studies   (Mining Engineering)   THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver)     April, 2014     © Paul David Cordy, 2014   ii Abstract  Artisanal miners in more than 70 countries extract gold using mercury, which is often evaporated in densely populated urban areas. This work explores the behaviour of these emissions, and the potential implications for human health. Maps of urban mercury concentrations are used to evaluate the impact of mercury reduction interventions and estimate the distribution of health hazard. Atmospheric dispersion modeling is also used to corroborate inferences about the behaviour of urban mercury vapour that are derived from observations, and to simulate hazard distributions. Miners decompose the amalgams (with 50 to 60% mercury) and melt the raw gold in shops located near the centre of each town without sufficient condensers or filters. The average concentrations measured by mobile mercury vapour analyzer transects taken repeatedly over several weeks were 1.25 µgm -3 in 2010 in Segovia and 0.331 µgm -3 in Andacollo (2009). Mobile mercury measurements and atmospheric dispersion modeling both indicate that mercury emissions from gold shops, though high, dissipate rapidly in space and time.  Mobile mercury mapping along streets can detect most frequent emitters with only a few weeks of mobile sampling. Observations of concentrations greater than  1 µgm -3 indicate that within the past 5 minutes amalgam was being burned within a 200 metre radius. Measurements from towers show the temporal variability of mercury concentrations, and show that large quantities of mercury are available for long-range atmospheric transport.  By World Health Organization (WHO) standards, these towns are exposed to a significant health hazard, and globally, the millions of miners as well as non-miners who live in similar towns are at serious risk of neurological and renal disease. Various direct and indirect indicators of gold production and mercury reduction also show that mitigation efforts by the United Nations Industrial Organization (UNIDO) in Colombia have reduced urban airborne mercury concentrations by approximately 50% in Segovia, Antioquia, despite a 30% increase in gold production during that three year period. This is attributable to the adoption of retorts by miners and regulations banning new processing centres to the rural periphery.     iii Preface  All of the research design and execution relating to atmospheric mercury sampling, modeling, plotting, analysis, and documentation of gold shop practices at all field sites are entirely my work. Ben Crawford, a UBC Atmospheric Science PhD candidate, provided his talents as a meteorology technician. Ben prepared and assembled the meteorological instruments. He also programmed and monitored the computation and storage of variables in the data logger. We chose the tower site and deployed the instruments together, and we both conducted roving mercury surveys and helped to keep the tower based mercury analyzer functioning.  Victor Hugo Gonzales Carrasco managed the production of gold shop filters and retorts in Andacollo based on my designs and under my supervision. Victor Hugo assisted me as I led the public outreach and consultation sessions with the gold shop workers in which the condenser designs were altered to suit their situation and needs. In this endeavor we were also greatly assisted by Claudia Jara, Dr. Daniel Moraga, and the Municipality of Andacollo. Descriptions of mining engineering practices in Andacollo are equally attributable to myself and Marcello Veiga. Measurements in Suriname and Peru were obtained in collaboration with Dennis Wip and the United States Environmental Protection Agency (USEPA) respectively, and the sampling methodology in both instances was opportunistic according to their respective goals at the time. I am responsible for all of the assimilation, analysis, and explanation of the data and results of all of the aforementioned field work. Chapter 2 details the work and collaborations described above, which have been published by Elsevier in Environmental Research in 2013 as Characterization, mapping, and mitigation of mercury vapour emissions from artisanal mining gold shops (excluding meteorology and modeling which has yet to be published).  Cordy, P., Veiga, M., Crawford, B., Garcia, O., Gonzalez, V., Moraga, D., Wip, D., 2013. Characterization, mapping, and mitigation of mercury vapour emissions from artisanal mining gold shops. Environmental Research, vol. 125, pp. 82-91.   iv I alone am responsible for the research design, data processing, analysis, figures, writing and manuscript preparation for this article.  Characterization of the mining practices in Colombia in both publications related to this thesis were done by Marcello Veiga, Oseas Garcia, and the staff of the United Nations Industrial Development Organization (UNIDO), CorAntioquia (Corporación Autónoma Regional del Centro de Antioquia), as were mercury reduction initiatives, estimations of changes in gold production and mercury mass balance.  Much of the research elaborated in chapter 4 has already been published in Elsevier’s Science of the Total Environment in 2011 as an article entitled Mercury contamination from artisanal gold mining in Antioquia, Colombia: The world's highest per capita mercury pollution.   Cordy, P., Veiga, M.M., Salih, I., Al-Saadi, S., Console, S., Garcia, O., Mesa, L.A., Velásquez-López, P.C., Roeser, M. 2011. Mercury contamination from artisanal gold mining in Antioquia, Colombia: The world's highest per capita mercury pollution, The Science of The Total Environment, vol. 410, no. 1, pp. 154-160.  Preparation of this manuscript, including analysis, was shared equally by Marcello Veiga and I, and we also shared non-atmospheric data collection with the other coauthors. Some of the opportunistic sampling inside gold shops and non-mining buildings was done by Marcello Veiga and UNIDO-CorAntioquia staff, but I am responsible for all of the mobile and tower measurements either directly or by drivers and technicians trained and directed by me. Descriptions of interventions and the preliminary UNIDO Colombia project findings came courtesy of Oseas Garcia. All measurements and descriptions relating to gold shops in Colombia are my work, as are all of the atmospheric mercury mapping and tower measurements, and all subsequent analysis. A comparison of 2010 and 2011 urban mercury concentrations in Segovia was published in the aforementioned gold shop mercury paper, however the updated results from 2012 and for towns near Segovia appear for the first time in this dissertation. As in all related publications, the mercury mapping and analysis of sequential years is entirely my work.     v Table of Contents Abstract .............................................................................................................................. ii Preface .............................................................................................................................. iii Table of Contents ............................................................................................................... v List of Tables ................................................................................................................. viii List of Figures .................................................................................................................... x List of Symbols, Units, and Constants ........................................................................... xiii List of Abbreviations ...................................................................................................... xiv Acknowledgements .......................................................................................................... xv Dedication ....................................................................................................................... xvi 1 Introduction ................................................................................................................. 1 1.1 Mercury emissions from artisanal and small scale miners ................................... 1 1.2 Mercury mapping and human health .................................................................... 7 1.3 Mercury dispersion modeling ............................................................................. 10 1.4 Mercury mitigation ............................................................................................. 12 1.5 Objectives and research questions ...................................................................... 14 1.6 Approach ............................................................................................................. 15 1.7 Applied contributions of the research ................................................................. 16 1.8 Limitations of the research ................................................................................. 17 1.9 Thesis structure ................................................................................................... 18 2 Field study sites and gold shop contamination .......................................................... 19 2.1 Mercury vapour sampling methods .................................................................... 20 2.2 Gold shop mercury contamination ...................................................................... 21 2.3 Andacollo, Chile ................................................................................................. 22 2.3.1 Andacollo gold shops ................................................................................... 30 2.4 Segovia, Colombia .............................................................................................. 32 2.4.1 Segovia gold shops ....................................................................................... 35 2.5 Puerto Maldonado, Peru ..................................................................................... 44 2.5.1 Puerto Maldonado gold shops ...................................................................... 47 2.6 Ponce Enriquez, Ecuador .................................................................................... 48 2.6.1 Ponce Enriquez gold shops .......................................................................... 49 2.7 Paramaribo, Suriname ......................................................................................... 50 2.7.1 Paramaribo gold shops ................................................................................. 51 2.8 Field sites summary ............................................................................................ 52 3 Urban mercury mapping and monitoring .................................................................. 54 3.1 Synopsis .............................................................................................................. 54 3.2 Mercury monitoring and mapping methods ....................................................... 54 3.2.1 Sampling and analysis .................................................................................. 54 3.2.2 Sampling error .............................................................................................. 57 3.2.3 Comparing air quality measurements with air quality standards ................. 58 3.3 Results ................................................................................................................. 59 3.3.1 Tower measurements .................................................................................... 63  vi 3.3.2 Mercury mapping ......................................................................................... 66 3.4 Discussion ........................................................................................................... 71 3.5 Summary ............................................................................................................. 73 4 Modeling mercury dispersion .................................................................................... 75 4.1 Synopsis .............................................................................................................. 75 4.2 Meteorology methods ......................................................................................... 76 4.2.1 Instruments ................................................................................................... 76 4.2.2 The instrument tower ................................................................................... 79 4.2.3 Data quality control ...................................................................................... 80 4.3 Dispersion modeling methods ............................................................................ 82 4.3.1 Boundary layer meteorology ........................................................................ 82 4.3.2 Sensible heat flux ......................................................................................... 83 4.3.3 Friction velocity ........................................................................................... 84 4.3.4 The slab model ............................................................................................. 85 4.3.5 Urban roughness ........................................................................................... 86 4.3.6 Urban structure model .................................................................................. 87 4.3.7 Stability ........................................................................................................ 89 4.3.8 CalPUFF ....................................................................................................... 90 4.3.9 Emissions ..................................................................................................... 92 4.3.10 Sensitivity testing ....................................................................................... 94 4.4 Error .................................................................................................................... 94 4.4.1 Emissions error ............................................................................................. 94 4.4.2 Meteorology error ........................................................................................ 96 4.4.3 Model error ................................................................................................... 97 4.5 Data inventory .................................................................................................... 98 4.6 Model evaluation ................................................................................................ 99 4.7 Results ............................................................................................................... 100 4.7.1 Event averages ............................................................................................ 101 4.7.2 Spatial average analysis ............................................................................. 102 4.7.3 Tower series ............................................................................................... 104 4.7.4 Spacetime series ......................................................................................... 112 4.8 Discussion ......................................................................................................... 119 4.9 Summary ........................................................................................................... 122 5 Mitigation of mercury contamination ...................................................................... 124 5.1 Synopsis ............................................................................................................ 124 5.2 Methods ............................................................................................................ 126 5.2.1 Evaluating mercury mitigation measures ................................................... 127 5.3 Results ............................................................................................................... 130 5.3.1 Mercury imports and sales ......................................................................... 130 5.3.2 Mercury losses ............................................................................................ 131 5.3.3 Mercury emissions from milling ................................................................ 133 5.3.4 Cyanidation of mercury contaminated tailings .......................................... 134 5.3.5 Interventions ............................................................................................... 135 5.3.6 Intervention impact assessment .................................................................. 138 5.4 Discussion ......................................................................................................... 145 5.4.1 UNIDO’s final project evaluation .............................................................. 149 5.5 Summary ........................................................................................................... 150  vii 6 Conclusions ............................................................................................................. 152 6.1 Original contributions of this study .................................................................. 154 6.2 Limitations of this research .............................................................................. 154 6.3 Recommendations for reducing mercury emissions and human exposure ....... 155 6.4 Recommendations for future research .............................................................. 156 6.5 Summary ........................................................................................................... 158 References ...................................................................................................................... 163      viii List of Tables Table 1.1: Summary of previous studies of ambient gold shop mercury concentrations. ............................. 8	  Table 2.1: Typical mercury concentrations (µgm-3, 10 second averaging time) in and near gold shops in Peru, Ecuador, Suriname (measured with Lumex; 10 second averages), and Colombia (measured using Jerome; five second averages). “Outside” is within 10 m of the front door; “entrance” is just inside of the front entrance. ................................................................................................................ 22	  Table 2.2: Average mass of gold and mercury in 62 amalgams sold at three different gold shops in Andacollo from April to July, 2009. Amalgam weights were measured before and after burning. All this mercury was released to the environment due to lack of filters in gold shops. ........................... 31	  Table 2.3: Summary statistics of mercury concentrations measured adjacent to burn/melt events in progress in Segovia using the Jerome analyzer (five second averages for each measurement interval, averaged over the entire burn duration). Note that measurement often ceased part way through the event due to saturation of the Jerome sensor. Commonly this followed readings of  >999 µgm-3. ... 41	  Table 2.4: Atmospheric mercury in selected parts of Segovia, taken using the Jerome analyzer (5 second averages). ............................................................................................................................................ 42	  Table 2.5: Average mercury concentrations (µgm-3) measured within the urban centers of Segovia, Remedios, Zaragoza, El Bagre, and Nechi. (Averaging times: 5 s (Jerome) and 10 s (Lumex)). ..... 43	  Table 3.1: Mercury concentrations in dense urban areas with amalgam burning in Andacollo, Chile and Segovia, Colombia. ............................................................................................................................ 70	  Table 4.1: Tower heights of instrumentation and meteorological variables. ............................................... 77	  Table 4.2: Summary of quality control limits and thresholds used for high-frequency eddy covariance data. .................................................................................................................................................... 81	  Table 4.3: Andacollo building measurements, in metres. ............................................................................ 87	  Table 4.4: Modeled building dimensions at various directions ................................................................... 88	  Table 4.5: Stability v. Monin-Obukhov length for zo = 0.5. ........................................................................ 89	  Table 4.6: CalPUFF parameters ................................................................................................................... 91	  Table 4.7: Original timing and masses of amalgam burns recorded in the principal Andacollo gold shop during mercury vapour sampling (April and May 2009). .................................................................. 93	  Table 4.8: CalPUFF parameters that were set to physically reasonable values, and the degree of correlation between model output obtained using the chosen value and that which results from testing the sensitivity of the model within the stated range. ................................................................................ 94	  Table 4.9: Summary of mercury measurements and known emissions. ...................................................... 98	  Table 4.10: Correlations (r, α = 0.05) of modeled versus observed concentration for known mercury releases from the principal gold shop, and adjustments made to the timing of emissions are shown. Cases in which there was no overlap between model and observations are marked n/a. ................ 100	  Table 4.11: Timing, masses, and emissions rates of amalgam burn events that were detected at both stadium tower analyzer locations. .................................................................................................... 104	  Table 4.12: Timing, masses, and emissions rates of amalgam burn events that were detected at only the NE stadium tower analyzer location. ............................................................................................... 107	  Table 4.13: Timing, masses, and emissions rates of amalgam burn events that were detected at only the NE stadium tower analyzer location and by the mobile analyzer. ................................................... 112	  Table 5.1: Changes in mercury concentrations averaged over each year’s field campaign in the urban cores of Antioquia. ..................................................................................................................................... 126	  Table 5.2: Countries exporting mercury in 2009 to Colombia (UN COMTRADE, 2009). ...................... 130	  Table 5.3: Number of entables in 5 municipalities in Antioquia ............................................................... 131	  Table 5.4: Comparison of airborne mercury concentrations resulting from milling at a slower speed in different entables (µgm-3). Fast refers to speeds of 54 to 58 rpm and slow refers to speeds near 40 rpm. .................................................................................................................................................. 133	  Table 5.5: Urban Segovia production capacity change as a proxy of gold production. ............................ 138	  Table 5.6: Summary of changes in average mercury concentration [Hg] in the urban cores of Segovia, La Cruzada, and Remedios, Colombia. Mercury concentrations are averaged over the entire mobile sampling dataset of each year’s field campaign. .............................................................................. 139	  Table 5.7: Changes in average mercury [Hg] concentration in the Segovia urban core; in bulk and separated by hazard level. Average values are taken over the entire field campaign in each year (‘total observation time’). ................................................................................................................. 139	   ix Table 5.8: Results of Kolmogorov-Smirnov tests for successive years of mercury vapour data in the Segovia urban core. .......................................................................................................................... 140	  Table 5.9: Results of Kolmogorov-Smirnov tests for successive years of mercury vapour data in Remedios. ......................................................................................................................................... 143	  Table 5.10: Results of Kolmogorov-Smirnov tests for successive years of mercury vapour data in La Cruzada. ............................................................................................................................................ 145	  Table 5.11: Pre- and post-treatment mineral processing data in 35 out of at least 323 entables across Antioquia (unverified preliminary UNIDO data). Large increases in mercury free ore processing capacity were implemented with the assistance of UNIDO. ............................................................ 149	      x List of Figures  Figure 1.1: The global mercury budget (adapted from Mason et al. 2012). Total inventories (numbers in white boxes) are in tonnes, and fluxes in tonnes per year. The percentage values in brackets are the estimated increases in inventories in the past 100 years due to anthropogenic activities (UNEP 2013). .................................................................................................................................................... 1 Figure 1.2: All mercury amalgamation finishes with panning the mercury laden mineral concentrate in a batea (gold pan) and pouring the mercury out. © Paul Cordy photo. .................................................. 2 Figure 1.3: Squeezing excess mercury through a cloth yields a tiny gold amalgam that is roughly 50% mercury. Its estimated value was under 200 USD, but they had burned 60 litres of gasoline to obtain it. The miners estimated that their net profit for the day was 20-30 USD. © Paul Cordy photo. ........ 2 Figure 1.4: River dredge and sluice at work in Colombia. The operator (left) swings the dredge back and forth along the riverbed to dredge new material. Miners wash the carpets (right) onto the blocked sluice bed, just after adding mercury to the concentrate sand. Stirring the concentrate helps drive off the fine, light material and mix in the mercury. © Paul Cordy photo. ................................................. 3 Figure 1.5: Quickly shovel panning the river sediment at the location where the mercury laden concentrate was panned reveals that some mercury (the silver spot on the right hand side of the black sand) is being released directly to the river by this process. © Paul Cordy photo. ........................................... 3 Figure 1.6: A series of 3 excavators in Colombia dig out the alluvium and pass it up to the sluice, where water jets make it into a slurry. The Z-sluice is lined with carpets that concentrate the gold, but unlike the dredge miner these people add 50 kilograms of mercury to the carpets to trap more gold. As a result they lose considerable volumes of mercury to attrition. © Paul Cordy photo. .................. 4 Figure 1.7: Hard rock artisanal mining in Bolivia: ore is mined from quartz veins in mine shafts first excavated in the colonial era (visible as giant cave mouths in the cliffs in the top left image. Note the cable car carrying ore down from these mine openings high above the glacier at an elevation of 5000 metres). Mercury is added to the ball mills at the beginning of milling (top left), and it flows out with the slurry (bottom right), later to be concentrated by sluicing and panning as in the alluvial examples above. The lower two photos show the ball mill circuit where ore is ground progressively finer and then discharged into pans to be moved manually to subsequent milling or sluicing. The bottom left photo shows the large primary mill for coarse material, the wheelbarrow used to take the discharge of that material to the four smaller mills where the material is milled to a smaller size with mercury. © Paul Cordy photo. ............................................................................................................. 5 Figure 1.8: Gold shops are usually nested among other businesses in busy town centres, as in this picture of the main street of Puerto Maldonado, in the Peruvian Amazon. The two doors on the right are gold shops, the one in the middle is a hotel, a place to buy cell phone credits and the home of these two children. On the left is a bus company. In the Universal Oro shop there is a silver fume hood whose chimney empties into the room just above the shop operator’s head. © Paul Cordy photo. .... 6 Figure 1.9:  bubbler style gold shop filter installed in Andacollo. ................................................................ 9 Figure 3.1: Various mobile mercury monitoring solutions. Mercury concentration could be monitored on the laptop during sampling (as shown in the blue line plot), and a separate car battery was used to maintain constant electricity to the analyzer. The goal was always to have the hose inlet at 2 metres height with the inlet facing away from the vehicle, though the exact configuration varied in all studies. The truck was used in Antioquia, the car was our main vehicle in Andacollo, and we soon abandoned the bike carts in Andacollo when a loaned vehicle was available. © Paul Cordy photo. 55 Figure 3.2: The now demolished church tower site in the centre of Segovia with view of a typical Segovia street canyon (left), and the Andacollo stadium tower mercury monitoring sites. © Paul Cordy photo. .................................................................................................................................................. 56 Figure 3.3: Lognormal probability plots for Andacollo (top) and Segovia (bottom). Tower observations are shown in blue and mobile observations are shown in red. Dashed lines indicate perfect lognormality. ............................................................................................................................................................ 60 Figure 3.4: Geometric mean, 90th percentile, and 10th percentile values of mercury concentrations for various averaging periods in Andacollo. The top plot shows Andacollo’s tower observations (mean: 0.02 µgm-3), and the bottom plot shows mobile observations (mean: 0.07 µgm-3). The expected value of the annual average is the mean of the averages of all averaging periods. ............................ 61 Figure 3.5: Geometric mean, 90th percentile, and 10th percentile values of mercury concentrations for various averaging periods in Segovia. The top plot shows Segovia’s tower observations (mean: 0.1  xi µgm-3) and the bottom plot shows mobile observations (mean: 0.2 µgm-3). The expected value of the annual average is the mean of the averages of all averaging periods. ............................................... 62 Figure 3.6: Time series of mercury concentrations taken at two towers, both 18 metres above ground in Andacollo. The towers are approximately 50 metres (black) and 150 metres (gray) downwind of the principal gold shop. Dashed lines show the boundaries of individual time series, which are discontinuous. Purple underlines in the upper series indicate the segments shown at higher resolution in the lowest series. Extreme mercury concentrations are accommodated by plotting linearly up to 1 µgm-3 and logarithmically above (1 µgm-3 as an annual average is hazardous according to the WHO (2003)). Roughly one third of data occurred at night and are not shown as only background values were observed at night. Blue arrows indicate the wind speed and direction. ............................................................................................................................................................ 64 Figure 3.7: Mercury observations from the 17 metre tall church tower on the main square in Segovia. Dashed lines show the boundaries of individual time series, which are discontinuous. The horizontal bar in each plot represents one hour. Purple underlines in the upper two series indicate the segments shown at higher resolution in the bottom two series. Extreme mercury concentrations are accommodated by plotting linearly up to 1 µgm-3 and logarithmically above (1 µgm-3 as an annual average is hazardous according to the WHO (2003)). ....................................................................... 65 Figure 3.8: Average mercury concentrations along the Andacollo, Chile gold shop neighbourhood streets. The red and orange road lengths in the upper right corner of this image are road lengths which were only sampled on one or two occasions. The average of all one second observations measured over the entire field campaign in the area bounded by white dashes was found to have an average mercury vapour concentration of 0.331 µgm-3 with a standard deviation of 1.53. ............................ 67 Figure 3.9: Mercury concentrations in the Segovia, Colombia urban core; data are averages of three consecutive years of two-week measurement campaigns (2010, 2011, and 2012). The area shown was found to have an average mercury vapour concentration of 1.25 µgm-3 and standard deviation of 3.27 in 2010. Not all mercury sources are shown. Data north of UTM 782900 are not included in the average concentration as those segments were not consistently sampled. ......................................... 68 Figure 4.1: Instrument boom on the day of installation. From left to right: cup and vane anemometer, temperature and relative humidity sensor, net radiometer, sonic anemometer. © Paul Cordy photo.77 Figure 4.2: Google Earth imagery of Andacollo’s gold shop neighbourhood. Wind roses indicate the average wind speed (arrows) and relative frequency (black) in each 20 degree increment of direction over the entire field study. The cross bar on the north pointing wind speed arrow indicates 1 ms-1. 78 Figure 4.3: The meteorological tower looking east (left) and south (right). © Paul Cordy photo. ............. 79 Figure 4.4: Partial panorama of Andacollo taken from the tower at 6 metres height. This view looking west from South to North encompasses all directions from which the wind most frequently blows. © Paul Cordy photo. ............................................................................................................................... 80 Figure 4.5: Mixing layer evolution (from Stull 1988). ................................................................................ 82 Figure 4.6: Mixing layer growth model output for the Andacollo field campaign. ..................................... 85 Figure 4.7: View of Andacollo’s gold shop neighbourhood from a hill to the North. Note the Basilica to the lower right beside the mercury contaminated tailings pile, and the piles stretching along the central valley. © Paul Cordy photo. ................................................................................................... 86 Figure 4.8: Typical buildings in Andacollo. The sign reads “We are neighbours who take care of each other.” © Paul Cordy photo. ............................................................................................................... 88 Figure 4.12:  Logarithmic scatterplot and linear correlation of spatial averages. ...................................... 103 Figure 4.14: Logarithmic scatterplots for events 1 through 5. ................................................................... 106 Figure 4.15: Model (line) and observed (circles) mercury concentrations for event 6 (right), and the logarithmic scatter plot for this event. Blue arrows indicate wind speed and direction of five minute averages of wind, and the black arrow is the hourly average used in the model. ............................ 108 Figure 4.16: Model (line) and observed (circles) mercury concentrations for event 7 (right), and the logarithmic scatter plot for this event. Blue arrows indicate wind speed and direction of five minute averages of wind, and the black arrow is the hourly average wind used in the model. ................... 110 Figure 4.17: Model (line) and observed (circles) mercury concentrations for event 8 (right), and the logarithmic scatter plot for this event. Blue arrows indicate wind speed and direction of five minute averages of wind, and the black arrow is the hourly average wind used in the model. ................... 111 Figure 4.18: 3D plot (event 9) of mercury observations (top left), showing mobile observations as a wandering trace that passes back and forth through the mercury plume. Black arrows indicate five minute averages of wind speed and direction, and red indicates the hourly average wind. The red  xii star indicates the stack position at the start of the mercury emissions. The cross marks the tower location. All data are synchronized to the vertical time scale bar shown in the middle. Observed and modeled mercury concentrations at the NE tower are shown as colours (middle) and on time/concentration axes (right). The bottom series of plots show 2D time slices of mobile analyzer concentrations and contour maps of modeled concentrations. The arrows represent wind speed and direction; red represents the hourly average wind, black represents the five minute average of wind of the current time slice. ................................................................................................................... 113 Figure 4.19: Logarithmic scatterplot for event 9. ...................................................................................... 114 Figure 5.1: Cocos are small ball mills that grind and amalgamate the whole ore. © Paul Cordy photo. .. 126 Figure 5.2: Zinc roasting, the old way (top left), and with retorts. Pouring or pumping water over the retort speeds cooling and condensing. © Paul Cordy photo. ..................................................................... 136 Figure 5.3: Mercury recovered by the zinc retorts. © Paul Cordy photo. ................................................. 136 Figure 5.4: Empirical cumulative distributions of Segovia urban core one second mercury concentrations, on which the Kolmogorov-Smirnov test is based. Distributions further to the right have average values that are larger than those farther left. The 2010 distribution clearly contains higher concentrations of mercury. ............................................................................................................... 140 Figure 5.5: Mercury concentrations in the urban core of Segovia during short term measurement campaigns in successive years. Each colour represents the average of all concentrations measured within 10 metres of that point. Not all mercury sources are shown. ................................................ 141 Figure 5.6: Mercury concentrations in Remedios during the field campaigns of years shown. Each colour represents the average of all concentrations measured within 10 metres of that point. Not all mercury sources are shown. ............................................................................................................................ 142 Figure 5.7: Empirical cumulative distributions of Remedios mercury vapour concentrations, on which the Kolmogorov-Smirnov test is based. The 2010 distribution is farthest left, indicating that there were on average lower mercury concentrations in that year. .................................................................... 143 Figure 5.8: Mercury concentrations in La Cruzada during field campaigns of years shown. Each colour represents the average of all concentrations measured within 10 metres of that point. Not all mercury sources are shown. ............................................................................................................................ 144 Figure 5.9: Empirical cumulative distributions of La Cruzada mercury vapour concentrations, on which the Kolmogorov-Smirnov test is based. The 2010 distribution is farthest left, indicating that there were on average lower mercury concentrations in that year. ........................................................... 144     xiii List of Symbols, Units, and Constants   𝐶?  Empirical constant [0.2] 𝐶?𝐶??  Empirical constant [0.5] cp Jg-1K Specific heat capacity of air 𝑒 kPa Water vapour pressure  𝑒? kPa Water vapour pressure at 273 oK [0.611] 𝑒? kPa Saturation vapour pressure of water  𝑔 ms-2 Gravitational constant [9.81] ℎ m Mixing layer height  𝜅  Von Karman constant [0.4] 𝐿 m Monin-Obukhov length Lv J Latent heat of vaporization for liquid water [2.5x106] P kPa Atmospheric pressure Q* Wm-2 Net radiation  Qh Wm-2 Surface heat flux  𝑟 g/kg Mixing ratio of water in air  RH % Relative humidity Rv JK-1kg Gas constant for water vapour [461] 𝑇 oK Temperature 𝑇? oK Melting temperature of water [273] 𝑢∗ ms-1 Friction velocity u, v, w ms-1 Horizontal and vertical components of wind velocity 𝑢′𝑤′ and v’w’ m2s-2 Mean covariances of horizontal and vertical motions of air 𝑤?𝑇? mKs-1 Mean covariance of vertical air motion and temperature 𝑧? m Roughness length [m] zd m Zero displacement height [m] 𝑧? m Mean building height [m] zr m Blending height [m] Δθ oK Temperature difference at the top of the entrainment layer γ Km-1 Environmental lapse rate 𝜌 gm-3 Density of dry air     xiv List of Abbreviations  ABL Atmospheric boundary layer AMAP Arctic Monitoring and Assessment Program ASGM Artisanal and small scale mining CalPUFF United States Environmental Protection Agency dispersion model CERM3 Center for Environmental Research in Minerals, Metals, and Materials CETEM Centro de Technologia Mineral (Brazil) CorAntioquia Corporación Autónoma Regional del Centro de Antioquia DANE Departamento Administrativo Nacional de Estadistica (Colombia) ENAMI Empresa Nacional de Minería (Chile) GAMA Gestión Ambiental del Medio Ambiente (Peru/Switzerland) GEM Gaseous elemental mercury GOM Gaseous oxidized mercury GPS Global Positioning Satellite LOAEL Lowest Observed Adverse Effect Level MeHg Methylated mercury PVC Polyvinyl chloride RPM Revolutions per minute TWA Time Weighted Average UN COMTRADE United Nations Commodity Trade Statistics Database UNDP United Nations Development Program UNEP United Nations Environment Program UNIDO United Nations Industrial Development Organization USD United States dollars USEPA United States Environmental Protection Agency WHO World Health Organization WMO World Meteorological Organization     xv Acknowledgements  I am thankful for the collaboration of the UNIDO Mercury Project in Colombia, the Corporación Autónoma Regional del Centro de Antioquia (CorAntioquia) and the Government of Antioquia. In Andacollo, the cooperation of Victor Hugo Gonzalez Carrasco, Juan Cortes, Dr. Daniel Moraga, Claudia Jara, the Andacollo Artisanal Miners’ Association, the Municipality of Andacollo, and the employees and management at TECK’s Carmen de Andacollo operation were essential. In Suriname, Dennis Wip facilitated access to gold shops. Many thanks are also due to Oseas Garcia, Monika Roeser, Jesus Rua, Nick Lins, Adriana Maria Perez, Nicolás López Correa, Sandra Milena Osorio, Gustavo Vidales, Jesús María Bedoda, Cesar Ortega Tobón, Jaime Jaramillo Arbelaez, Padre Oscar Velaya (Iglesia de Segovia), David Gallego, Jorge Jaramillo Pereira, Edgar Velez Durango, Fanny Enriquez, Oscar Dario Amaya, Jesus M. Bodoya, Elias Pinto, Beatriz Duque Montoya, Pedro Perico, Julian Franco, and Augusto Posada. CERM3 UBC, CorAntioquia and Kevin Telmer provided mercury analyzers. Considerable scientific support was generously offered throughout this project by Marcello Veiga, Douw Steyn, Kevin Telmer, Ben Crawford, Andreas Christen, Jane Dennison, Sarah Gustin, and Maya Nakajima. This research was funded by NSERC, UNIDO, TECK, the Government of Antioquia, United States Department of State, and UBC.       xvi Dedication     To my parents, for the millions of ways in which you made this possible. 1 1  Introduction 1.1 Mercury emissions from artisanal and small scale miners  Between 10 and 15 million artisanal and small scale gold miners (ASGM) use mercury to extract gold in less developed nations worldwide, sometimes illegally and often producing severe impacts on landscapes and living systems. ASGM are the largest contributor to global atmospheric mercury emissions annually (37%), most of which is emitted in South America, Sub-Saharan Africa and East and Southeast Asia (United Nations Environment Program, UNEP 2013). ASGM emit an estimated 727 tonnes of mercury per year (estimated range 410–1040 tonnes per year) directly to the atmosphere. A significant but unknown portion of the amount released into the hydrosphere is later emitted to the atmosphere when it volatilizes.   Figure 1.1: The global mercury budget (adapted from Mason et al. 2012). Total inventories (numbers in white boxes) are in tonnes, and fluxes in tonnes per year. The percentage values in brackets are the estimated increases in inventories in the past 100 years due to anthropogenic activities (UNEP 2013).  2  Results from historical gold rushes suggest that over a period of 100 years at least 70% of known inputs to the hydrosphere were subsequently released to the atmosphere (Schuster et al. 2011).  The present gold rush in less developed nations around the world continues with little change in the technology that is used. Mercury is the oldest, cheapest and fastest means of extracting gold, and therefore amalgamation is the method of choice for artisanal miners globally. New technology adopted tends to be that which increases production but often does not increase efficiency or mitigation of environmental impacts: powerful excavators and bulldozers, diesel generators and pumps, electric rock drills and mine ventilation systems. These tools and the productivity they generate greatly accelerate the rate of landscape disturbance and generation of toxic wastes and residues, including mercury.    Figure 1.2: All mercury amalgamation finishes with panning the mercury laden mineral concentrate in a batea (gold pan) and pouring the mercury out. © Paul Cordy photo.   Figure 1.3: Squeezing excess mercury through a cloth yields a tiny gold amalgam that is roughly 50% mercury. Its estimated value was under 200 USD, but they had burned 60 litres of gasoline to obtain it. The miners estimated that their net profit for the day was 20-30 USD. © Paul Cordy photo.  3 Miners are often operating illegally, as they lack the organization and technical ability to apply for legal mining title and environmental licenses, thus preventing them from entering the formal economy and investing in cleaner technologies. Figure 1.2 and 1.3 illustrate the stereotypical artisanal miner, an independent alluvial miner who pans the river with the simplest technology and adds mercury to amalgamate the grains of gold dust. In fact this man operates a gasoline powered river dredge, which pumps sediment from the riverbed onto a sluice deck with metal gratings and special miner’s carpets that capture the gold from the slurry stream (Figure 1.4). The gold concentrate is then washed out of the carpets, and mercury is added to the concentrate and panned to further concentrate it.   Figure 1.4: River dredge and sluice at work in Colombia. The operator (left) swings the dredge back and forth along the riverbed to dredge new material. Miners wash the carpets (right) onto the blocked sluice bed, just after adding mercury to the concentrate sand. Stirring the concentrate helps drive off the fine, light material and mix in the mercury. © Paul Cordy photo.  Figure 1.5: Quickly shovel panning the river sediment at the location where the mercury laden concentrate was panned reveals that some mercury (the silver spot on the right hand side of the black sand) is being released directly to the river by this process. © Paul Cordy photo.  4 Like most artisanal and small scale gold miners, this miner does little to prevent the release of mercury (Figure 1.5). However, applying mercury only to the concentrate produces much less contamination than adding mercury to the sluice itself. The application of mercury to all of the primary material without concentration is known as whole ore amalgamation. In larger operations with higher operating expenses, mercury costs are dwarfed by fuel, labour and equipment costs. Thus larger operations tend to be relatively worse polluters as excess mercury is applied in the hope of maximizing gold recovery. This produces significant mercury losses from steady attrition of the material into the effluent, likely taking some of the gold with it (Veiga and Baker 1994) (Figure 1.6).   Figure 1.6: A series of 3 excavators in Colombia dig out the alluvium and pass it up to the sluice, where water jets make it into a slurry. The Z-sluice is lined with carpets that concentrate the gold, but unlike the dredge miner these people add 50 kilograms of mercury to the carpets to trap more gold. As a result they lose considerable volumes of mercury to attrition. © Paul Cordy photo.  Sometimes sluice beds are lined with copper amalgamation plates to capture and retain more gold. Miners apply a silver nitrate solution to the surface of the copper plates to form a consistent surface of silver-mercury amalgam that increases gold trapping, but mercury is still lost by attrition as ore particles scrape some of the mercury off the surface of the plates. The surface of each plate is ‘activated’ using mild acids or detergents to increase the surface tension and coalescence of the mercury, both of which increase amalgamation efficiency (Veiga and Baker 2004).  5 Whole ore amalgamation in hard rock mining (Figure 1.7) consists of adding mercury when milling the ore to maximize the mixing and amalgamation time.    Figure 1.7: Hard rock artisanal mining in Bolivia: ore is mined from quartz veins in mine shafts first excavated in the colonial era (visible as giant cave mouths in the cliffs in the top left image. Note the cable car carrying ore down from these mine openings high above the glacier at an elevation of 5000 metres). Mercury is added to the ball mills at the beginning of milling (top left), and it flows out with the slurry (bottom right), later to be concentrated by sluicing and panning as in the alluvial examples above. The lower two photos show the ball mill circuit where ore is ground progressively finer and then discharged into pans to be moved manually to subsequent milling or sluicing. The bottom left photo shows the large primary mill for coarse material, the wheelbarrow used to take the discharge of that material to the four smaller mills where the material is milled to a smaller size with mercury. © Paul Cordy photo.  6 This means that the mercury is pulverized for several hours along with the rock, and vaporized by the heat generated during grinding. Fine mercury droplets are then lost when the slurry is discharged. In both alluvial and hard rock contexts, whole ore amalgamation ensures that all of the mine tailings are contaminated, and all effluents contain high concentrations of mercury (Telmer and Veiga 2009). Although simple, relatively inexpensive and rudimentary methods can improve both gold recovery and reduce contamination, miners have little access to technical assistance and are reluctant to change their methods without direct evidence that new methods will be effective with their ore. Finally, regardless of the manner in which it is obtained, the amalgam must be heated to evaporate the mercury. Often this is done with a blowtorch at the mine site to get a sense of the quantity and quality of their gold before selling it, and sometimes this is done in a retort that condenses and collects the mercury for reuse.    Figure 1.8: Gold shops are usually nested among other businesses in busy town centres, as in this picture of the main street of Puerto Maldonado, in the Peruvian Amazon. The two doors on the right are gold shops, the one in the middle is a hotel, a place to buy cell phone credits and the home of these two children. On the left is a bus company. In the Universal Oro shop there is a silver fume hood whose chimney empties into the room just above the shop operator’s head. © Paul Cordy photo.  7 Burning amalgam produces a “doré” which is mostly gold but still contains other impurities such as copper, silver, lead, and typically 2 to 5% mercury (Veiga and Hinton 2002, Veiga and Baker 2004), whereas raw amalgam can contain 40 to 50% mercury in the study areas presented here. Mercury in the amalgam is evaporated in gold shops located in urban centres where the emissions pose significant risk to human health, even among those not directly involved in the gold trade. Gold is always heated or melted inside the gold shop with a propane torch to release any residual mercury that might be left over from the amalgamation and evaporation processes. In some towns in less developed countries, the whole amalgam is decomposed in gold shops, which exacerbates the human health risk. The resulting airborne contamination is especially hazardous, since gold shops are usually centrally located in populated areas for economic and security reasons (Figure 1.9).    1.2 Mercury mapping and human health  Maps of mercury contamination have been made for old mercury mining sites in Slovenia, Spain, USA, and Clor-Alkali plants in Sweden (Wängberg et al. 2005, Gosar et al. 2006, Grönlund et al. 2005). They show airborne concentrations of 0.01 to 1 µgm-3. The highest concentrations (>20 µgm-3) were recorded in Almadén, Spain, one of the largest mercury mines in the world, which was mined for more than 5000 years (Ferrara et al. 1998). These values, though extreme for the above contexts, are typical of those found in urban artisanal mining centres, especially where amalgam is burned without capturing the evaporated mercury (Table 1.1).  Normally in the Southern Hemisphere, atmospheric levels of Hg in rural areas are about 0.0009 to 0.0015 µgm-3, and about 0.01 to 0.02 µgm-3 in urban areas (Pirrone and Mahaffey 2005). The World Health Organization (WHO 2003) considers exposure to an annual average mercury concentration of 0.2 µgm-3 to be tolerable, whereas an annual average of 1 µgm-3 or greater is considered hazardous to human health. Arithmetic averages are commonly used in meteorological and public health applications (WMO 2013, WHO 2007, UNESCO 2013 Scire 2000) and this research follows that precedent to  8 facilitate comparison with established norms, except where geometric means are explicitly mentioned.  Table 1.1: Summary of previous studies of ambient gold shop mercury concentrations. Location [Hg] (µgm-3) City/Country Source  Rural site 0.01-0.94 Alta Floresta, Brazil Marins et al. 2000 Urban site 0.03-1.85 Alta Floresta, Brazil Marins et al. 2000 Near gold shops 1.4-18.6 Pocone, Brazil Marins et al. 1991 Street near burning 3.2 Porto Velho, Brazil Malm et al. 1990 Inside gold shop 5.25 Porto Velho, Brazil Malm et al. 1995 Inside gold shops 26.3-53.4 El Callao, Venezuela Garcia et al. 2006 Inside gold shops 0.14-2.3 Alta Floresta, Brazil Marins et al. 2000 Inside burning shop 10.7-17.5 Rondonia, Brazil Malm et al. 1991 Inside burning shop 0.07-40.6 Alta Floresta, Brazil Hacon et al. 1995   Mercury vapour impacts human health most directly, causing problems with the respiratory tract in short-term exposure to high levels of mercury vapour. Symptoms of chronic exposure include chest pains, dyspnoea, cough, haemoptysis, impairment of pulmonary function and interstitial pneumonitis (Stopford, 1979; Levin et al., 1988). Instantaneous exposure to mercury vapours at concentrations of 1200 to 8500 µgm-3 for several hours can be fatal (Jones, 1971, Solis et al. 2000). WHO (2003) states that mild subclinical signs of central nervous system toxicity can be observed among people who have been exposed occupationally to elemental mercury at a concentration of 20 µg/m3 or above for several years. While direct vapour exposure has been linked to physiological harm in miners (Tomicic et al. 2009), less is known about the effects of extremely acute but intermittent exposure in non-miners. On an ecosystem level, Tschakert and Singha (2007), Tschakert (2010), Howard et al. (2011), and numerous others have shown that mercury, especially methylated mercury, can accumulate in sediments, biota, and humans, causing long-term neurological and renal degeneration in the latter.  The cheapest and most immediate way to reduce the mercury vapour exposure of gold shop workers is to improve ventilation in the shops (Drake 2001, Malm et al. 1998, Bastos et al. 2004, Garcia et al. 2006). However, this strategy does not necessarily  9 address the emissions themselves, or public exposure to mercury. Anecdotal reports from individuals in communities with intense artisanal gold mining suggest unusually high incidences of cerebral palsy and learning deficits in children, as well as kidney failure in adults (Daniel Moraga, Chile, personal communication 2009, Oseas Garcia, Colombia, personal communication 2011). Investigation of these claims would require methods that show the variability of airborne mercury concentrations in space and time, as well as epidemiological studies to describe disease patterns among affected populations. Gold shop owners, businesses, governments and non-governmental organizations have independently developed many different mercury capture devices in various places. There are several basic designs, most of which are powered by electric fans or compressors. “Bubbler” condensers draw the vapour through a container of water (UNIDO 2007, Figure 1.9).  “Baffle” condensers instead use baffle plates (Argonne National Laboratory (ANL) 2007) to capture mercury aerosols without using water. “Scrubber” condensers draw the vapour through a rain or mist sprayed into a section of the exhaust pipe, such as is used for fly ash in smokestacks. Conventional fibre filters are also used, but safe disposal of used filter media is a significant problem. The United States Environmental Protection Agency (USEPA) tested the efficiency of one UNIDO bubbler (84% efficient) and four of their own baffle devices (average efficiency: 76%; range:  37-91%) by sampling the fume hood exhaust before and after burning amalgam of known mercury content (ANL 2007).  Figure 1.9:  bubbler style gold shop filter installed in Andacollo. spigotSerrated tube end submerged in waterwater line just above serrationscapped hosefor water re!llfanfume hood 10 Though it is important to capture as much mercury as possible, even a 90% reduction in mercury concentrations during amalgam (with 50% Hg) burns would likely still exceed the WHO lowest observed effect level of 20 µgm-3 for 8-hour worker exposure (WHO 2003). Therefore, even the most efficient mercury capture devices are likely insufficient to reduce mercury exposure to acceptable levels, at least for workers in direct contact with the vapours. Until now, mercury emissions from gold shops had not been characterized in terms of quantity or character, or in terms of persistence or dispersion in the atmosphere.  1.3 Mercury dispersion modeling  The nature and extent of mercury dispersion from artisanal mining activity has received little attention in the scientific literature and remains a significant source of error in global circulation models (UNEP 2013). Though it would be beneficial to use dispersion models to estimate patterns of local human exposure and explore the extent to which mercury is transported in the atmosphere, there has never been any attempt to apply dispersion modeling to artisanal mining emissions.   Marins et al. (2000) and Hacon et al. (1995) monitored soil burdens and airborne mercury concentrations at various distances from gold shops in the Brazilian Amazon and concluded that they are not a major source of mercury contamination in rural areas far (>2 km) from urban gold shops.  Garcia et al. (2006) also showed that the airborne and soil concentrations of mercury decrease sharply and in concert as one moves away from gold shops, but remain significantly above background up to 1 km away and possibly more.  Wang, Shi & Wei (2003) also support this finding, but others claim that the majority of Hg emitted from gold shops is deposited well within 1 km of the source (Lacerda 2003, Malm et al. 1995).  Some argue that the high boiling point and atomic weight of mercury cause its rapid removal from the atmosphere. However, mercury concentrations in soil surrounding gold shops in the Brazilian Amazon attenuate rapidly (Centro de Technologia Mineral (CETEM 1992), and rough integration of the soil concentrations indicate that soil mercury accounts for less than 2% of the 70 tonnes  emitted annually by the shops (Telmer et al. 2009).  11 It would be reasonable to suspect that the rest of the mercury in this case must be riding the powerful tropical thermals and mixing with the ~2 km deep atmospheric boundary layer (Stull 1988), where it is diluted, transported, and eventually contributes to raising the global mercury background. This interpretation is supported by measurements of mercury in the atmosphere made by airplane over the Amazon Basin (Artaxo et al., 2000) which indicate that gold mining areas contribute 63% of the total atmospheric Hg over the Amazon. Gaseous elemental mercury (GEM) is by far the most common form of mercury in the atmosphere, and it has an atmospheric lifetime of several months to a year. Most natural and anthropogenic sources emit GEM, which reacts relatively slowly with common atmospheric oxidants such as ozone (O3) (Mason et al. 2012). It reacts much faster with radicals such as OH- and Br- (or BrO), but these occur in very low concentrations in the troposphere and knowledge of the oxidants and rates of reaction is still limited. Hence GEM is subject to long-distance transport, whereas gaseous oxidized mercury (GOM) is, for the most part, dry or wet deposited close to the point of emission or formation (Subir et al. 2011, 2012). Thus mercury deposition to terrestrial and marine ecosystems is dominated by GOM, both via the direct deposition of gas phase species and through wet deposition of oxidized mercury compounds in precipitation (UNEP 2013). There is mounting evidence to support the idea that much of the methyl mercury (MeHg) found in biota originates by methylation of mercury in the water column, and the most important source of mercury to the world oceans is deposition from the atmosphere. It has been suggested that there is possibly a linear relationship between the inorganic mercury concentration in the ocean and the amount of MeHg formed in the upper waters of the ocean (Mason et al. 2012). Most models focus on global (Ryaboshapko et al. 2007, Travnikov 2005, Dastoor and Laroque 2004, Gbor et al. 2006), and continental scales (Pirrone et al. 2003, Cohen et al. 2004) but little effort has been devoted to local scale models of Hg transport in the air.  The few models that have simulated Hg transport at scales <100 km (Constantinou et al. 1995, Hodgson et al. 2007) focus on low concentrations emitted by chlor-alkali plants, industry, or natural sources.   Long-range transport of Hg is better understood, (Arctic Monitoring and Assessment Programme (AMAP) 2008, UNEP 2013) although many focus on predicting  12 annual average concentrations and fluxes on regional or global scales (Travnikov 2005, Pai et al. 1997, Ryaboshapko et al. 2002). Ryaboshapko et al. (2007) note that according to models, only 5 to 7 days would be required for significant intercontinental transport of mercury to occur, and that most models require only 1-2 years of spin up to reproduce observations of concentrations and fluxes with reasonable accuracy. Large-scale modeling provides useful insights, however it tends to neglect small-scale air circulations and planetary boundary layer structure. There are no studies that verify local scale mercury dispersion models, although CalPUFF (a Gaussian puff based dispersion model that is widely used for regulatory applications) has been used for microscale dispersion of other pollutants (Heckel et al. 2011, Oshan et al 2006).  In theory, atmospheric modeling should be able to simulate patterns and intensities of urban artisanal mining mercury contamination with reasonable accuracy if a sufficiently simple field context can be found. Modeling dispersion at fine scales in complex urban environments is still a very significant challenge in atmospheric science. Emissions data are difficult to obtain given that artisanal mining tends to be informal or illegal, and urban structure is often irregular and complex in artisanal mining towns. These are among the challenges in generating and verifying predictions of mercury concentrations. If mercury dispersion modeling could be done with reasonable accuracy, it could be used to better understand the persistence and behaviour of mercury plumes, estimate public health hazard, and guide mercury mitigation (for instance by assisting with site selection for relocation of mercury emitting industries).  1.4 Mercury mitigation  The UNIDO Global Mercury Project (GMP) is the largest development project ever to focus on artisanal and small scale mining. This project, in association with various other funding agencies, undertook interventions in Brazil, Venezuela, Guinea, Mozambique, Sudan, Tanzania, Zimbabwe, Indonesia, Lao PDR (Spiegel and Veiga 2006), and most recently, Colombia (Cordy et al. 2013). The GMP asserts that it should be possible to achieve at least a 50% reduction of mercury consumption in ASGM by 2017 by focusing on the following priority actions (Spiegel and Veiga 2006):  13  1. Elimination of whole ore amalgamation by introducing mercury-free concentration process prior to amalgamation.  2. Reducing mercury use in the amalgamation of concentrates through closed circuit processes in which mercury is efficiently recycled.  3. Elimination of mercury vapour releases through the use of retorts to contain and recycle emissions.  4. Introduction of completely mercury free techniques wherever the ore characteristics allow it.  The GMP tested and demonstrated the value of these actions over a decade in 10 countries. It is estimated that the gold shop filters, retorts, and mercury reactivation practices introduced in Indonesia and Brazil prevent 1-2 tonnes of mercury from being released annually in each country (McDaniels et al, 2010). Other achievements include formal training programs for thousands of miners worldwide, public health campaigns to educate governments and non-miners about the hazards of mercury, and dissemination of mineral processing techniques that improve gold recovery without the use of mercury. Although most of the mercury emitted by artisanal miners is lost in the process of whole ore amalgamation, remediation programs often focus on promoting the use of retorts because amalgam burning is the primary health hazard for miners and is also a significant source of contamination (Veiga and Baker 2004). The GAMA project (Gestion Ambiental para Mineria Artesanal) in Peru and Ecuador developed and built a series of communal retorts that are large, fixed structures which condense vapours by blowing them through several tubes immersed in a large water tank.  Mercury can also be replaced entirely by using cyanide in the same ball mills that miners use already (Veiga et al. 2009). This mill leaching process requires very little additional investment and time, yet potentially obtains much more gold than mercury amalgamation. Furthermore, milling with hydrogen peroxide before discharging the tailings destroys the cyanide, thereby greatly reducing environmental contamination. Increasingly, artisanal mining development projects focus on integrating miners into the formal economy. As most miners operate illegally, they have neither access to  14 credit or technical assistance, nor incentive to invest in cleaner production technologies or training. GAMA made some progress toward formalization of artisanal miners in Peru by assisting the government in revising mining laws such that they acknowledge artisanal miners and lower the barriers to entry into the formal economy (GAMA 2009). However, even artisanal mining operations that have the money and organization to formalize and eliminate mercury use may remain illegal because they cannot obtain legal mining title or negotiate with title-holders (Spiegel and Veiga 2010). GAMA also made progress in this respect, by helping miners reach formal agreements with existing mining title-holders.  Gran Colombia Gold, a large gold mine in Segovia, Colombia, also entered into partnerships with the artisanal miners who were illegally operating in their deposit. They agreed to provide mine safety, social security, health insurance, and to help them to incorporate in exchange for selling their ore directly to Gran Colombia’s processing plant. This is a rare example of cooperation between formal and informal mining sectors that holds promise. Technical assistance, elimination of whole ore amalgamation, mercury vapour capture, formalization, and alternative extraction technologies all contributed to the success of the GMP project in Colombia, where this research shows that airborne concentrations of mercury have been significantly reduced (Cordy et al. 2013), and an estimated 10 tonnes of mercury was recovered or prevented from release into the environment (Oseas Garcia, personal communication 2012).  Large and successful as these projects have been, there remain millions of people worldwide who live, work, shop, and go to school in the urban centres of mining areas and are exposed to mercury vapours. The degree and distribution of this hazard have never been mapped or modeled. Mitigation of mercury vapour emissions is difficult to implement and there have been no studies that measure reductions in environmental contamination in the receiving air, water or landscape. This work aims to fill these knowledge gaps.  1.5 Objectives and research questions  The goal of this work is to provide the first detailed analysis of urban mercury emissions from artisanal and small scale mining at the urban scale. It investigates the source characteristics, distributions and dispersion of mercury vapour in space and time,  15 as well as strategies for mitigation. These are the key questions that guide this research:  o How are mercury emissions from urban gold shops generated, released, and dispersed?  o What level of risk do urban atmospheric mercury emissions pose to people, and how can this risk be measured and communicated?  o Can atmospheric dispersion models be used to estimate the distribution and behaviour of urban mercury vapour at neighbourhood to urban scales, and what are the principal challenges to such modeling?  o How can these emissions be mitigated, and is it possible to measure the impact of mercury reduction initiatives?  1.6 Approach   This dissertation uses field measurements and dispersion modeling to better understand the nature and behaviour of urban mercury vapour emissions from artisanal mining, and to assess the impact of mercury mitigation strategies. It begins by illustrating the character and severity of the problem in and around select artisanal mining gold shops through site visits and opportunistic mercury sampling.  I investigated gold shops in Andacollo, Chile; Segovia, Colombia; Puerto Maldonado, Peru; Ponce Enriquez, Ecuador; and Paramaribo, Suriname. The first three towns provide small glimpses into the variety of manifestations of this phenomenon across South America, whereas I study Andacollo and Segovia in greater depth. Both of these towns are located high in the Andes and have centuries-old traditions of gold mining and raw amalgam burning. Organizational, behavioural, and production data were collected through interviews with miners and local authorities, and confirmed by direct observations of mercury use as well as information from governments and the United Nations.   16  I measured airborne mercury in Andacollo, Chile, and Segovia, Colombia, using regular automobile (and occasionally bicycle) transects and measurements from fixed towers. The goal was to characterize the nature and behavior of emissions in urban areas, and the distribution of human exposure. Modeling of gold shop emissions in Andacollo provides a deeper understanding of the shape, extent, and persistence of mercury plumes, and corroborates conclusions about the characteristics of gold shop mercury emissions that are drawn from observations.  Finally, the mapping techniques elaborated here were used to detect reductions in mercury vapour concentrations during the past three years of UNIDO’s mitigation campaign in Segovia.  1.7 Applied contributions of the research  This work will be of greatest interest to people and organizations who are engaged in interventions that aim to mitigate mercury emissions from artisanal and small scale mining. These people will benefit from having a detailed understanding of gold shop practices in various countries, and the resulting risk they pose to human health for both workers and residents. Mercury released from mining is commonly thought to be a problem for miners and receiving environments where gold is extracted (usually away from urban centres). This dissertation clearly shows that living in urban mining towns also presents a serious health risk to non-mining populations. Epidemiologists will want to use these monitoring techniques in conjunction with human health assessments to establish a relationship between mercury distributions, daily habits, residence location, mercury body burdens, and health outcomes. Those engaged in reducing mercury emissions will be most interested in knowing that it is possible to succeed in doing so even in highly complex and insecure situations. Previously, mitigation efforts had to be satisfied with indirect indications of project success, such as reporting the number of miners trained and estimating the mass of mercury that could be recovered if mitigation measures are consistently used (McDaniels et al. 2010). This work provides a new means of directly measuring decreases in  17 environmental contamination from urban mercury sources, and it is the first to demonstrate the effectiveness of mitigation efforts.   Atmospheric scientists will be interested in seeing a new micro-scale application of a standard regulatory dispersion model, and to see the limitations of this model that are revealed by the unique source characteristics of urban mining emissions. This work also demonstrates that urban mining emissions present daily mercury tracer experiments around the world that could be used to study pollutant dispersion in complex urban environments. Researchers interested in taking up this challenge would benefit from knowing of the difficulties and opportunities they are likely to encounter. In the event that mapping of mercury concentrations is not practical, dispersion modeling will offer those interested in human health risk a means of estimating mercury distributions given the location of sources and estimates of gold production. If mercury emitters are clandestine, the mercury mapping techniques used here can reveal locations of regular emitters to within a few hundred metres or less.   1.8 Limitations of the research  Although this work shows that there are several commonalities to artisanal gold shop practices, it is far from a comprehensive study of South American gold shops. I have established rough bounds for the magnitude of the problem by studying locations with exceptionally high and relatively low rates of gold commercialization, as well as the resulting frequency and intensity of emissions in each place, however there is likely to be a great deal of variability in the severity and characteristics of mercury contamination at other sites. Monitoring techniques used here will not be practical in all situations, and may not reliably detect the frequency and distribution of emissions from individuals who only occasionally burn mercury in their homes or offices.  This work also endeavours to probe the limits of scale in dispersion modeling in an urban environment. Urban dispersion complexity is only roughly parameterized; therefore one would not expect to see a high level of predictive accuracy even if observation data were dense and continuous. However, sparse roving observations further  18 complicate model evaluation, and therefore this research does not provide a confident quantitative validation of the dispersion techniques presented. Instead it elucidates the challenges presented by urban mining pollution dispersion and establishes the analytical techniques that are necessary for conducting this kind of work. In spite of these difficulties, applying dispersion models to new situations in which mercury is emitted could nevertheless provide important qualitative information of human health hazard. Given that miners and gold shops in Colombia openly burn amalgam and use whole ore amalgamation within a small and concentrated urban zone, there is the greatest potential for reduction of mercury consumption and contamination there. Other more sparse sites that already use somewhat cleaner techniques are unlikely to produce such dramatic reductions as shown here, though the alternative processing methods, policy changes and mercury capture devices could potentially be broadly applied. The main limiting factor to success of remediation initiatives tends to be the degree of cooperation from governments and mining associations (McDaniels et al. 2010).  1.9 Thesis structure   This dissertation begins with descriptions of the field study locations in which the research took place, followed by separate chapters for each of the main components of this work: mapping, modeling, and mitigation. They are presented in this order because the analytical techniques and findings presented in each chapter are used in or contribute to the understanding of subsequent chapters. Each chapter is quite distinct, and therefore I introduce the background and methods at the beginning of each chapter so that they are fresh in the reader’s mind when evaluating the results.     19 2 Field study sites and gold shop contamination   All of the sites investigated are major centres of artisanal mining where amalgam is burned in urban areas. Though they vary in the intensity of emissions and the amount of gold commercialized, they all exhibit rudimentary mining practices and little or no measures to limit the release of contaminants (Cordy et al. 2011 and 2013, Swenson et al. 2011, Higueras et al. 2005, Veiga 1997, Velasquez 2010). All sites (Figure 2.1) except for Andacollo, Chile, and Puerto Maldonado, Peru, were visited on the behest of UNIDO in order to assess the degree of contamination and study the mining methods used. Studies in Peru were undertaken in cooperation with the USEPA who were investigating the site for installation of gold shop filters. Together, these sites provide a broad understanding of the many commonalities of urban sites with frequent mercury emitters as well as a sampling of the many particularities of such sites. In each town, gold shops and miners could recall visits by internationally sponsored mercury mitigation initiatives and scientists conducting mercury studies. Most of these previous studies focus on characterizing the scale of contamination, distributing retorts, training miners, and measuring mercury body burdens in people and fish, with little attention devoted to gold shops (notable exceptions: USEPA (ANL 2007), GAMA (2009), UNIDO (2007), and some UNIDO work shown here (Cordy et al. 2013)). Miners and gold shop operators are usually very open and congenial, and happy to share their experiences and show us their activities. Sometimes they are slightly bemused and befuddled by foreign interest in mercury, which they often regard as a non-issue by virtue of the fact that they Figure 2.1: Map of field sites in South America.  20 have been using it for hundreds of years without being aware of its potentially debilitating effects. Yet measurements taken in all five field sites reveal this to be a widespread phenomenon with unique regional variations and myriad attempts at remediation whose effectiveness is questionable.  2.1 Mercury vapour sampling methods  The mercury sampling methodology varied depending on the research team, country, and degree of access. Mercury vapour data from in and around gold shops were summarized by considering sample locations that were common to all shops: on the street in front of the gold shop, just inside the front entrance, in the room where amalgam is burned or raw gold melted, and in the office (if one existed). In Colombia, we also sampled a variety of different non-mining buildings and public places. Security and/or health concerns as well as high variability in the receptiveness of business owners and miners required the adoption of an opportunistic sampling program instead of a more systematic approach.  Nevertheless, these measurements provide an impression of the acute exposures in a variety of settings.  Total mercury concentrations were measured using various Lumex RA 915+ mercury analyzers (“Lumex” from CERM3 of The University of British Columbia, Corporación Autónoma Regional del Centro de Antioquia, and the Artisanal Gold Council) and on some occasions a Jerome 431X analyzer (”Jerome” from the Corporación Autónoma Regional del Centro de Antioquia). The Lumex uses Zeeman atomic absorption spectrometry with high frequency modulation of light polarization (Sholupov et al. 2004), allowing 60 Hz sampling frequency from which one second average concentrations are computed internally, and which gives a precision of 0.001 µgm-3. The Jerome analyzer uses a thin gold film whose resistance changes relative to the concentration of mercury in the sample air stream (Singhvi et al. 2001), and gives a precision of 1 µgm-3 for a five minute sampling average. The low (high) detection limits of the Lumex and Jerome are 0.002 (50) µgm-3 and 3 (999) µgm-3 respectively. In statistical calculations that include exceedances of the upper detection limit of the Jerome, the limit value of 999 µgm-3 was used; therefore, these figures should be  21 considered a low estimate. To avoid instrument contamination, the Lumex was not used in situations that could exceed its upper limit (>50 µgm-3 in the more sensitive mode). In cases where this value was exceeded, the observed value is used as if it were true. Hence, these extreme values should be considered low estimates. The Jerome was used to measure concentrations inside gold shops, mineral processing centres, and some non-mining buildings. The Lumex has an internal calibration mechanism that was used at the beginning and end of each measurement period. The Lumex instrument’s self-calibration draws inlet air through a filter medium that eliminates mercury while leaving the remaining chemistry unchanged (Sholupov et al. 2004). At least ten seconds was allowed after this filter was applied so that the signal could stabilize, and then the instrument was recalibrated for at least thirty seconds before removing the filter and waiting at least ten more seconds for signal re-stabilization. This process was frequently repeated in low mercury areas during sampling on foot in order to avoid instrument drift.  Errors in measurement of mercury concentrations are primarily due to response lag. Given a step change in concentration, the Lumex will register 90% of the real concentration in 20 seconds at the instrument’s mean flow rate of 10 litres per second (Joseph Siperstein, Ohio Lumex 2011, personal communication).  Values obtained while sampling in and around gold shops and other locations on foot are ten second averages in the case of the Lumex (which the device computes internally) and five second averages in the case of the Jerome (which measures mercury accumulated on the sensor over a five second period).  All mercury signals are assumed to be attributable to artisanal gold mining activities, as there are no other mercury emitting industries (such as coal burning or chlor-alkali production) nearby and geogenic emissions produce airborne concentrations orders of magnitude lower than those that result from amalgam burning.   2.2 Gold shop mercury contamination  Gold shops are centrally located in the urban centres of mining towns in order to attract more business and because of the relative safety afforded by busy streets and proximity to police stations. Therefore, they are also located in areas that maximally  22 expose urban residents to mercury hazard. Gold shops always burn amalgam and/or raw gold (doré) before purchasing from miners to drive off any excess mercury. Miners insist on maintaining visual contact with their gold for fear of deception. Some miners resist the use of retorts to decompose amalgams because the service is provided at the point of sale anyway. In addition, some burn openly because they feel that retorted gold is inferior, since its colour (usually red or brown) may bring a lower negotiated value. Shop operators burn the raw amalgams with a propane torch in an open pan in front of the miners, and neither perceives the high concentrations of mercury vapour because it is a colourless and odorless gas. Where mercury capture devices exist, open-faced fume hoods still allow significant mercury releases into the shop.  Table 2.1: Typical mercury concentrations (µgm-3, 10 second averaging time) in and near gold shops in Peru, Ecuador, Suriname (measured with Lumex; 10 second averages), and Colombia (measured using Jerome; five second averages). “Outside” is within 10 m of the front door; “entrance” is just inside of the front entrance. Country/  Relative location   filter type  outside entrance burn room office Peru (n=8) mean 10.4 1.4 135.3 15.1 bubbler range 0.8-50+ 0.3-2.4 3-44 0.6-50+ Suriname (N=4) mean 1.3 26.0 28.7 2.9 bubbler  range 0.04-25 9-50+ 13-50+ 41.0 Ecuador (n=2) mean 13.6 5.5 41.6 27.7 no filter range 0.1-3 1.5-10 38-50+ 5-40 Colombia (n=27) mean 23.2 n/a 121.8 n/a fibre filters range <3 -95 n/a <3 -  >999 n/a  2.3 Andacollo, Chile  The town of Andacollo is located in the Northern-central part of Chile, ~50 km east of the cities of Coquimbo and La Serena, and ~500 km from Santiago (Figures 2.2 to 2.3). Andacollo has been mined for copper and gold since the time of the Inca Empire (Higueras et al. 2005).  At the end of the 16th century, quartz vein gold ore started to be mined and processed using the Miller pans (also known  as “Chilean mills” or “trapiches”) that consist of two cement wheels with steel rims weighing 1 tonne/wheel  23 that crush and grind the ore to a grain size less than 0.1 mm (Figure 2.5). As recently as 15 years ago there were an estimated 5000 artisanal gold and copper miners (“pirquiñeros”) in Andacollo, though since the prime mineral properties were awarded to foreign mining companies and the gold prices plummeted in the late 90s, the number of artisanal gold miners dropped to a minimum of around 200 in 2008. This number has likely increased since then.  Unfortunately, despite reduced gold production, there remains a significant exposure risk to the local population because many gold shops are located in residential areas and miners burn small amounts of amalgam in their homes.  Three gold shops are located within a two-block radius of an elementary school.  There are two medium-size mining companies in the town. Dayton Mining uses cyanidation in heap leaching to extract 27 kg of gold per day, and Carmen de Andacollo Mine uses sulfuric acid heap leaching to extract copper from oxidized ores. Other mines around Andacollo are leaching or floating copper minerals. There has never been a detailed inventory of gold production, processing plants, the number of people involved, or the amount of mercury consumed. Higueras et al (2005) reported that, in 2004, there were 8 processing plants with 30 trapiches operating in Andacollo. At the time of this study it was estimated that there were at least 10 to 15 processing plants, all of which use mercury. One of the plants is also a tourist attraction and the artisanal miners have an entire educational program to demonstrate to the visitors how they mine (using dynamite) and process the ore (using amalgamation). Miners excavate the gold-mineralized quartz veins using dynamite, and the ore is removed with wheelbarrows, buckets and pulleys.  Miners manually pulverize a part of the ore, pan it in a small batea or shovel and visually estimate the concentration of gold in the vein to be mined (Figure 2.4).  Figure 2.2: Map of northern Chile.  24   Figures 2.3: Andacollo from the Southeast (top), and from the North (bottom). Dayton and Carmen de Andacollo are large Canadian owned mines. The gold shops are primarily located in the area around the stadium. Tan and greenish piles are mercury contaminated tailings. © Paul Cordy photo.  Figure 2.4: Miners chase 1-2 cm wide quartz veins underground, the vein is visible in the top left as a thin red tinged line. Top right, miners grind the vein ore by hand and pan it in a shovel (bottom right) to determine the approximate grade. Gold is clearly visible in the centre of the lower right image. © Paul Cordy photo.  25  Figure 2.5: A typical Andacollo processing centre, with Chilean mills. They are no longer powered by mules, however urine is still used to restore the amalgamation efficiency of the copper amalgam plates, as the foreground miner is doing with his bare hands. © Paul Cordy photo.  Their estimated grades usually range from 30 to 60 g of gold per tonne. The extraction process is very rudimentary and it is estimated that a typical miner may extract on average 120 kg per day of ore. According to the Andacollo Pirquiñeros Association, miners in Andacollo produce between 0.3 to 0.6 tonnes of gold per year. Miners rent a processing plant for 1500 pesos/hour (or USD 3.3/hour, 1 USD = 460 pesos in 2008) to grind, concentrate and amalgamate the gold. The initial crushing procedure is done manually in order to reduce the ore to pieces below 10 cm so that they can be fed to the trapiches. Mercury is added to the trapiches during the grinding process and a large part is pulverized and lost with tailings. Part of the mercury forms an amalgam with free gold and is trapped on the copper amalgamation plates that are placed along the edge of the trapiche basin. Tailings are sold to other plants that use flotation in order to recover residual gold associated with sulfides (Figure 2.6). The ground tailings (below 0.1mm) from the trapiches are either floated using xanthate and pine oil or panned. No gravity concentration is used in the town. A small trapiche can process a bag of 60 kg of ore per hour. Plants usually have 4 to 10 trapiches with a capacity of 3 to 5 tonnes of ore per hour.   26   Figure 2.6: Copper amalgam plates line the milling basin and traps fine gold in suspension in the slurry (left). Flotation cells (right) concentrate the gold bearing sulfide minerals that stick to oily froth. © Paul Cordy photo. Amalgam is removed from the copper amalgamation plates by scraping the surface of the plates with a piece of plastic, and applying large amounts of urine. The amalgam collected in the bottom of the milling pan during the grinding process is recovered by panning. Excess mercury is filtered through a piece of cloth, and the resulting amalgam sold in a shop. If the amalgam is large enough to contain amounts of mercury that are worthwhile recovering, they will burn the amalgam in a local variety of retort at the mine site or in the yard of their house before selling it (Figure 2.7).  Miners estimate that they lose 1 gram of mercury per gram of gold produced, but based on the types of practices they are using, the ratio of Hglost:Auproduced is probably at least 10:1 (Veiga and Baker 2004). Higueras et al. (2005) reported that 30 to 50% of the mercury introduced in the process is lost, and that mercury vapour concentrations reached 100 µgm-3 during their Lumex measurements in the processing plants. This concentration is typical within metres of a site burning mercury in open-air.  27  Figure 2.7: The traditional Andacollo retort. The amalgam is placed in the inner pipe and a wood fire is built inside the large lid (triangular holes feed air to the fire). The basin underneath is filled with water which condenses the mercury vapour. © Paul Cordy photo. Higueras et al (2005) reported mercury concentrations of 0.05 µgm-3 in downtown Andacollo, and background mercury concentration of 0.007 to 0.01 µgm-3 in the region due to naturally occurring mercury minerals.  We were told that most miners usually do not bother to burn their amalgam, as they receive that service for free in one of the local gold shops. During our first visit in 2008, representatives of the Chilean Ministry of Environment (Comisión Nacional del Medio Ambiente: CONAMA) with whom we were touring the site told us that no one would tell them the location of the shops and there was no official record as they are not regulated. In later field missions, unaccompanied by government officials, we found that the principal gold shop was located beside an elementary school, and people were far less secretive about gold shop locations. Eventually, I was able to get some gold shop operators to record the mass of their amalgam before and after burning.   28  Figure 2.8: On Wednesdays, a fruit vendor sets up his stand within 200 metres of two gold shop smokestacks, one of which is visible on the right. The wind commonly flows from the stack directly towards this location. © Paul Cordy photo.  One of the plants we toured uses trapiches and flotation, and the owner also has his own mine. He charges other miners the equivalent of USD 21.7 per day to process their ore. He mills 60 kg/h of ore in the trapiche without adding mercury to the milling pan, instead only placing copper amalgamation plates around the mill. The ground material goes to a steel flotation cell operating with xanthate (USD 9.3/L in 2008) and pine oil (USD 4.4 /L in 2008) to produce 80 kg of sulfide concentrate daily (grades from 50 to 100 grams per tonne). The flotation technique is also rough, and they are visibly losing sulfides in the tailings. The concentrate is sun-dried in shallow pans and sold to the Government-owned company Empresa Nacional de Minería (ENAMI) in Coquimbo. In 2008, a gram of gold of average purity for the region is sold locally for about 10,500 pesos (USD 22.8/g). When miners sell their flotation concentrate to ENAMI, they are better paid since the company analyzes the gold in the concentrate and pays extra for the silver and copper contents. We were told in 2008 that ENAMI paid USD 32.6 (or 15,000 pesos) per gram of gold found in the concentrate from typical concentrate. On February 28, 2008, the international gold price was USD 970/oz or 31.2/g in New York (Kitco 2008).   29  Figure 2.9: Centuries of mercury contaminated mine tailings are stacked all around town, including right beside the Basilica. © Paul Cordy photo.  There were three main mercury dealers in Andacollo in 2008, selling 1 kg of Hg for 38,000 pesos or USD 82.6 when the international price was around USD 22/kg. It is believed that gold shops also sell mercury to the miners. Considering the typical Hglost:Auproduced ratio of 10:1 for operations using trapiches and copper plates, and considering the gold production in Andacollo is estimated to be between 0.3 and 0.6 tonnes annually, the amount of mercury consumed and released in Andacollo might be around 3-6 tonnes annually. Health of local residents is likely to be affected by this contamination; however the only evidence is that of a recent unpublished work by Floria Pancetti, from the Chilean North Catholic University, entitled “Prevalence of neurological diseases and neuropsychological damage in the population of artisanal gold workers exposed to mercury.” This study revealed that of 36 Andacollo miners examined by medical doctors, 26 showed neurological symptoms. In the same study they observe that out of a total population of about 14000 people, 30 children of Andacollo are afflicted with cerebral palsy (the base rate in the United States is 2 to 2.5 for every 1000 births (Krigger 2006)).  30 2.3.1 Andacollo gold shops  The gold shop operators themselves are well aware of the dangers of mercury inhalation, but seem not to believe that there are harmful quantities of mercury in the gold they purchase from miners. Large quantities of amalgam, for which recovery of mercury is economically significant, are sometimes burned using retorts.  Normally, they burn the amalgam without retorts under fume hoods that have no fans.  The mercury vapour either escapes via the chimney into the urban atmosphere or stays in the enclosed burning rooms, where mercury concentrations exceed the detection limit of the Lumex Vapour Analyzer (>50  µgm-3 at 3 metres from the amalgam burning). Impurities are eliminated by boiling the doré in nitric acid, which solubilizes metals like silver and copper. In one gold shop where this practice is used, the nitric acid fumes over time dissolved the fume hood (a steel oil drum with an iron chimney, Figure 2.10) and tin roof above it, causing the heavy metal chimney to collapse and damage the adjacent roof. The neighbours near this shop complain that on cold days they can smell the acid mist and feel it burn their nasal passages and throat.    Figure 2.10: The burning room of the principal gold shop in Andacollo. The left hand fume hood was used for boiling acid, as is evident from the heavier corrosion relative to the mercury fume hood. Before the corrosion-weakened smokestack collapsed (top left), and afterwards (bottom left). Corrosion of the roof and hood (middle) explains why the stack collapsed, damaging the neighbour’s roof.  A bubbler style condenser (right) was installed during the 2009 field campaign in the principal gold shop in Andacollo. Many aspects of the design, such as the fume hood face opening, are a compromise solution at best. The fan (lower right) exerts negative pressure on the water basin, which draws air out of the fume hood. The blue pipe exhausts to the outside at the same height as before. © Paul Cordy photo.  31  The operator of that gold shop, which is also adjacent to the soccer stadium, told us that he no longer uses acid dissolution if there is a soccer game on, as people had complained of similar irritation of their airways during games. This is all preliminary information, and a detailed health and engineering investigation would be needed to properly evaluate mercury emissions and public health risk.  The key information is therefore the timing and mass of emissions, which are understandably sensitive and private data. The openness and cooperation of the miners (Pirquinieros) and their miners association (Associacion de Pirquineros de Andacollo) was therefore essential, and we are very grateful to the Association’s leadership for encouraging all gold shops to collect amalgam mass and burn time data for us. Three shops provided some data, and one gold shop provided full records of 28 events spread out over the entire field campaign, followed by two more months of intermittent records. This particular gold shop is described by most residents as the “principal” gold shop and I will continue to refer to it as such.  Table 2.2: Average mass of gold and mercury in 62 amalgams sold at three different gold shops in Andacollo from April to July, 2009. Amalgam weights were measured before and after burning. All this mercury was released to the environment due to lack of filters in gold shops.   Average mass (g) Range Total amalgam mass 22.3 1.5-138 Gold recovered 9.93 0.7-118.8 Mercury released 12.3 0.8-77.0 % change 55.4 1.5-76.3 Burn time (minutes) 12.6 5-30 Total burn time (hours:minutes) Hg emitted (g) Average Au:Hg ratio Average % Hg 6:34 620.63 1.25 64  We built and demonstrated several kinds of retorts and gold shop filters for use in the shops and at the mine site. The Indonesian style bubbler that we installed in the  32 principle gold shop (Figure 2.10) is still in use, however we have been unable to obtain information on the amount of mercury recovered in the reservoir of the condenser. Table 2.2 shows the average of all recorded amalgam weights from April to July of 2009.  2.4 Segovia, Colombia    Figure 2.11: Map of Colombia (left), and the gold producing towns in Antioquia (right).  Mercury contamination within dense urban mining centres in the remote, insecure, and mountainous region of Antioquia, Colombia (Figure 2.11) is influenced by a complex array of socioeconomic and political difficulties in the country, and this makes it an exemplary case study. Despite a gradual reduction of poverty during recent years, the UN Human Development Report (UNDP, 2010) states that 16% of Colombians live on less than USD 1.25 per day while the Colombian statistics department  (DANE) estimates that at least 62% of the rural population lived in poverty in 2006 (Perfetti, 2009). Additionally, decades of political, military and gang-related conflict in Colombia have led to the forced displacement of millions over these years, leading to the highest number of internally displaced persons in the world (UNHCR, 2008). The vast majority  33 of the displaced are peasant farmers forced to flee rural conflict zones for the relative security and possible employment opportunities offered by urban centres (IDMC, 2009). The rural areas have also witnessed an increasing number of illegal plantations of coca with relatively few cattle farms. Only 26% of the 45.6 million Colombians live in the rural areas. Gold mining provides rural inhabitants in Colombia with an alternative to poverty, but without technical assistance their mining practices tend to be very rudimentary and generate severe environmental and health impacts.  A great majority of mining activities in Antioquia are illegal, as they lack permits either from the mining or the environmental authorities. As a result of the presence of guerrilla groups in the rural areas, gold ore mined in the surrounding hills is processed in the urban environment. Miners accumulate as little as two tonnes of ore to take to the processing centres, or ‘entables’, which are typically located alongside residences, schools and stores. Commonly miners in entables use whole ore amalgamation instead of pre-concentrating the ore, and mercury is added to the whole high grade material in small ball mills, known as “cocos” (Figure 2.12).    Figure 2.12: Rows of coco amalgamation mills are constantly spinning day and night to mill the ore with mercury. The miners concentrate the mineral by panning in tubs of water or an amalgamation pond, sometimes after sluicing with carpets. They then pour the liquid mercury into a cloth to squeeze out the excess, and sell the resulting amalgam. © Paul Cordy photo.  34  Figure 2.13: Panning the milled ore concentrates the mercury and allows easy separation (left). Squeezing the amalgam in cloth, and catching the excess for reuse. © Paul Cordy photo.  The entables charge the miners a nominal fee of USD 0.50 to 1.00 per coco with the condition that the miners leave their tailings to be further leached with cyanide. Miners obtain only the gold extracted by amalgamation, which leaves significant residual gold for the entable owners. Some entables have 5 to 10 cocos and others as many as 80. These steel ball mills are locally made with capacity of processing 50 to 70 kg of ore. We estimated that in 2010 there were between 2500 and 2700 cocos in 323 mineral processing centres in 5 gold producing municipalities in Antioquia (including Segovia, Cordy et al. 2011). The severity of mercury contamination from artisanal gold mining in Colombia has been recognized by a number of researchers and local authorities for over a decade (Ingeominas, 1995; Veiga, 1997; Olivero et al., 2002; Marrugo-Negrete et al., 2008). Telmer and Veiga (2009) estimated that the annual mercury emissions to air, land, and water from artisanal gold miners in Colombia in 2007 were between 50 and 100 tonnes, however given recent gold value increases and gold rush activity, current mercury emissions are likely much higher. Colombia is likely the world's third largest source of mercury emissions from artisanal and small scale mining after China (240 to 650 tonnes of Hg per year) and Indonesia (130 to 160 tonnes of Hg per year) (Telmer and Veiga, 2009) and is one of the world's highest per capita mercury polluters (Cordy et al. 2011).   Antioquia has between 15,000 and 30,000 artisanal gold miners who produce approximately 10 to 20 tonnes of gold annually, although it is impossible to know the exact amount because much of it is smuggled out of the country. Departmental officials estimate the total gold production of the Department of Antioquia in 2008, from  35 companies and artisanal miners, to be approximately 26 tonnes or 63% of the Colombian gold production (41.5 tonnes per year). Miners normally do not use retorts to recover mercury because they believe that gold is lost in the process and because the price of mercury is low in the region (USD 56/kg in 2009, 180 USD/kg in April 2012) although mercury’s rapid price increase is making them more interested in recovery. Amalgams with approximately 50% of mercury are frequently burned without an adequate condensing system in the urban environment across Antioquia. The municipalities in this study produce the majority of gold in the Antioquia region.  In the five municipalities studied, there are about 98 gold shops (Remedios = 14, Segovia = 57, Zaragoza = 12, El Bagre = 10, Nechi = 5). Amalgams are burned and sometimes melted under the scrutiny of the miners. Another type of business (“fundición”), also located in urban areas, charges miners to burn the amalgam to evaporate the mercury and melt the gold before selling it to a gold shop. This gives the miners a clearer idea of the purity of their gold (based on colour) and thus a stronger position when bargaining a price with the gold shops. Two fundiciónes we visited have Pinta style fumehood and fibre condensers, though the exhaust from these shops still has 100 times higher mercury concentration than the WHO (2003) limit for annual average public exposure (1 µgm-3).   2.4.1 Segovia gold shops   Most gold shops and some mineral processing centres in Antioquia use locally developed mercury capture devices with open fume hoods of unknown efficiency (Figure 2.14). Observed contamination indicates that all condensing systems are insufficient except the fully closed retorts. The most common design (the ‘Pinta’ model) condenses the vapour in a circuit of PVC tubes at room temperature, followed by a series of barrels filled with fibre insulation. This system is named after its inventor, Mr. Pinta, a local Segovia gold buyer who is concerned about his health and that of his neighbours.  His wife told us that before using the filters, the people refining gold would be able to wipe their eyebrows after a burn and produce droplets of mercury from them.    36  Figure 2.14: The Pinta condenser system (left), and a typical Segovia amalgam, wrapped in foil to stop pieces of metal from exploding off the amalgam as the mercury tries to escape. © Paul Cordy photo.   Figure 2.15: Mercury collectors below Pinta fume hoods.  The bottle on the left is full of recovered mercury gathered in the funnel above it. © Paul Cordy photo.  Workers no longer experience this, however the concentrations of mercury in shops that use the Pinta system are still excessive (0.6 - 100 µgm-3).   37 The wooden fume hood used by Mr. Pinta contains a torch and crucible, and the door has a switch that does not permit burning unless the door is closed (usually other gold shops with this system remove this switch so that the miners can always see their amalgam). Below the fume hood there is a funnel and a jar that captures liquid mercury that condenses inside the fume hood.  This mercury is frequently emptied and re-sold.    A fan sucks the vapour through a long circuit of tubes (Figure 2.16) where the mercury cools and condenses by virtue of the length of the circuit. Mercury captured in various parts of this circuit is also collected and reused. After passing through the tubes, the mercury vapour then passes through four barrels full of rolled up plastic fibre insulation (Figure 2.17). The insulation is periodically changed and disposed of by burning (the creators of this condenser considered burying the insulation but they were worried that poor people would dig it up and use it in pillows and mattresses).  Burning these probably re-emits the mercury and produces very high concentration mercury vapour plumes. The contaminated sheets of matted plastic fibre insulation have a very high surface area, and would probably be fairly efficient at filtering mercury.    Figure 2.16: Filtration barrels of the Pinta system.  Note the series of multiple barrels, tubes, and fans.     © Paul Cordy photo.  38  Figure 2.17: Rolled-up plastic insulation is used as a filter medium (left), and dust that was captured by the Pinta system (right).  It is granular and solid, but very heavy and likely contains a high proportion of mercury. © Paul Cordy photo.  They would also become a mercury reservoir which likely produces a constant source of low level contamination emitting from the chimney and from poorly sealed containers.  The dust shown on the right in Figure 2.17 collects below the tube series assembly.  It is unclear how this dry dust is produced separately from the other liquid mercury traps.  The shop seems to store this material indefinitely, as there was a 20 litre bucket full of it on site.   While the Pinta system clearly captures large masses of mercury, the high contamination levels in shops that use it show that it is woefully inadequate for protecting human health and creates other very serious waste disposal and environmental contamination issues. Some local mercury scrubbers use a shower of water from perforated tubes inside a box that the mercury vapours flow through on the way to the chimney.  The spray is meant to cool and scrub mercury from the passing air.  An alkaline solution is sprayed to neutralize acid fumes from assays and purification. The holes used seem too large to produce a mist, and therefore the system could allow a large proportion of mercury or nitric acid vapour to pass.  Though most of these systems are only used in gold shops, there is one metal assayist and refiner who is building a system for capturing nitric acid  39 fumes. To avoid the high cost of stainless steel he uses regular steel coated with fiberglass (Figure 2.18). The shower solution is caustic and therefore removes acid that contacts the droplets. Such alkaline solutions should never be used to filter mercury vapour because mercury is soluble in alkaline conditions and thus the water would become contaminated (Velasquez et al. 2011).      Figure 2.18: Colombian acid scrubber; a fiberglass coated box in which water is sprayed out of the tubes seen running along the top corners of the box (arrows). The size of the holes in the PVC pipe are shown at right (fingers for scale). © Paul Cordy photo.  Unlike the Pinta system, the scrubber system seems to be built by individual gold shop owners except for one exceptional case shown that was made and installed by a manufacturer in nearby Caucasia (Figure 2.19). The efficiency of the scrubber system could perhaps be improved by using smaller holes, spray nozzles, or the Venturi effect. It should be noted that metallic mercury is almost insoluble at neutral pH, so any wet spray condenser would be limited to the cooling effect of the water and the degree to which mercury aerosols are formed and removed by gravity. The advantage of any water-based system is that the recovered mercury is prevented from evaporating by a barrier of water, and contamination of that water is limited due to the insolubility of mercury.   40   According to the owner of the Cacao gold shop in Segovia, the University of Antioquia developed a different kind of condenser in their shop several years ago out of a 4-inch pipe wrapped in refrigeration coils. The pipe is lined with frost over the 50-70 centimetres that is covered in coils (Figure 2.). The owner believes that this configuration cools the vapours, but this likely only occurs in the few millimetres of air that is adjacent to the coils, as the transit times are likely less than one second to a few seconds at most.      Figure 2.20: Refrigerator coil system. The refrigeration coils wrap the tube inside the blue box below the fan. It was unclear why the chimney was pointed down with a hole in the top. The mercury concentrations in this location was >999 µgm-3 (measured with the Jerome, five second average.)         © Paul Cordy photo.  Figure 2.19: La 70 gold shop’s brand new ‘scrubber’ condenser model. Water is pumped from the lower reservoir, sprayed into the box through which the fumes flow, and is then collected and reused.  The fume hood has a door that increases air inflow velocity and decreases losses through the hood face. © Paul Cordy photo.   41 The mercury concentrations near the outlet of this refrigerator system exceeded the detection limit of the Jerome (>999 µgm-3) even though they were not burning at the time. The owner claims to recover 1-1.5 kilograms of mercury every three months, adding that he only burns retorted gold. The owner would not speculate as to what his monthly gold throughput was.  Shop owners claim to recover large amounts of mercury from each system, although the exact efficiency is unknown in each case. Table 2.3 shows masses of gold burned and the resulting concentrations. Processing amalgamation tailings with cyanide dissolves mercury and gold, and this mercury is emitted when miners evaporate the zinc shavings onto which the gold and mercury are precipitated (this process is described in detail in Chapter 5).  Table 2.3: Summary statistics of mercury concentrations measured adjacent to burn/melt events in progress in Segovia using the Jerome analyzer (five second averages for each measurement interval, averaged over the entire burn duration). Note that measurement often ceased part way through the event due to saturation of the Jerome sensor. Commonly this followed readings of  >999 µgm-3.   Gold/ zinc precipitate Melting Amalgam Burning Average weight of material being burned (g) 5556 114 Average [Hg] (µgm-3) 295 194 Standard deviation [Hg] (µgm-3) 137 171 Range [Hg] (µgm-3) 0 - >999* 0.004 - >999* Number of mercury evaporation events observed 12 14 Mean burn duration/ measurement interval (minutes) 78/10 8/1 *Note that measurement often ceased part way through the event due to saturation of the Jerome spectrometer.  A summary of opportunistic sampling of mercury levels analyzed in the ambient air in select municipalities is shown in Table 2.5, some individual measurements in various contexts are shown in Table 2.4, and they are consistent with previous findings in the region (Cordy et al. 2011). Five second average mercury concentrations on the streets and indoors are commonly far above the WHO (2003) annual average guidelines. The employees of the gold shops, given their extremely contaminated workplace, are likely  42 the most intoxicated individuals, but toxicological and neuropsychological exams have not yet been done.  Table 2.4: Atmospheric mercury in selected parts of Segovia, taken using the Jerome analyzer (5 second averages). Location  [Hg] (µgm-3)  Comments In front of El Dorado gold shop 79 amalgam burn in progress Inside open doorway of El Dorado 537 15 minutes after above burn  In front of compra Cacao 17 no amalgam burn at that moment Inside Cacao gold shop >999 beside the refrigerator coil chimney in Cacao Street between Cacao and El Rey 5 no amalgam burn at that moment In front of compra El Rey 7 as above Inside El Rey gold shop 50 no burn since yesterday In front of compra Carrasco 46 no burn for 3 days Inside compra Carrasco 31 as above Inside entable Bolivar 29 near cocos In front of compra La Real 3 this shop has a wet scrubber condensing system  Inside compra La Real 217 no amalgam burn at that moment In front of no name shop 25 this shop opened 20 days previous. Inside no name shop 127 this shop has a pinta condenser Inside compra Jaito Hugo 435 in gold melting area Inside compra HIH 162 this shop has a wet scrubber condensing system  Between HIH and Pepe gold shops 8 no amalgam burn at that moment In front of Pepe gold shop 4 as above Inside Pepe gold shop 28 this shop has a pinta condenser Entrance of a commercial building 10 more than 5 gold shops within one block. Inside the commercial building 6 inside a pharmacy. Commercial building second floor 4-5 inside a dentist’s office Inside the municipal offices 8 several gold shops on the adjacent street. Main plaza 0.1 - 0.2 on a holiday, no entable was working. Calle Real  0.15 - 0.3 as above. Calle 48  1 - 1.5 as above. Entable Guamo entrance 15 - 20 about 40 cocos were working at the time. Entable Guamo 80 measured near a coco. Inside another entable 943 near a coco; probable recent amalgam burn. Calle Castillos 40 near an elementary school. In front of a gold shop  3-5 no amalgam burn at that moment. Street near fundición on Calle 47A  3-5 amalgam burn in progress in fumehood with a “new” pinta mercury condensing system. In front of the above fundición 20 - 30 same as above.  Inside above fundición 40 the operators believe that the filters capture all mercury vapour. Near the nose of the operator (in the same fundición) 60 a child was sleeping near this worker while he was burning amalgam. At the exhaust of the condensing system 100 the condenser is clearly insufficient. During zinc precipitate burning 616 no filtration.  43 In addition to significant local alluvial gold sold in the town of Caucasia, Antioquia, miners bring amalgams from other municipalities to sell in Caucasia’s gold shops because taxes on gold are lower and therefore they get a better price. This causes not only loss of royalties by the municipality where the gold is mined but also significant pollution in a very populated city (second largest in Antioquia). Average values for Antioquia towns are hazardous (Table 2.5). Table 2.5: Average mercury concentrations (µgm-3) measured within the urban centers of Segovia, Remedios, Zaragoza, El Bagre, and Nechi. (Averaging times: 5 s (Jerome) and 10 s (Lumex)).  Average (µgm-3) Standard deviation Range N Analyzer Non-mining buildings 6.30 7.03 <3 – 40 31 Jerome Streets (not directly in front of Hg sources) 1.78 2.91 <3 – 12 48 Jerome Streets in front of Hg sources 13.4 17.4 <3 – 79 48 Jerome Inside gold shops/“entables” (not during burn/melt events) 95.6 191 3 - >999 77 Jerome Top of church tower 0.31 1.14 <0.01 – 39 65 hours Lumex Segovia vehicle survey (2010) 1.25 3.52 <0.01 – 79 26 hours Lumex Remedios vehicle survey (2010) 0.25 1.01 <0.01 – 32 7.6 hours Lumex   In one particularly informative interaction in Caucasia, one gold shop owner was shown that the mercury concentrations in his shop at that moment were 200 µgm-3 even though no one was burning amalgam. The owner did not believe in the measurements and said that he had recently analyzed his urine and the levels of mercury came back to “normal” after two weeks of treatment with diuretics in Bogotá. We explained to him that mercury in urine can decrease to a background level (for example 5 µg.L-1) in less than fifteen days, but this does not mean that mercury had been eliminated from his brain and other tissues. What he did not know is that the total mercury levels in urine would not be expected to correlate with neurological findings even once exposure has stopped (Veiga and Baker, 2004). Mercury concentrations inside entables and fundiciones often exceed the detection limit of the Jerome (>999 µgm-3), which suggests that workers’ exposure may at times approach fatal levels (1200 – 8500 µgm-3 according to Jones 1979).  44 Combined with the casual culture of mercury use and its long history as a normal and essential activity, nebulous associations between mercury and local health outcomes make it difficult to argue that mercury is toxic enough to merit the immediate and drastic action that is called for. The situation in Antioquia is clearly hazardous, but only neuropsychological tests would confirm the high level of intoxication of the population living near the gold shops and entables.  2.5 Puerto Maldonado, Peru   The gold rush in Peru is most active in the department of Madre de Dios, whose 85,000 km2 area is primarily Amazon lowland rain forest and one of the most biodiverse areas in the world (Brooks et al. 2006). Increasing gold prices have brought rapid growth in Madre de Dios deforestation rates (1915 ha/year, 2006–2009) as well as in Peruvian mercury imports (42% increase, 2006 to 2009) 130 t/yr (Swenson et al. 2011) (Figure 2.21). Peru is the largest gold producer on the continent (165 tonnes in 2012, USGS 2013).  Figure 2.21: The Madre de Dios department of Peru. Labeled mining areas are A: Guacamayo, B: Colorado-Puquiri, and C: Huaypetue. (Swenson et al. 2011).  45   Figure 2.22: Gold and deforestation in Madre de Dios, and mercury imports to Peru (Swenson et al. 2011).  Official artisanal gold production in Peru in 2002 was 18.7 tonnes, which represented 12% of Peru’s total gold production. Official artisanal gold production in Madre de Dios in 2002 was 15.5 tonnes, however the real number is likely to be far higher as a large portion of the region’s production is smuggled out of the country illegally (Hentschel et al. 2003). In 2004, artisanal gold mining accounted for 9 percent (14.8 t) of the gold produced in Peru (Gurmendi, 2006). Figure 2.22 shows that as gold prices have risen, mercury imports and the area impacted by mining have increased exponentially between 2006 and 2010, and this can be clearly seen in the satellite imagery in Figure 2.23.   Conflicts between artisanal miners and communities are becoming increasingly acute in Madre de Dios, and miners have begun to invade the traditional territories of Native communities. Some native people have also started their own artisanal mining operations (frequently contracting non-native mine workers), leading to further conflicts as some indigenous communities consider their territories as “off-limits” for mining legislation. Some of them claim exclusive mining rights but neglect their fiscal and environmental obligations (Pautrat 2000).  46  Figure 2.23: Satellite images of recent mining activity (Landsat TM bands 5, 4, 3); A) Guacamayo (12°51′S, 70°00′W) along the Inter Oceanic Highway and (B) Colorado-Puquiri (12°44′S, 70°32′W) in the protected area buffer zone. (Swenson et al. 2011). Pink patches indicate mining activity.  The main urban centre for this complex region is Puerto Maldonado, which is a river port on the newly completed Trans-Oceanic Highway connecting Peru, Bolivia, and Brazil. Puerto Maldonado is therefore an important commercial and administrative centre and also has a thriving gold trade.   47 2.5.1 Puerto Maldonado gold shops  In Puerto Maldonado, Peru, past mercury remediation interventions mainly focused on the use of retorts by miners in the field, but also raised awareness of gold shop emissions. The regional government mandated that all gold shops should have a mercury condenser of the kind developed by GAMA (2009). This is based on their communal retort: a large concrete water tank with small diameter tubes in which the fan driven mercury vapour is cooled without contacting the water. Several of these retorts are still functional and in use in coastal Ecuador and Peru. GAMA coiled these tubes up into a water tower that could fit in small shops. Municipal officials mentioned that fume hoods and condensers were adopted ubiquitously for compliance, but that the most expensive components, such as fans, were omitted. In many cases, there is only a fume hood without an exterior exhaust tube, or the tubes end in a sealed water bucket with no outlet (Figure 2.2.24). Remarkably, despite a lack of external exhaust, the extreme mercury concentrations in the air of these shops (usually greater than 25 µgm-3) dissipates quite rapidly once burning stops. Roll-up metal security barriers are the only shop doors, so during business hours the shop is wide open and mercury vapour is quickly mixed with outside air.  Figure 2.24: Various fume hood and condenser designs in Puerto Maldonado. Typical fume hoods; some have tubes (left) connecting them to a water bucket (centre), but there is no exhaust tube and therefore there is no chance that air will be pumped through the water. Often the fume hood exhausts just above the operator’s head (right). © Paul Cordy photo.   48 Unfortunately, the gold shops are across the street from the main market and are themselves a social gathering spot. Thus, due to the combination of mercury dispersion into the outside air with a high concentration of street dust, there are likely large numbers of people who are not involved in the gold trade but who are nevertheless exposed to elevated levels of mercury. Though some gold shop workers are concerned about their health, other people seem unconvinced of the dangers: minutes after a burn event in one gold shop during which mercury concentrations exceeded 50 µgm-3, a woman was seen breastfeeding her baby beside the fume hood (Figure 2.25). Beginning in 2008, the USEPA installed its mercury condensers (ANL 2007) in local shops.            2.6 Ponce Enriquez, Ecuador  Southern Ecuador has been an important and productive mining region for hundreds of years (Figure 2.26). Recent high gold prices and serial agricultural crises caused by the El Niño phenomenon have intensified informal mining activities in Ecuador (LaPlante et al. 2002). The vast majority (85%) of the gold production in the country Figure 2.26: Map of Southern Ecuador. From Velasquez 2010. Figure 2.25: People are unaware of the risk to children posed by mercury contamination in Puerto Maldonado gold shops. The ubiquitous ‘El Espanol’ brand mercury bottle is visible on the shop counter inside the clear plastic box that covers the electronic scale. © Paul Cordy photo.   49 comes from the artisanal small-scale mining sector, and approximately 70% of gold produced is sold to the black market (Veiga et al. 2009). There are approximately 100,000 artisanal and small-scale gold miners in Ecuador, and the annual mercury emission from these operations is very high (Velasquez 2010). A survey accomplished by the Swedish Environmental System estimated discharges of mercury to be 40 kg per year in the River Siete and 160 kg per year in the River Calera (Prodeminca 1999) in the Portovelo-Zaruma region. The underground ore in this region is mined manually, and miners take 40 to 50 kg bags of ore with grades ranging from 3 to 30 g/tonne of gold to processing centres that use both mercury and cyanide (Veiga et al. 2009). Miners grind the ore, concentrate it in sluice boxes and amalgamate the concentrate for free or for a low fee in these processing centres, which are locally referred to as “beneficios.” In partial payment for the rental of processing facilities, the miners give the tailings to the beneficio plant owner, who extracts residual gold by milling and cyanidation. The ground material is discharged on 7–10 m long cement sluice boxes covered with locally made wool carpets. A large amount of concentrate is panned in a water box and then amalgamated using about 100 g of mercury for 3 kg of concentrate. This is a relatively small amount of mercury to add to that amount of concentrate, compared to artisanal mining operations in other Latin American countries such as Venezuela and Brazil (Veiga et al. 2005, Veiga and Hinton 2002). The amalgam is most often burned at the processing centre where gold is extracted, and many operations have communal retorts designed by GAMA (2009).  The analysis of the amalgamation systems in 8 centres indicated that 12 to 40% of the total mercury used in the process is evaporated when amalgams are burned, 40 to 60% of mercury is recovered and 1 to 35% of mercury is lost with the tailings (Velasquez, 2010).  2.6.1 Ponce Enriquez gold shops  In Ponce Enriquez, Southern Ecuador, amalgam is decomposed and gold melted without any mercury condensing or filtration systems. One second averaged mercury concentrations in the ambient air of these shops were high (ten second averages were  50 between 10 and 40 µgm-3, see Table 2.1), even though no amalgam was being burned at the time of the measurements. This is probably evasion of mercury accumulated during a long history of burning indoors (Malm et al. 1998). In one shop, mercury was kept in a plastic sandwich bag in the top drawer of the office desk. The shop owners were willing to have mercury measurements taken inside their shops, but they did not expect to find mercury contamination because the miners were supposed to be using retorts in the field. They claimed that since foreigners came and distributed retorts, there were no more mercury problems. In truth, retorts are often not adopted because miners fear losing gold or because the colour of retorted gold is sometimes different, which impacts price negotiation.   2.7 Paramaribo, Suriname   For hundreds of years, members of Maroon tribes, descendants of former slaves who escaped into the jungle, mined the alluvial gold deposits of Suriname with hand tools and concentrated using only gold pans. Recently, a large influx of Brazilian miners (“garimpeiros”) transformed artisanal mining by introducing hydraulic monitors, bulldozers, excavators, dredges, and other semi-mechanized mining methods that have increased gold production by boosting the throughput of alluvial material (Veiga 1997). Garimpeiros are being lured to Suriname by rumors in the Brazilian gold fields that Suriname is the new “El Dorado.” In Brazil, the exhaustion of easily obtained alluvial gold and increased enforcement in illegal mining areas compels miners to look elsewhere for their livelihood. By contrast, Brazilians can obtain inexpensive work permits in Suriname, and Surinamese people are often happy to partner with them given their superior, though still wasteful and destructive, mining technology Figure 2.27: Map of Suriname.  51 and practices. In 2001 it was estimated that gold miners cleared 48–96 km2 of old-growth forest in Suriname annually and it was predicted to reach 750–2280 km2 by 2010.  Analysis of abandoned mining sites suggests that forest recovery following mining is slow and qualitatively inferior compared to regeneration following other land uses (Peterson et al. 2001).  2.7.1 Paramaribo gold shops  Local officials claim that artisanal gold mining in Suriname has changed dramatically with the influx of Brazilian miners (Veiga 1997). They apparently brought with them industrial scale extraction methods as well as an artisanal bubbler design. Gold shop owners in Suriname are legally obliged to use a mercury capture device when operating in the capital city of Paramaribo (Figure 2.28). However, mercury concentrations exceeding 10 µgm-3 were commonly found inside gold shops (Table 2.1), suggesting that the local bubbler systems are ineffective. Most gold shops use an air compressor to pump the fumes through water, which probably produces insufficient pressure to prevent gases escaping the fume hood.  Well-ventilated shops had lower mercury concentrations in peripheral rooms compared to the gold workrooms, whereas poorly ventilated shops had similar mercury concentrations throughout.   Figure 2.28: Compressor driven mercury condensers in Paramaribo. Ducts carry vapour from fume hoods (right and left) and push them through large tanks of water (centre). © Paul Cordy photo.  52 Mercury concentrations outside the front door of the shops varied with wind speed and direction, but ten second averages could jump to 10 µgm-3 or more, so there is likely to be significant intermittent collateral exposure among people walking past or working near the shops.  2.8 Field sites summary  All gold shops in this study show hazardous to extreme levels of mercury contamination, and artisanal filtration systems (when present) give a false impression that the contamination is being controlled. There is clearly a need for targeted education of miners and gold shop workers, as well as programs to establish local traditions for using and maintaining more efficient gold shop condensers. The ultimate goal for any initiative aimed at reducing mercury emissions from gold shops must be the complete elimination of processing centres and raw amalgam burning from populated areas, in combination with strategies that promote ongoing monitoring and enforcement of regulations concerning gold shops. There is no system for filtration of mercury vapour that is sufficiently efficient to burn amalgam without adversely affecting human health, therefore amalgam burning must not be permitted in the urban core. Even if this goal is achieved, it is likely that shop walls and other surfaces will emit their mercury burden for months to years after relocating mercury sources. It is also imperative to improve or replace locally developed mercury capture devices to prevent contamination from residual mercury in doré. One such improvement would be to add glass doors on the fume hoods such that miners can still see the amalgam being burned, but the vapour is forced to escape via the filters. At the moment, the best mercury capture system is the UNIDO bubbler design, but they are merely starting points that demand further improvement.  Many solutions to this problem have been developed and tested (Jonsson et al. 2009, Hilson et al. 2007, Zolnikov 2012, Amankwa et al. 2010), but broad adoption of retorts, condensers, filters and other alternatives is still elusive. The Puerto Maldonado case shows that, despite a well-engineered solution and wide apparent compliance, regulations do not guarantee harm reduction. Successful initiatives require a regular presence of  53 trainers and field technicians to monitor and encourage proper installation, usage and maintenance in the short term, and to train municipal officials in long-term regulation enforcement (Clifford 2011, Sippl and Selin 2012). Though most gold shops report their income and pay taxes, they might under-report the gold bought and mercury recovered in a way that could prevent accurate tracking of the effectiveness of condensers and regulations. It is therefore important to find other proxies of gold production and implement a formal audit program wherever possible. In Suriname, Colombia, Peru and Ecuador, mercury emission reduction initiatives have had unintended consequences, as they convince gold shop owners that use of retorts in the field eliminates all of the mercury in gold that they purchase, and/or that installation of fume hoods reduces mercury vapour in their shops to tolerable levels. Mitigation of mercury contamination therefore requires sustained initiatives that are atypical of foreign-led development projects and are often beyond the technical and financial resources of local governments (Zolnikov 2012, Clifford 2011). Many other issues further complicate matters, such as child labour (Hilson 2012), migrant labour (Jonsson et al. 2009), scarcity of energy and water resources (Valdivia and Ugaya 2011), and conflict with agrarian producers.  We found no case in which mercury was adequately recovered and shop operators were exposed to high levels of contamination in all cases. Furthermore, the mercury emissions from gold shops likely present a significant hazard to the health of local residents who are not involved in the gold trade. It is therefore necessary to quantify the frequency, distribution, and intensity of urban mercury emissions     54 3 Urban mercury mapping and monitoring 3.1 Synopsis  Artisanal miners sell their gold to shops that are usually located in the urban core, where the mercury-gold amalgam is burned or melted to evaporate the mercury that was added during ore processing. People living and working near these gold shops are exposed to intermittent and extreme concentrations of mercury vapour. In the urban centres of Segovia, Colombia, and Andacollo, Chile, the average concentrations (averaged over the entire field campaign) as measured by mobile mercury vapour analyzer transects taken repeatedly over several weeks were 1.25 and 0.331 µgm -3 respectively. These values are greater than the World Health Organization (2003) health hazard limit (1 µgm -3), and tolerable level (0.2 µgm -3) respectively. Therefore, the residents of Segovia are exposed to significant health hazard, as are Andacollans living in close proximity to gold shops. Globally, it is likely that millions of miners, as well as non-miners who live near gold shops, are at serious risk of neurological and renal deficits. Maps of average mercury concentrations show the spatial distribution of the hazard in relation to residential buildings and schools. Measurements from towers show the temporal variability of mercury concentrations, and suggest that large quantities of mercury are available for long-range atmospheric transport. This is the first full description of artisanal mining gold shop practices and of the character, severity, and distribution of mercury emissions within urban mining centres. Most of the research reported here has been published in Cordy et al. (2013).  3.2 Mercury monitoring and mapping methods 3.2.1 Sampling and analysis  Spatially resolved mercury measurements were taken while driving a car along a set route around the urban areas with the Lumex inlet at a height of two metres (Figure 3.1). Most transects were taken during daylight hours without rain, ensuring the greatest possible coverage in time of day and day of week. Positions were taken using a Garmin 62s GPS device, and both the Lumex and GPS logged data at a frequency of one second.  55 The mercury data were corrected for transit time in the 2 metre sampling tube. In order to obtain the highest possible spatial resolution, the target sampling speed was 20 km/h; however the actual speed varied depending on traffic. The Lumex was calibrated at the beginning and end of each run, and instrument drift was removed based on these calibrations. Calibrations were only done in absence of strong mercury signals, when the concentrations were stable and the mean concentration below 0.1 µgm-3. These periods (generally five minutes or less in duration) are used to determine local background and remove instrument drift. Drift was manually eliminated by subtracting linear trends and constant biases with reference to the beginning and ending calibrations.   Figure 3.1: Various mobile mercury monitoring solutions. Mercury concentration could be monitored on the laptop during sampling (as shown in the blue line plot), and a separate car battery was used to maintain constant electricity to the analyzer. The goal was always to have the hose inlet at 2 metres height with the inlet facing away from the vehicle, though the exact configuration varied in all studies. The truck was used in Antioquia, the car was our main vehicle in Andacollo, and we soon abandoned the bike carts in Andacollo when a loaned vehicle was available. © Paul Cordy photo.  56 The transect routes were designed to minimize duplication of pathways and maximize coverage of the neighbourhood containing the gold shops within a one hour period. The exact route taken during each transect varied due to road access and traffic issues, and deliberate asides were occasionally undertaken between runs of the set transect routes to sample areas outside the gold shop neighbourhoods.  Average mercury concentrations measured within a 10 metre radius of each point along the surveyed roads were computed for display, though only non-overlapping 20 metre separated averages were used in statistical analysis. The mean value of these 20 metre point averages in the urban cores of Andacollo and Segovia are referred to here as spatial averages. The median one-second concentration sample sizes for each 20 metres of road in Andacollo, Chile and Segovia, Colombia are 109 and 97, respectively. Tower measurements in Segovia were recorded at a height of 17 metres in a condemned church bell tower adjacent to the main square (Figure 3.2).    Figure 3.2: The now demolished church tower site in the centre of Segovia with view of a typical Segovia street canyon (left), and the Andacollo stadium tower mercury monitoring sites. © Paul Cordy photo.  In Andacollo, the Lumex was mounted at a height of 18 metres on one or two of three soccer stadium light towers immediately downwind of the most frequented gold  57 shop in town (the analyzer was placed on the NW tower on or before May 6th and on the NE tower thereafter. When there were two analyzers on towers, they were placed on the NE and SE towers). Chilean measurements were taken in collaboration with Daniel Moraga of the Catholic University of Coquimbo in 2009. Colombian measurements in this chapter were taken for UNIDO’s mercury remediation project in Segovia in April 2010. Data from 2011 and 2012 are shown in the chapter on mitigation.  The average elevations of Andacollo, Chile, and Segovia, Colombia, are approximately 1000 and 700 metres respectively, so the difference in atmospheric pressure is assumed to be insignificant.  3.2.2 Sampling error  Errors in measurement of mercury concentrations are primarily due to response lag. Given a step change in concentration, the Lumex will register 90% of the real concentration in 20 seconds at the instrument’s mean flow rate of 10 litres per second (personal communication, Joseph Siperstein, Ohio Lumex 2011). During this time, the transect vehicle would have moved approximately 110 metres, and the tower mounted Lumex would be taking in air from ~35 metres upwind (given mean wind speed of 1.77 ms-1 over the study period in Andacollo). The Lumex registers changes in concentration immediately, though lagging concentration values introduce some degree of spatiotemporal averaging into the raw data. These instrument errors result in underestimation of peak concentration values, and overestimation of low concentrations that immediately follow strong signals. This would tend to spread out strong signals and cut off signal peaks. Measurements of mercury concentration from tower locations suggest that real concentrations can be expected to vary on similar scales (on the order of magnitude of 10 metres and 10 seconds) given the complex turbulent flows in urban areas. Given the relative infrequency of strong mercury signals as compared with low concentration values, it is likely that underestimates of high concentrations are more frequent than overestimates of low concentrations. The biases introduced by these errors are minimized  58 subsequently by the spatial averaging used to produce maps, and the net effect of these errors on tower and spatial measurements is to underestimate mercury concentrations. Errors of unknown magnitude may be caused by variations in traffic flows, transect vehicle driver habits, changes in transect patterns to avoid road construction, or events such as parades, and differences in instantaneous speeds for similar locations in different transects. These are also assumed to be minimized by averaging. GPS position errors are also assumed to be on the order of 1 metre and unlikely to contribute significant error. Physically impossible ‘spikes’ in the GPS position data were removed by visual inspection and linearly interpolated for gap filling.  Instrument drift is a potential source of sampling error, although it was assumed to be approximately linear over most observation periods, which were normally about one hour, but varied from thirty minutes up to 8-10 hours in the case of tower based monitoring. Initial and final instrument calibrations were used to remove linear drift. Given that signal noise varies within a narrow range, this was also used to visually check that the drift removal was appropriate. On numerous occasions the Lumex was run overnight or longer in mercury-free air in order to understand the long-term behaviour of instrument drift. The Lumex readings usually rise or fall steadily by up to 0.2 µgm-3 over several hours, so that within one hour of continuous operation the drift is more or less linear. Data series with nonlinear, irregular, or extreme drift were rejected from this analysis. At least once per day of transect measurement, the analyzer was run for 30 minutes or more at a site in the desert outside Andacollo that was more than 500 m from any building and more than1 km away from any known mercury sources. This ‘local background’ site always yielded stable values between 0.02 and 0.03 µgm-3.  All Jerome measurements are assumed to be accurate within the constraints of its precision (1 µgm-3).   3.2.3 Comparing air quality measurements with air quality standards  The WHO (2003) standards for mercury exposure are given as annual average concentrations, however the data presented here span no more than one month. I therefore computed descriptive statistics for various averaging periods and then used  59 trends in these values to extrapolate reasonable estimates for annual averages as in Larsen (1971). Tower measurements at night in both Andacollo and Segovia observed mostly background concentrations, and gold shops in both places are generally closed between 7pm and 7am. Given that mobile measurements were only taken during the day, estimates of annual average based on daytime measurements must therefore be reduced by half to account for night hours (which are also assumed to have remained at background concentration because most emissions occur during the day. Strictly speaking the annual average should be the sum of day concentration and night concentration divided by two, but average night concentrations are much less than during the day).  Averaging periods containing fewer than 50% valid concentration values are not used in the analysis, and periods do not overlap in order to avoid double-counting data. Given that the observation time series start and end at arbitrary times, the averaging periods also start at the beginning of each time series in order to maximize the number of averaging periods that exceed the threshold of valid values. Up to 6 hour averages are possible to compute using temporally continuous periods of data, but since there are few periods longer than this in which observations were made in more than 50% of the time, longer period averages were computed from the set of all data strung end to end as if it were a continuous and contiguous series. The data are assumed to be stationary and lognormally distributed, and the geometric mean and standard deviation   𝜇? = 𝑥? ? ?            𝜎? = 𝑒𝑥𝑝 ™ ???? ??  are used for extrapolation as these are appropriate for lognormal distributions. Given the geometric mean and standard deviations for each averaging period, I computed the 10th and 90th percentile concentrations (using 𝑥 =   𝜇?𝜎?? and the geometric z-scores of -1.28 and 1.28 respectively (Larsen 1971)).   3.3 Results   The one second interval mercury concentration data from Andacollo and Segovia (both tower and mobile measurements) range over five orders of magnitude and are  60 lognormally distributed from the 10th percentile to the 90th percentile in Segovia, to the 80th percentile at the Andacollo tower (Figure 3.3). Andacollo mobile data somewhat deviates from lognormality. Probability plots of longer-term averages (30 seconds, 1 minute, 15 minutes, 1hour, etc.) plot successively closer to lognormality in both Andacollo and Segovia.  Figure 3.3: Lognormal probability plots for Andacollo (top) and Segovia (bottom). Tower observations are shown in blue and mobile observations are shown in red. Dashed lines indicate perfect lognormality.  61  Figure 3.5 show that the mean value of the observations is the same for all averaging periods, and that the least squares fit line of the 90th and 10th percentile concentrations converge on the mean value before the annual averaging period.     Figure 3.4: Geometric mean, 90th percentile, and 10th percentile values of mercury concentrations for various averaging periods in Andacollo. The top plot shows Andacollo’s tower observations (mean: 0.02 µgm-3), and the bottom plot shows mobile observations (mean: 0.07 µgm-3). The expected value of the annual average is the mean of the averages of all averaging periods.   62    Figure 3.5: Geometric mean, 90th percentile, and 10th percentile values of mercury concentrations for various averaging periods in Segovia. The top plot shows Segovia’s tower observations (mean: 0.1 µgm-3) and the bottom plot shows mobile observations (mean: 0.2 µgm-3). The expected value of the annual average is the mean of the averages of all averaging periods.   63 Given this, I assume that if the data included series of uninterrupted observations over several years it would still produce average values that are similar to those which are used in this study (assuming that conditions and habits at the field sites remained the same). Of course, the assumptions of stationarity, representativeness of the data collected, and lognormality are hard to justify in detail, nor is it necessarily correct that mercury emission events never occur at night. Even more problematic is the fact that the Lumex analyzer as used does not accurately measure concentrations greater than 50 µgm-3 and it takes 20 seconds for the analyzer to register 90% of the real value of a step change in concentration.    Given all of these uncertainties that lead to underestimations of the real concentrations, it is prudent to apply a factor of safety when assessing the hazard represented by average values of mercury concentration presented here. Given that the estimates of annual average concentration derived using daytime values are roughly double those that would result from inclusion of night data, comparison of the WHO annual average standards with estimates derived from daytime data only is equivalent to a factor of safety of 2.  3.3.1 Tower measurements  Lumex measurements from towers in both places show that mercury is dispersing to heights where it can be exported to the regional scale, though further studies are needed to confirm and quantify long-range transport. In Andacollo (Figure 3.6), positioning of the tower immediately downwind of the principal gold shop resulted in peaks of very high mercury concentration. Viewed at a higher resolution, the peaks do not appear as sustained values, but rather as a period of extreme variability, due to turbulence, varying wind direction, and perhaps narrow plume size. For some events, such as on both Saturday mornings and most midweek periods sampled, the mercury concentrations before and after each event are low and stable. This suggests that these are probably discrete plumes emanating from a single source, and not several overlapping events. If this is true, it suggests that any given mercury vapour release could potentially contaminate the surrounding area for between approximately 15 to 30 minutes (this duration is confirmed by the record of amalgam burning events in Table 2.2).    66 The duration, persistence, and timing of these events will also vary according to wind speed and direction, such that each event may only represent a fraction of the actual duration and concentration of the mercury vapour released. On Friday and Saturday afternoons, mercury concentrations are much more sustained, and it appears that there are many signals that overlap. This is consistent with the fact that, according to many miners, gold shop operators, and local residents, more amalgam are burned on these days as miners return from the mines with the week’s amalgam to be sold in anticipation of the weekend. Figure 3.7 shows that Segovia mercury signals are also more sustained at high concentrations, and more frequent. Even though the church tower in Segovia was upwind from most of the mercury sources during monitoring, there were at least 20 plumes in which mercury concentrations exceeded 1 µgm-3 and 6 events that exceeded 10 µgm-3 (in 65 hours of observations over four days). These events typically lasted about 20 minutes, though peak concentrations rarely lasted for more than a few minutes at a time, as dilution seemed to be rapid. The consistently strong signal from May 16, 2010, 13:30 to 14:30 lasted for an entire hour however, indicating that hazardous concentrations are occasionally sustained for much longer. The lower plots in Figure 3.7 show typical plumes at higher resolution. These show that the plumes have more consistently high concentration than those found in Andacollo. More sustained concentrations could be attributable to the larger size of amalgam burned, or perhaps the meteorology of the site. Certainly, the frequent and consistently high concentrations measured at the tower in Segovia reflect the much larger number, weight, and quantity of amalgam being burned in a larger number of shops than is the case in Andacollo. The two long quiescent periods in the series are night observations, and they suggest that the assumption that there are few emissions at night may be reasonable. Continuous monitoring is needed to confirm that, however.  3.3.2 Mercury mapping  Roving mercury monitoring in Andacollo, Chile and Segovia, Colombia shows that extreme concentrations occur frequently, especially near known sources (Figure 3.8 and Figure 3.9). The mercury concentrations shown in Figure 3.8 are averages over all one  67 second interval Lumex measurements recorded within a ten metre radius of the centre of each coloured dot. These data were collected between April 20 and May 16 of 2009, and the median cumulative measurement time at each point was 97 seconds.    Figure 3.8: Average mercury concentrations along the Andacollo, Chile gold shop neighbourhood streets. The red and orange road lengths in the upper right corner of this image are road lengths which were only sampled on one or two occasions. The average of all one second observations measured over the entire field campaign in the area bounded by white dashes was found to have an average mercury vapour concentration of 0.331 µgm-3 with a standard deviation of 1.53.   68 At walking speed, the average person takes two breaths in the time it takes to walk twenty metres, so the spatial averages represented by each coloured dot can be roughly interpreted as the expected instantaneous exposure of a pedestrian during the daylight hours of the field campaign.  In reality, probability would dictate that many times when people pass each of those points they are exposed to much lower levels of mercury or none at all, however it is useful to consider who the people in this area are and what their habits might be.    Figure 3.9: Mercury concentrations in the Segovia, Colombia urban core; data are averages of three consecutive years of two-week measurement campaigns (2010, 2011, and 2012). The area shown was found to have an average mercury vapour concentration of 1.25 µgm-3 and standard deviation of 3.27 in 2010. Not all mercury sources are shown. Data north of UTM 782900 are not included in the average concentration as those segments were not consistently sampled.  69  For instance, students from all over the area walk to the school in the centre of this neighbourhood, and more than half of them live on the south end of town and thus have to pass fairly close to the principal gold shop. These children are frequently passing through ground level mercury plumes in the street, and having them waft through the schoolyard and classrooms from a block away without realizing it. Consider also all the heavy breathing happening as adults and children alike spend hours on the soccer field when mercury plumes frequently disperse across it. Therefore the spatial averages shown can be thought of as maps of the risk of being exposed to a mercury plume, and as indications of the long-term exposure of the residents near each point, but individual exposures will vary greatly according to habits, profession, and location of residence among other things.  Figure 3.9 also shows that mercury contamination in Segovia is more intense and widespread than in Andacollo (Figure 3.8). Concentrations within 100 m of schools are lower in Andacollo than in Segovia, however the effect of this kind of exposure on developing minds is not well understood (WHO 2003). Segovia also has many more gold shops, though not all sources are shown, and the concentrations are highest where sources are clustered most tightly together. Obviously, there are many mercury emissions that occur in locations other than those with known emitters. In both Andacollo and Segovia, people often tell of miners burning amalgam in their kitchens, or processing with small mills in their backyard. One miner we met invited us to watch him burn a 120 gram amalgam that he had accumulated over the past four days of mining at his small mine in the adjacent mountains and 8 hours of processing in a plant on the edge of town. He did so using the local rudimentary retort in the inner courtyard of his house and allowed us to use the Jerome (which takes five second average concentrations). During the 25 minutes it took to heat the retort with a torch, the mercury concentrations were on average about 21 µgm-3 at a distance of two metres (ten observations, σ = 6 µgm-3). When the retort was opened, concentrations rose to an average of 136 µgm-3 (six observations, σ = 21 µgm-3). Events like these could explain some of the signals found away from known sources. There are also, however, many processing centres and possibly small gold shops that are not official or are undocumented by local authorities.    70 Nevertheless, average concentrations greater than or equal to 1 µgm-3 are almost invariably found within 200 metres of a known mercury source in concentration maps of both towns, and almost all sources in Segovia are within 50-100 m of a location with an average concentration greater than 3 µgm-3 (as is the principal gold shop in Andacollo).  Table 3.1 compares Andacollo and Segovia in terms of contamination. The average values of the tower and mobile sampling in this table are time averages, meaning they are the average value of all one second interval measurements of mercury recorded during the entire field campaign. In the case of the tower this is therefore a simple time average, but in the case of mobile measurements this is an average over space and time, so there is possibly some bias toward areas that were sampled more frequently. This bias was minimized by adhering to the standard transect routes whenever possible, and by designing the transects so as to reduce duplication. Segovia is larger but the mean concentrations in the streets and at the tower, as well as the mean amalgam mass are more than triple those found in Andacollo. Lumex values were derived from mapping and tower monitoring, and Jerome values were obtained by opportunistic point sampling in streets and inside buildings.  Table 3.1: Mercury concentrations in dense urban areas with amalgam burning in Andacollo, Chile and Segovia, Colombia.  Andacollo (Lumex) Segovia (Lumex) Segovia (Jerome) Mean amalgam mass* 30.4 g  114 g    Number of known emitters 6  87    Core sampling area (km2) 0.338  0.769    Measurement type/ location Tower (18 m) Mobile surveya Tower (17 m) Mobile surveya Inside non-mining buildings >50m from emitters Average (µgm-3) 0.051 0.331 0.305 1.25 6.30 1.78 Range (µgm-3) <0.01 – >50 <0.01 – >50 <0.01 – 39 <0.01 – >50 <1 –40 <1 –12 Standard Deviation 0.51 1.53 1.14 3.27 7.03 2.91 Total time (hours) 183 31 65 26   Number of data points     31 48  *Each ‘amalgam’ in this case is an individual gold/mercury mix which is heated for ~5 to 15 minutes. aSpatial statistics are taken over Segovia urban core (2010) and Andacollo’s (2009) residential/gold shop neighbourhood only.  71  Had there been equal observation density at night, the above average values and maps of average concentrations would be roughly half of those shown. It is clear that the average street level concentrations in Segovia exceed WHO (2003) tolerable levels (0.2 µgm-3), even if halved. For the remainder of this study, we use the above averages without reducing them by half, as it is equivalent to applying a factor of safety of 2 to account for the underestimation of peak values and the inherent uncertainty in the observation methods. By this logic, both study areas of Andacollo and Segovia exhibit mean concentrations that are potentially hazardous for residents according to the WHO standard for hazard (1 µgm-3), and even air at the top of the tower in Segovia exceeds tolerable levels.   3.4 Discussion  Finding that average airborne mercury concentrations in Andacollo and Segovia are possibly high enough to be hazardous to human health is unsurprising given that even non-miners exhibit symptoms of mercury poisoning (Corral et al. 2013). This corroborating epidemiological evidence suggests that it is more appropriate to directly compare short term averages that are heavily weighted toward daytime values with annual average health standards, instead of recalibrating the averages to account for nighttime concentration lows. However, it does not unequivocally indicate that elevated street level concentrations cause health problems, as there may be more complex pathways by which individuals’ dosages may exceed tolerable limits. It may be that the airborne mercury in the streets, sampled as described here, is not an accurate measure of human exposure, except perhaps in the case of slow taxi drivers. While the streets allow for the movement of mercury plumes at the ground level, the turbulence and flow of air would tend to quickly disperse. Yet most of the time people are in shops, offices, bars, homes, medical clinics, schools, and other indoor spaces. Indoor spaces potentially have much storage potential for mercury contamination, especially as many people are accustomed to using air conditioning. The previous chapter showed that the average value of all indoor sampling of non-mining buildings in downtown core buildings across  72 Antioquia was 6 times the WHO hazard limit. It is therefore possible that there are other factors that control the epidemiology of mercury intoxication in these towns, and street concentrations are an imperfect proxy. For instance, local people often tell us that the smallest scale miners who pan rivers or tailings of other operations (downstream of mines and mineral processing plants, rivers are themselves full of tailings) will frequently burn small quantities of amalgam in their kitchen. Even though panning may only be a supplemental income (often women are panners because they have less professional opportunites, especially as mothers), and so they and their children are at much higher risk than other people because they burn in their homes. Curtains, furniture, and walls could also absorb and slowly re-emit mercury over longer periods. Likely, their neighbours are also at a higher risk. Even less direct pathways are possible, as Table 2.4 shows that people may unknowingly be exposed to concentrations that are many times greater than the WHO hazard level in places such as a pharmacy or municipal hall, among other buildings, where people tend to spend long periods waiting or doing business. Furthermore, different types of clothing absorb and emit mercury at different rates; during the Andacollo field study we accidentally found that a wool sweater absorbed mercury during gold shop visits, and subsequently produced airborne concentrations exceeding 1 µgm-3 at a distance of 50 cm several days after the initial exposure (quick drying synthetic fabrics tended not to absorb mercury, however we did not formally study the properties of different textiles). Also, one gold shop operator we met in the street in Andacollo (when concentrations were otherwise near background) produced concentrations exceeding 5 µgm-3 at one metre from the analyzer inlet, even though he claimed not to have burned amalgam for many hours. Therefore it is possible that short-term exposure to high concentration mercury vapour may persist on peoples’ clothing without their knowledge, thereby greatly extending the time in which they are exposed. Furthermore, the WHO hazard level is only defined for adults, and there are many children who live and attend school in areas where average concentrations exceed tolerable levels in both towns. A factor of safety of two is the minimum appropriate level of caution given the uncertainty of the measurements and methods for assessing public health risks used here, therefore the daytime average of mercury concentration is a reasonable minimum estimate of mercury hazard.  73 3.5 Summary  This is the first thorough description of urban ambient mercury concentrations related to artisanal gold mining. The mercury mapping, monitoring and data analysis systems presented here can be used to detect, monitor, and increase awareness of the health hazard produced by artisanal gold shops and urban mercury emissions. Airborne mercury concentrations (averaged over several mobile mercury transects) which are greater than 1 µgm-3 strongly indicate the presence of a significant mercury emitting operation within 200 metres, and levels greater than 3 µgm-3 indicate a potential source within 100 metres. Mercury emission events are extreme and intermittent, usually lasting from 5 to 30 minutes, however the average concentrations in Andacollo and Segovia are nevertheless intolerably high and possibly hazardous. Mercury signals dissipate rapidly, therefore the mobile mercury sampling methods described here can be used to pinpoint locations that frequently emit mercury vapour even if they are not known in advance. Table 3.1, Figure 3.8 and Figure 3.9 show that, on average, daytime concentrations in the residential area of Andacollo exceed the WHO tolerable level (0.2 µgm-3) and in the urban core of Segovia they exceed the WHO hazard level (1 µgm-3) (WHO 2003). Cutting these values in half would produce a more rigorous estimate of expected annual average values, but sampling uncertainty is such that the daily value is a more appropriate measure of risk given the observation uncertainty. Therefore the average resident of these areas is possibly at risk of mercury poisoning, and this is likely the case for many tens of thousands of residents of artisanal mining towns worldwide. The mapping techniques presented here would enable any such site to be evaluated for the distribution and intensity of mercury contamination, and these could be used to guide interventions and estimate human exposures. However, a precise calibration of the results of the monitoring methods used here is necessary, as the precise level of hazard that they represent can only be determined by matching continuous long term monitoring with thorough epidemiological studies. Yet epidemiology studies are complex and expensive, and there are preliminary data that suggest that in both Andacollo and Segovia, miners and non-miners alike suffer the effects of mercury poisoning (Corral et al. 2013, UNIDO 2013). Therefore the mapping and monitoring methods described here can indicate potential mercury hazard areas, though further study is required to designate this with precision.  74 Even without contemplating a full epidemiological study, the methods shown here are time consuming and require costly and specialized equipment. Where the emission patterns and location of sources are known or can be estimated, it may be easier and cheaper to use dispersion modeling to assess and visualize human health hazards from mercury vapour emissions.     75 4 Modeling mercury dispersion 4.1 Synopsis  The previous section goes some way towards understanding urban mercury concentrations, and this section combines this information with data on mercury emissions and meteorology, a high resolution dispersion model and a system for testing its validity. If shown to be accurate, dispersion modeling could help estimate health hazard distributions in artisanal mining areas and simulate local scale mercury transport. CalPUFF was chosen to model mercury dispersion because it is widely used as a regulatory pollution dispersion model, and it can simulate dispersion at sub-hourly scales at which mercury plumes from artisanal gold mining activities are emitted and disperse. The nature and extent of mercury dispersion from artisanal gold mining activity has received little attention in the scientific literature and remains a significant source of error in global mercury models (AMAP 2008, UNEP 2013). Atmospheric mercury transport at the local scale is poorly understood (Lindberg et al. 2007), and there is almost no data on emissions from artisanal mining gold shops. Furthermore, it is not always possible to mount a monitoring campaign similar to that described in the previous section. It would therefore be useful to use dispersion models to estimate patterns of human exposure, to explore the extent to which mercury is transported in the atmosphere at the local scale, and to test the validity of dispersion modeling in this context. In this chapter I use high frequency meteorology data to model the dispersion of mercury plumes of known source strength and timing. This is the first attempt to apply dispersion modeling to artisanal mining mercury emissions. The goal is to show that a high-resolution puff dispersion model can reasonably estimate ambient urban mercury concentrations from gold shops, though higher resolution urban models and more field data are needed in order to properly capture the complexity of dispersion in an urban environment like Andacollo. Despite many challenges, Andacollo is in some ways an excellent site to study individual mercury emission events, as they are often relatively sparse in space and time (during our field campaign), thereby increasing the likelihood that an individual signal can be attributed to a known event. Therefore the timing and mass of mercury emitted is the most critical data. Such information is extremely rare or non-existent because gold shop  76 operators are often understandably reticent to share it openly. This reinforces the importance of local engagement and positive community relations. Detailed amalgam burning records shown in this study enable investigation of mercury dispersion at a finer scale than has previously been attempted and in a very different context to which dispersion models are typically applied. The purpose of this chapter is to show that atmospheric dispersion modeling can reproduce patterns and intensities of urban artisanal mercury contamination with reasonable accuracy. Modeling dispersion at fine scales in complex urban environments is still a very significant challenge in atmospheric science (Hanna et al. 2006). Errors in the emissions and urban structure models, as well as having only sparse observations with which to verify the dispersion model, further increase the difficulty in generating and verifying predictions of mercury concentrations. In spite of all this, the results clearly show that CalPUFF produces realistic, if oversimplified, representations of mercury plume behaviour.  4.2 Meteorology methods  Andacollo, Chile is an excellent natural laboratory in which to study mercury dispersion. The weather is usually clear, with sunny days and moderate winds, reasonably flat local terrain, fairly homogenous urban structures around the gold shops, and strong, frequent emissions. Mercury vapour measurement methods are described fully in the preceding chapter. The following describes the micrometeorology techniques and model parameters that were used. The instrumental and analytical techniques here provide essential raw inputs to drive the dispersion model.  4.2.1 Instruments   Table 4.1 lists the instruments used and variables derived from their measurements. The high frequency raw data was used to compute 30 and 60 minute averages of the variables shown in Table 4.1, and five minute averages for the wind speed and direction.  77 Data from the temperature probe (which measures ambient air temperature (T) and relative humidity (RH)) and the radiometer (which measures the net radiation (Q*) from the sky and ground) were visually inspected to ensure quality. RM Young 12005 cup & vane anemometer (which measures wind speed and direction (u)) was installed as a backup and data from this instrument are not used in this analysis. Instead, a high frequency sonic anemometer operating at 20 Hz (producing twenty data points per second) provided u and v (orthogonal horizontal wind speeds), w (vertical wind speed), 𝑢∗ (a measure of the stress generated as air passes over rough surfaces) and Qh (the sensible heat flux, or the transfer rate of heat from the surface to the air) (Figure 4.1).  Table 4.1: Tower heights of instrumentation and meteorological variables. Instrument Variables Units Height (m) RM Young 81000 sonic anemometer u, Qh,  T(virtual), 𝑢∗ ms-1, Wm-2 oK, ms-1 12 Kipp & Zonen NR-Lite net radiometer Q* Wm-2 12 Vaisala HMP 35C T/RH T, RH oK, % 12 RM Young 12005 cup & vane anemometer u ms-1 12 Campbell Scientific CR1000 logger  - - 0   Figure 4.1: Instrument boom on the day of installation. From left to right: cup and vane anemometer, temperature and relative humidity sensor, net radiometer, sonic anemometer. © Paul Cordy photo.  78  Figure 4.2: Google Earth imagery of Andacollo’s gold shop neighbourhood. Wind roses indicate the average wind speed (arrows) and relative frequency (black) in each 20 degree increment of direction over the entire field study. The cross bar on the north pointing wind speed arrow indicates 1 ms-1.   79 4.2.2 The instrument tower  Instrument siting is challenging in dense and inhomogeneous urban areas, but the World Meteorological Organization provides some guidance (WMO 2010). It involves a complex set of tradeoffs, but in general one must choose a tower site where the source area of the instruments is reasonably uniform even as it changes size relative to stability, and changes position relative to wind speed and direction. Despite challenges in finding a property owner who would let us build a tower at a site that was suitable, we found a reasonably representative and uniform fetch over which the air would pass before arriving at the instruments (Figure 4.2) The source area for the radiometer also includes a fairly representative mixture of road surface, patchy vegetation/dirt, and dusty tin roofing, though road surface and dirt yard are somewhat over-represented. The height of the tower must also be greater than the blending height zr, which means it must be at least 1.5 times the average height of the roughness elements. Given that most buildings are 3-6 metres tall and the average building height is 3.5 metres, we built a 12 metre tower (Figure 4.3, Figure 4.4). The tower was made of metal tubing slotted into a concrete flagpole stand.    Figure 4.3: The meteorological tower looking east (left) and south (right). © Paul Cordy photo.   80  Figure 4.4: Partial panorama of Andacollo taken from the tower at 6 metres height. This view looking west from South to North encompasses all directions from which the wind most frequently blows. © Paul Cordy photo. Two six metre lengths of tubing were used, whose diameters allowed one to fit snugly into the other. A straight line was scratched into the inner tube to enable exact orientation of the sensors. Once fixed to the stand, the instruments were fixed to the top of the inner tube from a scaffold. The inner tube was then pulled up out of the base tube and bolted in place, thus raising the sensors to a full height of 12 metres. Four guy wires attached at mid height and at the top of the tower kept it stable and vertical, and straightness and orientation were verified with a compass and inclinometer.   4.2.3 Data quality control  All sensors were calibrated and tested in the laboratory before being deployed in the field, nevertheless raw data can contain spurious values. The following describes the automated systems that were used to flag data values that were not physically valid.  4.2.3.1 High-frequency spike detection  Electromagnetic noise and short-term data 'spikes' were filtered out of high frequency data using a dynamic iterative standard deviation filter (e.g. Vickers and Mahrt 1997). First, individual 20 Hz data points were flagged if they fall outside a physically justified, realistic data range for each variable (Table 4.2). Individual 20 Hz data points were then flagged as spikes and withheld from further processing if they were above or below a variable-specific standard deviation threshold from a 30 minute mean (Table 4.2). Consecutive passes were then performed with the standard deviation threshold raised by 0.3 each time until no spikes were detected. Spikes must also be less than 0.3 seconds in duration, otherwise they were considered real.  81  4.2.3.2 High-frequency statistics check  Statistics of 30 minute standard deviation, skewness, and kurtosis were calculated for the vertical (w) and horizontal (u and v) winds as well as temperature (T). Empirical limits were determined for each variable and data from periods with values outside these limits were flagged as questionable (Table 4.2).   Table 4.2: Summary of quality control limits and thresholds used for high-frequency eddy covariance data.  u v w T Physically-based min/max thresholds -30/30 -30/30 -30/30 -30/30 Spike threshold (# of standard deviations) 6 6 8 8 Standard deviation (min/max) 0.05/4.0 0.05/4.0 0.02/1.5 0.01/2.0 Skewness (min/max) -3/3 -3/3 -2.0/2.0 -2.5/2.5 Kurtosis (min/max) -2.0/5.0 -2.0/5.0 2.0/15.0 2.0/15.0  4.2.3.3 Block average calculation and rotation   After the initial high frequency data quality control filters were applied, mean values and higher-order moments (including covariances) were calculated if more than 75% of data in that period were valid values. Three averaging periods were computed: 5 minutes, 30 minutes, and 60 minutes. The criteria were set differently for all 20 Hz observations of u, v, w, and T. Wind components were rotated two times so that the x-axis of the new sonic coordinate system is aligned with the mean 30 minute wind direction, and the mean vertical wind w was zero (McMillen 1988, Finnigan et al. 2002). Following Reynold's decomposition, 5, 30, and 60 minute statistics are calculated based on a simple block-average.     82 4.3 Dispersion modeling methods 4.3.1 Boundary layer meteorology  The daily evolution of the atmospheric boundary layer (ABL) controls the dilution and transport of airborne contaminants like mercury (Figure 4.5). This is the part of the atmosphere near the Earth’s surface that is structurally, thermodynamically, and compositionally altered by the properties of the underlying surface (Stull 1988).   Figure 4.5: Mixing layer evolution (from Stull 1988).  Before sunrise, a quiescent residual layer of air that had been homogenized by the previous day’s mixing overlies a stable surface layer (50-500m) (Lena and Desiato 1999) that develops as the air near the ground cools during the night. During the day, solar warming of the ground creates a strong temperature gradient in the surface layer, the lower ~5% of the mixing layer where turbulence and heating is most intense. Conduction and turbulence drive heat upward. These effects are quantified as the sensible heat flux (Qh, in Wm-2). Warm, buoyant air is displaced by cooler air around it, generating eddies and entraining the air above. Turbulence is also produced by stress that is generated as air  83 moves over friction elements on the ground (this effect is quantified as the friction velocity: 𝑢∗, in ms-1).  The turbulent “mixing layer” thickens rapidly in the morning by consuming the stable and residual layers, reaching a maximum height of 1-2 km in the afternoon, when heating is the strongest. Turbulence in the mixing layer largely homogenizes its temperature and composition, including pollutant concentrations. It has been shown that for calm, sunny days 20% of surface heat flux powers entrainment (Stull 1988). This energy is dissipated in the 10-100m thick entrainment layer as warmer, quiescent air is mixed into the cooler air below. This temperature difference at the top of the entrainment layer (symbolized by Δθ, in oK) functionally defines the height of the mixing layer (ℎ, in metres) and is an important control on mixing layer growth. This capping inversion acts as a lid that separates the mixing layer from the air above. The environmental lapse rate (γ, in oKm-1) can be measured from a sounding (vertical weather profile), but I use 𝛾 = 0.006  as in Nath and Patil (2006), and Batchvarova and Gryning (1994), because that is close to the value of the standard atmospheric lapse rate (Stull 2000). Mesoscale subsidence and advection are sometimes included in boundary layer models (Steyn and Oke 1982), but these terms are omitted from the Batchvarova and Gryning (1994) model used here.   4.3.2 Sensible heat flux   The sensible heat flux (Qh) is calculated as: 𝑄? =   𝑤?𝑇? ∙ 𝑐? ∙ 𝜌 where            𝑐? = 1004.67 ∙ 1+ 0.84 ∙ 𝑟        and            𝑟 =    ?. ™? ∙????   𝑤?𝑇? is the covariance of vertical air motion and temperature, cp (Jg-1K) is the specific heat capacity of air, ρ (gL-1) is the density of air, 𝑟 (gkg-1) is the mixing ratio of water in air. The mixing ratio varies with vapour pressure, 𝑒 (kPa), which is derived from the relative humidity and the saturation vapour pressure of water:  84 𝑒𝑒? =    𝑅𝐻100%  where the saturation vapour pressure (es in kPa) is given by the Clausius-Clayperon equation:  ln 𝑒?𝑒? =    𝐿?𝑅? ∙ 1𝑇? − 1𝑇   in which 𝑒? = 0.611 kPa and 𝑇? = 273 oK are constants. Lv = 2.5x106 Jkg-1 is the latent heat of vaporization for liquid water and Rv = 461 JK-1kg is the gas constant for water vapour. Qh values are also corrected to account for high frequency flux losses based on sonic anemometer acoustic path length (Moore 1986). In these corrections, the instrument-specific path length of the sonic is (0.147 m), the zero-plane displacement (zd) was estimated to be 2.2 m, or 2/3 the average height of the canopy (effective canopy height = 3.5 m), and the Monin-Obukhov length (which is explained in section 4.3.7). Pressure was fixed at 98 kPa (average pressure for 1000 metres elevation) for lack of a digital barometer. Acoustic temperatures were computed using Lanzinger and Langmack (2005). Table 4.1 lists the instruments used and variables derived from their measurements. The high frequency raw data were used to predict hourly averages of the variables shown in Table 4.1. The wind speed and direction during known mercury emission events was replaced by the vector average of all 5 minute wind averages during that period to increase the likelihood that the hourly averages are representative of the wind conditions during each emission event.  4.3.3 Friction velocity   The friction velocity (𝑢∗) is used to parameterize the surface stress (Stull 1998), and is given by: 𝑢∗ =    𝑢′𝑤′? + 𝑣′𝑤′? ? ?  85 where 𝑢′𝑤′ and v’w’ are the 30 minute average covariances of horizontal and vertical motions of air, and reflect the nature of eddies at the height where these quantities are measured. The friction velocity is an important velocity scaling parameter for determining mixing heights.  4.3.4 The slab model  Figure 4.6 shows the mixing layer heights predicted using a slab model of mixing layer growth as in Batchvarova and Gryning (1994).   Figure 4.6: Mixing layer growth model output for the Andacollo field campaign.  Most slab models of mixing layer growth (Steyn and Oke 1982, Nath and Patil 2006) use the surface heat flux to model the entrainment rate, and thus mixing layer growth. Slab models generate a reasonably accurate model of mixing layer evolution using the potential temperature gradient above the mixing layer (γ), kinematic surface heat flux (Qh) and friction velocity (𝑢∗) to predict mixing layer heights (ℎ):   𝜕ℎ𝜕𝑡 = 𝑄?𝛾ℎ    ∙ 1+ 2𝐶? + 2𝐶?𝐶?? ∙ 𝑢∗?𝑇𝑔𝛾ℎ?   86 where 𝐶? = 0.2 and 𝐶?𝐶?? = 0.5 are empirical constants set by Batchvarova and Gryning (1994), and 𝛾 =  0.006. This equation is solved iteratively on an hourly basis, and ??™ . Δ𝑡 is added to h to give the mixing layer height for the next hour. Nocturnal mixing heights are more difficult to model, so the initial mixing height each day was set arbitrarily to 100 metres, which is a plausible height according to previous field studies (Mahrt et al. 1979, Poulos et al. 2002).   4.3.5 Urban roughness   Urban structures generate turbulence as air moves over them, and therefore it is important to quantify the relative roughness of an urban surface. The roughness length (𝑧?, in metres) is a key parameter that must be calculated or estimated for any given surface in order to model turbulence and dispersion. More rigorous methods of determining 𝑧? use micrometeorology measurements or empirical relationships based on the average morphology of friction elements. Grimmond and Oke (1999) assimilated many of these studies from various urban surfaces and developed a qualitative visual method for estimating several key aerodynamic parameters for a range of heights and densities of structures.   Figure 4.7: View of Andacollo’s gold shop neighbourhood from a hill to the North. Note the Basilica to the lower right beside the mercury contaminated tailings pile, and the piles stretching along the central valley. © Paul Cordy photo.  87 They suggest that oblique aerial photography is best for quickly assessing the mix of elements, their packing, and heights. Figure 4.7 was taken from a hill north of Andacollo, and the evening shadows provide excellent relief for observing the structure of the gold shop neighbourhood. I also used Google Earth imagery and field measurements of individual building dimensions (Table 4.3) to estimate Andacollo’s 𝑧? as ~0.5 according to the Grimmond and Oke (1999) method. 𝑧? = 0.5 is a mid-range value for low height and density urban surfaces with one or two story single houses, gardens, small trees, mixed houses and small shops, warehouses, light industrial development and few trees.  The mean building height (𝑧?) is 3.5 m (Table 4.3).    Table 4.3: Andacollo building measurements, in metres. height width height width 3.2 4.3 3.3 4.5 3.4 5.1 4.9 4.7 2.5 4.1 4.8 4.8 2.9 4.2 2.3 4.1 3.2 4.4 3.1 4.1 5.2 5.5 3.1 4.9 2.5 4.6 3.1 4.3 3.2 4.5 3.3 4.9 3.6 5.0 5.0 5.1 5.4 5.6 2.5 4.2 2.7 4.2 mean height and width 3.0 4.3 3.5 4.6      Although the gold shop neighbourhood has slight inclines that increase at the edges, I modeled it as a flat surface with an elevation of 1000 metres to match the area’s actual mean elevation. Figure 4.8 shows typical buildings in the area. 4.3.6 Urban structure model  Effects of adjacent buildings on dispersion were accounted for in the coarse CalPUFF method, which is to list the height of all nearby buildings in 36 directions at 10 degree increments. Table 4.4 displays the parameters of the urban landscape model.  88 Table 4.4: Modeled building dimensions at various directions relative to the principal gold shop. Azimuth (o) Height (m) Length (m) 20 3.5 40 70 3.5 20 180 4 40 270 3.5 16     Figure 4.8: Typical buildings in Andacollo. The sign reads “We are neighbours who take care of each other.” © Paul Cordy photo. In the CalPUFF building downwash model, which simulates the turbulent effects of nearby buildings on pollutant dispersion, the building dimensions are very approximate. The forty metre segments are meant to represent the stadium fence as long buildings. The other two segments are large because the individual houses in those directions are bunched tightly and effectively create a single building. All other directions have no buildings along them that are close enough to warrant inclusion. Clearly this is a  89 highly simplified and approximate representation of the buildings that would cause building downwash, but a more exact description is beyond the scope of this work.  4.3.7 Stability  The stability of air in the boundary layer is determined by the interplay between buoyancy and shear. The Monin-Obukhov length (L, in metres) is a useful surface layer scaling parameter that incorporates Qh as a buoyancy term and 𝑢∗ for shear.  𝐿 =   − 𝑢∗?𝑇κg𝑄?  where 𝜅 = 0.4  is the von Karman constant. L is roughly proportional to the height above the surface at which buoyant factors first dominate over shear production of turbulence (Stull 1988). Negative values imply instability, with buoyancy being dominant, and positive values indicate stability.  The CalPUFF dispersion model uses the Pasquil-Gifford (PG) stability scheme. It assigns a categorical value based on mean wind speed, insolation, and cloud cover. I assigned PG categories for each hour by using Golder’s (1972) nomogram, which relates L to PG for any given 𝑧? (Table 4.5).   Table 4.5: Stability v. Monin-Obukhov length for zo = 0.5.  Pasquil-Gifford stability 1/L A < -0.105 B ≥ -0.105 to < -0.0402 C ≥ -0.0402 to < -0.01 D ≥ -0.01 to < 0.01 E ≥ 0.01 to < 0.045 F ≥ 0.045  90 4.3.8 CalPUFF  The previous chapter showed that urban artisanal mining mercury emissions are intermittent, short lived, and extremely concentrated at the source. Also, the source is often not well defined by a stack. There are no dispersion models designed to handle this kind of pollution event, however a Gaussian puff model is perhaps better suited than any other. If one considers mercury emission events to be a chain of puffs, then one can track the puffs as they disperse downwind. Supposing statistically stationary and homogeneous turbulence, the growth of a puff released to the atmosphere can be calculated as a function of time (Caputo et al. 2003), and its movement relates to the wind speed, direction, and profile as well as the buoyancy, atmospheric stability, and release height of the puff. CalPUFF is a generalized non-steady state air quality model designed for regulatory use (Scire 2000). It models pollution emissions as a series of Gaussian puffs that are individually tracked as they are influenced by changes in the meteorological and landscape parameter fields while dispersing downwind. Puffs also shear in response to changes in wind direction with height, and are diverted by terrain features. The resulting concentrations at receptor locations are computed for a given temporal resolution measured in seconds. The timing and duration of emissions events can also be specified at a resolution of one second, with a spatial dispersion resolution of tens of metres. This study employs CalPUFF to model mercury dispersion using meteorological tower data, emissions inventories from participating gold shops, and mixing height model output as discussed above. Model receptors located at 20 metre intervals along the mercury sampling transect road lengths and at the stadium towers are used to generate maps and time series of mercury concentrations that can be compared with mercury concentrations as measured in the field by Lumex observations within 10 metres of the receptor locations. The spatial domain of the model was limited to the gold shop neighbourhood in which the roving measurements were made, which is an area of approximately 2 km2.  Normally, CalPUFF computes mixing heights using gridded meteorological and landscape fields. For simplicity, and due to the very small scale of the urban area being modeled, a spatially homogenous meteorological field was projected from a single meteorological station (using the ISCMET option), and constant landscape parameters  91 were used. It could be argued that constant meteorological and landscape parameters effectively reduce CalPUFF to a simple Gaussian plume and therefore the use of a more complex model is unjustified. However CalPUFF has many other positive attributes that make it a sensible choice. As the standard model for assessing regulatory compliance in the United States, it is the most recent in a long line of regulatory models that have been tested in many contexts and is likely to perform reasonably well when applied in new situations (Scire 2000). It also allows inputs, outputs, and model resolutions in seconds, which is useful when the emissions usually occur over five to twenty minutes. Table 4.6 lists key parameter values used in CalPUFF.    Table 4.6: CalPUFF parameters  Parameter Value Description/units Datum WGS-84  Base time zone  – 4 hours time relative to Greenwich mean time NX and NY 200 number of grid points in x and y direction NZ 10 number of grid points in z direction Grid spacing 0.01 km Origin (x) 298.82954 UTM Origin (y) 6652.5179 UTM Model time step 10 seconds Post processing time step 10 seconds Wind speed profile ISC urban-1 wind profile power law exponents  Land use type Urban  Roughness length 0.5 m Leaf area index 0.01  Dispersion regime urban  Tower base elevation 1000 m Tower height 12 m N. latitude -30.232 degrees W. longitude -71.0789 degrees  The mercury plumes were modeled as puffs (individual parcels of Gaussian distributed contamination) rather than slugs (elongated puffs that take the place of many puffs for computational simplicity). Several assumptions were made in the absence of local parameter values or to simplify modeling. Mercury was assumed to have no chemical transformations, no wet or dry removal, nor did I use any terrain adjustments. Default parameter values were used where not mentioned explicitly. Model background  92 mercury for the Andacollo gold shop neighbourhood was set to 0.025 µgm-3, as determined by taking the mean of all time periods in each time series between start calibrations and the first appreciable signal, and between the last mercury signal and end calibrations. The variability of the background data (mean: 0.0248 µgm-3 standard deviation: 0.0191) could reduce the overall strength of correlations, especially as observations are subject to variability in instrument calibration and drift removal errors, whereas the background of the model output is always 0.025 µgm-3.  4.3.9 Emissions   There were 28 recorded mercury emission events during the Andacollo field campaign for which timing and amalgam weight difference were measured (Table 4.7; these are hereafter referred to as “known emissions”). The stack exit temperature was not measured directly because the roof and chimney were unsafe and mercury exposure extreme (Figure 4.9). Propane torches of the kind used in Andacollo can produce temperatures up to 1200 oK (ANL 2007) but most would be considerably cooler, especially given their poor operating condition. At one atmosphere of pressure, mercury evaporates at 629 oK, therefore the initial vapour temperature must be higher. Ambient air mixing from the open face of the fume hood lowers the temperature of gases in the stack, and ANL (2007) show that aerosols are forming in the first few metres of the exhaust pipe, therefore temperatures must already be below mercury’s boiling point at the margins of air flow. Given that CETEM (1992) showed that only 2% of mercury emissions contributes to soil contamination within 1km of the source, it is reasonable to assume that most of the exhaust gases from gold shops are emitted and only a small percentage precipitate in the stack. I therefore set the stack exit temperature to 700 oK and modeled the mercury  Figure 4.9: Smokestack of the principal gold shop. © Paul Cordy photo.  93 emissions as if all of it remained entirely in vapour form. Mercury was assumed to be emitted at a constant rate during the entire burn, with emission rates given by the total mass of mercury emitted divided by burn time. The stack diameter was 25 cm with a rain cover that is explicitly modeled by CalPUFF. Rough calculation of the stack exit velocity was done using the above values with the ideal gas law, Archimedes’ buoyancy equation, and Newton’s laws. Table 4.8 gives the values for all parameters that were not set to the default value.  Table 4.7: Original timing and masses of amalgam burns recorded in the principal Andacollo gold shop during mercury vapour sampling (April and May 2009). Day  Start time End time Start mass (g) Final mass (g) Hg mass released(g) Burn time (min) Emission rate (g/s)  Notes 114 1400 1430 38 9.2 28.8 30 1.60x10-2  122 1000 1010 4 1.2 2.8 10 4.67x10-3  122 1105 1115 3.7 1.4 2.3 10 3.83x10-3  124 1300 1315 20 5.8 14.2 15 1.58x10-2  124 1831 1837 24.5 5.8 18.7 6 5.19x10-2  125 1310 1320 50 16.6 33.4 10 5.57x10-2  125 1400 1410 3 1.5 1.5 10 2.50x10-3  125 1640 1700 19 5 14 20 1.17x10-2  125 1820 1830 79.2 78 1.2 10 2.00x10-3  125 1831 1842 138 118.8 19.2 11 2.91x10-2 Acid boil after 126 1200 1210 8 3 5 10 1.22x10-2  126 1230 1240 4.5 2.3 2.2 10 3.67x10-3  127 930 940 38 9.5 28.5 10 4.75x10-2  128 1730 1740 122 45 77 10 1.28x10-1  129 1000 1015 28 7 21 15 2.33x10-2  129 1030 1040 44.5 15.8 28.7 10 4.78x10-2  129 1220 1230 0.2 0 0.2 10 3.33x10-4 Acid boil only 129 1310 1315 12.5 4.3 8.2 5 2.73x10-2  129 1900 1905 12.5 4.5 8 5 2.67x10-2  131 1200 1215 18.8 18 0.8 15 8.89x10-4  132 1500 1510 18 5 13 10 2.17x10-2  132 1800 1810 27 8.2 18.8 10 3.13x10-2  132 1820 1830 12 4.2 7.8 10 1.30x10-2  133 1520 1530 10 3.3 6.7 10 1.12x10-2  133 1540 1550 27.9 12.5 15.4 10 2.57x10-2  133 1640 1645 18.2 8.3 9.9 5 3.30x10-2  133 1720 1730 20.3 7 13.3 10 2.22x10-2  133 1735 1740 18 6.4 11.6 5 3.87x10-2   94 4.3.10 Sensitivity testing   Model sensitivity to the aforementioned parameters (Table 4.8) is negligible within the stated ranges (only values above background are used in sensitivity testing). For all parameters, correlations greater than 0.98 (p-value <<0.01) were found between model output produced with the parameter value used in this study and the model output produced by the sensitivity testing variant of the same parameter. Stack exit temperature and velocity were tested for their impact on dispersion model output, as well as zo. The other parameters were only tested for their impact on the mixing height model. True values of the variables are almost certainly within the specified ranges, so I suspect that they contribute relatively little error in this analysis relative to much more uncertain parameters such as emissions timing.  Table 4.8: CalPUFF parameters that were set to physically reasonable values, and the degree of correlation between model output obtained using the chosen value and that which results from testing the sensitivity of the model within the stated range.  Variable Chosen value Range Correlation Stack exit temperature / velocity 700 oK / 10 ms-1 500-900 / 8-12 0.991-0.997 Roughness length (zo) 0.5 0.2-0.8 1 Nocturnal mixing layer height 100 m 50-200 0.999-0.998 Environmental lapse rate (γ) 0.006 0.004-0.008 0.999  4.4  Error 4.4.1 Emissions error   The timing and mass of mercury emitted were recorded on an amalgam roasting log form that I provided to each gold shop operator, who was also given a watch that we synchronized with our instruments. The form had columns for the date, burn initiation time in minutes, mass of amalgam before and after burning, burn termination time, and a box for comments such as ‘nitric acid boil only’. I also gave the operator of the principal gold shop a digital scale with which to verify the mass determined using his old analog scale. The mass values agreed perfectly within the precision of both methods (+/-0.1 g). I believe that the masses in the principal gold shop were accurately recorded without any  95 deliberate alteration because the operator was instructed by the shop owner to provide all of the information requested, and as an employee of the shop owner he had no obvious personal stake in misleading our study. Given that measuring the mercury after the burn was something that had to be done in any case (and there were direct costs for inaccuracy), that is likely to be very accurate, as is the mass of the gold before burning. However, the exact timing of the burn is perhaps less reliable. The amalgam roasting logs show time values that mostly end in 0 or 5 minutes, so I assume this to indicate the maximum expected precision of the emission times. It is plausible that the value is very different from the actual value by as much as fifteen minutes if one considers the nature of the gold shop operator’s job.  In a typical sale, the client walks up to the shop where the operator stands at the door waving to neighbours and passers-by. A miner walks up and they head inside to the burning room, making small talk as the miner pulls from his pocket a bit of crumpled paper with one or two crumbling little bits of amalgam in it. The operator may or may not always have tared either balance with a new piece of paper (as instructed), but instead just used the paper that the miner had used to wrap the amalgam. Small talk continues as the operator breaks up and flattens the amalgam in the roasting pan, then fires up the torch. Typically, the amalgam is roasted for 5 to 15 minutes, and on rare occasions (such as if the colour of the gold afterward suggests that there are lots of impurities) the operator might boil the doré in nitric acid to leach out the undesirable metals. The operator would then weigh the gold, and negotiate a price based on its mass and colour, knowledge of the miner’s deposit and previous gold quality, as well as their personal history. Once the transaction is completed they chat some more and the operator walks the client to the door to finish the conversation and see them off. I suspect that the operator sometimes came back to note the emission times long after the actual event and wrote plausible estimates. Finally, it is worth noting that the gold shop operator had been roasting amalgam in poorly ventilated spaces for more than 25 years. Judging by his penmanship, simple math skills and wit, he is remarkably resilient to this exposure, however even some native Spanish speakers found his speech at times difficult to understand.  Therefore the emissions model is one of the most significant possible sources of  96 error, and I estimate that this could affect the precision of emissions timing by 5 to 15 minutes. The masses of amalgam are likely to be quite precise, as the measuring devices have the correct precision and agree with one another while the weight of a small piece of paper is lower than the 0.1 gram precision of the scales.  Stack exit temperature, stack dimensions, and exit velocity parameters were estimated visually and logically because there were no reasonable and safe means to measure them exactly. Just as it is impractical and excessively risky from a health perspective to post an observer in the gold shop to watch each burn event, climbing onto the flimsy roof with a ladder to reach the stack exit and measure temperature and exit velocity was out of the question. However, the model is not very sensitive to these parameters (see Table 4.8). 4.4.2 Meteorology error   Pressure was not explicitly measured in this study, and we assumed it to have an average value for the elevation of the gold shop neighbourhood: 97.7 kPa. We used 30 minute values of moisture calculated from the relative humidity sensor to correct the sonic virtual temperatures, when ideally we would have used fast-response moisture data from an infrared gas analyzer instead. We estimate that both of the above would cause errors of up to 2% in the calculation of sensible heat flux (Qh) (Crawford et al. 2010). Furthermore, while the net radiation should always balance the sum of the sensible heat flux, latent heat flux, and storage (mostly through conduction in the ground), it is well known that most studies underestimate these fluxes by up to 20% because they cannot capture the full spectrum of turbulence frequencies. We expect similar energy balance closure problems in our case.   The choice of tower site is a possible source of errors in estimating heat flux and turbulence parameterization, and the variability of errors will depend on the wind speed and direction because of the variability in land cover and urban structures in all directions. The most affected wind directions are from the east, as the tower is on the eastern edge of town and the ground east of there is smooth open desert with few manmade structures. Fortunately, the source area for the turbulence measurements is fairly representative of the gold shop neighbourhood as a whole when the wind blows  97 from most directions, and especially from the north (the most favorable direction for detecting mercury burn events from the principal gold shop at the stadium towers). The radiometer was recalibrated after the Andacollo campaign and the calibration factor was found to have changed by 2% relative to the original factor determined at the factory, therefore the radiometer readings could also be off by about 2%. All other sensors were operating optimally within manufacturer specifications. 4.4.3 Model error 4.4.3.1 Urban roughness   CalPUFF only accepts very limited information about the urban landscape: the roughness length (zo), and the nearest building dimensions and distances in all 10o separated directions. A constant elevation was used despite the fact that the study area slopes downward at the edges, which could contribute some error in modeled ground level concentration. Furthermore, there is no information about the alignment of street canyons relative to the wind, and the stadium is modeled as several thin buildings in a row. I therefore expect the model will not be able to simulate the microscale variability of mercury in this landscape very accurately. In particular, CalPUFF does not simulate recirculation in street canyons that block or temporarily sequester contaminants, or accelerated transport in wind parallel streets. Furthermore, there is likely to be significant storage and slow release of contamination by the surrounding buildings. It is probable that the gold shops themselves, with their extreme concentrations indoors, leak contamination into the street and other buildings for some time after the burn event stops, and other buildings might cause similar lags in dispersal. These effects are not accounted for in the model, and likely cause significant errors at ground level. 4.4.3.2 zo and Pasquill categories   Varying zo has little effect on the mixing height. Pasquill-Gifford stability category is also chosen based on this zo, but is not sensitive to variations of zo within a range that is reasonable according to Grimmond and Oke (1999).      98 4.5 Data inventory   Table 4.9 summarizes the amount of each type of data that was collected. “Measurement days” are defined as days in which meteorological measurements were made at the same time as mercury vapour measurements (in this case it is not the number of measurement hours expressed in days). One “measurement period” is any hour or part of an hour in which the specified type or combination of measurements were made. Recall that “known emission events” are amalgam roasting events whose timing and change in amalgam weight were recorded on an amalgam roasting log form provided to each gold shop operator. Fortunately, almost all known mercury emission events coincided with measurements, however in some cases there were no observations of elevated concentrations because the mercury analyzer did not happen to sample the mercury plume at the positions they were in. Also, many measurements were made during periods and at locations in which observed concentrations were consistent with a mercury emission plume but there were no known emissions events to which they could be attributed. Considering that we were never alerted when emissions would be or were taking place, we were quite lucky that our strategy of simply measuring as much and as systematically as possible worked as well as it did.  Table 4.9: Summary of mercury measurements and known emissions. Measurement type Hours Measurement days   Tower 238.93 16    Mobile 67.84 12    Total 306.77     Combined measurement type Number of measurement periods Known emission events during measurements Mobile only 31 2 Tower/mobile 41 6 2 towers 40 8 1 tower 176 8 Total 294 24  Principal shop All shops Number of known emissions  28 34 Total Hg emissions (grams) 412.2 620.6 Total emission time (hours) 5 6.6  99 4.6 Model evaluation  My goal was to test whether the modeled and observed mercury signals were coincident in space and time, and whether the predicted plume concentrations were similar (ideally, to within a factor of three). First, I assessed the overall accuracy of timing, position, and concentration taken together by pairing observed and modeled values at a high resolution.  Regular periodic gridded model output and mobile monitoring data were compared using a fixed set of receptor locations that are coincident with the 20 metre separated observation point averaging locations. In order to match the timing of the model, receptor output data at each location were compared with observations taken within 10 seconds of the model output time and 10 metres of the receptor location. Each tower location is also a model receptor, and 10 second averages of tower observations are paired with the equivalent model average at the same time. Matching the observed data to the model output in these high-resolution pairs facilitates various analyses. Scatterplots and correlations for these pairs are used to assess model accuracy during each individual amalgam burn event.   Of course it is extremely difficult to be accurate at that resolution, therefore I also take the average of all data pairs during each amalgam burn event (which I describe as “event averages”).   There are several mercury emission events in which a strong mercury signal appears shortly before or after the emission event time listed in the amalgam roasting log. In cases where these observed signals seem to match the model output well if not for the apparent mismatch in time, the timing of the modeled emissions for those events was adjusted to correspond with the observed signal. In this way I hope to evaluate the maximum possible accuracy of the model at a precision of +/-15 minutes or less, which is close to both the estimated signal temporal error and mean amalgam roasting duration for Andacollo. Therefore event times were adjusted by up to 15 minutes as listed in Table 4.10. The wind data for event 6 was also adjusted because the wind was calm and the anemometers did not register any valid values. This resulted in average wind speed and direction for the hour that were skewed toward periods where the wind was stronger. The mercury observations seemed to indicate that calm winds were allowing the plume to  100 linger so the wind speed was set to 0.5 m/s and given a direction that makes it hit the tower sensor directly. For each analysis in which adjustments were made, the results from the original amalgam roast timing and meteorological parameters are compared with the results of adjusted inputs. For all analyses I reject the null hypothesis of no correlation if the p value is less than 0.05.   4.7 Results  There were 28 known mercury emission events from the principal Andacollo gold shop for which we obtained a full record of weights and burn times. Of these, four occurred in the absence of any measurements and 13 events were not sampled in the path of the plume predicted by the model. Significant mercury signals occur coincidentally with model output in the remaining 11 events (see Table 4.10). ‘Adjusted’ refers to the model output that resulted from altering the emissions timing, and ‘Original’ output uses unaltered timing. The model output 10 second averages for all receptors along the streets and at the tower locations, and 10 second averages of tower observations were computed for comparison.  Table 4.10: Correlations (r, α = 0.05) of modeled versus observed concentration for known mercury releases from the principal gold shop, and adjustments made to the timing of emissions are shown. Cases in which there was no overlap between model and observations are marked n/a.   Start time adjustment Finish time adjustment Original Adjusted Event number r p r p 1 -5 0 -0.13 0.2 0.30 <0.01 2 0 0 n/a n/a 0.44 0.04 3 -5 0 -0.44 0.02 0.52 <0.01 4 10 5 -0.07 0.56 0.10 0.24 5 -5 0 -0.14 0.31 0.25 0.03 6 -15 -15 n/a n/a 0.54 <0.01 7 -5 -5 n/a n/a 0.27 0.06 8 0 0 n/a n/a -0.13 0.38 9 0 0 0.01 0.88 0.06 0.43 10 15 15 -0.04 0.75 -0.04 0.75 11 5 0 -0.4 0.04 0.03 0.02  101  I maintained the one-second resolution for the street level observations in order to preserve the spatial resolution, accepting that the temporal average period mismatch is not ideal. Without time adjustments, the correlations are poor; the model signals are mostly misaligned, but often roughly match the intensity and duration of signals observed within 5-15 minutes. Some timing adjustments therefore produce strong agreements between the model and observations.   4.7.1 Event averages  Most modeling efforts focus on longer term averages because extensive micro-scale variability in wind and turbulence is difficult to observe and simulate. Yet the low spatiotemporal coverage and high discontinuity of the observations presented here are insufficient for comparison with model outputs of long-term averages. Furthermore, the brief, intense, and intermittent nature of the emissions demand high-resolution comparison in order to treat them as the discrete events that they are. Therefore I averaged the observations and model output over all high-frequency pairs over the time in which the mercury amalgam burning was taking place. A separate such average was computed for each mercury amalgam burning event, which lasted between 5 and 15 minutes, and I refer to these as ‘event averages’.    UPXFSPOMZFigure 4.10: Logarithmic scatterplot of mercury concentration averages for each amalgam burn event. Circles mark events in which mobile and tower observations are included in the average, and crosses indicate events with only tower observations.   102  Figure 4.10 shows that the linear correlation between modeled and observed event average concentrations is fair (0.43) and significant (<<0.01).  4.7.2 Spatial average analysis  The spatial average as defined here is the average of all values recorded at each receptor location, cumulated over the entire time in which amalgam burn events were taking place as listed in the amalgam roasting log (+/- five minutes, as that is the precision of the emission timing). The spatial average modeled mercury concentrations for the set of receptor locations (including towers) show no correlation with the observations (Figure 4.11). This is unsurprising, as the spatiotemporal coverage is low, we only have emissions data from one of at least 6 or 7 sources in the area, and the data in Table 4.10 show several events where correlations were negligible or negative.   TQBUJBMBWFSBHFMPDBUJPOTUPXFSMPDBUJPOTQSJODJQBMHPMETIPQFigure 4.11: Average modeled (contours) versus spatial averages of observed (circles) mercury concentrations for all model receptor locations over all amalgam burning events. Colours plot the base ten logarithm of average mercury concentration for each receptor location.   103 However, it is clear that at select points in space or during known burn events there is strong model agreement, and the average spatial model output clearly marks the location of the principal gold shop. Both modeled and observed concentrations attenuate with distance from the principal shop, though observations are more variable, perhaps due to undocumented emissions in the study area, small sample size, or some effect of urban structure that increases persistence or range of signals at the rover observation height. Interestingly, the immediate vicinity of the principal gold shop appears to have much lower average observed concentrations than just one receptor farther away, but the model does not capture that phenomenon. This is likely the effect of the chimney being high, rooflines low, and the street canyons narrow; pollutants likely don’t have time to mix down to street level until ten to one hundred metres away. The model seems not to account for this effect despite modeling plume rise and roughly specifying urban structure. Spatial modeled and observed averages do not correlate (Figure 4.12), but the following section shows that if the timing of the emissions are adjusted there are many individual events where the model accurately predicts tower concentrations.   Figure 4.12:  Logarithmic scatterplot and linear correlation of spatial averages.    104 4.7.3 Tower series  The soccer stadium towers immediately downwind of the principal gold shop provide the best cases with which to compare modeled and observed mercury concentrations. The best of these is a brief period in which 5 amalgam were burned in sequence during particularly fortuitous wind conditions. This first time series of tower observations all occurred within the same two hour period on Wednesday, May 13, 2009, and the mercury plumes were observed by two analyzers at 18 metres height on stadium towers arrayed downwind of the principal gold shop. The entire duration of the tower series is shown in Figures 4.13 and 4.14, though some parts of individual series for the two towers are missing due to instrument difficulties. The amalgam roasting log kept by the shop operator confirmed that five discrete mercury release events occurred during this period and the observations of high concentration show a similar number of major peak value clusters.  There may be a better set of adjustments that would optimize the correlations, however, the minimal adjustments used here are consistent with the precision of the amalgam log and all reasonable adjustment parameters yield similar correlations. It is interesting to note that the peak of both modeled and observed mercury concentrations are lower at the more distant SE tower. This suggests that dilution is rapid and that the model may be accurately predicting the dilution rate.  Table 4.11: Timing, masses, and emissions rates of amalgam burn events that were detected at both stadium tower analyzer locations. Event Start time Finish time Mass before burn (g) Mass after burn (g) Hg emitted (g) Burn time Emissions rate (gs-1) 1 1520 1530 10 3.3 6.7 10 0.0112 1* 1515 1530 10 3.3 6.7 15 0.0047 2 1540 1550 27.9 12.5 15.4 10 0.0257 3 1640 1645 18.2 8.3 9.9 5 0.033 3* 1635 1645 18.2 8.3 9.9 10 0.0165 4 1720 1730 20.3 7 13.3 10 0.0222 4* 1700 1710 20.3 7 13.3 10 0.0222 5 1735 1740 18 6.4 11.6 5 0.0387 5* 1715 1725 18 6.4 11.6 10 0.0193 * indicates events for which emission timing was adjusted to fit the observations.   105    Figure 4.13: Time series of events 1 to 5, in chronological order. Modeled (lines) versus observed (dots) ten second average mercury concentrations for the two stadium towers in Andacollo are shown with five minute average wind speed and direction in blue arrows and hourly averages in black (Wednesday, May 13, 2009). Only the time adjusted model output is shown.    106   Figure 4.14: Logarithmic scatterplots for events 1 through 5.  Three more events were observed while there was a single mercury analyzer in the Northeast stadium tower. Event 6 occurred on Saturday, May 9, 2009 (Figure 4.15). If the recorded duration is used but the event is modeled as beginning at 9:45 instead of 10:00, this event has the closest agreement between the model and observations in this study. The sudden appearance of the model signal and its slow attenuation also roughly match the observed signal.    107 Table 4.12: Timing, masses, and emissions rates of amalgam burn events that were detected at only the NE stadium tower analyzer location. Event Start time Finish time Mass before burn (g) Mass after burn (g) Hg emitted (g) Burn time Emission rate (gs-1) 6 1000 1015 28 7 6.7 15 0.0233 6* 950 1005 28 7 6.7 15 0.0233 7 1200 1210 8 3 5 10 0.00833 7* 1155 1205 8 3 5 10 0.00833 8 1230 1240    4.5 2.3 2.2 10 0.00367   If the amalgam log timing is used without adjustment, there is no overlap of observed and modeled mercury signals for this event. The winds during the event were very light and variable in terms of direction, and this is likely the cause of the missing values of five-minute average wind in the middle of the series (either instrument error near the sensor’s low detection limit or removal of data by the high frequency statistics quality control). The hourly average is therefore not representative of this moment at the margin of the hour where there is a gap in wind observations, and it was replaced with an average wind speed and direction that is plausible given the five minute wind observations adjacent to the data gap and the observed mercury concentrations. The wind was set to 0.5 m/s and the direction is pointed straight at the NE stadium tower where the analyzer was located on this occasion. Changes in the timing and wind velocity fall within a reasonable range given the precision of the amalgam burn time observations and the variability of wind during this event. The 15-minute time shift exceeds the apparent precision of the emission time records (which are largely recorded in five minute units), however the shop operator likely wrote the times in after completing the negotiations, purchase, and possibly several minutes of small talk with the client.  For several reasons, it is more likely that the adjusted timing is correct than to have obtained such remarkable agreement between model and observations due to some other emissions wafting in from nearby. First, the wind was too light to bring a plume from another source to the tower analyzer, unless it emanated from the same set of buildings as the principal gold shop. Second, the observations and model output both suggest that plumes are discrete and limited in space and time, and that it is unlikely that extreme mercury values would appear at the tower from a distant source. Furthermore,  108 recorded events are relatively infrequent, as are observations of high concentrations in general, and therefore the joint probability of both events randomly coinciding is even lower.   Figure 4.15: Model (line) and observed (circles) mercury concentrations for event 6 (right), and the logarithmic scatter plot for this event. Blue arrows indicate wind speed and direction of five minute averages of wind, and the black arrow is the hourly average used in the model.  109 Finally, the tower concentrations are constant and near local background for a half hour before and after events 6-8, which supports the notion that there were no errant undocumented emission events in the vicinity at this time. The strongest argument in favour of the adjusted modeled event corresponding to the observed tower signals is the fact that the two time series are so smoothly variable with closely matched growth and decay of concentrations.   It is possible for events to reach the tower from afar, but that would require wind speeds that could carry the signals some distance and higher wind speeds would produce more turbulence and variable signals at the tower. Furthermore, the amalgam size and mercury emission rates would have to have been such that they mimicked quite precisely the concentration pattern of much closer emissions despite dilution during transport.  Therefore given the degree of agreement, calmness of wind, and implausibility of randomly aligning emissions from alternate sources, I consider this remarkably close correlation between the model and the observations as strong evidence of the validity of CalPUFF. Even if the adjustments aligned the model output with the signal from another mercury burn event, it had to have come from the principal gold shop for the pattern and concentration to be distributed as they are, and the amalgam probably falls within the range of normal weights for Andacollo. Therefore even if there is an event mismatch, this case shows that CalPUFF can achieve quite close agreement with observations in light wind with what we would assume is a fairly typical local amalgam. In event 7 (Wednesday, May 6, 2009, Figure 4.16), the tower observed concentrations that are typical of a normal mercury plume core, but in this case the concentrations are higher than the model indicates.   Nevertheless, the model predicts average concentrations given the dispersive character of the air at that moment, so it is fair to consider this event a correct prediction in terms of both timing and intensity, especially as the observations show no other signals within fifteen minutes before or after event 7.   110  Figure 4.16: Model (line) and observed (circles) mercury concentrations for event 7 (right), and the logarithmic scatter plot for this event. Blue arrows indicate wind speed and direction of five minute averages of wind, and the black arrow is the hourly average wind used in the model.  In event 8 (also on Wednesday, May 6, 2009, Figure 4.17) there is a fair concentration match, however the confidence in this assessment is low because the observed concentrations are within the range of random variability for the site. This event offers no insight into the accuracy of the model.  111  Figure 4.17: Model (line) and observed (circles) mercury concentrations for event 8 (right), and the logarithmic scatter plot for this event. Blue arrows indicate wind speed and direction of five minute averages of wind, and the black arrow is the hourly average wind used in the model.    112 4.7.4 Spacetime series  When known mercury emissions coincided with mobile mercury sampling, model and observed data during the event were plotted with two horizontal spatial dimensions and time in the vertical dimension (elevation is ignored, as in the dispersion model). Mobile observations are represented by coloured circles, and street level (2 metre height) model output is plotted as coloured contours. The tower observations are displayed as a column of colours displayed on the side with all time horizons matched between three dimensional plot and the tower plot. Table 4.13 shows timing and emissions data for these events.  Table 4.13: Timing, masses, and emissions rates of amalgam burn events that were detected at only the NE stadium tower analyzer location and by the mobile analyzer. Event Start time Finish time Mass before burn (g) Mass after burn (g) Hg emitted (g) Burn time Emission rate (gs-1) 9 1730 1742 122 45 77 15 0.128 10 1220 1245 1.2 1.0 0.2 10 0.000333 11 1310 1315 12.5 4.3 8.2 5 0.0273 11* 1305 1315 12.5 4.3 8.2 10 0.0137   Periodic contour maps of concentration plotted at regular time intervals define the street level plume through which the mobile sampling data wanders. This enables visual identification of coinciding observations from tower and roving analyzers along with the model predictions at those times and locations. Event 9 (Friday, May 8, 2009) occurred while I was conducting mobile mercury sampling and was also captured by an analyzer on the NE stadium tower. There appears to be a signal at 17:35 that is slightly upwind of the principal gold shop. It is possible that this is from yet another source of contamination; however an interesting alternative hypothesis is that it emanates from the primary shop and has moved upwind by diffusion or transport along street canyons. This might at first seem implausible, however it is important to keep in mind that the winds at street level are liable to be lower speed and more complex. Regardless of the  113 true attribution of the early signal, the concentrations are consistent with the maximum downwind plume concentrations at that distance from the source (Figure 4.18).      Figure 4.18: 3D plot (event 9) of mercury observations (top left), showing mobile observations as a wandering trace that passes back and forth through the mercury plume. Black arrows indicate five minute averages of wind speed and direction, and red indicates the hourly average wind. The red star indicates the stack position at the start of the mercury emissions. The cross marks the tower location. All data are synchronized to the vertical time scale bar shown in the middle. Observed and modeled mercury concentrations at the NE tower are shown as colours (middle) and on time/concentration axes (right). The bottom series of plots show 2D time slices of mobile analyzer concentrations and contour maps of modeled concentrations. The arrows represent wind speed and direction; red represents the hourly average wind, black represents the five minute average of wind of the current time slice.   114     Figure 4.19: Logarithmic scatterplot for event 9.   Interestingly, there is no observed signal adjacent to the gold shop at the time of emission, but immediately afterward and ~100 metres downwind the observations agree quite well with the model plume. This rough agreement continues as the analyzer wanders around in the immediate area downwind of the primary shop. At the end of the sampling run, I parked in front of the primary shop and went inside to speak to the shop operator. He had recently burned amalgam, and he had already noted the masses and timing. This event increased my confidence that he was noting the relevant information at or near the time of the event. The amalgam (weight after burning: 45 g) was sold for 553500 pesos, or 1383.75 USD at a rate of 12300 pesos (30.75 USD) per gram (all of these values are from May 2009). The signal observed by the analyzer in the vehicle while I was inside the shop is also interesting (it forms a vertical column at the end of the mobile trace between 17:46 and 17:49). It indicates the degree to which these discrete events can linger in the adjacent streets. It is unclear whether this is recirculation in the street canyon, or lagging diffusion and dispersion out of doors, windows, cracks, and the inner yard of the block of buildings that the shop is part of. I suspect it is primarily due to slow emanation of  115 pollutants from buildings, but our model does not attempt to simulate these fugitive emissions.   The correlation between observed and modeled values is not statistically significant (p = 0.431). The tower signal first appears at around 17:28, suggesting that either the burn began before the time recorded by the operator or there are plumes from other sources being detected. I believe that the former is more likely, as the burn times recorded are clearly approximate. Adjusting the timing does not improve correlations so there was no adjustment made in the analysis of this event.  Visually, the accuracy appears even higher than the statistics suggest, so it is possible that the model accuracy could be much higher if there had been denser sampling.  However, this event also suggests that the incomplete models of the urban landscape (no street canyon effects; very limited and approximate urban structure model), use of hourly average winds, and the pollutant release mechanism (only stack emissions are modeled, and emissions from other parts of the building are ignored) preclude accurate predictions at the micro-scale. It is also interesting to note that the temporal and spatial extent of the modeled plume at 2 metres above ground matches reasonably well with observations despite the low correlations of model and observed data that are exactly coincident in space and time. The observed mercury plume at 2 metres elevation is more spread out, and high concentrations persist for longer than the modeled plume. However, between 17:25 and 17:45 the concentrations at various distances from the plume are fairly consistent with model predictions for the plume at those distances, and the concentrations attenuate over similar time scales. Even though it is difficult to prove quantitatively, this event suggests that CalPUFF creates a reasonable simulation of the general behaviour of mercury plumes. Qualitatively, the signals are coincident in space and time by the coarse standards of this study, though precise mismatches lead to uncorrelated model and observed signal averages. This event therefore illustrates the difficulty of validating CalPUFF’s use in micro-scale applications; very near misses are statistically equivalent to complete misses (if one demands space and time pairing).  The tower signals appear to be even more variable than those of the vehicle survey for this event. Nevertheless, the five minute average winds clearly show that the  116 wind direction at 17:30 was pointed straight at the tower and this provides very reliable evidence that the emitted event was occurring at this time and that the high concentrations observed at the tower were most likely emitted from the principal shop.  The model output for event 9 seems visually to correlate much more with the observations than the analysis suggests, and there are many points mere metres and seconds away from matching model data. This is fine scale error that would probably average out with higher spatial sampling density and longer field campaigns.  It is significant that even though this is one of the largest amalgam burned during this study, the model predicts that concentrations fall below 0.1 µgm-3 500 m away from the source. All of the modeled plumes show this kind of rapid dilution, which strengthens the conclusion of rapid dilution inferred from observations. In event 10, which occurred on Saturday, May 9, 2009 (Figures 4.20, 4.21), there are observed signals that are contemporaneous with the model output, though the concentrations and spatial extent do not match. This is clearly a case where some other emission(s) are dispersing about town that either came from another shop or were not recorded in the principal one. It is unsurprising that this would happen on a Saturday, as miners usually spend most of the week mining and processing and then sell their gold on the weekend. Weekends are also much more likely to have emission events in private homes as well as gold shops. This is an interesting event because it shows the degree to which unrecorded emission events can possibly cause spurious modeling results, and reinforces the good fortune that event 6 occurred on the same day, but in a much more quiescent period with respect to undocumented mercury signals.   However, it is difficult to determine where the emissions originated for event 10 and to what degree the storage of high mercury concentrations in buildings and street canyon recirculation contribute to the observations. The correlation between model and observations for this event are high due to the fact that the model predicts near background concentrations and the nearest observations in space and time happen to agree in spite of high concentrations observed outside the area of the modeled plume. The results in this case are therefore inconclusive as the plume seems to have been overprinted by signals originating from other sources.     117  Figure 4.20: Spacetime plot of mercury concentrations for event 10. All time horizons are synchronized with the left hand time scale. Coloured bars are observed (left) and modeled (right) concentrations at the NE tower. Black arrows show five minute average wind speed and direction, red arrow shows hourly average wind. The red star indicates the stack position at the start of the mercury emissions. The cross marks the tower location. Periodic contour plots at three minute time intervals show ground level mercury concentrations. Mobile mercury measurements are shown as coloured points that meander through the modeled plume.   Figure 4:21: Logarithic scatterplot for event 10.   118 In event 11, as in event 10, which happened an hour earlier on the same day (Saturday, May 9, 2009, Figures 22 and 23), it seems that there are sources of contamination other than the recorded one. There is good agreement between the model and observed tower concentrations, however this could just be a coincidence, and this case is not considered a model success.    Figure 4.22: Spacetime plot of mercury concentrations for event 11. All time horizons are synchronized with the left hand time scale. Coloured bars are observed (left) and modeled (right) concentrations at the NE tower. Black arrows show five minute average wind speed and direction, red arrow shows hourly average wind. The red star indicates the stack position at the start of the mercury emissions. The cross marks the tower location. Periodic contour plots at three minute time intervals show ground level mercury concentrations. Mobile mercury measurements are shown as coloured points that meander through the modeled plume.  119   Figure 4.23: Logarithmic scatterplot for event 11.  4.8 Discussion  In general, the model performs surprisingly well in terrain and at scales for which it was not explicitly intended, though it is unable to accurately predict instantaneous micro-scale street concentrations. Of the 28 known mercury emissions events, we captured relevant observations in 24 cases. Five of these 24 emission events show significant and moderate to high correlations between high-resolution pairs of model output and observations. At each receptor location (see Figure 4.11), spatial averages of modeled concentrations are modestly correlated with the spatial average of observations. The original and adjusted correlations (Table 4.10) provide a coarse representation of model performance that one could expect if the exact mercury emissions timing were known with perfect certainty. On the upper end of performance there are certainly finer adjustments that could further improve correlations, but the gains are likely to be small. If the strict timing is used the model produces no significant correlations, but adjusted data show remarkable agreements between modeled and observed concentrations of mercury vapour at the individual event scale. Most adjustments are moderate, plausible, and  120 justifiable given the particular details of each event and adjustment. Some events correlate especially well, as in event 6. This event occurred during calm winds, which is a situation that should be much easier to model (it consists mostly of diffusion, buoyancy and dilution, with little advection or shear turbulence). Even if the adjustment in this case matched the model output with a completely different plume for which no data were recorded, the calm wind, high concentration and sustained nature of the signal strongly indicate the origin to be the principal gold shop. In this case it could still be said that the model accurately describes mercury emission events from that gold shop, even if they are not precisely the same event. During mobile sampling we were always impressed by how fleeting the emissions were, and how remarkably limited in space and time. One could observe signals greater than 10 µgm-3 and then circle around the block and at the same location find no signal at all minutes later. The concentrations observed while vehicle sampling attenuate in space and time rapidly enough to suggest that observed concentrations over 1 µgm-3 must have originated within 200 metres of the observation location (as in Chapter 3). However, given the limitations of our sampling methods, there is no way to know for certain what the origin of any given signal is simply by observing it on any given mobile sampling run, and so the skeptical observer is left wondering whether a given signal could have wafted in from much greater distances. CalPUFF suggests that this is not the case, and that the qualitative observation that high concentrations must indicate close proximity to the source of the emissions is in fact consistent with simulated mercury plumes.  Three-dimensional plots offer a simultaneous view of the shape of the modeled mercury plume with respect to wind conditions, the variability of wind, and the degree to which those two factors could cause a plume with very high pollutant concentration to completely fail to register on a fixed tower location that was only a few degrees off the mean wind direction. The latter is one of the main difficulties of working at this scale with extreme events. The statistical behaviour of the model could be very different from the observed individual mercury plume, and models are designed to provide long term averages where extreme concentrations and plume position errors average out over time, and monitoring locations are fixed (ideally at some height on a tower). The event  121 averages show that sparse extreme values in this case produce average model concentrations that are roughly consistent with the observations.  Statistically, as well as visually, CalPUFF predicts the intensity and distribution of mercury contamination fairly accurately, even if instantaneous predictions often fail. Therefore CalPUFF could be useful for predicting human health hazard if one had reliable estimates of emission amounts, frequencies and locations of all gold shops in an area. However, it would only be a hazard prediction proxy and it would need calibration. In Andacollo, emissions frequencies are so low that the spatial average model output concentrations are below 1 µgm-3 at almost all receptor locations over the entire three week time period during which the 28 known mercury releases occurred. Therefore, if the goal is to evaluate human health hazard, it would be better to use the frequency of exceedence of health limits rather than averages when events are rare. The average observations (Figure 4.11) clearly mark the location of the principal gold shop. The transfer of mercury from the plume to the street canyon might not be as efficient as in the model, and much of the contamination may skim over the streets at roof height or be swept along them more rapidly proportional to the along-street component of wind. Another likely cause of the poor correlation, other than low sampling density, is that the model averages include only known emission events, but mercury vapour observations are likely to record emission events that were not reported to us. Also, whereas model output is a continuous series, the observations are patchy and only occasionally are the same points sampled multiple times during an event or in any given half-hour period. Together, all of these factors contribute to the difficulty in obtaining more numerous strong correlations between model and observations.  Important questions remain: how much risk are the residents of this neighbourhood exposed to? How can we translate the averages and exceedence frequencies into estimates of risk for residents and workers in the area? These are epidemiological questions that require information about residence/employment history and health outcomes over time in order to calibrate the model output intensity to the public health risk intensity. The map of spatial averages of all mercury concentrations observed at each location (Figure 4.11) provides additional insight into the average exposure of local  122 residents, and could therefore be used to help calibrate the modeled risk maps. Higher sampling density and frequency of emissions could produce a much more reliable and robust evaluation of CalPUFF as an urban mercury dispersion model and better estimate public health hazard, however this was not possible given the available resources.  4.9 Summary  Despite considerable challenges in terms of data collection, estimation of model parameters, instrument installation, among many others, I have shown that CalPUFF simulates real micro-scale mercury release events with reasonable accuracy in Andacollo. Some of the modeled mercury release event periods were adjusted to match the occurrence of the events in time, however this was done within plausible limits and with good justification. Therefore I believe that the results of the adjusted model inputs (which effectively reduce the time precision of the model predictions or in one case compensates for spurious wind data) better represent the accuracy one could reasonably expect under similar circumstances when exact timing is known. Strong correlations between modeled and observed mercury concentrations can be achieved for individual mercury release events, especially under favorable meteorological conditions such as calm winds. CalPUFF accurately predicts mercury concentrations at the tower in most events where there is little noise from other sources and the emission rates are high. Mobile mercury observations correlate poorly with model results partly due to sparse data but also because of the oversimplified urban structure model and perhaps also the way in which urban dispersion is parameterized.  The data collected here are in quantity only barely sufficient to make an assessment of CalPUFF’s validity at this scale and in this context, however this is the best and most significant study of its kind by virtue of being the only study of its kind. Regardless of its shortcomings, it is an important first step in understanding fine scale dispersion of mercury and the methods developed here, along with the challenges identified in this work, will greatly facilitate the planning and execution of similar efforts in the future.  Given that the model can be fairly accurate, the 3D images of model output and observations vividly illustrate the probable behaviour of mercury releases from urban  123 gold shops. The model confirms the observations that these mercury plumes are ephemeral events that last only slightly longer than it takes to burn the amalgam itself and then quickly disperse. We can now confidently say that mercury emitted by gold shops in Andacollo reaches extreme concentrations for very short periods and then dissipates rapidly, and that any observation of greater than 1 µgm-3 is a reliable indication that there is an amalgam being burned within 200 metres and five minutes of the observation location and time. CalPUFF is unable to reliably predict instantaneous street concentrations, however maps produced by CalPUFF are still informative with respect to hazard, if the hazard levels can be calibrated. The following chapter shows that mobile mercury mapping can even be used to assess the impact of mercury reduction interventions.    124 5 Mitigation of mercury contamination 5.1 Synopsis  This chapter demonstrates the effectiveness of UNIDO’s mercury mitigation project in Antioquia by using the mercury mapping methods described earlier and an analysis of mercury use. It sets Colombia in the global context, showing where mercury originates, how it is distributed, used, recycled, and lost. Colombia is in the top ranks of mercury emitters (Telmer and Veiga 2009), but the United Nations have begun to mitigate this critical situation. Based on mercury mass balance in 15 entables (Cordy et al. 2011), 50% of the mercury added to small ball mills (“cocos”) is lost: 46% with tailings and 4% when amalgam is burned. In just 5 cities of Antioquia (with a total of 150,000 inhabitants: Segovia, Remedios, Zaragoza, El Bagre, and Nechí, (Figure 2.11), there are 323 entables producing 10-20 tonnes of gold annually. Considering the average levels of mercury losses estimated by mass balance and interviews of entable owners, the mercury lost in these five municipalities must be around 93 tonnes annually. Five second average urban atmospheric mercury levels range from 0.02 µgm-3 to 1000 µgm-3 (inside gold shops) with 10 µgm-3 being common in residential areas (the WHO limit for annual average public exposure is 1 µgm-3 (WHO 2003)). The total mass of mercury released into the environment could be as high as 150 tonnes annually, making Colombia the world’s highest per capita mercury polluter. Entables must be removed from urban centers and technical assistance is badly needed to improve their technology and reduce emissions. The five municipalities in this study produce the majority of gold in Antioquia, and this is the first formal description of gold production methods and pathways for mercury release to the environment in this area. Atmospheric mercury concentrations in the urban core of Segovia, Colombia have decreased from 2010 and 2012, despite a 30% increase in gold production. Table 5.1 shows the changes in mercury concentrations averaged over each year’s mobile mercury measurements. Analysis of the data presented in this chapter show increases in average mercury concentrations in the neighbouring towns of Remedios and La Cruzada, and almost 50% reduction in Segovia. These opposing trends reflect the fact that most new  125 processing capacity was built in the hills outside of Segovia, and gold shops in Remedios and La Cruzada are closer to this new capacity than downtown Segovia. Therefore much of the reduction observed in Segovia is displaced emissions, as sources have been relocated to adjacent rural areas. Differences in average mercury concentration over the entire Segovia urban area were verified by comparison with separate averages of extreme values. Though results in Segovia are encouraging, there is considerable uncertainty given the short and intermittent measurement campaigns. Unknown temporal variability of gold production and amalgam burning are also significant sources of uncertainty. There is much greater uncertainty in assessments of changes in mercury contamination in La Cruzada and Remedios because mercury measurements are fewer and cover a much smaller area. However, given that both subsequent campaigns show a similar decrease in mercury concentrations over similar time periods in Segovia (as well as increases in the other two towns), we can have much greater confidence that these decreases are not artifacts of random processes. My results show that UNIDO’s efforts have had significant impact in terms of reducing mercury concentrations in Segovia Improvements in Segovia are likely attributable to the use of retorts in the processing centres, improvements in gold shop filters, and regulations by the Office of the Secretary of Mines of Antioquia which ban new mercury sources in the urban core.  Retorts recover mercury for re-use, thereby vastly decreasing the amount of mercury evaporated in open burning. Further improvement efforts should focus on installing high temperature glass doors on gold shop fume hoods (thus allowing operators to close the door without miners worrying that they are being cheated), and decreasing the milling speeds to increase grinding efficiency and reduce evaporation due to frictional heat generation. Long term goals should include extending the ban on burning amalgam in the urban core to existing emitters and relocating processing centres away from residential and commercial areas. Even if all contemporary sources of mercury were eliminated, concentrations would persist for some time as a result of evaporation from contaminated soils and surfaces. The research in this chapter has in part been published in Cordy et al. 2011 and 2013.    126 Table 5.1: Changes in mercury concentrations averaged over each year’s field campaign in the urban cores of Antioquia. Temporal averages of atmospheric mercury concentration (µgm-3) 2010 2012 % change Segovia 1.25 0.660 -47 Remedios 0.244 0.436 +78 La Cruzada 0.557 1.28 +131      Figure 5.1: Cocos are small ball mills that grind and amalgamate the whole ore. © Paul Cordy photo.  5.2 Methods  The data presented here were collected through interviews with miners and local authorities, and confirmed with direct observations. Data about imports and sales of mercury in Colombia, and specifically in Antioquia, were obtained from the Colombian Ministry of Commerce (Bogotá), the UN COMTRADE (United Nations Commodity Trade Statistics Database) and direct interviews with mercury vendors and users. Interviews with entable owners were conducted over a two-year period in Segovia, Remedios, and La Cruzada (the largest mercury users in the region) in order to obtain self-report data on the amount of mercury purchased per month by each entable. These entables provide mercury to the miners, and occasionally use mercury to process their own ores if they have their own mine. Thus, when attempting to arrive at the value  127 of mercury purchased (which is a direct indicator of mercury consumption) per month, one can expect the values to be higher than reported due to the fact that some miners buy mercury from a separate distributor and bring it to the entables. Another source of error is interviewer bias, as people are naturally inclined to underreport their use of mercury (which they know to be a contaminant), whether consciously or unconsciously.  In order to check the self-reported data, a mass balance of mercury losses was conducted in 15 entables. Mercury was weighed before and after amalgamation using a scale with a detection limit of 0.01 mg. The losses by evaporation were measured by the weight of the amalgam before and after burning, as in the previous section. These data underestimate the true value as raw gold commonly retains 2 to 5% of residual mercury after amalgam burning (Veiga and Baker, 2004), to be released when the gold is later melted into ingots. Based on these self-reported and mass balance data, it is possible to estimate the mercury consumption rates for the remaining entables that were not interviewed in these and other municipalities. Measurement of mercury concentrations in air were described fully in section 3.2. They consist of coincident GPS and Lumex observations made from a moving vehicle. Average mercury concentrations measured within a 10 metre radius of each point along the surveyed roads were computed for display, though only non-overlapping 20 metre separated averages were used in statistical analysis. The mean value of these point averages are shown in colour on maps of Segovia’s roads. The median one second concentration sample sizes for each 20 metres of road in Segovia were 97.  5.2.1 Evaluating mercury mitigation measures  Since 2009, UNIDO has worked to reduce mercury contamination throughout the department of Antioquia, Colombia, where Segovia is located. They helped introduce municipal regulations that require retorts or gold shop filters of some kind be used when burning amalgam, and that prevent new gold shops and processing centres from opening. This intervention was accompanied by an energetic outreach and monitoring campaign, in which UNIDO representatives continually circulated among gold shops and processing centres demonstrating cleaner technologies that recover more gold, advocating for the use  128 of retorts, and collecting data on gold production and mercury consumption. The average monthly tonnage of material processed per ball mill was derived from years of direct observation by UNIDO staff. It is assumed that each ball mill is in use most of the time, which is consistent with observations of miners’ work habits and processing centre activity over three years. Changes in processing infrastructure and services were quantified in a mining census that UNIDO undertook as part of the aforementioned activities during July and August of 2010 and 2011, and these data are used as a proxy for real production. It is difficult to assess increases in production using tax revenue and self-reported data that are available because producers and shops will tend to underestimate their wealth and mercury emissions, as the latter could prompt punitive action by the government. Nevertheless, documented increases in the number of processing centres and in the number of amalgamation ball mills in each centre are a response to increased demand for processing capacity, and therefore mercury use. Spatial measurements of mercury concentrations (as described in Chapter 3) in Colombia were taken over three field campaigns (April 10 to 21, 2010, September 21 to October 7, 2011, and November 10 to 20, 2012) over a three-year period, to determine whether UNIDO’s remediation efforts were effective. The Kolmogorov-Smirnov test (Massey 1951) was used to see if the two datasets were significantly different, and if so, which showed greater contamination. This non-parametric test accommodates different dataset sizes and lognormal distributions, and uses the largest difference between the two empirical cumulative distribution functions to reject the null hypothesis that both samples are from the same distribution. Seasonal variation or behaviour changes on the part of the miners are assumed to be insignificant in this analysis. Several observations support this assumption: 1) Monthly production varies little as there is no seasonality to underground mining in tropical hills. 2) The weather was similar in all field campaigns as they were each done on the shoulder of the rainy season (comparison of historic weather for the nearby city of Medellin during each campaign show small maximum differences in average temperature, humidity, wind speed, and the percentage of days with rain among the campaigns: 1.8 oC, 2.4%, 1.0 ms-1, and 4.6% respectively (Wunderground 2013)). Early in each campaign, I gave interviews  129 on local radio stations to promote UNIDO activities and increase public understanding of mercury detection, risk mitigation, and health effects. These radio sessions were intended to ensure broad awareness of the upcoming mercury studies. During the field campaign, rumors of a foreigner with a mercury analyzer probably circulated quickly, and theoretically could have motivated changes in the behaviour of miners and gold shops that would have altered the results of the mercury measurements. By the 2011 campaign, many miners knew of, or had seen, the mercury maps made in 2010. However, it is unlikely that independent miners would collectively change their behaviour and travel far to sell their gold or withhold it for two weeks in order to skew the data of an air pollution study. Firstly, miners often lack savings and therefore depend on their weekly production income to provide for their families, constraining their ability or incentive to alter gold output, even temporarily. Most miners sell gold as soon as they obtain it, and shops probably would not turn them away for an entire week. Furthermore, all entable mills were running constantly during the entire study (as usual), so there appeared to be no decrease in mineral processing. There were no announcements made regarding the 2012 measurement campaign either by way of meetings or radio. Finally, the people of Segovia generally seem to have more faith in the near impossibility of enforcing regulations than fear of their consequences. As a result of these income restraints and personal beliefs, it is assumed that changes in mercury emitter behaviours were not significant contributors to the observed reduction in mercury concentrations. Given that this assumption is impossible to test and there was no control municipality, a large and statistically significant reduction is needed to have confidence in the effectiveness of mitigation measures. Finally, since extreme values can skew averages, both low and high values were averaged separately to see how the relative change in average mercury concentrations differed. The WHO (2003) annual average health hazard limit for mercury in air (1 µgm-3) was used as the boundary between low and high, and it was assumed to be equivalent to shorter term averages as explained in Chapter 3. Road closures and other random circumstances forced changes in mercury detection circuits, leading to slight differences in mercury map coverages, therefore only road segments common to both datasets were used in statistical analysis although all are displayed in the maps.  It is important to acknowledge the openness and participation of the gold trade  130 community in the five municipalities studied, and especially in Segovia. Meetings I had with them were always well attended and jovial, and cleaner technologies like retorts are being installed and used as requested by UNIDO. Furthermore they are welcoming to frequent external mercury use observation and reporting by UNIDO staff. This cooperation is a vital element in tracking the effectiveness of remediation measures and resulting declines in mercury vapour emissions.  5.3 Results 5.3.1 Mercury imports and sales  Mercury can be legally imported to Colombia, but local miners in Antioquia have also reportedly been buying mercury from illegal suppliers who bring it from Peru. According to the data obtained from the Colombian Ministry of Commerce and confirmed by the UN COMTRADE, 130.96 tonnes of metallic mercury were imported by Colombia in 2009, a significant increase from 57 tonnes in 2001. The countries exporting mercury into Colombia are outlined in Table 5.2, with the majority of exports originating from the Netherlands and Germany.  Table 5.2: Countries exporting mercury in 2009 to Colombia (UN COMTRADE, 2009). Country tonnes of Hg/a Netherlands 47.95 Germany 35.58 Mexico 24.76 Spain 14.49 United Kingdom 4.14 United States 3.44 Italy 0.6 Total 130.96  Table 5.3 shows the staggering number of mineral processing centres in the region and their emissions rates based on interviews with the operators. Approximately 1.37 tonnes of mercury is purchased monthly by all 69 Segovia entables investigated in this study. Thus on average each entable purchases approximately 19.86 kg of mercury per month and releases it to the environment. By using this average, the total mercury purchased  131 across all 94 entables in Segovia would be approximately 22.40 tonnes annually. This estimate is 4.38 tonnes higher than the sum of estimates of local mercury vendors. The difference could be attributable to self-reporting errors or omissions, to mercury brought from outside the entable by individual workers, and to the consumption of the nearby municipality of Remedios where mercury from Segovia is also sold. There are five entables that consume more than 100 kg Hg per month or 73% of all mercury used in gold mining in Segovia.  Table 5.3: Number of entables in 5 municipalities in Antioquia (Data from the UNIDO Colombia Mercury Project and Cordy et al. 2011) Municipality Number of entables Remedios/La Cruzada 24 Segovia 94 Zaragoza 47 El Bagre 123 Nechi 35 Total 323  If we assume that the average entable consumption of 24.10 kg per month for all 323 active entables then the total amount of mercury released is around 93.4 tonnes annually. This estimate assumes that the entables do not stockpile any mercury, i.e. all mercury purchased monthly is lost to the environment as vapour (when amalgam are burned in the gold shops), in water effluent, or in the tailings.  5.3.2 Mercury losses  Assessment of 15 entables revealed that, on average, 60 kg (range: 47 to 73) of ore is added to each coco together with 80 g (range: 50 to 110 g) of mercury. The alkalinity of the pulp is adjusted to pH 11 with quick lime (Ca(OH)2) and milled for about four hours. During the grinding process, a portion of mercury is pulverized, oxidized and loses coalescence. The material is then discharged, washed and the amalgam is recovered by panning. Excess mercury is squeezed off in a piece of fabric and the resulting amalgam is  132 burned in a fume hood, commonly without any filter or condenser. If the ore is very rich in gold, the tailings are collected for a second amalgamation in mills for one or two hours but the pH is reduced to 5 with lime juice. Molasses and sodium bicarbonate are also added to the mixture together with mercury recycled from the first amalgamation. It seems that the change in pH provides some coalescence to mercury droplets, however mercury loss is visible when the final tailing is panned. This whole procedure varies depending on the entable and type of ore being processed (ore grade, sulphide content, gold grain size, etc.). Some people add guava leaves, toothpaste, detergents, and other ingredients they believe avoid mercury losses or improve amalgamation. The amalgamation tailings, contaminated with mercury, are then collected in a concrete pool, where it settles, decants, and awaits leaching with cyanide by stagnant percolation.   The mercury mass balance reveals that around 50% of excess mercury is recovered by squeezing the amalgam through a cloth, indicating that the miners put too much mercury into the cocos. On average, 46.3% of mercury is pulverized during grinding and lost with tailings. The amount of mercury lost when amalgam are burned without retorts is 3.76% of the initial mercury added into the cocos. It was extremely difficult to know who uses retorts, but it is assumed for the purposes of this study that evaporated mercury was not recovered. This is reasonable, as the data presented here were collected in the early stages of the UNIDO campaign and most local miners brought raw amalgam to the gold shops, which had no mercury recovery devices, or devices that were insufficient at best. An unknown quantity of mercury is lost as vapour during milling due to heat produced by excessive milling speeds.  An alternative calculation of the mercury consumption in the region can be obtained based on the mass balance data from Cordy et al. (2011). There are 2600 cocos in the five municipalities studied and each coco receives on average 78 g of mercury of which it loses about 50%. Considering that each coco runs twice, and in some cases, three times per day, 360 days per year, the unrecovered mercury losses might be between 73 and 110 tonnes annually in the five municipalities studied. The average Hglost: Auproduced ratio obtained for Segovia was around 15 which is compatible with operations in other countries using whole ore amalgamation (Veiga and Baker 2004; Castilhos et al, 2006). This ratio has been used as a parameter to quantify  133 mercury releases from artisanal gold mining operations. Although the ratio of Hglost:Auproduced is an imprecise measure of an individual entable, due to fluctuations in gold production that can result in overestimation, it is adequate as an average of many operations (Veiga and Baker, 2004).   5.3.3 Mercury emissions from milling  A small, informal study of milling speeds and mercury concentrations suggests that by decreasing milling speed from the usual 54 to 58 rpm to 40 rpm, mercury concentrations emitted when the coco is opened decreased by around 60% (Table 5.4) (Oseas Garcia, personal communication 2012). The concentrations listed are five second averages taken using the Jerome analyzer located within 30 cm of the mill door at the moment it was opened.  Table 5.4: Comparison of airborne mercury concentrations resulting from milling at a slower speed in different entables (µgm-3). Fast refers to speeds of 54 to 58 rpm and slow refers to speeds near 40 rpm. Entable  [Hg] fast mill [Hg] slow mill El Galeon 176 86 Los Toritos  36 El Vecino  37 Rafael Gaviria 262 144 Colon 985 573 El Tejar 146 95 Relampago 54 23 El Cogote 137 42  Amalgamating during milling emits significant amounts of mercury vapour at any speed, but lower milling speeds mitigate this while improve milling efficiency at the same time. 40 rpm is just below the critical speed of mills with the diameter of the average coco (0.5 m). Milling below the critical speed allows better tumbling of the steel balls, whereas above the critical speed the balls are pinned to the mill walls by centripetal forces (Djordjevic 2004).   134 5.3.4 Cyanidation of mercury contaminated tailings  On average 46.3% (and in some cases as much as 82%) of the mercury introduced in the system is lost with tailings. Tailings with up to 5,000 mg/kg of Hg were analyzed in one entable in Segovia. The tailings, with residual gold, are then subjected to cyanidation in vat leaching with subsequent gold precipitation with zinc shavings. This usually happens in the entables where owners excavate tanks in the ground and line them impermeable plastic. Permeable cotton rags are placed under the tailings to keep particulates out of the cyanide solution. The filtered solution passes through a wooden box or PVC pipes filled with zinc shavings where gold and residual mercury are precipitated. The operators in Antioquia do not realize that re-dissolution of gold happens when zinc precipitation is not conducted in a vacuum, and usually a part of the gold is lost in the effluent (Velasquez 2010). After precipitating the gold on the zinc shavings, the solution is pumped back into the percolation tank. This leaching cycle occurs from eight to thirty days, depending on the grade of the tailings. This is far too long and indicates a lack of aeration in the vat. The amount of mercury retained by the zinc is not known, but in similar operations in Ecuador it was observed that about 28% of the mercury in the tailings introduced in the cyanidation tanks is precipitated on the zinc shavings (Velasquez et al, 2011). The rest is therefore sitting in the cyanide contaminated tailings or the barren solution, which is discharged without further treatment. Both situations pose an intolerable risk, given that the mobility, bioavailability, and toxicity of mercury are amplified by ionization and formation of organic complexes (Jensen and Jernelov, 1969). At the end of the leaching process, when miners visually notice that there is no more gold being precipitated on the zinc shavings (fresh shavings do not turn black), they discharge the pulp with gold-barren cyanide solution, still rich in zinc and mercury, into local creeks. The zinc shavings, rich in gold, are burned in open gas furnaces or wood fires, which contaminates a wide area with zinc vapour. Mercury, lead and other heavy metals in the tailings, partially leached by cyanide and precipitated by the zinc, are also released into the urban air when the zinc is evaporated. After burning the zinc off, gold is melted and sold to the gold shops.  Mercury dissolves in cyanide slower than gold. Mercury in cyanide forms soluble complexes such as [Hg(CN)4]2-, which is stable at pH > 8.5, and Hg(CN)2 (aq), which is  135 stable at pH < 7.8 (Flynn and Haslem, 1995). Velasquez et al. (2011) studied dissolution of mercury in an agitated gold leaching tank and found that while more than 92% of the gold is dissolved in five days, only 27% of mercury becomes soluble with cyanide. We expect that similar processes in Antioquia would yield similar results, suggesting that tailings dumped into the rivers probably still contain mercuric-cyanide complexes in solution as well as metallic mercury droplets.  Near many sites where mercury contaminated tailings are leached with cyanide, fish contain high levels of mercury (McDaniels et al, 2010). Rodrigues et al (2004) analyzed 31 samples of carnivorous fish from a small lagoon at the Brazilian Amazon that was receiving effluents from a heap-leaching cyanidation operation. The average total mercury concentration of the samples was 4 ± 5 mg/kg and a small fish (15 cm) sample showed levels of 21.9 mg Hg/kg, probably a new world record (Sousa and Veiga, 2008). In Antioquia, the creeks adjacent to the entables appear to have no aquatic life, though they might be too silted with tailings to see if there was any. Where there is life further downstream, we expect to find alarming mercury contamination in fish.  5.3.5 Interventions  From 2009 until 2012, UNIDO, CorAntioquia, and Government of Antioquia representatives continually circulated among gold shops and processing centres, demonstrating cleaner technologies that recover more gold and advocating for the use of retorts (Figure 5.2 and 5.3). They also informed people of the concentrations of mercury vapour found in their operations and the implications for their health. This included showing the mercury concentration maps from the 2010 mobile measurement campaign to miners, non-miners, government officials, as well as entable and gold shop operators. They also worked hard to convince people to eliminate the practice of whole ore amalgamation. UNIDO demonstrated that by optimizing size to which the ore is ground and then using centrifuges, sluices, and shaker tables to concentrate the gold, miners could greatly reduce the amount of mass to which mercury or cyanide is applied. This would, in turn, greatly reduce the amount of mercury that ends up in the tailings or vaporized during milling or zinc roasting.   136      Figure 5.2: Zinc roasting, the old way (top left), and with retorts. Pouring or pumping water over the retort speeds cooling and condensing. © Paul Cordy photo.  Figure 5.3: Mercury recovered by the zinc retorts. © Paul Cordy photo. Throughout the campaign, emphasis was placed on the monetary savings of reducing mercury consumption and superior gold recovery that could be obtained using cleaner techniques. Their philosophy was to primarily focus on the incentives of their interventions, and their only punitive method was to encourage local governments to ban  137 all new mercury emitters in the urban core and require all operations to use retorts. This regulation includes those who burn zinc, and they developed a zinc roasting retort designed specifically for this use (Figure 5.2). UNIDO also demonstrated that slower milling speeds reduce mercury vaporization while increasing gold recovery, and promoted this throughout the region.  A few entables were also convinced to largely eliminate mercury amalgamation in favor of cyanidation in agitated tanks (which are far superior in terms of speed and efficiency than the still percolation ditches they normally use (Velasquez et al. 2011). This change is highly significant in light of the social challenges that must be overcome to change the entable system.  Entable-style processing centres are a widespread phenomenon in South America, as they arise naturally where there are high concentrations of miners. One miner gets organized and wealthy enough to acquire land and equipment, and builds a plant to process his minerals and rent it out when he is down in the mine. The miners renting his plant leave their amalgamation tailings behind once they have recovered the 40% or less of gold that amalgamation can separate. Once enough tailings build up, the plant owner begins to use cyanidation. Very soon the plant owner makes enough money that just operating a plant is far more lucrative, safer and easier than actually mining. No one feels cheated because the miners start out knowing each other and renting the plant time from their friends and people they trust. Also, many miners are not aware of how low their gold recovery is, nor that most of the gold in their ore can only be obtained with cyanide. Yet if one were to suddenly switch to cyanidation only, the miners would not have the technical skills to operate a cyanidation circuit without training. If then trained and knowledgeable about the true content of their ore, they may not want to hand over 60% of their profits as a rental fee. Plant operators have a significant stake in keeping the status quo.  One solution to this dilemma would be to keep a small number of amalgamation mills that miners can use as an artisanal assay. Miners would take only a small portion of their ore and process it in the normal way that they understand and trust: whole ore amalgamation. Based on their recovery by this method, they could negotiate a value for the rest of the ore, and the plant operator would then buy the un-processed ore from the  138 miner. This way, not only would the miner be spared the thought of a 60% loss of potential profit, but also they would no longer have to spend hours processing their material. Everyone would likely perceive it as a win-win situation even though the miner is still being swindled out of most of his gold. UNIDO staff have found a group of miners who, having learned of potentially significant improvements in recovery, have pooled their resources to open a plant of their own based on the amalgamation assay and bulk cyanidation model (Oseas Garcia, personal communication 2012).  5.3.6 Intervention impact assessment  UNIDO’s mercury mitigation initiatives in Segovia fought an uphill battle against growing profitability of gold during the study period. High gold prices have spurred production increases that could easily overwhelm mercury mitigation successes as far as total mercury released. Despite recent regulations banning new mercury burning businesses within the urban core, Segovia saw a 30% increase in the number of gold shops operating in the city centre, though half of these were licensed shops that had not been operating when the regulations became law. Only two new processing centres opened between 2010 and 2012, but combined with increased numbers of mills that were installed in licensed centres, the total number of amalgamation mills (“cocos”) in Segovia increased by at least 30% over that time (Table 5.5).   Table 5.5: Urban Segovia production capacity change as a proxy of gold production.   2010 2012 % increase Gold shops 40 53 32 Processing centres (entables) 94 96 2 Amalgamation mills (cocos) 1968 2590 32  In the rural periphery of Segovia, milling capacity increased by more than 600% between 2010 and 2012, probably in response to restrictions in the urban zone. Most of the amalgam produced is burned, sold, and melted in the urban core of Segovia and adjacent towns, though gold processed in the rural periphery of these towns is more likely  139 to be sold in Remedios and La Cruzada. UNIDO estimates that combined urban and rural gold production has increased by 30%, and they could reasonably have expected mercury contamination to increase by a similar amount if there had been no intervention. Conversely, Table 5.6 and 5.7 show that average mercury concentrations in Segovia have decreased by approximately 20 to 50%.  Table 5.6: Summary of changes in average mercury concentration [Hg] in the urban cores of Segovia, La Cruzada, and Remedios, Colombia. Mercury concentrations are averaged over the entire mobile sampling dataset of each year’s field campaign.  Table 5.7: Changes in average mercury [Hg] concentration in the Segovia urban core; in bulk and separated by hazard level. Average values are taken over the entire field campaign in each year (‘total observation time’).  2010 2011 2010-11 % change  2012 2010-2012 % change % of observations [Hg] >1* µgm-3 24% 18% -6   11.8 -12 Total observation Time (hh:mm) 26:18 56:10 +113  21:18 -18 Time averages (µgm-3)   [Hg] 1.25 0.977 -22 0.660 -47 [Hg] >1 4.32 4.24 -2 3.80 -12 [Hg] <1 0.286 0.265 -7 0.238 -17 *An annual average of 1 µgm-3 is the World Health Organization health hazard limit.  The two-sample Kolmogorov-Smirnov test was used to verify the significance of the difference among the mercury concentrations found in the three different field campaigns. When comparing each year with each other, the Kolmogorov-Smirnov test rejected the null hypothesis that the data are from the same distribution (Figure 5.4). The [Hg] Time averages (µgm-3) 2010 2011 2010-11 % change  2012 2010-2012 % change Segovia 1.24 0.977 -22 0.66 -47 Remedios 0.245 0.323 +32 0.436 +78 La Cruzada 0.557 0.818 +47 1.284 +131  140 Kolmogorov-Smirnov test statistic is the maximum separation of the cumulative distribution functions of the datasets, and this figure is largest when comparing 2010 and 2012 in Segovia, than for either other year pairs. The alternative hypothesis therefore supports the conclusion that mercury concentrations have decreased over the months between each field campaign, and the declining successive averages corroborate this.  Table 5.8: Results of Kolmogorov-Smirnov tests for successive years of mercury vapour data in the Segovia urban core. Years compared P value Test statistic 2010-2011 <0.01 0.0983 2010-2012 <0.01 0.1661 2011-2012 <0.01 0.0257   Figure 5.4: Empirical cumulative distributions of Segovia urban core one second mercury concentrations, on which the Kolmogorov-Smirnov test is based. Distributions further to the right have average values that are larger than those farther left. The 2010 distribution clearly contains higher concentrations of mercury.     142 In Figure 5.4 and 5.5, significant differences in the distribution and severity of mercury contamination are clearly visible, as are chronic problem areas that are still extremely polluted. The contrast between the declines in concentrations in Segovia and the increases in nearby Remedios and La Cruzada (Figure 5.6, 5.7, 5.8 and 5.9) is almost certainly an effect of new regulations banning new mercury sources in the Segovia urban core. With an estimated 600% increase in hinterland mineral processing capacity, more miners are outside of Segovia when they obtain their amalgam and are therefore more likely to sell to a nearer gold shop and stay out of the chaos, dust, poor roads, traffic, and hassle of downtown Segovia.   Figure 5.6: Mercury concentrations in Remedios during the field campaigns of years shown. Each colour represents the average of all concentrations measured within 10 metres of that point. Not all mercury sources are shown.   143  In a way UNIDO has not entirely eliminated the emissions; some has been displaced to the rural periphery. Certainly this is an important part of the observed changes in contamination, but the increases in two tiny satellite towns may not significantly offset the much larger declines in the sheer number of sources and frequency of emissions in Segovia. Questions remain as to the strength of emissions in the hinterland.  Table 5.9: Results of Kolmogorov-Smirnov tests for successive years of mercury vapour data in Remedios. Years compared P value Test statistic 2010-2011 <0.01 0.3495 2010-2012 <0.01 0.2678 2011-2012 <0.01 0.0943     Figure 5.7: Empirical cumulative distributions of Remedios mercury vapour concentrations, on which the Kolmogorov-Smirnov test is based. The 2010 distribution is farthest left, indicating that there were on average lower mercury concentrations in that year.     144   Figure 5.8: Mercury concentrations in La Cruzada during field campaigns of years shown. Each colour represents the average of all concentrations measured within 10 metres of that point. Not all mercury sources are shown.   Figure 5.9: Empirical cumulative distributions of La Cruzada mercury vapour concentrations, on which the Kolmogorov-Smirnov test is based. The 2010 distribution is farthest left, indicating that there were on average lower mercury concentrations in that year.  145 Table 5.10: Results of Kolmogorov-Smirnov tests for successive years of mercury vapour data in La Cruzada. Years compared P value Test statistic 2010-2011 <0.01 0.1611 2010-2012 <0.01 0.1327 2011-2012 <0.01 0.0954  5.4 Discussion  Inhalation of Hg vapour is the primary exposure pathway for miners, gold shop workers and people living near areas where mercury processes are used in the five municipalities studied in Antioquia, Colombia. Normally, authorities and researchers are more concerned about the ingestion of fish contaminated with methylmercury, because methylmercury is more toxic than metallic mercury and mercury vapour pollution is perceived to be exclusively an occupational hazard. Information about mercury contamination of neighbours living near processing centers or gold shops has been limited to a few anecdotal reports, and people who live where amalgamation has been done for hundreds of years are a difficult audience to convince of mercury related dangers. In spite of the contamination, rural communities whose economy depends almost entirely on artisanal gold mining have no interest in banning it. Closing the entables would have far more immediate consequences than environmental and chronic health effects of mining. For these reasons, UNIDO has focused on increasing production with cleaner and mercury free techniques in order to earn the confidence of miners. Public pressure and advocacy related to environmental injustice were not perceived to have a likely motivational influence toward behaviour change. Adults seem to be reasonably tolerant of mercury vapour, and this leads many to doubt the severity of the contamination and risk. However, both Segovia and Remedios’ mayors remarked to us that there is an unusually high incidence of kidney failure in their towns. Children and pregnant women are especially vulnerable and municipal officials noted that there is a high incidence of cerebral palsy in Segovia. School teachers notice that Segovian students have greater learning difficulties and poorer memory than students in other towns. The lowest observable adverse effect levels for most contaminants are known to be much lower among children than adults; in fact, many new studies have  146 found that for some heavy metals, there is no safe level of exposure, as even very low amounts can produce unexpectedly significant cognitive impairments (Lanphear et al. 2005). The anecdotal health claims in Segovia warrant further investigation, but given the high levels of mercury reported in this investigation, adherence to the precautionary principle would require mercury burning activities be banned or relocated. Even after this happens it is likely that shop walls and other surfaces will emit their mercury burden for months to years after relocating mercury sources.   These interventions will result in a costly process that will involve the whole community and take many years. It is a much more difficult social enterprise than a technical one. I was struck by something one of the miners said on my most recent field venture in Segovia. He told me that at first there was resentment and fear in the mining community: greater attention to illegality and environmental contamination is never welcome, even if individuals feel safe knowing that everyone is polluting the same way and they can’t all be stopped at once. But there was more to it than that. Segovians are proud of their mining heritage and don’t like people telling them that their methods are inefficient and contaminating… their mining practices are part of their culture and they are proud of their culture as it has centuries old mining roots. But now he says that since he’s making more money using cleaner techniques he doesn’t care about culture as far as mining is concerned. He didn’t mention health, enforcement, or the environment. On January 19th, 2013 over 140 nations signed the Minamata Convention on Mercury, a legally binding treaty that aims to eliminate mercury use and trade globally. This treaty singles out artisanal mining as a high priority problem as it is now arguably the largest source of mercury emissions globally (UNEP 2013). The United States and Europe lead the charge, both having already banned mercury exports. Therefore, more than 80% of Colombia’s mercury imports have suddenly dried up (Table 5.2), and it is likely to be a similar story across the continent. It has taken over three years of intensive field work on the part of UNIDO’s team to reduce the use and release of mercury by 50% in target municipalities in Antioquia (as per preliminary data from select processing centres that were monitored; this figure is therefore optimistic at best). Even if we are excessively generous and apply this reduction factor to the entire department of  147 Antioquia, these miners are still set to experience a supply shortfall of 30%. Where will they get their mercury? How much will it cost? On February 29, 2012, the Andean Community (CAN 2013) held its first meeting dedicated to coordinating efforts to deal with illegal mining; a practice that has been referred to as artisanal mining in this document. Generally, both labels functionally define the same diverse category of people. Convened at the behest of the governments of Colombia and Ecuador, and including the governments of Peru and Bolivia, the meeting began with each delegation giving an outline of the illegal mining situation in their own country, and a description of the current policies and legislation that exist or are in the pipeline to address the problem. They all agreed that illegal mining is destroying thousands of hectares of Andean forests and ecosystems every year, polluting river basins, threatening biodiversity and harming human health. To quote the CAN website directly:  “Participants explored possible forms of cooperation under the framework of the CAN, in particular a proposal to develop and implement a Community legal norm that would facilitate the prevention of and fight against illegal mining. This norm would cover all aspects related to: the exploration, exploitation and marketing of minerals; environmental protection; the control of chemicals used such as mercury and cyanide; the training for artisanal miners; and social and health aspects.”   Though it is positive to see training for artisanal miners included as a related aspect, the words “fight against illegal mining” clearly set the tone for the intended legislative cooperation. I predict that if they live up to this language, mercury export bans and the fight against illegal mining will deepen the rift between artisanal miners and society, driving them further into the hands of criminal organizations for chemical and technological supplies, and for defense against the states in which they operate. Segovia used to lie at the heart of Pablo Escobar’s Medellin cocaine cartel, and gold was naturally incorporated into criminality. Buying gold from miners is the most efficient way to launder money as it is untraceable and easy to smuggle. Colombian officials still believe that gold production funds armed groups in the region.   148  The link between organized crime and artisanal mining, however, is not unique to Colombia. Early in my field work, Barrick Gold, a large Canadian mining firm, was interested in exploring the possibility of building USEPA gold shop filters in towns near its Lagunas Norte mine, high in the Andes east of Trujillo, Peru. Together with some Barrick representatives, in unmarked vehicles, we visited nearby Huamachuco, whose mayor (names and dates omitted to protect identities) was very eager to meet with us and discuss artisanal mining as an issue. It was immediately clear that the situation was impossible. “They don’t use mercury, and we can’t even go in there!” he said, referring to the hill less than 3 km from the centre of town. They have better guns than the police, and they shoot if we get too close!” The hill is encrusted with cyanidation vats, tailings piles, and mine adits. “It is illegal for them to use cyanide, but they get it from the formal mine down the road.” I had seen that formal, legal mine on the way to Lagunas Norte. The river below it is choked with rainbow oxidized fine sediments and polychrome crystals along the creek bed edges; not a normal state for a steep, coarse bedded Andean creek. “That mine is operated by the Buenaventura family [one of the wealthiest in Peru]. They hide the cyanide in lorries under piles of potatoes and ship them up here with guns and cash. Then the lorries go into the jungle, deliver more guns and chemicals, and return to the coast with drugs and gold.” Formalization is a critical step in folding artisanal miners into society, but legality is no guarantee of compliance and criminality requires enforcement. On the other hand, criminalization gives artisanal miners few options and no technical assistance, and they will defend their livelihood violently because the only alternative is desperate poverty. Yet government power is transmitted through the laws it creates, so criminalization is the natural solution to human perpetrated ecological and social disasters, especially in the absence of any evidence of effective coercive measures.  There is a desperate need for greater understanding of the hazards of mercury contamination from artisanal mining, as well as for examples of successful interventions and how they may be implemented and monitored. UNIDO has been promoting cleaner technologies by emphasizing increased production and lower costs over decreased contamination. They also emphasized formalization. Theirs is the first mercury remediation initiative to produce measured reductions in environmental contamination, in part because this is the first study to regularly monitor airborne concentrations throughout  149 an intervention. The Andean Community needs to learn from the UNIDO example, and the research presented here offers a detailed and data driven illustration.  5.4.1 UNIDO’s final project evaluation  At the time of writing, UNIDO had not yet finalized their analyses, however the preliminary results suggest impressive levels of success. As we have not yet had access to the raw data or personally investigated these claims in the field we cannot make conclusions as to the veracity of their assessments. However, given that they follow the same methods and procedures described here we expect these data and conclusions to be fairly robust. Also, it is not yet possible to directly compare the data on individual towns in this final evaluation, as these assessments were made by considering a subset of 35 entables in Segovia, Remedios, and Zaragoza. At the beginning and end of the campaign, UNIDO personnel directly measured the quantities of mercury used and recovered in these entables over a three month period. Clearly there has been a significant change in the entables shown in Table 5.11. Miners in these facilities consume about half the mercury that they did before the interventions.   Table 5.11: Pre- and post-treatment mineral processing data in 35 out of at least 323 entables across Antioquia (unverified preliminary UNIDO data). Large increases in mercury free ore processing capacity were implemented with the assistance of UNIDO.   Before After % Change Mercury consumed (kg/month) 752 389 -48% Mercury consumed (g/tonne) 1151 617 -46% Amalgamated ore (kg/month) 640 630 -2% Ore processed without mercury (kg/month) 610 4838 +693% Number of miners processing ore 1081 1036 -4%   150 Though the rate of ore amalgamation has only declined slightly, most of the new processing capacity is clearly mercury-free and requires far less labour from miners themselves as they are selling their unprocessed ore to cyanidation operations. UNIDO estimates that in 22 of the 35 evaluation entables for which they conducted a three month mercury mass balance, they have recycled 5.5 tonnes of mercury per year and prevented the consumption of over 4 tonnes of mercury per year. As Table 5.11 shows, UNIDO and its partners helped to vastly increase the processing capacity of the entables using mercury free techniques. Entables studied here cut their mercury use almost in half by improving their amalgamation practices, even though they also increased processing capacity by over 600%. Repair, maintenance, and the space required to increase coco capacity by that amount would be impossible for these entables, and large cyanidation tanks take up much less space relative to their capacity. Each additional mass of ore processed means higher profit for these entables, therefore there is a clear and compelling economic case for switching to mercury free techniques.   5.5 Summary  I have shown that time-averaged concentrations in the Segovia urban core have decreased by almost 50% over the three years of mercury reduction initiatives. Using changes in gold processing infrastructure as a proxy, gold production in Segovia was estimated to have increased by 30%. Therefore it is likely that UNIDO’s efforts have effectively reduced urban airborne mercury concentrations by 80% if one assumes that production and contamination should have increased in direct proportion in the absence of interventions. Of course, this mercury wasn’t eliminated completely, and increases in concentrations in surrounding towns suggest that much of that mercury is still being released elsewhere. Regardless, this is a significant public health achievement, as the increases in the tiny towns of Remedios and La Cruzada do not offset the decreases in Segovia, and the number of people exposed is much greater in Segovia. This work shows that early in UNIDO’s intervention, over 90 tonnes of mercury per year was lost to the environment in the five main gold producing municipalities of Antioquia alone. The estimated 30% increase in gold production would have produced about 120 tonnes of  151 mercury emissions per year in the five intervention municipalities in Antioquia, though the real number is likely to be lower thanks to increased mercury efficiency and recycling, as well as alternative processing methods. Observed reductions in airborne mercury in Segovia are attributable to use of retorts, decreased milling speeds, introduction of pre-concentration and mercury-free techniques, and to regulations that restrict new processing centres to non-urban areas. In order to unequivocally declare the effectiveness of interventions, the mass balances, production estimates, and atmospheric monitoring in this study must be repeated to establish a trend.  If the observed reduction in mercury vapour in Segovia is attributable to UNIDO’s efforts, as is likely from the data presented here, it is a significant achievement in a relatively short time, and it is the first time that reductions in mercury contamination have been measured rather than inferred from expected reductions in mercury consumption. These are laudable efforts by the gold trade community and UNIDO, both in terms of a technical and a social achievement. They must be continued, praised, and expanded. However, mercury emissions of any intensity are unacceptable in urban centres and must eventually be banned altogether.    152 6 Conclusions   This is the first detailed analysis of urban mercury emissions from artisanal and small scale mining, and the first attempt to map, model, and measure the effectiveness of mitigation. The questions raised at the beginning of this dissertation have been addressed in the following ways:  o What is the nature of mercury emissions from urban gold shops?   Mercury emission events from gold shops are elevated and intermittent, and usually result in peak airborne concentrations that exceed 50 µgm-3 and last from 5 to 30 minutes, followed by rapid dispersal. Concentrations averaged over several mobile monitoring transects which exceed 1 µgm-3 are strongly indicative of sources within 200 metres.  o What level of risk do urban atmospheric mercury emissions pose to people, and how can this risk be measured and communicated?  Mercury transect maps show that the estimated annual average mercury concentration in Andacollo, Chile and Segovia, Colombia were as high as 0.331 µgm-3 and 1.25 µgm-3 respectively, and also highlight the probable locations of frequent emitters. Both values exceed the WHO tolerable limits for exposure, assuming that the daytime average over the entirety of each field campaign is a reasonable proxy value for the annual average for that area.  o Can dispersion models be used to estimate the distribution and behaviour of urban mercury vapour at fine scales, and what are the principal challenges in generating accurate models?  CalPUFF was used to simulate micro-scale mercury release events for the first time and with reasonable accuracy. Reliable source data is essential for generating realistic plumes, and it is only obtained by building trust with the emitters. High quality and  153 density observation data are needed for model validation. Observations and dispersion models both indicate that mercury emitted from amalgam burning is rapidly diluted and may be transported large distances and eventually contribute to steadily rising global background mercury concentrations.  o How can these emissions be mitigated, and is it possible to measure the impact of mercury reduction initiatives?  Successive years of mercury mapping show that atmospheric concentrations of mercury in Segovia decreased approximately 50% despite a 30% increase in production. These reductions are attributable to adoption of retorts, decreased milling speeds, introduction of pre-concentration and mercury-free techniques, as well as to regulations that restrict new processing centres to non-urban areas.  Over 90 tonnes of mercury is emitted to air, water, and soils in the Department of Antioquia, Colombia, which is slightly less than the emissions from coal fired electrical generation in China (96.7 tonnes, UNEP 2013). Perhaps as little as 4% is emitted directly to the atmosphere; nevertheless this is more than twenty times greater than the emissions of all Colombian coal fired power plants combined (0.177 tonnes per year.)  Others have found that mercury concentrations are low in air and soil farther than 1 km from mercury sources (Lacerda 2003, Malm et al. 1995) and inferred that little mercury is transported beyond the local scale. It is therefore significant that both the observed and modeled concentrations show rapid dilution to concentrations that are far below the saturation vapour pressure of mercury. In the absence of significant chemical changes that would produce mercury species with high dry deposition rates, large amounts of evaporated mercury are being exported from the local area. The observation that mercury signals are not detectable beyond 1 km from the source indicates rapid dilution and not rapid deposition close to the source. Although downwind concentrations decline to tolerable levels or less beyond 500 m, mercury exported from the local scale is free to disperse around the globe or deposit hundreds of kilometers from the source.  154 Clearly there is a great need for mitigating mercury emissions from artisanal mining. This work not only shows that mitigation programs can be successful, but also provides a framework for assessing the severity of health impacts and magnitude of changes in airborne mercury contamination.  6.1 Original contributions of this study   The following knowledge appears for the first time in this dissertation:  o A detailed description of gold shop practices and levels of mercury contamination in a variety of contexts.  o Maps of mercury vapour in urban areas, and their application in estimating human exposure and detection of changes in contamination over time.  o A framework for analyzing and visualizing mercury source and atmospheric observation data.  o A micro-scale dispersion model of mercury emitted by artisanal miners, and comparison of model output to observations.  o Detection of reductions in environmental concentrations of mercury associated with a mercury abatement project.   6.2 Limitations of this research  The weakest part of this research is the small size of the datasets, however there are no other data with which to estimate human health risk, test urban dispersion models, or evaluate mercury interventions. This work establishes methods for collecting, analyzing, and interpreting these data so that others may find it easier to undertake similar research that can confirm or question the results found here. Using a few weeks of monitoring to derive yearly average values pushes the limits of reason, and sustained monitoring for a year or more is needed to explore the validity of such extrapolation. Furthermore, connecting potential exposure of the general population to individual dosages and  155 physiological responses is highly problematic without an extended epidemiological campaign.   There are too few instances of coincidence between observations and mercury releases to make high confidence statements about the accuracy of dispersion models.  Modeling sub-hourly variability in meteorological conditions and a more precise urban structure model are needed to precisely match the small scale on which the emissions occur and disperse. Nevertheless, it is useful to picture the general behaviour of amalgam burning emissions, and to set a benchmark for the minimum expected accuracy of models as applied to urban mercury micrometeorology. It is very difficult to ensure the transparency and reliability of processing and production information on which mitigation impact assessments are based, especially in such a complex and informal operating environment. Gathering such data requires sustained and dedicated effort by trusted and technically proficient staff over many months, and it was beyond the scope of my work to audit UNIDO’s estimates of reductions in mercury use. Furthermore, airborne mercury monitoring in Colombia was undertaken as isolated consulting work without the long term vision of assessing the progress of mitigation initiatives, and therefore the timing, extent, and frequency of monitoring was far less than ideal. Further uncertainty arises from the habits of miners themselves, who are subject to changing laws and national attitudes to mining, volatile commodity prices, gang wars, legal and property disputes, and the general instability of gold rushes. In spite of this uncertainty, the end result is highly evocative and powerful. Myriad competing objectives and pressures in development projects limit the scope for well-planned multi-year research and ongoing project assessment, therefore the results presented here are unlikely to be surpassed in terms of thoroughness and rigorousness for some time.  6.3 Recommendations for reducing mercury emissions and human exposure  The ultimate goal should be the removal of all mercury emitting activities in urban areas, and replacement of contaminating practices with gravity concentration and responsible cyanidation where appropriate. In the short term, emitters could raise their smokestacks  156 to increase dilution and transport, in addition to improving mercury condensers and using retorts. Elimination of whole ore amalgamation is the key to making drastic reductions in mercury use, and widespread miner training is sorely needed to support technical improvements. The principal barrier to progress is the insecurity that comes from informality and illegality. Even properly trained miners cannot invest in better equipment if they could be evicted from their work site, whether by the government or criminal organizations. At its root, this is a development issue, and it is unproductive to focus solely on the illegality and environmental devastation. Demonizing artisanal miners will only drive them further into the hands of organized crime, and governments should engage miners and provide them with a pathway to formality and investment, along with the security, transparency, and training they require to develop.   6.4 Recommendations for future research  These studies need to be repeated in other mercury emitting urban areas to corroborate or challenge these findings and to determine which environmental factors influence the persistence and behaviour of mercury emissions. Furthermore, continuous monitoring for a year or more is necessary to determine the true annual average concentrations and evaluate the annual average estimation methods used here. Mercury mapping studies need to be paired with epidemiological studies to elucidate the effects of this hazard on human health. International health norms need to be refined and updated by further research into the effect of intermittent extreme exposure, as it is unsure whether the health hazard in urban areas where mercury is emitted can be adequately represented by LOAEL or present international health standards. The comparison of model and observation need to be replicated with higher observation density and more complete source records for all emitters in the study area. Urban micrometeorologists should find a way to use this inadvertent tracer experiment, which is constantly occurring all over the world, to study fine scale dispersion of extreme concentrations very near to the source.  Future research should also extend the time series of mercury maps and gold production estimates as long as possible in Segovia. The mitigation and mapping  157 techniques introduced here need to be applied in other artisanal mining communities to test the robustness of these results and further reduce mercury contamination.     158 6.5 Summary  Most artisanal mining research focuses almost entirely on issues in remote areas near the mine site; however there are important human health impacts in urban areas in mining regions that have largely been ignored. I have provided a vivid illustration of the problem in two very different but equally instructive contexts, Andacollo, Chile, and Antioquia, Colombia, which had not previously been investigated in any detail. They frame the problem nicely by bounding the relative intensities of this phenomenon.  According to the local miner’s association in Andacollo, the opening of the Canadian industrial gold mine 15 years ago eliminated around 90% of the mercury emissions because they had taken over roughly that proportion of the local artisanal gold production (the miners had not formalized or obtained the legal title to their mineral deposits). Now the gold shops and the frequency of their amalgam burns are low enough to be able to distinguish individual signals, particularly on weekdays when there are fewer events. The model and observations indicate that the mercury dissipates rapidly and that locations greater than 500 metres away are unlikely to be at significant risk to adults according to WHO (2003) guidelines. Mean observed street level concentrations (averaged over all sampling time) support this, as most locations in the area had an average lower than 1 µgm-3, and the average value over the entire campaign in the gold shop neighbourhood of Andacollo was 0.331 µgm-3. Although the true estimate of annual average concentration would be half of this value to account for night hours in which emissions mostly do not occur, I argue that using averages of mostly daytime data is equivalent to using a factor of safety of two when assigning hazard. The mapping techniques used here provide an imperfect proxy for health hazard for instrumentation and methodological reasons, and the real exposures would easily be double the average concentration values found here. Segovia and the other mining towns of Antioquia define the extreme opposite end of the contamination spectrum. Though average airborne concentrations of mercury over the entirety of each field campaign in Segovia have declined from 1.25 to 0.660 µgm-3 in the period from 2010 to 2012, concentrations remain near the WHO hazard limit and mining production is steadily increasing. Individual signals may dissipate rapidly, but the sheer mass and frequency of raw amalgam burns is sufficient to regularly produce  159 hazardous concentrations in non-mining buildings across a wide area within the urban centres of Segovia.  Individual mercury signals are ephemeral and acute, lasting only about as long as the amalgam burn that produces them and then returning fairly quickly to tolerable levels. For this reason it may be more instructive to regulate emissions and communicate health hazards based on exceedences of instantaneous ceiling limits or short-term exposure limits. This is common practice in the case of particulate emissions that can vary greatly depending on weather but still have an acute impact on certain days even if the long-term averages are low (WHO 2005). However, health impacts of the extreme and intermittent kind seen in urban mining towns have not been studied and it is not clear how one would establish such guidelines. The duration of exposure at various intensities is a key question to address when studying or attempting to establish guidelines, and little is understood about how extremely variable exposure integrates physiologically into a hazardous dose.  If they could be established, CalPUFF could model mercury exceedences with fair accuracy in a simple urban environment, as it can generate model plumes that match real plumes quite well in terms of intensity and position. However, the model fails to predict instantaneous street level concentrations with accuracy in Andacollo and any analysis of risk using the cumulative average model output mercury concentration would have to be calibrated with rigorous epidemiological studies. In practice this is unlikely to happen soon, just as in retrospect it was perhaps unlikely that the mission in Andacollo would have provided all the necessary elements to realize this study. Most sites are too complex in terms of social challenges and urban structure, with too many intermingling sources and too little trust or cooperative spirit on the part of emitters.  Sporadic sampling nevertheless provided mercury vapour data for almost all of the known amalgam burn events in Andacollo. On most of those occasions the model produced signals at the receptor locations where observations showed elevated concentrations in the same time period, and forecasted background in most instances where there were no observed signals. However it is easy to be accurate in predicting that extreme events which are rare and limited in space and time will not occur, so I prefer to focus on the detection of definite mercury signals. In five of 16 instances in which the model predicts high concentrations at a tower, the model output is modestly but  160 significantly correlated with the observations for that time and place, and the event duration averages for all of those agree within a factor of three or less. In truth, such a meticulously developed and tested model that has been applied to a wide variety of pollutants, contexts, and scales over decades was likely to work fairly well in a new context. However there were no previous studies that applied CalPUFF at such fine scales, nor have any ever been applied to mercury emitted from artisanal mining below continental scales. This work provides a framework for analyzing and visualizing this data, and outlines the kinds of difficulties to be faced by anyone choosing to take up this mercury micrometeorology challenge.  Together, the gold processing, mapping and modeling studies presented here provide a new and higher resolution understanding of a phenomenon of significant local and global human and environmental health impact. I have shown that mercury emissions pose a significant human health risk near the source, but are also exported in large amounts that rapidly dilute. Generating a time series of mercury maps of a single area provides a striking visual image of atmospheric mercury contamination and the impacts of remediation initiatives. Normally, mercury abatement project leaders must be satisfied with knowing how many trainers and miners were trained to use cleaner techniques, and perhaps (though unlikely) some mercury balances before and after the initiative. This is the first study to detect any improvement in environmental quality from a mercury abatement project.   I have communicated the Colombia studies presented here in dozens of radio and television interviews, presentations in workshops, UN forums, government meetings at all levels, and even greater numbers of presentations of my data have been made by other people. Though it is difficult to measure, my Colombian and UN colleagues are convinced that showing my maps of urban mercury contamination significantly contributed to their success in convincing miners, communities, governments, and donors of the gravity of the mercury problem.  Even a single map of mercury, made in a few hours over a few days, can raise peoples’ awareness of the magnitude of the problem. Recently I travelled to the Chocó region adjacent to Antioquia, though it might as well have been on another continent as with regards to culture, ethnicity, and connection to the rest of Colombia. UNIDO  161 recently started a new mercury project in Chocó and I was invited to do some mobile monitoring. I was driven out to all the remote mining towns and, unsurprisingly, found extreme street concentrations in all urban centres that had gold shops. On the way back I insisted on taking a few hours to do some mobile sampling in Quibdo, the capital city of Chocó. As usual, I drove in circles with several other scientists and government officials, watching the mercury spikes register on the laptop. Within three or four blocks of the Ministry of Environment of Chocó, where my accompanying colleagues work, several gold shops were emitting strong and frequent mercury plumes. One these colleagues, Harry Mosquera (personal communication 2012), was startled by this knowledge, and said “It’s amazing! For years we have been so focused on the environmental and health impacts for miners as they amalgamate in their gold pans right in the river or burn mercury out in the forest without a retort… but the problem is right here! It is right under our noses, but no one thought to look because we assumed that the contamination was out there and the gold brought into Quibdo no longer contained mercury!”  The simple monitoring and visualization techniques I have established here can be used by anyone with a vapour analyzer, GPS, and a vehicle. If vehicle surveys are not possible, dispersion modeling demonstrated in this study could also be used to predict likely patterns of mercury contamination if the emitter locations and gold production are known or estimated. Alternatively, one could compare it with Andacollo, a low frequency, low intensity mercury hazard area, and Segovia, the extreme case, and qualitatively propose average concentrations and distributions relative to these boundary cases. Mercury is a silent menace, and it is important to illustrate the degree to which people are at risk even though they cannot detect it with their own senses.  The most surprising of my findings was the strong and consistent evidence for decreases in mercury contamination in Segovia. I had expected the degree of change resulting from UNIDO’s initiatives would be too small to detect, and that daily, weekly, and monthly variability in gold production would exceed the impact of interventions. However the sheer magnitude of the difference and the fact that similar decreases were found in the second and third field seasons makes it highly unlikely that the observed differences are due only to random chance. While results of further repetitions of these field campaigns could regress to some mean concentration value that might be higher  162 than that observed in 2012, I suspect that concentrations would be far lower than if there had been no intervention. Future research should extend the time series of mercury maps and gold production estimates as long as possible. These mitigation and mapping techniques can be applied in other artisanal mining communities to test the robustness of these results and further reduce mercury contamination.  Mercury remediation and replacement is possible, as this dissertation has shown, though not at the pace that is necessary for artisanal miners to adapt to the swell of international pressure being exerted on them. Mapping these emissions provides an important tool for understanding and illustrating the distribution of mercury contamination, and modeling them can provide similar insights where field measurements are absent. Mitigation of mercury emissions is an important and pressing goal, though elimination will take years of building relationships with, transferring technology to, and formalizing the operations of artisanal miners. It is up to the international community to avoid further marginalization of artisanal mining and help them lift their communities out of poverty in a legal and environmentally sustainable way. I hope that my dissertation helps to inform this cause.    163 References  Amankwah, R.K., Styles, M.T., Nartey, R.S., 2010. The application of direct smelting of gold concentrates as an alternative to mercury amalgamation in small-scale gold mining operations in Ghana. International Journal of Environment and Pollution Vol. 41, no. 3-4, pp. 304-315.   AMAP/UNEP, 2008. Technical background report to the global atmospheric mercury assessment. http://www.chem.unep.ch/mercury/Atmospheric_Emissions/ Technical_background_report.pdf. Accessed May 2012.   Argonne National Laboratory, Environmental Science Division, 2007. Technology for Mercury Emission Reductions in Small-Scale Gold Refining Facilities: Construction, Installation, and Testing. http://www.artisanalmining.org/index.cfm?page=page_disp&pid=4264, pp. 30. Accessed May 2012.  Artaxo, P., Calixto de Campos, R., Fernandes, E. T., V Martins, J., Xiao, Z., Lindqvist, O., Maenhaut, W. 2000. Large scale mercury and trace element measurements in the Amazon basin. Atmospheric Environment, vol. 34 no. 24, pp 4085-4096.  Bastos, W.R., de Freitas Fonseca, M., Pinto, F.N., de Freitas Rebelo, M., Silva dos Santos, S., Glória da Silveira, E., Torres, J.P.M., Malm, O. and Pfeiffer, W.C., 2004. Mercury persistence in indoor environments in the Amazon Region, Brazil, Environmental Research, vol. 96, no. 2, pp. 235-238.  Batchvarova, E. & Gryning, S. 1994. An applied model for the height of the daytime mixed layer and the entrainment zone. Boundary Layer Meteorology, vol. 71, no. 3, pp. 311-323.  Brooks T.M., Mittermeier R.A., da Fonseca G.A.B., Gerlach J., Hoffmann M., 2006. Global Biodiversity Conservation Priorities. Science vol. 313, pp. 58–61.   164 Brooks W.E., Sandoval E., Yepez M.A., Howell H., 2007. Peru Mercury Inventory 2006: U.S. Geological Survey Open-File Report 2007-1252. http://pubs.usgs.gov/of/2007/1252/. Accessed May 2013, pp. 55.  Caputo, M., Giménez, M., and Schlamp, M., 2003. Intercomparison of atmospheric dispersion models. Atmospheric Environment vol. 37 no. 18. pp. 2435-2449.  Castilhos ZC, Rodrigues-Filho S, Rodrigues AP, Villas-Bôas RC, Siegel S, Veiga MM, Beinhoff C., 2006. Mercury contamination in fish from gold mining areas in indonesia and human health risk assessment. Science of the Total Environment, vol. 368, pp. 320–325.  CETEM, 1992. Preliminary diagnosis of the environmental impacts caused by gold prospecting in Alta Floresta.  Farid, LH, (ed.) Centro de Tecnologia Mineral, Rio de Janeiro. 185 pp.  Clifford, M.J., 2011. Pork knocking in the land of many waters: Artisanal and small-scale mining in Guyana. Resource Policy. Vol. 36, no. 4, pp. 354-362.  Constantinou, E., X. A. Wu, and C. Seigneur, 1995. Development and application of a reactive plume model for mercury emissions. Water, Air, & Soil Pollution, vol. 80, no. 1, pp. 325-335.  Communidad Andina, 2013. http://www.comunidadandina.org/prensa/notas/np29-2-12.htm. Accessed April 6, 2013.  Cordy, P., Veiga, M., Crawford, B., Garcia, O., Gonzalez, V., Moraga, D., Wip, D., 2013. Characterization, mapping, and mitigation of mercury vapour emissions from artisanal mining gold shops. Environmental research, vol. 125, pp.82-91.      165 Cordy, P., Veiga, M.M., Salih, I., Al-Saadi, S., Console, S., Garcia, O., Mesa, L.A., Velásquez-López, P.C., Roeser, M. 2011. Mercury contamination from artisanal gold mining in Antioquia, Colombia: The world's highest per capita mercury pollution, Science of The Total Environment, vol. 410, no. 1, pp. 154-160.  Corral, S., Sáez, D., Lam, G., Lillo, P., Sandoval, R., Lancellotti, D., and Pancetti, F. (2013). Neurological and neuropsychological deterioration in artisanal gold miners from the town of Andacollo, Chile. Toxicological & Environmental Chemistry, vol. 95, no. 2, pp. 344-358.   Crawford, B., Christen, A., Ketler, R., 2010. EPiCC technical report #1: Processing and quality control procedures of turbulent flux measurements during the Vancouver EPiCC experiment. University of British Columbia, 15 pp.  Dastoor, A.P., and Larocque, Y., 2004. Global circulation of atmospheric mercury: a modelling study. Atmospheric Environment, vol. 38, no. 1, pp. 147-161.  Djordjevic, N., Shi, F. N., and Morrison, R., 2004. Determination of lifter design, speed and filling effects in AG mills by 3D DEM. Minerals engineering, vol. 17, no. 11, 1135-1142.  Drake, P.L., 2001. Occupational exposure to airborne mercury during gold mining operations near El Callao, Venezuela. International Archives of Occupational and Environmental Health, vol.74, no.3, pp. 206-212.  Ferrara, R., Maserti, B.E., Andersson, M., Edner, H., Ragnarson, P., Svanberg, S. and Hernandez, A., 1998. Atmospheric mercury concentrations and fluxes in the Almadén district (Spain). Atmospheric Environment, vol. 32, no. 22, pp. 3897-3904.  Finnigan, J.J., R. Clement, Y. Malhi, R. Leuning, and H.A. Cleugh, 2002. A re-evaluation of long-term flux measurement techniques, Part I: Averaging and coordinate rotation. Boundary-Layer Meteorology, vol. 107, pp. 1-48.  166  Flynn, C. M., and McGill, S. M., 1995. Cyanide Chemistry: Precious Metals Processing and Waste Treatment. US Department of the Interior, Bureau of Mines. NTIS, 282 pp.  Gestion Ambiental de Mineria Artesanal (GAMA), 2009, Retorta Communal. http://www.gama-peru.org/gama/tiki-index.php?page=Publicaciones. Accessed February 2013.  Garcia, A.S., Contreras, F., Adams, M. and Santos, F., 2006. Atmospheric mercury emissions from polluted gold mining areas (Venezuela). Env Geochem. and Health, vol.28 no.6, pp. 529-540.   Gbor, P. K., Wen, D., Meng, F., Yang, F., Zhang, B., and Sloan, J. J., 2006. Improved model for mercury emission, transport and deposition. Atmospheric Environment, vol. 40, no. 5, pp. 973-983.  Golder, D., 1972. Relations among stability parameters in the surface layer." Boundary-Layer Meteorology, vol. 3, no. 1, pp. 47-58.  Gosar, M., Šajn, R. & Biester, H., 2006. Binding of mercury in soils and attic dust in the  Idrija mercury mine area (Slovenia), Science of the Total Environment, vol. 369, no. 1-3, pp. 150-162.  Gran Colombia Gold ltd. 2013. http://www.grancolombiagold.com/operations-and projects/segovia/overview/default.aspx. Accessed April 6, 2013.  Grimmond, C. S. B., & Oke, T. R. 1999. Aerodynamic properties of urban areas derived from analysis of surface form. Journal of Applied Meteorology, 38, no. 9, 1262-1292.      167 Grönlund, R., Edner, H., Svanberg, S., Kotnik, J. and Horvat, M., 2005. Mercury emissions  from the Idrija mercury mine measured by differential absorption lidar techniques and a  point monitoring absorption spectrometer, Atmospheric Environment, vol. 39, no. 22, pp. 4067-4074.  Gurmendi, A.G., 2006, The mineral industry of Peru, in Area Reports—International –Latin America and Canada: U.S. Geological Survey Minerals Yearbook 2004, v. III, p. 15.1-15.16. http://minerals.usgs.gov/minerals/pubs/country/2004/pemyb04.pdf. Accessed July 2013.  Hacon, S., Artaxo, P., Gerab, F., Yamasoe, M.A., Campos, R.C., Conti, L.F. and De Lacerda, L. D., 1995. Atmospheric mercury and trace elements in the region of Alta Floresta in the Amazon Basin. Water, Air, & Soil Pollution, vol. 80, no. 1, pp. 273-283.   Hanna, S. R., Brown, M. J., Camelli, F. E., Chan, S. T., Coirier, W. J., Kim, S., Reynolds, R. M., 2006. Detailed simulations of atmospheric flow and dispersion in downtown Manhattan: An application of five computational fluid dynamics models. Bulletin of the American Meteorological Society, vol. 87, no. 12, pp. 1713-1726.  Heckel, P. F., & LeMasters, G. K., 2011. The Use of AERMOD Air Pollution Dispersion Models to Estimate Residential Ambient Concentrations of Elemental Mercury. Water, Air, & Soil Pollution, vol. 219 no. 1-4, pp. 377-388.  Hentschel, T., Hruschka, F., and Priester, M., 2003. Artisanal and small-scale mining: challenges and opportunities. Iied. http://books.google.ca/books?hl=en&lr=&id=kamklwIa6acC&oi=fnd&pg=PR7&dq=Hentschel,+T.,+Hruschka,+F.,+%26+Priester,+M.+%282003%29.+Artisanal+and+small-scale+mining:+challenges+and+opportunities.+Iied.&ots=tPoOexwUHj&sig=yUrFObbAPOpeNeaWxl1WhoOs5BE#v=onepage&q&f=false. Accessed July 2013.   168 Higueras, P., Oyarzun, R., Lillo, J., Oyarzún, J and Maturana, H., 2005. Atmospheric mercury data for the Coquimbo region, Chile: influence of mineral deposits and metal recovery practices.  Atm. Env, vol.39 no.39, pp. 7587-7596.  Hilson, G., 2012. Family Hardship and Cultural Values: Child Labor in Malian Small-Scale Gold Mining Communities. World Development, vol. 40, no. 8 pp. 1663-1674.  Hilson, G, Hilson, C.J., Pardie, S., 2007. Improving awareness of mercury pollution in small-scale gold mining communities: Challenges and ways forward in rural Ghana. Environmental Research, vol. 103, no. 2, pp. 275-287.  Hodgson, S., Nieuwenhuijsen, M. J., Colvile, R., and Jarup, L. 2007. Assessment of exposure to mercury from industrial emissions: comparing “distance as a proxy” and dispersion modelling approaches. Occupational and Environmental Medicine, vol. 64, no. 6, pp. 380-388.  Howard J, Trotz MA, Thomas K, 2011. Total mercury loadings in sediment from gold mining and conservation areas in Guyana.  Environmental Monitoring and Assessment. vol. 179, pp. 555-573.  Instituto de Investigaciones en Geociéncias, Mineria y Química (INGEOMINAS), 1995. Colombia, a Gold Mine. Ministry of Mines and Energy, 25 pp.  Internal Displacement Monitoring Centre, 2009. IDMC submission to the Human Rights Committee, Internal Displacement Monitoring Centre (IDMC). http://www.internal-displacement.org/countries/colombia.  Published 15 August. Accessed June 2013.  Jensen S, Jernelov A., 1969. Biological methylation of mercury in aquatic organisms. Nature, vol. 223 pp. 753-754.  Jones HR., 1979. Mercury pollution control. Noyes Data Co., New Jersey, 251 pp.  169  Jonsson, J.P., Appel, P.W., Chibunda, R.T., 2009. A matter of approach: the retort's potential to reduce mercury consumption within small-scale gold mining settlements in Tanzania. Journal of Cleaner Production, vol. 17, no. 1, pp. 77-86.  Kitco.com, www.kitco.com, accessed February 2008.  Koracin, D., and Berkowicz, R. 1988. Nocturnal boundary-layer height: Observations by acoustic sounders and predictions in terms of surface-layer parameters. Boundary-Layer Meteorology, vol. 43 no. 1-2, 65-83.  Krigger, K.W., 2006. Cerebral palsy: an overview. American Family Physician. Vol. 73, no. 1, pp. 91-100.  Lacerda L, 2003. Updating global Hg emissions from small-scale gold mining and assessing its environmental impacts, Environmental Geology, vol. 43, no. 3, pp. 308-314.  Lanzinger, E., & Langmack, H. Measuring air temperature by using an ultrasonic anemometer. 2005. TECO-2005, Bucharest, Romania, available at: http://www. wmo. int/pages/prog/www/IMOP/publications/IOM-82-TECO_2005/Posters. pp 3. Accessed May, 2011.  Lanphear BP, Hornung R, Khoury J, Yolton K, Baghurst P, Bellinger DC, Canfield RL, Dietrich KN, Bornschein R, Greene T, Rothenberg SJ, Needleman HL, Schnaas L, Wasserman G, Graziano J. 2005. Low-level environmental lead exposure and children’s intellectual function: An international pooled analysis.  Environ Health Perspect, vol. 113, pp. 894-899.     170 Laplante AR, Spiller DE. In: Mular AL, Halbe DN, Barratt DJ, editors. Mineral processing plant design, practice, and control. Bench-sale and pilot plant testwork for gravity concentration circuit design, vol. 1. Littleton, CO: SME – Society for Mining, Metallurgy and Exploration; 2002. p. 160–75.   Larsen, R. I., 1971. Mathematical model for relating air quality measurements to air quality standards. Government Printing Office, Washington, DC (USA).  Lena, F., and Desiato, F., 1999. Intercomparison of nocturnal mixing height estimate methods for urban air pollution modelling. Atmospheric Environment, vol. 33 no. 15, pp. 2385-2393.  Levin M, Jacobs J, Polos PG. 1988. Acute mercury poisoning and mercurial pneumonitis from gold ore purification. Chest, vol 94, pp. 554-56.  Lindberg, S., Bullock, R., Ebinghaus, R., Engstrom, D., Feng, X., Fitzgerald, W., and Seigneur, C., 2007. A synthesis of progress and uncertainties in attributing the sources of mercury in deposition. AMBIO: A Journal of the Human Environment, vol. 36 no. 1, 19-33.  Lindberg SE, Turner RR, Meyers TP, Taylor GE, Schroeder WH, 1991. Atmospheric concentrations and deposition of mercury to a deciduous forest at Walker Branch Watershed, Tennessee, USA. Water, Air, and Soil Pollution, vol. 56, pp. 577–594.  Lindqvist O., 1991a. Mercury in the Swedish environment: 9. Mercury in terrestrial ecosystems bioavailability and effects. Water, Air, and Soil Pollution, vol. 55 no. 1–2, pp. 101–108.  Lindqvist O., 1991b. Mercury in the Swedish environment: 4. Emissions of mercury to the atmosphere. Water, Air, and Soil Pollution, vol. 55 no.1–2, pp. 23–32.  Mahrt, L., Heald, R. C., Lenschow, D. H., Stankov, B. B., & Troen, I. B., 1979. An observational study of the structure of the nocturnal boundary layer. Boundary-Layer Meteorology, vol. 17, no. 2, 247-264.  171  Malm, O., de Freitas Fonseca, M., Hissnauer Miguel, P., Rodrigues Bastos, W. and Neves Pinto, F., 1998. Use of epiphyte plants as biomonitors to map atmospheric mercury in a gold trade center city, Amazon, Brazil. Sci Tot. Env., vol.213 no.1-3, pp. 57-64.   Malm, O., Castro, M.B., Bastos, W.R., Branches, F.J.P., Guimarães, J.R.D., Zuffo, C.E. and Pfeiffer, W.C. 1995, "An assessment of Hg pollution in different goldmining areas, Amazon Brazil", Science of The Total Environment, vol. 175, no. 2, pp. 127-140.  Malm, O., Pfeiffer, W. C., and Souza, C. M. M., 1991. Main pathways of mercury in the Madeira river area, Rondonia, Brazil. CEP Consultants, (ed.) Edinburgh, UK In Proceedings of the VIII International Conference on Heavy Metals in the Environment. pp. 515-518.  Malm, O., Pfeiffer, W.C., Souza, C.M.M., Reuther, R., 1990. Mercury Pollution Due to Gold Mining in the  Madeira River Basin, Brazil. Ambio, vol. 19, no. 1, pp. 11-15.  Marins, R.V., de Andrade, J.B., Pereira, P.A., Paiva, E.C. and Paraquetti, H.M., 2000. Sampling techniques for the assessment of anthropogenic vapour and particulate mercury in the Brazilian Amazon atmosphere. Journal of Environmental Monitoring, vol. 2, pp. 325-328.  Marrugo-Negrete J, Benitez LN, Olivero-Verbel J., 2008.  Distribution of mercury in several environmental compartments in an aquatic ecosystem impacted by gold mining in Northern Colombia. Archives of Environmental Contamination and Toxicology, vol. 55 pp. 305–316.  Mason, R.P., A.L. Choi, W.F. Fitzgerald, C.R. Hammerschmidt, C.H. Lamborg, A.L. Soerensen and E.M. Sunderland, 2012. Mercury biogeochemical cycling in the ocean and policy implications. Environmental Research, vol. 119, pp. 101-117.   172 Massey, F.J., 1951. The Kolmogorov-Smirnov Test for Goodness of Fit. J. Am. Stat. Ass., vol. 46, no.253, pp. 68-78.  McDaniels J, Chouinard R, Veiga MM. 2010. Appraising the Global Mercury Project: an adaptive management approach to combating mercury pollution in small-scale gold mining. International Journal of Environmental Pollution, vol. 41, no. 3-4 pp. 242-258.  McMillen, R.T., 1988. An eddy correlation technique with extended applicability to non-simple terrain. Boundary-Layer Meteorology, vol. 43, pp. 231-245.  Moore, C.J., 1986. Frequency response corrections for eddy correlation systems. Boundary-Layer. Meteorology, vol. 37, pp. 17-35.  Nath, S. & Patil, R.S. 2006. Prediction of air pollution concentration using an in situ real time mixing height model. Atmospheric Environment, vol. 40, no. 20, pp. 3816-3822.  Olivero J, Johnson B, Arguello E., 2002. Human exposure to mercury in San Jorge River Basin, Colombia (South America). Science of the Total Environment, vol. 289, pp. 41-47.  Oshan, R., Kumar, A., Masuraha, A., 2006. Application of the USEPA's CALPUFF model to an urban area. Environmental Progress, vol. 25, no. 1, pp. 1547-5921.  Pai, P., Karamchandani, P., and Seifneur, C., 1997. Simulation of the regional atmospheric transport and fate of mercury using a comprehensive Eulerian model. Atmospheric Environment, vol. 31 no. 17, pp. 2717-2732.  Pautrat, L. 2000. Caracterización de la Explotación Aurífera en el Departamento de Madre de Dios y su Influencia en la Biodiversidad. wwfperu.org.pe/SWA/Mineria/mineria1.html. Accessed June 2013.   173 Perfetti JJ. 2009. Crisis y pobreza rural en América Latina: el caso de Colombia .Documento de Trabajo n. 43. Programa Dinámicas Territoriales Rurales Rimisp – Centro Latinoamericano para el Desarrollo Rural, Santiago, Chile. 29pp. Available at: http://www.rimisp.org/FCKeditor/UserFiles/File/documentos/docs/pdf/. Accessed November 2009.  Peterson, G.D. and Heemskerk M., 2001. Deforestation and forest regeneration following small-scale gold mining in the Amazon: the case of Suriname. Environmental Conservation, vol. 28, pp. 117-126.  Pirrone, N., Cinnirella, S., Feng, X., Finkelman, R. B., Friedli, H. R., Leaner, J., ... & Telmer, K., 2010. Global mercury emissions to the atmosphere from anthropogenic and natural sources. Atmospheric Chemistry and Physics, vol. 10, no. 13, 5951-5964.  Pirrone, N., Ferrara, R., Hedgecock, I. M., Kallos, G., Mamane, Y., Munthe, J., Wangberg, I., 2003. Dynamic processes of mercury over the Mediterranean region: results from the Mediterranean Atmospheric Mercury Cycle System (MAMCS) project. Atmospheric Environment, vol. 37, pp. 21-39.  Pirrone, N., and Mahaffey, K. R. (Eds.), 2005. Dynamics of mercury pollution on regional and global scales: atmospheric processes and human exposures around the world. Springer.  Poulos, G. S., Blumen, W., Fritts, D. C., Lundquist, J. K., Sun, J., Burns, S. P., Jensen, M., 2002. CASES-99: A comprehensive investigation of the stable nocturnal boundary layer. Bulletin of the American Meteorological Society, vol. 83, no. 4, pp. 555-581.  PRODEMINCA. Proyecto de Desarrollo Minero y Control Ambiental. Prodeminca. Quito: Swidish Environmental Systems. Ministerio de Energı´a y Minas del Ecuador; 1999. 212 pp.   174 Rodrigues-Filho S, Castilhos ZC, Santos R, Yallouz AV, Nascimento F, Egler SG, Peregovich B, Almeida Riberio RA, 2004. Environmental and health assessment in two small-scale gold mining areas in Brazil (Sao Chico and Creporizinho). Technical Report by CETEM to UNIDO, RT2004-004-02. Rio de Janeiro, 156 pp.  Ryaboshapko, A., Bullock Jr, O. R., Christensen, J., Cohen, M., Dastoor, A., Ilyin, I., and Munthe, J., 2007. Intercomparison study of atmospheric mercury models: 1. Comparison of models with short-term measurements. Science of the Total Environment, vol. 376, no. 1, pp.228-240.  Ryaboshapko, A., Bullock, R., Ebinghaus, R., Ilyin, I., Lohman, K., Munthe, J., Wängberg, I., 2002. Comparison of mercury chemistry models. Atmospheric Environment, vol. 36, no. 24, pp. 3881-3898.  Schotanus, P., F.T.M. Nieuwstadt, H.A.R. Bruin., 1983. Temperature measurement with a sonic anemometer and its application to heat and moisture fluxes. Boundary-Layer Meteorology, vol. 26, pp. 81-93.  Schuster, P.F., R.G. Striegl, G.R. Aiken, D.P. Krabbenhoft, J.F. Dewild, K. Butler, B. Kamark and M. Dornblaser, 2011. Mercury export from the Yukon River Basin and potential response to a changing climate. Environmental Science and Technology, vol. 245, pp. 9262-9267.    Scire, J. S., Strimaitis, D. G., and Yamartino, R. J., 2000. A user’s guide for the CALPUFF dispersion model. Earth Tech, Inc, 521.  Sholupov, S., Pogarev, S., Ryzhov, V., Mashyanov, N. and Stroganov, A. 2004, Zeeman atomic absorption spectrometer RA-915+ for direct determination of mercury in air and complex matrix samples, Fuel Processing Technology, vol. 85, no. 6-7, pp. 473-485.    175 Singhvi, R., Turpin, R., Kalnicky, D.J., Patel, J., 2001. Comparison of field and laboratory methods for monitoring metallic mercury vapor in indoor air. Journal of Hazardous Materials, vol. 83, no. 1-2, pp. 1-10.  Sippl, K., Selin, H., 2012. Global Policy For Local Livelihoods: Phasing Out Mercury in Artisanal and Small-Scale Gold Mining. Environment, vol. 54, no. 3, pp. 18-29.  Solis, M. T., Yuen, E., Cortez, P. S., and Goebel, P. J., 2000. Family poisoned by mercury vapor inhalation. The American Journal of Emergency Medicine, vol. 18, no. 5, pp. 599-602. Sousa, R. N., & Veiga, M. M., 2008. Garimpo Canaan: a successful case of reclamation of an artisanal gold mine in the Amazon. https://circle.ubc.ca/bitstream/handle/2429/8310/15Sousa+Veiga+Paper.pdf?sequence=1. Accessed May 2012. Sousa R.N., Veiga M.M., Van Zyl D., Telmer, K., Spiegel S., Selder J., 2011. Policies and regulations for Brazil’s artisanal gold mining sector: analysis and recommendations. Journal of Cleaner Production, vol. 19, pp. 742-750. Spiegel, S. J., & Veiga, M. M., 2006. Knowledge sharing in artisanal mining communities: mercury, technology and sustainability in developing countries.  https://circle.ubc.ca/bitstream/handle/2429/8455/13%20Spiegel%20Veiga%20%20Paper.pdf?se. Accessed November 13, 2012. Spiegel, S. J., & Veiga, M. M., 2010. International guidelines on mercury management in small-scale gold mining. Journal of Cleaner Production, vol. 18 no.4, 375-385.  Steyn, D. G., and T. R. Oke., 1982. The depth of the daytime mixed layer at two coastal sites: A model and its validation. Boundary-Layer Meteorology, vol. 24, no. 2, pp. 161-180.  Stopford W. 1979. Industrial exposure to mercury. The Biogeochemistry of Mercury in the Environment. J.O. Nriagu (Ed.). Elsevier, Biomedical Press, Amsterdam. p.367-397.  176  Stull, Roland B. An introduction to boundary layer meteorology, vol. 13. Springer, 1988.  Stull, Roland B. 2000. Meteorology: For Scientists and Engineers. Brooks/Cole, 2000.  Subir, M., Ariya, P. A., and Dastoor, A. P., 2011. A review of uncertainties in atmospheric modeling of mercury chemistry I. Uncertainties in existing kinetic parameters–Fundamental limitations and the importance of heterogeneous chemistry. Atmospheric Environment, vol. 45 no. 32, pp. 5664-5676.  Subir, M., Ariya, P. A., and Dastoor, A. P., 2012. A review of the sources of uncertainties in atmospheric mercury modeling II. Mercury surface and heterogeneous chemistry–A missing link. Atmospheric Environment, vol. 46, 1-10.  Swenson, J. J., Carter, C. E., Domec, J. C., and Delgado, C. I., 2011. Gold mining in the Peruvian Amazon: global prices, deforestation, and mercury imports. PloS one, vol. 6 no. 4, pp. 242-258.  Telmer K, Veiga MM. World emissions of mercury from small scale artisanal gold mining and the knowledge gaps about them. In: Pirrone, N., Mason, R. (Eds.), Mercury Fate and Transport in the Global Atmosphere: Measurements Models and Policy Implications. UNEP United Nations Environmental Program, pp. 96-129; 2009.  Tomicic C, Vernez D, Belem T., 2009. Human mercury exposure associated with small-scale gold mining in Burkina Faso.  International Archives of Occupational and Environmental Health. Vol. 84, pp. 539-546.  Travnikov, O., 2005. Contribution of the intercontinental atmospheric transport to mercury pollution in the Northern Hemisphere." Atmospheric Environment vol.39, 7541-7548.    177 Tschakert P. 2010. Mercury in fish: a critical examination of gold mining and human   contamination in Ghana. International Journal of Environment and Pollution, vol. 41, pp. 214-228.  Tschakert P, Singha K., 2007. Contaminated identities: Mercury and marginalization in Ghana's artisanal mining sector. Geoforum, vol. 38, pp. 1304-1321.  United Nations Comtrade, 2009. http://comtrade.un.org/. Accessed May 2009.  UNDP-United Nations Development Programme. 2010. Human development report. Overcoming barriers: human mobility and development. http://hdrstats.undp.org/en/ countries/profiles/COL.html2010. Accessed August 2010.  United Nations Environment Program (UNEP), 2012. Minamata convention on mercury. http://www.unep.org/hazardoussubstances/MercuryNot/MercuryNegotiations/tabid/3320/language/en-US/Default.aspx. Accessed April 6, 2013.  United Nations Environment Program, 2013. Global mercury assessment. http://www.unep.org/hazardoussubstances/Mercury/Informationmaterials/ReportsandPublications/tabid/3593/Default.aspx. UNEP Chemicals Branch, Geneva, Switzerland. Accessed September 2013.   United Nations High Commission on Refugees (UNHCR), 2008). Country Profiles: Colombia. Retrieved from http://www.unhcr.org/statistics/RefScout/refscout.html# . Accessed November 1, 2009.  United Nations Industrial Development Organization (UNIDO), 2007. Yayasan Tambuhak Sinta. (2007).  Yayasan Tambuhak Sinta GMP final report to UNIDO.  Kereng Pangi, Indonesia. 54 pp. http://www.globalmercuryproject.org/countries/indonesia/indonesia.htm.  Accessed May 2012.  178  United States Environmental Protection Agency (USEPA), 1984. Mercury health effects updates: Health issue assessment. Final report. Washington, DC, US Environmental Protection Agency, Office of Health and Environmental Assessment. (Document No. EPA 600/8-84-019F).  United States Geological Survey, 2013. Mineral Commodity Summaries, January 2013.  minerals.usgs.gov/minerals/pubs/commodity/gold/mcs-2013-gold.pdf . Accessed November 1, 2013.  Valdivia, S. M. and Ugaya, C. M. L., 2011. Life Cycle Inventories of Gold Artisanal and Small-Scale Mining Activities in Peru. Journal of Industrial Ecology, vol. 15, pp. 922–936.  Veiga M. M. 1997. Introducing new technologies for abatement of global mercury pollution in Latin America. UNIDO/UBC/CETEM. Rio de Janeiro, 94 npp.  Veiga, M. M., 1997. Artisanal gold mining activities in Suriname. United Nations Industrial Development Organization (UNIDO). http://142.103.159.167/countries/suriname/UNIDO%20Veiga%20Suriname1997-nomap.pdf. Accessed June 2013.  Veiga, M. M., Angeloci, G., Hitch, M., and Colon Velasquez-Lopez, P., 2014. Processing centres in artisanal gold mining. Journal of Cleaner Production, vol. 64, pp. 535-544.  Veiga M. M., Baker R., 2004. Protocols for environmental and health assessment of mercury released by artisanal and small-scale gold miners. GEF/UNDP/UNIDO Global Mercury Project. Vienna, 289 pp. http://www.undp.org/gef/documents/iw/practitioner/Protocols_for_Environmental_Health_Assess_of_Mercury-Released%20by-Artisanal-Small-Scale-Gold-Miners-1.pdf. Accessed May 2012.   179 Veiga M. M., Bermudez D., Pacheco-Ferreira H., Pedroso L.R.M, Gunson A.J., Berrios G., 2005. Mercury pollution from artisanal gold mining in Block B, El Callao, Bolívar State, Venezuela. In: Pirrone N, Mahaffey KR, editors. Dynamics of mercury pollution on regional and global scales, atmospheric processes and human exposures around the world. Norwell, MA, USA, Springer. pp. 421–50.  Veiga,M. M. and Hinton, J.J., 2002. Abandoned Artisanal Gold Mines in the Amazon: A Legacy of Mercury Pollution. Natural Resources Forum, vol. 26, no. 1, pp. 15-26. Veiga, M. M., Nunes, D., Klein, B., Shandro, J. A., Velasquez, P., & Sousa, R. N., 2009. Mill leaching: a viable substitute for mercury amalgamation in the artisanal gold mining sector?. Journal of Cleaner Production, vol. 17, no. 15, pp. 1373-1381.  Velasquez, C.P., 2010. Mercury in artisanal and small scale gold mining: identifying strategies to reduce environmental contamination in Southern Ecuador. https://circle.ubc.ca/handle/2429/29397. Accessed June 2013.  Velasquez PC, Veiga MM, Klein B, Shandro J, Hall K. 2011. Cyanidation of mercury-rich tailings in artisanal and small-scale gold mining: identifying strategies to manage environmental risks in Southern Ecuador. Journal of Cleaner Production, vol. 19, pp. 1125-1133.  Vickers, D. and L. Mahrt, 1997. Quality control and flux sampling problems for tower and aircraft data. Journal of Atmospheric and Oceanic Technology, vol. 14, pp. 512-526.  Wang, D., Shi, X. & Wei, S. 2003, Accumulation and transformation of atmospheric mercury in soil. The Science of The Total Environment, vol. 304, no. 1-3, pp. 209-214.  Wängberg, I., Barregard, L., Sällsten, G., Haeger-Eugensson, M., Munthe, J. & Sommar, J., 2005. Emissions, dispersion and human exposure of mercury from a Swedish chlor alkali plant. Atmospheric Environment, vol. 39, no. 39, pp. 7451-7458.   180 Wilson, K., Goldstein, A., Falge, E., Aubinet, M., Baldocchi, D., Berbigier, P., Bernhofer, C., Ceulemans, R., 2002. Energy balance closure at FLUXNET sites, Agricultural and Forest Meteorology, vol. 113, no. 1–4, pp. pp. 552-565.  Wilson, S.J., Steenhuisen, F., Pacyna, J.M. & Pacyna, E.G., 2006. Mapping the spatial distribution of global anthropogenic mercury atmospheric emission inventories. Atmospheric Environment, vol. 40, no. 24, pp. 4621-4632.  World Health Organization, 2003. Elemental mercury and Inorganic Mercury  Compounds:  Human Health Aspects.  http://www.inchem.org/documents/cicads/cicads/cicad50.htm.  Accessed May 2012.  World Health Organization Air quality guidelines for particulate matter, ozone, nitrogen dioxide and sulfur dioxide - Global update, 2005. - Summary of risk assessment. http://www.who.int/phe/health_topics/outdoorair/outdoorair_aqg/en/index.html. Accessed April, 2013.  World Meteorological Organization, 2010. WMO guide to meteorological instruments and methods of observation. http://www.wmo.int/pages/prog/www/IMOP/CIMO-Guide.html. No. 8. Accessed May, 2012. World Meteorological Organization, 2013. MeteoTerm terminology database. http://www.wmo.int/pages/prog/lsp/meteoterm_wmo_en.html. Accessed April 2013. Wunderground, 2013. http://www.wunderground.com/. Accessed April 2013. Yu, T., 1978. Determining the height of the nocturnal boundary layer. Journal of Applied Meteorology, vol. 17, pp. 28–33.   Zolnikov, T.R., 2012.  Limitations in small artisanal gold mining addressed by educational components paired with alternative mining methods. Science of the Total Environment, vol. 419, pp. 1-6. 

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.24.1-0103432/manifest

Comment

Related Items